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