]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - drivers/md/md.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
block: push down BKL into .open and .release
[mirror_ubuntu-zesty-kernel.git] / drivers / md / md.c
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/smp_lock.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
54 #include "md.h"
55 #include "bitmap.h"
56
57 #define DEBUG 0
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
59
60
61 #ifndef MODULE
62 static void autostart_arrays(int part);
63 #endif
64
65 static LIST_HEAD(pers_list);
66 static DEFINE_SPINLOCK(pers_lock);
67
68 static void md_print_devices(void);
69
70 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
71
72 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
73
74 /*
75 * Default number of read corrections we'll attempt on an rdev
76 * before ejecting it from the array. We divide the read error
77 * count by 2 for every hour elapsed between read errors.
78 */
79 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
80 /*
81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
82 * is 1000 KB/sec, so the extra system load does not show up that much.
83 * Increase it if you want to have more _guaranteed_ speed. Note that
84 * the RAID driver will use the maximum available bandwidth if the IO
85 * subsystem is idle. There is also an 'absolute maximum' reconstruction
86 * speed limit - in case reconstruction slows down your system despite
87 * idle IO detection.
88 *
89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
90 * or /sys/block/mdX/md/sync_speed_{min,max}
91 */
92
93 static int sysctl_speed_limit_min = 1000;
94 static int sysctl_speed_limit_max = 200000;
95 static inline int speed_min(mddev_t *mddev)
96 {
97 return mddev->sync_speed_min ?
98 mddev->sync_speed_min : sysctl_speed_limit_min;
99 }
100
101 static inline int speed_max(mddev_t *mddev)
102 {
103 return mddev->sync_speed_max ?
104 mddev->sync_speed_max : sysctl_speed_limit_max;
105 }
106
107 static struct ctl_table_header *raid_table_header;
108
109 static ctl_table raid_table[] = {
110 {
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = proc_dointvec,
116 },
117 {
118 .procname = "speed_limit_max",
119 .data = &sysctl_speed_limit_max,
120 .maxlen = sizeof(int),
121 .mode = S_IRUGO|S_IWUSR,
122 .proc_handler = proc_dointvec,
123 },
124 { }
125 };
126
127 static ctl_table raid_dir_table[] = {
128 {
129 .procname = "raid",
130 .maxlen = 0,
131 .mode = S_IRUGO|S_IXUGO,
132 .child = raid_table,
133 },
134 { }
135 };
136
137 static ctl_table raid_root_table[] = {
138 {
139 .procname = "dev",
140 .maxlen = 0,
141 .mode = 0555,
142 .child = raid_dir_table,
143 },
144 { }
145 };
146
147 static const struct block_device_operations md_fops;
148
149 static int start_readonly;
150
151 /*
152 * We have a system wide 'event count' that is incremented
153 * on any 'interesting' event, and readers of /proc/mdstat
154 * can use 'poll' or 'select' to find out when the event
155 * count increases.
156 *
157 * Events are:
158 * start array, stop array, error, add device, remove device,
159 * start build, activate spare
160 */
161 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
162 static atomic_t md_event_count;
163 void md_new_event(mddev_t *mddev)
164 {
165 atomic_inc(&md_event_count);
166 wake_up(&md_event_waiters);
167 }
168 EXPORT_SYMBOL_GPL(md_new_event);
169
170 /* Alternate version that can be called from interrupts
171 * when calling sysfs_notify isn't needed.
172 */
173 static void md_new_event_inintr(mddev_t *mddev)
174 {
175 atomic_inc(&md_event_count);
176 wake_up(&md_event_waiters);
177 }
178
179 /*
180 * Enables to iterate over all existing md arrays
181 * all_mddevs_lock protects this list.
182 */
183 static LIST_HEAD(all_mddevs);
184 static DEFINE_SPINLOCK(all_mddevs_lock);
185
186
187 /*
188 * iterates through all used mddevs in the system.
189 * We take care to grab the all_mddevs_lock whenever navigating
190 * the list, and to always hold a refcount when unlocked.
191 * Any code which breaks out of this loop while own
192 * a reference to the current mddev and must mddev_put it.
193 */
194 #define for_each_mddev(mddev,tmp) \
195 \
196 for (({ spin_lock(&all_mddevs_lock); \
197 tmp = all_mddevs.next; \
198 mddev = NULL;}); \
199 ({ if (tmp != &all_mddevs) \
200 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
201 spin_unlock(&all_mddevs_lock); \
202 if (mddev) mddev_put(mddev); \
203 mddev = list_entry(tmp, mddev_t, all_mddevs); \
204 tmp != &all_mddevs;}); \
205 ({ spin_lock(&all_mddevs_lock); \
206 tmp = tmp->next;}) \
207 )
208
209
210 /* Rather than calling directly into the personality make_request function,
211 * IO requests come here first so that we can check if the device is
212 * being suspended pending a reconfiguration.
213 * We hold a refcount over the call to ->make_request. By the time that
214 * call has finished, the bio has been linked into some internal structure
215 * and so is visible to ->quiesce(), so we don't need the refcount any more.
216 */
217 static int md_make_request(struct request_queue *q, struct bio *bio)
218 {
219 const int rw = bio_data_dir(bio);
220 mddev_t *mddev = q->queuedata;
221 int rv;
222 int cpu;
223
224 if (mddev == NULL || mddev->pers == NULL) {
225 bio_io_error(bio);
226 return 0;
227 }
228 rcu_read_lock();
229 if (mddev->suspended || mddev->barrier) {
230 DEFINE_WAIT(__wait);
231 for (;;) {
232 prepare_to_wait(&mddev->sb_wait, &__wait,
233 TASK_UNINTERRUPTIBLE);
234 if (!mddev->suspended && !mddev->barrier)
235 break;
236 rcu_read_unlock();
237 schedule();
238 rcu_read_lock();
239 }
240 finish_wait(&mddev->sb_wait, &__wait);
241 }
242 atomic_inc(&mddev->active_io);
243 rcu_read_unlock();
244
245 rv = mddev->pers->make_request(mddev, bio);
246
247 cpu = part_stat_lock();
248 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
249 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
250 bio_sectors(bio));
251 part_stat_unlock();
252
253 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
254 wake_up(&mddev->sb_wait);
255
256 return rv;
257 }
258
259 /* mddev_suspend makes sure no new requests are submitted
260 * to the device, and that any requests that have been submitted
261 * are completely handled.
262 * Once ->stop is called and completes, the module will be completely
263 * unused.
264 */
265 static void mddev_suspend(mddev_t *mddev)
266 {
267 BUG_ON(mddev->suspended);
268 mddev->suspended = 1;
269 synchronize_rcu();
270 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
271 mddev->pers->quiesce(mddev, 1);
272 }
273
274 static void mddev_resume(mddev_t *mddev)
275 {
276 mddev->suspended = 0;
277 wake_up(&mddev->sb_wait);
278 mddev->pers->quiesce(mddev, 0);
279 }
280
281 int mddev_congested(mddev_t *mddev, int bits)
282 {
283 if (mddev->barrier)
284 return 1;
285 return mddev->suspended;
286 }
287 EXPORT_SYMBOL(mddev_congested);
288
289 /*
290 * Generic barrier handling for md
291 */
292
293 #define POST_REQUEST_BARRIER ((void*)1)
294
295 static void md_end_barrier(struct bio *bio, int err)
296 {
297 mdk_rdev_t *rdev = bio->bi_private;
298 mddev_t *mddev = rdev->mddev;
299 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
300 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
301
302 rdev_dec_pending(rdev, mddev);
303
304 if (atomic_dec_and_test(&mddev->flush_pending)) {
305 if (mddev->barrier == POST_REQUEST_BARRIER) {
306 /* This was a post-request barrier */
307 mddev->barrier = NULL;
308 wake_up(&mddev->sb_wait);
309 } else
310 /* The pre-request barrier has finished */
311 schedule_work(&mddev->barrier_work);
312 }
313 bio_put(bio);
314 }
315
316 static void submit_barriers(mddev_t *mddev)
317 {
318 mdk_rdev_t *rdev;
319
320 rcu_read_lock();
321 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
322 if (rdev->raid_disk >= 0 &&
323 !test_bit(Faulty, &rdev->flags)) {
324 /* Take two references, one is dropped
325 * when request finishes, one after
326 * we reclaim rcu_read_lock
327 */
328 struct bio *bi;
329 atomic_inc(&rdev->nr_pending);
330 atomic_inc(&rdev->nr_pending);
331 rcu_read_unlock();
332 bi = bio_alloc(GFP_KERNEL, 0);
333 bi->bi_end_io = md_end_barrier;
334 bi->bi_private = rdev;
335 bi->bi_bdev = rdev->bdev;
336 atomic_inc(&mddev->flush_pending);
337 submit_bio(WRITE_BARRIER, bi);
338 rcu_read_lock();
339 rdev_dec_pending(rdev, mddev);
340 }
341 rcu_read_unlock();
342 }
343
344 static void md_submit_barrier(struct work_struct *ws)
345 {
346 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
347 struct bio *bio = mddev->barrier;
348
349 atomic_set(&mddev->flush_pending, 1);
350
351 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
352 bio_endio(bio, -EOPNOTSUPP);
353 else if (bio->bi_size == 0)
354 /* an empty barrier - all done */
355 bio_endio(bio, 0);
356 else {
357 bio->bi_rw &= ~REQ_HARDBARRIER;
358 if (mddev->pers->make_request(mddev, bio))
359 generic_make_request(bio);
360 mddev->barrier = POST_REQUEST_BARRIER;
361 submit_barriers(mddev);
362 }
363 if (atomic_dec_and_test(&mddev->flush_pending)) {
364 mddev->barrier = NULL;
365 wake_up(&mddev->sb_wait);
366 }
367 }
368
369 void md_barrier_request(mddev_t *mddev, struct bio *bio)
370 {
371 spin_lock_irq(&mddev->write_lock);
372 wait_event_lock_irq(mddev->sb_wait,
373 !mddev->barrier,
374 mddev->write_lock, /*nothing*/);
375 mddev->barrier = bio;
376 spin_unlock_irq(&mddev->write_lock);
377
378 atomic_set(&mddev->flush_pending, 1);
379 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
380
381 submit_barriers(mddev);
382
383 if (atomic_dec_and_test(&mddev->flush_pending))
384 schedule_work(&mddev->barrier_work);
385 }
386 EXPORT_SYMBOL(md_barrier_request);
387
388 static inline mddev_t *mddev_get(mddev_t *mddev)
389 {
390 atomic_inc(&mddev->active);
391 return mddev;
392 }
393
394 static void mddev_delayed_delete(struct work_struct *ws);
395
396 static void mddev_put(mddev_t *mddev)
397 {
398 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
399 return;
400 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
401 mddev->ctime == 0 && !mddev->hold_active) {
402 /* Array is not configured at all, and not held active,
403 * so destroy it */
404 list_del(&mddev->all_mddevs);
405 if (mddev->gendisk) {
406 /* we did a probe so need to clean up.
407 * Call schedule_work inside the spinlock
408 * so that flush_scheduled_work() after
409 * mddev_find will succeed in waiting for the
410 * work to be done.
411 */
412 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
413 schedule_work(&mddev->del_work);
414 } else
415 kfree(mddev);
416 }
417 spin_unlock(&all_mddevs_lock);
418 }
419
420 static void mddev_init(mddev_t *mddev)
421 {
422 mutex_init(&mddev->open_mutex);
423 mutex_init(&mddev->reconfig_mutex);
424 mutex_init(&mddev->bitmap_info.mutex);
425 INIT_LIST_HEAD(&mddev->disks);
426 INIT_LIST_HEAD(&mddev->all_mddevs);
427 init_timer(&mddev->safemode_timer);
428 atomic_set(&mddev->active, 1);
429 atomic_set(&mddev->openers, 0);
430 atomic_set(&mddev->active_io, 0);
431 spin_lock_init(&mddev->write_lock);
432 atomic_set(&mddev->flush_pending, 0);
433 init_waitqueue_head(&mddev->sb_wait);
434 init_waitqueue_head(&mddev->recovery_wait);
435 mddev->reshape_position = MaxSector;
436 mddev->resync_min = 0;
437 mddev->resync_max = MaxSector;
438 mddev->level = LEVEL_NONE;
439 }
440
441 static mddev_t * mddev_find(dev_t unit)
442 {
443 mddev_t *mddev, *new = NULL;
444
445 retry:
446 spin_lock(&all_mddevs_lock);
447
448 if (unit) {
449 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
450 if (mddev->unit == unit) {
451 mddev_get(mddev);
452 spin_unlock(&all_mddevs_lock);
453 kfree(new);
454 return mddev;
455 }
456
457 if (new) {
458 list_add(&new->all_mddevs, &all_mddevs);
459 spin_unlock(&all_mddevs_lock);
460 new->hold_active = UNTIL_IOCTL;
461 return new;
462 }
463 } else if (new) {
464 /* find an unused unit number */
465 static int next_minor = 512;
466 int start = next_minor;
467 int is_free = 0;
468 int dev = 0;
469 while (!is_free) {
470 dev = MKDEV(MD_MAJOR, next_minor);
471 next_minor++;
472 if (next_minor > MINORMASK)
473 next_minor = 0;
474 if (next_minor == start) {
475 /* Oh dear, all in use. */
476 spin_unlock(&all_mddevs_lock);
477 kfree(new);
478 return NULL;
479 }
480
481 is_free = 1;
482 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
483 if (mddev->unit == dev) {
484 is_free = 0;
485 break;
486 }
487 }
488 new->unit = dev;
489 new->md_minor = MINOR(dev);
490 new->hold_active = UNTIL_STOP;
491 list_add(&new->all_mddevs, &all_mddevs);
492 spin_unlock(&all_mddevs_lock);
493 return new;
494 }
495 spin_unlock(&all_mddevs_lock);
496
497 new = kzalloc(sizeof(*new), GFP_KERNEL);
498 if (!new)
499 return NULL;
500
501 new->unit = unit;
502 if (MAJOR(unit) == MD_MAJOR)
503 new->md_minor = MINOR(unit);
504 else
505 new->md_minor = MINOR(unit) >> MdpMinorShift;
506
507 mddev_init(new);
508
509 goto retry;
510 }
511
512 static inline int mddev_lock(mddev_t * mddev)
513 {
514 return mutex_lock_interruptible(&mddev->reconfig_mutex);
515 }
516
517 static inline int mddev_is_locked(mddev_t *mddev)
518 {
519 return mutex_is_locked(&mddev->reconfig_mutex);
520 }
521
522 static inline int mddev_trylock(mddev_t * mddev)
523 {
524 return mutex_trylock(&mddev->reconfig_mutex);
525 }
526
527 static struct attribute_group md_redundancy_group;
528
529 static void mddev_unlock(mddev_t * mddev)
530 {
531 if (mddev->to_remove) {
532 /* These cannot be removed under reconfig_mutex as
533 * an access to the files will try to take reconfig_mutex
534 * while holding the file unremovable, which leads to
535 * a deadlock.
536 * So hold open_mutex instead - we are allowed to take
537 * it while holding reconfig_mutex, and md_run can
538 * use it to wait for the remove to complete.
539 */
540 struct attribute_group *to_remove = mddev->to_remove;
541 mddev->to_remove = NULL;
542 mutex_lock(&mddev->open_mutex);
543 mutex_unlock(&mddev->reconfig_mutex);
544
545 if (to_remove != &md_redundancy_group)
546 sysfs_remove_group(&mddev->kobj, to_remove);
547 if (mddev->pers == NULL ||
548 mddev->pers->sync_request == NULL) {
549 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
550 if (mddev->sysfs_action)
551 sysfs_put(mddev->sysfs_action);
552 mddev->sysfs_action = NULL;
553 }
554 mutex_unlock(&mddev->open_mutex);
555 } else
556 mutex_unlock(&mddev->reconfig_mutex);
557
558 md_wakeup_thread(mddev->thread);
559 }
560
561 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
562 {
563 mdk_rdev_t *rdev;
564
565 list_for_each_entry(rdev, &mddev->disks, same_set)
566 if (rdev->desc_nr == nr)
567 return rdev;
568
569 return NULL;
570 }
571
572 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
573 {
574 mdk_rdev_t *rdev;
575
576 list_for_each_entry(rdev, &mddev->disks, same_set)
577 if (rdev->bdev->bd_dev == dev)
578 return rdev;
579
580 return NULL;
581 }
582
583 static struct mdk_personality *find_pers(int level, char *clevel)
584 {
585 struct mdk_personality *pers;
586 list_for_each_entry(pers, &pers_list, list) {
587 if (level != LEVEL_NONE && pers->level == level)
588 return pers;
589 if (strcmp(pers->name, clevel)==0)
590 return pers;
591 }
592 return NULL;
593 }
594
595 /* return the offset of the super block in 512byte sectors */
596 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
597 {
598 sector_t num_sectors = bdev->bd_inode->i_size / 512;
599 return MD_NEW_SIZE_SECTORS(num_sectors);
600 }
601
602 static int alloc_disk_sb(mdk_rdev_t * rdev)
603 {
604 if (rdev->sb_page)
605 MD_BUG();
606
607 rdev->sb_page = alloc_page(GFP_KERNEL);
608 if (!rdev->sb_page) {
609 printk(KERN_ALERT "md: out of memory.\n");
610 return -ENOMEM;
611 }
612
613 return 0;
614 }
615
616 static void free_disk_sb(mdk_rdev_t * rdev)
617 {
618 if (rdev->sb_page) {
619 put_page(rdev->sb_page);
620 rdev->sb_loaded = 0;
621 rdev->sb_page = NULL;
622 rdev->sb_start = 0;
623 rdev->sectors = 0;
624 }
625 }
626
627
628 static void super_written(struct bio *bio, int error)
629 {
630 mdk_rdev_t *rdev = bio->bi_private;
631 mddev_t *mddev = rdev->mddev;
632
633 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
634 printk("md: super_written gets error=%d, uptodate=%d\n",
635 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
636 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
637 md_error(mddev, rdev);
638 }
639
640 if (atomic_dec_and_test(&mddev->pending_writes))
641 wake_up(&mddev->sb_wait);
642 bio_put(bio);
643 }
644
645 static void super_written_barrier(struct bio *bio, int error)
646 {
647 struct bio *bio2 = bio->bi_private;
648 mdk_rdev_t *rdev = bio2->bi_private;
649 mddev_t *mddev = rdev->mddev;
650
651 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
652 error == -EOPNOTSUPP) {
653 unsigned long flags;
654 /* barriers don't appear to be supported :-( */
655 set_bit(BarriersNotsupp, &rdev->flags);
656 mddev->barriers_work = 0;
657 spin_lock_irqsave(&mddev->write_lock, flags);
658 bio2->bi_next = mddev->biolist;
659 mddev->biolist = bio2;
660 spin_unlock_irqrestore(&mddev->write_lock, flags);
661 wake_up(&mddev->sb_wait);
662 bio_put(bio);
663 } else {
664 bio_put(bio2);
665 bio->bi_private = rdev;
666 super_written(bio, error);
667 }
668 }
669
670 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
671 sector_t sector, int size, struct page *page)
672 {
673 /* write first size bytes of page to sector of rdev
674 * Increment mddev->pending_writes before returning
675 * and decrement it on completion, waking up sb_wait
676 * if zero is reached.
677 * If an error occurred, call md_error
678 *
679 * As we might need to resubmit the request if REQ_HARDBARRIER
680 * causes ENOTSUPP, we allocate a spare bio...
681 */
682 struct bio *bio = bio_alloc(GFP_NOIO, 1);
683 int rw = REQ_WRITE | REQ_SYNC | REQ_UNPLUG;
684
685 bio->bi_bdev = rdev->bdev;
686 bio->bi_sector = sector;
687 bio_add_page(bio, page, size, 0);
688 bio->bi_private = rdev;
689 bio->bi_end_io = super_written;
690 bio->bi_rw = rw;
691
692 atomic_inc(&mddev->pending_writes);
693 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
694 struct bio *rbio;
695 rw |= REQ_HARDBARRIER;
696 rbio = bio_clone(bio, GFP_NOIO);
697 rbio->bi_private = bio;
698 rbio->bi_end_io = super_written_barrier;
699 submit_bio(rw, rbio);
700 } else
701 submit_bio(rw, bio);
702 }
703
704 void md_super_wait(mddev_t *mddev)
705 {
706 /* wait for all superblock writes that were scheduled to complete.
707 * if any had to be retried (due to BARRIER problems), retry them
708 */
709 DEFINE_WAIT(wq);
710 for(;;) {
711 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
712 if (atomic_read(&mddev->pending_writes)==0)
713 break;
714 while (mddev->biolist) {
715 struct bio *bio;
716 spin_lock_irq(&mddev->write_lock);
717 bio = mddev->biolist;
718 mddev->biolist = bio->bi_next ;
719 bio->bi_next = NULL;
720 spin_unlock_irq(&mddev->write_lock);
721 submit_bio(bio->bi_rw, bio);
722 }
723 schedule();
724 }
725 finish_wait(&mddev->sb_wait, &wq);
726 }
727
728 static void bi_complete(struct bio *bio, int error)
729 {
730 complete((struct completion*)bio->bi_private);
731 }
732
733 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
734 struct page *page, int rw)
735 {
736 struct bio *bio = bio_alloc(GFP_NOIO, 1);
737 struct completion event;
738 int ret;
739
740 rw |= REQ_SYNC | REQ_UNPLUG;
741
742 bio->bi_bdev = bdev;
743 bio->bi_sector = sector;
744 bio_add_page(bio, page, size, 0);
745 init_completion(&event);
746 bio->bi_private = &event;
747 bio->bi_end_io = bi_complete;
748 submit_bio(rw, bio);
749 wait_for_completion(&event);
750
751 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
752 bio_put(bio);
753 return ret;
754 }
755 EXPORT_SYMBOL_GPL(sync_page_io);
756
757 static int read_disk_sb(mdk_rdev_t * rdev, int size)
758 {
759 char b[BDEVNAME_SIZE];
760 if (!rdev->sb_page) {
761 MD_BUG();
762 return -EINVAL;
763 }
764 if (rdev->sb_loaded)
765 return 0;
766
767
768 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
769 goto fail;
770 rdev->sb_loaded = 1;
771 return 0;
772
773 fail:
774 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
775 bdevname(rdev->bdev,b));
776 return -EINVAL;
777 }
778
779 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
780 {
781 return sb1->set_uuid0 == sb2->set_uuid0 &&
782 sb1->set_uuid1 == sb2->set_uuid1 &&
783 sb1->set_uuid2 == sb2->set_uuid2 &&
784 sb1->set_uuid3 == sb2->set_uuid3;
785 }
786
787 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
788 {
789 int ret;
790 mdp_super_t *tmp1, *tmp2;
791
792 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
793 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
794
795 if (!tmp1 || !tmp2) {
796 ret = 0;
797 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
798 goto abort;
799 }
800
801 *tmp1 = *sb1;
802 *tmp2 = *sb2;
803
804 /*
805 * nr_disks is not constant
806 */
807 tmp1->nr_disks = 0;
808 tmp2->nr_disks = 0;
809
810 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
811 abort:
812 kfree(tmp1);
813 kfree(tmp2);
814 return ret;
815 }
816
817
818 static u32 md_csum_fold(u32 csum)
819 {
820 csum = (csum & 0xffff) + (csum >> 16);
821 return (csum & 0xffff) + (csum >> 16);
822 }
823
824 static unsigned int calc_sb_csum(mdp_super_t * sb)
825 {
826 u64 newcsum = 0;
827 u32 *sb32 = (u32*)sb;
828 int i;
829 unsigned int disk_csum, csum;
830
831 disk_csum = sb->sb_csum;
832 sb->sb_csum = 0;
833
834 for (i = 0; i < MD_SB_BYTES/4 ; i++)
835 newcsum += sb32[i];
836 csum = (newcsum & 0xffffffff) + (newcsum>>32);
837
838
839 #ifdef CONFIG_ALPHA
840 /* This used to use csum_partial, which was wrong for several
841 * reasons including that different results are returned on
842 * different architectures. It isn't critical that we get exactly
843 * the same return value as before (we always csum_fold before
844 * testing, and that removes any differences). However as we
845 * know that csum_partial always returned a 16bit value on
846 * alphas, do a fold to maximise conformity to previous behaviour.
847 */
848 sb->sb_csum = md_csum_fold(disk_csum);
849 #else
850 sb->sb_csum = disk_csum;
851 #endif
852 return csum;
853 }
854
855
856 /*
857 * Handle superblock details.
858 * We want to be able to handle multiple superblock formats
859 * so we have a common interface to them all, and an array of
860 * different handlers.
861 * We rely on user-space to write the initial superblock, and support
862 * reading and updating of superblocks.
863 * Interface methods are:
864 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
865 * loads and validates a superblock on dev.
866 * if refdev != NULL, compare superblocks on both devices
867 * Return:
868 * 0 - dev has a superblock that is compatible with refdev
869 * 1 - dev has a superblock that is compatible and newer than refdev
870 * so dev should be used as the refdev in future
871 * -EINVAL superblock incompatible or invalid
872 * -othererror e.g. -EIO
873 *
874 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
875 * Verify that dev is acceptable into mddev.
876 * The first time, mddev->raid_disks will be 0, and data from
877 * dev should be merged in. Subsequent calls check that dev
878 * is new enough. Return 0 or -EINVAL
879 *
880 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
881 * Update the superblock for rdev with data in mddev
882 * This does not write to disc.
883 *
884 */
885
886 struct super_type {
887 char *name;
888 struct module *owner;
889 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
890 int minor_version);
891 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
892 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
893 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
894 sector_t num_sectors);
895 };
896
897 /*
898 * Check that the given mddev has no bitmap.
899 *
900 * This function is called from the run method of all personalities that do not
901 * support bitmaps. It prints an error message and returns non-zero if mddev
902 * has a bitmap. Otherwise, it returns 0.
903 *
904 */
905 int md_check_no_bitmap(mddev_t *mddev)
906 {
907 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
908 return 0;
909 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
910 mdname(mddev), mddev->pers->name);
911 return 1;
912 }
913 EXPORT_SYMBOL(md_check_no_bitmap);
914
915 /*
916 * load_super for 0.90.0
917 */
918 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
919 {
920 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
921 mdp_super_t *sb;
922 int ret;
923
924 /*
925 * Calculate the position of the superblock (512byte sectors),
926 * it's at the end of the disk.
927 *
928 * It also happens to be a multiple of 4Kb.
929 */
930 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
931
932 ret = read_disk_sb(rdev, MD_SB_BYTES);
933 if (ret) return ret;
934
935 ret = -EINVAL;
936
937 bdevname(rdev->bdev, b);
938 sb = (mdp_super_t*)page_address(rdev->sb_page);
939
940 if (sb->md_magic != MD_SB_MAGIC) {
941 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
942 b);
943 goto abort;
944 }
945
946 if (sb->major_version != 0 ||
947 sb->minor_version < 90 ||
948 sb->minor_version > 91) {
949 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
950 sb->major_version, sb->minor_version,
951 b);
952 goto abort;
953 }
954
955 if (sb->raid_disks <= 0)
956 goto abort;
957
958 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
959 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
960 b);
961 goto abort;
962 }
963
964 rdev->preferred_minor = sb->md_minor;
965 rdev->data_offset = 0;
966 rdev->sb_size = MD_SB_BYTES;
967
968 if (sb->level == LEVEL_MULTIPATH)
969 rdev->desc_nr = -1;
970 else
971 rdev->desc_nr = sb->this_disk.number;
972
973 if (!refdev) {
974 ret = 1;
975 } else {
976 __u64 ev1, ev2;
977 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
978 if (!uuid_equal(refsb, sb)) {
979 printk(KERN_WARNING "md: %s has different UUID to %s\n",
980 b, bdevname(refdev->bdev,b2));
981 goto abort;
982 }
983 if (!sb_equal(refsb, sb)) {
984 printk(KERN_WARNING "md: %s has same UUID"
985 " but different superblock to %s\n",
986 b, bdevname(refdev->bdev, b2));
987 goto abort;
988 }
989 ev1 = md_event(sb);
990 ev2 = md_event(refsb);
991 if (ev1 > ev2)
992 ret = 1;
993 else
994 ret = 0;
995 }
996 rdev->sectors = rdev->sb_start;
997
998 if (rdev->sectors < sb->size * 2 && sb->level > 1)
999 /* "this cannot possibly happen" ... */
1000 ret = -EINVAL;
1001
1002 abort:
1003 return ret;
1004 }
1005
1006 /*
1007 * validate_super for 0.90.0
1008 */
1009 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1010 {
1011 mdp_disk_t *desc;
1012 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1013 __u64 ev1 = md_event(sb);
1014
1015 rdev->raid_disk = -1;
1016 clear_bit(Faulty, &rdev->flags);
1017 clear_bit(In_sync, &rdev->flags);
1018 clear_bit(WriteMostly, &rdev->flags);
1019 clear_bit(BarriersNotsupp, &rdev->flags);
1020
1021 if (mddev->raid_disks == 0) {
1022 mddev->major_version = 0;
1023 mddev->minor_version = sb->minor_version;
1024 mddev->patch_version = sb->patch_version;
1025 mddev->external = 0;
1026 mddev->chunk_sectors = sb->chunk_size >> 9;
1027 mddev->ctime = sb->ctime;
1028 mddev->utime = sb->utime;
1029 mddev->level = sb->level;
1030 mddev->clevel[0] = 0;
1031 mddev->layout = sb->layout;
1032 mddev->raid_disks = sb->raid_disks;
1033 mddev->dev_sectors = sb->size * 2;
1034 mddev->events = ev1;
1035 mddev->bitmap_info.offset = 0;
1036 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1037
1038 if (mddev->minor_version >= 91) {
1039 mddev->reshape_position = sb->reshape_position;
1040 mddev->delta_disks = sb->delta_disks;
1041 mddev->new_level = sb->new_level;
1042 mddev->new_layout = sb->new_layout;
1043 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1044 } else {
1045 mddev->reshape_position = MaxSector;
1046 mddev->delta_disks = 0;
1047 mddev->new_level = mddev->level;
1048 mddev->new_layout = mddev->layout;
1049 mddev->new_chunk_sectors = mddev->chunk_sectors;
1050 }
1051
1052 if (sb->state & (1<<MD_SB_CLEAN))
1053 mddev->recovery_cp = MaxSector;
1054 else {
1055 if (sb->events_hi == sb->cp_events_hi &&
1056 sb->events_lo == sb->cp_events_lo) {
1057 mddev->recovery_cp = sb->recovery_cp;
1058 } else
1059 mddev->recovery_cp = 0;
1060 }
1061
1062 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1063 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1064 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1065 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1066
1067 mddev->max_disks = MD_SB_DISKS;
1068
1069 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1070 mddev->bitmap_info.file == NULL)
1071 mddev->bitmap_info.offset =
1072 mddev->bitmap_info.default_offset;
1073
1074 } else if (mddev->pers == NULL) {
1075 /* Insist on good event counter while assembling, except
1076 * for spares (which don't need an event count) */
1077 ++ev1;
1078 if (sb->disks[rdev->desc_nr].state & (
1079 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1080 if (ev1 < mddev->events)
1081 return -EINVAL;
1082 } else if (mddev->bitmap) {
1083 /* if adding to array with a bitmap, then we can accept an
1084 * older device ... but not too old.
1085 */
1086 if (ev1 < mddev->bitmap->events_cleared)
1087 return 0;
1088 } else {
1089 if (ev1 < mddev->events)
1090 /* just a hot-add of a new device, leave raid_disk at -1 */
1091 return 0;
1092 }
1093
1094 if (mddev->level != LEVEL_MULTIPATH) {
1095 desc = sb->disks + rdev->desc_nr;
1096
1097 if (desc->state & (1<<MD_DISK_FAULTY))
1098 set_bit(Faulty, &rdev->flags);
1099 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1100 desc->raid_disk < mddev->raid_disks */) {
1101 set_bit(In_sync, &rdev->flags);
1102 rdev->raid_disk = desc->raid_disk;
1103 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1104 /* active but not in sync implies recovery up to
1105 * reshape position. We don't know exactly where
1106 * that is, so set to zero for now */
1107 if (mddev->minor_version >= 91) {
1108 rdev->recovery_offset = 0;
1109 rdev->raid_disk = desc->raid_disk;
1110 }
1111 }
1112 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1113 set_bit(WriteMostly, &rdev->flags);
1114 } else /* MULTIPATH are always insync */
1115 set_bit(In_sync, &rdev->flags);
1116 return 0;
1117 }
1118
1119 /*
1120 * sync_super for 0.90.0
1121 */
1122 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1123 {
1124 mdp_super_t *sb;
1125 mdk_rdev_t *rdev2;
1126 int next_spare = mddev->raid_disks;
1127
1128
1129 /* make rdev->sb match mddev data..
1130 *
1131 * 1/ zero out disks
1132 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1133 * 3/ any empty disks < next_spare become removed
1134 *
1135 * disks[0] gets initialised to REMOVED because
1136 * we cannot be sure from other fields if it has
1137 * been initialised or not.
1138 */
1139 int i;
1140 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1141
1142 rdev->sb_size = MD_SB_BYTES;
1143
1144 sb = (mdp_super_t*)page_address(rdev->sb_page);
1145
1146 memset(sb, 0, sizeof(*sb));
1147
1148 sb->md_magic = MD_SB_MAGIC;
1149 sb->major_version = mddev->major_version;
1150 sb->patch_version = mddev->patch_version;
1151 sb->gvalid_words = 0; /* ignored */
1152 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1153 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1154 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1155 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1156
1157 sb->ctime = mddev->ctime;
1158 sb->level = mddev->level;
1159 sb->size = mddev->dev_sectors / 2;
1160 sb->raid_disks = mddev->raid_disks;
1161 sb->md_minor = mddev->md_minor;
1162 sb->not_persistent = 0;
1163 sb->utime = mddev->utime;
1164 sb->state = 0;
1165 sb->events_hi = (mddev->events>>32);
1166 sb->events_lo = (u32)mddev->events;
1167
1168 if (mddev->reshape_position == MaxSector)
1169 sb->minor_version = 90;
1170 else {
1171 sb->minor_version = 91;
1172 sb->reshape_position = mddev->reshape_position;
1173 sb->new_level = mddev->new_level;
1174 sb->delta_disks = mddev->delta_disks;
1175 sb->new_layout = mddev->new_layout;
1176 sb->new_chunk = mddev->new_chunk_sectors << 9;
1177 }
1178 mddev->minor_version = sb->minor_version;
1179 if (mddev->in_sync)
1180 {
1181 sb->recovery_cp = mddev->recovery_cp;
1182 sb->cp_events_hi = (mddev->events>>32);
1183 sb->cp_events_lo = (u32)mddev->events;
1184 if (mddev->recovery_cp == MaxSector)
1185 sb->state = (1<< MD_SB_CLEAN);
1186 } else
1187 sb->recovery_cp = 0;
1188
1189 sb->layout = mddev->layout;
1190 sb->chunk_size = mddev->chunk_sectors << 9;
1191
1192 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1193 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1194
1195 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1196 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1197 mdp_disk_t *d;
1198 int desc_nr;
1199 int is_active = test_bit(In_sync, &rdev2->flags);
1200
1201 if (rdev2->raid_disk >= 0 &&
1202 sb->minor_version >= 91)
1203 /* we have nowhere to store the recovery_offset,
1204 * but if it is not below the reshape_position,
1205 * we can piggy-back on that.
1206 */
1207 is_active = 1;
1208 if (rdev2->raid_disk < 0 ||
1209 test_bit(Faulty, &rdev2->flags))
1210 is_active = 0;
1211 if (is_active)
1212 desc_nr = rdev2->raid_disk;
1213 else
1214 desc_nr = next_spare++;
1215 rdev2->desc_nr = desc_nr;
1216 d = &sb->disks[rdev2->desc_nr];
1217 nr_disks++;
1218 d->number = rdev2->desc_nr;
1219 d->major = MAJOR(rdev2->bdev->bd_dev);
1220 d->minor = MINOR(rdev2->bdev->bd_dev);
1221 if (is_active)
1222 d->raid_disk = rdev2->raid_disk;
1223 else
1224 d->raid_disk = rdev2->desc_nr; /* compatibility */
1225 if (test_bit(Faulty, &rdev2->flags))
1226 d->state = (1<<MD_DISK_FAULTY);
1227 else if (is_active) {
1228 d->state = (1<<MD_DISK_ACTIVE);
1229 if (test_bit(In_sync, &rdev2->flags))
1230 d->state |= (1<<MD_DISK_SYNC);
1231 active++;
1232 working++;
1233 } else {
1234 d->state = 0;
1235 spare++;
1236 working++;
1237 }
1238 if (test_bit(WriteMostly, &rdev2->flags))
1239 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1240 }
1241 /* now set the "removed" and "faulty" bits on any missing devices */
1242 for (i=0 ; i < mddev->raid_disks ; i++) {
1243 mdp_disk_t *d = &sb->disks[i];
1244 if (d->state == 0 && d->number == 0) {
1245 d->number = i;
1246 d->raid_disk = i;
1247 d->state = (1<<MD_DISK_REMOVED);
1248 d->state |= (1<<MD_DISK_FAULTY);
1249 failed++;
1250 }
1251 }
1252 sb->nr_disks = nr_disks;
1253 sb->active_disks = active;
1254 sb->working_disks = working;
1255 sb->failed_disks = failed;
1256 sb->spare_disks = spare;
1257
1258 sb->this_disk = sb->disks[rdev->desc_nr];
1259 sb->sb_csum = calc_sb_csum(sb);
1260 }
1261
1262 /*
1263 * rdev_size_change for 0.90.0
1264 */
1265 static unsigned long long
1266 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1267 {
1268 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1269 return 0; /* component must fit device */
1270 if (rdev->mddev->bitmap_info.offset)
1271 return 0; /* can't move bitmap */
1272 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1273 if (!num_sectors || num_sectors > rdev->sb_start)
1274 num_sectors = rdev->sb_start;
1275 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1276 rdev->sb_page);
1277 md_super_wait(rdev->mddev);
1278 return num_sectors / 2; /* kB for sysfs */
1279 }
1280
1281
1282 /*
1283 * version 1 superblock
1284 */
1285
1286 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1287 {
1288 __le32 disk_csum;
1289 u32 csum;
1290 unsigned long long newcsum;
1291 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1292 __le32 *isuper = (__le32*)sb;
1293 int i;
1294
1295 disk_csum = sb->sb_csum;
1296 sb->sb_csum = 0;
1297 newcsum = 0;
1298 for (i=0; size>=4; size -= 4 )
1299 newcsum += le32_to_cpu(*isuper++);
1300
1301 if (size == 2)
1302 newcsum += le16_to_cpu(*(__le16*) isuper);
1303
1304 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1305 sb->sb_csum = disk_csum;
1306 return cpu_to_le32(csum);
1307 }
1308
1309 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1310 {
1311 struct mdp_superblock_1 *sb;
1312 int ret;
1313 sector_t sb_start;
1314 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1315 int bmask;
1316
1317 /*
1318 * Calculate the position of the superblock in 512byte sectors.
1319 * It is always aligned to a 4K boundary and
1320 * depeding on minor_version, it can be:
1321 * 0: At least 8K, but less than 12K, from end of device
1322 * 1: At start of device
1323 * 2: 4K from start of device.
1324 */
1325 switch(minor_version) {
1326 case 0:
1327 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1328 sb_start -= 8*2;
1329 sb_start &= ~(sector_t)(4*2-1);
1330 break;
1331 case 1:
1332 sb_start = 0;
1333 break;
1334 case 2:
1335 sb_start = 8;
1336 break;
1337 default:
1338 return -EINVAL;
1339 }
1340 rdev->sb_start = sb_start;
1341
1342 /* superblock is rarely larger than 1K, but it can be larger,
1343 * and it is safe to read 4k, so we do that
1344 */
1345 ret = read_disk_sb(rdev, 4096);
1346 if (ret) return ret;
1347
1348
1349 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1350
1351 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1352 sb->major_version != cpu_to_le32(1) ||
1353 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1354 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1355 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1356 return -EINVAL;
1357
1358 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1359 printk("md: invalid superblock checksum on %s\n",
1360 bdevname(rdev->bdev,b));
1361 return -EINVAL;
1362 }
1363 if (le64_to_cpu(sb->data_size) < 10) {
1364 printk("md: data_size too small on %s\n",
1365 bdevname(rdev->bdev,b));
1366 return -EINVAL;
1367 }
1368
1369 rdev->preferred_minor = 0xffff;
1370 rdev->data_offset = le64_to_cpu(sb->data_offset);
1371 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1372
1373 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1374 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1375 if (rdev->sb_size & bmask)
1376 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1377
1378 if (minor_version
1379 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1380 return -EINVAL;
1381
1382 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1383 rdev->desc_nr = -1;
1384 else
1385 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1386
1387 if (!refdev) {
1388 ret = 1;
1389 } else {
1390 __u64 ev1, ev2;
1391 struct mdp_superblock_1 *refsb =
1392 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1393
1394 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1395 sb->level != refsb->level ||
1396 sb->layout != refsb->layout ||
1397 sb->chunksize != refsb->chunksize) {
1398 printk(KERN_WARNING "md: %s has strangely different"
1399 " superblock to %s\n",
1400 bdevname(rdev->bdev,b),
1401 bdevname(refdev->bdev,b2));
1402 return -EINVAL;
1403 }
1404 ev1 = le64_to_cpu(sb->events);
1405 ev2 = le64_to_cpu(refsb->events);
1406
1407 if (ev1 > ev2)
1408 ret = 1;
1409 else
1410 ret = 0;
1411 }
1412 if (minor_version)
1413 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1414 le64_to_cpu(sb->data_offset);
1415 else
1416 rdev->sectors = rdev->sb_start;
1417 if (rdev->sectors < le64_to_cpu(sb->data_size))
1418 return -EINVAL;
1419 rdev->sectors = le64_to_cpu(sb->data_size);
1420 if (le64_to_cpu(sb->size) > rdev->sectors)
1421 return -EINVAL;
1422 return ret;
1423 }
1424
1425 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1426 {
1427 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1428 __u64 ev1 = le64_to_cpu(sb->events);
1429
1430 rdev->raid_disk = -1;
1431 clear_bit(Faulty, &rdev->flags);
1432 clear_bit(In_sync, &rdev->flags);
1433 clear_bit(WriteMostly, &rdev->flags);
1434 clear_bit(BarriersNotsupp, &rdev->flags);
1435
1436 if (mddev->raid_disks == 0) {
1437 mddev->major_version = 1;
1438 mddev->patch_version = 0;
1439 mddev->external = 0;
1440 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1441 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1442 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1443 mddev->level = le32_to_cpu(sb->level);
1444 mddev->clevel[0] = 0;
1445 mddev->layout = le32_to_cpu(sb->layout);
1446 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1447 mddev->dev_sectors = le64_to_cpu(sb->size);
1448 mddev->events = ev1;
1449 mddev->bitmap_info.offset = 0;
1450 mddev->bitmap_info.default_offset = 1024 >> 9;
1451
1452 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1453 memcpy(mddev->uuid, sb->set_uuid, 16);
1454
1455 mddev->max_disks = (4096-256)/2;
1456
1457 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1458 mddev->bitmap_info.file == NULL )
1459 mddev->bitmap_info.offset =
1460 (__s32)le32_to_cpu(sb->bitmap_offset);
1461
1462 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1463 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1464 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1465 mddev->new_level = le32_to_cpu(sb->new_level);
1466 mddev->new_layout = le32_to_cpu(sb->new_layout);
1467 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1468 } else {
1469 mddev->reshape_position = MaxSector;
1470 mddev->delta_disks = 0;
1471 mddev->new_level = mddev->level;
1472 mddev->new_layout = mddev->layout;
1473 mddev->new_chunk_sectors = mddev->chunk_sectors;
1474 }
1475
1476 } else if (mddev->pers == NULL) {
1477 /* Insist of good event counter while assembling, except for
1478 * spares (which don't need an event count) */
1479 ++ev1;
1480 if (rdev->desc_nr >= 0 &&
1481 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1482 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1483 if (ev1 < mddev->events)
1484 return -EINVAL;
1485 } else if (mddev->bitmap) {
1486 /* If adding to array with a bitmap, then we can accept an
1487 * older device, but not too old.
1488 */
1489 if (ev1 < mddev->bitmap->events_cleared)
1490 return 0;
1491 } else {
1492 if (ev1 < mddev->events)
1493 /* just a hot-add of a new device, leave raid_disk at -1 */
1494 return 0;
1495 }
1496 if (mddev->level != LEVEL_MULTIPATH) {
1497 int role;
1498 if (rdev->desc_nr < 0 ||
1499 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1500 role = 0xffff;
1501 rdev->desc_nr = -1;
1502 } else
1503 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1504 switch(role) {
1505 case 0xffff: /* spare */
1506 break;
1507 case 0xfffe: /* faulty */
1508 set_bit(Faulty, &rdev->flags);
1509 break;
1510 default:
1511 if ((le32_to_cpu(sb->feature_map) &
1512 MD_FEATURE_RECOVERY_OFFSET))
1513 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1514 else
1515 set_bit(In_sync, &rdev->flags);
1516 rdev->raid_disk = role;
1517 break;
1518 }
1519 if (sb->devflags & WriteMostly1)
1520 set_bit(WriteMostly, &rdev->flags);
1521 } else /* MULTIPATH are always insync */
1522 set_bit(In_sync, &rdev->flags);
1523
1524 return 0;
1525 }
1526
1527 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1528 {
1529 struct mdp_superblock_1 *sb;
1530 mdk_rdev_t *rdev2;
1531 int max_dev, i;
1532 /* make rdev->sb match mddev and rdev data. */
1533
1534 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1535
1536 sb->feature_map = 0;
1537 sb->pad0 = 0;
1538 sb->recovery_offset = cpu_to_le64(0);
1539 memset(sb->pad1, 0, sizeof(sb->pad1));
1540 memset(sb->pad2, 0, sizeof(sb->pad2));
1541 memset(sb->pad3, 0, sizeof(sb->pad3));
1542
1543 sb->utime = cpu_to_le64((__u64)mddev->utime);
1544 sb->events = cpu_to_le64(mddev->events);
1545 if (mddev->in_sync)
1546 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1547 else
1548 sb->resync_offset = cpu_to_le64(0);
1549
1550 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1551
1552 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1553 sb->size = cpu_to_le64(mddev->dev_sectors);
1554 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1555 sb->level = cpu_to_le32(mddev->level);
1556 sb->layout = cpu_to_le32(mddev->layout);
1557
1558 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1559 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1560 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1561 }
1562
1563 if (rdev->raid_disk >= 0 &&
1564 !test_bit(In_sync, &rdev->flags)) {
1565 sb->feature_map |=
1566 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1567 sb->recovery_offset =
1568 cpu_to_le64(rdev->recovery_offset);
1569 }
1570
1571 if (mddev->reshape_position != MaxSector) {
1572 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1573 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1574 sb->new_layout = cpu_to_le32(mddev->new_layout);
1575 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1576 sb->new_level = cpu_to_le32(mddev->new_level);
1577 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1578 }
1579
1580 max_dev = 0;
1581 list_for_each_entry(rdev2, &mddev->disks, same_set)
1582 if (rdev2->desc_nr+1 > max_dev)
1583 max_dev = rdev2->desc_nr+1;
1584
1585 if (max_dev > le32_to_cpu(sb->max_dev)) {
1586 int bmask;
1587 sb->max_dev = cpu_to_le32(max_dev);
1588 rdev->sb_size = max_dev * 2 + 256;
1589 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1590 if (rdev->sb_size & bmask)
1591 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1592 }
1593 for (i=0; i<max_dev;i++)
1594 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1595
1596 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1597 i = rdev2->desc_nr;
1598 if (test_bit(Faulty, &rdev2->flags))
1599 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1600 else if (test_bit(In_sync, &rdev2->flags))
1601 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1602 else if (rdev2->raid_disk >= 0)
1603 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1604 else
1605 sb->dev_roles[i] = cpu_to_le16(0xffff);
1606 }
1607
1608 sb->sb_csum = calc_sb_1_csum(sb);
1609 }
1610
1611 static unsigned long long
1612 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1613 {
1614 struct mdp_superblock_1 *sb;
1615 sector_t max_sectors;
1616 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1617 return 0; /* component must fit device */
1618 if (rdev->sb_start < rdev->data_offset) {
1619 /* minor versions 1 and 2; superblock before data */
1620 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1621 max_sectors -= rdev->data_offset;
1622 if (!num_sectors || num_sectors > max_sectors)
1623 num_sectors = max_sectors;
1624 } else if (rdev->mddev->bitmap_info.offset) {
1625 /* minor version 0 with bitmap we can't move */
1626 return 0;
1627 } else {
1628 /* minor version 0; superblock after data */
1629 sector_t sb_start;
1630 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1631 sb_start &= ~(sector_t)(4*2 - 1);
1632 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1633 if (!num_sectors || num_sectors > max_sectors)
1634 num_sectors = max_sectors;
1635 rdev->sb_start = sb_start;
1636 }
1637 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1638 sb->data_size = cpu_to_le64(num_sectors);
1639 sb->super_offset = rdev->sb_start;
1640 sb->sb_csum = calc_sb_1_csum(sb);
1641 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1642 rdev->sb_page);
1643 md_super_wait(rdev->mddev);
1644 return num_sectors / 2; /* kB for sysfs */
1645 }
1646
1647 static struct super_type super_types[] = {
1648 [0] = {
1649 .name = "0.90.0",
1650 .owner = THIS_MODULE,
1651 .load_super = super_90_load,
1652 .validate_super = super_90_validate,
1653 .sync_super = super_90_sync,
1654 .rdev_size_change = super_90_rdev_size_change,
1655 },
1656 [1] = {
1657 .name = "md-1",
1658 .owner = THIS_MODULE,
1659 .load_super = super_1_load,
1660 .validate_super = super_1_validate,
1661 .sync_super = super_1_sync,
1662 .rdev_size_change = super_1_rdev_size_change,
1663 },
1664 };
1665
1666 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1667 {
1668 mdk_rdev_t *rdev, *rdev2;
1669
1670 rcu_read_lock();
1671 rdev_for_each_rcu(rdev, mddev1)
1672 rdev_for_each_rcu(rdev2, mddev2)
1673 if (rdev->bdev->bd_contains ==
1674 rdev2->bdev->bd_contains) {
1675 rcu_read_unlock();
1676 return 1;
1677 }
1678 rcu_read_unlock();
1679 return 0;
1680 }
1681
1682 static LIST_HEAD(pending_raid_disks);
1683
1684 /*
1685 * Try to register data integrity profile for an mddev
1686 *
1687 * This is called when an array is started and after a disk has been kicked
1688 * from the array. It only succeeds if all working and active component devices
1689 * are integrity capable with matching profiles.
1690 */
1691 int md_integrity_register(mddev_t *mddev)
1692 {
1693 mdk_rdev_t *rdev, *reference = NULL;
1694
1695 if (list_empty(&mddev->disks))
1696 return 0; /* nothing to do */
1697 if (blk_get_integrity(mddev->gendisk))
1698 return 0; /* already registered */
1699 list_for_each_entry(rdev, &mddev->disks, same_set) {
1700 /* skip spares and non-functional disks */
1701 if (test_bit(Faulty, &rdev->flags))
1702 continue;
1703 if (rdev->raid_disk < 0)
1704 continue;
1705 /*
1706 * If at least one rdev is not integrity capable, we can not
1707 * enable data integrity for the md device.
1708 */
1709 if (!bdev_get_integrity(rdev->bdev))
1710 return -EINVAL;
1711 if (!reference) {
1712 /* Use the first rdev as the reference */
1713 reference = rdev;
1714 continue;
1715 }
1716 /* does this rdev's profile match the reference profile? */
1717 if (blk_integrity_compare(reference->bdev->bd_disk,
1718 rdev->bdev->bd_disk) < 0)
1719 return -EINVAL;
1720 }
1721 /*
1722 * All component devices are integrity capable and have matching
1723 * profiles, register the common profile for the md device.
1724 */
1725 if (blk_integrity_register(mddev->gendisk,
1726 bdev_get_integrity(reference->bdev)) != 0) {
1727 printk(KERN_ERR "md: failed to register integrity for %s\n",
1728 mdname(mddev));
1729 return -EINVAL;
1730 }
1731 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1732 mdname(mddev));
1733 return 0;
1734 }
1735 EXPORT_SYMBOL(md_integrity_register);
1736
1737 /* Disable data integrity if non-capable/non-matching disk is being added */
1738 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1739 {
1740 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1741 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1742
1743 if (!bi_mddev) /* nothing to do */
1744 return;
1745 if (rdev->raid_disk < 0) /* skip spares */
1746 return;
1747 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1748 rdev->bdev->bd_disk) >= 0)
1749 return;
1750 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1751 blk_integrity_unregister(mddev->gendisk);
1752 }
1753 EXPORT_SYMBOL(md_integrity_add_rdev);
1754
1755 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1756 {
1757 char b[BDEVNAME_SIZE];
1758 struct kobject *ko;
1759 char *s;
1760 int err;
1761
1762 if (rdev->mddev) {
1763 MD_BUG();
1764 return -EINVAL;
1765 }
1766
1767 /* prevent duplicates */
1768 if (find_rdev(mddev, rdev->bdev->bd_dev))
1769 return -EEXIST;
1770
1771 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1772 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1773 rdev->sectors < mddev->dev_sectors)) {
1774 if (mddev->pers) {
1775 /* Cannot change size, so fail
1776 * If mddev->level <= 0, then we don't care
1777 * about aligning sizes (e.g. linear)
1778 */
1779 if (mddev->level > 0)
1780 return -ENOSPC;
1781 } else
1782 mddev->dev_sectors = rdev->sectors;
1783 }
1784
1785 /* Verify rdev->desc_nr is unique.
1786 * If it is -1, assign a free number, else
1787 * check number is not in use
1788 */
1789 if (rdev->desc_nr < 0) {
1790 int choice = 0;
1791 if (mddev->pers) choice = mddev->raid_disks;
1792 while (find_rdev_nr(mddev, choice))
1793 choice++;
1794 rdev->desc_nr = choice;
1795 } else {
1796 if (find_rdev_nr(mddev, rdev->desc_nr))
1797 return -EBUSY;
1798 }
1799 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1800 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1801 mdname(mddev), mddev->max_disks);
1802 return -EBUSY;
1803 }
1804 bdevname(rdev->bdev,b);
1805 while ( (s=strchr(b, '/')) != NULL)
1806 *s = '!';
1807
1808 rdev->mddev = mddev;
1809 printk(KERN_INFO "md: bind<%s>\n", b);
1810
1811 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1812 goto fail;
1813
1814 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1815 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1816 kobject_del(&rdev->kobj);
1817 goto fail;
1818 }
1819 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, NULL, "state");
1820
1821 list_add_rcu(&rdev->same_set, &mddev->disks);
1822 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1823
1824 /* May as well allow recovery to be retried once */
1825 mddev->recovery_disabled = 0;
1826
1827 return 0;
1828
1829 fail:
1830 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1831 b, mdname(mddev));
1832 return err;
1833 }
1834
1835 static void md_delayed_delete(struct work_struct *ws)
1836 {
1837 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1838 kobject_del(&rdev->kobj);
1839 kobject_put(&rdev->kobj);
1840 }
1841
1842 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1843 {
1844 char b[BDEVNAME_SIZE];
1845 if (!rdev->mddev) {
1846 MD_BUG();
1847 return;
1848 }
1849 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1850 list_del_rcu(&rdev->same_set);
1851 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1852 rdev->mddev = NULL;
1853 sysfs_remove_link(&rdev->kobj, "block");
1854 sysfs_put(rdev->sysfs_state);
1855 rdev->sysfs_state = NULL;
1856 /* We need to delay this, otherwise we can deadlock when
1857 * writing to 'remove' to "dev/state". We also need
1858 * to delay it due to rcu usage.
1859 */
1860 synchronize_rcu();
1861 INIT_WORK(&rdev->del_work, md_delayed_delete);
1862 kobject_get(&rdev->kobj);
1863 schedule_work(&rdev->del_work);
1864 }
1865
1866 /*
1867 * prevent the device from being mounted, repartitioned or
1868 * otherwise reused by a RAID array (or any other kernel
1869 * subsystem), by bd_claiming the device.
1870 */
1871 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1872 {
1873 int err = 0;
1874 struct block_device *bdev;
1875 char b[BDEVNAME_SIZE];
1876
1877 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1878 if (IS_ERR(bdev)) {
1879 printk(KERN_ERR "md: could not open %s.\n",
1880 __bdevname(dev, b));
1881 return PTR_ERR(bdev);
1882 }
1883 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1884 if (err) {
1885 printk(KERN_ERR "md: could not bd_claim %s.\n",
1886 bdevname(bdev, b));
1887 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1888 return err;
1889 }
1890 if (!shared)
1891 set_bit(AllReserved, &rdev->flags);
1892 rdev->bdev = bdev;
1893 return err;
1894 }
1895
1896 static void unlock_rdev(mdk_rdev_t *rdev)
1897 {
1898 struct block_device *bdev = rdev->bdev;
1899 rdev->bdev = NULL;
1900 if (!bdev)
1901 MD_BUG();
1902 bd_release(bdev);
1903 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1904 }
1905
1906 void md_autodetect_dev(dev_t dev);
1907
1908 static void export_rdev(mdk_rdev_t * rdev)
1909 {
1910 char b[BDEVNAME_SIZE];
1911 printk(KERN_INFO "md: export_rdev(%s)\n",
1912 bdevname(rdev->bdev,b));
1913 if (rdev->mddev)
1914 MD_BUG();
1915 free_disk_sb(rdev);
1916 #ifndef MODULE
1917 if (test_bit(AutoDetected, &rdev->flags))
1918 md_autodetect_dev(rdev->bdev->bd_dev);
1919 #endif
1920 unlock_rdev(rdev);
1921 kobject_put(&rdev->kobj);
1922 }
1923
1924 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1925 {
1926 unbind_rdev_from_array(rdev);
1927 export_rdev(rdev);
1928 }
1929
1930 static void export_array(mddev_t *mddev)
1931 {
1932 mdk_rdev_t *rdev, *tmp;
1933
1934 rdev_for_each(rdev, tmp, mddev) {
1935 if (!rdev->mddev) {
1936 MD_BUG();
1937 continue;
1938 }
1939 kick_rdev_from_array(rdev);
1940 }
1941 if (!list_empty(&mddev->disks))
1942 MD_BUG();
1943 mddev->raid_disks = 0;
1944 mddev->major_version = 0;
1945 }
1946
1947 static void print_desc(mdp_disk_t *desc)
1948 {
1949 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1950 desc->major,desc->minor,desc->raid_disk,desc->state);
1951 }
1952
1953 static void print_sb_90(mdp_super_t *sb)
1954 {
1955 int i;
1956
1957 printk(KERN_INFO
1958 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1959 sb->major_version, sb->minor_version, sb->patch_version,
1960 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1961 sb->ctime);
1962 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1963 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1964 sb->md_minor, sb->layout, sb->chunk_size);
1965 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1966 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1967 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1968 sb->failed_disks, sb->spare_disks,
1969 sb->sb_csum, (unsigned long)sb->events_lo);
1970
1971 printk(KERN_INFO);
1972 for (i = 0; i < MD_SB_DISKS; i++) {
1973 mdp_disk_t *desc;
1974
1975 desc = sb->disks + i;
1976 if (desc->number || desc->major || desc->minor ||
1977 desc->raid_disk || (desc->state && (desc->state != 4))) {
1978 printk(" D %2d: ", i);
1979 print_desc(desc);
1980 }
1981 }
1982 printk(KERN_INFO "md: THIS: ");
1983 print_desc(&sb->this_disk);
1984 }
1985
1986 static void print_sb_1(struct mdp_superblock_1 *sb)
1987 {
1988 __u8 *uuid;
1989
1990 uuid = sb->set_uuid;
1991 printk(KERN_INFO
1992 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1993 "md: Name: \"%s\" CT:%llu\n",
1994 le32_to_cpu(sb->major_version),
1995 le32_to_cpu(sb->feature_map),
1996 uuid,
1997 sb->set_name,
1998 (unsigned long long)le64_to_cpu(sb->ctime)
1999 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2000
2001 uuid = sb->device_uuid;
2002 printk(KERN_INFO
2003 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2004 " RO:%llu\n"
2005 "md: Dev:%08x UUID: %pU\n"
2006 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2007 "md: (MaxDev:%u) \n",
2008 le32_to_cpu(sb->level),
2009 (unsigned long long)le64_to_cpu(sb->size),
2010 le32_to_cpu(sb->raid_disks),
2011 le32_to_cpu(sb->layout),
2012 le32_to_cpu(sb->chunksize),
2013 (unsigned long long)le64_to_cpu(sb->data_offset),
2014 (unsigned long long)le64_to_cpu(sb->data_size),
2015 (unsigned long long)le64_to_cpu(sb->super_offset),
2016 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2017 le32_to_cpu(sb->dev_number),
2018 uuid,
2019 sb->devflags,
2020 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2021 (unsigned long long)le64_to_cpu(sb->events),
2022 (unsigned long long)le64_to_cpu(sb->resync_offset),
2023 le32_to_cpu(sb->sb_csum),
2024 le32_to_cpu(sb->max_dev)
2025 );
2026 }
2027
2028 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2029 {
2030 char b[BDEVNAME_SIZE];
2031 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2032 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2033 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2034 rdev->desc_nr);
2035 if (rdev->sb_loaded) {
2036 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2037 switch (major_version) {
2038 case 0:
2039 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2040 break;
2041 case 1:
2042 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2043 break;
2044 }
2045 } else
2046 printk(KERN_INFO "md: no rdev superblock!\n");
2047 }
2048
2049 static void md_print_devices(void)
2050 {
2051 struct list_head *tmp;
2052 mdk_rdev_t *rdev;
2053 mddev_t *mddev;
2054 char b[BDEVNAME_SIZE];
2055
2056 printk("\n");
2057 printk("md: **********************************\n");
2058 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2059 printk("md: **********************************\n");
2060 for_each_mddev(mddev, tmp) {
2061
2062 if (mddev->bitmap)
2063 bitmap_print_sb(mddev->bitmap);
2064 else
2065 printk("%s: ", mdname(mddev));
2066 list_for_each_entry(rdev, &mddev->disks, same_set)
2067 printk("<%s>", bdevname(rdev->bdev,b));
2068 printk("\n");
2069
2070 list_for_each_entry(rdev, &mddev->disks, same_set)
2071 print_rdev(rdev, mddev->major_version);
2072 }
2073 printk("md: **********************************\n");
2074 printk("\n");
2075 }
2076
2077
2078 static void sync_sbs(mddev_t * mddev, int nospares)
2079 {
2080 /* Update each superblock (in-memory image), but
2081 * if we are allowed to, skip spares which already
2082 * have the right event counter, or have one earlier
2083 * (which would mean they aren't being marked as dirty
2084 * with the rest of the array)
2085 */
2086 mdk_rdev_t *rdev;
2087
2088 /* First make sure individual recovery_offsets are correct */
2089 list_for_each_entry(rdev, &mddev->disks, same_set) {
2090 if (rdev->raid_disk >= 0 &&
2091 mddev->delta_disks >= 0 &&
2092 !test_bit(In_sync, &rdev->flags) &&
2093 mddev->curr_resync_completed > rdev->recovery_offset)
2094 rdev->recovery_offset = mddev->curr_resync_completed;
2095
2096 }
2097 list_for_each_entry(rdev, &mddev->disks, same_set) {
2098 if (rdev->sb_events == mddev->events ||
2099 (nospares &&
2100 rdev->raid_disk < 0 &&
2101 rdev->sb_events+1 == mddev->events)) {
2102 /* Don't update this superblock */
2103 rdev->sb_loaded = 2;
2104 } else {
2105 super_types[mddev->major_version].
2106 sync_super(mddev, rdev);
2107 rdev->sb_loaded = 1;
2108 }
2109 }
2110 }
2111
2112 static void md_update_sb(mddev_t * mddev, int force_change)
2113 {
2114 mdk_rdev_t *rdev;
2115 int sync_req;
2116 int nospares = 0;
2117
2118 mddev->utime = get_seconds();
2119 if (mddev->external)
2120 return;
2121 repeat:
2122 spin_lock_irq(&mddev->write_lock);
2123
2124 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2125 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2126 force_change = 1;
2127 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2128 /* just a clean<-> dirty transition, possibly leave spares alone,
2129 * though if events isn't the right even/odd, we will have to do
2130 * spares after all
2131 */
2132 nospares = 1;
2133 if (force_change)
2134 nospares = 0;
2135 if (mddev->degraded)
2136 /* If the array is degraded, then skipping spares is both
2137 * dangerous and fairly pointless.
2138 * Dangerous because a device that was removed from the array
2139 * might have a event_count that still looks up-to-date,
2140 * so it can be re-added without a resync.
2141 * Pointless because if there are any spares to skip,
2142 * then a recovery will happen and soon that array won't
2143 * be degraded any more and the spare can go back to sleep then.
2144 */
2145 nospares = 0;
2146
2147 sync_req = mddev->in_sync;
2148
2149 /* If this is just a dirty<->clean transition, and the array is clean
2150 * and 'events' is odd, we can roll back to the previous clean state */
2151 if (nospares
2152 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2153 && mddev->can_decrease_events
2154 && mddev->events != 1) {
2155 mddev->events--;
2156 mddev->can_decrease_events = 0;
2157 } else {
2158 /* otherwise we have to go forward and ... */
2159 mddev->events ++;
2160 mddev->can_decrease_events = nospares;
2161 }
2162
2163 if (!mddev->events) {
2164 /*
2165 * oops, this 64-bit counter should never wrap.
2166 * Either we are in around ~1 trillion A.C., assuming
2167 * 1 reboot per second, or we have a bug:
2168 */
2169 MD_BUG();
2170 mddev->events --;
2171 }
2172
2173 /*
2174 * do not write anything to disk if using
2175 * nonpersistent superblocks
2176 */
2177 if (!mddev->persistent) {
2178 if (!mddev->external)
2179 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2180
2181 spin_unlock_irq(&mddev->write_lock);
2182 wake_up(&mddev->sb_wait);
2183 return;
2184 }
2185 sync_sbs(mddev, nospares);
2186 spin_unlock_irq(&mddev->write_lock);
2187
2188 dprintk(KERN_INFO
2189 "md: updating %s RAID superblock on device (in sync %d)\n",
2190 mdname(mddev),mddev->in_sync);
2191
2192 bitmap_update_sb(mddev->bitmap);
2193 list_for_each_entry(rdev, &mddev->disks, same_set) {
2194 char b[BDEVNAME_SIZE];
2195 dprintk(KERN_INFO "md: ");
2196 if (rdev->sb_loaded != 1)
2197 continue; /* no noise on spare devices */
2198 if (test_bit(Faulty, &rdev->flags))
2199 dprintk("(skipping faulty ");
2200
2201 dprintk("%s ", bdevname(rdev->bdev,b));
2202 if (!test_bit(Faulty, &rdev->flags)) {
2203 md_super_write(mddev,rdev,
2204 rdev->sb_start, rdev->sb_size,
2205 rdev->sb_page);
2206 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2207 bdevname(rdev->bdev,b),
2208 (unsigned long long)rdev->sb_start);
2209 rdev->sb_events = mddev->events;
2210
2211 } else
2212 dprintk(")\n");
2213 if (mddev->level == LEVEL_MULTIPATH)
2214 /* only need to write one superblock... */
2215 break;
2216 }
2217 md_super_wait(mddev);
2218 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2219
2220 spin_lock_irq(&mddev->write_lock);
2221 if (mddev->in_sync != sync_req ||
2222 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2223 /* have to write it out again */
2224 spin_unlock_irq(&mddev->write_lock);
2225 goto repeat;
2226 }
2227 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2228 spin_unlock_irq(&mddev->write_lock);
2229 wake_up(&mddev->sb_wait);
2230 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2231 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2232
2233 }
2234
2235 /* words written to sysfs files may, or may not, be \n terminated.
2236 * We want to accept with case. For this we use cmd_match.
2237 */
2238 static int cmd_match(const char *cmd, const char *str)
2239 {
2240 /* See if cmd, written into a sysfs file, matches
2241 * str. They must either be the same, or cmd can
2242 * have a trailing newline
2243 */
2244 while (*cmd && *str && *cmd == *str) {
2245 cmd++;
2246 str++;
2247 }
2248 if (*cmd == '\n')
2249 cmd++;
2250 if (*str || *cmd)
2251 return 0;
2252 return 1;
2253 }
2254
2255 struct rdev_sysfs_entry {
2256 struct attribute attr;
2257 ssize_t (*show)(mdk_rdev_t *, char *);
2258 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2259 };
2260
2261 static ssize_t
2262 state_show(mdk_rdev_t *rdev, char *page)
2263 {
2264 char *sep = "";
2265 size_t len = 0;
2266
2267 if (test_bit(Faulty, &rdev->flags)) {
2268 len+= sprintf(page+len, "%sfaulty",sep);
2269 sep = ",";
2270 }
2271 if (test_bit(In_sync, &rdev->flags)) {
2272 len += sprintf(page+len, "%sin_sync",sep);
2273 sep = ",";
2274 }
2275 if (test_bit(WriteMostly, &rdev->flags)) {
2276 len += sprintf(page+len, "%swrite_mostly",sep);
2277 sep = ",";
2278 }
2279 if (test_bit(Blocked, &rdev->flags)) {
2280 len += sprintf(page+len, "%sblocked", sep);
2281 sep = ",";
2282 }
2283 if (!test_bit(Faulty, &rdev->flags) &&
2284 !test_bit(In_sync, &rdev->flags)) {
2285 len += sprintf(page+len, "%sspare", sep);
2286 sep = ",";
2287 }
2288 return len+sprintf(page+len, "\n");
2289 }
2290
2291 static ssize_t
2292 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2293 {
2294 /* can write
2295 * faulty - simulates and error
2296 * remove - disconnects the device
2297 * writemostly - sets write_mostly
2298 * -writemostly - clears write_mostly
2299 * blocked - sets the Blocked flag
2300 * -blocked - clears the Blocked flag
2301 * insync - sets Insync providing device isn't active
2302 */
2303 int err = -EINVAL;
2304 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2305 md_error(rdev->mddev, rdev);
2306 err = 0;
2307 } else if (cmd_match(buf, "remove")) {
2308 if (rdev->raid_disk >= 0)
2309 err = -EBUSY;
2310 else {
2311 mddev_t *mddev = rdev->mddev;
2312 kick_rdev_from_array(rdev);
2313 if (mddev->pers)
2314 md_update_sb(mddev, 1);
2315 md_new_event(mddev);
2316 err = 0;
2317 }
2318 } else if (cmd_match(buf, "writemostly")) {
2319 set_bit(WriteMostly, &rdev->flags);
2320 err = 0;
2321 } else if (cmd_match(buf, "-writemostly")) {
2322 clear_bit(WriteMostly, &rdev->flags);
2323 err = 0;
2324 } else if (cmd_match(buf, "blocked")) {
2325 set_bit(Blocked, &rdev->flags);
2326 err = 0;
2327 } else if (cmd_match(buf, "-blocked")) {
2328 clear_bit(Blocked, &rdev->flags);
2329 wake_up(&rdev->blocked_wait);
2330 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2331 md_wakeup_thread(rdev->mddev->thread);
2332
2333 err = 0;
2334 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2335 set_bit(In_sync, &rdev->flags);
2336 err = 0;
2337 }
2338 if (!err && rdev->sysfs_state)
2339 sysfs_notify_dirent(rdev->sysfs_state);
2340 return err ? err : len;
2341 }
2342 static struct rdev_sysfs_entry rdev_state =
2343 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2344
2345 static ssize_t
2346 errors_show(mdk_rdev_t *rdev, char *page)
2347 {
2348 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2349 }
2350
2351 static ssize_t
2352 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2353 {
2354 char *e;
2355 unsigned long n = simple_strtoul(buf, &e, 10);
2356 if (*buf && (*e == 0 || *e == '\n')) {
2357 atomic_set(&rdev->corrected_errors, n);
2358 return len;
2359 }
2360 return -EINVAL;
2361 }
2362 static struct rdev_sysfs_entry rdev_errors =
2363 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2364
2365 static ssize_t
2366 slot_show(mdk_rdev_t *rdev, char *page)
2367 {
2368 if (rdev->raid_disk < 0)
2369 return sprintf(page, "none\n");
2370 else
2371 return sprintf(page, "%d\n", rdev->raid_disk);
2372 }
2373
2374 static ssize_t
2375 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2376 {
2377 char *e;
2378 int err;
2379 char nm[20];
2380 int slot = simple_strtoul(buf, &e, 10);
2381 if (strncmp(buf, "none", 4)==0)
2382 slot = -1;
2383 else if (e==buf || (*e && *e!= '\n'))
2384 return -EINVAL;
2385 if (rdev->mddev->pers && slot == -1) {
2386 /* Setting 'slot' on an active array requires also
2387 * updating the 'rd%d' link, and communicating
2388 * with the personality with ->hot_*_disk.
2389 * For now we only support removing
2390 * failed/spare devices. This normally happens automatically,
2391 * but not when the metadata is externally managed.
2392 */
2393 if (rdev->raid_disk == -1)
2394 return -EEXIST;
2395 /* personality does all needed checks */
2396 if (rdev->mddev->pers->hot_add_disk == NULL)
2397 return -EINVAL;
2398 err = rdev->mddev->pers->
2399 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2400 if (err)
2401 return err;
2402 sprintf(nm, "rd%d", rdev->raid_disk);
2403 sysfs_remove_link(&rdev->mddev->kobj, nm);
2404 rdev->raid_disk = -1;
2405 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2406 md_wakeup_thread(rdev->mddev->thread);
2407 } else if (rdev->mddev->pers) {
2408 mdk_rdev_t *rdev2;
2409 /* Activating a spare .. or possibly reactivating
2410 * if we ever get bitmaps working here.
2411 */
2412
2413 if (rdev->raid_disk != -1)
2414 return -EBUSY;
2415
2416 if (rdev->mddev->pers->hot_add_disk == NULL)
2417 return -EINVAL;
2418
2419 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2420 if (rdev2->raid_disk == slot)
2421 return -EEXIST;
2422
2423 rdev->raid_disk = slot;
2424 if (test_bit(In_sync, &rdev->flags))
2425 rdev->saved_raid_disk = slot;
2426 else
2427 rdev->saved_raid_disk = -1;
2428 err = rdev->mddev->pers->
2429 hot_add_disk(rdev->mddev, rdev);
2430 if (err) {
2431 rdev->raid_disk = -1;
2432 return err;
2433 } else
2434 sysfs_notify_dirent(rdev->sysfs_state);
2435 sprintf(nm, "rd%d", rdev->raid_disk);
2436 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2437 printk(KERN_WARNING
2438 "md: cannot register "
2439 "%s for %s\n",
2440 nm, mdname(rdev->mddev));
2441
2442 /* don't wakeup anyone, leave that to userspace. */
2443 } else {
2444 if (slot >= rdev->mddev->raid_disks)
2445 return -ENOSPC;
2446 rdev->raid_disk = slot;
2447 /* assume it is working */
2448 clear_bit(Faulty, &rdev->flags);
2449 clear_bit(WriteMostly, &rdev->flags);
2450 set_bit(In_sync, &rdev->flags);
2451 sysfs_notify_dirent(rdev->sysfs_state);
2452 }
2453 return len;
2454 }
2455
2456
2457 static struct rdev_sysfs_entry rdev_slot =
2458 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2459
2460 static ssize_t
2461 offset_show(mdk_rdev_t *rdev, char *page)
2462 {
2463 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2464 }
2465
2466 static ssize_t
2467 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2468 {
2469 char *e;
2470 unsigned long long offset = simple_strtoull(buf, &e, 10);
2471 if (e==buf || (*e && *e != '\n'))
2472 return -EINVAL;
2473 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2474 return -EBUSY;
2475 if (rdev->sectors && rdev->mddev->external)
2476 /* Must set offset before size, so overlap checks
2477 * can be sane */
2478 return -EBUSY;
2479 rdev->data_offset = offset;
2480 return len;
2481 }
2482
2483 static struct rdev_sysfs_entry rdev_offset =
2484 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2485
2486 static ssize_t
2487 rdev_size_show(mdk_rdev_t *rdev, char *page)
2488 {
2489 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2490 }
2491
2492 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2493 {
2494 /* check if two start/length pairs overlap */
2495 if (s1+l1 <= s2)
2496 return 0;
2497 if (s2+l2 <= s1)
2498 return 0;
2499 return 1;
2500 }
2501
2502 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2503 {
2504 unsigned long long blocks;
2505 sector_t new;
2506
2507 if (strict_strtoull(buf, 10, &blocks) < 0)
2508 return -EINVAL;
2509
2510 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2511 return -EINVAL; /* sector conversion overflow */
2512
2513 new = blocks * 2;
2514 if (new != blocks * 2)
2515 return -EINVAL; /* unsigned long long to sector_t overflow */
2516
2517 *sectors = new;
2518 return 0;
2519 }
2520
2521 static ssize_t
2522 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2523 {
2524 mddev_t *my_mddev = rdev->mddev;
2525 sector_t oldsectors = rdev->sectors;
2526 sector_t sectors;
2527
2528 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2529 return -EINVAL;
2530 if (my_mddev->pers && rdev->raid_disk >= 0) {
2531 if (my_mddev->persistent) {
2532 sectors = super_types[my_mddev->major_version].
2533 rdev_size_change(rdev, sectors);
2534 if (!sectors)
2535 return -EBUSY;
2536 } else if (!sectors)
2537 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2538 rdev->data_offset;
2539 }
2540 if (sectors < my_mddev->dev_sectors)
2541 return -EINVAL; /* component must fit device */
2542
2543 rdev->sectors = sectors;
2544 if (sectors > oldsectors && my_mddev->external) {
2545 /* need to check that all other rdevs with the same ->bdev
2546 * do not overlap. We need to unlock the mddev to avoid
2547 * a deadlock. We have already changed rdev->sectors, and if
2548 * we have to change it back, we will have the lock again.
2549 */
2550 mddev_t *mddev;
2551 int overlap = 0;
2552 struct list_head *tmp;
2553
2554 mddev_unlock(my_mddev);
2555 for_each_mddev(mddev, tmp) {
2556 mdk_rdev_t *rdev2;
2557
2558 mddev_lock(mddev);
2559 list_for_each_entry(rdev2, &mddev->disks, same_set)
2560 if (test_bit(AllReserved, &rdev2->flags) ||
2561 (rdev->bdev == rdev2->bdev &&
2562 rdev != rdev2 &&
2563 overlaps(rdev->data_offset, rdev->sectors,
2564 rdev2->data_offset,
2565 rdev2->sectors))) {
2566 overlap = 1;
2567 break;
2568 }
2569 mddev_unlock(mddev);
2570 if (overlap) {
2571 mddev_put(mddev);
2572 break;
2573 }
2574 }
2575 mddev_lock(my_mddev);
2576 if (overlap) {
2577 /* Someone else could have slipped in a size
2578 * change here, but doing so is just silly.
2579 * We put oldsectors back because we *know* it is
2580 * safe, and trust userspace not to race with
2581 * itself
2582 */
2583 rdev->sectors = oldsectors;
2584 return -EBUSY;
2585 }
2586 }
2587 return len;
2588 }
2589
2590 static struct rdev_sysfs_entry rdev_size =
2591 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2592
2593
2594 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2595 {
2596 unsigned long long recovery_start = rdev->recovery_offset;
2597
2598 if (test_bit(In_sync, &rdev->flags) ||
2599 recovery_start == MaxSector)
2600 return sprintf(page, "none\n");
2601
2602 return sprintf(page, "%llu\n", recovery_start);
2603 }
2604
2605 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2606 {
2607 unsigned long long recovery_start;
2608
2609 if (cmd_match(buf, "none"))
2610 recovery_start = MaxSector;
2611 else if (strict_strtoull(buf, 10, &recovery_start))
2612 return -EINVAL;
2613
2614 if (rdev->mddev->pers &&
2615 rdev->raid_disk >= 0)
2616 return -EBUSY;
2617
2618 rdev->recovery_offset = recovery_start;
2619 if (recovery_start == MaxSector)
2620 set_bit(In_sync, &rdev->flags);
2621 else
2622 clear_bit(In_sync, &rdev->flags);
2623 return len;
2624 }
2625
2626 static struct rdev_sysfs_entry rdev_recovery_start =
2627 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2628
2629 static struct attribute *rdev_default_attrs[] = {
2630 &rdev_state.attr,
2631 &rdev_errors.attr,
2632 &rdev_slot.attr,
2633 &rdev_offset.attr,
2634 &rdev_size.attr,
2635 &rdev_recovery_start.attr,
2636 NULL,
2637 };
2638 static ssize_t
2639 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2640 {
2641 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2642 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2643 mddev_t *mddev = rdev->mddev;
2644 ssize_t rv;
2645
2646 if (!entry->show)
2647 return -EIO;
2648
2649 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2650 if (!rv) {
2651 if (rdev->mddev == NULL)
2652 rv = -EBUSY;
2653 else
2654 rv = entry->show(rdev, page);
2655 mddev_unlock(mddev);
2656 }
2657 return rv;
2658 }
2659
2660 static ssize_t
2661 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2662 const char *page, size_t length)
2663 {
2664 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2665 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2666 ssize_t rv;
2667 mddev_t *mddev = rdev->mddev;
2668
2669 if (!entry->store)
2670 return -EIO;
2671 if (!capable(CAP_SYS_ADMIN))
2672 return -EACCES;
2673 rv = mddev ? mddev_lock(mddev): -EBUSY;
2674 if (!rv) {
2675 if (rdev->mddev == NULL)
2676 rv = -EBUSY;
2677 else
2678 rv = entry->store(rdev, page, length);
2679 mddev_unlock(mddev);
2680 }
2681 return rv;
2682 }
2683
2684 static void rdev_free(struct kobject *ko)
2685 {
2686 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2687 kfree(rdev);
2688 }
2689 static const struct sysfs_ops rdev_sysfs_ops = {
2690 .show = rdev_attr_show,
2691 .store = rdev_attr_store,
2692 };
2693 static struct kobj_type rdev_ktype = {
2694 .release = rdev_free,
2695 .sysfs_ops = &rdev_sysfs_ops,
2696 .default_attrs = rdev_default_attrs,
2697 };
2698
2699 /*
2700 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2701 *
2702 * mark the device faulty if:
2703 *
2704 * - the device is nonexistent (zero size)
2705 * - the device has no valid superblock
2706 *
2707 * a faulty rdev _never_ has rdev->sb set.
2708 */
2709 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2710 {
2711 char b[BDEVNAME_SIZE];
2712 int err;
2713 mdk_rdev_t *rdev;
2714 sector_t size;
2715
2716 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2717 if (!rdev) {
2718 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2719 return ERR_PTR(-ENOMEM);
2720 }
2721
2722 if ((err = alloc_disk_sb(rdev)))
2723 goto abort_free;
2724
2725 err = lock_rdev(rdev, newdev, super_format == -2);
2726 if (err)
2727 goto abort_free;
2728
2729 kobject_init(&rdev->kobj, &rdev_ktype);
2730
2731 rdev->desc_nr = -1;
2732 rdev->saved_raid_disk = -1;
2733 rdev->raid_disk = -1;
2734 rdev->flags = 0;
2735 rdev->data_offset = 0;
2736 rdev->sb_events = 0;
2737 rdev->last_read_error.tv_sec = 0;
2738 rdev->last_read_error.tv_nsec = 0;
2739 atomic_set(&rdev->nr_pending, 0);
2740 atomic_set(&rdev->read_errors, 0);
2741 atomic_set(&rdev->corrected_errors, 0);
2742
2743 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2744 if (!size) {
2745 printk(KERN_WARNING
2746 "md: %s has zero or unknown size, marking faulty!\n",
2747 bdevname(rdev->bdev,b));
2748 err = -EINVAL;
2749 goto abort_free;
2750 }
2751
2752 if (super_format >= 0) {
2753 err = super_types[super_format].
2754 load_super(rdev, NULL, super_minor);
2755 if (err == -EINVAL) {
2756 printk(KERN_WARNING
2757 "md: %s does not have a valid v%d.%d "
2758 "superblock, not importing!\n",
2759 bdevname(rdev->bdev,b),
2760 super_format, super_minor);
2761 goto abort_free;
2762 }
2763 if (err < 0) {
2764 printk(KERN_WARNING
2765 "md: could not read %s's sb, not importing!\n",
2766 bdevname(rdev->bdev,b));
2767 goto abort_free;
2768 }
2769 }
2770
2771 INIT_LIST_HEAD(&rdev->same_set);
2772 init_waitqueue_head(&rdev->blocked_wait);
2773
2774 return rdev;
2775
2776 abort_free:
2777 if (rdev->sb_page) {
2778 if (rdev->bdev)
2779 unlock_rdev(rdev);
2780 free_disk_sb(rdev);
2781 }
2782 kfree(rdev);
2783 return ERR_PTR(err);
2784 }
2785
2786 /*
2787 * Check a full RAID array for plausibility
2788 */
2789
2790
2791 static void analyze_sbs(mddev_t * mddev)
2792 {
2793 int i;
2794 mdk_rdev_t *rdev, *freshest, *tmp;
2795 char b[BDEVNAME_SIZE];
2796
2797 freshest = NULL;
2798 rdev_for_each(rdev, tmp, mddev)
2799 switch (super_types[mddev->major_version].
2800 load_super(rdev, freshest, mddev->minor_version)) {
2801 case 1:
2802 freshest = rdev;
2803 break;
2804 case 0:
2805 break;
2806 default:
2807 printk( KERN_ERR \
2808 "md: fatal superblock inconsistency in %s"
2809 " -- removing from array\n",
2810 bdevname(rdev->bdev,b));
2811 kick_rdev_from_array(rdev);
2812 }
2813
2814
2815 super_types[mddev->major_version].
2816 validate_super(mddev, freshest);
2817
2818 i = 0;
2819 rdev_for_each(rdev, tmp, mddev) {
2820 if (mddev->max_disks &&
2821 (rdev->desc_nr >= mddev->max_disks ||
2822 i > mddev->max_disks)) {
2823 printk(KERN_WARNING
2824 "md: %s: %s: only %d devices permitted\n",
2825 mdname(mddev), bdevname(rdev->bdev, b),
2826 mddev->max_disks);
2827 kick_rdev_from_array(rdev);
2828 continue;
2829 }
2830 if (rdev != freshest)
2831 if (super_types[mddev->major_version].
2832 validate_super(mddev, rdev)) {
2833 printk(KERN_WARNING "md: kicking non-fresh %s"
2834 " from array!\n",
2835 bdevname(rdev->bdev,b));
2836 kick_rdev_from_array(rdev);
2837 continue;
2838 }
2839 if (mddev->level == LEVEL_MULTIPATH) {
2840 rdev->desc_nr = i++;
2841 rdev->raid_disk = rdev->desc_nr;
2842 set_bit(In_sync, &rdev->flags);
2843 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2844 rdev->raid_disk = -1;
2845 clear_bit(In_sync, &rdev->flags);
2846 }
2847 }
2848 }
2849
2850 /* Read a fixed-point number.
2851 * Numbers in sysfs attributes should be in "standard" units where
2852 * possible, so time should be in seconds.
2853 * However we internally use a a much smaller unit such as
2854 * milliseconds or jiffies.
2855 * This function takes a decimal number with a possible fractional
2856 * component, and produces an integer which is the result of
2857 * multiplying that number by 10^'scale'.
2858 * all without any floating-point arithmetic.
2859 */
2860 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2861 {
2862 unsigned long result = 0;
2863 long decimals = -1;
2864 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2865 if (*cp == '.')
2866 decimals = 0;
2867 else if (decimals < scale) {
2868 unsigned int value;
2869 value = *cp - '0';
2870 result = result * 10 + value;
2871 if (decimals >= 0)
2872 decimals++;
2873 }
2874 cp++;
2875 }
2876 if (*cp == '\n')
2877 cp++;
2878 if (*cp)
2879 return -EINVAL;
2880 if (decimals < 0)
2881 decimals = 0;
2882 while (decimals < scale) {
2883 result *= 10;
2884 decimals ++;
2885 }
2886 *res = result;
2887 return 0;
2888 }
2889
2890
2891 static void md_safemode_timeout(unsigned long data);
2892
2893 static ssize_t
2894 safe_delay_show(mddev_t *mddev, char *page)
2895 {
2896 int msec = (mddev->safemode_delay*1000)/HZ;
2897 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2898 }
2899 static ssize_t
2900 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2901 {
2902 unsigned long msec;
2903
2904 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2905 return -EINVAL;
2906 if (msec == 0)
2907 mddev->safemode_delay = 0;
2908 else {
2909 unsigned long old_delay = mddev->safemode_delay;
2910 mddev->safemode_delay = (msec*HZ)/1000;
2911 if (mddev->safemode_delay == 0)
2912 mddev->safemode_delay = 1;
2913 if (mddev->safemode_delay < old_delay)
2914 md_safemode_timeout((unsigned long)mddev);
2915 }
2916 return len;
2917 }
2918 static struct md_sysfs_entry md_safe_delay =
2919 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2920
2921 static ssize_t
2922 level_show(mddev_t *mddev, char *page)
2923 {
2924 struct mdk_personality *p = mddev->pers;
2925 if (p)
2926 return sprintf(page, "%s\n", p->name);
2927 else if (mddev->clevel[0])
2928 return sprintf(page, "%s\n", mddev->clevel);
2929 else if (mddev->level != LEVEL_NONE)
2930 return sprintf(page, "%d\n", mddev->level);
2931 else
2932 return 0;
2933 }
2934
2935 static ssize_t
2936 level_store(mddev_t *mddev, const char *buf, size_t len)
2937 {
2938 char clevel[16];
2939 ssize_t rv = len;
2940 struct mdk_personality *pers;
2941 long level;
2942 void *priv;
2943 mdk_rdev_t *rdev;
2944
2945 if (mddev->pers == NULL) {
2946 if (len == 0)
2947 return 0;
2948 if (len >= sizeof(mddev->clevel))
2949 return -ENOSPC;
2950 strncpy(mddev->clevel, buf, len);
2951 if (mddev->clevel[len-1] == '\n')
2952 len--;
2953 mddev->clevel[len] = 0;
2954 mddev->level = LEVEL_NONE;
2955 return rv;
2956 }
2957
2958 /* request to change the personality. Need to ensure:
2959 * - array is not engaged in resync/recovery/reshape
2960 * - old personality can be suspended
2961 * - new personality will access other array.
2962 */
2963
2964 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2965 return -EBUSY;
2966
2967 if (!mddev->pers->quiesce) {
2968 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2969 mdname(mddev), mddev->pers->name);
2970 return -EINVAL;
2971 }
2972
2973 /* Now find the new personality */
2974 if (len == 0 || len >= sizeof(clevel))
2975 return -EINVAL;
2976 strncpy(clevel, buf, len);
2977 if (clevel[len-1] == '\n')
2978 len--;
2979 clevel[len] = 0;
2980 if (strict_strtol(clevel, 10, &level))
2981 level = LEVEL_NONE;
2982
2983 if (request_module("md-%s", clevel) != 0)
2984 request_module("md-level-%s", clevel);
2985 spin_lock(&pers_lock);
2986 pers = find_pers(level, clevel);
2987 if (!pers || !try_module_get(pers->owner)) {
2988 spin_unlock(&pers_lock);
2989 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2990 return -EINVAL;
2991 }
2992 spin_unlock(&pers_lock);
2993
2994 if (pers == mddev->pers) {
2995 /* Nothing to do! */
2996 module_put(pers->owner);
2997 return rv;
2998 }
2999 if (!pers->takeover) {
3000 module_put(pers->owner);
3001 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3002 mdname(mddev), clevel);
3003 return -EINVAL;
3004 }
3005
3006 list_for_each_entry(rdev, &mddev->disks, same_set)
3007 rdev->new_raid_disk = rdev->raid_disk;
3008
3009 /* ->takeover must set new_* and/or delta_disks
3010 * if it succeeds, and may set them when it fails.
3011 */
3012 priv = pers->takeover(mddev);
3013 if (IS_ERR(priv)) {
3014 mddev->new_level = mddev->level;
3015 mddev->new_layout = mddev->layout;
3016 mddev->new_chunk_sectors = mddev->chunk_sectors;
3017 mddev->raid_disks -= mddev->delta_disks;
3018 mddev->delta_disks = 0;
3019 module_put(pers->owner);
3020 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3021 mdname(mddev), clevel);
3022 return PTR_ERR(priv);
3023 }
3024
3025 /* Looks like we have a winner */
3026 mddev_suspend(mddev);
3027 mddev->pers->stop(mddev);
3028
3029 if (mddev->pers->sync_request == NULL &&
3030 pers->sync_request != NULL) {
3031 /* need to add the md_redundancy_group */
3032 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3033 printk(KERN_WARNING
3034 "md: cannot register extra attributes for %s\n",
3035 mdname(mddev));
3036 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3037 }
3038 if (mddev->pers->sync_request != NULL &&
3039 pers->sync_request == NULL) {
3040 /* need to remove the md_redundancy_group */
3041 if (mddev->to_remove == NULL)
3042 mddev->to_remove = &md_redundancy_group;
3043 }
3044
3045 if (mddev->pers->sync_request == NULL &&
3046 mddev->external) {
3047 /* We are converting from a no-redundancy array
3048 * to a redundancy array and metadata is managed
3049 * externally so we need to be sure that writes
3050 * won't block due to a need to transition
3051 * clean->dirty
3052 * until external management is started.
3053 */
3054 mddev->in_sync = 0;
3055 mddev->safemode_delay = 0;
3056 mddev->safemode = 0;
3057 }
3058
3059 list_for_each_entry(rdev, &mddev->disks, same_set) {
3060 char nm[20];
3061 if (rdev->raid_disk < 0)
3062 continue;
3063 if (rdev->new_raid_disk > mddev->raid_disks)
3064 rdev->new_raid_disk = -1;
3065 if (rdev->new_raid_disk == rdev->raid_disk)
3066 continue;
3067 sprintf(nm, "rd%d", rdev->raid_disk);
3068 sysfs_remove_link(&mddev->kobj, nm);
3069 }
3070 list_for_each_entry(rdev, &mddev->disks, same_set) {
3071 if (rdev->raid_disk < 0)
3072 continue;
3073 if (rdev->new_raid_disk == rdev->raid_disk)
3074 continue;
3075 rdev->raid_disk = rdev->new_raid_disk;
3076 if (rdev->raid_disk < 0)
3077 clear_bit(In_sync, &rdev->flags);
3078 else {
3079 char nm[20];
3080 sprintf(nm, "rd%d", rdev->raid_disk);
3081 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3082 printk("md: cannot register %s for %s after level change\n",
3083 nm, mdname(mddev));
3084 }
3085 }
3086
3087 module_put(mddev->pers->owner);
3088 mddev->pers = pers;
3089 mddev->private = priv;
3090 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3091 mddev->level = mddev->new_level;
3092 mddev->layout = mddev->new_layout;
3093 mddev->chunk_sectors = mddev->new_chunk_sectors;
3094 mddev->delta_disks = 0;
3095 if (mddev->pers->sync_request == NULL) {
3096 /* this is now an array without redundancy, so
3097 * it must always be in_sync
3098 */
3099 mddev->in_sync = 1;
3100 del_timer_sync(&mddev->safemode_timer);
3101 }
3102 pers->run(mddev);
3103 mddev_resume(mddev);
3104 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3105 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3106 md_wakeup_thread(mddev->thread);
3107 sysfs_notify(&mddev->kobj, NULL, "level");
3108 md_new_event(mddev);
3109 return rv;
3110 }
3111
3112 static struct md_sysfs_entry md_level =
3113 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3114
3115
3116 static ssize_t
3117 layout_show(mddev_t *mddev, char *page)
3118 {
3119 /* just a number, not meaningful for all levels */
3120 if (mddev->reshape_position != MaxSector &&
3121 mddev->layout != mddev->new_layout)
3122 return sprintf(page, "%d (%d)\n",
3123 mddev->new_layout, mddev->layout);
3124 return sprintf(page, "%d\n", mddev->layout);
3125 }
3126
3127 static ssize_t
3128 layout_store(mddev_t *mddev, const char *buf, size_t len)
3129 {
3130 char *e;
3131 unsigned long n = simple_strtoul(buf, &e, 10);
3132
3133 if (!*buf || (*e && *e != '\n'))
3134 return -EINVAL;
3135
3136 if (mddev->pers) {
3137 int err;
3138 if (mddev->pers->check_reshape == NULL)
3139 return -EBUSY;
3140 mddev->new_layout = n;
3141 err = mddev->pers->check_reshape(mddev);
3142 if (err) {
3143 mddev->new_layout = mddev->layout;
3144 return err;
3145 }
3146 } else {
3147 mddev->new_layout = n;
3148 if (mddev->reshape_position == MaxSector)
3149 mddev->layout = n;
3150 }
3151 return len;
3152 }
3153 static struct md_sysfs_entry md_layout =
3154 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3155
3156
3157 static ssize_t
3158 raid_disks_show(mddev_t *mddev, char *page)
3159 {
3160 if (mddev->raid_disks == 0)
3161 return 0;
3162 if (mddev->reshape_position != MaxSector &&
3163 mddev->delta_disks != 0)
3164 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3165 mddev->raid_disks - mddev->delta_disks);
3166 return sprintf(page, "%d\n", mddev->raid_disks);
3167 }
3168
3169 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3170
3171 static ssize_t
3172 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3173 {
3174 char *e;
3175 int rv = 0;
3176 unsigned long n = simple_strtoul(buf, &e, 10);
3177
3178 if (!*buf || (*e && *e != '\n'))
3179 return -EINVAL;
3180
3181 if (mddev->pers)
3182 rv = update_raid_disks(mddev, n);
3183 else if (mddev->reshape_position != MaxSector) {
3184 int olddisks = mddev->raid_disks - mddev->delta_disks;
3185 mddev->delta_disks = n - olddisks;
3186 mddev->raid_disks = n;
3187 } else
3188 mddev->raid_disks = n;
3189 return rv ? rv : len;
3190 }
3191 static struct md_sysfs_entry md_raid_disks =
3192 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3193
3194 static ssize_t
3195 chunk_size_show(mddev_t *mddev, char *page)
3196 {
3197 if (mddev->reshape_position != MaxSector &&
3198 mddev->chunk_sectors != mddev->new_chunk_sectors)
3199 return sprintf(page, "%d (%d)\n",
3200 mddev->new_chunk_sectors << 9,
3201 mddev->chunk_sectors << 9);
3202 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3203 }
3204
3205 static ssize_t
3206 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3207 {
3208 char *e;
3209 unsigned long n = simple_strtoul(buf, &e, 10);
3210
3211 if (!*buf || (*e && *e != '\n'))
3212 return -EINVAL;
3213
3214 if (mddev->pers) {
3215 int err;
3216 if (mddev->pers->check_reshape == NULL)
3217 return -EBUSY;
3218 mddev->new_chunk_sectors = n >> 9;
3219 err = mddev->pers->check_reshape(mddev);
3220 if (err) {
3221 mddev->new_chunk_sectors = mddev->chunk_sectors;
3222 return err;
3223 }
3224 } else {
3225 mddev->new_chunk_sectors = n >> 9;
3226 if (mddev->reshape_position == MaxSector)
3227 mddev->chunk_sectors = n >> 9;
3228 }
3229 return len;
3230 }
3231 static struct md_sysfs_entry md_chunk_size =
3232 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3233
3234 static ssize_t
3235 resync_start_show(mddev_t *mddev, char *page)
3236 {
3237 if (mddev->recovery_cp == MaxSector)
3238 return sprintf(page, "none\n");
3239 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3240 }
3241
3242 static ssize_t
3243 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3244 {
3245 char *e;
3246 unsigned long long n = simple_strtoull(buf, &e, 10);
3247
3248 if (mddev->pers)
3249 return -EBUSY;
3250 if (cmd_match(buf, "none"))
3251 n = MaxSector;
3252 else if (!*buf || (*e && *e != '\n'))
3253 return -EINVAL;
3254
3255 mddev->recovery_cp = n;
3256 return len;
3257 }
3258 static struct md_sysfs_entry md_resync_start =
3259 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3260
3261 /*
3262 * The array state can be:
3263 *
3264 * clear
3265 * No devices, no size, no level
3266 * Equivalent to STOP_ARRAY ioctl
3267 * inactive
3268 * May have some settings, but array is not active
3269 * all IO results in error
3270 * When written, doesn't tear down array, but just stops it
3271 * suspended (not supported yet)
3272 * All IO requests will block. The array can be reconfigured.
3273 * Writing this, if accepted, will block until array is quiescent
3274 * readonly
3275 * no resync can happen. no superblocks get written.
3276 * write requests fail
3277 * read-auto
3278 * like readonly, but behaves like 'clean' on a write request.
3279 *
3280 * clean - no pending writes, but otherwise active.
3281 * When written to inactive array, starts without resync
3282 * If a write request arrives then
3283 * if metadata is known, mark 'dirty' and switch to 'active'.
3284 * if not known, block and switch to write-pending
3285 * If written to an active array that has pending writes, then fails.
3286 * active
3287 * fully active: IO and resync can be happening.
3288 * When written to inactive array, starts with resync
3289 *
3290 * write-pending
3291 * clean, but writes are blocked waiting for 'active' to be written.
3292 *
3293 * active-idle
3294 * like active, but no writes have been seen for a while (100msec).
3295 *
3296 */
3297 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3298 write_pending, active_idle, bad_word};
3299 static char *array_states[] = {
3300 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3301 "write-pending", "active-idle", NULL };
3302
3303 static int match_word(const char *word, char **list)
3304 {
3305 int n;
3306 for (n=0; list[n]; n++)
3307 if (cmd_match(word, list[n]))
3308 break;
3309 return n;
3310 }
3311
3312 static ssize_t
3313 array_state_show(mddev_t *mddev, char *page)
3314 {
3315 enum array_state st = inactive;
3316
3317 if (mddev->pers)
3318 switch(mddev->ro) {
3319 case 1:
3320 st = readonly;
3321 break;
3322 case 2:
3323 st = read_auto;
3324 break;
3325 case 0:
3326 if (mddev->in_sync)
3327 st = clean;
3328 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3329 st = write_pending;
3330 else if (mddev->safemode)
3331 st = active_idle;
3332 else
3333 st = active;
3334 }
3335 else {
3336 if (list_empty(&mddev->disks) &&
3337 mddev->raid_disks == 0 &&
3338 mddev->dev_sectors == 0)
3339 st = clear;
3340 else
3341 st = inactive;
3342 }
3343 return sprintf(page, "%s\n", array_states[st]);
3344 }
3345
3346 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3347 static int md_set_readonly(mddev_t * mddev, int is_open);
3348 static int do_md_run(mddev_t * mddev);
3349 static int restart_array(mddev_t *mddev);
3350
3351 static ssize_t
3352 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3353 {
3354 int err = -EINVAL;
3355 enum array_state st = match_word(buf, array_states);
3356 switch(st) {
3357 case bad_word:
3358 break;
3359 case clear:
3360 /* stopping an active array */
3361 if (atomic_read(&mddev->openers) > 0)
3362 return -EBUSY;
3363 err = do_md_stop(mddev, 0, 0);
3364 break;
3365 case inactive:
3366 /* stopping an active array */
3367 if (mddev->pers) {
3368 if (atomic_read(&mddev->openers) > 0)
3369 return -EBUSY;
3370 err = do_md_stop(mddev, 2, 0);
3371 } else
3372 err = 0; /* already inactive */
3373 break;
3374 case suspended:
3375 break; /* not supported yet */
3376 case readonly:
3377 if (mddev->pers)
3378 err = md_set_readonly(mddev, 0);
3379 else {
3380 mddev->ro = 1;
3381 set_disk_ro(mddev->gendisk, 1);
3382 err = do_md_run(mddev);
3383 }
3384 break;
3385 case read_auto:
3386 if (mddev->pers) {
3387 if (mddev->ro == 0)
3388 err = md_set_readonly(mddev, 0);
3389 else if (mddev->ro == 1)
3390 err = restart_array(mddev);
3391 if (err == 0) {
3392 mddev->ro = 2;
3393 set_disk_ro(mddev->gendisk, 0);
3394 }
3395 } else {
3396 mddev->ro = 2;
3397 err = do_md_run(mddev);
3398 }
3399 break;
3400 case clean:
3401 if (mddev->pers) {
3402 restart_array(mddev);
3403 spin_lock_irq(&mddev->write_lock);
3404 if (atomic_read(&mddev->writes_pending) == 0) {
3405 if (mddev->in_sync == 0) {
3406 mddev->in_sync = 1;
3407 if (mddev->safemode == 1)
3408 mddev->safemode = 0;
3409 if (mddev->persistent)
3410 set_bit(MD_CHANGE_CLEAN,
3411 &mddev->flags);
3412 }
3413 err = 0;
3414 } else
3415 err = -EBUSY;
3416 spin_unlock_irq(&mddev->write_lock);
3417 } else
3418 err = -EINVAL;
3419 break;
3420 case active:
3421 if (mddev->pers) {
3422 restart_array(mddev);
3423 if (mddev->external)
3424 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3425 wake_up(&mddev->sb_wait);
3426 err = 0;
3427 } else {
3428 mddev->ro = 0;
3429 set_disk_ro(mddev->gendisk, 0);
3430 err = do_md_run(mddev);
3431 }
3432 break;
3433 case write_pending:
3434 case active_idle:
3435 /* these cannot be set */
3436 break;
3437 }
3438 if (err)
3439 return err;
3440 else {
3441 sysfs_notify_dirent(mddev->sysfs_state);
3442 return len;
3443 }
3444 }
3445 static struct md_sysfs_entry md_array_state =
3446 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3447
3448 static ssize_t
3449 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3450 return sprintf(page, "%d\n",
3451 atomic_read(&mddev->max_corr_read_errors));
3452 }
3453
3454 static ssize_t
3455 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3456 {
3457 char *e;
3458 unsigned long n = simple_strtoul(buf, &e, 10);
3459
3460 if (*buf && (*e == 0 || *e == '\n')) {
3461 atomic_set(&mddev->max_corr_read_errors, n);
3462 return len;
3463 }
3464 return -EINVAL;
3465 }
3466
3467 static struct md_sysfs_entry max_corr_read_errors =
3468 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3469 max_corrected_read_errors_store);
3470
3471 static ssize_t
3472 null_show(mddev_t *mddev, char *page)
3473 {
3474 return -EINVAL;
3475 }
3476
3477 static ssize_t
3478 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3479 {
3480 /* buf must be %d:%d\n? giving major and minor numbers */
3481 /* The new device is added to the array.
3482 * If the array has a persistent superblock, we read the
3483 * superblock to initialise info and check validity.
3484 * Otherwise, only checking done is that in bind_rdev_to_array,
3485 * which mainly checks size.
3486 */
3487 char *e;
3488 int major = simple_strtoul(buf, &e, 10);
3489 int minor;
3490 dev_t dev;
3491 mdk_rdev_t *rdev;
3492 int err;
3493
3494 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3495 return -EINVAL;
3496 minor = simple_strtoul(e+1, &e, 10);
3497 if (*e && *e != '\n')
3498 return -EINVAL;
3499 dev = MKDEV(major, minor);
3500 if (major != MAJOR(dev) ||
3501 minor != MINOR(dev))
3502 return -EOVERFLOW;
3503
3504
3505 if (mddev->persistent) {
3506 rdev = md_import_device(dev, mddev->major_version,
3507 mddev->minor_version);
3508 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3509 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3510 mdk_rdev_t, same_set);
3511 err = super_types[mddev->major_version]
3512 .load_super(rdev, rdev0, mddev->minor_version);
3513 if (err < 0)
3514 goto out;
3515 }
3516 } else if (mddev->external)
3517 rdev = md_import_device(dev, -2, -1);
3518 else
3519 rdev = md_import_device(dev, -1, -1);
3520
3521 if (IS_ERR(rdev))
3522 return PTR_ERR(rdev);
3523 err = bind_rdev_to_array(rdev, mddev);
3524 out:
3525 if (err)
3526 export_rdev(rdev);
3527 return err ? err : len;
3528 }
3529
3530 static struct md_sysfs_entry md_new_device =
3531 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3532
3533 static ssize_t
3534 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3535 {
3536 char *end;
3537 unsigned long chunk, end_chunk;
3538
3539 if (!mddev->bitmap)
3540 goto out;
3541 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3542 while (*buf) {
3543 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3544 if (buf == end) break;
3545 if (*end == '-') { /* range */
3546 buf = end + 1;
3547 end_chunk = simple_strtoul(buf, &end, 0);
3548 if (buf == end) break;
3549 }
3550 if (*end && !isspace(*end)) break;
3551 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3552 buf = skip_spaces(end);
3553 }
3554 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3555 out:
3556 return len;
3557 }
3558
3559 static struct md_sysfs_entry md_bitmap =
3560 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3561
3562 static ssize_t
3563 size_show(mddev_t *mddev, char *page)
3564 {
3565 return sprintf(page, "%llu\n",
3566 (unsigned long long)mddev->dev_sectors / 2);
3567 }
3568
3569 static int update_size(mddev_t *mddev, sector_t num_sectors);
3570
3571 static ssize_t
3572 size_store(mddev_t *mddev, const char *buf, size_t len)
3573 {
3574 /* If array is inactive, we can reduce the component size, but
3575 * not increase it (except from 0).
3576 * If array is active, we can try an on-line resize
3577 */
3578 sector_t sectors;
3579 int err = strict_blocks_to_sectors(buf, &sectors);
3580
3581 if (err < 0)
3582 return err;
3583 if (mddev->pers) {
3584 err = update_size(mddev, sectors);
3585 md_update_sb(mddev, 1);
3586 } else {
3587 if (mddev->dev_sectors == 0 ||
3588 mddev->dev_sectors > sectors)
3589 mddev->dev_sectors = sectors;
3590 else
3591 err = -ENOSPC;
3592 }
3593 return err ? err : len;
3594 }
3595
3596 static struct md_sysfs_entry md_size =
3597 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3598
3599
3600 /* Metdata version.
3601 * This is one of
3602 * 'none' for arrays with no metadata (good luck...)
3603 * 'external' for arrays with externally managed metadata,
3604 * or N.M for internally known formats
3605 */
3606 static ssize_t
3607 metadata_show(mddev_t *mddev, char *page)
3608 {
3609 if (mddev->persistent)
3610 return sprintf(page, "%d.%d\n",
3611 mddev->major_version, mddev->minor_version);
3612 else if (mddev->external)
3613 return sprintf(page, "external:%s\n", mddev->metadata_type);
3614 else
3615 return sprintf(page, "none\n");
3616 }
3617
3618 static ssize_t
3619 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3620 {
3621 int major, minor;
3622 char *e;
3623 /* Changing the details of 'external' metadata is
3624 * always permitted. Otherwise there must be
3625 * no devices attached to the array.
3626 */
3627 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3628 ;
3629 else if (!list_empty(&mddev->disks))
3630 return -EBUSY;
3631
3632 if (cmd_match(buf, "none")) {
3633 mddev->persistent = 0;
3634 mddev->external = 0;
3635 mddev->major_version = 0;
3636 mddev->minor_version = 90;
3637 return len;
3638 }
3639 if (strncmp(buf, "external:", 9) == 0) {
3640 size_t namelen = len-9;
3641 if (namelen >= sizeof(mddev->metadata_type))
3642 namelen = sizeof(mddev->metadata_type)-1;
3643 strncpy(mddev->metadata_type, buf+9, namelen);
3644 mddev->metadata_type[namelen] = 0;
3645 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3646 mddev->metadata_type[--namelen] = 0;
3647 mddev->persistent = 0;
3648 mddev->external = 1;
3649 mddev->major_version = 0;
3650 mddev->minor_version = 90;
3651 return len;
3652 }
3653 major = simple_strtoul(buf, &e, 10);
3654 if (e==buf || *e != '.')
3655 return -EINVAL;
3656 buf = e+1;
3657 minor = simple_strtoul(buf, &e, 10);
3658 if (e==buf || (*e && *e != '\n') )
3659 return -EINVAL;
3660 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3661 return -ENOENT;
3662 mddev->major_version = major;
3663 mddev->minor_version = minor;
3664 mddev->persistent = 1;
3665 mddev->external = 0;
3666 return len;
3667 }
3668
3669 static struct md_sysfs_entry md_metadata =
3670 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3671
3672 static ssize_t
3673 action_show(mddev_t *mddev, char *page)
3674 {
3675 char *type = "idle";
3676 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3677 type = "frozen";
3678 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3679 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3680 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3681 type = "reshape";
3682 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3683 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3684 type = "resync";
3685 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3686 type = "check";
3687 else
3688 type = "repair";
3689 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3690 type = "recover";
3691 }
3692 return sprintf(page, "%s\n", type);
3693 }
3694
3695 static ssize_t
3696 action_store(mddev_t *mddev, const char *page, size_t len)
3697 {
3698 if (!mddev->pers || !mddev->pers->sync_request)
3699 return -EINVAL;
3700
3701 if (cmd_match(page, "frozen"))
3702 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3703 else
3704 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3705
3706 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3707 if (mddev->sync_thread) {
3708 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3709 md_unregister_thread(mddev->sync_thread);
3710 mddev->sync_thread = NULL;
3711 mddev->recovery = 0;
3712 }
3713 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3714 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3715 return -EBUSY;
3716 else if (cmd_match(page, "resync"))
3717 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3718 else if (cmd_match(page, "recover")) {
3719 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3720 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3721 } else if (cmd_match(page, "reshape")) {
3722 int err;
3723 if (mddev->pers->start_reshape == NULL)
3724 return -EINVAL;
3725 err = mddev->pers->start_reshape(mddev);
3726 if (err)
3727 return err;
3728 sysfs_notify(&mddev->kobj, NULL, "degraded");
3729 } else {
3730 if (cmd_match(page, "check"))
3731 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3732 else if (!cmd_match(page, "repair"))
3733 return -EINVAL;
3734 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3735 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3736 }
3737 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3738 md_wakeup_thread(mddev->thread);
3739 sysfs_notify_dirent(mddev->sysfs_action);
3740 return len;
3741 }
3742
3743 static ssize_t
3744 mismatch_cnt_show(mddev_t *mddev, char *page)
3745 {
3746 return sprintf(page, "%llu\n",
3747 (unsigned long long) mddev->resync_mismatches);
3748 }
3749
3750 static struct md_sysfs_entry md_scan_mode =
3751 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3752
3753
3754 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3755
3756 static ssize_t
3757 sync_min_show(mddev_t *mddev, char *page)
3758 {
3759 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3760 mddev->sync_speed_min ? "local": "system");
3761 }
3762
3763 static ssize_t
3764 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3765 {
3766 int min;
3767 char *e;
3768 if (strncmp(buf, "system", 6)==0) {
3769 mddev->sync_speed_min = 0;
3770 return len;
3771 }
3772 min = simple_strtoul(buf, &e, 10);
3773 if (buf == e || (*e && *e != '\n') || min <= 0)
3774 return -EINVAL;
3775 mddev->sync_speed_min = min;
3776 return len;
3777 }
3778
3779 static struct md_sysfs_entry md_sync_min =
3780 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3781
3782 static ssize_t
3783 sync_max_show(mddev_t *mddev, char *page)
3784 {
3785 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3786 mddev->sync_speed_max ? "local": "system");
3787 }
3788
3789 static ssize_t
3790 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3791 {
3792 int max;
3793 char *e;
3794 if (strncmp(buf, "system", 6)==0) {
3795 mddev->sync_speed_max = 0;
3796 return len;
3797 }
3798 max = simple_strtoul(buf, &e, 10);
3799 if (buf == e || (*e && *e != '\n') || max <= 0)
3800 return -EINVAL;
3801 mddev->sync_speed_max = max;
3802 return len;
3803 }
3804
3805 static struct md_sysfs_entry md_sync_max =
3806 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3807
3808 static ssize_t
3809 degraded_show(mddev_t *mddev, char *page)
3810 {
3811 return sprintf(page, "%d\n", mddev->degraded);
3812 }
3813 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3814
3815 static ssize_t
3816 sync_force_parallel_show(mddev_t *mddev, char *page)
3817 {
3818 return sprintf(page, "%d\n", mddev->parallel_resync);
3819 }
3820
3821 static ssize_t
3822 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3823 {
3824 long n;
3825
3826 if (strict_strtol(buf, 10, &n))
3827 return -EINVAL;
3828
3829 if (n != 0 && n != 1)
3830 return -EINVAL;
3831
3832 mddev->parallel_resync = n;
3833
3834 if (mddev->sync_thread)
3835 wake_up(&resync_wait);
3836
3837 return len;
3838 }
3839
3840 /* force parallel resync, even with shared block devices */
3841 static struct md_sysfs_entry md_sync_force_parallel =
3842 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3843 sync_force_parallel_show, sync_force_parallel_store);
3844
3845 static ssize_t
3846 sync_speed_show(mddev_t *mddev, char *page)
3847 {
3848 unsigned long resync, dt, db;
3849 if (mddev->curr_resync == 0)
3850 return sprintf(page, "none\n");
3851 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3852 dt = (jiffies - mddev->resync_mark) / HZ;
3853 if (!dt) dt++;
3854 db = resync - mddev->resync_mark_cnt;
3855 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3856 }
3857
3858 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3859
3860 static ssize_t
3861 sync_completed_show(mddev_t *mddev, char *page)
3862 {
3863 unsigned long max_sectors, resync;
3864
3865 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3866 return sprintf(page, "none\n");
3867
3868 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3869 max_sectors = mddev->resync_max_sectors;
3870 else
3871 max_sectors = mddev->dev_sectors;
3872
3873 resync = mddev->curr_resync_completed;
3874 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3875 }
3876
3877 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3878
3879 static ssize_t
3880 min_sync_show(mddev_t *mddev, char *page)
3881 {
3882 return sprintf(page, "%llu\n",
3883 (unsigned long long)mddev->resync_min);
3884 }
3885 static ssize_t
3886 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3887 {
3888 unsigned long long min;
3889 if (strict_strtoull(buf, 10, &min))
3890 return -EINVAL;
3891 if (min > mddev->resync_max)
3892 return -EINVAL;
3893 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3894 return -EBUSY;
3895
3896 /* Must be a multiple of chunk_size */
3897 if (mddev->chunk_sectors) {
3898 sector_t temp = min;
3899 if (sector_div(temp, mddev->chunk_sectors))
3900 return -EINVAL;
3901 }
3902 mddev->resync_min = min;
3903
3904 return len;
3905 }
3906
3907 static struct md_sysfs_entry md_min_sync =
3908 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3909
3910 static ssize_t
3911 max_sync_show(mddev_t *mddev, char *page)
3912 {
3913 if (mddev->resync_max == MaxSector)
3914 return sprintf(page, "max\n");
3915 else
3916 return sprintf(page, "%llu\n",
3917 (unsigned long long)mddev->resync_max);
3918 }
3919 static ssize_t
3920 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3921 {
3922 if (strncmp(buf, "max", 3) == 0)
3923 mddev->resync_max = MaxSector;
3924 else {
3925 unsigned long long max;
3926 if (strict_strtoull(buf, 10, &max))
3927 return -EINVAL;
3928 if (max < mddev->resync_min)
3929 return -EINVAL;
3930 if (max < mddev->resync_max &&
3931 mddev->ro == 0 &&
3932 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3933 return -EBUSY;
3934
3935 /* Must be a multiple of chunk_size */
3936 if (mddev->chunk_sectors) {
3937 sector_t temp = max;
3938 if (sector_div(temp, mddev->chunk_sectors))
3939 return -EINVAL;
3940 }
3941 mddev->resync_max = max;
3942 }
3943 wake_up(&mddev->recovery_wait);
3944 return len;
3945 }
3946
3947 static struct md_sysfs_entry md_max_sync =
3948 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3949
3950 static ssize_t
3951 suspend_lo_show(mddev_t *mddev, char *page)
3952 {
3953 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3954 }
3955
3956 static ssize_t
3957 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3958 {
3959 char *e;
3960 unsigned long long new = simple_strtoull(buf, &e, 10);
3961
3962 if (mddev->pers == NULL ||
3963 mddev->pers->quiesce == NULL)
3964 return -EINVAL;
3965 if (buf == e || (*e && *e != '\n'))
3966 return -EINVAL;
3967 if (new >= mddev->suspend_hi ||
3968 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3969 mddev->suspend_lo = new;
3970 mddev->pers->quiesce(mddev, 2);
3971 return len;
3972 } else
3973 return -EINVAL;
3974 }
3975 static struct md_sysfs_entry md_suspend_lo =
3976 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3977
3978
3979 static ssize_t
3980 suspend_hi_show(mddev_t *mddev, char *page)
3981 {
3982 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3983 }
3984
3985 static ssize_t
3986 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3987 {
3988 char *e;
3989 unsigned long long new = simple_strtoull(buf, &e, 10);
3990
3991 if (mddev->pers == NULL ||
3992 mddev->pers->quiesce == NULL)
3993 return -EINVAL;
3994 if (buf == e || (*e && *e != '\n'))
3995 return -EINVAL;
3996 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3997 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3998 mddev->suspend_hi = new;
3999 mddev->pers->quiesce(mddev, 1);
4000 mddev->pers->quiesce(mddev, 0);
4001 return len;
4002 } else
4003 return -EINVAL;
4004 }
4005 static struct md_sysfs_entry md_suspend_hi =
4006 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4007
4008 static ssize_t
4009 reshape_position_show(mddev_t *mddev, char *page)
4010 {
4011 if (mddev->reshape_position != MaxSector)
4012 return sprintf(page, "%llu\n",
4013 (unsigned long long)mddev->reshape_position);
4014 strcpy(page, "none\n");
4015 return 5;
4016 }
4017
4018 static ssize_t
4019 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4020 {
4021 char *e;
4022 unsigned long long new = simple_strtoull(buf, &e, 10);
4023 if (mddev->pers)
4024 return -EBUSY;
4025 if (buf == e || (*e && *e != '\n'))
4026 return -EINVAL;
4027 mddev->reshape_position = new;
4028 mddev->delta_disks = 0;
4029 mddev->new_level = mddev->level;
4030 mddev->new_layout = mddev->layout;
4031 mddev->new_chunk_sectors = mddev->chunk_sectors;
4032 return len;
4033 }
4034
4035 static struct md_sysfs_entry md_reshape_position =
4036 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4037 reshape_position_store);
4038
4039 static ssize_t
4040 array_size_show(mddev_t *mddev, char *page)
4041 {
4042 if (mddev->external_size)
4043 return sprintf(page, "%llu\n",
4044 (unsigned long long)mddev->array_sectors/2);
4045 else
4046 return sprintf(page, "default\n");
4047 }
4048
4049 static ssize_t
4050 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4051 {
4052 sector_t sectors;
4053
4054 if (strncmp(buf, "default", 7) == 0) {
4055 if (mddev->pers)
4056 sectors = mddev->pers->size(mddev, 0, 0);
4057 else
4058 sectors = mddev->array_sectors;
4059
4060 mddev->external_size = 0;
4061 } else {
4062 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4063 return -EINVAL;
4064 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4065 return -E2BIG;
4066
4067 mddev->external_size = 1;
4068 }
4069
4070 mddev->array_sectors = sectors;
4071 set_capacity(mddev->gendisk, mddev->array_sectors);
4072 if (mddev->pers)
4073 revalidate_disk(mddev->gendisk);
4074
4075 return len;
4076 }
4077
4078 static struct md_sysfs_entry md_array_size =
4079 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4080 array_size_store);
4081
4082 static struct attribute *md_default_attrs[] = {
4083 &md_level.attr,
4084 &md_layout.attr,
4085 &md_raid_disks.attr,
4086 &md_chunk_size.attr,
4087 &md_size.attr,
4088 &md_resync_start.attr,
4089 &md_metadata.attr,
4090 &md_new_device.attr,
4091 &md_safe_delay.attr,
4092 &md_array_state.attr,
4093 &md_reshape_position.attr,
4094 &md_array_size.attr,
4095 &max_corr_read_errors.attr,
4096 NULL,
4097 };
4098
4099 static struct attribute *md_redundancy_attrs[] = {
4100 &md_scan_mode.attr,
4101 &md_mismatches.attr,
4102 &md_sync_min.attr,
4103 &md_sync_max.attr,
4104 &md_sync_speed.attr,
4105 &md_sync_force_parallel.attr,
4106 &md_sync_completed.attr,
4107 &md_min_sync.attr,
4108 &md_max_sync.attr,
4109 &md_suspend_lo.attr,
4110 &md_suspend_hi.attr,
4111 &md_bitmap.attr,
4112 &md_degraded.attr,
4113 NULL,
4114 };
4115 static struct attribute_group md_redundancy_group = {
4116 .name = NULL,
4117 .attrs = md_redundancy_attrs,
4118 };
4119
4120
4121 static ssize_t
4122 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4123 {
4124 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4125 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4126 ssize_t rv;
4127
4128 if (!entry->show)
4129 return -EIO;
4130 rv = mddev_lock(mddev);
4131 if (!rv) {
4132 rv = entry->show(mddev, page);
4133 mddev_unlock(mddev);
4134 }
4135 return rv;
4136 }
4137
4138 static ssize_t
4139 md_attr_store(struct kobject *kobj, struct attribute *attr,
4140 const char *page, size_t length)
4141 {
4142 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4143 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4144 ssize_t rv;
4145
4146 if (!entry->store)
4147 return -EIO;
4148 if (!capable(CAP_SYS_ADMIN))
4149 return -EACCES;
4150 rv = mddev_lock(mddev);
4151 if (mddev->hold_active == UNTIL_IOCTL)
4152 mddev->hold_active = 0;
4153 if (!rv) {
4154 rv = entry->store(mddev, page, length);
4155 mddev_unlock(mddev);
4156 }
4157 return rv;
4158 }
4159
4160 static void md_free(struct kobject *ko)
4161 {
4162 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4163
4164 if (mddev->sysfs_state)
4165 sysfs_put(mddev->sysfs_state);
4166
4167 if (mddev->gendisk) {
4168 del_gendisk(mddev->gendisk);
4169 put_disk(mddev->gendisk);
4170 }
4171 if (mddev->queue)
4172 blk_cleanup_queue(mddev->queue);
4173
4174 kfree(mddev);
4175 }
4176
4177 static const struct sysfs_ops md_sysfs_ops = {
4178 .show = md_attr_show,
4179 .store = md_attr_store,
4180 };
4181 static struct kobj_type md_ktype = {
4182 .release = md_free,
4183 .sysfs_ops = &md_sysfs_ops,
4184 .default_attrs = md_default_attrs,
4185 };
4186
4187 int mdp_major = 0;
4188
4189 static void mddev_delayed_delete(struct work_struct *ws)
4190 {
4191 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4192
4193 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4194 kobject_del(&mddev->kobj);
4195 kobject_put(&mddev->kobj);
4196 }
4197
4198 static int md_alloc(dev_t dev, char *name)
4199 {
4200 static DEFINE_MUTEX(disks_mutex);
4201 mddev_t *mddev = mddev_find(dev);
4202 struct gendisk *disk;
4203 int partitioned;
4204 int shift;
4205 int unit;
4206 int error;
4207
4208 if (!mddev)
4209 return -ENODEV;
4210
4211 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4212 shift = partitioned ? MdpMinorShift : 0;
4213 unit = MINOR(mddev->unit) >> shift;
4214
4215 /* wait for any previous instance if this device
4216 * to be completed removed (mddev_delayed_delete).
4217 */
4218 flush_scheduled_work();
4219
4220 mutex_lock(&disks_mutex);
4221 error = -EEXIST;
4222 if (mddev->gendisk)
4223 goto abort;
4224
4225 if (name) {
4226 /* Need to ensure that 'name' is not a duplicate.
4227 */
4228 mddev_t *mddev2;
4229 spin_lock(&all_mddevs_lock);
4230
4231 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4232 if (mddev2->gendisk &&
4233 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4234 spin_unlock(&all_mddevs_lock);
4235 goto abort;
4236 }
4237 spin_unlock(&all_mddevs_lock);
4238 }
4239
4240 error = -ENOMEM;
4241 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4242 if (!mddev->queue)
4243 goto abort;
4244 mddev->queue->queuedata = mddev;
4245
4246 /* Can be unlocked because the queue is new: no concurrency */
4247 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4248
4249 blk_queue_make_request(mddev->queue, md_make_request);
4250
4251 disk = alloc_disk(1 << shift);
4252 if (!disk) {
4253 blk_cleanup_queue(mddev->queue);
4254 mddev->queue = NULL;
4255 goto abort;
4256 }
4257 disk->major = MAJOR(mddev->unit);
4258 disk->first_minor = unit << shift;
4259 if (name)
4260 strcpy(disk->disk_name, name);
4261 else if (partitioned)
4262 sprintf(disk->disk_name, "md_d%d", unit);
4263 else
4264 sprintf(disk->disk_name, "md%d", unit);
4265 disk->fops = &md_fops;
4266 disk->private_data = mddev;
4267 disk->queue = mddev->queue;
4268 /* Allow extended partitions. This makes the
4269 * 'mdp' device redundant, but we can't really
4270 * remove it now.
4271 */
4272 disk->flags |= GENHD_FL_EXT_DEVT;
4273 add_disk(disk);
4274 mddev->gendisk = disk;
4275 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4276 &disk_to_dev(disk)->kobj, "%s", "md");
4277 if (error) {
4278 /* This isn't possible, but as kobject_init_and_add is marked
4279 * __must_check, we must do something with the result
4280 */
4281 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4282 disk->disk_name);
4283 error = 0;
4284 }
4285 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4286 printk(KERN_DEBUG "pointless warning\n");
4287 abort:
4288 mutex_unlock(&disks_mutex);
4289 if (!error) {
4290 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4291 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, NULL, "array_state");
4292 }
4293 mddev_put(mddev);
4294 return error;
4295 }
4296
4297 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4298 {
4299 md_alloc(dev, NULL);
4300 return NULL;
4301 }
4302
4303 static int add_named_array(const char *val, struct kernel_param *kp)
4304 {
4305 /* val must be "md_*" where * is not all digits.
4306 * We allocate an array with a large free minor number, and
4307 * set the name to val. val must not already be an active name.
4308 */
4309 int len = strlen(val);
4310 char buf[DISK_NAME_LEN];
4311
4312 while (len && val[len-1] == '\n')
4313 len--;
4314 if (len >= DISK_NAME_LEN)
4315 return -E2BIG;
4316 strlcpy(buf, val, len+1);
4317 if (strncmp(buf, "md_", 3) != 0)
4318 return -EINVAL;
4319 return md_alloc(0, buf);
4320 }
4321
4322 static void md_safemode_timeout(unsigned long data)
4323 {
4324 mddev_t *mddev = (mddev_t *) data;
4325
4326 if (!atomic_read(&mddev->writes_pending)) {
4327 mddev->safemode = 1;
4328 if (mddev->external)
4329 sysfs_notify_dirent(mddev->sysfs_state);
4330 }
4331 md_wakeup_thread(mddev->thread);
4332 }
4333
4334 static int start_dirty_degraded;
4335
4336 static int md_run(mddev_t *mddev)
4337 {
4338 int err;
4339 mdk_rdev_t *rdev;
4340 struct mdk_personality *pers;
4341
4342 if (list_empty(&mddev->disks))
4343 /* cannot run an array with no devices.. */
4344 return -EINVAL;
4345
4346 if (mddev->pers)
4347 return -EBUSY;
4348
4349 /* These two calls synchronise us with the
4350 * sysfs_remove_group calls in mddev_unlock,
4351 * so they must have completed.
4352 */
4353 mutex_lock(&mddev->open_mutex);
4354 mutex_unlock(&mddev->open_mutex);
4355
4356 /*
4357 * Analyze all RAID superblock(s)
4358 */
4359 if (!mddev->raid_disks) {
4360 if (!mddev->persistent)
4361 return -EINVAL;
4362 analyze_sbs(mddev);
4363 }
4364
4365 if (mddev->level != LEVEL_NONE)
4366 request_module("md-level-%d", mddev->level);
4367 else if (mddev->clevel[0])
4368 request_module("md-%s", mddev->clevel);
4369
4370 /*
4371 * Drop all container device buffers, from now on
4372 * the only valid external interface is through the md
4373 * device.
4374 */
4375 list_for_each_entry(rdev, &mddev->disks, same_set) {
4376 if (test_bit(Faulty, &rdev->flags))
4377 continue;
4378 sync_blockdev(rdev->bdev);
4379 invalidate_bdev(rdev->bdev);
4380
4381 /* perform some consistency tests on the device.
4382 * We don't want the data to overlap the metadata,
4383 * Internal Bitmap issues have been handled elsewhere.
4384 */
4385 if (rdev->data_offset < rdev->sb_start) {
4386 if (mddev->dev_sectors &&
4387 rdev->data_offset + mddev->dev_sectors
4388 > rdev->sb_start) {
4389 printk("md: %s: data overlaps metadata\n",
4390 mdname(mddev));
4391 return -EINVAL;
4392 }
4393 } else {
4394 if (rdev->sb_start + rdev->sb_size/512
4395 > rdev->data_offset) {
4396 printk("md: %s: metadata overlaps data\n",
4397 mdname(mddev));
4398 return -EINVAL;
4399 }
4400 }
4401 sysfs_notify_dirent(rdev->sysfs_state);
4402 }
4403
4404 spin_lock(&pers_lock);
4405 pers = find_pers(mddev->level, mddev->clevel);
4406 if (!pers || !try_module_get(pers->owner)) {
4407 spin_unlock(&pers_lock);
4408 if (mddev->level != LEVEL_NONE)
4409 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4410 mddev->level);
4411 else
4412 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4413 mddev->clevel);
4414 return -EINVAL;
4415 }
4416 mddev->pers = pers;
4417 spin_unlock(&pers_lock);
4418 if (mddev->level != pers->level) {
4419 mddev->level = pers->level;
4420 mddev->new_level = pers->level;
4421 }
4422 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4423
4424 if (mddev->reshape_position != MaxSector &&
4425 pers->start_reshape == NULL) {
4426 /* This personality cannot handle reshaping... */
4427 mddev->pers = NULL;
4428 module_put(pers->owner);
4429 return -EINVAL;
4430 }
4431
4432 if (pers->sync_request) {
4433 /* Warn if this is a potentially silly
4434 * configuration.
4435 */
4436 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4437 mdk_rdev_t *rdev2;
4438 int warned = 0;
4439
4440 list_for_each_entry(rdev, &mddev->disks, same_set)
4441 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4442 if (rdev < rdev2 &&
4443 rdev->bdev->bd_contains ==
4444 rdev2->bdev->bd_contains) {
4445 printk(KERN_WARNING
4446 "%s: WARNING: %s appears to be"
4447 " on the same physical disk as"
4448 " %s.\n",
4449 mdname(mddev),
4450 bdevname(rdev->bdev,b),
4451 bdevname(rdev2->bdev,b2));
4452 warned = 1;
4453 }
4454 }
4455
4456 if (warned)
4457 printk(KERN_WARNING
4458 "True protection against single-disk"
4459 " failure might be compromised.\n");
4460 }
4461
4462 mddev->recovery = 0;
4463 /* may be over-ridden by personality */
4464 mddev->resync_max_sectors = mddev->dev_sectors;
4465
4466 mddev->barriers_work = 1;
4467 mddev->ok_start_degraded = start_dirty_degraded;
4468
4469 if (start_readonly && mddev->ro == 0)
4470 mddev->ro = 2; /* read-only, but switch on first write */
4471
4472 err = mddev->pers->run(mddev);
4473 if (err)
4474 printk(KERN_ERR "md: pers->run() failed ...\n");
4475 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4476 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4477 " but 'external_size' not in effect?\n", __func__);
4478 printk(KERN_ERR
4479 "md: invalid array_size %llu > default size %llu\n",
4480 (unsigned long long)mddev->array_sectors / 2,
4481 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4482 err = -EINVAL;
4483 mddev->pers->stop(mddev);
4484 }
4485 if (err == 0 && mddev->pers->sync_request) {
4486 err = bitmap_create(mddev);
4487 if (err) {
4488 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4489 mdname(mddev), err);
4490 mddev->pers->stop(mddev);
4491 }
4492 }
4493 if (err) {
4494 module_put(mddev->pers->owner);
4495 mddev->pers = NULL;
4496 bitmap_destroy(mddev);
4497 return err;
4498 }
4499 if (mddev->pers->sync_request) {
4500 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4501 printk(KERN_WARNING
4502 "md: cannot register extra attributes for %s\n",
4503 mdname(mddev));
4504 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
4505 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4506 mddev->ro = 0;
4507
4508 atomic_set(&mddev->writes_pending,0);
4509 atomic_set(&mddev->max_corr_read_errors,
4510 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4511 mddev->safemode = 0;
4512 mddev->safemode_timer.function = md_safemode_timeout;
4513 mddev->safemode_timer.data = (unsigned long) mddev;
4514 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4515 mddev->in_sync = 1;
4516
4517 list_for_each_entry(rdev, &mddev->disks, same_set)
4518 if (rdev->raid_disk >= 0) {
4519 char nm[20];
4520 sprintf(nm, "rd%d", rdev->raid_disk);
4521 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4522 printk("md: cannot register %s for %s\n",
4523 nm, mdname(mddev));
4524 }
4525
4526 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4527
4528 if (mddev->flags)
4529 md_update_sb(mddev, 0);
4530
4531 md_wakeup_thread(mddev->thread);
4532 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4533
4534 md_new_event(mddev);
4535 sysfs_notify_dirent(mddev->sysfs_state);
4536 if (mddev->sysfs_action)
4537 sysfs_notify_dirent(mddev->sysfs_action);
4538 sysfs_notify(&mddev->kobj, NULL, "degraded");
4539 return 0;
4540 }
4541
4542 static int do_md_run(mddev_t *mddev)
4543 {
4544 int err;
4545
4546 err = md_run(mddev);
4547 if (err)
4548 goto out;
4549
4550 set_capacity(mddev->gendisk, mddev->array_sectors);
4551 revalidate_disk(mddev->gendisk);
4552 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4553 out:
4554 return err;
4555 }
4556
4557 static int restart_array(mddev_t *mddev)
4558 {
4559 struct gendisk *disk = mddev->gendisk;
4560
4561 /* Complain if it has no devices */
4562 if (list_empty(&mddev->disks))
4563 return -ENXIO;
4564 if (!mddev->pers)
4565 return -EINVAL;
4566 if (!mddev->ro)
4567 return -EBUSY;
4568 mddev->safemode = 0;
4569 mddev->ro = 0;
4570 set_disk_ro(disk, 0);
4571 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4572 mdname(mddev));
4573 /* Kick recovery or resync if necessary */
4574 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4575 md_wakeup_thread(mddev->thread);
4576 md_wakeup_thread(mddev->sync_thread);
4577 sysfs_notify_dirent(mddev->sysfs_state);
4578 return 0;
4579 }
4580
4581 /* similar to deny_write_access, but accounts for our holding a reference
4582 * to the file ourselves */
4583 static int deny_bitmap_write_access(struct file * file)
4584 {
4585 struct inode *inode = file->f_mapping->host;
4586
4587 spin_lock(&inode->i_lock);
4588 if (atomic_read(&inode->i_writecount) > 1) {
4589 spin_unlock(&inode->i_lock);
4590 return -ETXTBSY;
4591 }
4592 atomic_set(&inode->i_writecount, -1);
4593 spin_unlock(&inode->i_lock);
4594
4595 return 0;
4596 }
4597
4598 void restore_bitmap_write_access(struct file *file)
4599 {
4600 struct inode *inode = file->f_mapping->host;
4601
4602 spin_lock(&inode->i_lock);
4603 atomic_set(&inode->i_writecount, 1);
4604 spin_unlock(&inode->i_lock);
4605 }
4606
4607 static void md_clean(mddev_t *mddev)
4608 {
4609 mddev->array_sectors = 0;
4610 mddev->external_size = 0;
4611 mddev->dev_sectors = 0;
4612 mddev->raid_disks = 0;
4613 mddev->recovery_cp = 0;
4614 mddev->resync_min = 0;
4615 mddev->resync_max = MaxSector;
4616 mddev->reshape_position = MaxSector;
4617 mddev->external = 0;
4618 mddev->persistent = 0;
4619 mddev->level = LEVEL_NONE;
4620 mddev->clevel[0] = 0;
4621 mddev->flags = 0;
4622 mddev->ro = 0;
4623 mddev->metadata_type[0] = 0;
4624 mddev->chunk_sectors = 0;
4625 mddev->ctime = mddev->utime = 0;
4626 mddev->layout = 0;
4627 mddev->max_disks = 0;
4628 mddev->events = 0;
4629 mddev->can_decrease_events = 0;
4630 mddev->delta_disks = 0;
4631 mddev->new_level = LEVEL_NONE;
4632 mddev->new_layout = 0;
4633 mddev->new_chunk_sectors = 0;
4634 mddev->curr_resync = 0;
4635 mddev->resync_mismatches = 0;
4636 mddev->suspend_lo = mddev->suspend_hi = 0;
4637 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4638 mddev->recovery = 0;
4639 mddev->in_sync = 0;
4640 mddev->degraded = 0;
4641 mddev->barriers_work = 0;
4642 mddev->safemode = 0;
4643 mddev->bitmap_info.offset = 0;
4644 mddev->bitmap_info.default_offset = 0;
4645 mddev->bitmap_info.chunksize = 0;
4646 mddev->bitmap_info.daemon_sleep = 0;
4647 mddev->bitmap_info.max_write_behind = 0;
4648 }
4649
4650 static void md_stop_writes(mddev_t *mddev)
4651 {
4652 if (mddev->sync_thread) {
4653 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4654 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4655 md_unregister_thread(mddev->sync_thread);
4656 mddev->sync_thread = NULL;
4657 }
4658
4659 del_timer_sync(&mddev->safemode_timer);
4660
4661 bitmap_flush(mddev);
4662 md_super_wait(mddev);
4663
4664 if (!mddev->in_sync || mddev->flags) {
4665 /* mark array as shutdown cleanly */
4666 mddev->in_sync = 1;
4667 md_update_sb(mddev, 1);
4668 }
4669 }
4670
4671 static void md_stop(mddev_t *mddev)
4672 {
4673 md_stop_writes(mddev);
4674
4675 mddev->pers->stop(mddev);
4676 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4677 mddev->to_remove = &md_redundancy_group;
4678 module_put(mddev->pers->owner);
4679 mddev->pers = NULL;
4680 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4681 }
4682
4683 static int md_set_readonly(mddev_t *mddev, int is_open)
4684 {
4685 int err = 0;
4686 mutex_lock(&mddev->open_mutex);
4687 if (atomic_read(&mddev->openers) > is_open) {
4688 printk("md: %s still in use.\n",mdname(mddev));
4689 err = -EBUSY;
4690 goto out;
4691 }
4692 if (mddev->pers) {
4693 md_stop_writes(mddev);
4694
4695 err = -ENXIO;
4696 if (mddev->ro==1)
4697 goto out;
4698 mddev->ro = 1;
4699 set_disk_ro(mddev->gendisk, 1);
4700 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4701 sysfs_notify_dirent(mddev->sysfs_state);
4702 err = 0;
4703 }
4704 out:
4705 mutex_unlock(&mddev->open_mutex);
4706 return err;
4707 }
4708
4709 /* mode:
4710 * 0 - completely stop and dis-assemble array
4711 * 2 - stop but do not disassemble array
4712 */
4713 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4714 {
4715 int err = 0;
4716 struct gendisk *disk = mddev->gendisk;
4717 mdk_rdev_t *rdev;
4718
4719 mutex_lock(&mddev->open_mutex);
4720 if (atomic_read(&mddev->openers) > is_open) {
4721 printk("md: %s still in use.\n",mdname(mddev));
4722 err = -EBUSY;
4723 } else if (mddev->pers) {
4724
4725 if (mddev->ro)
4726 set_disk_ro(disk, 0);
4727
4728 md_stop(mddev);
4729 mddev->queue->merge_bvec_fn = NULL;
4730 mddev->queue->unplug_fn = NULL;
4731 mddev->queue->backing_dev_info.congested_fn = NULL;
4732
4733 /* tell userspace to handle 'inactive' */
4734 sysfs_notify_dirent(mddev->sysfs_state);
4735
4736 list_for_each_entry(rdev, &mddev->disks, same_set)
4737 if (rdev->raid_disk >= 0) {
4738 char nm[20];
4739 sprintf(nm, "rd%d", rdev->raid_disk);
4740 sysfs_remove_link(&mddev->kobj, nm);
4741 }
4742
4743 set_capacity(disk, 0);
4744 revalidate_disk(disk);
4745
4746 if (mddev->ro)
4747 mddev->ro = 0;
4748
4749 err = 0;
4750 }
4751 mutex_unlock(&mddev->open_mutex);
4752 if (err)
4753 return err;
4754 /*
4755 * Free resources if final stop
4756 */
4757 if (mode == 0) {
4758
4759 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4760
4761 bitmap_destroy(mddev);
4762 if (mddev->bitmap_info.file) {
4763 restore_bitmap_write_access(mddev->bitmap_info.file);
4764 fput(mddev->bitmap_info.file);
4765 mddev->bitmap_info.file = NULL;
4766 }
4767 mddev->bitmap_info.offset = 0;
4768
4769 export_array(mddev);
4770
4771 md_clean(mddev);
4772 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4773 if (mddev->hold_active == UNTIL_STOP)
4774 mddev->hold_active = 0;
4775
4776 }
4777 err = 0;
4778 blk_integrity_unregister(disk);
4779 md_new_event(mddev);
4780 sysfs_notify_dirent(mddev->sysfs_state);
4781 return err;
4782 }
4783
4784 #ifndef MODULE
4785 static void autorun_array(mddev_t *mddev)
4786 {
4787 mdk_rdev_t *rdev;
4788 int err;
4789
4790 if (list_empty(&mddev->disks))
4791 return;
4792
4793 printk(KERN_INFO "md: running: ");
4794
4795 list_for_each_entry(rdev, &mddev->disks, same_set) {
4796 char b[BDEVNAME_SIZE];
4797 printk("<%s>", bdevname(rdev->bdev,b));
4798 }
4799 printk("\n");
4800
4801 err = do_md_run(mddev);
4802 if (err) {
4803 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4804 do_md_stop(mddev, 0, 0);
4805 }
4806 }
4807
4808 /*
4809 * lets try to run arrays based on all disks that have arrived
4810 * until now. (those are in pending_raid_disks)
4811 *
4812 * the method: pick the first pending disk, collect all disks with
4813 * the same UUID, remove all from the pending list and put them into
4814 * the 'same_array' list. Then order this list based on superblock
4815 * update time (freshest comes first), kick out 'old' disks and
4816 * compare superblocks. If everything's fine then run it.
4817 *
4818 * If "unit" is allocated, then bump its reference count
4819 */
4820 static void autorun_devices(int part)
4821 {
4822 mdk_rdev_t *rdev0, *rdev, *tmp;
4823 mddev_t *mddev;
4824 char b[BDEVNAME_SIZE];
4825
4826 printk(KERN_INFO "md: autorun ...\n");
4827 while (!list_empty(&pending_raid_disks)) {
4828 int unit;
4829 dev_t dev;
4830 LIST_HEAD(candidates);
4831 rdev0 = list_entry(pending_raid_disks.next,
4832 mdk_rdev_t, same_set);
4833
4834 printk(KERN_INFO "md: considering %s ...\n",
4835 bdevname(rdev0->bdev,b));
4836 INIT_LIST_HEAD(&candidates);
4837 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4838 if (super_90_load(rdev, rdev0, 0) >= 0) {
4839 printk(KERN_INFO "md: adding %s ...\n",
4840 bdevname(rdev->bdev,b));
4841 list_move(&rdev->same_set, &candidates);
4842 }
4843 /*
4844 * now we have a set of devices, with all of them having
4845 * mostly sane superblocks. It's time to allocate the
4846 * mddev.
4847 */
4848 if (part) {
4849 dev = MKDEV(mdp_major,
4850 rdev0->preferred_minor << MdpMinorShift);
4851 unit = MINOR(dev) >> MdpMinorShift;
4852 } else {
4853 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4854 unit = MINOR(dev);
4855 }
4856 if (rdev0->preferred_minor != unit) {
4857 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4858 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4859 break;
4860 }
4861
4862 md_probe(dev, NULL, NULL);
4863 mddev = mddev_find(dev);
4864 if (!mddev || !mddev->gendisk) {
4865 if (mddev)
4866 mddev_put(mddev);
4867 printk(KERN_ERR
4868 "md: cannot allocate memory for md drive.\n");
4869 break;
4870 }
4871 if (mddev_lock(mddev))
4872 printk(KERN_WARNING "md: %s locked, cannot run\n",
4873 mdname(mddev));
4874 else if (mddev->raid_disks || mddev->major_version
4875 || !list_empty(&mddev->disks)) {
4876 printk(KERN_WARNING
4877 "md: %s already running, cannot run %s\n",
4878 mdname(mddev), bdevname(rdev0->bdev,b));
4879 mddev_unlock(mddev);
4880 } else {
4881 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4882 mddev->persistent = 1;
4883 rdev_for_each_list(rdev, tmp, &candidates) {
4884 list_del_init(&rdev->same_set);
4885 if (bind_rdev_to_array(rdev, mddev))
4886 export_rdev(rdev);
4887 }
4888 autorun_array(mddev);
4889 mddev_unlock(mddev);
4890 }
4891 /* on success, candidates will be empty, on error
4892 * it won't...
4893 */
4894 rdev_for_each_list(rdev, tmp, &candidates) {
4895 list_del_init(&rdev->same_set);
4896 export_rdev(rdev);
4897 }
4898 mddev_put(mddev);
4899 }
4900 printk(KERN_INFO "md: ... autorun DONE.\n");
4901 }
4902 #endif /* !MODULE */
4903
4904 static int get_version(void __user * arg)
4905 {
4906 mdu_version_t ver;
4907
4908 ver.major = MD_MAJOR_VERSION;
4909 ver.minor = MD_MINOR_VERSION;
4910 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4911
4912 if (copy_to_user(arg, &ver, sizeof(ver)))
4913 return -EFAULT;
4914
4915 return 0;
4916 }
4917
4918 static int get_array_info(mddev_t * mddev, void __user * arg)
4919 {
4920 mdu_array_info_t info;
4921 int nr,working,insync,failed,spare;
4922 mdk_rdev_t *rdev;
4923
4924 nr=working=insync=failed=spare=0;
4925 list_for_each_entry(rdev, &mddev->disks, same_set) {
4926 nr++;
4927 if (test_bit(Faulty, &rdev->flags))
4928 failed++;
4929 else {
4930 working++;
4931 if (test_bit(In_sync, &rdev->flags))
4932 insync++;
4933 else
4934 spare++;
4935 }
4936 }
4937
4938 info.major_version = mddev->major_version;
4939 info.minor_version = mddev->minor_version;
4940 info.patch_version = MD_PATCHLEVEL_VERSION;
4941 info.ctime = mddev->ctime;
4942 info.level = mddev->level;
4943 info.size = mddev->dev_sectors / 2;
4944 if (info.size != mddev->dev_sectors / 2) /* overflow */
4945 info.size = -1;
4946 info.nr_disks = nr;
4947 info.raid_disks = mddev->raid_disks;
4948 info.md_minor = mddev->md_minor;
4949 info.not_persistent= !mddev->persistent;
4950
4951 info.utime = mddev->utime;
4952 info.state = 0;
4953 if (mddev->in_sync)
4954 info.state = (1<<MD_SB_CLEAN);
4955 if (mddev->bitmap && mddev->bitmap_info.offset)
4956 info.state = (1<<MD_SB_BITMAP_PRESENT);
4957 info.active_disks = insync;
4958 info.working_disks = working;
4959 info.failed_disks = failed;
4960 info.spare_disks = spare;
4961
4962 info.layout = mddev->layout;
4963 info.chunk_size = mddev->chunk_sectors << 9;
4964
4965 if (copy_to_user(arg, &info, sizeof(info)))
4966 return -EFAULT;
4967
4968 return 0;
4969 }
4970
4971 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4972 {
4973 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4974 char *ptr, *buf = NULL;
4975 int err = -ENOMEM;
4976
4977 if (md_allow_write(mddev))
4978 file = kmalloc(sizeof(*file), GFP_NOIO);
4979 else
4980 file = kmalloc(sizeof(*file), GFP_KERNEL);
4981
4982 if (!file)
4983 goto out;
4984
4985 /* bitmap disabled, zero the first byte and copy out */
4986 if (!mddev->bitmap || !mddev->bitmap->file) {
4987 file->pathname[0] = '\0';
4988 goto copy_out;
4989 }
4990
4991 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4992 if (!buf)
4993 goto out;
4994
4995 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4996 if (IS_ERR(ptr))
4997 goto out;
4998
4999 strcpy(file->pathname, ptr);
5000
5001 copy_out:
5002 err = 0;
5003 if (copy_to_user(arg, file, sizeof(*file)))
5004 err = -EFAULT;
5005 out:
5006 kfree(buf);
5007 kfree(file);
5008 return err;
5009 }
5010
5011 static int get_disk_info(mddev_t * mddev, void __user * arg)
5012 {
5013 mdu_disk_info_t info;
5014 mdk_rdev_t *rdev;
5015
5016 if (copy_from_user(&info, arg, sizeof(info)))
5017 return -EFAULT;
5018
5019 rdev = find_rdev_nr(mddev, info.number);
5020 if (rdev) {
5021 info.major = MAJOR(rdev->bdev->bd_dev);
5022 info.minor = MINOR(rdev->bdev->bd_dev);
5023 info.raid_disk = rdev->raid_disk;
5024 info.state = 0;
5025 if (test_bit(Faulty, &rdev->flags))
5026 info.state |= (1<<MD_DISK_FAULTY);
5027 else if (test_bit(In_sync, &rdev->flags)) {
5028 info.state |= (1<<MD_DISK_ACTIVE);
5029 info.state |= (1<<MD_DISK_SYNC);
5030 }
5031 if (test_bit(WriteMostly, &rdev->flags))
5032 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5033 } else {
5034 info.major = info.minor = 0;
5035 info.raid_disk = -1;
5036 info.state = (1<<MD_DISK_REMOVED);
5037 }
5038
5039 if (copy_to_user(arg, &info, sizeof(info)))
5040 return -EFAULT;
5041
5042 return 0;
5043 }
5044
5045 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5046 {
5047 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5048 mdk_rdev_t *rdev;
5049 dev_t dev = MKDEV(info->major,info->minor);
5050
5051 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5052 return -EOVERFLOW;
5053
5054 if (!mddev->raid_disks) {
5055 int err;
5056 /* expecting a device which has a superblock */
5057 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5058 if (IS_ERR(rdev)) {
5059 printk(KERN_WARNING
5060 "md: md_import_device returned %ld\n",
5061 PTR_ERR(rdev));
5062 return PTR_ERR(rdev);
5063 }
5064 if (!list_empty(&mddev->disks)) {
5065 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5066 mdk_rdev_t, same_set);
5067 err = super_types[mddev->major_version]
5068 .load_super(rdev, rdev0, mddev->minor_version);
5069 if (err < 0) {
5070 printk(KERN_WARNING
5071 "md: %s has different UUID to %s\n",
5072 bdevname(rdev->bdev,b),
5073 bdevname(rdev0->bdev,b2));
5074 export_rdev(rdev);
5075 return -EINVAL;
5076 }
5077 }
5078 err = bind_rdev_to_array(rdev, mddev);
5079 if (err)
5080 export_rdev(rdev);
5081 return err;
5082 }
5083
5084 /*
5085 * add_new_disk can be used once the array is assembled
5086 * to add "hot spares". They must already have a superblock
5087 * written
5088 */
5089 if (mddev->pers) {
5090 int err;
5091 if (!mddev->pers->hot_add_disk) {
5092 printk(KERN_WARNING
5093 "%s: personality does not support diskops!\n",
5094 mdname(mddev));
5095 return -EINVAL;
5096 }
5097 if (mddev->persistent)
5098 rdev = md_import_device(dev, mddev->major_version,
5099 mddev->minor_version);
5100 else
5101 rdev = md_import_device(dev, -1, -1);
5102 if (IS_ERR(rdev)) {
5103 printk(KERN_WARNING
5104 "md: md_import_device returned %ld\n",
5105 PTR_ERR(rdev));
5106 return PTR_ERR(rdev);
5107 }
5108 /* set save_raid_disk if appropriate */
5109 if (!mddev->persistent) {
5110 if (info->state & (1<<MD_DISK_SYNC) &&
5111 info->raid_disk < mddev->raid_disks)
5112 rdev->raid_disk = info->raid_disk;
5113 else
5114 rdev->raid_disk = -1;
5115 } else
5116 super_types[mddev->major_version].
5117 validate_super(mddev, rdev);
5118 rdev->saved_raid_disk = rdev->raid_disk;
5119
5120 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5121 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5122 set_bit(WriteMostly, &rdev->flags);
5123 else
5124 clear_bit(WriteMostly, &rdev->flags);
5125
5126 rdev->raid_disk = -1;
5127 err = bind_rdev_to_array(rdev, mddev);
5128 if (!err && !mddev->pers->hot_remove_disk) {
5129 /* If there is hot_add_disk but no hot_remove_disk
5130 * then added disks for geometry changes,
5131 * and should be added immediately.
5132 */
5133 super_types[mddev->major_version].
5134 validate_super(mddev, rdev);
5135 err = mddev->pers->hot_add_disk(mddev, rdev);
5136 if (err)
5137 unbind_rdev_from_array(rdev);
5138 }
5139 if (err)
5140 export_rdev(rdev);
5141 else
5142 sysfs_notify_dirent(rdev->sysfs_state);
5143
5144 md_update_sb(mddev, 1);
5145 if (mddev->degraded)
5146 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5147 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5148 md_wakeup_thread(mddev->thread);
5149 return err;
5150 }
5151
5152 /* otherwise, add_new_disk is only allowed
5153 * for major_version==0 superblocks
5154 */
5155 if (mddev->major_version != 0) {
5156 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5157 mdname(mddev));
5158 return -EINVAL;
5159 }
5160
5161 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5162 int err;
5163 rdev = md_import_device(dev, -1, 0);
5164 if (IS_ERR(rdev)) {
5165 printk(KERN_WARNING
5166 "md: error, md_import_device() returned %ld\n",
5167 PTR_ERR(rdev));
5168 return PTR_ERR(rdev);
5169 }
5170 rdev->desc_nr = info->number;
5171 if (info->raid_disk < mddev->raid_disks)
5172 rdev->raid_disk = info->raid_disk;
5173 else
5174 rdev->raid_disk = -1;
5175
5176 if (rdev->raid_disk < mddev->raid_disks)
5177 if (info->state & (1<<MD_DISK_SYNC))
5178 set_bit(In_sync, &rdev->flags);
5179
5180 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5181 set_bit(WriteMostly, &rdev->flags);
5182
5183 if (!mddev->persistent) {
5184 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5185 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5186 } else
5187 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5188 rdev->sectors = rdev->sb_start;
5189
5190 err = bind_rdev_to_array(rdev, mddev);
5191 if (err) {
5192 export_rdev(rdev);
5193 return err;
5194 }
5195 }
5196
5197 return 0;
5198 }
5199
5200 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5201 {
5202 char b[BDEVNAME_SIZE];
5203 mdk_rdev_t *rdev;
5204
5205 rdev = find_rdev(mddev, dev);
5206 if (!rdev)
5207 return -ENXIO;
5208
5209 if (rdev->raid_disk >= 0)
5210 goto busy;
5211
5212 kick_rdev_from_array(rdev);
5213 md_update_sb(mddev, 1);
5214 md_new_event(mddev);
5215
5216 return 0;
5217 busy:
5218 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5219 bdevname(rdev->bdev,b), mdname(mddev));
5220 return -EBUSY;
5221 }
5222
5223 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5224 {
5225 char b[BDEVNAME_SIZE];
5226 int err;
5227 mdk_rdev_t *rdev;
5228
5229 if (!mddev->pers)
5230 return -ENODEV;
5231
5232 if (mddev->major_version != 0) {
5233 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5234 " version-0 superblocks.\n",
5235 mdname(mddev));
5236 return -EINVAL;
5237 }
5238 if (!mddev->pers->hot_add_disk) {
5239 printk(KERN_WARNING
5240 "%s: personality does not support diskops!\n",
5241 mdname(mddev));
5242 return -EINVAL;
5243 }
5244
5245 rdev = md_import_device(dev, -1, 0);
5246 if (IS_ERR(rdev)) {
5247 printk(KERN_WARNING
5248 "md: error, md_import_device() returned %ld\n",
5249 PTR_ERR(rdev));
5250 return -EINVAL;
5251 }
5252
5253 if (mddev->persistent)
5254 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5255 else
5256 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5257
5258 rdev->sectors = rdev->sb_start;
5259
5260 if (test_bit(Faulty, &rdev->flags)) {
5261 printk(KERN_WARNING
5262 "md: can not hot-add faulty %s disk to %s!\n",
5263 bdevname(rdev->bdev,b), mdname(mddev));
5264 err = -EINVAL;
5265 goto abort_export;
5266 }
5267 clear_bit(In_sync, &rdev->flags);
5268 rdev->desc_nr = -1;
5269 rdev->saved_raid_disk = -1;
5270 err = bind_rdev_to_array(rdev, mddev);
5271 if (err)
5272 goto abort_export;
5273
5274 /*
5275 * The rest should better be atomic, we can have disk failures
5276 * noticed in interrupt contexts ...
5277 */
5278
5279 rdev->raid_disk = -1;
5280
5281 md_update_sb(mddev, 1);
5282
5283 /*
5284 * Kick recovery, maybe this spare has to be added to the
5285 * array immediately.
5286 */
5287 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5288 md_wakeup_thread(mddev->thread);
5289 md_new_event(mddev);
5290 return 0;
5291
5292 abort_export:
5293 export_rdev(rdev);
5294 return err;
5295 }
5296
5297 static int set_bitmap_file(mddev_t *mddev, int fd)
5298 {
5299 int err;
5300
5301 if (mddev->pers) {
5302 if (!mddev->pers->quiesce)
5303 return -EBUSY;
5304 if (mddev->recovery || mddev->sync_thread)
5305 return -EBUSY;
5306 /* we should be able to change the bitmap.. */
5307 }
5308
5309
5310 if (fd >= 0) {
5311 if (mddev->bitmap)
5312 return -EEXIST; /* cannot add when bitmap is present */
5313 mddev->bitmap_info.file = fget(fd);
5314
5315 if (mddev->bitmap_info.file == NULL) {
5316 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5317 mdname(mddev));
5318 return -EBADF;
5319 }
5320
5321 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5322 if (err) {
5323 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5324 mdname(mddev));
5325 fput(mddev->bitmap_info.file);
5326 mddev->bitmap_info.file = NULL;
5327 return err;
5328 }
5329 mddev->bitmap_info.offset = 0; /* file overrides offset */
5330 } else if (mddev->bitmap == NULL)
5331 return -ENOENT; /* cannot remove what isn't there */
5332 err = 0;
5333 if (mddev->pers) {
5334 mddev->pers->quiesce(mddev, 1);
5335 if (fd >= 0)
5336 err = bitmap_create(mddev);
5337 if (fd < 0 || err) {
5338 bitmap_destroy(mddev);
5339 fd = -1; /* make sure to put the file */
5340 }
5341 mddev->pers->quiesce(mddev, 0);
5342 }
5343 if (fd < 0) {
5344 if (mddev->bitmap_info.file) {
5345 restore_bitmap_write_access(mddev->bitmap_info.file);
5346 fput(mddev->bitmap_info.file);
5347 }
5348 mddev->bitmap_info.file = NULL;
5349 }
5350
5351 return err;
5352 }
5353
5354 /*
5355 * set_array_info is used two different ways
5356 * The original usage is when creating a new array.
5357 * In this usage, raid_disks is > 0 and it together with
5358 * level, size, not_persistent,layout,chunksize determine the
5359 * shape of the array.
5360 * This will always create an array with a type-0.90.0 superblock.
5361 * The newer usage is when assembling an array.
5362 * In this case raid_disks will be 0, and the major_version field is
5363 * use to determine which style super-blocks are to be found on the devices.
5364 * The minor and patch _version numbers are also kept incase the
5365 * super_block handler wishes to interpret them.
5366 */
5367 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5368 {
5369
5370 if (info->raid_disks == 0) {
5371 /* just setting version number for superblock loading */
5372 if (info->major_version < 0 ||
5373 info->major_version >= ARRAY_SIZE(super_types) ||
5374 super_types[info->major_version].name == NULL) {
5375 /* maybe try to auto-load a module? */
5376 printk(KERN_INFO
5377 "md: superblock version %d not known\n",
5378 info->major_version);
5379 return -EINVAL;
5380 }
5381 mddev->major_version = info->major_version;
5382 mddev->minor_version = info->minor_version;
5383 mddev->patch_version = info->patch_version;
5384 mddev->persistent = !info->not_persistent;
5385 /* ensure mddev_put doesn't delete this now that there
5386 * is some minimal configuration.
5387 */
5388 mddev->ctime = get_seconds();
5389 return 0;
5390 }
5391 mddev->major_version = MD_MAJOR_VERSION;
5392 mddev->minor_version = MD_MINOR_VERSION;
5393 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5394 mddev->ctime = get_seconds();
5395
5396 mddev->level = info->level;
5397 mddev->clevel[0] = 0;
5398 mddev->dev_sectors = 2 * (sector_t)info->size;
5399 mddev->raid_disks = info->raid_disks;
5400 /* don't set md_minor, it is determined by which /dev/md* was
5401 * openned
5402 */
5403 if (info->state & (1<<MD_SB_CLEAN))
5404 mddev->recovery_cp = MaxSector;
5405 else
5406 mddev->recovery_cp = 0;
5407 mddev->persistent = ! info->not_persistent;
5408 mddev->external = 0;
5409
5410 mddev->layout = info->layout;
5411 mddev->chunk_sectors = info->chunk_size >> 9;
5412
5413 mddev->max_disks = MD_SB_DISKS;
5414
5415 if (mddev->persistent)
5416 mddev->flags = 0;
5417 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5418
5419 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5420 mddev->bitmap_info.offset = 0;
5421
5422 mddev->reshape_position = MaxSector;
5423
5424 /*
5425 * Generate a 128 bit UUID
5426 */
5427 get_random_bytes(mddev->uuid, 16);
5428
5429 mddev->new_level = mddev->level;
5430 mddev->new_chunk_sectors = mddev->chunk_sectors;
5431 mddev->new_layout = mddev->layout;
5432 mddev->delta_disks = 0;
5433
5434 return 0;
5435 }
5436
5437 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5438 {
5439 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5440
5441 if (mddev->external_size)
5442 return;
5443
5444 mddev->array_sectors = array_sectors;
5445 }
5446 EXPORT_SYMBOL(md_set_array_sectors);
5447
5448 static int update_size(mddev_t *mddev, sector_t num_sectors)
5449 {
5450 mdk_rdev_t *rdev;
5451 int rv;
5452 int fit = (num_sectors == 0);
5453
5454 if (mddev->pers->resize == NULL)
5455 return -EINVAL;
5456 /* The "num_sectors" is the number of sectors of each device that
5457 * is used. This can only make sense for arrays with redundancy.
5458 * linear and raid0 always use whatever space is available. We can only
5459 * consider changing this number if no resync or reconstruction is
5460 * happening, and if the new size is acceptable. It must fit before the
5461 * sb_start or, if that is <data_offset, it must fit before the size
5462 * of each device. If num_sectors is zero, we find the largest size
5463 * that fits.
5464
5465 */
5466 if (mddev->sync_thread)
5467 return -EBUSY;
5468 if (mddev->bitmap)
5469 /* Sorry, cannot grow a bitmap yet, just remove it,
5470 * grow, and re-add.
5471 */
5472 return -EBUSY;
5473 list_for_each_entry(rdev, &mddev->disks, same_set) {
5474 sector_t avail = rdev->sectors;
5475
5476 if (fit && (num_sectors == 0 || num_sectors > avail))
5477 num_sectors = avail;
5478 if (avail < num_sectors)
5479 return -ENOSPC;
5480 }
5481 rv = mddev->pers->resize(mddev, num_sectors);
5482 if (!rv)
5483 revalidate_disk(mddev->gendisk);
5484 return rv;
5485 }
5486
5487 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5488 {
5489 int rv;
5490 /* change the number of raid disks */
5491 if (mddev->pers->check_reshape == NULL)
5492 return -EINVAL;
5493 if (raid_disks <= 0 ||
5494 (mddev->max_disks && raid_disks >= mddev->max_disks))
5495 return -EINVAL;
5496 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5497 return -EBUSY;
5498 mddev->delta_disks = raid_disks - mddev->raid_disks;
5499
5500 rv = mddev->pers->check_reshape(mddev);
5501 return rv;
5502 }
5503
5504
5505 /*
5506 * update_array_info is used to change the configuration of an
5507 * on-line array.
5508 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5509 * fields in the info are checked against the array.
5510 * Any differences that cannot be handled will cause an error.
5511 * Normally, only one change can be managed at a time.
5512 */
5513 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5514 {
5515 int rv = 0;
5516 int cnt = 0;
5517 int state = 0;
5518
5519 /* calculate expected state,ignoring low bits */
5520 if (mddev->bitmap && mddev->bitmap_info.offset)
5521 state |= (1 << MD_SB_BITMAP_PRESENT);
5522
5523 if (mddev->major_version != info->major_version ||
5524 mddev->minor_version != info->minor_version ||
5525 /* mddev->patch_version != info->patch_version || */
5526 mddev->ctime != info->ctime ||
5527 mddev->level != info->level ||
5528 /* mddev->layout != info->layout || */
5529 !mddev->persistent != info->not_persistent||
5530 mddev->chunk_sectors != info->chunk_size >> 9 ||
5531 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5532 ((state^info->state) & 0xfffffe00)
5533 )
5534 return -EINVAL;
5535 /* Check there is only one change */
5536 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5537 cnt++;
5538 if (mddev->raid_disks != info->raid_disks)
5539 cnt++;
5540 if (mddev->layout != info->layout)
5541 cnt++;
5542 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5543 cnt++;
5544 if (cnt == 0)
5545 return 0;
5546 if (cnt > 1)
5547 return -EINVAL;
5548
5549 if (mddev->layout != info->layout) {
5550 /* Change layout
5551 * we don't need to do anything at the md level, the
5552 * personality will take care of it all.
5553 */
5554 if (mddev->pers->check_reshape == NULL)
5555 return -EINVAL;
5556 else {
5557 mddev->new_layout = info->layout;
5558 rv = mddev->pers->check_reshape(mddev);
5559 if (rv)
5560 mddev->new_layout = mddev->layout;
5561 return rv;
5562 }
5563 }
5564 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5565 rv = update_size(mddev, (sector_t)info->size * 2);
5566
5567 if (mddev->raid_disks != info->raid_disks)
5568 rv = update_raid_disks(mddev, info->raid_disks);
5569
5570 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5571 if (mddev->pers->quiesce == NULL)
5572 return -EINVAL;
5573 if (mddev->recovery || mddev->sync_thread)
5574 return -EBUSY;
5575 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5576 /* add the bitmap */
5577 if (mddev->bitmap)
5578 return -EEXIST;
5579 if (mddev->bitmap_info.default_offset == 0)
5580 return -EINVAL;
5581 mddev->bitmap_info.offset =
5582 mddev->bitmap_info.default_offset;
5583 mddev->pers->quiesce(mddev, 1);
5584 rv = bitmap_create(mddev);
5585 if (rv)
5586 bitmap_destroy(mddev);
5587 mddev->pers->quiesce(mddev, 0);
5588 } else {
5589 /* remove the bitmap */
5590 if (!mddev->bitmap)
5591 return -ENOENT;
5592 if (mddev->bitmap->file)
5593 return -EINVAL;
5594 mddev->pers->quiesce(mddev, 1);
5595 bitmap_destroy(mddev);
5596 mddev->pers->quiesce(mddev, 0);
5597 mddev->bitmap_info.offset = 0;
5598 }
5599 }
5600 md_update_sb(mddev, 1);
5601 return rv;
5602 }
5603
5604 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5605 {
5606 mdk_rdev_t *rdev;
5607
5608 if (mddev->pers == NULL)
5609 return -ENODEV;
5610
5611 rdev = find_rdev(mddev, dev);
5612 if (!rdev)
5613 return -ENODEV;
5614
5615 md_error(mddev, rdev);
5616 return 0;
5617 }
5618
5619 /*
5620 * We have a problem here : there is no easy way to give a CHS
5621 * virtual geometry. We currently pretend that we have a 2 heads
5622 * 4 sectors (with a BIG number of cylinders...). This drives
5623 * dosfs just mad... ;-)
5624 */
5625 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5626 {
5627 mddev_t *mddev = bdev->bd_disk->private_data;
5628
5629 geo->heads = 2;
5630 geo->sectors = 4;
5631 geo->cylinders = mddev->array_sectors / 8;
5632 return 0;
5633 }
5634
5635 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5636 unsigned int cmd, unsigned long arg)
5637 {
5638 int err = 0;
5639 void __user *argp = (void __user *)arg;
5640 mddev_t *mddev = NULL;
5641 int ro;
5642
5643 if (!capable(CAP_SYS_ADMIN))
5644 return -EACCES;
5645
5646 /*
5647 * Commands dealing with the RAID driver but not any
5648 * particular array:
5649 */
5650 switch (cmd)
5651 {
5652 case RAID_VERSION:
5653 err = get_version(argp);
5654 goto done;
5655
5656 case PRINT_RAID_DEBUG:
5657 err = 0;
5658 md_print_devices();
5659 goto done;
5660
5661 #ifndef MODULE
5662 case RAID_AUTORUN:
5663 err = 0;
5664 autostart_arrays(arg);
5665 goto done;
5666 #endif
5667 default:;
5668 }
5669
5670 /*
5671 * Commands creating/starting a new array:
5672 */
5673
5674 mddev = bdev->bd_disk->private_data;
5675
5676 if (!mddev) {
5677 BUG();
5678 goto abort;
5679 }
5680
5681 err = mddev_lock(mddev);
5682 if (err) {
5683 printk(KERN_INFO
5684 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5685 err, cmd);
5686 goto abort;
5687 }
5688
5689 switch (cmd)
5690 {
5691 case SET_ARRAY_INFO:
5692 {
5693 mdu_array_info_t info;
5694 if (!arg)
5695 memset(&info, 0, sizeof(info));
5696 else if (copy_from_user(&info, argp, sizeof(info))) {
5697 err = -EFAULT;
5698 goto abort_unlock;
5699 }
5700 if (mddev->pers) {
5701 err = update_array_info(mddev, &info);
5702 if (err) {
5703 printk(KERN_WARNING "md: couldn't update"
5704 " array info. %d\n", err);
5705 goto abort_unlock;
5706 }
5707 goto done_unlock;
5708 }
5709 if (!list_empty(&mddev->disks)) {
5710 printk(KERN_WARNING
5711 "md: array %s already has disks!\n",
5712 mdname(mddev));
5713 err = -EBUSY;
5714 goto abort_unlock;
5715 }
5716 if (mddev->raid_disks) {
5717 printk(KERN_WARNING
5718 "md: array %s already initialised!\n",
5719 mdname(mddev));
5720 err = -EBUSY;
5721 goto abort_unlock;
5722 }
5723 err = set_array_info(mddev, &info);
5724 if (err) {
5725 printk(KERN_WARNING "md: couldn't set"
5726 " array info. %d\n", err);
5727 goto abort_unlock;
5728 }
5729 }
5730 goto done_unlock;
5731
5732 default:;
5733 }
5734
5735 /*
5736 * Commands querying/configuring an existing array:
5737 */
5738 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5739 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5740 if ((!mddev->raid_disks && !mddev->external)
5741 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5742 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5743 && cmd != GET_BITMAP_FILE) {
5744 err = -ENODEV;
5745 goto abort_unlock;
5746 }
5747
5748 /*
5749 * Commands even a read-only array can execute:
5750 */
5751 switch (cmd)
5752 {
5753 case GET_ARRAY_INFO:
5754 err = get_array_info(mddev, argp);
5755 goto done_unlock;
5756
5757 case GET_BITMAP_FILE:
5758 err = get_bitmap_file(mddev, argp);
5759 goto done_unlock;
5760
5761 case GET_DISK_INFO:
5762 err = get_disk_info(mddev, argp);
5763 goto done_unlock;
5764
5765 case RESTART_ARRAY_RW:
5766 err = restart_array(mddev);
5767 goto done_unlock;
5768
5769 case STOP_ARRAY:
5770 err = do_md_stop(mddev, 0, 1);
5771 goto done_unlock;
5772
5773 case STOP_ARRAY_RO:
5774 err = md_set_readonly(mddev, 1);
5775 goto done_unlock;
5776
5777 case BLKROSET:
5778 if (get_user(ro, (int __user *)(arg))) {
5779 err = -EFAULT;
5780 goto done_unlock;
5781 }
5782 err = -EINVAL;
5783
5784 /* if the bdev is going readonly the value of mddev->ro
5785 * does not matter, no writes are coming
5786 */
5787 if (ro)
5788 goto done_unlock;
5789
5790 /* are we are already prepared for writes? */
5791 if (mddev->ro != 1)
5792 goto done_unlock;
5793
5794 /* transitioning to readauto need only happen for
5795 * arrays that call md_write_start
5796 */
5797 if (mddev->pers) {
5798 err = restart_array(mddev);
5799 if (err == 0) {
5800 mddev->ro = 2;
5801 set_disk_ro(mddev->gendisk, 0);
5802 }
5803 }
5804 goto done_unlock;
5805 }
5806
5807 /*
5808 * The remaining ioctls are changing the state of the
5809 * superblock, so we do not allow them on read-only arrays.
5810 * However non-MD ioctls (e.g. get-size) will still come through
5811 * here and hit the 'default' below, so only disallow
5812 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5813 */
5814 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5815 if (mddev->ro == 2) {
5816 mddev->ro = 0;
5817 sysfs_notify_dirent(mddev->sysfs_state);
5818 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5819 md_wakeup_thread(mddev->thread);
5820 } else {
5821 err = -EROFS;
5822 goto abort_unlock;
5823 }
5824 }
5825
5826 switch (cmd)
5827 {
5828 case ADD_NEW_DISK:
5829 {
5830 mdu_disk_info_t info;
5831 if (copy_from_user(&info, argp, sizeof(info)))
5832 err = -EFAULT;
5833 else
5834 err = add_new_disk(mddev, &info);
5835 goto done_unlock;
5836 }
5837
5838 case HOT_REMOVE_DISK:
5839 err = hot_remove_disk(mddev, new_decode_dev(arg));
5840 goto done_unlock;
5841
5842 case HOT_ADD_DISK:
5843 err = hot_add_disk(mddev, new_decode_dev(arg));
5844 goto done_unlock;
5845
5846 case SET_DISK_FAULTY:
5847 err = set_disk_faulty(mddev, new_decode_dev(arg));
5848 goto done_unlock;
5849
5850 case RUN_ARRAY:
5851 err = do_md_run(mddev);
5852 goto done_unlock;
5853
5854 case SET_BITMAP_FILE:
5855 err = set_bitmap_file(mddev, (int)arg);
5856 goto done_unlock;
5857
5858 default:
5859 err = -EINVAL;
5860 goto abort_unlock;
5861 }
5862
5863 done_unlock:
5864 abort_unlock:
5865 if (mddev->hold_active == UNTIL_IOCTL &&
5866 err != -EINVAL)
5867 mddev->hold_active = 0;
5868 mddev_unlock(mddev);
5869
5870 return err;
5871 done:
5872 if (err)
5873 MD_BUG();
5874 abort:
5875 return err;
5876 }
5877 #ifdef CONFIG_COMPAT
5878 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5879 unsigned int cmd, unsigned long arg)
5880 {
5881 switch (cmd) {
5882 case HOT_REMOVE_DISK:
5883 case HOT_ADD_DISK:
5884 case SET_DISK_FAULTY:
5885 case SET_BITMAP_FILE:
5886 /* These take in integer arg, do not convert */
5887 break;
5888 default:
5889 arg = (unsigned long)compat_ptr(arg);
5890 break;
5891 }
5892
5893 return md_ioctl(bdev, mode, cmd, arg);
5894 }
5895 #endif /* CONFIG_COMPAT */
5896
5897 static int md_open(struct block_device *bdev, fmode_t mode)
5898 {
5899 /*
5900 * Succeed if we can lock the mddev, which confirms that
5901 * it isn't being stopped right now.
5902 */
5903 mddev_t *mddev = mddev_find(bdev->bd_dev);
5904 int err;
5905
5906 lock_kernel();
5907 if (mddev->gendisk != bdev->bd_disk) {
5908 /* we are racing with mddev_put which is discarding this
5909 * bd_disk.
5910 */
5911 mddev_put(mddev);
5912 /* Wait until bdev->bd_disk is definitely gone */
5913 flush_scheduled_work();
5914 /* Then retry the open from the top */
5915 unlock_kernel();
5916 return -ERESTARTSYS;
5917 }
5918 BUG_ON(mddev != bdev->bd_disk->private_data);
5919
5920 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5921 goto out;
5922
5923 err = 0;
5924 atomic_inc(&mddev->openers);
5925 mutex_unlock(&mddev->open_mutex);
5926
5927 check_disk_size_change(mddev->gendisk, bdev);
5928 out:
5929 unlock_kernel();
5930 return err;
5931 }
5932
5933 static int md_release(struct gendisk *disk, fmode_t mode)
5934 {
5935 mddev_t *mddev = disk->private_data;
5936
5937 BUG_ON(!mddev);
5938 lock_kernel();
5939 atomic_dec(&mddev->openers);
5940 mddev_put(mddev);
5941 unlock_kernel();
5942
5943 return 0;
5944 }
5945 static const struct block_device_operations md_fops =
5946 {
5947 .owner = THIS_MODULE,
5948 .open = md_open,
5949 .release = md_release,
5950 .ioctl = md_ioctl,
5951 #ifdef CONFIG_COMPAT
5952 .compat_ioctl = md_compat_ioctl,
5953 #endif
5954 .getgeo = md_getgeo,
5955 };
5956
5957 static int md_thread(void * arg)
5958 {
5959 mdk_thread_t *thread = arg;
5960
5961 /*
5962 * md_thread is a 'system-thread', it's priority should be very
5963 * high. We avoid resource deadlocks individually in each
5964 * raid personality. (RAID5 does preallocation) We also use RR and
5965 * the very same RT priority as kswapd, thus we will never get
5966 * into a priority inversion deadlock.
5967 *
5968 * we definitely have to have equal or higher priority than
5969 * bdflush, otherwise bdflush will deadlock if there are too
5970 * many dirty RAID5 blocks.
5971 */
5972
5973 allow_signal(SIGKILL);
5974 while (!kthread_should_stop()) {
5975
5976 /* We need to wait INTERRUPTIBLE so that
5977 * we don't add to the load-average.
5978 * That means we need to be sure no signals are
5979 * pending
5980 */
5981 if (signal_pending(current))
5982 flush_signals(current);
5983
5984 wait_event_interruptible_timeout
5985 (thread->wqueue,
5986 test_bit(THREAD_WAKEUP, &thread->flags)
5987 || kthread_should_stop(),
5988 thread->timeout);
5989
5990 clear_bit(THREAD_WAKEUP, &thread->flags);
5991
5992 thread->run(thread->mddev);
5993 }
5994
5995 return 0;
5996 }
5997
5998 void md_wakeup_thread(mdk_thread_t *thread)
5999 {
6000 if (thread) {
6001 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6002 set_bit(THREAD_WAKEUP, &thread->flags);
6003 wake_up(&thread->wqueue);
6004 }
6005 }
6006
6007 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6008 const char *name)
6009 {
6010 mdk_thread_t *thread;
6011
6012 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6013 if (!thread)
6014 return NULL;
6015
6016 init_waitqueue_head(&thread->wqueue);
6017
6018 thread->run = run;
6019 thread->mddev = mddev;
6020 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6021 thread->tsk = kthread_run(md_thread, thread,
6022 "%s_%s",
6023 mdname(thread->mddev),
6024 name ?: mddev->pers->name);
6025 if (IS_ERR(thread->tsk)) {
6026 kfree(thread);
6027 return NULL;
6028 }
6029 return thread;
6030 }
6031
6032 void md_unregister_thread(mdk_thread_t *thread)
6033 {
6034 if (!thread)
6035 return;
6036 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6037
6038 kthread_stop(thread->tsk);
6039 kfree(thread);
6040 }
6041
6042 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6043 {
6044 if (!mddev) {
6045 MD_BUG();
6046 return;
6047 }
6048
6049 if (!rdev || test_bit(Faulty, &rdev->flags))
6050 return;
6051
6052 if (mddev->external)
6053 set_bit(Blocked, &rdev->flags);
6054 /*
6055 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6056 mdname(mddev),
6057 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6058 __builtin_return_address(0),__builtin_return_address(1),
6059 __builtin_return_address(2),__builtin_return_address(3));
6060 */
6061 if (!mddev->pers)
6062 return;
6063 if (!mddev->pers->error_handler)
6064 return;
6065 mddev->pers->error_handler(mddev,rdev);
6066 if (mddev->degraded)
6067 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6068 sysfs_notify_dirent(rdev->sysfs_state);
6069 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6070 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6071 md_wakeup_thread(mddev->thread);
6072 md_new_event_inintr(mddev);
6073 }
6074
6075 /* seq_file implementation /proc/mdstat */
6076
6077 static void status_unused(struct seq_file *seq)
6078 {
6079 int i = 0;
6080 mdk_rdev_t *rdev;
6081
6082 seq_printf(seq, "unused devices: ");
6083
6084 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6085 char b[BDEVNAME_SIZE];
6086 i++;
6087 seq_printf(seq, "%s ",
6088 bdevname(rdev->bdev,b));
6089 }
6090 if (!i)
6091 seq_printf(seq, "<none>");
6092
6093 seq_printf(seq, "\n");
6094 }
6095
6096
6097 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6098 {
6099 sector_t max_sectors, resync, res;
6100 unsigned long dt, db;
6101 sector_t rt;
6102 int scale;
6103 unsigned int per_milli;
6104
6105 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6106
6107 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6108 max_sectors = mddev->resync_max_sectors;
6109 else
6110 max_sectors = mddev->dev_sectors;
6111
6112 /*
6113 * Should not happen.
6114 */
6115 if (!max_sectors) {
6116 MD_BUG();
6117 return;
6118 }
6119 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6120 * in a sector_t, and (max_sectors>>scale) will fit in a
6121 * u32, as those are the requirements for sector_div.
6122 * Thus 'scale' must be at least 10
6123 */
6124 scale = 10;
6125 if (sizeof(sector_t) > sizeof(unsigned long)) {
6126 while ( max_sectors/2 > (1ULL<<(scale+32)))
6127 scale++;
6128 }
6129 res = (resync>>scale)*1000;
6130 sector_div(res, (u32)((max_sectors>>scale)+1));
6131
6132 per_milli = res;
6133 {
6134 int i, x = per_milli/50, y = 20-x;
6135 seq_printf(seq, "[");
6136 for (i = 0; i < x; i++)
6137 seq_printf(seq, "=");
6138 seq_printf(seq, ">");
6139 for (i = 0; i < y; i++)
6140 seq_printf(seq, ".");
6141 seq_printf(seq, "] ");
6142 }
6143 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6144 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6145 "reshape" :
6146 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6147 "check" :
6148 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6149 "resync" : "recovery"))),
6150 per_milli/10, per_milli % 10,
6151 (unsigned long long) resync/2,
6152 (unsigned long long) max_sectors/2);
6153
6154 /*
6155 * dt: time from mark until now
6156 * db: blocks written from mark until now
6157 * rt: remaining time
6158 *
6159 * rt is a sector_t, so could be 32bit or 64bit.
6160 * So we divide before multiply in case it is 32bit and close
6161 * to the limit.
6162 * We scale the divisor (db) by 32 to avoid loosing precision
6163 * near the end of resync when the number of remaining sectors
6164 * is close to 'db'.
6165 * We then divide rt by 32 after multiplying by db to compensate.
6166 * The '+1' avoids division by zero if db is very small.
6167 */
6168 dt = ((jiffies - mddev->resync_mark) / HZ);
6169 if (!dt) dt++;
6170 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6171 - mddev->resync_mark_cnt;
6172
6173 rt = max_sectors - resync; /* number of remaining sectors */
6174 sector_div(rt, db/32+1);
6175 rt *= dt;
6176 rt >>= 5;
6177
6178 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6179 ((unsigned long)rt % 60)/6);
6180
6181 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6182 }
6183
6184 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6185 {
6186 struct list_head *tmp;
6187 loff_t l = *pos;
6188 mddev_t *mddev;
6189
6190 if (l >= 0x10000)
6191 return NULL;
6192 if (!l--)
6193 /* header */
6194 return (void*)1;
6195
6196 spin_lock(&all_mddevs_lock);
6197 list_for_each(tmp,&all_mddevs)
6198 if (!l--) {
6199 mddev = list_entry(tmp, mddev_t, all_mddevs);
6200 mddev_get(mddev);
6201 spin_unlock(&all_mddevs_lock);
6202 return mddev;
6203 }
6204 spin_unlock(&all_mddevs_lock);
6205 if (!l--)
6206 return (void*)2;/* tail */
6207 return NULL;
6208 }
6209
6210 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6211 {
6212 struct list_head *tmp;
6213 mddev_t *next_mddev, *mddev = v;
6214
6215 ++*pos;
6216 if (v == (void*)2)
6217 return NULL;
6218
6219 spin_lock(&all_mddevs_lock);
6220 if (v == (void*)1)
6221 tmp = all_mddevs.next;
6222 else
6223 tmp = mddev->all_mddevs.next;
6224 if (tmp != &all_mddevs)
6225 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6226 else {
6227 next_mddev = (void*)2;
6228 *pos = 0x10000;
6229 }
6230 spin_unlock(&all_mddevs_lock);
6231
6232 if (v != (void*)1)
6233 mddev_put(mddev);
6234 return next_mddev;
6235
6236 }
6237
6238 static void md_seq_stop(struct seq_file *seq, void *v)
6239 {
6240 mddev_t *mddev = v;
6241
6242 if (mddev && v != (void*)1 && v != (void*)2)
6243 mddev_put(mddev);
6244 }
6245
6246 struct mdstat_info {
6247 int event;
6248 };
6249
6250 static int md_seq_show(struct seq_file *seq, void *v)
6251 {
6252 mddev_t *mddev = v;
6253 sector_t sectors;
6254 mdk_rdev_t *rdev;
6255 struct mdstat_info *mi = seq->private;
6256 struct bitmap *bitmap;
6257
6258 if (v == (void*)1) {
6259 struct mdk_personality *pers;
6260 seq_printf(seq, "Personalities : ");
6261 spin_lock(&pers_lock);
6262 list_for_each_entry(pers, &pers_list, list)
6263 seq_printf(seq, "[%s] ", pers->name);
6264
6265 spin_unlock(&pers_lock);
6266 seq_printf(seq, "\n");
6267 mi->event = atomic_read(&md_event_count);
6268 return 0;
6269 }
6270 if (v == (void*)2) {
6271 status_unused(seq);
6272 return 0;
6273 }
6274
6275 if (mddev_lock(mddev) < 0)
6276 return -EINTR;
6277
6278 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6279 seq_printf(seq, "%s : %sactive", mdname(mddev),
6280 mddev->pers ? "" : "in");
6281 if (mddev->pers) {
6282 if (mddev->ro==1)
6283 seq_printf(seq, " (read-only)");
6284 if (mddev->ro==2)
6285 seq_printf(seq, " (auto-read-only)");
6286 seq_printf(seq, " %s", mddev->pers->name);
6287 }
6288
6289 sectors = 0;
6290 list_for_each_entry(rdev, &mddev->disks, same_set) {
6291 char b[BDEVNAME_SIZE];
6292 seq_printf(seq, " %s[%d]",
6293 bdevname(rdev->bdev,b), rdev->desc_nr);
6294 if (test_bit(WriteMostly, &rdev->flags))
6295 seq_printf(seq, "(W)");
6296 if (test_bit(Faulty, &rdev->flags)) {
6297 seq_printf(seq, "(F)");
6298 continue;
6299 } else if (rdev->raid_disk < 0)
6300 seq_printf(seq, "(S)"); /* spare */
6301 sectors += rdev->sectors;
6302 }
6303
6304 if (!list_empty(&mddev->disks)) {
6305 if (mddev->pers)
6306 seq_printf(seq, "\n %llu blocks",
6307 (unsigned long long)
6308 mddev->array_sectors / 2);
6309 else
6310 seq_printf(seq, "\n %llu blocks",
6311 (unsigned long long)sectors / 2);
6312 }
6313 if (mddev->persistent) {
6314 if (mddev->major_version != 0 ||
6315 mddev->minor_version != 90) {
6316 seq_printf(seq," super %d.%d",
6317 mddev->major_version,
6318 mddev->minor_version);
6319 }
6320 } else if (mddev->external)
6321 seq_printf(seq, " super external:%s",
6322 mddev->metadata_type);
6323 else
6324 seq_printf(seq, " super non-persistent");
6325
6326 if (mddev->pers) {
6327 mddev->pers->status(seq, mddev);
6328 seq_printf(seq, "\n ");
6329 if (mddev->pers->sync_request) {
6330 if (mddev->curr_resync > 2) {
6331 status_resync(seq, mddev);
6332 seq_printf(seq, "\n ");
6333 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6334 seq_printf(seq, "\tresync=DELAYED\n ");
6335 else if (mddev->recovery_cp < MaxSector)
6336 seq_printf(seq, "\tresync=PENDING\n ");
6337 }
6338 } else
6339 seq_printf(seq, "\n ");
6340
6341 if ((bitmap = mddev->bitmap)) {
6342 unsigned long chunk_kb;
6343 unsigned long flags;
6344 spin_lock_irqsave(&bitmap->lock, flags);
6345 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6346 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6347 "%lu%s chunk",
6348 bitmap->pages - bitmap->missing_pages,
6349 bitmap->pages,
6350 (bitmap->pages - bitmap->missing_pages)
6351 << (PAGE_SHIFT - 10),
6352 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6353 chunk_kb ? "KB" : "B");
6354 if (bitmap->file) {
6355 seq_printf(seq, ", file: ");
6356 seq_path(seq, &bitmap->file->f_path, " \t\n");
6357 }
6358
6359 seq_printf(seq, "\n");
6360 spin_unlock_irqrestore(&bitmap->lock, flags);
6361 }
6362
6363 seq_printf(seq, "\n");
6364 }
6365 mddev_unlock(mddev);
6366
6367 return 0;
6368 }
6369
6370 static const struct seq_operations md_seq_ops = {
6371 .start = md_seq_start,
6372 .next = md_seq_next,
6373 .stop = md_seq_stop,
6374 .show = md_seq_show,
6375 };
6376
6377 static int md_seq_open(struct inode *inode, struct file *file)
6378 {
6379 int error;
6380 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6381 if (mi == NULL)
6382 return -ENOMEM;
6383
6384 error = seq_open(file, &md_seq_ops);
6385 if (error)
6386 kfree(mi);
6387 else {
6388 struct seq_file *p = file->private_data;
6389 p->private = mi;
6390 mi->event = atomic_read(&md_event_count);
6391 }
6392 return error;
6393 }
6394
6395 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6396 {
6397 struct seq_file *m = filp->private_data;
6398 struct mdstat_info *mi = m->private;
6399 int mask;
6400
6401 poll_wait(filp, &md_event_waiters, wait);
6402
6403 /* always allow read */
6404 mask = POLLIN | POLLRDNORM;
6405
6406 if (mi->event != atomic_read(&md_event_count))
6407 mask |= POLLERR | POLLPRI;
6408 return mask;
6409 }
6410
6411 static const struct file_operations md_seq_fops = {
6412 .owner = THIS_MODULE,
6413 .open = md_seq_open,
6414 .read = seq_read,
6415 .llseek = seq_lseek,
6416 .release = seq_release_private,
6417 .poll = mdstat_poll,
6418 };
6419
6420 int register_md_personality(struct mdk_personality *p)
6421 {
6422 spin_lock(&pers_lock);
6423 list_add_tail(&p->list, &pers_list);
6424 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6425 spin_unlock(&pers_lock);
6426 return 0;
6427 }
6428
6429 int unregister_md_personality(struct mdk_personality *p)
6430 {
6431 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6432 spin_lock(&pers_lock);
6433 list_del_init(&p->list);
6434 spin_unlock(&pers_lock);
6435 return 0;
6436 }
6437
6438 static int is_mddev_idle(mddev_t *mddev, int init)
6439 {
6440 mdk_rdev_t * rdev;
6441 int idle;
6442 int curr_events;
6443
6444 idle = 1;
6445 rcu_read_lock();
6446 rdev_for_each_rcu(rdev, mddev) {
6447 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6448 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6449 (int)part_stat_read(&disk->part0, sectors[1]) -
6450 atomic_read(&disk->sync_io);
6451 /* sync IO will cause sync_io to increase before the disk_stats
6452 * as sync_io is counted when a request starts, and
6453 * disk_stats is counted when it completes.
6454 * So resync activity will cause curr_events to be smaller than
6455 * when there was no such activity.
6456 * non-sync IO will cause disk_stat to increase without
6457 * increasing sync_io so curr_events will (eventually)
6458 * be larger than it was before. Once it becomes
6459 * substantially larger, the test below will cause
6460 * the array to appear non-idle, and resync will slow
6461 * down.
6462 * If there is a lot of outstanding resync activity when
6463 * we set last_event to curr_events, then all that activity
6464 * completing might cause the array to appear non-idle
6465 * and resync will be slowed down even though there might
6466 * not have been non-resync activity. This will only
6467 * happen once though. 'last_events' will soon reflect
6468 * the state where there is little or no outstanding
6469 * resync requests, and further resync activity will
6470 * always make curr_events less than last_events.
6471 *
6472 */
6473 if (init || curr_events - rdev->last_events > 64) {
6474 rdev->last_events = curr_events;
6475 idle = 0;
6476 }
6477 }
6478 rcu_read_unlock();
6479 return idle;
6480 }
6481
6482 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6483 {
6484 /* another "blocks" (512byte) blocks have been synced */
6485 atomic_sub(blocks, &mddev->recovery_active);
6486 wake_up(&mddev->recovery_wait);
6487 if (!ok) {
6488 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6489 md_wakeup_thread(mddev->thread);
6490 // stop recovery, signal do_sync ....
6491 }
6492 }
6493
6494
6495 /* md_write_start(mddev, bi)
6496 * If we need to update some array metadata (e.g. 'active' flag
6497 * in superblock) before writing, schedule a superblock update
6498 * and wait for it to complete.
6499 */
6500 void md_write_start(mddev_t *mddev, struct bio *bi)
6501 {
6502 int did_change = 0;
6503 if (bio_data_dir(bi) != WRITE)
6504 return;
6505
6506 BUG_ON(mddev->ro == 1);
6507 if (mddev->ro == 2) {
6508 /* need to switch to read/write */
6509 mddev->ro = 0;
6510 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6511 md_wakeup_thread(mddev->thread);
6512 md_wakeup_thread(mddev->sync_thread);
6513 did_change = 1;
6514 }
6515 atomic_inc(&mddev->writes_pending);
6516 if (mddev->safemode == 1)
6517 mddev->safemode = 0;
6518 if (mddev->in_sync) {
6519 spin_lock_irq(&mddev->write_lock);
6520 if (mddev->in_sync) {
6521 mddev->in_sync = 0;
6522 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6523 md_wakeup_thread(mddev->thread);
6524 did_change = 1;
6525 }
6526 spin_unlock_irq(&mddev->write_lock);
6527 }
6528 if (did_change)
6529 sysfs_notify_dirent(mddev->sysfs_state);
6530 wait_event(mddev->sb_wait,
6531 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6532 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6533 }
6534
6535 void md_write_end(mddev_t *mddev)
6536 {
6537 if (atomic_dec_and_test(&mddev->writes_pending)) {
6538 if (mddev->safemode == 2)
6539 md_wakeup_thread(mddev->thread);
6540 else if (mddev->safemode_delay)
6541 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6542 }
6543 }
6544
6545 /* md_allow_write(mddev)
6546 * Calling this ensures that the array is marked 'active' so that writes
6547 * may proceed without blocking. It is important to call this before
6548 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6549 * Must be called with mddev_lock held.
6550 *
6551 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6552 * is dropped, so return -EAGAIN after notifying userspace.
6553 */
6554 int md_allow_write(mddev_t *mddev)
6555 {
6556 if (!mddev->pers)
6557 return 0;
6558 if (mddev->ro)
6559 return 0;
6560 if (!mddev->pers->sync_request)
6561 return 0;
6562
6563 spin_lock_irq(&mddev->write_lock);
6564 if (mddev->in_sync) {
6565 mddev->in_sync = 0;
6566 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6567 if (mddev->safemode_delay &&
6568 mddev->safemode == 0)
6569 mddev->safemode = 1;
6570 spin_unlock_irq(&mddev->write_lock);
6571 md_update_sb(mddev, 0);
6572 sysfs_notify_dirent(mddev->sysfs_state);
6573 } else
6574 spin_unlock_irq(&mddev->write_lock);
6575
6576 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6577 return -EAGAIN;
6578 else
6579 return 0;
6580 }
6581 EXPORT_SYMBOL_GPL(md_allow_write);
6582
6583 #define SYNC_MARKS 10
6584 #define SYNC_MARK_STEP (3*HZ)
6585 void md_do_sync(mddev_t *mddev)
6586 {
6587 mddev_t *mddev2;
6588 unsigned int currspeed = 0,
6589 window;
6590 sector_t max_sectors,j, io_sectors;
6591 unsigned long mark[SYNC_MARKS];
6592 sector_t mark_cnt[SYNC_MARKS];
6593 int last_mark,m;
6594 struct list_head *tmp;
6595 sector_t last_check;
6596 int skipped = 0;
6597 mdk_rdev_t *rdev;
6598 char *desc;
6599
6600 /* just incase thread restarts... */
6601 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6602 return;
6603 if (mddev->ro) /* never try to sync a read-only array */
6604 return;
6605
6606 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6607 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6608 desc = "data-check";
6609 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6610 desc = "requested-resync";
6611 else
6612 desc = "resync";
6613 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6614 desc = "reshape";
6615 else
6616 desc = "recovery";
6617
6618 /* we overload curr_resync somewhat here.
6619 * 0 == not engaged in resync at all
6620 * 2 == checking that there is no conflict with another sync
6621 * 1 == like 2, but have yielded to allow conflicting resync to
6622 * commense
6623 * other == active in resync - this many blocks
6624 *
6625 * Before starting a resync we must have set curr_resync to
6626 * 2, and then checked that every "conflicting" array has curr_resync
6627 * less than ours. When we find one that is the same or higher
6628 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6629 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6630 * This will mean we have to start checking from the beginning again.
6631 *
6632 */
6633
6634 do {
6635 mddev->curr_resync = 2;
6636
6637 try_again:
6638 if (kthread_should_stop())
6639 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6640
6641 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6642 goto skip;
6643 for_each_mddev(mddev2, tmp) {
6644 if (mddev2 == mddev)
6645 continue;
6646 if (!mddev->parallel_resync
6647 && mddev2->curr_resync
6648 && match_mddev_units(mddev, mddev2)) {
6649 DEFINE_WAIT(wq);
6650 if (mddev < mddev2 && mddev->curr_resync == 2) {
6651 /* arbitrarily yield */
6652 mddev->curr_resync = 1;
6653 wake_up(&resync_wait);
6654 }
6655 if (mddev > mddev2 && mddev->curr_resync == 1)
6656 /* no need to wait here, we can wait the next
6657 * time 'round when curr_resync == 2
6658 */
6659 continue;
6660 /* We need to wait 'interruptible' so as not to
6661 * contribute to the load average, and not to
6662 * be caught by 'softlockup'
6663 */
6664 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6665 if (!kthread_should_stop() &&
6666 mddev2->curr_resync >= mddev->curr_resync) {
6667 printk(KERN_INFO "md: delaying %s of %s"
6668 " until %s has finished (they"
6669 " share one or more physical units)\n",
6670 desc, mdname(mddev), mdname(mddev2));
6671 mddev_put(mddev2);
6672 if (signal_pending(current))
6673 flush_signals(current);
6674 schedule();
6675 finish_wait(&resync_wait, &wq);
6676 goto try_again;
6677 }
6678 finish_wait(&resync_wait, &wq);
6679 }
6680 }
6681 } while (mddev->curr_resync < 2);
6682
6683 j = 0;
6684 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6685 /* resync follows the size requested by the personality,
6686 * which defaults to physical size, but can be virtual size
6687 */
6688 max_sectors = mddev->resync_max_sectors;
6689 mddev->resync_mismatches = 0;
6690 /* we don't use the checkpoint if there's a bitmap */
6691 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6692 j = mddev->resync_min;
6693 else if (!mddev->bitmap)
6694 j = mddev->recovery_cp;
6695
6696 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6697 max_sectors = mddev->dev_sectors;
6698 else {
6699 /* recovery follows the physical size of devices */
6700 max_sectors = mddev->dev_sectors;
6701 j = MaxSector;
6702 rcu_read_lock();
6703 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6704 if (rdev->raid_disk >= 0 &&
6705 !test_bit(Faulty, &rdev->flags) &&
6706 !test_bit(In_sync, &rdev->flags) &&
6707 rdev->recovery_offset < j)
6708 j = rdev->recovery_offset;
6709 rcu_read_unlock();
6710 }
6711
6712 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6713 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6714 " %d KB/sec/disk.\n", speed_min(mddev));
6715 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6716 "(but not more than %d KB/sec) for %s.\n",
6717 speed_max(mddev), desc);
6718
6719 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6720
6721 io_sectors = 0;
6722 for (m = 0; m < SYNC_MARKS; m++) {
6723 mark[m] = jiffies;
6724 mark_cnt[m] = io_sectors;
6725 }
6726 last_mark = 0;
6727 mddev->resync_mark = mark[last_mark];
6728 mddev->resync_mark_cnt = mark_cnt[last_mark];
6729
6730 /*
6731 * Tune reconstruction:
6732 */
6733 window = 32*(PAGE_SIZE/512);
6734 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6735 window/2,(unsigned long long) max_sectors/2);
6736
6737 atomic_set(&mddev->recovery_active, 0);
6738 last_check = 0;
6739
6740 if (j>2) {
6741 printk(KERN_INFO
6742 "md: resuming %s of %s from checkpoint.\n",
6743 desc, mdname(mddev));
6744 mddev->curr_resync = j;
6745 }
6746 mddev->curr_resync_completed = mddev->curr_resync;
6747
6748 while (j < max_sectors) {
6749 sector_t sectors;
6750
6751 skipped = 0;
6752
6753 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6754 ((mddev->curr_resync > mddev->curr_resync_completed &&
6755 (mddev->curr_resync - mddev->curr_resync_completed)
6756 > (max_sectors >> 4)) ||
6757 (j - mddev->curr_resync_completed)*2
6758 >= mddev->resync_max - mddev->curr_resync_completed
6759 )) {
6760 /* time to update curr_resync_completed */
6761 blk_unplug(mddev->queue);
6762 wait_event(mddev->recovery_wait,
6763 atomic_read(&mddev->recovery_active) == 0);
6764 mddev->curr_resync_completed =
6765 mddev->curr_resync;
6766 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6767 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6768 }
6769
6770 while (j >= mddev->resync_max && !kthread_should_stop()) {
6771 /* As this condition is controlled by user-space,
6772 * we can block indefinitely, so use '_interruptible'
6773 * to avoid triggering warnings.
6774 */
6775 flush_signals(current); /* just in case */
6776 wait_event_interruptible(mddev->recovery_wait,
6777 mddev->resync_max > j
6778 || kthread_should_stop());
6779 }
6780
6781 if (kthread_should_stop())
6782 goto interrupted;
6783
6784 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6785 currspeed < speed_min(mddev));
6786 if (sectors == 0) {
6787 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6788 goto out;
6789 }
6790
6791 if (!skipped) { /* actual IO requested */
6792 io_sectors += sectors;
6793 atomic_add(sectors, &mddev->recovery_active);
6794 }
6795
6796 j += sectors;
6797 if (j>1) mddev->curr_resync = j;
6798 mddev->curr_mark_cnt = io_sectors;
6799 if (last_check == 0)
6800 /* this is the earliers that rebuilt will be
6801 * visible in /proc/mdstat
6802 */
6803 md_new_event(mddev);
6804
6805 if (last_check + window > io_sectors || j == max_sectors)
6806 continue;
6807
6808 last_check = io_sectors;
6809
6810 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6811 break;
6812
6813 repeat:
6814 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6815 /* step marks */
6816 int next = (last_mark+1) % SYNC_MARKS;
6817
6818 mddev->resync_mark = mark[next];
6819 mddev->resync_mark_cnt = mark_cnt[next];
6820 mark[next] = jiffies;
6821 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6822 last_mark = next;
6823 }
6824
6825
6826 if (kthread_should_stop())
6827 goto interrupted;
6828
6829
6830 /*
6831 * this loop exits only if either when we are slower than
6832 * the 'hard' speed limit, or the system was IO-idle for
6833 * a jiffy.
6834 * the system might be non-idle CPU-wise, but we only care
6835 * about not overloading the IO subsystem. (things like an
6836 * e2fsck being done on the RAID array should execute fast)
6837 */
6838 blk_unplug(mddev->queue);
6839 cond_resched();
6840
6841 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6842 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6843
6844 if (currspeed > speed_min(mddev)) {
6845 if ((currspeed > speed_max(mddev)) ||
6846 !is_mddev_idle(mddev, 0)) {
6847 msleep(500);
6848 goto repeat;
6849 }
6850 }
6851 }
6852 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6853 /*
6854 * this also signals 'finished resyncing' to md_stop
6855 */
6856 out:
6857 blk_unplug(mddev->queue);
6858
6859 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6860
6861 /* tell personality that we are finished */
6862 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6863
6864 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6865 mddev->curr_resync > 2) {
6866 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6867 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6868 if (mddev->curr_resync >= mddev->recovery_cp) {
6869 printk(KERN_INFO
6870 "md: checkpointing %s of %s.\n",
6871 desc, mdname(mddev));
6872 mddev->recovery_cp = mddev->curr_resync;
6873 }
6874 } else
6875 mddev->recovery_cp = MaxSector;
6876 } else {
6877 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6878 mddev->curr_resync = MaxSector;
6879 rcu_read_lock();
6880 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6881 if (rdev->raid_disk >= 0 &&
6882 mddev->delta_disks >= 0 &&
6883 !test_bit(Faulty, &rdev->flags) &&
6884 !test_bit(In_sync, &rdev->flags) &&
6885 rdev->recovery_offset < mddev->curr_resync)
6886 rdev->recovery_offset = mddev->curr_resync;
6887 rcu_read_unlock();
6888 }
6889 }
6890 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6891
6892 skip:
6893 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6894 /* We completed so min/max setting can be forgotten if used. */
6895 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6896 mddev->resync_min = 0;
6897 mddev->resync_max = MaxSector;
6898 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6899 mddev->resync_min = mddev->curr_resync_completed;
6900 mddev->curr_resync = 0;
6901 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6902 mddev->curr_resync_completed = 0;
6903 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6904 wake_up(&resync_wait);
6905 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6906 md_wakeup_thread(mddev->thread);
6907 return;
6908
6909 interrupted:
6910 /*
6911 * got a signal, exit.
6912 */
6913 printk(KERN_INFO
6914 "md: md_do_sync() got signal ... exiting\n");
6915 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6916 goto out;
6917
6918 }
6919 EXPORT_SYMBOL_GPL(md_do_sync);
6920
6921
6922 static int remove_and_add_spares(mddev_t *mddev)
6923 {
6924 mdk_rdev_t *rdev;
6925 int spares = 0;
6926
6927 mddev->curr_resync_completed = 0;
6928
6929 list_for_each_entry(rdev, &mddev->disks, same_set)
6930 if (rdev->raid_disk >= 0 &&
6931 !test_bit(Blocked, &rdev->flags) &&
6932 (test_bit(Faulty, &rdev->flags) ||
6933 ! test_bit(In_sync, &rdev->flags)) &&
6934 atomic_read(&rdev->nr_pending)==0) {
6935 if (mddev->pers->hot_remove_disk(
6936 mddev, rdev->raid_disk)==0) {
6937 char nm[20];
6938 sprintf(nm,"rd%d", rdev->raid_disk);
6939 sysfs_remove_link(&mddev->kobj, nm);
6940 rdev->raid_disk = -1;
6941 }
6942 }
6943
6944 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6945 list_for_each_entry(rdev, &mddev->disks, same_set) {
6946 if (rdev->raid_disk >= 0 &&
6947 !test_bit(In_sync, &rdev->flags) &&
6948 !test_bit(Blocked, &rdev->flags))
6949 spares++;
6950 if (rdev->raid_disk < 0
6951 && !test_bit(Faulty, &rdev->flags)) {
6952 rdev->recovery_offset = 0;
6953 if (mddev->pers->
6954 hot_add_disk(mddev, rdev) == 0) {
6955 char nm[20];
6956 sprintf(nm, "rd%d", rdev->raid_disk);
6957 if (sysfs_create_link(&mddev->kobj,
6958 &rdev->kobj, nm))
6959 printk(KERN_WARNING
6960 "md: cannot register "
6961 "%s for %s\n",
6962 nm, mdname(mddev));
6963 spares++;
6964 md_new_event(mddev);
6965 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6966 } else
6967 break;
6968 }
6969 }
6970 }
6971 return spares;
6972 }
6973 /*
6974 * This routine is regularly called by all per-raid-array threads to
6975 * deal with generic issues like resync and super-block update.
6976 * Raid personalities that don't have a thread (linear/raid0) do not
6977 * need this as they never do any recovery or update the superblock.
6978 *
6979 * It does not do any resync itself, but rather "forks" off other threads
6980 * to do that as needed.
6981 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6982 * "->recovery" and create a thread at ->sync_thread.
6983 * When the thread finishes it sets MD_RECOVERY_DONE
6984 * and wakeups up this thread which will reap the thread and finish up.
6985 * This thread also removes any faulty devices (with nr_pending == 0).
6986 *
6987 * The overall approach is:
6988 * 1/ if the superblock needs updating, update it.
6989 * 2/ If a recovery thread is running, don't do anything else.
6990 * 3/ If recovery has finished, clean up, possibly marking spares active.
6991 * 4/ If there are any faulty devices, remove them.
6992 * 5/ If array is degraded, try to add spares devices
6993 * 6/ If array has spares or is not in-sync, start a resync thread.
6994 */
6995 void md_check_recovery(mddev_t *mddev)
6996 {
6997 mdk_rdev_t *rdev;
6998
6999
7000 if (mddev->bitmap)
7001 bitmap_daemon_work(mddev);
7002
7003 if (mddev->ro)
7004 return;
7005
7006 if (signal_pending(current)) {
7007 if (mddev->pers->sync_request && !mddev->external) {
7008 printk(KERN_INFO "md: %s in immediate safe mode\n",
7009 mdname(mddev));
7010 mddev->safemode = 2;
7011 }
7012 flush_signals(current);
7013 }
7014
7015 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7016 return;
7017 if ( ! (
7018 (mddev->flags && !mddev->external) ||
7019 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7020 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7021 (mddev->external == 0 && mddev->safemode == 1) ||
7022 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7023 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7024 ))
7025 return;
7026
7027 if (mddev_trylock(mddev)) {
7028 int spares = 0;
7029
7030 if (mddev->ro) {
7031 /* Only thing we do on a ro array is remove
7032 * failed devices.
7033 */
7034 remove_and_add_spares(mddev);
7035 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7036 goto unlock;
7037 }
7038
7039 if (!mddev->external) {
7040 int did_change = 0;
7041 spin_lock_irq(&mddev->write_lock);
7042 if (mddev->safemode &&
7043 !atomic_read(&mddev->writes_pending) &&
7044 !mddev->in_sync &&
7045 mddev->recovery_cp == MaxSector) {
7046 mddev->in_sync = 1;
7047 did_change = 1;
7048 if (mddev->persistent)
7049 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7050 }
7051 if (mddev->safemode == 1)
7052 mddev->safemode = 0;
7053 spin_unlock_irq(&mddev->write_lock);
7054 if (did_change)
7055 sysfs_notify_dirent(mddev->sysfs_state);
7056 }
7057
7058 if (mddev->flags)
7059 md_update_sb(mddev, 0);
7060
7061 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7062 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7063 /* resync/recovery still happening */
7064 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7065 goto unlock;
7066 }
7067 if (mddev->sync_thread) {
7068 /* resync has finished, collect result */
7069 md_unregister_thread(mddev->sync_thread);
7070 mddev->sync_thread = NULL;
7071 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7072 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7073 /* success...*/
7074 /* activate any spares */
7075 if (mddev->pers->spare_active(mddev))
7076 sysfs_notify(&mddev->kobj, NULL,
7077 "degraded");
7078 }
7079 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7080 mddev->pers->finish_reshape)
7081 mddev->pers->finish_reshape(mddev);
7082 md_update_sb(mddev, 1);
7083
7084 /* if array is no-longer degraded, then any saved_raid_disk
7085 * information must be scrapped
7086 */
7087 if (!mddev->degraded)
7088 list_for_each_entry(rdev, &mddev->disks, same_set)
7089 rdev->saved_raid_disk = -1;
7090
7091 mddev->recovery = 0;
7092 /* flag recovery needed just to double check */
7093 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7094 sysfs_notify_dirent(mddev->sysfs_action);
7095 md_new_event(mddev);
7096 goto unlock;
7097 }
7098 /* Set RUNNING before clearing NEEDED to avoid
7099 * any transients in the value of "sync_action".
7100 */
7101 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7102 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7103 /* Clear some bits that don't mean anything, but
7104 * might be left set
7105 */
7106 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7107 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7108
7109 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7110 goto unlock;
7111 /* no recovery is running.
7112 * remove any failed drives, then
7113 * add spares if possible.
7114 * Spare are also removed and re-added, to allow
7115 * the personality to fail the re-add.
7116 */
7117
7118 if (mddev->reshape_position != MaxSector) {
7119 if (mddev->pers->check_reshape == NULL ||
7120 mddev->pers->check_reshape(mddev) != 0)
7121 /* Cannot proceed */
7122 goto unlock;
7123 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7124 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7125 } else if ((spares = remove_and_add_spares(mddev))) {
7126 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7127 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7128 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7129 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7130 } else if (mddev->recovery_cp < MaxSector) {
7131 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7132 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7133 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7134 /* nothing to be done ... */
7135 goto unlock;
7136
7137 if (mddev->pers->sync_request) {
7138 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7139 /* We are adding a device or devices to an array
7140 * which has the bitmap stored on all devices.
7141 * So make sure all bitmap pages get written
7142 */
7143 bitmap_write_all(mddev->bitmap);
7144 }
7145 mddev->sync_thread = md_register_thread(md_do_sync,
7146 mddev,
7147 "resync");
7148 if (!mddev->sync_thread) {
7149 printk(KERN_ERR "%s: could not start resync"
7150 " thread...\n",
7151 mdname(mddev));
7152 /* leave the spares where they are, it shouldn't hurt */
7153 mddev->recovery = 0;
7154 } else
7155 md_wakeup_thread(mddev->sync_thread);
7156 sysfs_notify_dirent(mddev->sysfs_action);
7157 md_new_event(mddev);
7158 }
7159 unlock:
7160 if (!mddev->sync_thread) {
7161 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7162 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7163 &mddev->recovery))
7164 if (mddev->sysfs_action)
7165 sysfs_notify_dirent(mddev->sysfs_action);
7166 }
7167 mddev_unlock(mddev);
7168 }
7169 }
7170
7171 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7172 {
7173 sysfs_notify_dirent(rdev->sysfs_state);
7174 wait_event_timeout(rdev->blocked_wait,
7175 !test_bit(Blocked, &rdev->flags),
7176 msecs_to_jiffies(5000));
7177 rdev_dec_pending(rdev, mddev);
7178 }
7179 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7180
7181 static int md_notify_reboot(struct notifier_block *this,
7182 unsigned long code, void *x)
7183 {
7184 struct list_head *tmp;
7185 mddev_t *mddev;
7186
7187 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7188
7189 printk(KERN_INFO "md: stopping all md devices.\n");
7190
7191 for_each_mddev(mddev, tmp)
7192 if (mddev_trylock(mddev)) {
7193 /* Force a switch to readonly even array
7194 * appears to still be in use. Hence
7195 * the '100'.
7196 */
7197 md_set_readonly(mddev, 100);
7198 mddev_unlock(mddev);
7199 }
7200 /*
7201 * certain more exotic SCSI devices are known to be
7202 * volatile wrt too early system reboots. While the
7203 * right place to handle this issue is the given
7204 * driver, we do want to have a safe RAID driver ...
7205 */
7206 mdelay(1000*1);
7207 }
7208 return NOTIFY_DONE;
7209 }
7210
7211 static struct notifier_block md_notifier = {
7212 .notifier_call = md_notify_reboot,
7213 .next = NULL,
7214 .priority = INT_MAX, /* before any real devices */
7215 };
7216
7217 static void md_geninit(void)
7218 {
7219 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7220
7221 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7222 }
7223
7224 static int __init md_init(void)
7225 {
7226 if (register_blkdev(MD_MAJOR, "md"))
7227 return -1;
7228 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7229 unregister_blkdev(MD_MAJOR, "md");
7230 return -1;
7231 }
7232 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7233 md_probe, NULL, NULL);
7234 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7235 md_probe, NULL, NULL);
7236
7237 register_reboot_notifier(&md_notifier);
7238 raid_table_header = register_sysctl_table(raid_root_table);
7239
7240 md_geninit();
7241 return 0;
7242 }
7243
7244
7245 #ifndef MODULE
7246
7247 /*
7248 * Searches all registered partitions for autorun RAID arrays
7249 * at boot time.
7250 */
7251
7252 static LIST_HEAD(all_detected_devices);
7253 struct detected_devices_node {
7254 struct list_head list;
7255 dev_t dev;
7256 };
7257
7258 void md_autodetect_dev(dev_t dev)
7259 {
7260 struct detected_devices_node *node_detected_dev;
7261
7262 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7263 if (node_detected_dev) {
7264 node_detected_dev->dev = dev;
7265 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7266 } else {
7267 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7268 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7269 }
7270 }
7271
7272
7273 static void autostart_arrays(int part)
7274 {
7275 mdk_rdev_t *rdev;
7276 struct detected_devices_node *node_detected_dev;
7277 dev_t dev;
7278 int i_scanned, i_passed;
7279
7280 i_scanned = 0;
7281 i_passed = 0;
7282
7283 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7284
7285 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7286 i_scanned++;
7287 node_detected_dev = list_entry(all_detected_devices.next,
7288 struct detected_devices_node, list);
7289 list_del(&node_detected_dev->list);
7290 dev = node_detected_dev->dev;
7291 kfree(node_detected_dev);
7292 rdev = md_import_device(dev,0, 90);
7293 if (IS_ERR(rdev))
7294 continue;
7295
7296 if (test_bit(Faulty, &rdev->flags)) {
7297 MD_BUG();
7298 continue;
7299 }
7300 set_bit(AutoDetected, &rdev->flags);
7301 list_add(&rdev->same_set, &pending_raid_disks);
7302 i_passed++;
7303 }
7304
7305 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7306 i_scanned, i_passed);
7307
7308 autorun_devices(part);
7309 }
7310
7311 #endif /* !MODULE */
7312
7313 static __exit void md_exit(void)
7314 {
7315 mddev_t *mddev;
7316 struct list_head *tmp;
7317
7318 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7319 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7320
7321 unregister_blkdev(MD_MAJOR,"md");
7322 unregister_blkdev(mdp_major, "mdp");
7323 unregister_reboot_notifier(&md_notifier);
7324 unregister_sysctl_table(raid_table_header);
7325 remove_proc_entry("mdstat", NULL);
7326 for_each_mddev(mddev, tmp) {
7327 export_array(mddev);
7328 mddev->hold_active = 0;
7329 }
7330 }
7331
7332 subsys_initcall(md_init);
7333 module_exit(md_exit)
7334
7335 static int get_ro(char *buffer, struct kernel_param *kp)
7336 {
7337 return sprintf(buffer, "%d", start_readonly);
7338 }
7339 static int set_ro(const char *val, struct kernel_param *kp)
7340 {
7341 char *e;
7342 int num = simple_strtoul(val, &e, 10);
7343 if (*val && (*e == '\0' || *e == '\n')) {
7344 start_readonly = num;
7345 return 0;
7346 }
7347 return -EINVAL;
7348 }
7349
7350 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7351 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7352
7353 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7354
7355 EXPORT_SYMBOL(register_md_personality);
7356 EXPORT_SYMBOL(unregister_md_personality);
7357 EXPORT_SYMBOL(md_error);
7358 EXPORT_SYMBOL(md_done_sync);
7359 EXPORT_SYMBOL(md_write_start);
7360 EXPORT_SYMBOL(md_write_end);
7361 EXPORT_SYMBOL(md_register_thread);
7362 EXPORT_SYMBOL(md_unregister_thread);
7363 EXPORT_SYMBOL(md_wakeup_thread);
7364 EXPORT_SYMBOL(md_check_recovery);
7365 MODULE_LICENSE("GPL");
7366 MODULE_DESCRIPTION("MD RAID framework");
7367 MODULE_ALIAS("md");
7368 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
ubuntu-zesty-kernel mirror