?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fs/ecryptfs/file.c: introduce missing free
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blob336792ed58d7d9f674b4f7b993532ce0789cc911
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 mddev_t *mddev = q->queuedata;
219 int rv;
220 if (mddev == NULL || mddev->pers == NULL) {
221 bio_io_error(bio);
222 return 0;
223 }
224 rcu_read_lock();
225 if (mddev->suspended || mddev->barrier) {
226 DEFINE_WAIT(__wait);
227 for (;;) {
228 prepare_to_wait(&mddev->sb_wait, &__wait,
229 TASK_UNINTERRUPTIBLE);
230 if (!mddev->suspended && !mddev->barrier)
231 break;
232 rcu_read_unlock();
233 schedule();
234 rcu_read_lock();
235 }
236 finish_wait(&mddev->sb_wait, &__wait);
237 }
238 atomic_inc(&mddev->active_io);
239 rcu_read_unlock();
240 rv = mddev->pers->make_request(q, bio);
241 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
242 wake_up(&mddev->sb_wait);
243
244 return rv;
245 }
246
247 static void mddev_suspend(mddev_t *mddev)
248 {
249 BUG_ON(mddev->suspended);
250 mddev->suspended = 1;
251 synchronize_rcu();
252 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
253 mddev->pers->quiesce(mddev, 1);
254 md_unregister_thread(mddev->thread);
255 mddev->thread = NULL;
256 /* we now know that no code is executing in the personality module,
257 * except possibly the tail end of a ->bi_end_io function, but that
258 * is certain to complete before the module has a chance to get
259 * unloaded
260 */
261 }
262
263 static void mddev_resume(mddev_t *mddev)
264 {
265 mddev->suspended = 0;
266 wake_up(&mddev->sb_wait);
267 mddev->pers->quiesce(mddev, 0);
268 }
269
270 int mddev_congested(mddev_t *mddev, int bits)
271 {
272 if (mddev->barrier)
273 return 1;
274 return mddev->suspended;
275 }
276 EXPORT_SYMBOL(mddev_congested);
277
278 /*
279 * Generic barrier handling for md
280 */
281
282 #define POST_REQUEST_BARRIER ((void*)1)
283
284 static void md_end_barrier(struct bio *bio, int err)
285 {
286 mdk_rdev_t *rdev = bio->bi_private;
287 mddev_t *mddev = rdev->mddev;
288 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
289 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
290
291 rdev_dec_pending(rdev, mddev);
292
293 if (atomic_dec_and_test(&mddev->flush_pending)) {
294 if (mddev->barrier == POST_REQUEST_BARRIER) {
295 /* This was a post-request barrier */
296 mddev->barrier = NULL;
297 wake_up(&mddev->sb_wait);
298 } else
299 /* The pre-request barrier has finished */
300 schedule_work(&mddev->barrier_work);
301 }
302 bio_put(bio);
303 }
304
305 static void submit_barriers(mddev_t *mddev)
306 {
307 mdk_rdev_t *rdev;
308
309 rcu_read_lock();
310 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
311 if (rdev->raid_disk >= 0 &&
312 !test_bit(Faulty, &rdev->flags)) {
313 /* Take two references, one is dropped
314 * when request finishes, one after
315 * we reclaim rcu_read_lock
316 */
317 struct bio *bi;
318 atomic_inc(&rdev->nr_pending);
319 atomic_inc(&rdev->nr_pending);
320 rcu_read_unlock();
321 bi = bio_alloc(GFP_KERNEL, 0);
322 bi->bi_end_io = md_end_barrier;
323 bi->bi_private = rdev;
324 bi->bi_bdev = rdev->bdev;
325 atomic_inc(&mddev->flush_pending);
326 submit_bio(WRITE_BARRIER, bi);
327 rcu_read_lock();
328 rdev_dec_pending(rdev, mddev);
329 }
330 rcu_read_unlock();
331 }
332
333 static void md_submit_barrier(struct work_struct *ws)
334 {
335 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
336 struct bio *bio = mddev->barrier;
337
338 atomic_set(&mddev->flush_pending, 1);
339
340 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
341 bio_endio(bio, -EOPNOTSUPP);
342 else if (bio->bi_size == 0)
343 /* an empty barrier - all done */
344 bio_endio(bio, 0);
345 else {
346 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
347 if (mddev->pers->make_request(mddev->queue, bio))
348 generic_make_request(bio);
349 mddev->barrier = POST_REQUEST_BARRIER;
350 submit_barriers(mddev);
351 }
352 if (atomic_dec_and_test(&mddev->flush_pending)) {
353 mddev->barrier = NULL;
354 wake_up(&mddev->sb_wait);
355 }
356 }
357
358 void md_barrier_request(mddev_t *mddev, struct bio *bio)
359 {
360 spin_lock_irq(&mddev->write_lock);
361 wait_event_lock_irq(mddev->sb_wait,
362 !mddev->barrier,
363 mddev->write_lock, /*nothing*/);
364 mddev->barrier = bio;
365 spin_unlock_irq(&mddev->write_lock);
366
367 atomic_set(&mddev->flush_pending, 1);
368 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
369
370 submit_barriers(mddev);
371
372 if (atomic_dec_and_test(&mddev->flush_pending))
373 schedule_work(&mddev->barrier_work);
374 }
375 EXPORT_SYMBOL(md_barrier_request);
376
377 static inline mddev_t *mddev_get(mddev_t *mddev)
378 {
379 atomic_inc(&mddev->active);
380 return mddev;
381 }
382
383 static void mddev_delayed_delete(struct work_struct *ws);
384
385 static void mddev_put(mddev_t *mddev)
386 {
387 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
388 return;
389 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
390 mddev->ctime == 0 && !mddev->hold_active) {
391 /* Array is not configured at all, and not held active,
392 * so destroy it */
393 list_del(&mddev->all_mddevs);
394 if (mddev->gendisk) {
395 /* we did a probe so need to clean up.
396 * Call schedule_work inside the spinlock
397 * so that flush_scheduled_work() after
398 * mddev_find will succeed in waiting for the
399 * work to be done.
400 */
401 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
402 schedule_work(&mddev->del_work);
403 } else
404 kfree(mddev);
405 }
406 spin_unlock(&all_mddevs_lock);
407 }
408
409 static mddev_t * mddev_find(dev_t unit)
410 {
411 mddev_t *mddev, *new = NULL;
412
413 retry:
414 spin_lock(&all_mddevs_lock);
415
416 if (unit) {
417 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
418 if (mddev->unit == unit) {
419 mddev_get(mddev);
420 spin_unlock(&all_mddevs_lock);
421 kfree(new);
422 return mddev;
423 }
424
425 if (new) {
426 list_add(&new->all_mddevs, &all_mddevs);
427 spin_unlock(&all_mddevs_lock);
428 new->hold_active = UNTIL_IOCTL;
429 return new;
430 }
431 } else if (new) {
432 /* find an unused unit number */
433 static int next_minor = 512;
434 int start = next_minor;
435 int is_free = 0;
436 int dev = 0;
437 while (!is_free) {
438 dev = MKDEV(MD_MAJOR, next_minor);
439 next_minor++;
440 if (next_minor > MINORMASK)
441 next_minor = 0;
442 if (next_minor == start) {
443 /* Oh dear, all in use. */
444 spin_unlock(&all_mddevs_lock);
445 kfree(new);
446 return NULL;
447 }
448
449 is_free = 1;
450 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
451 if (mddev->unit == dev) {
452 is_free = 0;
453 break;
454 }
455 }
456 new->unit = dev;
457 new->md_minor = MINOR(dev);
458 new->hold_active = UNTIL_STOP;
459 list_add(&new->all_mddevs, &all_mddevs);
460 spin_unlock(&all_mddevs_lock);
461 return new;
462 }
463 spin_unlock(&all_mddevs_lock);
464
465 new = kzalloc(sizeof(*new), GFP_KERNEL);
466 if (!new)
467 return NULL;
468
469 new->unit = unit;
470 if (MAJOR(unit) == MD_MAJOR)
471 new->md_minor = MINOR(unit);
472 else
473 new->md_minor = MINOR(unit) >> MdpMinorShift;
474
475 mutex_init(&new->open_mutex);
476 mutex_init(&new->reconfig_mutex);
477 mutex_init(&new->bitmap_info.mutex);
478 INIT_LIST_HEAD(&new->disks);
479 INIT_LIST_HEAD(&new->all_mddevs);
480 init_timer(&new->safemode_timer);
481 atomic_set(&new->active, 1);
482 atomic_set(&new->openers, 0);
483 atomic_set(&new->active_io, 0);
484 spin_lock_init(&new->write_lock);
485 atomic_set(&new->flush_pending, 0);
486 init_waitqueue_head(&new->sb_wait);
487 init_waitqueue_head(&new->recovery_wait);
488 new->reshape_position = MaxSector;
489 new->resync_min = 0;
490 new->resync_max = MaxSector;
491 new->level = LEVEL_NONE;
492
493 goto retry;
494 }
495
496 static inline int mddev_lock(mddev_t * mddev)
497 {
498 return mutex_lock_interruptible(&mddev->reconfig_mutex);
499 }
500
501 static inline int mddev_is_locked(mddev_t *mddev)
502 {
503 return mutex_is_locked(&mddev->reconfig_mutex);
504 }
505
506 static inline int mddev_trylock(mddev_t * mddev)
507 {
508 return mutex_trylock(&mddev->reconfig_mutex);
509 }
510
511 static struct attribute_group md_redundancy_group;
512
513 static void mddev_unlock(mddev_t * mddev)
514 {
515 if (mddev->to_remove) {
516 /* These cannot be removed under reconfig_mutex as
517 * an access to the files will try to take reconfig_mutex
518 * while holding the file unremovable, which leads to
519 * a deadlock.
520 * So hold open_mutex instead - we are allowed to take
521 * it while holding reconfig_mutex, and md_run can
522 * use it to wait for the remove to complete.
523 */
524 struct attribute_group *to_remove = mddev->to_remove;
525 mddev->to_remove = NULL;
526 mutex_lock(&mddev->open_mutex);
527 mutex_unlock(&mddev->reconfig_mutex);
528
529 if (to_remove != &md_redundancy_group)
530 sysfs_remove_group(&mddev->kobj, to_remove);
531 if (mddev->pers == NULL ||
532 mddev->pers->sync_request == NULL) {
533 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
534 if (mddev->sysfs_action)
535 sysfs_put(mddev->sysfs_action);
536 mddev->sysfs_action = NULL;
537 }
538 mutex_unlock(&mddev->open_mutex);
539 } else
540 mutex_unlock(&mddev->reconfig_mutex);
541
542 md_wakeup_thread(mddev->thread);
543 }
544
545 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
546 {
547 mdk_rdev_t *rdev;
548
549 list_for_each_entry(rdev, &mddev->disks, same_set)
550 if (rdev->desc_nr == nr)
551 return rdev;
552
553 return NULL;
554 }
555
556 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
557 {
558 mdk_rdev_t *rdev;
559
560 list_for_each_entry(rdev, &mddev->disks, same_set)
561 if (rdev->bdev->bd_dev == dev)
562 return rdev;
563
564 return NULL;
565 }
566
567 static struct mdk_personality *find_pers(int level, char *clevel)
568 {
569 struct mdk_personality *pers;
570 list_for_each_entry(pers, &pers_list, list) {
571 if (level != LEVEL_NONE && pers->level == level)
572 return pers;
573 if (strcmp(pers->name, clevel)==0)
574 return pers;
575 }
576 return NULL;
577 }
578
579 /* return the offset of the super block in 512byte sectors */
580 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
581 {
582 sector_t num_sectors = bdev->bd_inode->i_size / 512;
583 return MD_NEW_SIZE_SECTORS(num_sectors);
584 }
585
586 static int alloc_disk_sb(mdk_rdev_t * rdev)
587 {
588 if (rdev->sb_page)
589 MD_BUG();
590
591 rdev->sb_page = alloc_page(GFP_KERNEL);
592 if (!rdev->sb_page) {
593 printk(KERN_ALERT "md: out of memory.\n");
594 return -ENOMEM;
595 }
596
597 return 0;
598 }
599
600 static void free_disk_sb(mdk_rdev_t * rdev)
601 {
602 if (rdev->sb_page) {
603 put_page(rdev->sb_page);
604 rdev->sb_loaded = 0;
605 rdev->sb_page = NULL;
606 rdev->sb_start = 0;
607 rdev->sectors = 0;
608 }
609 }
610
611
612 static void super_written(struct bio *bio, int error)
613 {
614 mdk_rdev_t *rdev = bio->bi_private;
615 mddev_t *mddev = rdev->mddev;
616
617 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
618 printk("md: super_written gets error=%d, uptodate=%d\n",
619 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
620 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
621 md_error(mddev, rdev);
622 }
623
624 if (atomic_dec_and_test(&mddev->pending_writes))
625 wake_up(&mddev->sb_wait);
626 bio_put(bio);
627 }
628
629 static void super_written_barrier(struct bio *bio, int error)
630 {
631 struct bio *bio2 = bio->bi_private;
632 mdk_rdev_t *rdev = bio2->bi_private;
633 mddev_t *mddev = rdev->mddev;
634
635 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
636 error == -EOPNOTSUPP) {
637 unsigned long flags;
638 /* barriers don't appear to be supported :-( */
639 set_bit(BarriersNotsupp, &rdev->flags);
640 mddev->barriers_work = 0;
641 spin_lock_irqsave(&mddev->write_lock, flags);
642 bio2->bi_next = mddev->biolist;
643 mddev->biolist = bio2;
644 spin_unlock_irqrestore(&mddev->write_lock, flags);
645 wake_up(&mddev->sb_wait);
646 bio_put(bio);
647 } else {
648 bio_put(bio2);
649 bio->bi_private = rdev;
650 super_written(bio, error);
651 }
652 }
653
654 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
655 sector_t sector, int size, struct page *page)
656 {
657 /* write first size bytes of page to sector of rdev
658 * Increment mddev->pending_writes before returning
659 * and decrement it on completion, waking up sb_wait
660 * if zero is reached.
661 * If an error occurred, call md_error
662 *
663 * As we might need to resubmit the request if BIO_RW_BARRIER
664 * causes ENOTSUPP, we allocate a spare bio...
665 */
666 struct bio *bio = bio_alloc(GFP_NOIO, 1);
667 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
668
669 bio->bi_bdev = rdev->bdev;
670 bio->bi_sector = sector;
671 bio_add_page(bio, page, size, 0);
672 bio->bi_private = rdev;
673 bio->bi_end_io = super_written;
674 bio->bi_rw = rw;
675
676 atomic_inc(&mddev->pending_writes);
677 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
678 struct bio *rbio;
679 rw |= (1<<BIO_RW_BARRIER);
680 rbio = bio_clone(bio, GFP_NOIO);
681 rbio->bi_private = bio;
682 rbio->bi_end_io = super_written_barrier;
683 submit_bio(rw, rbio);
684 } else
685 submit_bio(rw, bio);
686 }
687
688 void md_super_wait(mddev_t *mddev)
689 {
690 /* wait for all superblock writes that were scheduled to complete.
691 * if any had to be retried (due to BARRIER problems), retry them
692 */
693 DEFINE_WAIT(wq);
694 for(;;) {
695 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
696 if (atomic_read(&mddev->pending_writes)==0)
697 break;
698 while (mddev->biolist) {
699 struct bio *bio;
700 spin_lock_irq(&mddev->write_lock);
701 bio = mddev->biolist;
702 mddev->biolist = bio->bi_next ;
703 bio->bi_next = NULL;
704 spin_unlock_irq(&mddev->write_lock);
705 submit_bio(bio->bi_rw, bio);
706 }
707 schedule();
708 }
709 finish_wait(&mddev->sb_wait, &wq);
710 }
711
712 static void bi_complete(struct bio *bio, int error)
713 {
714 complete((struct completion*)bio->bi_private);
715 }
716
717 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
718 struct page *page, int rw)
719 {
720 struct bio *bio = bio_alloc(GFP_NOIO, 1);
721 struct completion event;
722 int ret;
723
724 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
725
726 bio->bi_bdev = bdev;
727 bio->bi_sector = sector;
728 bio_add_page(bio, page, size, 0);
729 init_completion(&event);
730 bio->bi_private = &event;
731 bio->bi_end_io = bi_complete;
732 submit_bio(rw, bio);
733 wait_for_completion(&event);
734
735 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
736 bio_put(bio);
737 return ret;
738 }
739 EXPORT_SYMBOL_GPL(sync_page_io);
740
741 static int read_disk_sb(mdk_rdev_t * rdev, int size)
742 {
743 char b[BDEVNAME_SIZE];
744 if (!rdev->sb_page) {
745 MD_BUG();
746 return -EINVAL;
747 }
748 if (rdev->sb_loaded)
749 return 0;
750
751
752 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
753 goto fail;
754 rdev->sb_loaded = 1;
755 return 0;
756
757 fail:
758 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
759 bdevname(rdev->bdev,b));
760 return -EINVAL;
761 }
762
763 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
764 {
765 return sb1->set_uuid0 == sb2->set_uuid0 &&
766 sb1->set_uuid1 == sb2->set_uuid1 &&
767 sb1->set_uuid2 == sb2->set_uuid2 &&
768 sb1->set_uuid3 == sb2->set_uuid3;
769 }
770
771 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
772 {
773 int ret;
774 mdp_super_t *tmp1, *tmp2;
775
776 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
777 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
778
779 if (!tmp1 || !tmp2) {
780 ret = 0;
781 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
782 goto abort;
783 }
784
785 *tmp1 = *sb1;
786 *tmp2 = *sb2;
787
788 /*
789 * nr_disks is not constant
790 */
791 tmp1->nr_disks = 0;
792 tmp2->nr_disks = 0;
793
794 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
795 abort:
796 kfree(tmp1);
797 kfree(tmp2);
798 return ret;
799 }
800
801
802 static u32 md_csum_fold(u32 csum)
803 {
804 csum = (csum & 0xffff) + (csum >> 16);
805 return (csum & 0xffff) + (csum >> 16);
806 }
807
808 static unsigned int calc_sb_csum(mdp_super_t * sb)
809 {
810 u64 newcsum = 0;
811 u32 *sb32 = (u32*)sb;
812 int i;
813 unsigned int disk_csum, csum;
814
815 disk_csum = sb->sb_csum;
816 sb->sb_csum = 0;
817
818 for (i = 0; i < MD_SB_BYTES/4 ; i++)
819 newcsum += sb32[i];
820 csum = (newcsum & 0xffffffff) + (newcsum>>32);
821
822
823 #ifdef CONFIG_ALPHA
824 /* This used to use csum_partial, which was wrong for several
825 * reasons including that different results are returned on
826 * different architectures. It isn't critical that we get exactly
827 * the same return value as before (we always csum_fold before
828 * testing, and that removes any differences). However as we
829 * know that csum_partial always returned a 16bit value on
830 * alphas, do a fold to maximise conformity to previous behaviour.
831 */
832 sb->sb_csum = md_csum_fold(disk_csum);
833 #else
834 sb->sb_csum = disk_csum;
835 #endif
836 return csum;
837 }
838
839
840 /*
841 * Handle superblock details.
842 * We want to be able to handle multiple superblock formats
843 * so we have a common interface to them all, and an array of
844 * different handlers.
845 * We rely on user-space to write the initial superblock, and support
846 * reading and updating of superblocks.
847 * Interface methods are:
848 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
849 * loads and validates a superblock on dev.
850 * if refdev != NULL, compare superblocks on both devices
851 * Return:
852 * 0 - dev has a superblock that is compatible with refdev
853 * 1 - dev has a superblock that is compatible and newer than refdev
854 * so dev should be used as the refdev in future
855 * -EINVAL superblock incompatible or invalid
856 * -othererror e.g. -EIO
857 *
858 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
859 * Verify that dev is acceptable into mddev.
860 * The first time, mddev->raid_disks will be 0, and data from
861 * dev should be merged in. Subsequent calls check that dev
862 * is new enough. Return 0 or -EINVAL
863 *
864 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
865 * Update the superblock for rdev with data in mddev
866 * This does not write to disc.
867 *
868 */
869
870 struct super_type {
871 char *name;
872 struct module *owner;
873 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
874 int minor_version);
875 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
876 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
877 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
878 sector_t num_sectors);
879 };
880
881 /*
882 * Check that the given mddev has no bitmap.
883 *
884 * This function is called from the run method of all personalities that do not
885 * support bitmaps. It prints an error message and returns non-zero if mddev
886 * has a bitmap. Otherwise, it returns 0.
887 *
888 */
889 int md_check_no_bitmap(mddev_t *mddev)
890 {
891 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
892 return 0;
893 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
894 mdname(mddev), mddev->pers->name);
895 return 1;
896 }
897 EXPORT_SYMBOL(md_check_no_bitmap);
898
899 /*
900 * load_super for 0.90.0
901 */
902 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
903 {
904 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
905 mdp_super_t *sb;
906 int ret;
907
908 /*
909 * Calculate the position of the superblock (512byte sectors),
910 * it's at the end of the disk.
911 *
912 * It also happens to be a multiple of 4Kb.
913 */
914 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
915
916 ret = read_disk_sb(rdev, MD_SB_BYTES);
917 if (ret) return ret;
918
919 ret = -EINVAL;
920
921 bdevname(rdev->bdev, b);
922 sb = (mdp_super_t*)page_address(rdev->sb_page);
923
924 if (sb->md_magic != MD_SB_MAGIC) {
925 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
926 b);
927 goto abort;
928 }
929
930 if (sb->major_version != 0 ||
931 sb->minor_version < 90 ||
932 sb->minor_version > 91) {
933 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
934 sb->major_version, sb->minor_version,
935 b);
936 goto abort;
937 }
938
939 if (sb->raid_disks <= 0)
940 goto abort;
941
942 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
943 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
944 b);
945 goto abort;
946 }
947
948 rdev->preferred_minor = sb->md_minor;
949 rdev->data_offset = 0;
950 rdev->sb_size = MD_SB_BYTES;
951
952 if (sb->level == LEVEL_MULTIPATH)
953 rdev->desc_nr = -1;
954 else
955 rdev->desc_nr = sb->this_disk.number;
956
957 if (!refdev) {
958 ret = 1;
959 } else {
960 __u64 ev1, ev2;
961 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
962 if (!uuid_equal(refsb, sb)) {
963 printk(KERN_WARNING "md: %s has different UUID to %s\n",
964 b, bdevname(refdev->bdev,b2));
965 goto abort;
966 }
967 if (!sb_equal(refsb, sb)) {
968 printk(KERN_WARNING "md: %s has same UUID"
969 " but different superblock to %s\n",
970 b, bdevname(refdev->bdev, b2));
971 goto abort;
972 }
973 ev1 = md_event(sb);
974 ev2 = md_event(refsb);
975 if (ev1 > ev2)
976 ret = 1;
977 else
978 ret = 0;
979 }
980 rdev->sectors = rdev->sb_start;
981
982 if (rdev->sectors < sb->size * 2 && sb->level > 1)
983 /* "this cannot possibly happen" ... */
984 ret = -EINVAL;
985
986 abort:
987 return ret;
988 }
989
990 /*
991 * validate_super for 0.90.0
992 */
993 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
994 {
995 mdp_disk_t *desc;
996 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
997 __u64 ev1 = md_event(sb);
998
999 rdev->raid_disk = -1;
1000 clear_bit(Faulty, &rdev->flags);
1001 clear_bit(In_sync, &rdev->flags);
1002 clear_bit(WriteMostly, &rdev->flags);
1003 clear_bit(BarriersNotsupp, &rdev->flags);
1004
1005 if (mddev->raid_disks == 0) {
1006 mddev->major_version = 0;
1007 mddev->minor_version = sb->minor_version;
1008 mddev->patch_version = sb->patch_version;
1009 mddev->external = 0;
1010 mddev->chunk_sectors = sb->chunk_size >> 9;
1011 mddev->ctime = sb->ctime;
1012 mddev->utime = sb->utime;
1013 mddev->level = sb->level;
1014 mddev->clevel[0] = 0;
1015 mddev->layout = sb->layout;
1016 mddev->raid_disks = sb->raid_disks;
1017 mddev->dev_sectors = sb->size * 2;
1018 mddev->events = ev1;
1019 mddev->bitmap_info.offset = 0;
1020 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1021
1022 if (mddev->minor_version >= 91) {
1023 mddev->reshape_position = sb->reshape_position;
1024 mddev->delta_disks = sb->delta_disks;
1025 mddev->new_level = sb->new_level;
1026 mddev->new_layout = sb->new_layout;
1027 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1028 } else {
1029 mddev->reshape_position = MaxSector;
1030 mddev->delta_disks = 0;
1031 mddev->new_level = mddev->level;
1032 mddev->new_layout = mddev->layout;
1033 mddev->new_chunk_sectors = mddev->chunk_sectors;
1034 }
1035
1036 if (sb->state & (1<<MD_SB_CLEAN))
1037 mddev->recovery_cp = MaxSector;
1038 else {
1039 if (sb->events_hi == sb->cp_events_hi &&
1040 sb->events_lo == sb->cp_events_lo) {
1041 mddev->recovery_cp = sb->recovery_cp;
1042 } else
1043 mddev->recovery_cp = 0;
1044 }
1045
1046 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1047 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1048 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1049 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1050
1051 mddev->max_disks = MD_SB_DISKS;
1052
1053 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1054 mddev->bitmap_info.file == NULL)
1055 mddev->bitmap_info.offset =
1056 mddev->bitmap_info.default_offset;
1057
1058 } else if (mddev->pers == NULL) {
1059 /* Insist on good event counter while assembling */
1060 ++ev1;
1061 if (ev1 < mddev->events)
1062 return -EINVAL;
1063 } else if (mddev->bitmap) {
1064 /* if adding to array with a bitmap, then we can accept an
1065 * older device ... but not too old.
1066 */
1067 if (ev1 < mddev->bitmap->events_cleared)
1068 return 0;
1069 } else {
1070 if (ev1 < mddev->events)
1071 /* just a hot-add of a new device, leave raid_disk at -1 */
1072 return 0;
1073 }
1074
1075 if (mddev->level != LEVEL_MULTIPATH) {
1076 desc = sb->disks + rdev->desc_nr;
1077
1078 if (desc->state & (1<<MD_DISK_FAULTY))
1079 set_bit(Faulty, &rdev->flags);
1080 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1081 desc->raid_disk < mddev->raid_disks */) {
1082 set_bit(In_sync, &rdev->flags);
1083 rdev->raid_disk = desc->raid_disk;
1084 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1085 /* active but not in sync implies recovery up to
1086 * reshape position. We don't know exactly where
1087 * that is, so set to zero for now */
1088 if (mddev->minor_version >= 91) {
1089 rdev->recovery_offset = 0;
1090 rdev->raid_disk = desc->raid_disk;
1091 }
1092 }
1093 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1094 set_bit(WriteMostly, &rdev->flags);
1095 } else /* MULTIPATH are always insync */
1096 set_bit(In_sync, &rdev->flags);
1097 return 0;
1098 }
1099
1100 /*
1101 * sync_super for 0.90.0
1102 */
1103 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1104 {
1105 mdp_super_t *sb;
1106 mdk_rdev_t *rdev2;
1107 int next_spare = mddev->raid_disks;
1108
1109
1110 /* make rdev->sb match mddev data..
1111 *
1112 * 1/ zero out disks
1113 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1114 * 3/ any empty disks < next_spare become removed
1115 *
1116 * disks[0] gets initialised to REMOVED because
1117 * we cannot be sure from other fields if it has
1118 * been initialised or not.
1119 */
1120 int i;
1121 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1122
1123 rdev->sb_size = MD_SB_BYTES;
1124
1125 sb = (mdp_super_t*)page_address(rdev->sb_page);
1126
1127 memset(sb, 0, sizeof(*sb));
1128
1129 sb->md_magic = MD_SB_MAGIC;
1130 sb->major_version = mddev->major_version;
1131 sb->patch_version = mddev->patch_version;
1132 sb->gvalid_words = 0; /* ignored */
1133 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1134 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1135 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1136 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1137
1138 sb->ctime = mddev->ctime;
1139 sb->level = mddev->level;
1140 sb->size = mddev->dev_sectors / 2;
1141 sb->raid_disks = mddev->raid_disks;
1142 sb->md_minor = mddev->md_minor;
1143 sb->not_persistent = 0;
1144 sb->utime = mddev->utime;
1145 sb->state = 0;
1146 sb->events_hi = (mddev->events>>32);
1147 sb->events_lo = (u32)mddev->events;
1148
1149 if (mddev->reshape_position == MaxSector)
1150 sb->minor_version = 90;
1151 else {
1152 sb->minor_version = 91;
1153 sb->reshape_position = mddev->reshape_position;
1154 sb->new_level = mddev->new_level;
1155 sb->delta_disks = mddev->delta_disks;
1156 sb->new_layout = mddev->new_layout;
1157 sb->new_chunk = mddev->new_chunk_sectors << 9;
1158 }
1159 mddev->minor_version = sb->minor_version;
1160 if (mddev->in_sync)
1161 {
1162 sb->recovery_cp = mddev->recovery_cp;
1163 sb->cp_events_hi = (mddev->events>>32);
1164 sb->cp_events_lo = (u32)mddev->events;
1165 if (mddev->recovery_cp == MaxSector)
1166 sb->state = (1<< MD_SB_CLEAN);
1167 } else
1168 sb->recovery_cp = 0;
1169
1170 sb->layout = mddev->layout;
1171 sb->chunk_size = mddev->chunk_sectors << 9;
1172
1173 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1174 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1175
1176 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1177 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1178 mdp_disk_t *d;
1179 int desc_nr;
1180 int is_active = test_bit(In_sync, &rdev2->flags);
1181
1182 if (rdev2->raid_disk >= 0 &&
1183 sb->minor_version >= 91)
1184 /* we have nowhere to store the recovery_offset,
1185 * but if it is not below the reshape_position,
1186 * we can piggy-back on that.
1187 */
1188 is_active = 1;
1189 if (rdev2->raid_disk < 0 ||
1190 test_bit(Faulty, &rdev2->flags))
1191 is_active = 0;
1192 if (is_active)
1193 desc_nr = rdev2->raid_disk;
1194 else
1195 desc_nr = next_spare++;
1196 rdev2->desc_nr = desc_nr;
1197 d = &sb->disks[rdev2->desc_nr];
1198 nr_disks++;
1199 d->number = rdev2->desc_nr;
1200 d->major = MAJOR(rdev2->bdev->bd_dev);
1201 d->minor = MINOR(rdev2->bdev->bd_dev);
1202 if (is_active)
1203 d->raid_disk = rdev2->raid_disk;
1204 else
1205 d->raid_disk = rdev2->desc_nr; /* compatibility */
1206 if (test_bit(Faulty, &rdev2->flags))
1207 d->state = (1<<MD_DISK_FAULTY);
1208 else if (is_active) {
1209 d->state = (1<<MD_DISK_ACTIVE);
1210 if (test_bit(In_sync, &rdev2->flags))
1211 d->state |= (1<<MD_DISK_SYNC);
1212 active++;
1213 working++;
1214 } else {
1215 d->state = 0;
1216 spare++;
1217 working++;
1218 }
1219 if (test_bit(WriteMostly, &rdev2->flags))
1220 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1221 }
1222 /* now set the "removed" and "faulty" bits on any missing devices */
1223 for (i=0 ; i < mddev->raid_disks ; i++) {
1224 mdp_disk_t *d = &sb->disks[i];
1225 if (d->state == 0 && d->number == 0) {
1226 d->number = i;
1227 d->raid_disk = i;
1228 d->state = (1<<MD_DISK_REMOVED);
1229 d->state |= (1<<MD_DISK_FAULTY);
1230 failed++;
1231 }
1232 }
1233 sb->nr_disks = nr_disks;
1234 sb->active_disks = active;
1235 sb->working_disks = working;
1236 sb->failed_disks = failed;
1237 sb->spare_disks = spare;
1238
1239 sb->this_disk = sb->disks[rdev->desc_nr];
1240 sb->sb_csum = calc_sb_csum(sb);
1241 }
1242
1243 /*
1244 * rdev_size_change for 0.90.0
1245 */
1246 static unsigned long long
1247 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1248 {
1249 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1250 return 0; /* component must fit device */
1251 if (rdev->mddev->bitmap_info.offset)
1252 return 0; /* can't move bitmap */
1253 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1254 if (!num_sectors || num_sectors > rdev->sb_start)
1255 num_sectors = rdev->sb_start;
1256 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1257 rdev->sb_page);
1258 md_super_wait(rdev->mddev);
1259 return num_sectors / 2; /* kB for sysfs */
1260 }
1261
1262
1263 /*
1264 * version 1 superblock
1265 */
1266
1267 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1268 {
1269 __le32 disk_csum;
1270 u32 csum;
1271 unsigned long long newcsum;
1272 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1273 __le32 *isuper = (__le32*)sb;
1274 int i;
1275
1276 disk_csum = sb->sb_csum;
1277 sb->sb_csum = 0;
1278 newcsum = 0;
1279 for (i=0; size>=4; size -= 4 )
1280 newcsum += le32_to_cpu(*isuper++);
1281
1282 if (size == 2)
1283 newcsum += le16_to_cpu(*(__le16*) isuper);
1284
1285 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1286 sb->sb_csum = disk_csum;
1287 return cpu_to_le32(csum);
1288 }
1289
1290 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1291 {
1292 struct mdp_superblock_1 *sb;
1293 int ret;
1294 sector_t sb_start;
1295 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1296 int bmask;
1297
1298 /*
1299 * Calculate the position of the superblock in 512byte sectors.
1300 * It is always aligned to a 4K boundary and
1301 * depeding on minor_version, it can be:
1302 * 0: At least 8K, but less than 12K, from end of device
1303 * 1: At start of device
1304 * 2: 4K from start of device.
1305 */
1306 switch(minor_version) {
1307 case 0:
1308 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1309 sb_start -= 8*2;
1310 sb_start &= ~(sector_t)(4*2-1);
1311 break;
1312 case 1:
1313 sb_start = 0;
1314 break;
1315 case 2:
1316 sb_start = 8;
1317 break;
1318 default:
1319 return -EINVAL;
1320 }
1321 rdev->sb_start = sb_start;
1322
1323 /* superblock is rarely larger than 1K, but it can be larger,
1324 * and it is safe to read 4k, so we do that
1325 */
1326 ret = read_disk_sb(rdev, 4096);
1327 if (ret) return ret;
1328
1329
1330 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1331
1332 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1333 sb->major_version != cpu_to_le32(1) ||
1334 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1335 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1336 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1337 return -EINVAL;
1338
1339 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1340 printk("md: invalid superblock checksum on %s\n",
1341 bdevname(rdev->bdev,b));
1342 return -EINVAL;
1343 }
1344 if (le64_to_cpu(sb->data_size) < 10) {
1345 printk("md: data_size too small on %s\n",
1346 bdevname(rdev->bdev,b));
1347 return -EINVAL;
1348 }
1349
1350 rdev->preferred_minor = 0xffff;
1351 rdev->data_offset = le64_to_cpu(sb->data_offset);
1352 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1353
1354 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1355 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1356 if (rdev->sb_size & bmask)
1357 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1358
1359 if (minor_version
1360 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1361 return -EINVAL;
1362
1363 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1364 rdev->desc_nr = -1;
1365 else
1366 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1367
1368 if (!refdev) {
1369 ret = 1;
1370 } else {
1371 __u64 ev1, ev2;
1372 struct mdp_superblock_1 *refsb =
1373 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1374
1375 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1376 sb->level != refsb->level ||
1377 sb->layout != refsb->layout ||
1378 sb->chunksize != refsb->chunksize) {
1379 printk(KERN_WARNING "md: %s has strangely different"
1380 " superblock to %s\n",
1381 bdevname(rdev->bdev,b),
1382 bdevname(refdev->bdev,b2));
1383 return -EINVAL;
1384 }
1385 ev1 = le64_to_cpu(sb->events);
1386 ev2 = le64_to_cpu(refsb->events);
1387
1388 if (ev1 > ev2)
1389 ret = 1;
1390 else
1391 ret = 0;
1392 }
1393 if (minor_version)
1394 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1395 le64_to_cpu(sb->data_offset);
1396 else
1397 rdev->sectors = rdev->sb_start;
1398 if (rdev->sectors < le64_to_cpu(sb->data_size))
1399 return -EINVAL;
1400 rdev->sectors = le64_to_cpu(sb->data_size);
1401 if (le64_to_cpu(sb->size) > rdev->sectors)
1402 return -EINVAL;
1403 return ret;
1404 }
1405
1406 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1407 {
1408 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1409 __u64 ev1 = le64_to_cpu(sb->events);
1410
1411 rdev->raid_disk = -1;
1412 clear_bit(Faulty, &rdev->flags);
1413 clear_bit(In_sync, &rdev->flags);
1414 clear_bit(WriteMostly, &rdev->flags);
1415 clear_bit(BarriersNotsupp, &rdev->flags);
1416
1417 if (mddev->raid_disks == 0) {
1418 mddev->major_version = 1;
1419 mddev->patch_version = 0;
1420 mddev->external = 0;
1421 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1422 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1423 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1424 mddev->level = le32_to_cpu(sb->level);
1425 mddev->clevel[0] = 0;
1426 mddev->layout = le32_to_cpu(sb->layout);
1427 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1428 mddev->dev_sectors = le64_to_cpu(sb->size);
1429 mddev->events = ev1;
1430 mddev->bitmap_info.offset = 0;
1431 mddev->bitmap_info.default_offset = 1024 >> 9;
1432
1433 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1434 memcpy(mddev->uuid, sb->set_uuid, 16);
1435
1436 mddev->max_disks = (4096-256)/2;
1437
1438 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1439 mddev->bitmap_info.file == NULL )
1440 mddev->bitmap_info.offset =
1441 (__s32)le32_to_cpu(sb->bitmap_offset);
1442
1443 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1444 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1445 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1446 mddev->new_level = le32_to_cpu(sb->new_level);
1447 mddev->new_layout = le32_to_cpu(sb->new_layout);
1448 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1449 } else {
1450 mddev->reshape_position = MaxSector;
1451 mddev->delta_disks = 0;
1452 mddev->new_level = mddev->level;
1453 mddev->new_layout = mddev->layout;
1454 mddev->new_chunk_sectors = mddev->chunk_sectors;
1455 }
1456
1457 } else if (mddev->pers == NULL) {
1458 /* Insist of good event counter while assembling */
1459 ++ev1;
1460 if (ev1 < mddev->events)
1461 return -EINVAL;
1462 } else if (mddev->bitmap) {
1463 /* If adding to array with a bitmap, then we can accept an
1464 * older device, but not too old.
1465 */
1466 if (ev1 < mddev->bitmap->events_cleared)
1467 return 0;
1468 } else {
1469 if (ev1 < mddev->events)
1470 /* just a hot-add of a new device, leave raid_disk at -1 */
1471 return 0;
1472 }
1473 if (mddev->level != LEVEL_MULTIPATH) {
1474 int role;
1475 if (rdev->desc_nr < 0 ||
1476 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1477 role = 0xffff;
1478 rdev->desc_nr = -1;
1479 } else
1480 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1481 switch(role) {
1482 case 0xffff: /* spare */
1483 break;
1484 case 0xfffe: /* faulty */
1485 set_bit(Faulty, &rdev->flags);
1486 break;
1487 default:
1488 if ((le32_to_cpu(sb->feature_map) &
1489 MD_FEATURE_RECOVERY_OFFSET))
1490 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1491 else
1492 set_bit(In_sync, &rdev->flags);
1493 rdev->raid_disk = role;
1494 break;
1495 }
1496 if (sb->devflags & WriteMostly1)
1497 set_bit(WriteMostly, &rdev->flags);
1498 } else /* MULTIPATH are always insync */
1499 set_bit(In_sync, &rdev->flags);
1500
1501 return 0;
1502 }
1503
1504 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1505 {
1506 struct mdp_superblock_1 *sb;
1507 mdk_rdev_t *rdev2;
1508 int max_dev, i;
1509 /* make rdev->sb match mddev and rdev data. */
1510
1511 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1512
1513 sb->feature_map = 0;
1514 sb->pad0 = 0;
1515 sb->recovery_offset = cpu_to_le64(0);
1516 memset(sb->pad1, 0, sizeof(sb->pad1));
1517 memset(sb->pad2, 0, sizeof(sb->pad2));
1518 memset(sb->pad3, 0, sizeof(sb->pad3));
1519
1520 sb->utime = cpu_to_le64((__u64)mddev->utime);
1521 sb->events = cpu_to_le64(mddev->events);
1522 if (mddev->in_sync)
1523 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1524 else
1525 sb->resync_offset = cpu_to_le64(0);
1526
1527 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1528
1529 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1530 sb->size = cpu_to_le64(mddev->dev_sectors);
1531 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1532 sb->level = cpu_to_le32(mddev->level);
1533 sb->layout = cpu_to_le32(mddev->layout);
1534
1535 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1536 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1537 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1538 }
1539
1540 if (rdev->raid_disk >= 0 &&
1541 !test_bit(In_sync, &rdev->flags)) {
1542 sb->feature_map |=
1543 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1544 sb->recovery_offset =
1545 cpu_to_le64(rdev->recovery_offset);
1546 }
1547
1548 if (mddev->reshape_position != MaxSector) {
1549 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1550 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1551 sb->new_layout = cpu_to_le32(mddev->new_layout);
1552 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1553 sb->new_level = cpu_to_le32(mddev->new_level);
1554 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1555 }
1556
1557 max_dev = 0;
1558 list_for_each_entry(rdev2, &mddev->disks, same_set)
1559 if (rdev2->desc_nr+1 > max_dev)
1560 max_dev = rdev2->desc_nr+1;
1561
1562 if (max_dev > le32_to_cpu(sb->max_dev)) {
1563 int bmask;
1564 sb->max_dev = cpu_to_le32(max_dev);
1565 rdev->sb_size = max_dev * 2 + 256;
1566 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1567 if (rdev->sb_size & bmask)
1568 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1569 }
1570 for (i=0; i<max_dev;i++)
1571 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1572
1573 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1574 i = rdev2->desc_nr;
1575 if (test_bit(Faulty, &rdev2->flags))
1576 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1577 else if (test_bit(In_sync, &rdev2->flags))
1578 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1579 else if (rdev2->raid_disk >= 0)
1580 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1581 else
1582 sb->dev_roles[i] = cpu_to_le16(0xffff);
1583 }
1584
1585 sb->sb_csum = calc_sb_1_csum(sb);
1586 }
1587
1588 static unsigned long long
1589 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1590 {
1591 struct mdp_superblock_1 *sb;
1592 sector_t max_sectors;
1593 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1594 return 0; /* component must fit device */
1595 if (rdev->sb_start < rdev->data_offset) {
1596 /* minor versions 1 and 2; superblock before data */
1597 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1598 max_sectors -= rdev->data_offset;
1599 if (!num_sectors || num_sectors > max_sectors)
1600 num_sectors = max_sectors;
1601 } else if (rdev->mddev->bitmap_info.offset) {
1602 /* minor version 0 with bitmap we can't move */
1603 return 0;
1604 } else {
1605 /* minor version 0; superblock after data */
1606 sector_t sb_start;
1607 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1608 sb_start &= ~(sector_t)(4*2 - 1);
1609 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1610 if (!num_sectors || num_sectors > max_sectors)
1611 num_sectors = max_sectors;
1612 rdev->sb_start = sb_start;
1613 }
1614 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1615 sb->data_size = cpu_to_le64(num_sectors);
1616 sb->super_offset = rdev->sb_start;
1617 sb->sb_csum = calc_sb_1_csum(sb);
1618 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1619 rdev->sb_page);
1620 md_super_wait(rdev->mddev);
1621 return num_sectors / 2; /* kB for sysfs */
1622 }
1623
1624 static struct super_type super_types[] = {
1625 [0] = {
1626 .name = "0.90.0",
1627 .owner = THIS_MODULE,
1628 .load_super = super_90_load,
1629 .validate_super = super_90_validate,
1630 .sync_super = super_90_sync,
1631 .rdev_size_change = super_90_rdev_size_change,
1632 },
1633 [1] = {
1634 .name = "md-1",
1635 .owner = THIS_MODULE,
1636 .load_super = super_1_load,
1637 .validate_super = super_1_validate,
1638 .sync_super = super_1_sync,
1639 .rdev_size_change = super_1_rdev_size_change,
1640 },
1641 };
1642
1643 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1644 {
1645 mdk_rdev_t *rdev, *rdev2;
1646
1647 rcu_read_lock();
1648 rdev_for_each_rcu(rdev, mddev1)
1649 rdev_for_each_rcu(rdev2, mddev2)
1650 if (rdev->bdev->bd_contains ==
1651 rdev2->bdev->bd_contains) {
1652 rcu_read_unlock();
1653 return 1;
1654 }
1655 rcu_read_unlock();
1656 return 0;
1657 }
1658
1659 static LIST_HEAD(pending_raid_disks);
1660
1661 /*
1662 * Try to register data integrity profile for an mddev
1663 *
1664 * This is called when an array is started and after a disk has been kicked
1665 * from the array. It only succeeds if all working and active component devices
1666 * are integrity capable with matching profiles.
1667 */
1668 int md_integrity_register(mddev_t *mddev)
1669 {
1670 mdk_rdev_t *rdev, *reference = NULL;
1671
1672 if (list_empty(&mddev->disks))
1673 return 0; /* nothing to do */
1674 if (blk_get_integrity(mddev->gendisk))
1675 return 0; /* already registered */
1676 list_for_each_entry(rdev, &mddev->disks, same_set) {
1677 /* skip spares and non-functional disks */
1678 if (test_bit(Faulty, &rdev->flags))
1679 continue;
1680 if (rdev->raid_disk < 0)
1681 continue;
1682 /*
1683 * If at least one rdev is not integrity capable, we can not
1684 * enable data integrity for the md device.
1685 */
1686 if (!bdev_get_integrity(rdev->bdev))
1687 return -EINVAL;
1688 if (!reference) {
1689 /* Use the first rdev as the reference */
1690 reference = rdev;
1691 continue;
1692 }
1693 /* does this rdev's profile match the reference profile? */
1694 if (blk_integrity_compare(reference->bdev->bd_disk,
1695 rdev->bdev->bd_disk) < 0)
1696 return -EINVAL;
1697 }
1698 /*
1699 * All component devices are integrity capable and have matching
1700 * profiles, register the common profile for the md device.
1701 */
1702 if (blk_integrity_register(mddev->gendisk,
1703 bdev_get_integrity(reference->bdev)) != 0) {
1704 printk(KERN_ERR "md: failed to register integrity for %s\n",
1705 mdname(mddev));
1706 return -EINVAL;
1707 }
1708 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1709 mdname(mddev));
1710 return 0;
1711 }
1712 EXPORT_SYMBOL(md_integrity_register);
1713
1714 /* Disable data integrity if non-capable/non-matching disk is being added */
1715 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1716 {
1717 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1718 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1719
1720 if (!bi_mddev) /* nothing to do */
1721 return;
1722 if (rdev->raid_disk < 0) /* skip spares */
1723 return;
1724 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1725 rdev->bdev->bd_disk) >= 0)
1726 return;
1727 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1728 blk_integrity_unregister(mddev->gendisk);
1729 }
1730 EXPORT_SYMBOL(md_integrity_add_rdev);
1731
1732 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1733 {
1734 char b[BDEVNAME_SIZE];
1735 struct kobject *ko;
1736 char *s;
1737 int err;
1738
1739 if (rdev->mddev) {
1740 MD_BUG();
1741 return -EINVAL;
1742 }
1743
1744 /* prevent duplicates */
1745 if (find_rdev(mddev, rdev->bdev->bd_dev))
1746 return -EEXIST;
1747
1748 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1749 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1750 rdev->sectors < mddev->dev_sectors)) {
1751 if (mddev->pers) {
1752 /* Cannot change size, so fail
1753 * If mddev->level <= 0, then we don't care
1754 * about aligning sizes (e.g. linear)
1755 */
1756 if (mddev->level > 0)
1757 return -ENOSPC;
1758 } else
1759 mddev->dev_sectors = rdev->sectors;
1760 }
1761
1762 /* Verify rdev->desc_nr is unique.
1763 * If it is -1, assign a free number, else
1764 * check number is not in use
1765 */
1766 if (rdev->desc_nr < 0) {
1767 int choice = 0;
1768 if (mddev->pers) choice = mddev->raid_disks;
1769 while (find_rdev_nr(mddev, choice))
1770 choice++;
1771 rdev->desc_nr = choice;
1772 } else {
1773 if (find_rdev_nr(mddev, rdev->desc_nr))
1774 return -EBUSY;
1775 }
1776 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1777 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1778 mdname(mddev), mddev->max_disks);
1779 return -EBUSY;
1780 }
1781 bdevname(rdev->bdev,b);
1782 while ( (s=strchr(b, '/')) != NULL)
1783 *s = '!';
1784
1785 rdev->mddev = mddev;
1786 printk(KERN_INFO "md: bind<%s>\n", b);
1787
1788 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1789 goto fail;
1790
1791 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1792 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1793 kobject_del(&rdev->kobj);
1794 goto fail;
1795 }
1796 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1797
1798 list_add_rcu(&rdev->same_set, &mddev->disks);
1799 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1800
1801 /* May as well allow recovery to be retried once */
1802 mddev->recovery_disabled = 0;
1803
1804 return 0;
1805
1806 fail:
1807 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1808 b, mdname(mddev));
1809 return err;
1810 }
1811
1812 static void md_delayed_delete(struct work_struct *ws)
1813 {
1814 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1815 kobject_del(&rdev->kobj);
1816 kobject_put(&rdev->kobj);
1817 }
1818
1819 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1820 {
1821 char b[BDEVNAME_SIZE];
1822 if (!rdev->mddev) {
1823 MD_BUG();
1824 return;
1825 }
1826 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1827 list_del_rcu(&rdev->same_set);
1828 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1829 rdev->mddev = NULL;
1830 sysfs_remove_link(&rdev->kobj, "block");
1831 sysfs_put(rdev->sysfs_state);
1832 rdev->sysfs_state = NULL;
1833 /* We need to delay this, otherwise we can deadlock when
1834 * writing to 'remove' to "dev/state". We also need
1835 * to delay it due to rcu usage.
1836 */
1837 synchronize_rcu();
1838 INIT_WORK(&rdev->del_work, md_delayed_delete);
1839 kobject_get(&rdev->kobj);
1840 schedule_work(&rdev->del_work);
1841 }
1842
1843 /*
1844 * prevent the device from being mounted, repartitioned or
1845 * otherwise reused by a RAID array (or any other kernel
1846 * subsystem), by bd_claiming the device.
1847 */
1848 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1849 {
1850 int err = 0;
1851 struct block_device *bdev;
1852 char b[BDEVNAME_SIZE];
1853
1854 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1855 if (IS_ERR(bdev)) {
1856 printk(KERN_ERR "md: could not open %s.\n",
1857 __bdevname(dev, b));
1858 return PTR_ERR(bdev);
1859 }
1860 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1861 if (err) {
1862 printk(KERN_ERR "md: could not bd_claim %s.\n",
1863 bdevname(bdev, b));
1864 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1865 return err;
1866 }
1867 if (!shared)
1868 set_bit(AllReserved, &rdev->flags);
1869 rdev->bdev = bdev;
1870 return err;
1871 }
1872
1873 static void unlock_rdev(mdk_rdev_t *rdev)
1874 {
1875 struct block_device *bdev = rdev->bdev;
1876 rdev->bdev = NULL;
1877 if (!bdev)
1878 MD_BUG();
1879 bd_release(bdev);
1880 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1881 }
1882
1883 void md_autodetect_dev(dev_t dev);
1884
1885 static void export_rdev(mdk_rdev_t * rdev)
1886 {
1887 char b[BDEVNAME_SIZE];
1888 printk(KERN_INFO "md: export_rdev(%s)\n",
1889 bdevname(rdev->bdev,b));
1890 if (rdev->mddev)
1891 MD_BUG();
1892 free_disk_sb(rdev);
1893 #ifndef MODULE
1894 if (test_bit(AutoDetected, &rdev->flags))
1895 md_autodetect_dev(rdev->bdev->bd_dev);
1896 #endif
1897 unlock_rdev(rdev);
1898 kobject_put(&rdev->kobj);
1899 }
1900
1901 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1902 {
1903 unbind_rdev_from_array(rdev);
1904 export_rdev(rdev);
1905 }
1906
1907 static void export_array(mddev_t *mddev)
1908 {
1909 mdk_rdev_t *rdev, *tmp;
1910
1911 rdev_for_each(rdev, tmp, mddev) {
1912 if (!rdev->mddev) {
1913 MD_BUG();
1914 continue;
1915 }
1916 kick_rdev_from_array(rdev);
1917 }
1918 if (!list_empty(&mddev->disks))
1919 MD_BUG();
1920 mddev->raid_disks = 0;
1921 mddev->major_version = 0;
1922 }
1923
1924 static void print_desc(mdp_disk_t *desc)
1925 {
1926 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1927 desc->major,desc->minor,desc->raid_disk,desc->state);
1928 }
1929
1930 static void print_sb_90(mdp_super_t *sb)
1931 {
1932 int i;
1933
1934 printk(KERN_INFO
1935 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1936 sb->major_version, sb->minor_version, sb->patch_version,
1937 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1938 sb->ctime);
1939 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1940 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1941 sb->md_minor, sb->layout, sb->chunk_size);
1942 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1943 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1944 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1945 sb->failed_disks, sb->spare_disks,
1946 sb->sb_csum, (unsigned long)sb->events_lo);
1947
1948 printk(KERN_INFO);
1949 for (i = 0; i < MD_SB_DISKS; i++) {
1950 mdp_disk_t *desc;
1951
1952 desc = sb->disks + i;
1953 if (desc->number || desc->major || desc->minor ||
1954 desc->raid_disk || (desc->state && (desc->state != 4))) {
1955 printk(" D %2d: ", i);
1956 print_desc(desc);
1957 }
1958 }
1959 printk(KERN_INFO "md: THIS: ");
1960 print_desc(&sb->this_disk);
1961 }
1962
1963 static void print_sb_1(struct mdp_superblock_1 *sb)
1964 {
1965 __u8 *uuid;
1966
1967 uuid = sb->set_uuid;
1968 printk(KERN_INFO
1969 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1970 "md: Name: \"%s\" CT:%llu\n",
1971 le32_to_cpu(sb->major_version),
1972 le32_to_cpu(sb->feature_map),
1973 uuid,
1974 sb->set_name,
1975 (unsigned long long)le64_to_cpu(sb->ctime)
1976 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1977
1978 uuid = sb->device_uuid;
1979 printk(KERN_INFO
1980 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1981 " RO:%llu\n"
1982 "md: Dev:%08x UUID: %pU\n"
1983 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1984 "md: (MaxDev:%u) \n",
1985 le32_to_cpu(sb->level),
1986 (unsigned long long)le64_to_cpu(sb->size),
1987 le32_to_cpu(sb->raid_disks),
1988 le32_to_cpu(sb->layout),
1989 le32_to_cpu(sb->chunksize),
1990 (unsigned long long)le64_to_cpu(sb->data_offset),
1991 (unsigned long long)le64_to_cpu(sb->data_size),
1992 (unsigned long long)le64_to_cpu(sb->super_offset),
1993 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1994 le32_to_cpu(sb->dev_number),
1995 uuid,
1996 sb->devflags,
1997 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1998 (unsigned long long)le64_to_cpu(sb->events),
1999 (unsigned long long)le64_to_cpu(sb->resync_offset),
2000 le32_to_cpu(sb->sb_csum),
2001 le32_to_cpu(sb->max_dev)
2002 );
2003 }
2004
2005 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2006 {
2007 char b[BDEVNAME_SIZE];
2008 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2009 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2010 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2011 rdev->desc_nr);
2012 if (rdev->sb_loaded) {
2013 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2014 switch (major_version) {
2015 case 0:
2016 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2017 break;
2018 case 1:
2019 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2020 break;
2021 }
2022 } else
2023 printk(KERN_INFO "md: no rdev superblock!\n");
2024 }
2025
2026 static void md_print_devices(void)
2027 {
2028 struct list_head *tmp;
2029 mdk_rdev_t *rdev;
2030 mddev_t *mddev;
2031 char b[BDEVNAME_SIZE];
2032
2033 printk("\n");
2034 printk("md: **********************************\n");
2035 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2036 printk("md: **********************************\n");
2037 for_each_mddev(mddev, tmp) {
2038
2039 if (mddev->bitmap)
2040 bitmap_print_sb(mddev->bitmap);
2041 else
2042 printk("%s: ", mdname(mddev));
2043 list_for_each_entry(rdev, &mddev->disks, same_set)
2044 printk("<%s>", bdevname(rdev->bdev,b));
2045 printk("\n");
2046
2047 list_for_each_entry(rdev, &mddev->disks, same_set)
2048 print_rdev(rdev, mddev->major_version);
2049 }
2050 printk("md: **********************************\n");
2051 printk("\n");
2052 }
2053
2054
2055 static void sync_sbs(mddev_t * mddev, int nospares)
2056 {
2057 /* Update each superblock (in-memory image), but
2058 * if we are allowed to, skip spares which already
2059 * have the right event counter, or have one earlier
2060 * (which would mean they aren't being marked as dirty
2061 * with the rest of the array)
2062 */
2063 mdk_rdev_t *rdev;
2064
2065 /* First make sure individual recovery_offsets are correct */
2066 list_for_each_entry(rdev, &mddev->disks, same_set) {
2067 if (rdev->raid_disk >= 0 &&
2068 !test_bit(In_sync, &rdev->flags) &&
2069 mddev->curr_resync_completed > rdev->recovery_offset)
2070 rdev->recovery_offset = mddev->curr_resync_completed;
2071
2072 }
2073 list_for_each_entry(rdev, &mddev->disks, same_set) {
2074 if (rdev->sb_events == mddev->events ||
2075 (nospares &&
2076 rdev->raid_disk < 0 &&
2077 (rdev->sb_events&1)==0 &&
2078 rdev->sb_events+1 == mddev->events)) {
2079 /* Don't update this superblock */
2080 rdev->sb_loaded = 2;
2081 } else {
2082 super_types[mddev->major_version].
2083 sync_super(mddev, rdev);
2084 rdev->sb_loaded = 1;
2085 }
2086 }
2087 }
2088
2089 static void md_update_sb(mddev_t * mddev, int force_change)
2090 {
2091 mdk_rdev_t *rdev;
2092 int sync_req;
2093 int nospares = 0;
2094
2095 mddev->utime = get_seconds();
2096 if (mddev->external)
2097 return;
2098 repeat:
2099 spin_lock_irq(&mddev->write_lock);
2100
2101 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2102 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2103 force_change = 1;
2104 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2105 /* just a clean<-> dirty transition, possibly leave spares alone,
2106 * though if events isn't the right even/odd, we will have to do
2107 * spares after all
2108 */
2109 nospares = 1;
2110 if (force_change)
2111 nospares = 0;
2112 if (mddev->degraded)
2113 /* If the array is degraded, then skipping spares is both
2114 * dangerous and fairly pointless.
2115 * Dangerous because a device that was removed from the array
2116 * might have a event_count that still looks up-to-date,
2117 * so it can be re-added without a resync.
2118 * Pointless because if there are any spares to skip,
2119 * then a recovery will happen and soon that array won't
2120 * be degraded any more and the spare can go back to sleep then.
2121 */
2122 nospares = 0;
2123
2124 sync_req = mddev->in_sync;
2125
2126 /* If this is just a dirty<->clean transition, and the array is clean
2127 * and 'events' is odd, we can roll back to the previous clean state */
2128 if (nospares
2129 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2130 && (mddev->events & 1)
2131 && mddev->events != 1)
2132 mddev->events--;
2133 else {
2134 /* otherwise we have to go forward and ... */
2135 mddev->events ++;
2136 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2137 /* .. if the array isn't clean, an 'even' event must also go
2138 * to spares. */
2139 if ((mddev->events&1)==0) {
2140 nospares = 0;
2141 sync_req = 2; /* force a second update to get the
2142 * even/odd in sync */
2143 }
2144 } else {
2145 /* otherwise an 'odd' event must go to spares */
2146 if ((mddev->events&1)) {
2147 nospares = 0;
2148 sync_req = 2; /* force a second update to get the
2149 * even/odd in sync */
2150 }
2151 }
2152 }
2153
2154 if (!mddev->events) {
2155 /*
2156 * oops, this 64-bit counter should never wrap.
2157 * Either we are in around ~1 trillion A.C., assuming
2158 * 1 reboot per second, or we have a bug:
2159 */
2160 MD_BUG();
2161 mddev->events --;
2162 }
2163
2164 /*
2165 * do not write anything to disk if using
2166 * nonpersistent superblocks
2167 */
2168 if (!mddev->persistent) {
2169 if (!mddev->external)
2170 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2171
2172 spin_unlock_irq(&mddev->write_lock);
2173 wake_up(&mddev->sb_wait);
2174 return;
2175 }
2176 sync_sbs(mddev, nospares);
2177 spin_unlock_irq(&mddev->write_lock);
2178
2179 dprintk(KERN_INFO
2180 "md: updating %s RAID superblock on device (in sync %d)\n",
2181 mdname(mddev),mddev->in_sync);
2182
2183 bitmap_update_sb(mddev->bitmap);
2184 list_for_each_entry(rdev, &mddev->disks, same_set) {
2185 char b[BDEVNAME_SIZE];
2186 dprintk(KERN_INFO "md: ");
2187 if (rdev->sb_loaded != 1)
2188 continue; /* no noise on spare devices */
2189 if (test_bit(Faulty, &rdev->flags))
2190 dprintk("(skipping faulty ");
2191
2192 dprintk("%s ", bdevname(rdev->bdev,b));
2193 if (!test_bit(Faulty, &rdev->flags)) {
2194 md_super_write(mddev,rdev,
2195 rdev->sb_start, rdev->sb_size,
2196 rdev->sb_page);
2197 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2198 bdevname(rdev->bdev,b),
2199 (unsigned long long)rdev->sb_start);
2200 rdev->sb_events = mddev->events;
2201
2202 } else
2203 dprintk(")\n");
2204 if (mddev->level == LEVEL_MULTIPATH)
2205 /* only need to write one superblock... */
2206 break;
2207 }
2208 md_super_wait(mddev);
2209 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2210
2211 spin_lock_irq(&mddev->write_lock);
2212 if (mddev->in_sync != sync_req ||
2213 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2214 /* have to write it out again */
2215 spin_unlock_irq(&mddev->write_lock);
2216 goto repeat;
2217 }
2218 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2219 spin_unlock_irq(&mddev->write_lock);
2220 wake_up(&mddev->sb_wait);
2221 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2222 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2223
2224 }
2225
2226 /* words written to sysfs files may, or may not, be \n terminated.
2227 * We want to accept with case. For this we use cmd_match.
2228 */
2229 static int cmd_match(const char *cmd, const char *str)
2230 {
2231 /* See if cmd, written into a sysfs file, matches
2232 * str. They must either be the same, or cmd can
2233 * have a trailing newline
2234 */
2235 while (*cmd && *str && *cmd == *str) {
2236 cmd++;
2237 str++;
2238 }
2239 if (*cmd == '\n')
2240 cmd++;
2241 if (*str || *cmd)
2242 return 0;
2243 return 1;
2244 }
2245
2246 struct rdev_sysfs_entry {
2247 struct attribute attr;
2248 ssize_t (*show)(mdk_rdev_t *, char *);
2249 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2250 };
2251
2252 static ssize_t
2253 state_show(mdk_rdev_t *rdev, char *page)
2254 {
2255 char *sep = "";
2256 size_t len = 0;
2257
2258 if (test_bit(Faulty, &rdev->flags)) {
2259 len+= sprintf(page+len, "%sfaulty",sep);
2260 sep = ",";
2261 }
2262 if (test_bit(In_sync, &rdev->flags)) {
2263 len += sprintf(page+len, "%sin_sync",sep);
2264 sep = ",";
2265 }
2266 if (test_bit(WriteMostly, &rdev->flags)) {
2267 len += sprintf(page+len, "%swrite_mostly",sep);
2268 sep = ",";
2269 }
2270 if (test_bit(Blocked, &rdev->flags)) {
2271 len += sprintf(page+len, "%sblocked", sep);
2272 sep = ",";
2273 }
2274 if (!test_bit(Faulty, &rdev->flags) &&
2275 !test_bit(In_sync, &rdev->flags)) {
2276 len += sprintf(page+len, "%sspare", sep);
2277 sep = ",";
2278 }
2279 return len+sprintf(page+len, "\n");
2280 }
2281
2282 static ssize_t
2283 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2284 {
2285 /* can write
2286 * faulty - simulates and error
2287 * remove - disconnects the device
2288 * writemostly - sets write_mostly
2289 * -writemostly - clears write_mostly
2290 * blocked - sets the Blocked flag
2291 * -blocked - clears the Blocked flag
2292 * insync - sets Insync providing device isn't active
2293 */
2294 int err = -EINVAL;
2295 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2296 md_error(rdev->mddev, rdev);
2297 err = 0;
2298 } else if (cmd_match(buf, "remove")) {
2299 if (rdev->raid_disk >= 0)
2300 err = -EBUSY;
2301 else {
2302 mddev_t *mddev = rdev->mddev;
2303 kick_rdev_from_array(rdev);
2304 if (mddev->pers)
2305 md_update_sb(mddev, 1);
2306 md_new_event(mddev);
2307 err = 0;
2308 }
2309 } else if (cmd_match(buf, "writemostly")) {
2310 set_bit(WriteMostly, &rdev->flags);
2311 err = 0;
2312 } else if (cmd_match(buf, "-writemostly")) {
2313 clear_bit(WriteMostly, &rdev->flags);
2314 err = 0;
2315 } else if (cmd_match(buf, "blocked")) {
2316 set_bit(Blocked, &rdev->flags);
2317 err = 0;
2318 } else if (cmd_match(buf, "-blocked")) {
2319 clear_bit(Blocked, &rdev->flags);
2320 wake_up(&rdev->blocked_wait);
2321 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2322 md_wakeup_thread(rdev->mddev->thread);
2323
2324 err = 0;
2325 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2326 set_bit(In_sync, &rdev->flags);
2327 err = 0;
2328 }
2329 if (!err && rdev->sysfs_state)
2330 sysfs_notify_dirent(rdev->sysfs_state);
2331 return err ? err : len;
2332 }
2333 static struct rdev_sysfs_entry rdev_state =
2334 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2335
2336 static ssize_t
2337 errors_show(mdk_rdev_t *rdev, char *page)
2338 {
2339 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2340 }
2341
2342 static ssize_t
2343 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2344 {
2345 char *e;
2346 unsigned long n = simple_strtoul(buf, &e, 10);
2347 if (*buf && (*e == 0 || *e == '\n')) {
2348 atomic_set(&rdev->corrected_errors, n);
2349 return len;
2350 }
2351 return -EINVAL;
2352 }
2353 static struct rdev_sysfs_entry rdev_errors =
2354 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2355
2356 static ssize_t
2357 slot_show(mdk_rdev_t *rdev, char *page)
2358 {
2359 if (rdev->raid_disk < 0)
2360 return sprintf(page, "none\n");
2361 else
2362 return sprintf(page, "%d\n", rdev->raid_disk);
2363 }
2364
2365 static ssize_t
2366 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2367 {
2368 char *e;
2369 int err;
2370 char nm[20];
2371 int slot = simple_strtoul(buf, &e, 10);
2372 if (strncmp(buf, "none", 4)==0)
2373 slot = -1;
2374 else if (e==buf || (*e && *e!= '\n'))
2375 return -EINVAL;
2376 if (rdev->mddev->pers && slot == -1) {
2377 /* Setting 'slot' on an active array requires also
2378 * updating the 'rd%d' link, and communicating
2379 * with the personality with ->hot_*_disk.
2380 * For now we only support removing
2381 * failed/spare devices. This normally happens automatically,
2382 * but not when the metadata is externally managed.
2383 */
2384 if (rdev->raid_disk == -1)
2385 return -EEXIST;
2386 /* personality does all needed checks */
2387 if (rdev->mddev->pers->hot_add_disk == NULL)
2388 return -EINVAL;
2389 err = rdev->mddev->pers->
2390 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2391 if (err)
2392 return err;
2393 sprintf(nm, "rd%d", rdev->raid_disk);
2394 sysfs_remove_link(&rdev->mddev->kobj, nm);
2395 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2396 md_wakeup_thread(rdev->mddev->thread);
2397 } else if (rdev->mddev->pers) {
2398 mdk_rdev_t *rdev2;
2399 /* Activating a spare .. or possibly reactivating
2400 * if we ever get bitmaps working here.
2401 */
2402
2403 if (rdev->raid_disk != -1)
2404 return -EBUSY;
2405
2406 if (rdev->mddev->pers->hot_add_disk == NULL)
2407 return -EINVAL;
2408
2409 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2410 if (rdev2->raid_disk == slot)
2411 return -EEXIST;
2412
2413 rdev->raid_disk = slot;
2414 if (test_bit(In_sync, &rdev->flags))
2415 rdev->saved_raid_disk = slot;
2416 else
2417 rdev->saved_raid_disk = -1;
2418 err = rdev->mddev->pers->
2419 hot_add_disk(rdev->mddev, rdev);
2420 if (err) {
2421 rdev->raid_disk = -1;
2422 return err;
2423 } else
2424 sysfs_notify_dirent(rdev->sysfs_state);
2425 sprintf(nm, "rd%d", rdev->raid_disk);
2426 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2427 printk(KERN_WARNING
2428 "md: cannot register "
2429 "%s for %s\n",
2430 nm, mdname(rdev->mddev));
2431
2432 /* don't wakeup anyone, leave that to userspace. */
2433 } else {
2434 if (slot >= rdev->mddev->raid_disks)
2435 return -ENOSPC;
2436 rdev->raid_disk = slot;
2437 /* assume it is working */
2438 clear_bit(Faulty, &rdev->flags);
2439 clear_bit(WriteMostly, &rdev->flags);
2440 set_bit(In_sync, &rdev->flags);
2441 sysfs_notify_dirent(rdev->sysfs_state);
2442 }
2443 return len;
2444 }
2445
2446
2447 static struct rdev_sysfs_entry rdev_slot =
2448 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2449
2450 static ssize_t
2451 offset_show(mdk_rdev_t *rdev, char *page)
2452 {
2453 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2454 }
2455
2456 static ssize_t
2457 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2458 {
2459 char *e;
2460 unsigned long long offset = simple_strtoull(buf, &e, 10);
2461 if (e==buf || (*e && *e != '\n'))
2462 return -EINVAL;
2463 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2464 return -EBUSY;
2465 if (rdev->sectors && rdev->mddev->external)
2466 /* Must set offset before size, so overlap checks
2467 * can be sane */
2468 return -EBUSY;
2469 rdev->data_offset = offset;
2470 return len;
2471 }
2472
2473 static struct rdev_sysfs_entry rdev_offset =
2474 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2475
2476 static ssize_t
2477 rdev_size_show(mdk_rdev_t *rdev, char *page)
2478 {
2479 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2480 }
2481
2482 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2483 {
2484 /* check if two start/length pairs overlap */
2485 if (s1+l1 <= s2)
2486 return 0;
2487 if (s2+l2 <= s1)
2488 return 0;
2489 return 1;
2490 }
2491
2492 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2493 {
2494 unsigned long long blocks;
2495 sector_t new;
2496
2497 if (strict_strtoull(buf, 10, &blocks) < 0)
2498 return -EINVAL;
2499
2500 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2501 return -EINVAL; /* sector conversion overflow */
2502
2503 new = blocks * 2;
2504 if (new != blocks * 2)
2505 return -EINVAL; /* unsigned long long to sector_t overflow */
2506
2507 *sectors = new;
2508 return 0;
2509 }
2510
2511 static ssize_t
2512 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2513 {
2514 mddev_t *my_mddev = rdev->mddev;
2515 sector_t oldsectors = rdev->sectors;
2516 sector_t sectors;
2517
2518 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2519 return -EINVAL;
2520 if (my_mddev->pers && rdev->raid_disk >= 0) {
2521 if (my_mddev->persistent) {
2522 sectors = super_types[my_mddev->major_version].
2523 rdev_size_change(rdev, sectors);
2524 if (!sectors)
2525 return -EBUSY;
2526 } else if (!sectors)
2527 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2528 rdev->data_offset;
2529 }
2530 if (sectors < my_mddev->dev_sectors)
2531 return -EINVAL; /* component must fit device */
2532
2533 rdev->sectors = sectors;
2534 if (sectors > oldsectors && my_mddev->external) {
2535 /* need to check that all other rdevs with the same ->bdev
2536 * do not overlap. We need to unlock the mddev to avoid
2537 * a deadlock. We have already changed rdev->sectors, and if
2538 * we have to change it back, we will have the lock again.
2539 */
2540 mddev_t *mddev;
2541 int overlap = 0;
2542 struct list_head *tmp;
2543
2544 mddev_unlock(my_mddev);
2545 for_each_mddev(mddev, tmp) {
2546 mdk_rdev_t *rdev2;
2547
2548 mddev_lock(mddev);
2549 list_for_each_entry(rdev2, &mddev->disks, same_set)
2550 if (test_bit(AllReserved, &rdev2->flags) ||
2551 (rdev->bdev == rdev2->bdev &&
2552 rdev != rdev2 &&
2553 overlaps(rdev->data_offset, rdev->sectors,
2554 rdev2->data_offset,
2555 rdev2->sectors))) {
2556 overlap = 1;
2557 break;
2558 }
2559 mddev_unlock(mddev);
2560 if (overlap) {
2561 mddev_put(mddev);
2562 break;
2563 }
2564 }
2565 mddev_lock(my_mddev);
2566 if (overlap) {
2567 /* Someone else could have slipped in a size
2568 * change here, but doing so is just silly.
2569 * We put oldsectors back because we *know* it is
2570 * safe, and trust userspace not to race with
2571 * itself
2572 */
2573 rdev->sectors = oldsectors;
2574 return -EBUSY;
2575 }
2576 }
2577 return len;
2578 }
2579
2580 static struct rdev_sysfs_entry rdev_size =
2581 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2582
2583
2584 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2585 {
2586 unsigned long long recovery_start = rdev->recovery_offset;
2587
2588 if (test_bit(In_sync, &rdev->flags) ||
2589 recovery_start == MaxSector)
2590 return sprintf(page, "none\n");
2591
2592 return sprintf(page, "%llu\n", recovery_start);
2593 }
2594
2595 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2596 {
2597 unsigned long long recovery_start;
2598
2599 if (cmd_match(buf, "none"))
2600 recovery_start = MaxSector;
2601 else if (strict_strtoull(buf, 10, &recovery_start))
2602 return -EINVAL;
2603
2604 if (rdev->mddev->pers &&
2605 rdev->raid_disk >= 0)
2606 return -EBUSY;
2607
2608 rdev->recovery_offset = recovery_start;
2609 if (recovery_start == MaxSector)
2610 set_bit(In_sync, &rdev->flags);
2611 else
2612 clear_bit(In_sync, &rdev->flags);
2613 return len;
2614 }
2615
2616 static struct rdev_sysfs_entry rdev_recovery_start =
2617 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2618
2619 static struct attribute *rdev_default_attrs[] = {
2620 &rdev_state.attr,
2621 &rdev_errors.attr,
2622 &rdev_slot.attr,
2623 &rdev_offset.attr,
2624 &rdev_size.attr,
2625 &rdev_recovery_start.attr,
2626 NULL,
2627 };
2628 static ssize_t
2629 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2630 {
2631 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2632 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2633 mddev_t *mddev = rdev->mddev;
2634 ssize_t rv;
2635
2636 if (!entry->show)
2637 return -EIO;
2638
2639 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2640 if (!rv) {
2641 if (rdev->mddev == NULL)
2642 rv = -EBUSY;
2643 else
2644 rv = entry->show(rdev, page);
2645 mddev_unlock(mddev);
2646 }
2647 return rv;
2648 }
2649
2650 static ssize_t
2651 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2652 const char *page, size_t length)
2653 {
2654 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2655 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2656 ssize_t rv;
2657 mddev_t *mddev = rdev->mddev;
2658
2659 if (!entry->store)
2660 return -EIO;
2661 if (!capable(CAP_SYS_ADMIN))
2662 return -EACCES;
2663 rv = mddev ? mddev_lock(mddev): -EBUSY;
2664 if (!rv) {
2665 if (rdev->mddev == NULL)
2666 rv = -EBUSY;
2667 else
2668 rv = entry->store(rdev, page, length);
2669 mddev_unlock(mddev);
2670 }
2671 return rv;
2672 }
2673
2674 static void rdev_free(struct kobject *ko)
2675 {
2676 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2677 kfree(rdev);
2678 }
2679 static const struct sysfs_ops rdev_sysfs_ops = {
2680 .show = rdev_attr_show,
2681 .store = rdev_attr_store,
2682 };
2683 static struct kobj_type rdev_ktype = {
2684 .release = rdev_free,
2685 .sysfs_ops = &rdev_sysfs_ops,
2686 .default_attrs = rdev_default_attrs,
2687 };
2688
2689 /*
2690 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2691 *
2692 * mark the device faulty if:
2693 *
2694 * - the device is nonexistent (zero size)
2695 * - the device has no valid superblock
2696 *
2697 * a faulty rdev _never_ has rdev->sb set.
2698 */
2699 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2700 {
2701 char b[BDEVNAME_SIZE];
2702 int err;
2703 mdk_rdev_t *rdev;
2704 sector_t size;
2705
2706 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2707 if (!rdev) {
2708 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2709 return ERR_PTR(-ENOMEM);
2710 }
2711
2712 if ((err = alloc_disk_sb(rdev)))
2713 goto abort_free;
2714
2715 err = lock_rdev(rdev, newdev, super_format == -2);
2716 if (err)
2717 goto abort_free;
2718
2719 kobject_init(&rdev->kobj, &rdev_ktype);
2720
2721 rdev->desc_nr = -1;
2722 rdev->saved_raid_disk = -1;
2723 rdev->raid_disk = -1;
2724 rdev->flags = 0;
2725 rdev->data_offset = 0;
2726 rdev->sb_events = 0;
2727 rdev->last_read_error.tv_sec = 0;
2728 rdev->last_read_error.tv_nsec = 0;
2729 atomic_set(&rdev->nr_pending, 0);
2730 atomic_set(&rdev->read_errors, 0);
2731 atomic_set(&rdev->corrected_errors, 0);
2732
2733 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2734 if (!size) {
2735 printk(KERN_WARNING
2736 "md: %s has zero or unknown size, marking faulty!\n",
2737 bdevname(rdev->bdev,b));
2738 err = -EINVAL;
2739 goto abort_free;
2740 }
2741
2742 if (super_format >= 0) {
2743 err = super_types[super_format].
2744 load_super(rdev, NULL, super_minor);
2745 if (err == -EINVAL) {
2746 printk(KERN_WARNING
2747 "md: %s does not have a valid v%d.%d "
2748 "superblock, not importing!\n",
2749 bdevname(rdev->bdev,b),
2750 super_format, super_minor);
2751 goto abort_free;
2752 }
2753 if (err < 0) {
2754 printk(KERN_WARNING
2755 "md: could not read %s's sb, not importing!\n",
2756 bdevname(rdev->bdev,b));
2757 goto abort_free;
2758 }
2759 }
2760
2761 INIT_LIST_HEAD(&rdev->same_set);
2762 init_waitqueue_head(&rdev->blocked_wait);
2763
2764 return rdev;
2765
2766 abort_free:
2767 if (rdev->sb_page) {
2768 if (rdev->bdev)
2769 unlock_rdev(rdev);
2770 free_disk_sb(rdev);
2771 }
2772 kfree(rdev);
2773 return ERR_PTR(err);
2774 }
2775
2776 /*
2777 * Check a full RAID array for plausibility
2778 */
2779
2780
2781 static void analyze_sbs(mddev_t * mddev)
2782 {
2783 int i;
2784 mdk_rdev_t *rdev, *freshest, *tmp;
2785 char b[BDEVNAME_SIZE];
2786
2787 freshest = NULL;
2788 rdev_for_each(rdev, tmp, mddev)
2789 switch (super_types[mddev->major_version].
2790 load_super(rdev, freshest, mddev->minor_version)) {
2791 case 1:
2792 freshest = rdev;
2793 break;
2794 case 0:
2795 break;
2796 default:
2797 printk( KERN_ERR \
2798 "md: fatal superblock inconsistency in %s"
2799 " -- removing from array\n",
2800 bdevname(rdev->bdev,b));
2801 kick_rdev_from_array(rdev);
2802 }
2803
2804
2805 super_types[mddev->major_version].
2806 validate_super(mddev, freshest);
2807
2808 i = 0;
2809 rdev_for_each(rdev, tmp, mddev) {
2810 if (rdev->desc_nr >= mddev->max_disks ||
2811 i > mddev->max_disks) {
2812 printk(KERN_WARNING
2813 "md: %s: %s: only %d devices permitted\n",
2814 mdname(mddev), bdevname(rdev->bdev, b),
2815 mddev->max_disks);
2816 kick_rdev_from_array(rdev);
2817 continue;
2818 }
2819 if (rdev != freshest)
2820 if (super_types[mddev->major_version].
2821 validate_super(mddev, rdev)) {
2822 printk(KERN_WARNING "md: kicking non-fresh %s"
2823 " from array!\n",
2824 bdevname(rdev->bdev,b));
2825 kick_rdev_from_array(rdev);
2826 continue;
2827 }
2828 if (mddev->level == LEVEL_MULTIPATH) {
2829 rdev->desc_nr = i++;
2830 rdev->raid_disk = rdev->desc_nr;
2831 set_bit(In_sync, &rdev->flags);
2832 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2833 rdev->raid_disk = -1;
2834 clear_bit(In_sync, &rdev->flags);
2835 }
2836 }
2837 }
2838
2839 /* Read a fixed-point number.
2840 * Numbers in sysfs attributes should be in "standard" units where
2841 * possible, so time should be in seconds.
2842 * However we internally use a a much smaller unit such as
2843 * milliseconds or jiffies.
2844 * This function takes a decimal number with a possible fractional
2845 * component, and produces an integer which is the result of
2846 * multiplying that number by 10^'scale'.
2847 * all without any floating-point arithmetic.
2848 */
2849 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2850 {
2851 unsigned long result = 0;
2852 long decimals = -1;
2853 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2854 if (*cp == '.')
2855 decimals = 0;
2856 else if (decimals < scale) {
2857 unsigned int value;
2858 value = *cp - '0';
2859 result = result * 10 + value;
2860 if (decimals >= 0)
2861 decimals++;
2862 }
2863 cp++;
2864 }
2865 if (*cp == '\n')
2866 cp++;
2867 if (*cp)
2868 return -EINVAL;
2869 if (decimals < 0)
2870 decimals = 0;
2871 while (decimals < scale) {
2872 result *= 10;
2873 decimals ++;
2874 }
2875 *res = result;
2876 return 0;
2877 }
2878
2879
2880 static void md_safemode_timeout(unsigned long data);
2881
2882 static ssize_t
2883 safe_delay_show(mddev_t *mddev, char *page)
2884 {
2885 int msec = (mddev->safemode_delay*1000)/HZ;
2886 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2887 }
2888 static ssize_t
2889 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2890 {
2891 unsigned long msec;
2892
2893 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2894 return -EINVAL;
2895 if (msec == 0)
2896 mddev->safemode_delay = 0;
2897 else {
2898 unsigned long old_delay = mddev->safemode_delay;
2899 mddev->safemode_delay = (msec*HZ)/1000;
2900 if (mddev->safemode_delay == 0)
2901 mddev->safemode_delay = 1;
2902 if (mddev->safemode_delay < old_delay)
2903 md_safemode_timeout((unsigned long)mddev);
2904 }
2905 return len;
2906 }
2907 static struct md_sysfs_entry md_safe_delay =
2908 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2909
2910 static ssize_t
2911 level_show(mddev_t *mddev, char *page)
2912 {
2913 struct mdk_personality *p = mddev->pers;
2914 if (p)
2915 return sprintf(page, "%s\n", p->name);
2916 else if (mddev->clevel[0])
2917 return sprintf(page, "%s\n", mddev->clevel);
2918 else if (mddev->level != LEVEL_NONE)
2919 return sprintf(page, "%d\n", mddev->level);
2920 else
2921 return 0;
2922 }
2923
2924 static ssize_t
2925 level_store(mddev_t *mddev, const char *buf, size_t len)
2926 {
2927 char level[16];
2928 ssize_t rv = len;
2929 struct mdk_personality *pers;
2930 void *priv;
2931 mdk_rdev_t *rdev;
2932
2933 if (mddev->pers == NULL) {
2934 if (len == 0)
2935 return 0;
2936 if (len >= sizeof(mddev->clevel))
2937 return -ENOSPC;
2938 strncpy(mddev->clevel, buf, len);
2939 if (mddev->clevel[len-1] == '\n')
2940 len--;
2941 mddev->clevel[len] = 0;
2942 mddev->level = LEVEL_NONE;
2943 return rv;
2944 }
2945
2946 /* request to change the personality. Need to ensure:
2947 * - array is not engaged in resync/recovery/reshape
2948 * - old personality can be suspended
2949 * - new personality will access other array.
2950 */
2951
2952 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2953 return -EBUSY;
2954
2955 if (!mddev->pers->quiesce) {
2956 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2957 mdname(mddev), mddev->pers->name);
2958 return -EINVAL;
2959 }
2960
2961 /* Now find the new personality */
2962 if (len == 0 || len >= sizeof(level))
2963 return -EINVAL;
2964 strncpy(level, buf, len);
2965 if (level[len-1] == '\n')
2966 len--;
2967 level[len] = 0;
2968
2969 request_module("md-%s", level);
2970 spin_lock(&pers_lock);
2971 pers = find_pers(LEVEL_NONE, level);
2972 if (!pers || !try_module_get(pers->owner)) {
2973 spin_unlock(&pers_lock);
2974 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2975 return -EINVAL;
2976 }
2977 spin_unlock(&pers_lock);
2978
2979 if (pers == mddev->pers) {
2980 /* Nothing to do! */
2981 module_put(pers->owner);
2982 return rv;
2983 }
2984 if (!pers->takeover) {
2985 module_put(pers->owner);
2986 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2987 mdname(mddev), level);
2988 return -EINVAL;
2989 }
2990
2991 /* ->takeover must set new_* and/or delta_disks
2992 * if it succeeds, and may set them when it fails.
2993 */
2994 priv = pers->takeover(mddev);
2995 if (IS_ERR(priv)) {
2996 mddev->new_level = mddev->level;
2997 mddev->new_layout = mddev->layout;
2998 mddev->new_chunk_sectors = mddev->chunk_sectors;
2999 mddev->raid_disks -= mddev->delta_disks;
3000 mddev->delta_disks = 0;
3001 module_put(pers->owner);
3002 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3003 mdname(mddev), level);
3004 return PTR_ERR(priv);
3005 }
3006
3007 /* Looks like we have a winner */
3008 mddev_suspend(mddev);
3009 mddev->pers->stop(mddev);
3010
3011 if (mddev->pers->sync_request == NULL &&
3012 pers->sync_request != NULL) {
3013 /* need to add the md_redundancy_group */
3014 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3015 printk(KERN_WARNING
3016 "md: cannot register extra attributes for %s\n",
3017 mdname(mddev));
3018 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3019 }
3020 if (mddev->pers->sync_request != NULL &&
3021 pers->sync_request == NULL) {
3022 /* need to remove the md_redundancy_group */
3023 if (mddev->to_remove == NULL)
3024 mddev->to_remove = &md_redundancy_group;
3025 }
3026
3027 module_put(mddev->pers->owner);
3028 /* Invalidate devices that are now superfluous */
3029 list_for_each_entry(rdev, &mddev->disks, same_set)
3030 if (rdev->raid_disk >= mddev->raid_disks) {
3031 rdev->raid_disk = -1;
3032 clear_bit(In_sync, &rdev->flags);
3033 }
3034 mddev->pers = pers;
3035 mddev->private = priv;
3036 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3037 mddev->level = mddev->new_level;
3038 mddev->layout = mddev->new_layout;
3039 mddev->chunk_sectors = mddev->new_chunk_sectors;
3040 mddev->delta_disks = 0;
3041 pers->run(mddev);
3042 mddev_resume(mddev);
3043 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3044 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3045 md_wakeup_thread(mddev->thread);
3046 return rv;
3047 }
3048
3049 static struct md_sysfs_entry md_level =
3050 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3051
3052
3053 static ssize_t
3054 layout_show(mddev_t *mddev, char *page)
3055 {
3056 /* just a number, not meaningful for all levels */
3057 if (mddev->reshape_position != MaxSector &&
3058 mddev->layout != mddev->new_layout)
3059 return sprintf(page, "%d (%d)\n",
3060 mddev->new_layout, mddev->layout);
3061 return sprintf(page, "%d\n", mddev->layout);
3062 }
3063
3064 static ssize_t
3065 layout_store(mddev_t *mddev, const char *buf, size_t len)
3066 {
3067 char *e;
3068 unsigned long n = simple_strtoul(buf, &e, 10);
3069
3070 if (!*buf || (*e && *e != '\n'))
3071 return -EINVAL;
3072
3073 if (mddev->pers) {
3074 int err;
3075 if (mddev->pers->check_reshape == NULL)
3076 return -EBUSY;
3077 mddev->new_layout = n;
3078 err = mddev->pers->check_reshape(mddev);
3079 if (err) {
3080 mddev->new_layout = mddev->layout;
3081 return err;
3082 }
3083 } else {
3084 mddev->new_layout = n;
3085 if (mddev->reshape_position == MaxSector)
3086 mddev->layout = n;
3087 }
3088 return len;
3089 }
3090 static struct md_sysfs_entry md_layout =
3091 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3092
3093
3094 static ssize_t
3095 raid_disks_show(mddev_t *mddev, char *page)
3096 {
3097 if (mddev->raid_disks == 0)
3098 return 0;
3099 if (mddev->reshape_position != MaxSector &&
3100 mddev->delta_disks != 0)
3101 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3102 mddev->raid_disks - mddev->delta_disks);
3103 return sprintf(page, "%d\n", mddev->raid_disks);
3104 }
3105
3106 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3107
3108 static ssize_t
3109 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3110 {
3111 char *e;
3112 int rv = 0;
3113 unsigned long n = simple_strtoul(buf, &e, 10);
3114
3115 if (!*buf || (*e && *e != '\n'))
3116 return -EINVAL;
3117
3118 if (mddev->pers)
3119 rv = update_raid_disks(mddev, n);
3120 else if (mddev->reshape_position != MaxSector) {
3121 int olddisks = mddev->raid_disks - mddev->delta_disks;
3122 mddev->delta_disks = n - olddisks;
3123 mddev->raid_disks = n;
3124 } else
3125 mddev->raid_disks = n;
3126 return rv ? rv : len;
3127 }
3128 static struct md_sysfs_entry md_raid_disks =
3129 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3130
3131 static ssize_t
3132 chunk_size_show(mddev_t *mddev, char *page)
3133 {
3134 if (mddev->reshape_position != MaxSector &&
3135 mddev->chunk_sectors != mddev->new_chunk_sectors)
3136 return sprintf(page, "%d (%d)\n",
3137 mddev->new_chunk_sectors << 9,
3138 mddev->chunk_sectors << 9);
3139 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3140 }
3141
3142 static ssize_t
3143 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3144 {
3145 char *e;
3146 unsigned long n = simple_strtoul(buf, &e, 10);
3147
3148 if (!*buf || (*e && *e != '\n'))
3149 return -EINVAL;
3150
3151 if (mddev->pers) {
3152 int err;
3153 if (mddev->pers->check_reshape == NULL)
3154 return -EBUSY;
3155 mddev->new_chunk_sectors = n >> 9;
3156 err = mddev->pers->check_reshape(mddev);
3157 if (err) {
3158 mddev->new_chunk_sectors = mddev->chunk_sectors;
3159 return err;
3160 }
3161 } else {
3162 mddev->new_chunk_sectors = n >> 9;
3163 if (mddev->reshape_position == MaxSector)
3164 mddev->chunk_sectors = n >> 9;
3165 }
3166 return len;
3167 }
3168 static struct md_sysfs_entry md_chunk_size =
3169 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3170
3171 static ssize_t
3172 resync_start_show(mddev_t *mddev, char *page)
3173 {
3174 if (mddev->recovery_cp == MaxSector)
3175 return sprintf(page, "none\n");
3176 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3177 }
3178
3179 static ssize_t
3180 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3181 {
3182 char *e;
3183 unsigned long long n = simple_strtoull(buf, &e, 10);
3184
3185 if (mddev->pers)
3186 return -EBUSY;
3187 if (cmd_match(buf, "none"))
3188 n = MaxSector;
3189 else if (!*buf || (*e && *e != '\n'))
3190 return -EINVAL;
3191
3192 mddev->recovery_cp = n;
3193 return len;
3194 }
3195 static struct md_sysfs_entry md_resync_start =
3196 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3197
3198 /*
3199 * The array state can be:
3200 *
3201 * clear
3202 * No devices, no size, no level
3203 * Equivalent to STOP_ARRAY ioctl
3204 * inactive
3205 * May have some settings, but array is not active
3206 * all IO results in error
3207 * When written, doesn't tear down array, but just stops it
3208 * suspended (not supported yet)
3209 * All IO requests will block. The array can be reconfigured.
3210 * Writing this, if accepted, will block until array is quiescent
3211 * readonly
3212 * no resync can happen. no superblocks get written.
3213 * write requests fail
3214 * read-auto
3215 * like readonly, but behaves like 'clean' on a write request.
3216 *
3217 * clean - no pending writes, but otherwise active.
3218 * When written to inactive array, starts without resync
3219 * If a write request arrives then
3220 * if metadata is known, mark 'dirty' and switch to 'active'.
3221 * if not known, block and switch to write-pending
3222 * If written to an active array that has pending writes, then fails.
3223 * active
3224 * fully active: IO and resync can be happening.
3225 * When written to inactive array, starts with resync
3226 *
3227 * write-pending
3228 * clean, but writes are blocked waiting for 'active' to be written.
3229 *
3230 * active-idle
3231 * like active, but no writes have been seen for a while (100msec).
3232 *
3233 */
3234 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3235 write_pending, active_idle, bad_word};
3236 static char *array_states[] = {
3237 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3238 "write-pending", "active-idle", NULL };
3239
3240 static int match_word(const char *word, char **list)
3241 {
3242 int n;
3243 for (n=0; list[n]; n++)
3244 if (cmd_match(word, list[n]))
3245 break;
3246 return n;
3247 }
3248
3249 static ssize_t
3250 array_state_show(mddev_t *mddev, char *page)
3251 {
3252 enum array_state st = inactive;
3253
3254 if (mddev->pers)
3255 switch(mddev->ro) {
3256 case 1:
3257 st = readonly;
3258 break;
3259 case 2:
3260 st = read_auto;
3261 break;
3262 case 0:
3263 if (mddev->in_sync)
3264 st = clean;
3265 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3266 st = write_pending;
3267 else if (mddev->safemode)
3268 st = active_idle;
3269 else
3270 st = active;
3271 }
3272 else {
3273 if (list_empty(&mddev->disks) &&
3274 mddev->raid_disks == 0 &&
3275 mddev->dev_sectors == 0)
3276 st = clear;
3277 else
3278 st = inactive;
3279 }
3280 return sprintf(page, "%s\n", array_states[st]);
3281 }
3282
3283 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3284 static int do_md_run(mddev_t * mddev);
3285 static int restart_array(mddev_t *mddev);
3286
3287 static ssize_t
3288 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3289 {
3290 int err = -EINVAL;
3291 enum array_state st = match_word(buf, array_states);
3292 switch(st) {
3293 case bad_word:
3294 break;
3295 case clear:
3296 /* stopping an active array */
3297 if (atomic_read(&mddev->openers) > 0)
3298 return -EBUSY;
3299 err = do_md_stop(mddev, 0, 0);
3300 break;
3301 case inactive:
3302 /* stopping an active array */
3303 if (mddev->pers) {
3304 if (atomic_read(&mddev->openers) > 0)
3305 return -EBUSY;
3306 err = do_md_stop(mddev, 2, 0);
3307 } else
3308 err = 0; /* already inactive */
3309 break;
3310 case suspended:
3311 break; /* not supported yet */
3312 case readonly:
3313 if (mddev->pers)
3314 err = do_md_stop(mddev, 1, 0);
3315 else {
3316 mddev->ro = 1;
3317 set_disk_ro(mddev->gendisk, 1);
3318 err = do_md_run(mddev);
3319 }
3320 break;
3321 case read_auto:
3322 if (mddev->pers) {
3323 if (mddev->ro == 0)
3324 err = do_md_stop(mddev, 1, 0);
3325 else if (mddev->ro == 1)
3326 err = restart_array(mddev);
3327 if (err == 0) {
3328 mddev->ro = 2;
3329 set_disk_ro(mddev->gendisk, 0);
3330 }
3331 } else {
3332 mddev->ro = 2;
3333 err = do_md_run(mddev);
3334 }
3335 break;
3336 case clean:
3337 if (mddev->pers) {
3338 restart_array(mddev);
3339 spin_lock_irq(&mddev->write_lock);
3340 if (atomic_read(&mddev->writes_pending) == 0) {
3341 if (mddev->in_sync == 0) {
3342 mddev->in_sync = 1;
3343 if (mddev->safemode == 1)
3344 mddev->safemode = 0;
3345 if (mddev->persistent)
3346 set_bit(MD_CHANGE_CLEAN,
3347 &mddev->flags);
3348 }
3349 err = 0;
3350 } else
3351 err = -EBUSY;
3352 spin_unlock_irq(&mddev->write_lock);
3353 } else
3354 err = -EINVAL;
3355 break;
3356 case active:
3357 if (mddev->pers) {
3358 restart_array(mddev);
3359 if (mddev->external)
3360 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3361 wake_up(&mddev->sb_wait);
3362 err = 0;
3363 } else {
3364 mddev->ro = 0;
3365 set_disk_ro(mddev->gendisk, 0);
3366 err = do_md_run(mddev);
3367 }
3368 break;
3369 case write_pending:
3370 case active_idle:
3371 /* these cannot be set */
3372 break;
3373 }
3374 if (err)
3375 return err;
3376 else {
3377 sysfs_notify_dirent(mddev->sysfs_state);
3378 return len;
3379 }
3380 }
3381 static struct md_sysfs_entry md_array_state =
3382 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3383
3384 static ssize_t
3385 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3386 return sprintf(page, "%d\n",
3387 atomic_read(&mddev->max_corr_read_errors));
3388 }
3389
3390 static ssize_t
3391 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3392 {
3393 char *e;
3394 unsigned long n = simple_strtoul(buf, &e, 10);
3395
3396 if (*buf && (*e == 0 || *e == '\n')) {
3397 atomic_set(&mddev->max_corr_read_errors, n);
3398 return len;
3399 }
3400 return -EINVAL;
3401 }
3402
3403 static struct md_sysfs_entry max_corr_read_errors =
3404 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3405 max_corrected_read_errors_store);
3406
3407 static ssize_t
3408 null_show(mddev_t *mddev, char *page)
3409 {
3410 return -EINVAL;
3411 }
3412
3413 static ssize_t
3414 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3415 {
3416 /* buf must be %d:%d\n? giving major and minor numbers */
3417 /* The new device is added to the array.
3418 * If the array has a persistent superblock, we read the
3419 * superblock to initialise info and check validity.
3420 * Otherwise, only checking done is that in bind_rdev_to_array,
3421 * which mainly checks size.
3422 */
3423 char *e;
3424 int major = simple_strtoul(buf, &e, 10);
3425 int minor;
3426 dev_t dev;
3427 mdk_rdev_t *rdev;
3428 int err;
3429
3430 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3431 return -EINVAL;
3432 minor = simple_strtoul(e+1, &e, 10);
3433 if (*e && *e != '\n')
3434 return -EINVAL;
3435 dev = MKDEV(major, minor);
3436 if (major != MAJOR(dev) ||
3437 minor != MINOR(dev))
3438 return -EOVERFLOW;
3439
3440
3441 if (mddev->persistent) {
3442 rdev = md_import_device(dev, mddev->major_version,
3443 mddev->minor_version);
3444 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3445 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3446 mdk_rdev_t, same_set);
3447 err = super_types[mddev->major_version]
3448 .load_super(rdev, rdev0, mddev->minor_version);
3449 if (err < 0)
3450 goto out;
3451 }
3452 } else if (mddev->external)
3453 rdev = md_import_device(dev, -2, -1);
3454 else
3455 rdev = md_import_device(dev, -1, -1);
3456
3457 if (IS_ERR(rdev))
3458 return PTR_ERR(rdev);
3459 err = bind_rdev_to_array(rdev, mddev);
3460 out:
3461 if (err)
3462 export_rdev(rdev);
3463 return err ? err : len;
3464 }
3465
3466 static struct md_sysfs_entry md_new_device =
3467 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3468
3469 static ssize_t
3470 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3471 {
3472 char *end;
3473 unsigned long chunk, end_chunk;
3474
3475 if (!mddev->bitmap)
3476 goto out;
3477 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3478 while (*buf) {
3479 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3480 if (buf == end) break;
3481 if (*end == '-') { /* range */
3482 buf = end + 1;
3483 end_chunk = simple_strtoul(buf, &end, 0);
3484 if (buf == end) break;
3485 }
3486 if (*end && !isspace(*end)) break;
3487 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3488 buf = skip_spaces(end);
3489 }
3490 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3491 out:
3492 return len;
3493 }
3494
3495 static struct md_sysfs_entry md_bitmap =
3496 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3497
3498 static ssize_t
3499 size_show(mddev_t *mddev, char *page)
3500 {
3501 return sprintf(page, "%llu\n",
3502 (unsigned long long)mddev->dev_sectors / 2);
3503 }
3504
3505 static int update_size(mddev_t *mddev, sector_t num_sectors);
3506
3507 static ssize_t
3508 size_store(mddev_t *mddev, const char *buf, size_t len)
3509 {
3510 /* If array is inactive, we can reduce the component size, but
3511 * not increase it (except from 0).
3512 * If array is active, we can try an on-line resize
3513 */
3514 sector_t sectors;
3515 int err = strict_blocks_to_sectors(buf, &sectors);
3516
3517 if (err < 0)
3518 return err;
3519 if (mddev->pers) {
3520 err = update_size(mddev, sectors);
3521 md_update_sb(mddev, 1);
3522 } else {
3523 if (mddev->dev_sectors == 0 ||
3524 mddev->dev_sectors > sectors)
3525 mddev->dev_sectors = sectors;
3526 else
3527 err = -ENOSPC;
3528 }
3529 return err ? err : len;
3530 }
3531
3532 static struct md_sysfs_entry md_size =
3533 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3534
3535
3536 /* Metdata version.
3537 * This is one of
3538 * 'none' for arrays with no metadata (good luck...)
3539 * 'external' for arrays with externally managed metadata,
3540 * or N.M for internally known formats
3541 */
3542 static ssize_t
3543 metadata_show(mddev_t *mddev, char *page)
3544 {
3545 if (mddev->persistent)
3546 return sprintf(page, "%d.%d\n",
3547 mddev->major_version, mddev->minor_version);
3548 else if (mddev->external)
3549 return sprintf(page, "external:%s\n", mddev->metadata_type);
3550 else
3551 return sprintf(page, "none\n");
3552 }
3553
3554 static ssize_t
3555 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3556 {
3557 int major, minor;
3558 char *e;
3559 /* Changing the details of 'external' metadata is
3560 * always permitted. Otherwise there must be
3561 * no devices attached to the array.
3562 */
3563 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3564 ;
3565 else if (!list_empty(&mddev->disks))
3566 return -EBUSY;
3567
3568 if (cmd_match(buf, "none")) {
3569 mddev->persistent = 0;
3570 mddev->external = 0;
3571 mddev->major_version = 0;
3572 mddev->minor_version = 90;
3573 return len;
3574 }
3575 if (strncmp(buf, "external:", 9) == 0) {
3576 size_t namelen = len-9;
3577 if (namelen >= sizeof(mddev->metadata_type))
3578 namelen = sizeof(mddev->metadata_type)-1;
3579 strncpy(mddev->metadata_type, buf+9, namelen);
3580 mddev->metadata_type[namelen] = 0;
3581 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3582 mddev->metadata_type[--namelen] = 0;
3583 mddev->persistent = 0;
3584 mddev->external = 1;
3585 mddev->major_version = 0;
3586 mddev->minor_version = 90;
3587 return len;
3588 }
3589 major = simple_strtoul(buf, &e, 10);
3590 if (e==buf || *e != '.')
3591 return -EINVAL;
3592 buf = e+1;
3593 minor = simple_strtoul(buf, &e, 10);
3594 if (e==buf || (*e && *e != '\n') )
3595 return -EINVAL;
3596 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3597 return -ENOENT;
3598 mddev->major_version = major;
3599 mddev->minor_version = minor;
3600 mddev->persistent = 1;
3601 mddev->external = 0;
3602 return len;
3603 }
3604
3605 static struct md_sysfs_entry md_metadata =
3606 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3607
3608 static ssize_t
3609 action_show(mddev_t *mddev, char *page)
3610 {
3611 char *type = "idle";
3612 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3613 type = "frozen";
3614 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3615 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3616 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3617 type = "reshape";
3618 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3619 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3620 type = "resync";
3621 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3622 type = "check";
3623 else
3624 type = "repair";
3625 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3626 type = "recover";
3627 }
3628 return sprintf(page, "%s\n", type);
3629 }
3630
3631 static ssize_t
3632 action_store(mddev_t *mddev, const char *page, size_t len)
3633 {
3634 if (!mddev->pers || !mddev->pers->sync_request)
3635 return -EINVAL;
3636
3637 if (cmd_match(page, "frozen"))
3638 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3639 else
3640 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3641
3642 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3643 if (mddev->sync_thread) {
3644 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3645 md_unregister_thread(mddev->sync_thread);
3646 mddev->sync_thread = NULL;
3647 mddev->recovery = 0;
3648 }
3649 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3650 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3651 return -EBUSY;
3652 else if (cmd_match(page, "resync"))
3653 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3654 else if (cmd_match(page, "recover")) {
3655 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3656 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3657 } else if (cmd_match(page, "reshape")) {
3658 int err;
3659 if (mddev->pers->start_reshape == NULL)
3660 return -EINVAL;
3661 err = mddev->pers->start_reshape(mddev);
3662 if (err)
3663 return err;
3664 sysfs_notify(&mddev->kobj, NULL, "degraded");
3665 } else {
3666 if (cmd_match(page, "check"))
3667 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3668 else if (!cmd_match(page, "repair"))
3669 return -EINVAL;
3670 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3671 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3672 }
3673 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3674 md_wakeup_thread(mddev->thread);
3675 sysfs_notify_dirent(mddev->sysfs_action);
3676 return len;
3677 }
3678
3679 static ssize_t
3680 mismatch_cnt_show(mddev_t *mddev, char *page)
3681 {
3682 return sprintf(page, "%llu\n",
3683 (unsigned long long) mddev->resync_mismatches);
3684 }
3685
3686 static struct md_sysfs_entry md_scan_mode =
3687 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3688
3689
3690 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3691
3692 static ssize_t
3693 sync_min_show(mddev_t *mddev, char *page)
3694 {
3695 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3696 mddev->sync_speed_min ? "local": "system");
3697 }
3698
3699 static ssize_t
3700 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3701 {
3702 int min;
3703 char *e;
3704 if (strncmp(buf, "system", 6)==0) {
3705 mddev->sync_speed_min = 0;
3706 return len;
3707 }
3708 min = simple_strtoul(buf, &e, 10);
3709 if (buf == e || (*e && *e != '\n') || min <= 0)
3710 return -EINVAL;
3711 mddev->sync_speed_min = min;
3712 return len;
3713 }
3714
3715 static struct md_sysfs_entry md_sync_min =
3716 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3717
3718 static ssize_t
3719 sync_max_show(mddev_t *mddev, char *page)
3720 {
3721 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3722 mddev->sync_speed_max ? "local": "system");
3723 }
3724
3725 static ssize_t
3726 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3727 {
3728 int max;
3729 char *e;
3730 if (strncmp(buf, "system", 6)==0) {
3731 mddev->sync_speed_max = 0;
3732 return len;
3733 }
3734 max = simple_strtoul(buf, &e, 10);
3735 if (buf == e || (*e && *e != '\n') || max <= 0)
3736 return -EINVAL;
3737 mddev->sync_speed_max = max;
3738 return len;
3739 }
3740
3741 static struct md_sysfs_entry md_sync_max =
3742 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3743
3744 static ssize_t
3745 degraded_show(mddev_t *mddev, char *page)
3746 {
3747 return sprintf(page, "%d\n", mddev->degraded);
3748 }
3749 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3750
3751 static ssize_t
3752 sync_force_parallel_show(mddev_t *mddev, char *page)
3753 {
3754 return sprintf(page, "%d\n", mddev->parallel_resync);
3755 }
3756
3757 static ssize_t
3758 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3759 {
3760 long n;
3761
3762 if (strict_strtol(buf, 10, &n))
3763 return -EINVAL;
3764
3765 if (n != 0 && n != 1)
3766 return -EINVAL;
3767
3768 mddev->parallel_resync = n;
3769
3770 if (mddev->sync_thread)
3771 wake_up(&resync_wait);
3772
3773 return len;
3774 }
3775
3776 /* force parallel resync, even with shared block devices */
3777 static struct md_sysfs_entry md_sync_force_parallel =
3778 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3779 sync_force_parallel_show, sync_force_parallel_store);
3780
3781 static ssize_t
3782 sync_speed_show(mddev_t *mddev, char *page)
3783 {
3784 unsigned long resync, dt, db;
3785 if (mddev->curr_resync == 0)
3786 return sprintf(page, "none\n");
3787 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3788 dt = (jiffies - mddev->resync_mark) / HZ;
3789 if (!dt) dt++;
3790 db = resync - mddev->resync_mark_cnt;
3791 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3792 }
3793
3794 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3795
3796 static ssize_t
3797 sync_completed_show(mddev_t *mddev, char *page)
3798 {
3799 unsigned long max_sectors, resync;
3800
3801 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3802 return sprintf(page, "none\n");
3803
3804 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3805 max_sectors = mddev->resync_max_sectors;
3806 else
3807 max_sectors = mddev->dev_sectors;
3808
3809 resync = mddev->curr_resync_completed;
3810 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3811 }
3812
3813 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3814
3815 static ssize_t
3816 min_sync_show(mddev_t *mddev, char *page)
3817 {
3818 return sprintf(page, "%llu\n",
3819 (unsigned long long)mddev->resync_min);
3820 }
3821 static ssize_t
3822 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3823 {
3824 unsigned long long min;
3825 if (strict_strtoull(buf, 10, &min))
3826 return -EINVAL;
3827 if (min > mddev->resync_max)
3828 return -EINVAL;
3829 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3830 return -EBUSY;
3831
3832 /* Must be a multiple of chunk_size */
3833 if (mddev->chunk_sectors) {
3834 sector_t temp = min;
3835 if (sector_div(temp, mddev->chunk_sectors))
3836 return -EINVAL;
3837 }
3838 mddev->resync_min = min;
3839
3840 return len;
3841 }
3842
3843 static struct md_sysfs_entry md_min_sync =
3844 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3845
3846 static ssize_t
3847 max_sync_show(mddev_t *mddev, char *page)
3848 {
3849 if (mddev->resync_max == MaxSector)
3850 return sprintf(page, "max\n");
3851 else
3852 return sprintf(page, "%llu\n",
3853 (unsigned long long)mddev->resync_max);
3854 }
3855 static ssize_t
3856 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3857 {
3858 if (strncmp(buf, "max", 3) == 0)
3859 mddev->resync_max = MaxSector;
3860 else {
3861 unsigned long long max;
3862 if (strict_strtoull(buf, 10, &max))
3863 return -EINVAL;
3864 if (max < mddev->resync_min)
3865 return -EINVAL;
3866 if (max < mddev->resync_max &&
3867 mddev->ro == 0 &&
3868 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3869 return -EBUSY;
3870
3871 /* Must be a multiple of chunk_size */
3872 if (mddev->chunk_sectors) {
3873 sector_t temp = max;
3874 if (sector_div(temp, mddev->chunk_sectors))
3875 return -EINVAL;
3876 }
3877 mddev->resync_max = max;
3878 }
3879 wake_up(&mddev->recovery_wait);
3880 return len;
3881 }
3882
3883 static struct md_sysfs_entry md_max_sync =
3884 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3885
3886 static ssize_t
3887 suspend_lo_show(mddev_t *mddev, char *page)
3888 {
3889 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3890 }
3891
3892 static ssize_t
3893 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3894 {
3895 char *e;
3896 unsigned long long new = simple_strtoull(buf, &e, 10);
3897
3898 if (mddev->pers == NULL ||
3899 mddev->pers->quiesce == NULL)
3900 return -EINVAL;
3901 if (buf == e || (*e && *e != '\n'))
3902 return -EINVAL;
3903 if (new >= mddev->suspend_hi ||
3904 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3905 mddev->suspend_lo = new;
3906 mddev->pers->quiesce(mddev, 2);
3907 return len;
3908 } else
3909 return -EINVAL;
3910 }
3911 static struct md_sysfs_entry md_suspend_lo =
3912 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3913
3914
3915 static ssize_t
3916 suspend_hi_show(mddev_t *mddev, char *page)
3917 {
3918 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3919 }
3920
3921 static ssize_t
3922 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3923 {
3924 char *e;
3925 unsigned long long new = simple_strtoull(buf, &e, 10);
3926
3927 if (mddev->pers == NULL ||
3928 mddev->pers->quiesce == NULL)
3929 return -EINVAL;
3930 if (buf == e || (*e && *e != '\n'))
3931 return -EINVAL;
3932 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3933 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3934 mddev->suspend_hi = new;
3935 mddev->pers->quiesce(mddev, 1);
3936 mddev->pers->quiesce(mddev, 0);
3937 return len;
3938 } else
3939 return -EINVAL;
3940 }
3941 static struct md_sysfs_entry md_suspend_hi =
3942 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3943
3944 static ssize_t
3945 reshape_position_show(mddev_t *mddev, char *page)
3946 {
3947 if (mddev->reshape_position != MaxSector)
3948 return sprintf(page, "%llu\n",
3949 (unsigned long long)mddev->reshape_position);
3950 strcpy(page, "none\n");
3951 return 5;
3952 }
3953
3954 static ssize_t
3955 reshape_position_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 if (mddev->pers)
3960 return -EBUSY;
3961 if (buf == e || (*e && *e != '\n'))
3962 return -EINVAL;
3963 mddev->reshape_position = new;
3964 mddev->delta_disks = 0;
3965 mddev->new_level = mddev->level;
3966 mddev->new_layout = mddev->layout;
3967 mddev->new_chunk_sectors = mddev->chunk_sectors;
3968 return len;
3969 }
3970
3971 static struct md_sysfs_entry md_reshape_position =
3972 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3973 reshape_position_store);
3974
3975 static ssize_t
3976 array_size_show(mddev_t *mddev, char *page)
3977 {
3978 if (mddev->external_size)
3979 return sprintf(page, "%llu\n",
3980 (unsigned long long)mddev->array_sectors/2);
3981 else
3982 return sprintf(page, "default\n");
3983 }
3984
3985 static ssize_t
3986 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3987 {
3988 sector_t sectors;
3989
3990 if (strncmp(buf, "default", 7) == 0) {
3991 if (mddev->pers)
3992 sectors = mddev->pers->size(mddev, 0, 0);
3993 else
3994 sectors = mddev->array_sectors;
3995
3996 mddev->external_size = 0;
3997 } else {
3998 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3999 return -EINVAL;
4000 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4001 return -E2BIG;
4002
4003 mddev->external_size = 1;
4004 }
4005
4006 mddev->array_sectors = sectors;
4007 set_capacity(mddev->gendisk, mddev->array_sectors);
4008 if (mddev->pers)
4009 revalidate_disk(mddev->gendisk);
4010
4011 return len;
4012 }
4013
4014 static struct md_sysfs_entry md_array_size =
4015 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4016 array_size_store);
4017
4018 static struct attribute *md_default_attrs[] = {
4019 &md_level.attr,
4020 &md_layout.attr,
4021 &md_raid_disks.attr,
4022 &md_chunk_size.attr,
4023 &md_size.attr,
4024 &md_resync_start.attr,
4025 &md_metadata.attr,
4026 &md_new_device.attr,
4027 &md_safe_delay.attr,
4028 &md_array_state.attr,
4029 &md_reshape_position.attr,
4030 &md_array_size.attr,
4031 &max_corr_read_errors.attr,
4032 NULL,
4033 };
4034
4035 static struct attribute *md_redundancy_attrs[] = {
4036 &md_scan_mode.attr,
4037 &md_mismatches.attr,
4038 &md_sync_min.attr,
4039 &md_sync_max.attr,
4040 &md_sync_speed.attr,
4041 &md_sync_force_parallel.attr,
4042 &md_sync_completed.attr,
4043 &md_min_sync.attr,
4044 &md_max_sync.attr,
4045 &md_suspend_lo.attr,
4046 &md_suspend_hi.attr,
4047 &md_bitmap.attr,
4048 &md_degraded.attr,
4049 NULL,
4050 };
4051 static struct attribute_group md_redundancy_group = {
4052 .name = NULL,
4053 .attrs = md_redundancy_attrs,
4054 };
4055
4056
4057 static ssize_t
4058 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4059 {
4060 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4061 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4062 ssize_t rv;
4063
4064 if (!entry->show)
4065 return -EIO;
4066 rv = mddev_lock(mddev);
4067 if (!rv) {
4068 rv = entry->show(mddev, page);
4069 mddev_unlock(mddev);
4070 }
4071 return rv;
4072 }
4073
4074 static ssize_t
4075 md_attr_store(struct kobject *kobj, struct attribute *attr,
4076 const char *page, size_t length)
4077 {
4078 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4079 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4080 ssize_t rv;
4081
4082 if (!entry->store)
4083 return -EIO;
4084 if (!capable(CAP_SYS_ADMIN))
4085 return -EACCES;
4086 rv = mddev_lock(mddev);
4087 if (mddev->hold_active == UNTIL_IOCTL)
4088 mddev->hold_active = 0;
4089 if (!rv) {
4090 rv = entry->store(mddev, page, length);
4091 mddev_unlock(mddev);
4092 }
4093 return rv;
4094 }
4095
4096 static void md_free(struct kobject *ko)
4097 {
4098 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4099
4100 if (mddev->sysfs_state)
4101 sysfs_put(mddev->sysfs_state);
4102
4103 if (mddev->gendisk) {
4104 del_gendisk(mddev->gendisk);
4105 put_disk(mddev->gendisk);
4106 }
4107 if (mddev->queue)
4108 blk_cleanup_queue(mddev->queue);
4109
4110 kfree(mddev);
4111 }
4112
4113 static const struct sysfs_ops md_sysfs_ops = {
4114 .show = md_attr_show,
4115 .store = md_attr_store,
4116 };
4117 static struct kobj_type md_ktype = {
4118 .release = md_free,
4119 .sysfs_ops = &md_sysfs_ops,
4120 .default_attrs = md_default_attrs,
4121 };
4122
4123 int mdp_major = 0;
4124
4125 static void mddev_delayed_delete(struct work_struct *ws)
4126 {
4127 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4128
4129 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4130 kobject_del(&mddev->kobj);
4131 kobject_put(&mddev->kobj);
4132 }
4133
4134 static int md_alloc(dev_t dev, char *name)
4135 {
4136 static DEFINE_MUTEX(disks_mutex);
4137 mddev_t *mddev = mddev_find(dev);
4138 struct gendisk *disk;
4139 int partitioned;
4140 int shift;
4141 int unit;
4142 int error;
4143
4144 if (!mddev)
4145 return -ENODEV;
4146
4147 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4148 shift = partitioned ? MdpMinorShift : 0;
4149 unit = MINOR(mddev->unit) >> shift;
4150
4151 /* wait for any previous instance if this device
4152 * to be completed removed (mddev_delayed_delete).
4153 */
4154 flush_scheduled_work();
4155
4156 mutex_lock(&disks_mutex);
4157 error = -EEXIST;
4158 if (mddev->gendisk)
4159 goto abort;
4160
4161 if (name) {
4162 /* Need to ensure that 'name' is not a duplicate.
4163 */
4164 mddev_t *mddev2;
4165 spin_lock(&all_mddevs_lock);
4166
4167 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4168 if (mddev2->gendisk &&
4169 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4170 spin_unlock(&all_mddevs_lock);
4171 goto abort;
4172 }
4173 spin_unlock(&all_mddevs_lock);
4174 }
4175
4176 error = -ENOMEM;
4177 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4178 if (!mddev->queue)
4179 goto abort;
4180 mddev->queue->queuedata = mddev;
4181
4182 /* Can be unlocked because the queue is new: no concurrency */
4183 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4184
4185 blk_queue_make_request(mddev->queue, md_make_request);
4186
4187 disk = alloc_disk(1 << shift);
4188 if (!disk) {
4189 blk_cleanup_queue(mddev->queue);
4190 mddev->queue = NULL;
4191 goto abort;
4192 }
4193 disk->major = MAJOR(mddev->unit);
4194 disk->first_minor = unit << shift;
4195 if (name)
4196 strcpy(disk->disk_name, name);
4197 else if (partitioned)
4198 sprintf(disk->disk_name, "md_d%d", unit);
4199 else
4200 sprintf(disk->disk_name, "md%d", unit);
4201 disk->fops = &md_fops;
4202 disk->private_data = mddev;
4203 disk->queue = mddev->queue;
4204 /* Allow extended partitions. This makes the
4205 * 'mdp' device redundant, but we can't really
4206 * remove it now.
4207 */
4208 disk->flags |= GENHD_FL_EXT_DEVT;
4209 add_disk(disk);
4210 mddev->gendisk = disk;
4211 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4212 &disk_to_dev(disk)->kobj, "%s", "md");
4213 if (error) {
4214 /* This isn't possible, but as kobject_init_and_add is marked
4215 * __must_check, we must do something with the result
4216 */
4217 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4218 disk->disk_name);
4219 error = 0;
4220 }
4221 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4222 printk(KERN_DEBUG "pointless warning\n");
4223 abort:
4224 mutex_unlock(&disks_mutex);
4225 if (!error) {
4226 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4227 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4228 }
4229 mddev_put(mddev);
4230 return error;
4231 }
4232
4233 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4234 {
4235 md_alloc(dev, NULL);
4236 return NULL;
4237 }
4238
4239 static int add_named_array(const char *val, struct kernel_param *kp)
4240 {
4241 /* val must be "md_*" where * is not all digits.
4242 * We allocate an array with a large free minor number, and
4243 * set the name to val. val must not already be an active name.
4244 */
4245 int len = strlen(val);
4246 char buf[DISK_NAME_LEN];
4247
4248 while (len && val[len-1] == '\n')
4249 len--;
4250 if (len >= DISK_NAME_LEN)
4251 return -E2BIG;
4252 strlcpy(buf, val, len+1);
4253 if (strncmp(buf, "md_", 3) != 0)
4254 return -EINVAL;
4255 return md_alloc(0, buf);
4256 }
4257
4258 static void md_safemode_timeout(unsigned long data)
4259 {
4260 mddev_t *mddev = (mddev_t *) data;
4261
4262 if (!atomic_read(&mddev->writes_pending)) {
4263 mddev->safemode = 1;
4264 if (mddev->external)
4265 sysfs_notify_dirent(mddev->sysfs_state);
4266 }
4267 md_wakeup_thread(mddev->thread);
4268 }
4269
4270 static int start_dirty_degraded;
4271
4272 static int do_md_run(mddev_t * mddev)
4273 {
4274 int err;
4275 mdk_rdev_t *rdev;
4276 struct gendisk *disk;
4277 struct mdk_personality *pers;
4278
4279 if (list_empty(&mddev->disks))
4280 /* cannot run an array with no devices.. */
4281 return -EINVAL;
4282
4283 if (mddev->pers)
4284 return -EBUSY;
4285
4286 /* These two calls synchronise us with the
4287 * sysfs_remove_group calls in mddev_unlock,
4288 * so they must have completed.
4289 */
4290 mutex_lock(&mddev->open_mutex);
4291 mutex_unlock(&mddev->open_mutex);
4292
4293 /*
4294 * Analyze all RAID superblock(s)
4295 */
4296 if (!mddev->raid_disks) {
4297 if (!mddev->persistent)
4298 return -EINVAL;
4299 analyze_sbs(mddev);
4300 }
4301
4302 if (mddev->level != LEVEL_NONE)
4303 request_module("md-level-%d", mddev->level);
4304 else if (mddev->clevel[0])
4305 request_module("md-%s", mddev->clevel);
4306
4307 /*
4308 * Drop all container device buffers, from now on
4309 * the only valid external interface is through the md
4310 * device.
4311 */
4312 list_for_each_entry(rdev, &mddev->disks, same_set) {
4313 if (test_bit(Faulty, &rdev->flags))
4314 continue;
4315 sync_blockdev(rdev->bdev);
4316 invalidate_bdev(rdev->bdev);
4317
4318 /* perform some consistency tests on the device.
4319 * We don't want the data to overlap the metadata,
4320 * Internal Bitmap issues have been handled elsewhere.
4321 */
4322 if (rdev->data_offset < rdev->sb_start) {
4323 if (mddev->dev_sectors &&
4324 rdev->data_offset + mddev->dev_sectors
4325 > rdev->sb_start) {
4326 printk("md: %s: data overlaps metadata\n",
4327 mdname(mddev));
4328 return -EINVAL;
4329 }
4330 } else {
4331 if (rdev->sb_start + rdev->sb_size/512
4332 > rdev->data_offset) {
4333 printk("md: %s: metadata overlaps data\n",
4334 mdname(mddev));
4335 return -EINVAL;
4336 }
4337 }
4338 sysfs_notify_dirent(rdev->sysfs_state);
4339 }
4340
4341 disk = mddev->gendisk;
4342
4343 spin_lock(&pers_lock);
4344 pers = find_pers(mddev->level, mddev->clevel);
4345 if (!pers || !try_module_get(pers->owner)) {
4346 spin_unlock(&pers_lock);
4347 if (mddev->level != LEVEL_NONE)
4348 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4349 mddev->level);
4350 else
4351 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4352 mddev->clevel);
4353 return -EINVAL;
4354 }
4355 mddev->pers = pers;
4356 spin_unlock(&pers_lock);
4357 if (mddev->level != pers->level) {
4358 mddev->level = pers->level;
4359 mddev->new_level = pers->level;
4360 }
4361 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4362
4363 if (mddev->reshape_position != MaxSector &&
4364 pers->start_reshape == NULL) {
4365 /* This personality cannot handle reshaping... */
4366 mddev->pers = NULL;
4367 module_put(pers->owner);
4368 return -EINVAL;
4369 }
4370
4371 if (pers->sync_request) {
4372 /* Warn if this is a potentially silly
4373 * configuration.
4374 */
4375 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4376 mdk_rdev_t *rdev2;
4377 int warned = 0;
4378
4379 list_for_each_entry(rdev, &mddev->disks, same_set)
4380 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4381 if (rdev < rdev2 &&
4382 rdev->bdev->bd_contains ==
4383 rdev2->bdev->bd_contains) {
4384 printk(KERN_WARNING
4385 "%s: WARNING: %s appears to be"
4386 " on the same physical disk as"
4387 " %s.\n",
4388 mdname(mddev),
4389 bdevname(rdev->bdev,b),
4390 bdevname(rdev2->bdev,b2));
4391 warned = 1;
4392 }
4393 }
4394
4395 if (warned)
4396 printk(KERN_WARNING
4397 "True protection against single-disk"
4398 " failure might be compromised.\n");
4399 }
4400
4401 mddev->recovery = 0;
4402 /* may be over-ridden by personality */
4403 mddev->resync_max_sectors = mddev->dev_sectors;
4404
4405 mddev->barriers_work = 1;
4406 mddev->ok_start_degraded = start_dirty_degraded;
4407
4408 if (start_readonly && mddev->ro == 0)
4409 mddev->ro = 2; /* read-only, but switch on first write */
4410
4411 err = mddev->pers->run(mddev);
4412 if (err)
4413 printk(KERN_ERR "md: pers->run() failed ...\n");
4414 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4415 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4416 " but 'external_size' not in effect?\n", __func__);
4417 printk(KERN_ERR
4418 "md: invalid array_size %llu > default size %llu\n",
4419 (unsigned long long)mddev->array_sectors / 2,
4420 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4421 err = -EINVAL;
4422 mddev->pers->stop(mddev);
4423 }
4424 if (err == 0 && mddev->pers->sync_request) {
4425 err = bitmap_create(mddev);
4426 if (err) {
4427 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4428 mdname(mddev), err);
4429 mddev->pers->stop(mddev);
4430 }
4431 }
4432 if (err) {
4433 module_put(mddev->pers->owner);
4434 mddev->pers = NULL;
4435 bitmap_destroy(mddev);
4436 return err;
4437 }
4438 if (mddev->pers->sync_request) {
4439 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4440 printk(KERN_WARNING
4441 "md: cannot register extra attributes for %s\n",
4442 mdname(mddev));
4443 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4444 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4445 mddev->ro = 0;
4446
4447 atomic_set(&mddev->writes_pending,0);
4448 atomic_set(&mddev->max_corr_read_errors,
4449 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4450 mddev->safemode = 0;
4451 mddev->safemode_timer.function = md_safemode_timeout;
4452 mddev->safemode_timer.data = (unsigned long) mddev;
4453 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4454 mddev->in_sync = 1;
4455
4456 list_for_each_entry(rdev, &mddev->disks, same_set)
4457 if (rdev->raid_disk >= 0) {
4458 char nm[20];
4459 sprintf(nm, "rd%d", rdev->raid_disk);
4460 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4461 printk("md: cannot register %s for %s\n",
4462 nm, mdname(mddev));
4463 }
4464
4465 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4466
4467 if (mddev->flags)
4468 md_update_sb(mddev, 0);
4469
4470 set_capacity(disk, mddev->array_sectors);
4471
4472 md_wakeup_thread(mddev->thread);
4473 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4474
4475 revalidate_disk(mddev->gendisk);
4476 mddev->changed = 1;
4477 md_new_event(mddev);
4478 sysfs_notify_dirent(mddev->sysfs_state);
4479 if (mddev->sysfs_action)
4480 sysfs_notify_dirent(mddev->sysfs_action);
4481 sysfs_notify(&mddev->kobj, NULL, "degraded");
4482 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4483 return 0;
4484 }
4485
4486 static int restart_array(mddev_t *mddev)
4487 {
4488 struct gendisk *disk = mddev->gendisk;
4489
4490 /* Complain if it has no devices */
4491 if (list_empty(&mddev->disks))
4492 return -ENXIO;
4493 if (!mddev->pers)
4494 return -EINVAL;
4495 if (!mddev->ro)
4496 return -EBUSY;
4497 mddev->safemode = 0;
4498 mddev->ro = 0;
4499 set_disk_ro(disk, 0);
4500 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4501 mdname(mddev));
4502 /* Kick recovery or resync if necessary */
4503 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4504 md_wakeup_thread(mddev->thread);
4505 md_wakeup_thread(mddev->sync_thread);
4506 sysfs_notify_dirent(mddev->sysfs_state);
4507 return 0;
4508 }
4509
4510 /* similar to deny_write_access, but accounts for our holding a reference
4511 * to the file ourselves */
4512 static int deny_bitmap_write_access(struct file * file)
4513 {
4514 struct inode *inode = file->f_mapping->host;
4515
4516 spin_lock(&inode->i_lock);
4517 if (atomic_read(&inode->i_writecount) > 1) {
4518 spin_unlock(&inode->i_lock);
4519 return -ETXTBSY;
4520 }
4521 atomic_set(&inode->i_writecount, -1);
4522 spin_unlock(&inode->i_lock);
4523
4524 return 0;
4525 }
4526
4527 void restore_bitmap_write_access(struct file *file)
4528 {
4529 struct inode *inode = file->f_mapping->host;
4530
4531 spin_lock(&inode->i_lock);
4532 atomic_set(&inode->i_writecount, 1);
4533 spin_unlock(&inode->i_lock);
4534 }
4535
4536 /* mode:
4537 * 0 - completely stop and dis-assemble array
4538 * 1 - switch to readonly
4539 * 2 - stop but do not disassemble array
4540 */
4541 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4542 {
4543 int err = 0;
4544 struct gendisk *disk = mddev->gendisk;
4545 mdk_rdev_t *rdev;
4546
4547 mutex_lock(&mddev->open_mutex);
4548 if (atomic_read(&mddev->openers) > is_open) {
4549 printk("md: %s still in use.\n",mdname(mddev));
4550 err = -EBUSY;
4551 } else if (mddev->pers) {
4552
4553 if (mddev->sync_thread) {
4554 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4555 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4556 md_unregister_thread(mddev->sync_thread);
4557 mddev->sync_thread = NULL;
4558 }
4559
4560 del_timer_sync(&mddev->safemode_timer);
4561
4562 switch(mode) {
4563 case 1: /* readonly */
4564 err = -ENXIO;
4565 if (mddev->ro==1)
4566 goto out;
4567 mddev->ro = 1;
4568 break;
4569 case 0: /* disassemble */
4570 case 2: /* stop */
4571 bitmap_flush(mddev);
4572 md_super_wait(mddev);
4573 if (mddev->ro)
4574 set_disk_ro(disk, 0);
4575
4576 mddev->pers->stop(mddev);
4577 mddev->queue->merge_bvec_fn = NULL;
4578 mddev->queue->unplug_fn = NULL;
4579 mddev->queue->backing_dev_info.congested_fn = NULL;
4580 module_put(mddev->pers->owner);
4581 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4582 mddev->to_remove = &md_redundancy_group;
4583 mddev->pers = NULL;
4584 /* tell userspace to handle 'inactive' */
4585 sysfs_notify_dirent(mddev->sysfs_state);
4586
4587 list_for_each_entry(rdev, &mddev->disks, same_set)
4588 if (rdev->raid_disk >= 0) {
4589 char nm[20];
4590 sprintf(nm, "rd%d", rdev->raid_disk);
4591 sysfs_remove_link(&mddev->kobj, nm);
4592 }
4593
4594 set_capacity(disk, 0);
4595 mddev->changed = 1;
4596
4597 if (mddev->ro)
4598 mddev->ro = 0;
4599 }
4600 if (!mddev->in_sync || mddev->flags) {
4601 /* mark array as shutdown cleanly */
4602 mddev->in_sync = 1;
4603 md_update_sb(mddev, 1);
4604 }
4605 if (mode == 1)
4606 set_disk_ro(disk, 1);
4607 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4608 err = 0;
4609 }
4610 out:
4611 mutex_unlock(&mddev->open_mutex);
4612 if (err)
4613 return err;
4614 /*
4615 * Free resources if final stop
4616 */
4617 if (mode == 0) {
4618
4619 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4620
4621 bitmap_destroy(mddev);
4622 if (mddev->bitmap_info.file) {
4623 restore_bitmap_write_access(mddev->bitmap_info.file);
4624 fput(mddev->bitmap_info.file);
4625 mddev->bitmap_info.file = NULL;
4626 }
4627 mddev->bitmap_info.offset = 0;
4628
4629 export_array(mddev);
4630
4631 mddev->array_sectors = 0;
4632 mddev->external_size = 0;
4633 mddev->dev_sectors = 0;
4634 mddev->raid_disks = 0;
4635 mddev->recovery_cp = 0;
4636 mddev->resync_min = 0;
4637 mddev->resync_max = MaxSector;
4638 mddev->reshape_position = MaxSector;
4639 mddev->external = 0;
4640 mddev->persistent = 0;
4641 mddev->level = LEVEL_NONE;
4642 mddev->clevel[0] = 0;
4643 mddev->flags = 0;
4644 mddev->ro = 0;
4645 mddev->metadata_type[0] = 0;
4646 mddev->chunk_sectors = 0;
4647 mddev->ctime = mddev->utime = 0;
4648 mddev->layout = 0;
4649 mddev->max_disks = 0;
4650 mddev->events = 0;
4651 mddev->delta_disks = 0;
4652 mddev->new_level = LEVEL_NONE;
4653 mddev->new_layout = 0;
4654 mddev->new_chunk_sectors = 0;
4655 mddev->curr_resync = 0;
4656 mddev->resync_mismatches = 0;
4657 mddev->suspend_lo = mddev->suspend_hi = 0;
4658 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4659 mddev->recovery = 0;
4660 mddev->in_sync = 0;
4661 mddev->changed = 0;
4662 mddev->degraded = 0;
4663 mddev->barriers_work = 0;
4664 mddev->safemode = 0;
4665 mddev->bitmap_info.offset = 0;
4666 mddev->bitmap_info.default_offset = 0;
4667 mddev->bitmap_info.chunksize = 0;
4668 mddev->bitmap_info.daemon_sleep = 0;
4669 mddev->bitmap_info.max_write_behind = 0;
4670 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4671 if (mddev->hold_active == UNTIL_STOP)
4672 mddev->hold_active = 0;
4673
4674 } else if (mddev->pers)
4675 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4676 mdname(mddev));
4677 err = 0;
4678 blk_integrity_unregister(disk);
4679 md_new_event(mddev);
4680 sysfs_notify_dirent(mddev->sysfs_state);
4681 return err;
4682 }
4683
4684 #ifndef MODULE
4685 static void autorun_array(mddev_t *mddev)
4686 {
4687 mdk_rdev_t *rdev;
4688 int err;
4689
4690 if (list_empty(&mddev->disks))
4691 return;
4692
4693 printk(KERN_INFO "md: running: ");
4694
4695 list_for_each_entry(rdev, &mddev->disks, same_set) {
4696 char b[BDEVNAME_SIZE];
4697 printk("<%s>", bdevname(rdev->bdev,b));
4698 }
4699 printk("\n");
4700
4701 err = do_md_run(mddev);
4702 if (err) {
4703 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4704 do_md_stop(mddev, 0, 0);
4705 }
4706 }
4707
4708 /*
4709 * lets try to run arrays based on all disks that have arrived
4710 * until now. (those are in pending_raid_disks)
4711 *
4712 * the method: pick the first pending disk, collect all disks with
4713 * the same UUID, remove all from the pending list and put them into
4714 * the 'same_array' list. Then order this list based on superblock
4715 * update time (freshest comes first), kick out 'old' disks and
4716 * compare superblocks. If everything's fine then run it.
4717 *
4718 * If "unit" is allocated, then bump its reference count
4719 */
4720 static void autorun_devices(int part)
4721 {
4722 mdk_rdev_t *rdev0, *rdev, *tmp;
4723 mddev_t *mddev;
4724 char b[BDEVNAME_SIZE];
4725
4726 printk(KERN_INFO "md: autorun ...\n");
4727 while (!list_empty(&pending_raid_disks)) {
4728 int unit;
4729 dev_t dev;
4730 LIST_HEAD(candidates);
4731 rdev0 = list_entry(pending_raid_disks.next,
4732 mdk_rdev_t, same_set);
4733
4734 printk(KERN_INFO "md: considering %s ...\n",
4735 bdevname(rdev0->bdev,b));
4736 INIT_LIST_HEAD(&candidates);
4737 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4738 if (super_90_load(rdev, rdev0, 0) >= 0) {
4739 printk(KERN_INFO "md: adding %s ...\n",
4740 bdevname(rdev->bdev,b));
4741 list_move(&rdev->same_set, &candidates);
4742 }
4743 /*
4744 * now we have a set of devices, with all of them having
4745 * mostly sane superblocks. It's time to allocate the
4746 * mddev.
4747 */
4748 if (part) {
4749 dev = MKDEV(mdp_major,
4750 rdev0->preferred_minor << MdpMinorShift);
4751 unit = MINOR(dev) >> MdpMinorShift;
4752 } else {
4753 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4754 unit = MINOR(dev);
4755 }
4756 if (rdev0->preferred_minor != unit) {
4757 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4758 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4759 break;
4760 }
4761
4762 md_probe(dev, NULL, NULL);
4763 mddev = mddev_find(dev);
4764 if (!mddev || !mddev->gendisk) {
4765 if (mddev)
4766 mddev_put(mddev);
4767 printk(KERN_ERR
4768 "md: cannot allocate memory for md drive.\n");
4769 break;
4770 }
4771 if (mddev_lock(mddev))
4772 printk(KERN_WARNING "md: %s locked, cannot run\n",
4773 mdname(mddev));
4774 else if (mddev->raid_disks || mddev->major_version
4775 || !list_empty(&mddev->disks)) {
4776 printk(KERN_WARNING
4777 "md: %s already running, cannot run %s\n",
4778 mdname(mddev), bdevname(rdev0->bdev,b));
4779 mddev_unlock(mddev);
4780 } else {
4781 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4782 mddev->persistent = 1;
4783 rdev_for_each_list(rdev, tmp, &candidates) {
4784 list_del_init(&rdev->same_set);
4785 if (bind_rdev_to_array(rdev, mddev))
4786 export_rdev(rdev);
4787 }
4788 autorun_array(mddev);
4789 mddev_unlock(mddev);
4790 }
4791 /* on success, candidates will be empty, on error
4792 * it won't...
4793 */
4794 rdev_for_each_list(rdev, tmp, &candidates) {
4795 list_del_init(&rdev->same_set);
4796 export_rdev(rdev);
4797 }
4798 mddev_put(mddev);
4799 }
4800 printk(KERN_INFO "md: ... autorun DONE.\n");
4801 }
4802 #endif /* !MODULE */
4803
4804 static int get_version(void __user * arg)
4805 {
4806 mdu_version_t ver;
4807
4808 ver.major = MD_MAJOR_VERSION;
4809 ver.minor = MD_MINOR_VERSION;
4810 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4811
4812 if (copy_to_user(arg, &ver, sizeof(ver)))
4813 return -EFAULT;
4814
4815 return 0;
4816 }
4817
4818 static int get_array_info(mddev_t * mddev, void __user * arg)
4819 {
4820 mdu_array_info_t info;
4821 int nr,working,insync,failed,spare;
4822 mdk_rdev_t *rdev;
4823
4824 nr=working=insync=failed=spare=0;
4825 list_for_each_entry(rdev, &mddev->disks, same_set) {
4826 nr++;
4827 if (test_bit(Faulty, &rdev->flags))
4828 failed++;
4829 else {
4830 working++;
4831 if (test_bit(In_sync, &rdev->flags))
4832 insync++;
4833 else
4834 spare++;
4835 }
4836 }
4837
4838 info.major_version = mddev->major_version;
4839 info.minor_version = mddev->minor_version;
4840 info.patch_version = MD_PATCHLEVEL_VERSION;
4841 info.ctime = mddev->ctime;
4842 info.level = mddev->level;
4843 info.size = mddev->dev_sectors / 2;
4844 if (info.size != mddev->dev_sectors / 2) /* overflow */
4845 info.size = -1;
4846 info.nr_disks = nr;
4847 info.raid_disks = mddev->raid_disks;
4848 info.md_minor = mddev->md_minor;
4849 info.not_persistent= !mddev->persistent;
4850
4851 info.utime = mddev->utime;
4852 info.state = 0;
4853 if (mddev->in_sync)
4854 info.state = (1<<MD_SB_CLEAN);
4855 if (mddev->bitmap && mddev->bitmap_info.offset)
4856 info.state = (1<<MD_SB_BITMAP_PRESENT);
4857 info.active_disks = insync;
4858 info.working_disks = working;
4859 info.failed_disks = failed;
4860 info.spare_disks = spare;
4861
4862 info.layout = mddev->layout;
4863 info.chunk_size = mddev->chunk_sectors << 9;
4864
4865 if (copy_to_user(arg, &info, sizeof(info)))
4866 return -EFAULT;
4867
4868 return 0;
4869 }
4870
4871 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4872 {
4873 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4874 char *ptr, *buf = NULL;
4875 int err = -ENOMEM;
4876
4877 if (md_allow_write(mddev))
4878 file = kmalloc(sizeof(*file), GFP_NOIO);
4879 else
4880 file = kmalloc(sizeof(*file), GFP_KERNEL);
4881
4882 if (!file)
4883 goto out;
4884
4885 /* bitmap disabled, zero the first byte and copy out */
4886 if (!mddev->bitmap || !mddev->bitmap->file) {
4887 file->pathname[0] = '\0';
4888 goto copy_out;
4889 }
4890
4891 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4892 if (!buf)
4893 goto out;
4894
4895 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4896 if (IS_ERR(ptr))
4897 goto out;
4898
4899 strcpy(file->pathname, ptr);
4900
4901 copy_out:
4902 err = 0;
4903 if (copy_to_user(arg, file, sizeof(*file)))
4904 err = -EFAULT;
4905 out:
4906 kfree(buf);
4907 kfree(file);
4908 return err;
4909 }
4910
4911 static int get_disk_info(mddev_t * mddev, void __user * arg)
4912 {
4913 mdu_disk_info_t info;
4914 mdk_rdev_t *rdev;
4915
4916 if (copy_from_user(&info, arg, sizeof(info)))
4917 return -EFAULT;
4918
4919 rdev = find_rdev_nr(mddev, info.number);
4920 if (rdev) {
4921 info.major = MAJOR(rdev->bdev->bd_dev);
4922 info.minor = MINOR(rdev->bdev->bd_dev);
4923 info.raid_disk = rdev->raid_disk;
4924 info.state = 0;
4925 if (test_bit(Faulty, &rdev->flags))
4926 info.state |= (1<<MD_DISK_FAULTY);
4927 else if (test_bit(In_sync, &rdev->flags)) {
4928 info.state |= (1<<MD_DISK_ACTIVE);
4929 info.state |= (1<<MD_DISK_SYNC);
4930 }
4931 if (test_bit(WriteMostly, &rdev->flags))
4932 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4933 } else {
4934 info.major = info.minor = 0;
4935 info.raid_disk = -1;
4936 info.state = (1<<MD_DISK_REMOVED);
4937 }
4938
4939 if (copy_to_user(arg, &info, sizeof(info)))
4940 return -EFAULT;
4941
4942 return 0;
4943 }
4944
4945 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4946 {
4947 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4948 mdk_rdev_t *rdev;
4949 dev_t dev = MKDEV(info->major,info->minor);
4950
4951 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4952 return -EOVERFLOW;
4953
4954 if (!mddev->raid_disks) {
4955 int err;
4956 /* expecting a device which has a superblock */
4957 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4958 if (IS_ERR(rdev)) {
4959 printk(KERN_WARNING
4960 "md: md_import_device returned %ld\n",
4961 PTR_ERR(rdev));
4962 return PTR_ERR(rdev);
4963 }
4964 if (!list_empty(&mddev->disks)) {
4965 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4966 mdk_rdev_t, same_set);
4967 err = super_types[mddev->major_version]
4968 .load_super(rdev, rdev0, mddev->minor_version);
4969 if (err < 0) {
4970 printk(KERN_WARNING
4971 "md: %s has different UUID to %s\n",
4972 bdevname(rdev->bdev,b),
4973 bdevname(rdev0->bdev,b2));
4974 export_rdev(rdev);
4975 return -EINVAL;
4976 }
4977 }
4978 err = bind_rdev_to_array(rdev, mddev);
4979 if (err)
4980 export_rdev(rdev);
4981 return err;
4982 }
4983
4984 /*
4985 * add_new_disk can be used once the array is assembled
4986 * to add "hot spares". They must already have a superblock
4987 * written
4988 */
4989 if (mddev->pers) {
4990 int err;
4991 if (!mddev->pers->hot_add_disk) {
4992 printk(KERN_WARNING
4993 "%s: personality does not support diskops!\n",
4994 mdname(mddev));
4995 return -EINVAL;
4996 }
4997 if (mddev->persistent)
4998 rdev = md_import_device(dev, mddev->major_version,
4999 mddev->minor_version);
5000 else
5001 rdev = md_import_device(dev, -1, -1);
5002 if (IS_ERR(rdev)) {
5003 printk(KERN_WARNING
5004 "md: md_import_device returned %ld\n",
5005 PTR_ERR(rdev));
5006 return PTR_ERR(rdev);
5007 }
5008 /* set save_raid_disk if appropriate */
5009 if (!mddev->persistent) {
5010 if (info->state & (1<<MD_DISK_SYNC) &&
5011 info->raid_disk < mddev->raid_disks)
5012 rdev->raid_disk = info->raid_disk;
5013 else
5014 rdev->raid_disk = -1;
5015 } else
5016 super_types[mddev->major_version].
5017 validate_super(mddev, rdev);
5018 rdev->saved_raid_disk = rdev->raid_disk;
5019
5020 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5021 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5022 set_bit(WriteMostly, &rdev->flags);
5023 else
5024 clear_bit(WriteMostly, &rdev->flags);
5025
5026 rdev->raid_disk = -1;
5027 err = bind_rdev_to_array(rdev, mddev);
5028 if (!err && !mddev->pers->hot_remove_disk) {
5029 /* If there is hot_add_disk but no hot_remove_disk
5030 * then added disks for geometry changes,
5031 * and should be added immediately.
5032 */
5033 super_types[mddev->major_version].
5034 validate_super(mddev, rdev);
5035 err = mddev->pers->hot_add_disk(mddev, rdev);
5036 if (err)
5037 unbind_rdev_from_array(rdev);
5038 }
5039 if (err)
5040 export_rdev(rdev);
5041 else
5042 sysfs_notify_dirent(rdev->sysfs_state);
5043
5044 md_update_sb(mddev, 1);
5045 if (mddev->degraded)
5046 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5047 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5048 md_wakeup_thread(mddev->thread);
5049 return err;
5050 }
5051
5052 /* otherwise, add_new_disk is only allowed
5053 * for major_version==0 superblocks
5054 */
5055 if (mddev->major_version != 0) {
5056 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5057 mdname(mddev));
5058 return -EINVAL;
5059 }
5060
5061 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5062 int err;
5063 rdev = md_import_device(dev, -1, 0);
5064 if (IS_ERR(rdev)) {
5065 printk(KERN_WARNING
5066 "md: error, md_import_device() returned %ld\n",
5067 PTR_ERR(rdev));
5068 return PTR_ERR(rdev);
5069 }
5070 rdev->desc_nr = info->number;
5071 if (info->raid_disk < mddev->raid_disks)
5072 rdev->raid_disk = info->raid_disk;
5073 else
5074 rdev->raid_disk = -1;
5075
5076 if (rdev->raid_disk < mddev->raid_disks)
5077 if (info->state & (1<<MD_DISK_SYNC))
5078 set_bit(In_sync, &rdev->flags);
5079
5080 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5081 set_bit(WriteMostly, &rdev->flags);
5082
5083 if (!mddev->persistent) {
5084 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5085 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5086 } else
5087 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5088 rdev->sectors = rdev->sb_start;
5089
5090 err = bind_rdev_to_array(rdev, mddev);
5091 if (err) {
5092 export_rdev(rdev);
5093 return err;
5094 }
5095 }
5096
5097 return 0;
5098 }
5099
5100 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5101 {
5102 char b[BDEVNAME_SIZE];
5103 mdk_rdev_t *rdev;
5104
5105 rdev = find_rdev(mddev, dev);
5106 if (!rdev)
5107 return -ENXIO;
5108
5109 if (rdev->raid_disk >= 0)
5110 goto busy;
5111
5112 kick_rdev_from_array(rdev);
5113 md_update_sb(mddev, 1);
5114 md_new_event(mddev);
5115
5116 return 0;
5117 busy:
5118 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5119 bdevname(rdev->bdev,b), mdname(mddev));
5120 return -EBUSY;
5121 }
5122
5123 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5124 {
5125 char b[BDEVNAME_SIZE];
5126 int err;
5127 mdk_rdev_t *rdev;
5128
5129 if (!mddev->pers)
5130 return -ENODEV;
5131
5132 if (mddev->major_version != 0) {
5133 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5134 " version-0 superblocks.\n",
5135 mdname(mddev));
5136 return -EINVAL;
5137 }
5138 if (!mddev->pers->hot_add_disk) {
5139 printk(KERN_WARNING
5140 "%s: personality does not support diskops!\n",
5141 mdname(mddev));
5142 return -EINVAL;
5143 }
5144
5145 rdev = md_import_device(dev, -1, 0);
5146 if (IS_ERR(rdev)) {
5147 printk(KERN_WARNING
5148 "md: error, md_import_device() returned %ld\n",
5149 PTR_ERR(rdev));
5150 return -EINVAL;
5151 }
5152
5153 if (mddev->persistent)
5154 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5155 else
5156 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5157
5158 rdev->sectors = rdev->sb_start;
5159
5160 if (test_bit(Faulty, &rdev->flags)) {
5161 printk(KERN_WARNING
5162 "md: can not hot-add faulty %s disk to %s!\n",
5163 bdevname(rdev->bdev,b), mdname(mddev));
5164 err = -EINVAL;
5165 goto abort_export;
5166 }
5167 clear_bit(In_sync, &rdev->flags);
5168 rdev->desc_nr = -1;
5169 rdev->saved_raid_disk = -1;
5170 err = bind_rdev_to_array(rdev, mddev);
5171 if (err)
5172 goto abort_export;
5173
5174 /*
5175 * The rest should better be atomic, we can have disk failures
5176 * noticed in interrupt contexts ...
5177 */
5178
5179 rdev->raid_disk = -1;
5180
5181 md_update_sb(mddev, 1);
5182
5183 /*
5184 * Kick recovery, maybe this spare has to be added to the
5185 * array immediately.
5186 */
5187 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5188 md_wakeup_thread(mddev->thread);
5189 md_new_event(mddev);
5190 return 0;
5191
5192 abort_export:
5193 export_rdev(rdev);
5194 return err;
5195 }
5196
5197 static int set_bitmap_file(mddev_t *mddev, int fd)
5198 {
5199 int err;
5200
5201 if (mddev->pers) {
5202 if (!mddev->pers->quiesce)
5203 return -EBUSY;
5204 if (mddev->recovery || mddev->sync_thread)
5205 return -EBUSY;
5206 /* we should be able to change the bitmap.. */
5207 }
5208
5209
5210 if (fd >= 0) {
5211 if (mddev->bitmap)
5212 return -EEXIST; /* cannot add when bitmap is present */
5213 mddev->bitmap_info.file = fget(fd);
5214
5215 if (mddev->bitmap_info.file == NULL) {
5216 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5217 mdname(mddev));
5218 return -EBADF;
5219 }
5220
5221 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5222 if (err) {
5223 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5224 mdname(mddev));
5225 fput(mddev->bitmap_info.file);
5226 mddev->bitmap_info.file = NULL;
5227 return err;
5228 }
5229 mddev->bitmap_info.offset = 0; /* file overrides offset */
5230 } else if (mddev->bitmap == NULL)
5231 return -ENOENT; /* cannot remove what isn't there */
5232 err = 0;
5233 if (mddev->pers) {
5234 mddev->pers->quiesce(mddev, 1);
5235 if (fd >= 0)
5236 err = bitmap_create(mddev);
5237 if (fd < 0 || err) {
5238 bitmap_destroy(mddev);
5239 fd = -1; /* make sure to put the file */
5240 }
5241 mddev->pers->quiesce(mddev, 0);
5242 }
5243 if (fd < 0) {
5244 if (mddev->bitmap_info.file) {
5245 restore_bitmap_write_access(mddev->bitmap_info.file);
5246 fput(mddev->bitmap_info.file);
5247 }
5248 mddev->bitmap_info.file = NULL;
5249 }
5250
5251 return err;
5252 }
5253
5254 /*
5255 * set_array_info is used two different ways
5256 * The original usage is when creating a new array.
5257 * In this usage, raid_disks is > 0 and it together with
5258 * level, size, not_persistent,layout,chunksize determine the
5259 * shape of the array.
5260 * This will always create an array with a type-0.90.0 superblock.
5261 * The newer usage is when assembling an array.
5262 * In this case raid_disks will be 0, and the major_version field is
5263 * use to determine which style super-blocks are to be found on the devices.
5264 * The minor and patch _version numbers are also kept incase the
5265 * super_block handler wishes to interpret them.
5266 */
5267 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5268 {
5269
5270 if (info->raid_disks == 0) {
5271 /* just setting version number for superblock loading */
5272 if (info->major_version < 0 ||
5273 info->major_version >= ARRAY_SIZE(super_types) ||
5274 super_types[info->major_version].name == NULL) {
5275 /* maybe try to auto-load a module? */
5276 printk(KERN_INFO
5277 "md: superblock version %d not known\n",
5278 info->major_version);
5279 return -EINVAL;
5280 }
5281 mddev->major_version = info->major_version;
5282 mddev->minor_version = info->minor_version;
5283 mddev->patch_version = info->patch_version;
5284 mddev->persistent = !info->not_persistent;
5285 /* ensure mddev_put doesn't delete this now that there
5286 * is some minimal configuration.
5287 */
5288 mddev->ctime = get_seconds();
5289 return 0;
5290 }
5291 mddev->major_version = MD_MAJOR_VERSION;
5292 mddev->minor_version = MD_MINOR_VERSION;
5293 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5294 mddev->ctime = get_seconds();
5295
5296 mddev->level = info->level;
5297 mddev->clevel[0] = 0;
5298 mddev->dev_sectors = 2 * (sector_t)info->size;
5299 mddev->raid_disks = info->raid_disks;
5300 /* don't set md_minor, it is determined by which /dev/md* was
5301 * openned
5302 */
5303 if (info->state & (1<<MD_SB_CLEAN))
5304 mddev->recovery_cp = MaxSector;
5305 else
5306 mddev->recovery_cp = 0;
5307 mddev->persistent = ! info->not_persistent;
5308 mddev->external = 0;
5309
5310 mddev->layout = info->layout;
5311 mddev->chunk_sectors = info->chunk_size >> 9;
5312
5313 mddev->max_disks = MD_SB_DISKS;
5314
5315 if (mddev->persistent)
5316 mddev->flags = 0;
5317 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5318
5319 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5320 mddev->bitmap_info.offset = 0;
5321
5322 mddev->reshape_position = MaxSector;
5323
5324 /*
5325 * Generate a 128 bit UUID
5326 */
5327 get_random_bytes(mddev->uuid, 16);
5328
5329 mddev->new_level = mddev->level;
5330 mddev->new_chunk_sectors = mddev->chunk_sectors;
5331 mddev->new_layout = mddev->layout;
5332 mddev->delta_disks = 0;
5333
5334 return 0;
5335 }
5336
5337 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5338 {
5339 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5340
5341 if (mddev->external_size)
5342 return;
5343
5344 mddev->array_sectors = array_sectors;
5345 }
5346 EXPORT_SYMBOL(md_set_array_sectors);
5347
5348 static int update_size(mddev_t *mddev, sector_t num_sectors)
5349 {
5350 mdk_rdev_t *rdev;
5351 int rv;
5352 int fit = (num_sectors == 0);
5353
5354 if (mddev->pers->resize == NULL)
5355 return -EINVAL;
5356 /* The "num_sectors" is the number of sectors of each device that
5357 * is used. This can only make sense for arrays with redundancy.
5358 * linear and raid0 always use whatever space is available. We can only
5359 * consider changing this number if no resync or reconstruction is
5360 * happening, and if the new size is acceptable. It must fit before the
5361 * sb_start or, if that is <data_offset, it must fit before the size
5362 * of each device. If num_sectors is zero, we find the largest size
5363 * that fits.
5364
5365 */
5366 if (mddev->sync_thread)
5367 return -EBUSY;
5368 if (mddev->bitmap)
5369 /* Sorry, cannot grow a bitmap yet, just remove it,
5370 * grow, and re-add.
5371 */
5372 return -EBUSY;
5373 list_for_each_entry(rdev, &mddev->disks, same_set) {
5374 sector_t avail = rdev->sectors;
5375
5376 if (fit && (num_sectors == 0 || num_sectors > avail))
5377 num_sectors = avail;
5378 if (avail < num_sectors)
5379 return -ENOSPC;
5380 }
5381 rv = mddev->pers->resize(mddev, num_sectors);
5382 if (!rv)
5383 revalidate_disk(mddev->gendisk);
5384 return rv;
5385 }
5386
5387 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5388 {
5389 int rv;
5390 /* change the number of raid disks */
5391 if (mddev->pers->check_reshape == NULL)
5392 return -EINVAL;
5393 if (raid_disks <= 0 ||
5394 raid_disks >= mddev->max_disks)
5395 return -EINVAL;
5396 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5397 return -EBUSY;
5398 mddev->delta_disks = raid_disks - mddev->raid_disks;
5399
5400 rv = mddev->pers->check_reshape(mddev);
5401 return rv;
5402 }
5403
5404
5405 /*
5406 * update_array_info is used to change the configuration of an
5407 * on-line array.
5408 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5409 * fields in the info are checked against the array.
5410 * Any differences that cannot be handled will cause an error.
5411 * Normally, only one change can be managed at a time.
5412 */
5413 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5414 {
5415 int rv = 0;
5416 int cnt = 0;
5417 int state = 0;
5418
5419 /* calculate expected state,ignoring low bits */
5420 if (mddev->bitmap && mddev->bitmap_info.offset)
5421 state |= (1 << MD_SB_BITMAP_PRESENT);
5422
5423 if (mddev->major_version != info->major_version ||
5424 mddev->minor_version != info->minor_version ||
5425 /* mddev->patch_version != info->patch_version || */
5426 mddev->ctime != info->ctime ||
5427 mddev->level != info->level ||
5428 /* mddev->layout != info->layout || */
5429 !mddev->persistent != info->not_persistent||
5430 mddev->chunk_sectors != info->chunk_size >> 9 ||
5431 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5432 ((state^info->state) & 0xfffffe00)
5433 )
5434 return -EINVAL;
5435 /* Check there is only one change */
5436 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5437 cnt++;
5438 if (mddev->raid_disks != info->raid_disks)
5439 cnt++;
5440 if (mddev->layout != info->layout)
5441 cnt++;
5442 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5443 cnt++;
5444 if (cnt == 0)
5445 return 0;
5446 if (cnt > 1)
5447 return -EINVAL;
5448
5449 if (mddev->layout != info->layout) {
5450 /* Change layout
5451 * we don't need to do anything at the md level, the
5452 * personality will take care of it all.
5453 */
5454 if (mddev->pers->check_reshape == NULL)
5455 return -EINVAL;
5456 else {
5457 mddev->new_layout = info->layout;
5458 rv = mddev->pers->check_reshape(mddev);
5459 if (rv)
5460 mddev->new_layout = mddev->layout;
5461 return rv;
5462 }
5463 }
5464 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5465 rv = update_size(mddev, (sector_t)info->size * 2);
5466
5467 if (mddev->raid_disks != info->raid_disks)
5468 rv = update_raid_disks(mddev, info->raid_disks);
5469
5470 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5471 if (mddev->pers->quiesce == NULL)
5472 return -EINVAL;
5473 if (mddev->recovery || mddev->sync_thread)
5474 return -EBUSY;
5475 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5476 /* add the bitmap */
5477 if (mddev->bitmap)
5478 return -EEXIST;
5479 if (mddev->bitmap_info.default_offset == 0)
5480 return -EINVAL;
5481 mddev->bitmap_info.offset =
5482 mddev->bitmap_info.default_offset;
5483 mddev->pers->quiesce(mddev, 1);
5484 rv = bitmap_create(mddev);
5485 if (rv)
5486 bitmap_destroy(mddev);
5487 mddev->pers->quiesce(mddev, 0);
5488 } else {
5489 /* remove the bitmap */
5490 if (!mddev->bitmap)
5491 return -ENOENT;
5492 if (mddev->bitmap->file)
5493 return -EINVAL;
5494 mddev->pers->quiesce(mddev, 1);
5495 bitmap_destroy(mddev);
5496 mddev->pers->quiesce(mddev, 0);
5497 mddev->bitmap_info.offset = 0;
5498 }
5499 }
5500 md_update_sb(mddev, 1);
5501 return rv;
5502 }
5503
5504 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5505 {
5506 mdk_rdev_t *rdev;
5507
5508 if (mddev->pers == NULL)
5509 return -ENODEV;
5510
5511 rdev = find_rdev(mddev, dev);
5512 if (!rdev)
5513 return -ENODEV;
5514
5515 md_error(mddev, rdev);
5516 return 0;
5517 }
5518
5519 /*
5520 * We have a problem here : there is no easy way to give a CHS
5521 * virtual geometry. We currently pretend that we have a 2 heads
5522 * 4 sectors (with a BIG number of cylinders...). This drives
5523 * dosfs just mad... ;-)
5524 */
5525 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5526 {
5527 mddev_t *mddev = bdev->bd_disk->private_data;
5528
5529 geo->heads = 2;
5530 geo->sectors = 4;
5531 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5532 return 0;
5533 }
5534
5535 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5536 unsigned int cmd, unsigned long arg)
5537 {
5538 int err = 0;
5539 void __user *argp = (void __user *)arg;
5540 mddev_t *mddev = NULL;
5541 int ro;
5542
5543 if (!capable(CAP_SYS_ADMIN))
5544 return -EACCES;
5545
5546 /*
5547 * Commands dealing with the RAID driver but not any
5548 * particular array:
5549 */
5550 switch (cmd)
5551 {
5552 case RAID_VERSION:
5553 err = get_version(argp);
5554 goto done;
5555
5556 case PRINT_RAID_DEBUG:
5557 err = 0;
5558 md_print_devices();
5559 goto done;
5560
5561 #ifndef MODULE
5562 case RAID_AUTORUN:
5563 err = 0;
5564 autostart_arrays(arg);
5565 goto done;
5566 #endif
5567 default:;
5568 }
5569
5570 /*
5571 * Commands creating/starting a new array:
5572 */
5573
5574 mddev = bdev->bd_disk->private_data;
5575
5576 if (!mddev) {
5577 BUG();
5578 goto abort;
5579 }
5580
5581 err = mddev_lock(mddev);
5582 if (err) {
5583 printk(KERN_INFO
5584 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5585 err, cmd);
5586 goto abort;
5587 }
5588
5589 switch (cmd)
5590 {
5591 case SET_ARRAY_INFO:
5592 {
5593 mdu_array_info_t info;
5594 if (!arg)
5595 memset(&info, 0, sizeof(info));
5596 else if (copy_from_user(&info, argp, sizeof(info))) {
5597 err = -EFAULT;
5598 goto abort_unlock;
5599 }
5600 if (mddev->pers) {
5601 err = update_array_info(mddev, &info);
5602 if (err) {
5603 printk(KERN_WARNING "md: couldn't update"
5604 " array info. %d\n", err);
5605 goto abort_unlock;
5606 }
5607 goto done_unlock;
5608 }
5609 if (!list_empty(&mddev->disks)) {
5610 printk(KERN_WARNING
5611 "md: array %s already has disks!\n",
5612 mdname(mddev));
5613 err = -EBUSY;
5614 goto abort_unlock;
5615 }
5616 if (mddev->raid_disks) {
5617 printk(KERN_WARNING
5618 "md: array %s already initialised!\n",
5619 mdname(mddev));
5620 err = -EBUSY;
5621 goto abort_unlock;
5622 }
5623 err = set_array_info(mddev, &info);
5624 if (err) {
5625 printk(KERN_WARNING "md: couldn't set"
5626 " array info. %d\n", err);
5627 goto abort_unlock;
5628 }
5629 }
5630 goto done_unlock;
5631
5632 default:;
5633 }
5634
5635 /*
5636 * Commands querying/configuring an existing array:
5637 */
5638 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5639 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5640 if ((!mddev->raid_disks && !mddev->external)
5641 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5642 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5643 && cmd != GET_BITMAP_FILE) {
5644 err = -ENODEV;
5645 goto abort_unlock;
5646 }
5647
5648 /*
5649 * Commands even a read-only array can execute:
5650 */
5651 switch (cmd)
5652 {
5653 case GET_ARRAY_INFO:
5654 err = get_array_info(mddev, argp);
5655 goto done_unlock;
5656
5657 case GET_BITMAP_FILE:
5658 err = get_bitmap_file(mddev, argp);
5659 goto done_unlock;
5660
5661 case GET_DISK_INFO:
5662 err = get_disk_info(mddev, argp);
5663 goto done_unlock;
5664
5665 case RESTART_ARRAY_RW:
5666 err = restart_array(mddev);
5667 goto done_unlock;
5668
5669 case STOP_ARRAY:
5670 err = do_md_stop(mddev, 0, 1);
5671 goto done_unlock;
5672
5673 case STOP_ARRAY_RO:
5674 err = do_md_stop(mddev, 1, 1);
5675 goto done_unlock;
5676
5677 case BLKROSET:
5678 if (get_user(ro, (int __user *)(arg))) {
5679 err = -EFAULT;
5680 goto done_unlock;
5681 }
5682 err = -EINVAL;
5683
5684 /* if the bdev is going readonly the value of mddev->ro
5685 * does not matter, no writes are coming
5686 */
5687 if (ro)
5688 goto done_unlock;
5689
5690 /* are we are already prepared for writes? */
5691 if (mddev->ro != 1)
5692 goto done_unlock;
5693
5694 /* transitioning to readauto need only happen for
5695 * arrays that call md_write_start
5696 */
5697 if (mddev->pers) {
5698 err = restart_array(mddev);
5699 if (err == 0) {
5700 mddev->ro = 2;
5701 set_disk_ro(mddev->gendisk, 0);
5702 }
5703 }
5704 goto done_unlock;
5705 }
5706
5707 /*
5708 * The remaining ioctls are changing the state of the
5709 * superblock, so we do not allow them on read-only arrays.
5710 * However non-MD ioctls (e.g. get-size) will still come through
5711 * here and hit the 'default' below, so only disallow
5712 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5713 */
5714 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5715 if (mddev->ro == 2) {
5716 mddev->ro = 0;
5717 sysfs_notify_dirent(mddev->sysfs_state);
5718 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5719 md_wakeup_thread(mddev->thread);
5720 } else {
5721 err = -EROFS;
5722 goto abort_unlock;
5723 }
5724 }
5725
5726 switch (cmd)
5727 {
5728 case ADD_NEW_DISK:
5729 {
5730 mdu_disk_info_t info;
5731 if (copy_from_user(&info, argp, sizeof(info)))
5732 err = -EFAULT;
5733 else
5734 err = add_new_disk(mddev, &info);
5735 goto done_unlock;
5736 }
5737
5738 case HOT_REMOVE_DISK:
5739 err = hot_remove_disk(mddev, new_decode_dev(arg));
5740 goto done_unlock;
5741
5742 case HOT_ADD_DISK:
5743 err = hot_add_disk(mddev, new_decode_dev(arg));
5744 goto done_unlock;
5745
5746 case SET_DISK_FAULTY:
5747 err = set_disk_faulty(mddev, new_decode_dev(arg));
5748 goto done_unlock;
5749
5750 case RUN_ARRAY:
5751 err = do_md_run(mddev);
5752 goto done_unlock;
5753
5754 case SET_BITMAP_FILE:
5755 err = set_bitmap_file(mddev, (int)arg);
5756 goto done_unlock;
5757
5758 default:
5759 err = -EINVAL;
5760 goto abort_unlock;
5761 }
5762
5763 done_unlock:
5764 abort_unlock:
5765 if (mddev->hold_active == UNTIL_IOCTL &&
5766 err != -EINVAL)
5767 mddev->hold_active = 0;
5768 mddev_unlock(mddev);
5769
5770 return err;
5771 done:
5772 if (err)
5773 MD_BUG();
5774 abort:
5775 return err;
5776 }
5777 #ifdef CONFIG_COMPAT
5778 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5779 unsigned int cmd, unsigned long arg)
5780 {
5781 switch (cmd) {
5782 case HOT_REMOVE_DISK:
5783 case HOT_ADD_DISK:
5784 case SET_DISK_FAULTY:
5785 case SET_BITMAP_FILE:
5786 /* These take in integer arg, do not convert */
5787 break;
5788 default:
5789 arg = (unsigned long)compat_ptr(arg);
5790 break;
5791 }
5792
5793 return md_ioctl(bdev, mode, cmd, arg);
5794 }
5795 #endif /* CONFIG_COMPAT */
5796
5797 static int md_open(struct block_device *bdev, fmode_t mode)
5798 {
5799 /*
5800 * Succeed if we can lock the mddev, which confirms that
5801 * it isn't being stopped right now.
5802 */
5803 mddev_t *mddev = mddev_find(bdev->bd_dev);
5804 int err;
5805
5806 if (mddev->gendisk != bdev->bd_disk) {
5807 /* we are racing with mddev_put which is discarding this
5808 * bd_disk.
5809 */
5810 mddev_put(mddev);
5811 /* Wait until bdev->bd_disk is definitely gone */
5812 flush_scheduled_work();
5813 /* Then retry the open from the top */
5814 return -ERESTARTSYS;
5815 }
5816 BUG_ON(mddev != bdev->bd_disk->private_data);
5817
5818 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5819 goto out;
5820
5821 err = 0;
5822 atomic_inc(&mddev->openers);
5823 mutex_unlock(&mddev->open_mutex);
5824
5825 check_disk_change(bdev);
5826 out:
5827 return err;
5828 }
5829
5830 static int md_release(struct gendisk *disk, fmode_t mode)
5831 {
5832 mddev_t *mddev = disk->private_data;
5833
5834 BUG_ON(!mddev);
5835 atomic_dec(&mddev->openers);
5836 mddev_put(mddev);
5837
5838 return 0;
5839 }
5840
5841 static int md_media_changed(struct gendisk *disk)
5842 {
5843 mddev_t *mddev = disk->private_data;
5844
5845 return mddev->changed;
5846 }
5847
5848 static int md_revalidate(struct gendisk *disk)
5849 {
5850 mddev_t *mddev = disk->private_data;
5851
5852 mddev->changed = 0;
5853 return 0;
5854 }
5855 static const struct block_device_operations md_fops =
5856 {
5857 .owner = THIS_MODULE,
5858 .open = md_open,
5859 .release = md_release,
5860 .ioctl = md_ioctl,
5861 #ifdef CONFIG_COMPAT
5862 .compat_ioctl = md_compat_ioctl,
5863 #endif
5864 .getgeo = md_getgeo,
5865 .media_changed = md_media_changed,
5866 .revalidate_disk= md_revalidate,
5867 };
5868
5869 static int md_thread(void * arg)
5870 {
5871 mdk_thread_t *thread = arg;
5872
5873 /*
5874 * md_thread is a 'system-thread', it's priority should be very
5875 * high. We avoid resource deadlocks individually in each
5876 * raid personality. (RAID5 does preallocation) We also use RR and
5877 * the very same RT priority as kswapd, thus we will never get
5878 * into a priority inversion deadlock.
5879 *
5880 * we definitely have to have equal or higher priority than
5881 * bdflush, otherwise bdflush will deadlock if there are too
5882 * many dirty RAID5 blocks.
5883 */
5884
5885 allow_signal(SIGKILL);
5886 while (!kthread_should_stop()) {
5887
5888 /* We need to wait INTERRUPTIBLE so that
5889 * we don't add to the load-average.
5890 * That means we need to be sure no signals are
5891 * pending
5892 */
5893 if (signal_pending(current))
5894 flush_signals(current);
5895
5896 wait_event_interruptible_timeout
5897 (thread->wqueue,
5898 test_bit(THREAD_WAKEUP, &thread->flags)
5899 || kthread_should_stop(),
5900 thread->timeout);
5901
5902 clear_bit(THREAD_WAKEUP, &thread->flags);
5903
5904 thread->run(thread->mddev);
5905 }
5906
5907 return 0;
5908 }
5909
5910 void md_wakeup_thread(mdk_thread_t *thread)
5911 {
5912 if (thread) {
5913 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5914 set_bit(THREAD_WAKEUP, &thread->flags);
5915 wake_up(&thread->wqueue);
5916 }
5917 }
5918
5919 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5920 const char *name)
5921 {
5922 mdk_thread_t *thread;
5923
5924 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5925 if (!thread)
5926 return NULL;
5927
5928 init_waitqueue_head(&thread->wqueue);
5929
5930 thread->run = run;
5931 thread->mddev = mddev;
5932 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5933 thread->tsk = kthread_run(md_thread, thread,
5934 "%s_%s",
5935 mdname(thread->mddev),
5936 name ?: mddev->pers->name);
5937 if (IS_ERR(thread->tsk)) {
5938 kfree(thread);
5939 return NULL;
5940 }
5941 return thread;
5942 }
5943
5944 void md_unregister_thread(mdk_thread_t *thread)
5945 {
5946 if (!thread)
5947 return;
5948 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5949
5950 kthread_stop(thread->tsk);
5951 kfree(thread);
5952 }
5953
5954 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5955 {
5956 if (!mddev) {
5957 MD_BUG();
5958 return;
5959 }
5960
5961 if (!rdev || test_bit(Faulty, &rdev->flags))
5962 return;
5963
5964 if (mddev->external)
5965 set_bit(Blocked, &rdev->flags);
5966 /*
5967 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5968 mdname(mddev),
5969 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5970 __builtin_return_address(0),__builtin_return_address(1),
5971 __builtin_return_address(2),__builtin_return_address(3));
5972 */
5973 if (!mddev->pers)
5974 return;
5975 if (!mddev->pers->error_handler)
5976 return;
5977 mddev->pers->error_handler(mddev,rdev);
5978 if (mddev->degraded)
5979 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5980 set_bit(StateChanged, &rdev->flags);
5981 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5982 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5983 md_wakeup_thread(mddev->thread);
5984 md_new_event_inintr(mddev);
5985 }
5986
5987 /* seq_file implementation /proc/mdstat */
5988
5989 static void status_unused(struct seq_file *seq)
5990 {
5991 int i = 0;
5992 mdk_rdev_t *rdev;
5993
5994 seq_printf(seq, "unused devices: ");
5995
5996 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5997 char b[BDEVNAME_SIZE];
5998 i++;
5999 seq_printf(seq, "%s ",
6000 bdevname(rdev->bdev,b));
6001 }
6002 if (!i)
6003 seq_printf(seq, "<none>");
6004
6005 seq_printf(seq, "\n");
6006 }
6007
6008
6009 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6010 {
6011 sector_t max_sectors, resync, res;
6012 unsigned long dt, db;
6013 sector_t rt;
6014 int scale;
6015 unsigned int per_milli;
6016
6017 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6018
6019 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6020 max_sectors = mddev->resync_max_sectors;
6021 else
6022 max_sectors = mddev->dev_sectors;
6023
6024 /*
6025 * Should not happen.
6026 */
6027 if (!max_sectors) {
6028 MD_BUG();
6029 return;
6030 }
6031 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6032 * in a sector_t, and (max_sectors>>scale) will fit in a
6033 * u32, as those are the requirements for sector_div.
6034 * Thus 'scale' must be at least 10
6035 */
6036 scale = 10;
6037 if (sizeof(sector_t) > sizeof(unsigned long)) {
6038 while ( max_sectors/2 > (1ULL<<(scale+32)))
6039 scale++;
6040 }
6041 res = (resync>>scale)*1000;
6042 sector_div(res, (u32)((max_sectors>>scale)+1));
6043
6044 per_milli = res;
6045 {
6046 int i, x = per_milli/50, y = 20-x;
6047 seq_printf(seq, "[");
6048 for (i = 0; i < x; i++)
6049 seq_printf(seq, "=");
6050 seq_printf(seq, ">");
6051 for (i = 0; i < y; i++)
6052 seq_printf(seq, ".");
6053 seq_printf(seq, "] ");
6054 }
6055 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6056 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6057 "reshape" :
6058 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6059 "check" :
6060 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6061 "resync" : "recovery"))),
6062 per_milli/10, per_milli % 10,
6063 (unsigned long long) resync/2,
6064 (unsigned long long) max_sectors/2);
6065
6066 /*
6067 * dt: time from mark until now
6068 * db: blocks written from mark until now
6069 * rt: remaining time
6070 *
6071 * rt is a sector_t, so could be 32bit or 64bit.
6072 * So we divide before multiply in case it is 32bit and close
6073 * to the limit.
6074 * We scale the divisor (db) by 32 to avoid loosing precision
6075 * near the end of resync when the number of remaining sectors
6076 * is close to 'db'.
6077 * We then divide rt by 32 after multiplying by db to compensate.
6078 * The '+1' avoids division by zero if db is very small.
6079 */
6080 dt = ((jiffies - mddev->resync_mark) / HZ);
6081 if (!dt) dt++;
6082 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6083 - mddev->resync_mark_cnt;
6084
6085 rt = max_sectors - resync; /* number of remaining sectors */
6086 sector_div(rt, db/32+1);
6087 rt *= dt;
6088 rt >>= 5;
6089
6090 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6091 ((unsigned long)rt % 60)/6);
6092
6093 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6094 }
6095
6096 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6097 {
6098 struct list_head *tmp;
6099 loff_t l = *pos;
6100 mddev_t *mddev;
6101
6102 if (l >= 0x10000)
6103 return NULL;
6104 if (!l--)
6105 /* header */
6106 return (void*)1;
6107
6108 spin_lock(&all_mddevs_lock);
6109 list_for_each(tmp,&all_mddevs)
6110 if (!l--) {
6111 mddev = list_entry(tmp, mddev_t, all_mddevs);
6112 mddev_get(mddev);
6113 spin_unlock(&all_mddevs_lock);
6114 return mddev;
6115 }
6116 spin_unlock(&all_mddevs_lock);
6117 if (!l--)
6118 return (void*)2;/* tail */
6119 return NULL;
6120 }
6121
6122 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6123 {
6124 struct list_head *tmp;
6125 mddev_t *next_mddev, *mddev = v;
6126
6127 ++*pos;
6128 if (v == (void*)2)
6129 return NULL;
6130
6131 spin_lock(&all_mddevs_lock);
6132 if (v == (void*)1)
6133 tmp = all_mddevs.next;
6134 else
6135 tmp = mddev->all_mddevs.next;
6136 if (tmp != &all_mddevs)
6137 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6138 else {
6139 next_mddev = (void*)2;
6140 *pos = 0x10000;
6141 }
6142 spin_unlock(&all_mddevs_lock);
6143
6144 if (v != (void*)1)
6145 mddev_put(mddev);
6146 return next_mddev;
6147
6148 }
6149
6150 static void md_seq_stop(struct seq_file *seq, void *v)
6151 {
6152 mddev_t *mddev = v;
6153
6154 if (mddev && v != (void*)1 && v != (void*)2)
6155 mddev_put(mddev);
6156 }
6157
6158 struct mdstat_info {
6159 int event;
6160 };
6161
6162 static int md_seq_show(struct seq_file *seq, void *v)
6163 {
6164 mddev_t *mddev = v;
6165 sector_t sectors;
6166 mdk_rdev_t *rdev;
6167 struct mdstat_info *mi = seq->private;
6168 struct bitmap *bitmap;
6169
6170 if (v == (void*)1) {
6171 struct mdk_personality *pers;
6172 seq_printf(seq, "Personalities : ");
6173 spin_lock(&pers_lock);
6174 list_for_each_entry(pers, &pers_list, list)
6175 seq_printf(seq, "[%s] ", pers->name);
6176
6177 spin_unlock(&pers_lock);
6178 seq_printf(seq, "\n");
6179 mi->event = atomic_read(&md_event_count);
6180 return 0;
6181 }
6182 if (v == (void*)2) {
6183 status_unused(seq);
6184 return 0;
6185 }
6186
6187 if (mddev_lock(mddev) < 0)
6188 return -EINTR;
6189
6190 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6191 seq_printf(seq, "%s : %sactive", mdname(mddev),
6192 mddev->pers ? "" : "in");
6193 if (mddev->pers) {
6194 if (mddev->ro==1)
6195 seq_printf(seq, " (read-only)");
6196 if (mddev->ro==2)
6197 seq_printf(seq, " (auto-read-only)");
6198 seq_printf(seq, " %s", mddev->pers->name);
6199 }
6200
6201 sectors = 0;
6202 list_for_each_entry(rdev, &mddev->disks, same_set) {
6203 char b[BDEVNAME_SIZE];
6204 seq_printf(seq, " %s[%d]",
6205 bdevname(rdev->bdev,b), rdev->desc_nr);
6206 if (test_bit(WriteMostly, &rdev->flags))
6207 seq_printf(seq, "(W)");
6208 if (test_bit(Faulty, &rdev->flags)) {
6209 seq_printf(seq, "(F)");
6210 continue;
6211 } else if (rdev->raid_disk < 0)
6212 seq_printf(seq, "(S)"); /* spare */
6213 sectors += rdev->sectors;
6214 }
6215
6216 if (!list_empty(&mddev->disks)) {
6217 if (mddev->pers)
6218 seq_printf(seq, "\n %llu blocks",
6219 (unsigned long long)
6220 mddev->array_sectors / 2);
6221 else
6222 seq_printf(seq, "\n %llu blocks",
6223 (unsigned long long)sectors / 2);
6224 }
6225 if (mddev->persistent) {
6226 if (mddev->major_version != 0 ||
6227 mddev->minor_version != 90) {
6228 seq_printf(seq," super %d.%d",
6229 mddev->major_version,
6230 mddev->minor_version);
6231 }
6232 } else if (mddev->external)
6233 seq_printf(seq, " super external:%s",
6234 mddev->metadata_type);
6235 else
6236 seq_printf(seq, " super non-persistent");
6237
6238 if (mddev->pers) {
6239 mddev->pers->status(seq, mddev);
6240 seq_printf(seq, "\n ");
6241 if (mddev->pers->sync_request) {
6242 if (mddev->curr_resync > 2) {
6243 status_resync(seq, mddev);
6244 seq_printf(seq, "\n ");
6245 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6246 seq_printf(seq, "\tresync=DELAYED\n ");
6247 else if (mddev->recovery_cp < MaxSector)
6248 seq_printf(seq, "\tresync=PENDING\n ");
6249 }
6250 } else
6251 seq_printf(seq, "\n ");
6252
6253 if ((bitmap = mddev->bitmap)) {
6254 unsigned long chunk_kb;
6255 unsigned long flags;
6256 spin_lock_irqsave(&bitmap->lock, flags);
6257 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6258 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6259 "%lu%s chunk",
6260 bitmap->pages - bitmap->missing_pages,
6261 bitmap->pages,
6262 (bitmap->pages - bitmap->missing_pages)
6263 << (PAGE_SHIFT - 10),
6264 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6265 chunk_kb ? "KB" : "B");
6266 if (bitmap->file) {
6267 seq_printf(seq, ", file: ");
6268 seq_path(seq, &bitmap->file->f_path, " \t\n");
6269 }
6270
6271 seq_printf(seq, "\n");
6272 spin_unlock_irqrestore(&bitmap->lock, flags);
6273 }
6274
6275 seq_printf(seq, "\n");
6276 }
6277 mddev_unlock(mddev);
6278
6279 return 0;
6280 }
6281
6282 static const struct seq_operations md_seq_ops = {
6283 .start = md_seq_start,
6284 .next = md_seq_next,
6285 .stop = md_seq_stop,
6286 .show = md_seq_show,
6287 };
6288
6289 static int md_seq_open(struct inode *inode, struct file *file)
6290 {
6291 int error;
6292 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6293 if (mi == NULL)
6294 return -ENOMEM;
6295
6296 error = seq_open(file, &md_seq_ops);
6297 if (error)
6298 kfree(mi);
6299 else {
6300 struct seq_file *p = file->private_data;
6301 p->private = mi;
6302 mi->event = atomic_read(&md_event_count);
6303 }
6304 return error;
6305 }
6306
6307 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6308 {
6309 struct seq_file *m = filp->private_data;
6310 struct mdstat_info *mi = m->private;
6311 int mask;
6312
6313 poll_wait(filp, &md_event_waiters, wait);
6314
6315 /* always allow read */
6316 mask = POLLIN | POLLRDNORM;
6317
6318 if (mi->event != atomic_read(&md_event_count))
6319 mask |= POLLERR | POLLPRI;
6320 return mask;
6321 }
6322
6323 static const struct file_operations md_seq_fops = {
6324 .owner = THIS_MODULE,
6325 .open = md_seq_open,
6326 .read = seq_read,
6327 .llseek = seq_lseek,
6328 .release = seq_release_private,
6329 .poll = mdstat_poll,
6330 };
6331
6332 int register_md_personality(struct mdk_personality *p)
6333 {
6334 spin_lock(&pers_lock);
6335 list_add_tail(&p->list, &pers_list);
6336 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6337 spin_unlock(&pers_lock);
6338 return 0;
6339 }
6340
6341 int unregister_md_personality(struct mdk_personality *p)
6342 {
6343 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6344 spin_lock(&pers_lock);
6345 list_del_init(&p->list);
6346 spin_unlock(&pers_lock);
6347 return 0;
6348 }
6349
6350 static int is_mddev_idle(mddev_t *mddev, int init)
6351 {
6352 mdk_rdev_t * rdev;
6353 int idle;
6354 int curr_events;
6355
6356 idle = 1;
6357 rcu_read_lock();
6358 rdev_for_each_rcu(rdev, mddev) {
6359 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6360 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6361 (int)part_stat_read(&disk->part0, sectors[1]) -
6362 atomic_read(&disk->sync_io);
6363 /* sync IO will cause sync_io to increase before the disk_stats
6364 * as sync_io is counted when a request starts, and
6365 * disk_stats is counted when it completes.
6366 * So resync activity will cause curr_events to be smaller than
6367 * when there was no such activity.
6368 * non-sync IO will cause disk_stat to increase without
6369 * increasing sync_io so curr_events will (eventually)
6370 * be larger than it was before. Once it becomes
6371 * substantially larger, the test below will cause
6372 * the array to appear non-idle, and resync will slow
6373 * down.
6374 * If there is a lot of outstanding resync activity when
6375 * we set last_event to curr_events, then all that activity
6376 * completing might cause the array to appear non-idle
6377 * and resync will be slowed down even though there might
6378 * not have been non-resync activity. This will only
6379 * happen once though. 'last_events' will soon reflect
6380 * the state where there is little or no outstanding
6381 * resync requests, and further resync activity will
6382 * always make curr_events less than last_events.
6383 *
6384 */
6385 if (init || curr_events - rdev->last_events > 64) {
6386 rdev->last_events = curr_events;
6387 idle = 0;
6388 }
6389 }
6390 rcu_read_unlock();
6391 return idle;
6392 }
6393
6394 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6395 {
6396 /* another "blocks" (512byte) blocks have been synced */
6397 atomic_sub(blocks, &mddev->recovery_active);
6398 wake_up(&mddev->recovery_wait);
6399 if (!ok) {
6400 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6401 md_wakeup_thread(mddev->thread);
6402 // stop recovery, signal do_sync ....
6403 }
6404 }
6405
6406
6407 /* md_write_start(mddev, bi)
6408 * If we need to update some array metadata (e.g. 'active' flag
6409 * in superblock) before writing, schedule a superblock update
6410 * and wait for it to complete.
6411 */
6412 void md_write_start(mddev_t *mddev, struct bio *bi)
6413 {
6414 int did_change = 0;
6415 if (bio_data_dir(bi) != WRITE)
6416 return;
6417
6418 BUG_ON(mddev->ro == 1);
6419 if (mddev->ro == 2) {
6420 /* need to switch to read/write */
6421 mddev->ro = 0;
6422 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6423 md_wakeup_thread(mddev->thread);
6424 md_wakeup_thread(mddev->sync_thread);
6425 did_change = 1;
6426 }
6427 atomic_inc(&mddev->writes_pending);
6428 if (mddev->safemode == 1)
6429 mddev->safemode = 0;
6430 if (mddev->in_sync) {
6431 spin_lock_irq(&mddev->write_lock);
6432 if (mddev->in_sync) {
6433 mddev->in_sync = 0;
6434 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6435 md_wakeup_thread(mddev->thread);
6436 did_change = 1;
6437 }
6438 spin_unlock_irq(&mddev->write_lock);
6439 }
6440 if (did_change)
6441 sysfs_notify_dirent(mddev->sysfs_state);
6442 wait_event(mddev->sb_wait,
6443 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6444 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6445 }
6446
6447 void md_write_end(mddev_t *mddev)
6448 {
6449 if (atomic_dec_and_test(&mddev->writes_pending)) {
6450 if (mddev->safemode == 2)
6451 md_wakeup_thread(mddev->thread);
6452 else if (mddev->safemode_delay)
6453 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6454 }
6455 }
6456
6457 /* md_allow_write(mddev)
6458 * Calling this ensures that the array is marked 'active' so that writes
6459 * may proceed without blocking. It is important to call this before
6460 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6461 * Must be called with mddev_lock held.
6462 *
6463 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6464 * is dropped, so return -EAGAIN after notifying userspace.
6465 */
6466 int md_allow_write(mddev_t *mddev)
6467 {
6468 if (!mddev->pers)
6469 return 0;
6470 if (mddev->ro)
6471 return 0;
6472 if (!mddev->pers->sync_request)
6473 return 0;
6474
6475 spin_lock_irq(&mddev->write_lock);
6476 if (mddev->in_sync) {
6477 mddev->in_sync = 0;
6478 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6479 if (mddev->safemode_delay &&
6480 mddev->safemode == 0)
6481 mddev->safemode = 1;
6482 spin_unlock_irq(&mddev->write_lock);
6483 md_update_sb(mddev, 0);
6484 sysfs_notify_dirent(mddev->sysfs_state);
6485 } else
6486 spin_unlock_irq(&mddev->write_lock);
6487
6488 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6489 return -EAGAIN;
6490 else
6491 return 0;
6492 }
6493 EXPORT_SYMBOL_GPL(md_allow_write);
6494
6495 #define SYNC_MARKS 10
6496 #define SYNC_MARK_STEP (3*HZ)
6497 void md_do_sync(mddev_t *mddev)
6498 {
6499 mddev_t *mddev2;
6500 unsigned int currspeed = 0,
6501 window;
6502 sector_t max_sectors,j, io_sectors;
6503 unsigned long mark[SYNC_MARKS];
6504 sector_t mark_cnt[SYNC_MARKS];
6505 int last_mark,m;
6506 struct list_head *tmp;
6507 sector_t last_check;
6508 int skipped = 0;
6509 mdk_rdev_t *rdev;
6510 char *desc;
6511
6512 /* just incase thread restarts... */
6513 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6514 return;
6515 if (mddev->ro) /* never try to sync a read-only array */
6516 return;
6517
6518 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6519 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6520 desc = "data-check";
6521 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6522 desc = "requested-resync";
6523 else
6524 desc = "resync";
6525 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6526 desc = "reshape";
6527 else
6528 desc = "recovery";
6529
6530 /* we overload curr_resync somewhat here.
6531 * 0 == not engaged in resync at all
6532 * 2 == checking that there is no conflict with another sync
6533 * 1 == like 2, but have yielded to allow conflicting resync to
6534 * commense
6535 * other == active in resync - this many blocks
6536 *
6537 * Before starting a resync we must have set curr_resync to
6538 * 2, and then checked that every "conflicting" array has curr_resync
6539 * less than ours. When we find one that is the same or higher
6540 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6541 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6542 * This will mean we have to start checking from the beginning again.
6543 *
6544 */
6545
6546 do {
6547 mddev->curr_resync = 2;
6548
6549 try_again:
6550 if (kthread_should_stop())
6551 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6552
6553 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6554 goto skip;
6555 for_each_mddev(mddev2, tmp) {
6556 if (mddev2 == mddev)
6557 continue;
6558 if (!mddev->parallel_resync
6559 && mddev2->curr_resync
6560 && match_mddev_units(mddev, mddev2)) {
6561 DEFINE_WAIT(wq);
6562 if (mddev < mddev2 && mddev->curr_resync == 2) {
6563 /* arbitrarily yield */
6564 mddev->curr_resync = 1;
6565 wake_up(&resync_wait);
6566 }
6567 if (mddev > mddev2 && mddev->curr_resync == 1)
6568 /* no need to wait here, we can wait the next
6569 * time 'round when curr_resync == 2
6570 */
6571 continue;
6572 /* We need to wait 'interruptible' so as not to
6573 * contribute to the load average, and not to
6574 * be caught by 'softlockup'
6575 */
6576 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6577 if (!kthread_should_stop() &&
6578 mddev2->curr_resync >= mddev->curr_resync) {
6579 printk(KERN_INFO "md: delaying %s of %s"
6580 " until %s has finished (they"
6581 " share one or more physical units)\n",
6582 desc, mdname(mddev), mdname(mddev2));
6583 mddev_put(mddev2);
6584 if (signal_pending(current))
6585 flush_signals(current);
6586 schedule();
6587 finish_wait(&resync_wait, &wq);
6588 goto try_again;
6589 }
6590 finish_wait(&resync_wait, &wq);
6591 }
6592 }
6593 } while (mddev->curr_resync < 2);
6594
6595 j = 0;
6596 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6597 /* resync follows the size requested by the personality,
6598 * which defaults to physical size, but can be virtual size
6599 */
6600 max_sectors = mddev->resync_max_sectors;
6601 mddev->resync_mismatches = 0;
6602 /* we don't use the checkpoint if there's a bitmap */
6603 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6604 j = mddev->resync_min;
6605 else if (!mddev->bitmap)
6606 j = mddev->recovery_cp;
6607
6608 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6609 max_sectors = mddev->dev_sectors;
6610 else {
6611 /* recovery follows the physical size of devices */
6612 max_sectors = mddev->dev_sectors;
6613 j = MaxSector;
6614 rcu_read_lock();
6615 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6616 if (rdev->raid_disk >= 0 &&
6617 !test_bit(Faulty, &rdev->flags) &&
6618 !test_bit(In_sync, &rdev->flags) &&
6619 rdev->recovery_offset < j)
6620 j = rdev->recovery_offset;
6621 rcu_read_unlock();
6622 }
6623
6624 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6625 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6626 " %d KB/sec/disk.\n", speed_min(mddev));
6627 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6628 "(but not more than %d KB/sec) for %s.\n",
6629 speed_max(mddev), desc);
6630
6631 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6632
6633 io_sectors = 0;
6634 for (m = 0; m < SYNC_MARKS; m++) {
6635 mark[m] = jiffies;
6636 mark_cnt[m] = io_sectors;
6637 }
6638 last_mark = 0;
6639 mddev->resync_mark = mark[last_mark];
6640 mddev->resync_mark_cnt = mark_cnt[last_mark];
6641
6642 /*
6643 * Tune reconstruction:
6644 */
6645 window = 32*(PAGE_SIZE/512);
6646 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6647 window/2,(unsigned long long) max_sectors/2);
6648
6649 atomic_set(&mddev->recovery_active, 0);
6650 last_check = 0;
6651
6652 if (j>2) {
6653 printk(KERN_INFO
6654 "md: resuming %s of %s from checkpoint.\n",
6655 desc, mdname(mddev));
6656 mddev->curr_resync = j;
6657 }
6658 mddev->curr_resync_completed = mddev->curr_resync;
6659
6660 while (j < max_sectors) {
6661 sector_t sectors;
6662
6663 skipped = 0;
6664
6665 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6666 ((mddev->curr_resync > mddev->curr_resync_completed &&
6667 (mddev->curr_resync - mddev->curr_resync_completed)
6668 > (max_sectors >> 4)) ||
6669 (j - mddev->curr_resync_completed)*2
6670 >= mddev->resync_max - mddev->curr_resync_completed
6671 )) {
6672 /* time to update curr_resync_completed */
6673 blk_unplug(mddev->queue);
6674 wait_event(mddev->recovery_wait,
6675 atomic_read(&mddev->recovery_active) == 0);
6676 mddev->curr_resync_completed =
6677 mddev->curr_resync;
6678 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6679 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6680 }
6681
6682 while (j >= mddev->resync_max && !kthread_should_stop()) {
6683 /* As this condition is controlled by user-space,
6684 * we can block indefinitely, so use '_interruptible'
6685 * to avoid triggering warnings.
6686 */
6687 flush_signals(current); /* just in case */
6688 wait_event_interruptible(mddev->recovery_wait,
6689 mddev->resync_max > j
6690 || kthread_should_stop());
6691 }
6692
6693 if (kthread_should_stop())
6694 goto interrupted;
6695
6696 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6697 currspeed < speed_min(mddev));
6698 if (sectors == 0) {
6699 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6700 goto out;
6701 }
6702
6703 if (!skipped) { /* actual IO requested */
6704 io_sectors += sectors;
6705 atomic_add(sectors, &mddev->recovery_active);
6706 }
6707
6708 j += sectors;
6709 if (j>1) mddev->curr_resync = j;
6710 mddev->curr_mark_cnt = io_sectors;
6711 if (last_check == 0)
6712 /* this is the earliers that rebuilt will be
6713 * visible in /proc/mdstat
6714 */
6715 md_new_event(mddev);
6716
6717 if (last_check + window > io_sectors || j == max_sectors)
6718 continue;
6719
6720 last_check = io_sectors;
6721
6722 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6723 break;
6724
6725 repeat:
6726 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6727 /* step marks */
6728 int next = (last_mark+1) % SYNC_MARKS;
6729
6730 mddev->resync_mark = mark[next];
6731 mddev->resync_mark_cnt = mark_cnt[next];
6732 mark[next] = jiffies;
6733 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6734 last_mark = next;
6735 }
6736
6737
6738 if (kthread_should_stop())
6739 goto interrupted;
6740
6741
6742 /*
6743 * this loop exits only if either when we are slower than
6744 * the 'hard' speed limit, or the system was IO-idle for
6745 * a jiffy.
6746 * the system might be non-idle CPU-wise, but we only care
6747 * about not overloading the IO subsystem. (things like an
6748 * e2fsck being done on the RAID array should execute fast)
6749 */
6750 blk_unplug(mddev->queue);
6751 cond_resched();
6752
6753 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6754 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6755
6756 if (currspeed > speed_min(mddev)) {
6757 if ((currspeed > speed_max(mddev)) ||
6758 !is_mddev_idle(mddev, 0)) {
6759 msleep(500);
6760 goto repeat;
6761 }
6762 }
6763 }
6764 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6765 /*
6766 * this also signals 'finished resyncing' to md_stop
6767 */
6768 out:
6769 blk_unplug(mddev->queue);
6770
6771 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6772
6773 /* tell personality that we are finished */
6774 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6775
6776 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6777 mddev->curr_resync > 2) {
6778 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6779 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6780 if (mddev->curr_resync >= mddev->recovery_cp) {
6781 printk(KERN_INFO
6782 "md: checkpointing %s of %s.\n",
6783 desc, mdname(mddev));
6784 mddev->recovery_cp = mddev->curr_resync;
6785 }
6786 } else
6787 mddev->recovery_cp = MaxSector;
6788 } else {
6789 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6790 mddev->curr_resync = MaxSector;
6791 rcu_read_lock();
6792 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6793 if (rdev->raid_disk >= 0 &&
6794 !test_bit(Faulty, &rdev->flags) &&
6795 !test_bit(In_sync, &rdev->flags) &&
6796 rdev->recovery_offset < mddev->curr_resync)
6797 rdev->recovery_offset = mddev->curr_resync;
6798 rcu_read_unlock();
6799 }
6800 }
6801 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6802
6803 skip:
6804 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6805 /* We completed so min/max setting can be forgotten if used. */
6806 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6807 mddev->resync_min = 0;
6808 mddev->resync_max = MaxSector;
6809 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6810 mddev->resync_min = mddev->curr_resync_completed;
6811 mddev->curr_resync = 0;
6812 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6813 mddev->curr_resync_completed = 0;
6814 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6815 wake_up(&resync_wait);
6816 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6817 md_wakeup_thread(mddev->thread);
6818 return;
6819
6820 interrupted:
6821 /*
6822 * got a signal, exit.
6823 */
6824 printk(KERN_INFO
6825 "md: md_do_sync() got signal ... exiting\n");
6826 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6827 goto out;
6828
6829 }
6830 EXPORT_SYMBOL_GPL(md_do_sync);
6831
6832
6833 static int remove_and_add_spares(mddev_t *mddev)
6834 {
6835 mdk_rdev_t *rdev;
6836 int spares = 0;
6837
6838 mddev->curr_resync_completed = 0;
6839
6840 list_for_each_entry(rdev, &mddev->disks, same_set)
6841 if (rdev->raid_disk >= 0 &&
6842 !test_bit(Blocked, &rdev->flags) &&
6843 (test_bit(Faulty, &rdev->flags) ||
6844 ! test_bit(In_sync, &rdev->flags)) &&
6845 atomic_read(&rdev->nr_pending)==0) {
6846 if (mddev->pers->hot_remove_disk(
6847 mddev, rdev->raid_disk)==0) {
6848 char nm[20];
6849 sprintf(nm,"rd%d", rdev->raid_disk);
6850 sysfs_remove_link(&mddev->kobj, nm);
6851 rdev->raid_disk = -1;
6852 }
6853 }
6854
6855 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6856 list_for_each_entry(rdev, &mddev->disks, same_set) {
6857 if (rdev->raid_disk >= 0 &&
6858 !test_bit(In_sync, &rdev->flags) &&
6859 !test_bit(Blocked, &rdev->flags))
6860 spares++;
6861 if (rdev->raid_disk < 0
6862 && !test_bit(Faulty, &rdev->flags)) {
6863 rdev->recovery_offset = 0;
6864 if (mddev->pers->
6865 hot_add_disk(mddev, rdev) == 0) {
6866 char nm[20];
6867 sprintf(nm, "rd%d", rdev->raid_disk);
6868 if (sysfs_create_link(&mddev->kobj,
6869 &rdev->kobj, nm))
6870 printk(KERN_WARNING
6871 "md: cannot register "
6872 "%s for %s\n",
6873 nm, mdname(mddev));
6874 spares++;
6875 md_new_event(mddev);
6876 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6877 } else
6878 break;
6879 }
6880 }
6881 }
6882 return spares;
6883 }
6884 /*
6885 * This routine is regularly called by all per-raid-array threads to
6886 * deal with generic issues like resync and super-block update.
6887 * Raid personalities that don't have a thread (linear/raid0) do not
6888 * need this as they never do any recovery or update the superblock.
6889 *
6890 * It does not do any resync itself, but rather "forks" off other threads
6891 * to do that as needed.
6892 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6893 * "->recovery" and create a thread at ->sync_thread.
6894 * When the thread finishes it sets MD_RECOVERY_DONE
6895 * and wakeups up this thread which will reap the thread and finish up.
6896 * This thread also removes any faulty devices (with nr_pending == 0).
6897 *
6898 * The overall approach is:
6899 * 1/ if the superblock needs updating, update it.
6900 * 2/ If a recovery thread is running, don't do anything else.
6901 * 3/ If recovery has finished, clean up, possibly marking spares active.
6902 * 4/ If there are any faulty devices, remove them.
6903 * 5/ If array is degraded, try to add spares devices
6904 * 6/ If array has spares or is not in-sync, start a resync thread.
6905 */
6906 void md_check_recovery(mddev_t *mddev)
6907 {
6908 mdk_rdev_t *rdev;
6909
6910
6911 if (mddev->bitmap)
6912 bitmap_daemon_work(mddev);
6913
6914 if (mddev->ro)
6915 return;
6916
6917 if (signal_pending(current)) {
6918 if (mddev->pers->sync_request && !mddev->external) {
6919 printk(KERN_INFO "md: %s in immediate safe mode\n",
6920 mdname(mddev));
6921 mddev->safemode = 2;
6922 }
6923 flush_signals(current);
6924 }
6925
6926 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6927 return;
6928 if ( ! (
6929 (mddev->flags && !mddev->external) ||
6930 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6931 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6932 (mddev->external == 0 && mddev->safemode == 1) ||
6933 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6934 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6935 ))
6936 return;
6937
6938 if (mddev_trylock(mddev)) {
6939 int spares = 0;
6940
6941 if (mddev->ro) {
6942 /* Only thing we do on a ro array is remove
6943 * failed devices.
6944 */
6945 remove_and_add_spares(mddev);
6946 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6947 goto unlock;
6948 }
6949
6950 if (!mddev->external) {
6951 int did_change = 0;
6952 spin_lock_irq(&mddev->write_lock);
6953 if (mddev->safemode &&
6954 !atomic_read(&mddev->writes_pending) &&
6955 !mddev->in_sync &&
6956 mddev->recovery_cp == MaxSector) {
6957 mddev->in_sync = 1;
6958 did_change = 1;
6959 if (mddev->persistent)
6960 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6961 }
6962 if (mddev->safemode == 1)
6963 mddev->safemode = 0;
6964 spin_unlock_irq(&mddev->write_lock);
6965 if (did_change)
6966 sysfs_notify_dirent(mddev->sysfs_state);
6967 }
6968
6969 if (mddev->flags)
6970 md_update_sb(mddev, 0);
6971
6972 list_for_each_entry(rdev, &mddev->disks, same_set)
6973 if (test_and_clear_bit(StateChanged, &rdev->flags))
6974 sysfs_notify_dirent(rdev->sysfs_state);
6975
6976
6977 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6978 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6979 /* resync/recovery still happening */
6980 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6981 goto unlock;
6982 }
6983 if (mddev->sync_thread) {
6984 /* resync has finished, collect result */
6985 md_unregister_thread(mddev->sync_thread);
6986 mddev->sync_thread = NULL;
6987 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6988 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6989 /* success...*/
6990 /* activate any spares */
6991 if (mddev->pers->spare_active(mddev))
6992 sysfs_notify(&mddev->kobj, NULL,
6993 "degraded");
6994 }
6995 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6996 mddev->pers->finish_reshape)
6997 mddev->pers->finish_reshape(mddev);
6998 md_update_sb(mddev, 1);
6999
7000 /* if array is no-longer degraded, then any saved_raid_disk
7001 * information must be scrapped
7002 */
7003 if (!mddev->degraded)
7004 list_for_each_entry(rdev, &mddev->disks, same_set)
7005 rdev->saved_raid_disk = -1;
7006
7007 mddev->recovery = 0;
7008 /* flag recovery needed just to double check */
7009 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7010 sysfs_notify_dirent(mddev->sysfs_action);
7011 md_new_event(mddev);
7012 goto unlock;
7013 }
7014 /* Set RUNNING before clearing NEEDED to avoid
7015 * any transients in the value of "sync_action".
7016 */
7017 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7018 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7019 /* Clear some bits that don't mean anything, but
7020 * might be left set
7021 */
7022 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7023 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7024
7025 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7026 goto unlock;
7027 /* no recovery is running.
7028 * remove any failed drives, then
7029 * add spares if possible.
7030 * Spare are also removed and re-added, to allow
7031 * the personality to fail the re-add.
7032 */
7033
7034 if (mddev->reshape_position != MaxSector) {
7035 if (mddev->pers->check_reshape == NULL ||
7036 mddev->pers->check_reshape(mddev) != 0)
7037 /* Cannot proceed */
7038 goto unlock;
7039 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7040 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7041 } else if ((spares = remove_and_add_spares(mddev))) {
7042 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7043 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7044 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7045 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7046 } else if (mddev->recovery_cp < MaxSector) {
7047 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7048 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7049 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7050 /* nothing to be done ... */
7051 goto unlock;
7052
7053 if (mddev->pers->sync_request) {
7054 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7055 /* We are adding a device or devices to an array
7056 * which has the bitmap stored on all devices.
7057 * So make sure all bitmap pages get written
7058 */
7059 bitmap_write_all(mddev->bitmap);
7060 }
7061 mddev->sync_thread = md_register_thread(md_do_sync,
7062 mddev,
7063 "resync");
7064 if (!mddev->sync_thread) {
7065 printk(KERN_ERR "%s: could not start resync"
7066 " thread...\n",
7067 mdname(mddev));
7068 /* leave the spares where they are, it shouldn't hurt */
7069 mddev->recovery = 0;
7070 } else
7071 md_wakeup_thread(mddev->sync_thread);
7072 sysfs_notify_dirent(mddev->sysfs_action);
7073 md_new_event(mddev);
7074 }
7075 unlock:
7076 if (!mddev->sync_thread) {
7077 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7078 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7079 &mddev->recovery))
7080 if (mddev->sysfs_action)
7081 sysfs_notify_dirent(mddev->sysfs_action);
7082 }
7083 mddev_unlock(mddev);
7084 }
7085 }
7086
7087 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7088 {
7089 sysfs_notify_dirent(rdev->sysfs_state);
7090 wait_event_timeout(rdev->blocked_wait,
7091 !test_bit(Blocked, &rdev->flags),
7092 msecs_to_jiffies(5000));
7093 rdev_dec_pending(rdev, mddev);
7094 }
7095 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7096
7097 static int md_notify_reboot(struct notifier_block *this,
7098 unsigned long code, void *x)
7099 {
7100 struct list_head *tmp;
7101 mddev_t *mddev;
7102
7103 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7104
7105 printk(KERN_INFO "md: stopping all md devices.\n");
7106
7107 for_each_mddev(mddev, tmp)
7108 if (mddev_trylock(mddev)) {
7109 /* Force a switch to readonly even array
7110 * appears to still be in use. Hence
7111 * the '100'.
7112 */
7113 do_md_stop(mddev, 1, 100);
7114 mddev_unlock(mddev);
7115 }
7116 /*
7117 * certain more exotic SCSI devices are known to be
7118 * volatile wrt too early system reboots. While the
7119 * right place to handle this issue is the given
7120 * driver, we do want to have a safe RAID driver ...
7121 */
7122 mdelay(1000*1);
7123 }
7124 return NOTIFY_DONE;
7125 }
7126
7127 static struct notifier_block md_notifier = {
7128 .notifier_call = md_notify_reboot,
7129 .next = NULL,
7130 .priority = INT_MAX, /* before any real devices */
7131 };
7132
7133 static void md_geninit(void)
7134 {
7135 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7136
7137 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7138 }
7139
7140 static int __init md_init(void)
7141 {
7142 if (register_blkdev(MD_MAJOR, "md"))
7143 return -1;
7144 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7145 unregister_blkdev(MD_MAJOR, "md");
7146 return -1;
7147 }
7148 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7149 md_probe, NULL, NULL);
7150 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7151 md_probe, NULL, NULL);
7152
7153 register_reboot_notifier(&md_notifier);
7154 raid_table_header = register_sysctl_table(raid_root_table);
7155
7156 md_geninit();
7157 return 0;
7158 }
7159
7160
7161 #ifndef MODULE
7162
7163 /*
7164 * Searches all registered partitions for autorun RAID arrays
7165 * at boot time.
7166 */
7167
7168 static LIST_HEAD(all_detected_devices);
7169 struct detected_devices_node {
7170 struct list_head list;
7171 dev_t dev;
7172 };
7173
7174 void md_autodetect_dev(dev_t dev)
7175 {
7176 struct detected_devices_node *node_detected_dev;
7177
7178 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7179 if (node_detected_dev) {
7180 node_detected_dev->dev = dev;
7181 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7182 } else {
7183 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7184 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7185 }
7186 }
7187
7188
7189 static void autostart_arrays(int part)
7190 {
7191 mdk_rdev_t *rdev;
7192 struct detected_devices_node *node_detected_dev;
7193 dev_t dev;
7194 int i_scanned, i_passed;
7195
7196 i_scanned = 0;
7197 i_passed = 0;
7198
7199 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7200
7201 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7202 i_scanned++;
7203 node_detected_dev = list_entry(all_detected_devices.next,
7204 struct detected_devices_node, list);
7205 list_del(&node_detected_dev->list);
7206 dev = node_detected_dev->dev;
7207 kfree(node_detected_dev);
7208 rdev = md_import_device(dev,0, 90);
7209 if (IS_ERR(rdev))
7210 continue;
7211
7212 if (test_bit(Faulty, &rdev->flags)) {
7213 MD_BUG();
7214 continue;
7215 }
7216 set_bit(AutoDetected, &rdev->flags);
7217 list_add(&rdev->same_set, &pending_raid_disks);
7218 i_passed++;
7219 }
7220
7221 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7222 i_scanned, i_passed);
7223
7224 autorun_devices(part);
7225 }
7226
7227 #endif /* !MODULE */
7228
7229 static __exit void md_exit(void)
7230 {
7231 mddev_t *mddev;
7232 struct list_head *tmp;
7233
7234 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7235 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7236
7237 unregister_blkdev(MD_MAJOR,"md");
7238 unregister_blkdev(mdp_major, "mdp");
7239 unregister_reboot_notifier(&md_notifier);
7240 unregister_sysctl_table(raid_table_header);
7241 remove_proc_entry("mdstat", NULL);
7242 for_each_mddev(mddev, tmp) {
7243 export_array(mddev);
7244 mddev->hold_active = 0;
7245 }
7246 }
7247
7248 subsys_initcall(md_init);
7249 module_exit(md_exit)
7250
7251 static int get_ro(char *buffer, struct kernel_param *kp)
7252 {
7253 return sprintf(buffer, "%d", start_readonly);
7254 }
7255 static int set_ro(const char *val, struct kernel_param *kp)
7256 {
7257 char *e;
7258 int num = simple_strtoul(val, &e, 10);
7259 if (*val && (*e == '\0' || *e == '\n')) {
7260 start_readonly = num;
7261 return 0;
7262 }
7263 return -EINVAL;
7264 }
7265
7266 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7267 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7268
7269 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7270
7271 EXPORT_SYMBOL(register_md_personality);
7272 EXPORT_SYMBOL(unregister_md_personality);
7273 EXPORT_SYMBOL(md_error);
7274 EXPORT_SYMBOL(md_done_sync);
7275 EXPORT_SYMBOL(md_write_start);
7276 EXPORT_SYMBOL(md_write_end);
7277 EXPORT_SYMBOL(md_register_thread);
7278 EXPORT_SYMBOL(md_unregister_thread);
7279 EXPORT_SYMBOL(md_wakeup_thread);
7280 EXPORT_SYMBOL(md_check_recovery);
7281 MODULE_LICENSE("GPL");
7282 MODULE_DESCRIPTION("MD RAID framework");
7283 MODULE_ALIAS("md");
7284 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree