Add linux-next specific files for 20110421
[linux-2.6/next.git] / drivers / md / md.c
blob7d6f7f18a92052e3258ffeae719c377b2df4fb78
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
7 Changes:
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>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
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.
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.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/mutex.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
54 #include "md.h"
55 #include "bitmap.h"
57 #define DEBUG 0
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
60 #ifndef MODULE
61 static void autostart_arrays(int part);
62 #endif
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
67 static void md_print_devices(void);
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70 static struct workqueue_struct *md_wq;
71 static struct workqueue_struct *md_misc_wq;
73 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
80 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
88 * idle IO detection.
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min = 1000;
95 static int sysctl_speed_limit_max = 200000;
96 static inline int speed_min(mddev_t *mddev)
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
102 static inline int speed_max(mddev_t *mddev)
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
108 static struct ctl_table_header *raid_table_header;
110 static ctl_table raid_table[] = {
112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
115 .mode = S_IRUGO|S_IWUSR,
116 .proc_handler = proc_dointvec,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = proc_dointvec,
128 static ctl_table raid_dir_table[] = {
130 .procname = "raid",
131 .maxlen = 0,
132 .mode = S_IRUGO|S_IXUGO,
133 .child = raid_table,
138 static ctl_table raid_root_table[] = {
140 .procname = "dev",
141 .maxlen = 0,
142 .mode = 0555,
143 .child = raid_dir_table,
148 static const struct block_device_operations md_fops;
150 static int start_readonly;
152 /* bio_clone_mddev
153 * like bio_clone, but with a local bio set
156 static void mddev_bio_destructor(struct bio *bio)
158 mddev_t *mddev, **mddevp;
160 mddevp = (void*)bio;
161 mddev = mddevp[-1];
163 bio_free(bio, mddev->bio_set);
166 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
167 mddev_t *mddev)
169 struct bio *b;
170 mddev_t **mddevp;
172 if (!mddev || !mddev->bio_set)
173 return bio_alloc(gfp_mask, nr_iovecs);
175 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
176 mddev->bio_set);
177 if (!b)
178 return NULL;
179 mddevp = (void*)b;
180 mddevp[-1] = mddev;
181 b->bi_destructor = mddev_bio_destructor;
182 return b;
184 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
186 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
187 mddev_t *mddev)
189 struct bio *b;
190 mddev_t **mddevp;
192 if (!mddev || !mddev->bio_set)
193 return bio_clone(bio, gfp_mask);
195 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
196 mddev->bio_set);
197 if (!b)
198 return NULL;
199 mddevp = (void*)b;
200 mddevp[-1] = mddev;
201 b->bi_destructor = mddev_bio_destructor;
202 __bio_clone(b, bio);
203 if (bio_integrity(bio)) {
204 int ret;
206 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
208 if (ret < 0) {
209 bio_put(b);
210 return NULL;
214 return b;
216 EXPORT_SYMBOL_GPL(bio_clone_mddev);
219 * We have a system wide 'event count' that is incremented
220 * on any 'interesting' event, and readers of /proc/mdstat
221 * can use 'poll' or 'select' to find out when the event
222 * count increases.
224 * Events are:
225 * start array, stop array, error, add device, remove device,
226 * start build, activate spare
228 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
229 static atomic_t md_event_count;
230 void md_new_event(mddev_t *mddev)
232 atomic_inc(&md_event_count);
233 wake_up(&md_event_waiters);
235 EXPORT_SYMBOL_GPL(md_new_event);
237 /* Alternate version that can be called from interrupts
238 * when calling sysfs_notify isn't needed.
240 static void md_new_event_inintr(mddev_t *mddev)
242 atomic_inc(&md_event_count);
243 wake_up(&md_event_waiters);
247 * Enables to iterate over all existing md arrays
248 * all_mddevs_lock protects this list.
250 static LIST_HEAD(all_mddevs);
251 static DEFINE_SPINLOCK(all_mddevs_lock);
255 * iterates through all used mddevs in the system.
256 * We take care to grab the all_mddevs_lock whenever navigating
257 * the list, and to always hold a refcount when unlocked.
258 * Any code which breaks out of this loop while own
259 * a reference to the current mddev and must mddev_put it.
261 #define for_each_mddev(mddev,tmp) \
263 for (({ spin_lock(&all_mddevs_lock); \
264 tmp = all_mddevs.next; \
265 mddev = NULL;}); \
266 ({ if (tmp != &all_mddevs) \
267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
268 spin_unlock(&all_mddevs_lock); \
269 if (mddev) mddev_put(mddev); \
270 mddev = list_entry(tmp, mddev_t, all_mddevs); \
271 tmp != &all_mddevs;}); \
272 ({ spin_lock(&all_mddevs_lock); \
273 tmp = tmp->next;}) \
277 /* Rather than calling directly into the personality make_request function,
278 * IO requests come here first so that we can check if the device is
279 * being suspended pending a reconfiguration.
280 * We hold a refcount over the call to ->make_request. By the time that
281 * call has finished, the bio has been linked into some internal structure
282 * and so is visible to ->quiesce(), so we don't need the refcount any more.
284 static int md_make_request(struct request_queue *q, struct bio *bio)
286 const int rw = bio_data_dir(bio);
287 mddev_t *mddev = q->queuedata;
288 int rv;
289 int cpu;
290 unsigned int sectors;
292 if (mddev == NULL || mddev->pers == NULL
293 || !mddev->ready) {
294 bio_io_error(bio);
295 return 0;
297 smp_rmb(); /* Ensure implications of 'active' are visible */
298 rcu_read_lock();
299 if (mddev->suspended) {
300 DEFINE_WAIT(__wait);
301 for (;;) {
302 prepare_to_wait(&mddev->sb_wait, &__wait,
303 TASK_UNINTERRUPTIBLE);
304 if (!mddev->suspended)
305 break;
306 rcu_read_unlock();
307 schedule();
308 rcu_read_lock();
310 finish_wait(&mddev->sb_wait, &__wait);
312 atomic_inc(&mddev->active_io);
313 rcu_read_unlock();
316 * save the sectors now since our bio can
317 * go away inside make_request
319 sectors = bio_sectors(bio);
320 rv = mddev->pers->make_request(mddev, bio);
322 cpu = part_stat_lock();
323 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
324 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
325 part_stat_unlock();
327 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
328 wake_up(&mddev->sb_wait);
330 return rv;
333 /* mddev_suspend makes sure no new requests are submitted
334 * to the device, and that any requests that have been submitted
335 * are completely handled.
336 * Once ->stop is called and completes, the module will be completely
337 * unused.
339 void mddev_suspend(mddev_t *mddev)
341 BUG_ON(mddev->suspended);
342 mddev->suspended = 1;
343 synchronize_rcu();
344 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
345 mddev->pers->quiesce(mddev, 1);
347 EXPORT_SYMBOL_GPL(mddev_suspend);
349 void mddev_resume(mddev_t *mddev)
351 mddev->suspended = 0;
352 wake_up(&mddev->sb_wait);
353 mddev->pers->quiesce(mddev, 0);
355 EXPORT_SYMBOL_GPL(mddev_resume);
357 int mddev_congested(mddev_t *mddev, int bits)
359 return mddev->suspended;
361 EXPORT_SYMBOL(mddev_congested);
364 * Generic flush handling for md
367 static void md_end_flush(struct bio *bio, int err)
369 mdk_rdev_t *rdev = bio->bi_private;
370 mddev_t *mddev = rdev->mddev;
372 rdev_dec_pending(rdev, mddev);
374 if (atomic_dec_and_test(&mddev->flush_pending)) {
375 /* The pre-request flush has finished */
376 queue_work(md_wq, &mddev->flush_work);
378 bio_put(bio);
381 static void md_submit_flush_data(struct work_struct *ws);
383 static void submit_flushes(struct work_struct *ws)
385 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
386 mdk_rdev_t *rdev;
388 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
389 atomic_set(&mddev->flush_pending, 1);
390 rcu_read_lock();
391 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
392 if (rdev->raid_disk >= 0 &&
393 !test_bit(Faulty, &rdev->flags)) {
394 /* Take two references, one is dropped
395 * when request finishes, one after
396 * we reclaim rcu_read_lock
398 struct bio *bi;
399 atomic_inc(&rdev->nr_pending);
400 atomic_inc(&rdev->nr_pending);
401 rcu_read_unlock();
402 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
403 bi->bi_end_io = md_end_flush;
404 bi->bi_private = rdev;
405 bi->bi_bdev = rdev->bdev;
406 atomic_inc(&mddev->flush_pending);
407 submit_bio(WRITE_FLUSH, bi);
408 rcu_read_lock();
409 rdev_dec_pending(rdev, mddev);
411 rcu_read_unlock();
412 if (atomic_dec_and_test(&mddev->flush_pending))
413 queue_work(md_wq, &mddev->flush_work);
416 static void md_submit_flush_data(struct work_struct *ws)
418 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
419 struct bio *bio = mddev->flush_bio;
421 if (bio->bi_size == 0)
422 /* an empty barrier - all done */
423 bio_endio(bio, 0);
424 else {
425 bio->bi_rw &= ~REQ_FLUSH;
426 if (mddev->pers->make_request(mddev, bio))
427 generic_make_request(bio);
430 mddev->flush_bio = NULL;
431 wake_up(&mddev->sb_wait);
434 void md_flush_request(mddev_t *mddev, struct bio *bio)
436 spin_lock_irq(&mddev->write_lock);
437 wait_event_lock_irq(mddev->sb_wait,
438 !mddev->flush_bio,
439 mddev->write_lock, /*nothing*/);
440 mddev->flush_bio = bio;
441 spin_unlock_irq(&mddev->write_lock);
443 INIT_WORK(&mddev->flush_work, submit_flushes);
444 queue_work(md_wq, &mddev->flush_work);
446 EXPORT_SYMBOL(md_flush_request);
448 /* Support for plugging.
449 * This mirrors the plugging support in request_queue, but does not
450 * require having a whole queue or request structures.
451 * We allocate an md_plug_cb for each md device and each thread it gets
452 * plugged on. This links tot the private plug_handle structure in the
453 * personality data where we keep a count of the number of outstanding
454 * plugs so other code can see if a plug is active.
456 struct md_plug_cb {
457 struct blk_plug_cb cb;
458 mddev_t *mddev;
461 static void plugger_unplug(struct blk_plug_cb *cb)
463 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
464 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
465 md_wakeup_thread(mdcb->mddev->thread);
466 kfree(mdcb);
469 /* Check that an unplug wakeup will come shortly.
470 * If not, wakeup the md thread immediately
472 int mddev_check_plugged(mddev_t *mddev)
474 struct blk_plug *plug = current->plug;
475 struct md_plug_cb *mdcb;
477 if (!plug)
478 return 0;
480 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
481 if (mdcb->cb.callback == plugger_unplug &&
482 mdcb->mddev == mddev) {
483 /* Already on the list, move to top */
484 if (mdcb != list_first_entry(&plug->cb_list,
485 struct md_plug_cb,
486 cb.list))
487 list_move(&mdcb->cb.list, &plug->cb_list);
488 return 1;
491 /* Not currently on the callback list */
492 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
493 if (!mdcb)
494 return 0;
496 mdcb->mddev = mddev;
497 mdcb->cb.callback = plugger_unplug;
498 atomic_inc(&mddev->plug_cnt);
499 list_add(&mdcb->cb.list, &plug->cb_list);
500 return 1;
502 EXPORT_SYMBOL_GPL(mddev_check_plugged);
504 static inline mddev_t *mddev_get(mddev_t *mddev)
506 atomic_inc(&mddev->active);
507 return mddev;
510 static void mddev_delayed_delete(struct work_struct *ws);
512 static void mddev_put(mddev_t *mddev)
514 struct bio_set *bs = NULL;
516 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
517 return;
518 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
519 mddev->ctime == 0 && !mddev->hold_active) {
520 /* Array is not configured at all, and not held active,
521 * so destroy it */
522 list_del(&mddev->all_mddevs);
523 bs = mddev->bio_set;
524 mddev->bio_set = NULL;
525 if (mddev->gendisk) {
526 /* We did a probe so need to clean up. Call
527 * queue_work inside the spinlock so that
528 * flush_workqueue() after mddev_find will
529 * succeed in waiting for the work to be done.
531 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
532 queue_work(md_misc_wq, &mddev->del_work);
533 } else
534 kfree(mddev);
536 spin_unlock(&all_mddevs_lock);
537 if (bs)
538 bioset_free(bs);
541 void mddev_init(mddev_t *mddev)
543 mutex_init(&mddev->open_mutex);
544 mutex_init(&mddev->reconfig_mutex);
545 mutex_init(&mddev->bitmap_info.mutex);
546 INIT_LIST_HEAD(&mddev->disks);
547 INIT_LIST_HEAD(&mddev->all_mddevs);
548 init_timer(&mddev->safemode_timer);
549 atomic_set(&mddev->active, 1);
550 atomic_set(&mddev->openers, 0);
551 atomic_set(&mddev->active_io, 0);
552 atomic_set(&mddev->plug_cnt, 0);
553 spin_lock_init(&mddev->write_lock);
554 atomic_set(&mddev->flush_pending, 0);
555 init_waitqueue_head(&mddev->sb_wait);
556 init_waitqueue_head(&mddev->recovery_wait);
557 mddev->reshape_position = MaxSector;
558 mddev->resync_min = 0;
559 mddev->resync_max = MaxSector;
560 mddev->level = LEVEL_NONE;
562 EXPORT_SYMBOL_GPL(mddev_init);
564 static mddev_t * mddev_find(dev_t unit)
566 mddev_t *mddev, *new = NULL;
568 if (unit && MAJOR(unit) != MD_MAJOR)
569 unit &= ~((1<<MdpMinorShift)-1);
571 retry:
572 spin_lock(&all_mddevs_lock);
574 if (unit) {
575 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
576 if (mddev->unit == unit) {
577 mddev_get(mddev);
578 spin_unlock(&all_mddevs_lock);
579 kfree(new);
580 return mddev;
583 if (new) {
584 list_add(&new->all_mddevs, &all_mddevs);
585 spin_unlock(&all_mddevs_lock);
586 new->hold_active = UNTIL_IOCTL;
587 return new;
589 } else if (new) {
590 /* find an unused unit number */
591 static int next_minor = 512;
592 int start = next_minor;
593 int is_free = 0;
594 int dev = 0;
595 while (!is_free) {
596 dev = MKDEV(MD_MAJOR, next_minor);
597 next_minor++;
598 if (next_minor > MINORMASK)
599 next_minor = 0;
600 if (next_minor == start) {
601 /* Oh dear, all in use. */
602 spin_unlock(&all_mddevs_lock);
603 kfree(new);
604 return NULL;
607 is_free = 1;
608 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
609 if (mddev->unit == dev) {
610 is_free = 0;
611 break;
614 new->unit = dev;
615 new->md_minor = MINOR(dev);
616 new->hold_active = UNTIL_STOP;
617 list_add(&new->all_mddevs, &all_mddevs);
618 spin_unlock(&all_mddevs_lock);
619 return new;
621 spin_unlock(&all_mddevs_lock);
623 new = kzalloc(sizeof(*new), GFP_KERNEL);
624 if (!new)
625 return NULL;
627 new->unit = unit;
628 if (MAJOR(unit) == MD_MAJOR)
629 new->md_minor = MINOR(unit);
630 else
631 new->md_minor = MINOR(unit) >> MdpMinorShift;
633 mddev_init(new);
635 goto retry;
638 static inline int mddev_lock(mddev_t * mddev)
640 return mutex_lock_interruptible(&mddev->reconfig_mutex);
643 static inline int mddev_is_locked(mddev_t *mddev)
645 return mutex_is_locked(&mddev->reconfig_mutex);
648 static inline int mddev_trylock(mddev_t * mddev)
650 return mutex_trylock(&mddev->reconfig_mutex);
653 static struct attribute_group md_redundancy_group;
655 static void mddev_unlock(mddev_t * mddev)
657 if (mddev->to_remove) {
658 /* These cannot be removed under reconfig_mutex as
659 * an access to the files will try to take reconfig_mutex
660 * while holding the file unremovable, which leads to
661 * a deadlock.
662 * So hold set sysfs_active while the remove in happeing,
663 * and anything else which might set ->to_remove or my
664 * otherwise change the sysfs namespace will fail with
665 * -EBUSY if sysfs_active is still set.
666 * We set sysfs_active under reconfig_mutex and elsewhere
667 * test it under the same mutex to ensure its correct value
668 * is seen.
670 struct attribute_group *to_remove = mddev->to_remove;
671 mddev->to_remove = NULL;
672 mddev->sysfs_active = 1;
673 mutex_unlock(&mddev->reconfig_mutex);
675 if (mddev->kobj.sd) {
676 if (to_remove != &md_redundancy_group)
677 sysfs_remove_group(&mddev->kobj, to_remove);
678 if (mddev->pers == NULL ||
679 mddev->pers->sync_request == NULL) {
680 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
681 if (mddev->sysfs_action)
682 sysfs_put(mddev->sysfs_action);
683 mddev->sysfs_action = NULL;
686 mddev->sysfs_active = 0;
687 } else
688 mutex_unlock(&mddev->reconfig_mutex);
690 md_wakeup_thread(mddev->thread);
693 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
695 mdk_rdev_t *rdev;
697 list_for_each_entry(rdev, &mddev->disks, same_set)
698 if (rdev->desc_nr == nr)
699 return rdev;
701 return NULL;
704 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
706 mdk_rdev_t *rdev;
708 list_for_each_entry(rdev, &mddev->disks, same_set)
709 if (rdev->bdev->bd_dev == dev)
710 return rdev;
712 return NULL;
715 static struct mdk_personality *find_pers(int level, char *clevel)
717 struct mdk_personality *pers;
718 list_for_each_entry(pers, &pers_list, list) {
719 if (level != LEVEL_NONE && pers->level == level)
720 return pers;
721 if (strcmp(pers->name, clevel)==0)
722 return pers;
724 return NULL;
727 /* return the offset of the super block in 512byte sectors */
728 static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
730 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
731 return MD_NEW_SIZE_SECTORS(num_sectors);
734 static int alloc_disk_sb(mdk_rdev_t * rdev)
736 if (rdev->sb_page)
737 MD_BUG();
739 rdev->sb_page = alloc_page(GFP_KERNEL);
740 if (!rdev->sb_page) {
741 printk(KERN_ALERT "md: out of memory.\n");
742 return -ENOMEM;
745 return 0;
748 static void free_disk_sb(mdk_rdev_t * rdev)
750 if (rdev->sb_page) {
751 put_page(rdev->sb_page);
752 rdev->sb_loaded = 0;
753 rdev->sb_page = NULL;
754 rdev->sb_start = 0;
755 rdev->sectors = 0;
760 static void super_written(struct bio *bio, int error)
762 mdk_rdev_t *rdev = bio->bi_private;
763 mddev_t *mddev = rdev->mddev;
765 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
766 printk("md: super_written gets error=%d, uptodate=%d\n",
767 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
768 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
769 md_error(mddev, rdev);
772 if (atomic_dec_and_test(&mddev->pending_writes))
773 wake_up(&mddev->sb_wait);
774 bio_put(bio);
777 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
778 sector_t sector, int size, struct page *page)
780 /* write first size bytes of page to sector of rdev
781 * Increment mddev->pending_writes before returning
782 * and decrement it on completion, waking up sb_wait
783 * if zero is reached.
784 * If an error occurred, call md_error
786 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
788 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
789 bio->bi_sector = sector;
790 bio_add_page(bio, page, size, 0);
791 bio->bi_private = rdev;
792 bio->bi_end_io = super_written;
794 atomic_inc(&mddev->pending_writes);
795 submit_bio(REQ_WRITE | REQ_SYNC | REQ_FLUSH | REQ_FUA, bio);
798 void md_super_wait(mddev_t *mddev)
800 /* wait for all superblock writes that were scheduled to complete */
801 DEFINE_WAIT(wq);
802 for(;;) {
803 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
804 if (atomic_read(&mddev->pending_writes)==0)
805 break;
806 schedule();
808 finish_wait(&mddev->sb_wait, &wq);
811 static void bi_complete(struct bio *bio, int error)
813 complete((struct completion*)bio->bi_private);
816 int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
817 struct page *page, int rw, bool metadata_op)
819 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
820 struct completion event;
821 int ret;
823 rw |= REQ_SYNC;
825 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
826 rdev->meta_bdev : rdev->bdev;
827 if (metadata_op)
828 bio->bi_sector = sector + rdev->sb_start;
829 else
830 bio->bi_sector = sector + rdev->data_offset;
831 bio_add_page(bio, page, size, 0);
832 init_completion(&event);
833 bio->bi_private = &event;
834 bio->bi_end_io = bi_complete;
835 submit_bio(rw, bio);
836 wait_for_completion(&event);
838 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
839 bio_put(bio);
840 return ret;
842 EXPORT_SYMBOL_GPL(sync_page_io);
844 static int read_disk_sb(mdk_rdev_t * rdev, int size)
846 char b[BDEVNAME_SIZE];
847 if (!rdev->sb_page) {
848 MD_BUG();
849 return -EINVAL;
851 if (rdev->sb_loaded)
852 return 0;
855 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
856 goto fail;
857 rdev->sb_loaded = 1;
858 return 0;
860 fail:
861 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
862 bdevname(rdev->bdev,b));
863 return -EINVAL;
866 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
868 return sb1->set_uuid0 == sb2->set_uuid0 &&
869 sb1->set_uuid1 == sb2->set_uuid1 &&
870 sb1->set_uuid2 == sb2->set_uuid2 &&
871 sb1->set_uuid3 == sb2->set_uuid3;
874 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
876 int ret;
877 mdp_super_t *tmp1, *tmp2;
879 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
880 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
882 if (!tmp1 || !tmp2) {
883 ret = 0;
884 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
885 goto abort;
888 *tmp1 = *sb1;
889 *tmp2 = *sb2;
892 * nr_disks is not constant
894 tmp1->nr_disks = 0;
895 tmp2->nr_disks = 0;
897 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
898 abort:
899 kfree(tmp1);
900 kfree(tmp2);
901 return ret;
905 static u32 md_csum_fold(u32 csum)
907 csum = (csum & 0xffff) + (csum >> 16);
908 return (csum & 0xffff) + (csum >> 16);
911 static unsigned int calc_sb_csum(mdp_super_t * sb)
913 u64 newcsum = 0;
914 u32 *sb32 = (u32*)sb;
915 int i;
916 unsigned int disk_csum, csum;
918 disk_csum = sb->sb_csum;
919 sb->sb_csum = 0;
921 for (i = 0; i < MD_SB_BYTES/4 ; i++)
922 newcsum += sb32[i];
923 csum = (newcsum & 0xffffffff) + (newcsum>>32);
926 #ifdef CONFIG_ALPHA
927 /* This used to use csum_partial, which was wrong for several
928 * reasons including that different results are returned on
929 * different architectures. It isn't critical that we get exactly
930 * the same return value as before (we always csum_fold before
931 * testing, and that removes any differences). However as we
932 * know that csum_partial always returned a 16bit value on
933 * alphas, do a fold to maximise conformity to previous behaviour.
935 sb->sb_csum = md_csum_fold(disk_csum);
936 #else
937 sb->sb_csum = disk_csum;
938 #endif
939 return csum;
944 * Handle superblock details.
945 * We want to be able to handle multiple superblock formats
946 * so we have a common interface to them all, and an array of
947 * different handlers.
948 * We rely on user-space to write the initial superblock, and support
949 * reading and updating of superblocks.
950 * Interface methods are:
951 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
952 * loads and validates a superblock on dev.
953 * if refdev != NULL, compare superblocks on both devices
954 * Return:
955 * 0 - dev has a superblock that is compatible with refdev
956 * 1 - dev has a superblock that is compatible and newer than refdev
957 * so dev should be used as the refdev in future
958 * -EINVAL superblock incompatible or invalid
959 * -othererror e.g. -EIO
961 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
962 * Verify that dev is acceptable into mddev.
963 * The first time, mddev->raid_disks will be 0, and data from
964 * dev should be merged in. Subsequent calls check that dev
965 * is new enough. Return 0 or -EINVAL
967 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
968 * Update the superblock for rdev with data in mddev
969 * This does not write to disc.
973 struct super_type {
974 char *name;
975 struct module *owner;
976 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
977 int minor_version);
978 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
979 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
980 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
981 sector_t num_sectors);
985 * Check that the given mddev has no bitmap.
987 * This function is called from the run method of all personalities that do not
988 * support bitmaps. It prints an error message and returns non-zero if mddev
989 * has a bitmap. Otherwise, it returns 0.
992 int md_check_no_bitmap(mddev_t *mddev)
994 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
995 return 0;
996 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
997 mdname(mddev), mddev->pers->name);
998 return 1;
1000 EXPORT_SYMBOL(md_check_no_bitmap);
1003 * load_super for 0.90.0
1005 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1007 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1008 mdp_super_t *sb;
1009 int ret;
1012 * Calculate the position of the superblock (512byte sectors),
1013 * it's at the end of the disk.
1015 * It also happens to be a multiple of 4Kb.
1017 rdev->sb_start = calc_dev_sboffset(rdev);
1019 ret = read_disk_sb(rdev, MD_SB_BYTES);
1020 if (ret) return ret;
1022 ret = -EINVAL;
1024 bdevname(rdev->bdev, b);
1025 sb = (mdp_super_t*)page_address(rdev->sb_page);
1027 if (sb->md_magic != MD_SB_MAGIC) {
1028 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1030 goto abort;
1033 if (sb->major_version != 0 ||
1034 sb->minor_version < 90 ||
1035 sb->minor_version > 91) {
1036 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1037 sb->major_version, sb->minor_version,
1039 goto abort;
1042 if (sb->raid_disks <= 0)
1043 goto abort;
1045 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1046 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1048 goto abort;
1051 rdev->preferred_minor = sb->md_minor;
1052 rdev->data_offset = 0;
1053 rdev->sb_size = MD_SB_BYTES;
1055 if (sb->level == LEVEL_MULTIPATH)
1056 rdev->desc_nr = -1;
1057 else
1058 rdev->desc_nr = sb->this_disk.number;
1060 if (!refdev) {
1061 ret = 1;
1062 } else {
1063 __u64 ev1, ev2;
1064 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1065 if (!uuid_equal(refsb, sb)) {
1066 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1067 b, bdevname(refdev->bdev,b2));
1068 goto abort;
1070 if (!sb_equal(refsb, sb)) {
1071 printk(KERN_WARNING "md: %s has same UUID"
1072 " but different superblock to %s\n",
1073 b, bdevname(refdev->bdev, b2));
1074 goto abort;
1076 ev1 = md_event(sb);
1077 ev2 = md_event(refsb);
1078 if (ev1 > ev2)
1079 ret = 1;
1080 else
1081 ret = 0;
1083 rdev->sectors = rdev->sb_start;
1085 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1086 /* "this cannot possibly happen" ... */
1087 ret = -EINVAL;
1089 abort:
1090 return ret;
1094 * validate_super for 0.90.0
1096 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1098 mdp_disk_t *desc;
1099 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1100 __u64 ev1 = md_event(sb);
1102 rdev->raid_disk = -1;
1103 clear_bit(Faulty, &rdev->flags);
1104 clear_bit(In_sync, &rdev->flags);
1105 clear_bit(WriteMostly, &rdev->flags);
1107 if (mddev->raid_disks == 0) {
1108 mddev->major_version = 0;
1109 mddev->minor_version = sb->minor_version;
1110 mddev->patch_version = sb->patch_version;
1111 mddev->external = 0;
1112 mddev->chunk_sectors = sb->chunk_size >> 9;
1113 mddev->ctime = sb->ctime;
1114 mddev->utime = sb->utime;
1115 mddev->level = sb->level;
1116 mddev->clevel[0] = 0;
1117 mddev->layout = sb->layout;
1118 mddev->raid_disks = sb->raid_disks;
1119 mddev->dev_sectors = sb->size * 2;
1120 mddev->events = ev1;
1121 mddev->bitmap_info.offset = 0;
1122 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1124 if (mddev->minor_version >= 91) {
1125 mddev->reshape_position = sb->reshape_position;
1126 mddev->delta_disks = sb->delta_disks;
1127 mddev->new_level = sb->new_level;
1128 mddev->new_layout = sb->new_layout;
1129 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1130 } else {
1131 mddev->reshape_position = MaxSector;
1132 mddev->delta_disks = 0;
1133 mddev->new_level = mddev->level;
1134 mddev->new_layout = mddev->layout;
1135 mddev->new_chunk_sectors = mddev->chunk_sectors;
1138 if (sb->state & (1<<MD_SB_CLEAN))
1139 mddev->recovery_cp = MaxSector;
1140 else {
1141 if (sb->events_hi == sb->cp_events_hi &&
1142 sb->events_lo == sb->cp_events_lo) {
1143 mddev->recovery_cp = sb->recovery_cp;
1144 } else
1145 mddev->recovery_cp = 0;
1148 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1149 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1150 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1151 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1153 mddev->max_disks = MD_SB_DISKS;
1155 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1156 mddev->bitmap_info.file == NULL)
1157 mddev->bitmap_info.offset =
1158 mddev->bitmap_info.default_offset;
1160 } else if (mddev->pers == NULL) {
1161 /* Insist on good event counter while assembling, except
1162 * for spares (which don't need an event count) */
1163 ++ev1;
1164 if (sb->disks[rdev->desc_nr].state & (
1165 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1166 if (ev1 < mddev->events)
1167 return -EINVAL;
1168 } else if (mddev->bitmap) {
1169 /* if adding to array with a bitmap, then we can accept an
1170 * older device ... but not too old.
1172 if (ev1 < mddev->bitmap->events_cleared)
1173 return 0;
1174 } else {
1175 if (ev1 < mddev->events)
1176 /* just a hot-add of a new device, leave raid_disk at -1 */
1177 return 0;
1180 if (mddev->level != LEVEL_MULTIPATH) {
1181 desc = sb->disks + rdev->desc_nr;
1183 if (desc->state & (1<<MD_DISK_FAULTY))
1184 set_bit(Faulty, &rdev->flags);
1185 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1186 desc->raid_disk < mddev->raid_disks */) {
1187 set_bit(In_sync, &rdev->flags);
1188 rdev->raid_disk = desc->raid_disk;
1189 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1190 /* active but not in sync implies recovery up to
1191 * reshape position. We don't know exactly where
1192 * that is, so set to zero for now */
1193 if (mddev->minor_version >= 91) {
1194 rdev->recovery_offset = 0;
1195 rdev->raid_disk = desc->raid_disk;
1198 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1199 set_bit(WriteMostly, &rdev->flags);
1200 } else /* MULTIPATH are always insync */
1201 set_bit(In_sync, &rdev->flags);
1202 return 0;
1206 * sync_super for 0.90.0
1208 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1210 mdp_super_t *sb;
1211 mdk_rdev_t *rdev2;
1212 int next_spare = mddev->raid_disks;
1215 /* make rdev->sb match mddev data..
1217 * 1/ zero out disks
1218 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1219 * 3/ any empty disks < next_spare become removed
1221 * disks[0] gets initialised to REMOVED because
1222 * we cannot be sure from other fields if it has
1223 * been initialised or not.
1225 int i;
1226 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1228 rdev->sb_size = MD_SB_BYTES;
1230 sb = (mdp_super_t*)page_address(rdev->sb_page);
1232 memset(sb, 0, sizeof(*sb));
1234 sb->md_magic = MD_SB_MAGIC;
1235 sb->major_version = mddev->major_version;
1236 sb->patch_version = mddev->patch_version;
1237 sb->gvalid_words = 0; /* ignored */
1238 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1239 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1240 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1241 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1243 sb->ctime = mddev->ctime;
1244 sb->level = mddev->level;
1245 sb->size = mddev->dev_sectors / 2;
1246 sb->raid_disks = mddev->raid_disks;
1247 sb->md_minor = mddev->md_minor;
1248 sb->not_persistent = 0;
1249 sb->utime = mddev->utime;
1250 sb->state = 0;
1251 sb->events_hi = (mddev->events>>32);
1252 sb->events_lo = (u32)mddev->events;
1254 if (mddev->reshape_position == MaxSector)
1255 sb->minor_version = 90;
1256 else {
1257 sb->minor_version = 91;
1258 sb->reshape_position = mddev->reshape_position;
1259 sb->new_level = mddev->new_level;
1260 sb->delta_disks = mddev->delta_disks;
1261 sb->new_layout = mddev->new_layout;
1262 sb->new_chunk = mddev->new_chunk_sectors << 9;
1264 mddev->minor_version = sb->minor_version;
1265 if (mddev->in_sync)
1267 sb->recovery_cp = mddev->recovery_cp;
1268 sb->cp_events_hi = (mddev->events>>32);
1269 sb->cp_events_lo = (u32)mddev->events;
1270 if (mddev->recovery_cp == MaxSector)
1271 sb->state = (1<< MD_SB_CLEAN);
1272 } else
1273 sb->recovery_cp = 0;
1275 sb->layout = mddev->layout;
1276 sb->chunk_size = mddev->chunk_sectors << 9;
1278 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1279 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1281 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1282 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1283 mdp_disk_t *d;
1284 int desc_nr;
1285 int is_active = test_bit(In_sync, &rdev2->flags);
1287 if (rdev2->raid_disk >= 0 &&
1288 sb->minor_version >= 91)
1289 /* we have nowhere to store the recovery_offset,
1290 * but if it is not below the reshape_position,
1291 * we can piggy-back on that.
1293 is_active = 1;
1294 if (rdev2->raid_disk < 0 ||
1295 test_bit(Faulty, &rdev2->flags))
1296 is_active = 0;
1297 if (is_active)
1298 desc_nr = rdev2->raid_disk;
1299 else
1300 desc_nr = next_spare++;
1301 rdev2->desc_nr = desc_nr;
1302 d = &sb->disks[rdev2->desc_nr];
1303 nr_disks++;
1304 d->number = rdev2->desc_nr;
1305 d->major = MAJOR(rdev2->bdev->bd_dev);
1306 d->minor = MINOR(rdev2->bdev->bd_dev);
1307 if (is_active)
1308 d->raid_disk = rdev2->raid_disk;
1309 else
1310 d->raid_disk = rdev2->desc_nr; /* compatibility */
1311 if (test_bit(Faulty, &rdev2->flags))
1312 d->state = (1<<MD_DISK_FAULTY);
1313 else if (is_active) {
1314 d->state = (1<<MD_DISK_ACTIVE);
1315 if (test_bit(In_sync, &rdev2->flags))
1316 d->state |= (1<<MD_DISK_SYNC);
1317 active++;
1318 working++;
1319 } else {
1320 d->state = 0;
1321 spare++;
1322 working++;
1324 if (test_bit(WriteMostly, &rdev2->flags))
1325 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1327 /* now set the "removed" and "faulty" bits on any missing devices */
1328 for (i=0 ; i < mddev->raid_disks ; i++) {
1329 mdp_disk_t *d = &sb->disks[i];
1330 if (d->state == 0 && d->number == 0) {
1331 d->number = i;
1332 d->raid_disk = i;
1333 d->state = (1<<MD_DISK_REMOVED);
1334 d->state |= (1<<MD_DISK_FAULTY);
1335 failed++;
1338 sb->nr_disks = nr_disks;
1339 sb->active_disks = active;
1340 sb->working_disks = working;
1341 sb->failed_disks = failed;
1342 sb->spare_disks = spare;
1344 sb->this_disk = sb->disks[rdev->desc_nr];
1345 sb->sb_csum = calc_sb_csum(sb);
1349 * rdev_size_change for 0.90.0
1351 static unsigned long long
1352 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1354 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1355 return 0; /* component must fit device */
1356 if (rdev->mddev->bitmap_info.offset)
1357 return 0; /* can't move bitmap */
1358 rdev->sb_start = calc_dev_sboffset(rdev);
1359 if (!num_sectors || num_sectors > rdev->sb_start)
1360 num_sectors = rdev->sb_start;
1361 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1362 rdev->sb_page);
1363 md_super_wait(rdev->mddev);
1364 return num_sectors;
1369 * version 1 superblock
1372 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1374 __le32 disk_csum;
1375 u32 csum;
1376 unsigned long long newcsum;
1377 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1378 __le32 *isuper = (__le32*)sb;
1379 int i;
1381 disk_csum = sb->sb_csum;
1382 sb->sb_csum = 0;
1383 newcsum = 0;
1384 for (i=0; size>=4; size -= 4 )
1385 newcsum += le32_to_cpu(*isuper++);
1387 if (size == 2)
1388 newcsum += le16_to_cpu(*(__le16*) isuper);
1390 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1391 sb->sb_csum = disk_csum;
1392 return cpu_to_le32(csum);
1395 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1397 struct mdp_superblock_1 *sb;
1398 int ret;
1399 sector_t sb_start;
1400 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1401 int bmask;
1404 * Calculate the position of the superblock in 512byte sectors.
1405 * It is always aligned to a 4K boundary and
1406 * depeding on minor_version, it can be:
1407 * 0: At least 8K, but less than 12K, from end of device
1408 * 1: At start of device
1409 * 2: 4K from start of device.
1411 switch(minor_version) {
1412 case 0:
1413 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1414 sb_start -= 8*2;
1415 sb_start &= ~(sector_t)(4*2-1);
1416 break;
1417 case 1:
1418 sb_start = 0;
1419 break;
1420 case 2:
1421 sb_start = 8;
1422 break;
1423 default:
1424 return -EINVAL;
1426 rdev->sb_start = sb_start;
1428 /* superblock is rarely larger than 1K, but it can be larger,
1429 * and it is safe to read 4k, so we do that
1431 ret = read_disk_sb(rdev, 4096);
1432 if (ret) return ret;
1435 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1437 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1438 sb->major_version != cpu_to_le32(1) ||
1439 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1440 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1441 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1442 return -EINVAL;
1444 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1445 printk("md: invalid superblock checksum on %s\n",
1446 bdevname(rdev->bdev,b));
1447 return -EINVAL;
1449 if (le64_to_cpu(sb->data_size) < 10) {
1450 printk("md: data_size too small on %s\n",
1451 bdevname(rdev->bdev,b));
1452 return -EINVAL;
1455 rdev->preferred_minor = 0xffff;
1456 rdev->data_offset = le64_to_cpu(sb->data_offset);
1457 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1459 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1460 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1461 if (rdev->sb_size & bmask)
1462 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1464 if (minor_version
1465 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1466 return -EINVAL;
1468 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1469 rdev->desc_nr = -1;
1470 else
1471 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1473 if (!refdev) {
1474 ret = 1;
1475 } else {
1476 __u64 ev1, ev2;
1477 struct mdp_superblock_1 *refsb =
1478 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1480 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1481 sb->level != refsb->level ||
1482 sb->layout != refsb->layout ||
1483 sb->chunksize != refsb->chunksize) {
1484 printk(KERN_WARNING "md: %s has strangely different"
1485 " superblock to %s\n",
1486 bdevname(rdev->bdev,b),
1487 bdevname(refdev->bdev,b2));
1488 return -EINVAL;
1490 ev1 = le64_to_cpu(sb->events);
1491 ev2 = le64_to_cpu(refsb->events);
1493 if (ev1 > ev2)
1494 ret = 1;
1495 else
1496 ret = 0;
1498 if (minor_version)
1499 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1500 le64_to_cpu(sb->data_offset);
1501 else
1502 rdev->sectors = rdev->sb_start;
1503 if (rdev->sectors < le64_to_cpu(sb->data_size))
1504 return -EINVAL;
1505 rdev->sectors = le64_to_cpu(sb->data_size);
1506 if (le64_to_cpu(sb->size) > rdev->sectors)
1507 return -EINVAL;
1508 return ret;
1511 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1513 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1514 __u64 ev1 = le64_to_cpu(sb->events);
1516 rdev->raid_disk = -1;
1517 clear_bit(Faulty, &rdev->flags);
1518 clear_bit(In_sync, &rdev->flags);
1519 clear_bit(WriteMostly, &rdev->flags);
1521 if (mddev->raid_disks == 0) {
1522 mddev->major_version = 1;
1523 mddev->patch_version = 0;
1524 mddev->external = 0;
1525 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1526 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1527 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1528 mddev->level = le32_to_cpu(sb->level);
1529 mddev->clevel[0] = 0;
1530 mddev->layout = le32_to_cpu(sb->layout);
1531 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1532 mddev->dev_sectors = le64_to_cpu(sb->size);
1533 mddev->events = ev1;
1534 mddev->bitmap_info.offset = 0;
1535 mddev->bitmap_info.default_offset = 1024 >> 9;
1537 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1538 memcpy(mddev->uuid, sb->set_uuid, 16);
1540 mddev->max_disks = (4096-256)/2;
1542 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1543 mddev->bitmap_info.file == NULL )
1544 mddev->bitmap_info.offset =
1545 (__s32)le32_to_cpu(sb->bitmap_offset);
1547 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1548 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1549 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1550 mddev->new_level = le32_to_cpu(sb->new_level);
1551 mddev->new_layout = le32_to_cpu(sb->new_layout);
1552 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1553 } else {
1554 mddev->reshape_position = MaxSector;
1555 mddev->delta_disks = 0;
1556 mddev->new_level = mddev->level;
1557 mddev->new_layout = mddev->layout;
1558 mddev->new_chunk_sectors = mddev->chunk_sectors;
1561 } else if (mddev->pers == NULL) {
1562 /* Insist of good event counter while assembling, except for
1563 * spares (which don't need an event count) */
1564 ++ev1;
1565 if (rdev->desc_nr >= 0 &&
1566 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1567 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1568 if (ev1 < mddev->events)
1569 return -EINVAL;
1570 } else if (mddev->bitmap) {
1571 /* If adding to array with a bitmap, then we can accept an
1572 * older device, but not too old.
1574 if (ev1 < mddev->bitmap->events_cleared)
1575 return 0;
1576 } else {
1577 if (ev1 < mddev->events)
1578 /* just a hot-add of a new device, leave raid_disk at -1 */
1579 return 0;
1581 if (mddev->level != LEVEL_MULTIPATH) {
1582 int role;
1583 if (rdev->desc_nr < 0 ||
1584 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1585 role = 0xffff;
1586 rdev->desc_nr = -1;
1587 } else
1588 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1589 switch(role) {
1590 case 0xffff: /* spare */
1591 break;
1592 case 0xfffe: /* faulty */
1593 set_bit(Faulty, &rdev->flags);
1594 break;
1595 default:
1596 if ((le32_to_cpu(sb->feature_map) &
1597 MD_FEATURE_RECOVERY_OFFSET))
1598 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1599 else
1600 set_bit(In_sync, &rdev->flags);
1601 rdev->raid_disk = role;
1602 break;
1604 if (sb->devflags & WriteMostly1)
1605 set_bit(WriteMostly, &rdev->flags);
1606 } else /* MULTIPATH are always insync */
1607 set_bit(In_sync, &rdev->flags);
1609 return 0;
1612 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1614 struct mdp_superblock_1 *sb;
1615 mdk_rdev_t *rdev2;
1616 int max_dev, i;
1617 /* make rdev->sb match mddev and rdev data. */
1619 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1621 sb->feature_map = 0;
1622 sb->pad0 = 0;
1623 sb->recovery_offset = cpu_to_le64(0);
1624 memset(sb->pad1, 0, sizeof(sb->pad1));
1625 memset(sb->pad2, 0, sizeof(sb->pad2));
1626 memset(sb->pad3, 0, sizeof(sb->pad3));
1628 sb->utime = cpu_to_le64((__u64)mddev->utime);
1629 sb->events = cpu_to_le64(mddev->events);
1630 if (mddev->in_sync)
1631 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1632 else
1633 sb->resync_offset = cpu_to_le64(0);
1635 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1637 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1638 sb->size = cpu_to_le64(mddev->dev_sectors);
1639 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1640 sb->level = cpu_to_le32(mddev->level);
1641 sb->layout = cpu_to_le32(mddev->layout);
1643 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1644 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1645 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1648 if (rdev->raid_disk >= 0 &&
1649 !test_bit(In_sync, &rdev->flags)) {
1650 sb->feature_map |=
1651 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1652 sb->recovery_offset =
1653 cpu_to_le64(rdev->recovery_offset);
1656 if (mddev->reshape_position != MaxSector) {
1657 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1658 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1659 sb->new_layout = cpu_to_le32(mddev->new_layout);
1660 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1661 sb->new_level = cpu_to_le32(mddev->new_level);
1662 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1665 max_dev = 0;
1666 list_for_each_entry(rdev2, &mddev->disks, same_set)
1667 if (rdev2->desc_nr+1 > max_dev)
1668 max_dev = rdev2->desc_nr+1;
1670 if (max_dev > le32_to_cpu(sb->max_dev)) {
1671 int bmask;
1672 sb->max_dev = cpu_to_le32(max_dev);
1673 rdev->sb_size = max_dev * 2 + 256;
1674 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1675 if (rdev->sb_size & bmask)
1676 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1677 } else
1678 max_dev = le32_to_cpu(sb->max_dev);
1680 for (i=0; i<max_dev;i++)
1681 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1683 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1684 i = rdev2->desc_nr;
1685 if (test_bit(Faulty, &rdev2->flags))
1686 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1687 else if (test_bit(In_sync, &rdev2->flags))
1688 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1689 else if (rdev2->raid_disk >= 0)
1690 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1691 else
1692 sb->dev_roles[i] = cpu_to_le16(0xffff);
1695 sb->sb_csum = calc_sb_1_csum(sb);
1698 static unsigned long long
1699 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1701 struct mdp_superblock_1 *sb;
1702 sector_t max_sectors;
1703 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1704 return 0; /* component must fit device */
1705 if (rdev->sb_start < rdev->data_offset) {
1706 /* minor versions 1 and 2; superblock before data */
1707 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1708 max_sectors -= rdev->data_offset;
1709 if (!num_sectors || num_sectors > max_sectors)
1710 num_sectors = max_sectors;
1711 } else if (rdev->mddev->bitmap_info.offset) {
1712 /* minor version 0 with bitmap we can't move */
1713 return 0;
1714 } else {
1715 /* minor version 0; superblock after data */
1716 sector_t sb_start;
1717 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1718 sb_start &= ~(sector_t)(4*2 - 1);
1719 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1720 if (!num_sectors || num_sectors > max_sectors)
1721 num_sectors = max_sectors;
1722 rdev->sb_start = sb_start;
1724 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1725 sb->data_size = cpu_to_le64(num_sectors);
1726 sb->super_offset = rdev->sb_start;
1727 sb->sb_csum = calc_sb_1_csum(sb);
1728 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1729 rdev->sb_page);
1730 md_super_wait(rdev->mddev);
1731 return num_sectors;
1734 static struct super_type super_types[] = {
1735 [0] = {
1736 .name = "0.90.0",
1737 .owner = THIS_MODULE,
1738 .load_super = super_90_load,
1739 .validate_super = super_90_validate,
1740 .sync_super = super_90_sync,
1741 .rdev_size_change = super_90_rdev_size_change,
1743 [1] = {
1744 .name = "md-1",
1745 .owner = THIS_MODULE,
1746 .load_super = super_1_load,
1747 .validate_super = super_1_validate,
1748 .sync_super = super_1_sync,
1749 .rdev_size_change = super_1_rdev_size_change,
1753 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1755 mdk_rdev_t *rdev, *rdev2;
1757 rcu_read_lock();
1758 rdev_for_each_rcu(rdev, mddev1)
1759 rdev_for_each_rcu(rdev2, mddev2)
1760 if (rdev->bdev->bd_contains ==
1761 rdev2->bdev->bd_contains) {
1762 rcu_read_unlock();
1763 return 1;
1765 rcu_read_unlock();
1766 return 0;
1769 static LIST_HEAD(pending_raid_disks);
1772 * Try to register data integrity profile for an mddev
1774 * This is called when an array is started and after a disk has been kicked
1775 * from the array. It only succeeds if all working and active component devices
1776 * are integrity capable with matching profiles.
1778 int md_integrity_register(mddev_t *mddev)
1780 mdk_rdev_t *rdev, *reference = NULL;
1782 if (list_empty(&mddev->disks))
1783 return 0; /* nothing to do */
1784 if (blk_get_integrity(mddev->gendisk))
1785 return 0; /* already registered */
1786 list_for_each_entry(rdev, &mddev->disks, same_set) {
1787 /* skip spares and non-functional disks */
1788 if (test_bit(Faulty, &rdev->flags))
1789 continue;
1790 if (rdev->raid_disk < 0)
1791 continue;
1792 if (!reference) {
1793 /* Use the first rdev as the reference */
1794 reference = rdev;
1795 continue;
1797 /* does this rdev's profile match the reference profile? */
1798 if (blk_integrity_compare(reference->bdev->bd_disk,
1799 rdev->bdev->bd_disk) < 0)
1800 return -EINVAL;
1802 if (!reference || !bdev_get_integrity(reference->bdev))
1803 return 0;
1805 * All component devices are integrity capable and have matching
1806 * profiles, register the common profile for the md device.
1808 if (blk_integrity_register(mddev->gendisk,
1809 bdev_get_integrity(reference->bdev)) != 0) {
1810 printk(KERN_ERR "md: failed to register integrity for %s\n",
1811 mdname(mddev));
1812 return -EINVAL;
1814 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1815 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1816 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1817 mdname(mddev));
1818 return -EINVAL;
1820 return 0;
1822 EXPORT_SYMBOL(md_integrity_register);
1824 /* Disable data integrity if non-capable/non-matching disk is being added */
1825 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1827 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1828 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1830 if (!bi_mddev) /* nothing to do */
1831 return;
1832 if (rdev->raid_disk < 0) /* skip spares */
1833 return;
1834 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1835 rdev->bdev->bd_disk) >= 0)
1836 return;
1837 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1838 blk_integrity_unregister(mddev->gendisk);
1840 EXPORT_SYMBOL(md_integrity_add_rdev);
1842 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1844 char b[BDEVNAME_SIZE];
1845 struct kobject *ko;
1846 char *s;
1847 int err;
1849 if (rdev->mddev) {
1850 MD_BUG();
1851 return -EINVAL;
1854 /* prevent duplicates */
1855 if (find_rdev(mddev, rdev->bdev->bd_dev))
1856 return -EEXIST;
1858 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1859 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1860 rdev->sectors < mddev->dev_sectors)) {
1861 if (mddev->pers) {
1862 /* Cannot change size, so fail
1863 * If mddev->level <= 0, then we don't care
1864 * about aligning sizes (e.g. linear)
1866 if (mddev->level > 0)
1867 return -ENOSPC;
1868 } else
1869 mddev->dev_sectors = rdev->sectors;
1872 /* Verify rdev->desc_nr is unique.
1873 * If it is -1, assign a free number, else
1874 * check number is not in use
1876 if (rdev->desc_nr < 0) {
1877 int choice = 0;
1878 if (mddev->pers) choice = mddev->raid_disks;
1879 while (find_rdev_nr(mddev, choice))
1880 choice++;
1881 rdev->desc_nr = choice;
1882 } else {
1883 if (find_rdev_nr(mddev, rdev->desc_nr))
1884 return -EBUSY;
1886 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1887 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1888 mdname(mddev), mddev->max_disks);
1889 return -EBUSY;
1891 bdevname(rdev->bdev,b);
1892 while ( (s=strchr(b, '/')) != NULL)
1893 *s = '!';
1895 rdev->mddev = mddev;
1896 printk(KERN_INFO "md: bind<%s>\n", b);
1898 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1899 goto fail;
1901 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1902 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1903 /* failure here is OK */;
1904 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1906 list_add_rcu(&rdev->same_set, &mddev->disks);
1907 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
1909 /* May as well allow recovery to be retried once */
1910 mddev->recovery_disabled = 0;
1912 return 0;
1914 fail:
1915 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1916 b, mdname(mddev));
1917 return err;
1920 static void md_delayed_delete(struct work_struct *ws)
1922 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1923 kobject_del(&rdev->kobj);
1924 kobject_put(&rdev->kobj);
1927 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1929 char b[BDEVNAME_SIZE];
1930 if (!rdev->mddev) {
1931 MD_BUG();
1932 return;
1934 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
1935 list_del_rcu(&rdev->same_set);
1936 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1937 rdev->mddev = NULL;
1938 sysfs_remove_link(&rdev->kobj, "block");
1939 sysfs_put(rdev->sysfs_state);
1940 rdev->sysfs_state = NULL;
1941 /* We need to delay this, otherwise we can deadlock when
1942 * writing to 'remove' to "dev/state". We also need
1943 * to delay it due to rcu usage.
1945 synchronize_rcu();
1946 INIT_WORK(&rdev->del_work, md_delayed_delete);
1947 kobject_get(&rdev->kobj);
1948 queue_work(md_misc_wq, &rdev->del_work);
1952 * prevent the device from being mounted, repartitioned or
1953 * otherwise reused by a RAID array (or any other kernel
1954 * subsystem), by bd_claiming the device.
1956 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1958 int err = 0;
1959 struct block_device *bdev;
1960 char b[BDEVNAME_SIZE];
1962 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1963 shared ? (mdk_rdev_t *)lock_rdev : rdev);
1964 if (IS_ERR(bdev)) {
1965 printk(KERN_ERR "md: could not open %s.\n",
1966 __bdevname(dev, b));
1967 return PTR_ERR(bdev);
1969 rdev->bdev = bdev;
1970 return err;
1973 static void unlock_rdev(mdk_rdev_t *rdev)
1975 struct block_device *bdev = rdev->bdev;
1976 rdev->bdev = NULL;
1977 if (!bdev)
1978 MD_BUG();
1979 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1982 void md_autodetect_dev(dev_t dev);
1984 static void export_rdev(mdk_rdev_t * rdev)
1986 char b[BDEVNAME_SIZE];
1987 printk(KERN_INFO "md: export_rdev(%s)\n",
1988 bdevname(rdev->bdev,b));
1989 if (rdev->mddev)
1990 MD_BUG();
1991 free_disk_sb(rdev);
1992 #ifndef MODULE
1993 if (test_bit(AutoDetected, &rdev->flags))
1994 md_autodetect_dev(rdev->bdev->bd_dev);
1995 #endif
1996 unlock_rdev(rdev);
1997 kobject_put(&rdev->kobj);
2000 static void kick_rdev_from_array(mdk_rdev_t * rdev)
2002 unbind_rdev_from_array(rdev);
2003 export_rdev(rdev);
2006 static void export_array(mddev_t *mddev)
2008 mdk_rdev_t *rdev, *tmp;
2010 rdev_for_each(rdev, tmp, mddev) {
2011 if (!rdev->mddev) {
2012 MD_BUG();
2013 continue;
2015 kick_rdev_from_array(rdev);
2017 if (!list_empty(&mddev->disks))
2018 MD_BUG();
2019 mddev->raid_disks = 0;
2020 mddev->major_version = 0;
2023 static void print_desc(mdp_disk_t *desc)
2025 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2026 desc->major,desc->minor,desc->raid_disk,desc->state);
2029 static void print_sb_90(mdp_super_t *sb)
2031 int i;
2033 printk(KERN_INFO
2034 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2035 sb->major_version, sb->minor_version, sb->patch_version,
2036 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2037 sb->ctime);
2038 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2039 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2040 sb->md_minor, sb->layout, sb->chunk_size);
2041 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2042 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2043 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2044 sb->failed_disks, sb->spare_disks,
2045 sb->sb_csum, (unsigned long)sb->events_lo);
2047 printk(KERN_INFO);
2048 for (i = 0; i < MD_SB_DISKS; i++) {
2049 mdp_disk_t *desc;
2051 desc = sb->disks + i;
2052 if (desc->number || desc->major || desc->minor ||
2053 desc->raid_disk || (desc->state && (desc->state != 4))) {
2054 printk(" D %2d: ", i);
2055 print_desc(desc);
2058 printk(KERN_INFO "md: THIS: ");
2059 print_desc(&sb->this_disk);
2062 static void print_sb_1(struct mdp_superblock_1 *sb)
2064 __u8 *uuid;
2066 uuid = sb->set_uuid;
2067 printk(KERN_INFO
2068 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2069 "md: Name: \"%s\" CT:%llu\n",
2070 le32_to_cpu(sb->major_version),
2071 le32_to_cpu(sb->feature_map),
2072 uuid,
2073 sb->set_name,
2074 (unsigned long long)le64_to_cpu(sb->ctime)
2075 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2077 uuid = sb->device_uuid;
2078 printk(KERN_INFO
2079 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2080 " RO:%llu\n"
2081 "md: Dev:%08x UUID: %pU\n"
2082 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2083 "md: (MaxDev:%u) \n",
2084 le32_to_cpu(sb->level),
2085 (unsigned long long)le64_to_cpu(sb->size),
2086 le32_to_cpu(sb->raid_disks),
2087 le32_to_cpu(sb->layout),
2088 le32_to_cpu(sb->chunksize),
2089 (unsigned long long)le64_to_cpu(sb->data_offset),
2090 (unsigned long long)le64_to_cpu(sb->data_size),
2091 (unsigned long long)le64_to_cpu(sb->super_offset),
2092 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2093 le32_to_cpu(sb->dev_number),
2094 uuid,
2095 sb->devflags,
2096 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2097 (unsigned long long)le64_to_cpu(sb->events),
2098 (unsigned long long)le64_to_cpu(sb->resync_offset),
2099 le32_to_cpu(sb->sb_csum),
2100 le32_to_cpu(sb->max_dev)
2104 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2106 char b[BDEVNAME_SIZE];
2107 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2108 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2109 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2110 rdev->desc_nr);
2111 if (rdev->sb_loaded) {
2112 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2113 switch (major_version) {
2114 case 0:
2115 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2116 break;
2117 case 1:
2118 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2119 break;
2121 } else
2122 printk(KERN_INFO "md: no rdev superblock!\n");
2125 static void md_print_devices(void)
2127 struct list_head *tmp;
2128 mdk_rdev_t *rdev;
2129 mddev_t *mddev;
2130 char b[BDEVNAME_SIZE];
2132 printk("\n");
2133 printk("md: **********************************\n");
2134 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2135 printk("md: **********************************\n");
2136 for_each_mddev(mddev, tmp) {
2138 if (mddev->bitmap)
2139 bitmap_print_sb(mddev->bitmap);
2140 else
2141 printk("%s: ", mdname(mddev));
2142 list_for_each_entry(rdev, &mddev->disks, same_set)
2143 printk("<%s>", bdevname(rdev->bdev,b));
2144 printk("\n");
2146 list_for_each_entry(rdev, &mddev->disks, same_set)
2147 print_rdev(rdev, mddev->major_version);
2149 printk("md: **********************************\n");
2150 printk("\n");
2154 static void sync_sbs(mddev_t * mddev, int nospares)
2156 /* Update each superblock (in-memory image), but
2157 * if we are allowed to, skip spares which already
2158 * have the right event counter, or have one earlier
2159 * (which would mean they aren't being marked as dirty
2160 * with the rest of the array)
2162 mdk_rdev_t *rdev;
2163 list_for_each_entry(rdev, &mddev->disks, same_set) {
2164 if (rdev->sb_events == mddev->events ||
2165 (nospares &&
2166 rdev->raid_disk < 0 &&
2167 rdev->sb_events+1 == mddev->events)) {
2168 /* Don't update this superblock */
2169 rdev->sb_loaded = 2;
2170 } else {
2171 super_types[mddev->major_version].
2172 sync_super(mddev, rdev);
2173 rdev->sb_loaded = 1;
2178 static void md_update_sb(mddev_t * mddev, int force_change)
2180 mdk_rdev_t *rdev;
2181 int sync_req;
2182 int nospares = 0;
2184 repeat:
2185 /* First make sure individual recovery_offsets are correct */
2186 list_for_each_entry(rdev, &mddev->disks, same_set) {
2187 if (rdev->raid_disk >= 0 &&
2188 mddev->delta_disks >= 0 &&
2189 !test_bit(In_sync, &rdev->flags) &&
2190 mddev->curr_resync_completed > rdev->recovery_offset)
2191 rdev->recovery_offset = mddev->curr_resync_completed;
2194 if (!mddev->persistent) {
2195 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2196 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2197 if (!mddev->external)
2198 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2199 wake_up(&mddev->sb_wait);
2200 return;
2203 spin_lock_irq(&mddev->write_lock);
2205 mddev->utime = get_seconds();
2207 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2208 force_change = 1;
2209 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2210 /* just a clean<-> dirty transition, possibly leave spares alone,
2211 * though if events isn't the right even/odd, we will have to do
2212 * spares after all
2214 nospares = 1;
2215 if (force_change)
2216 nospares = 0;
2217 if (mddev->degraded)
2218 /* If the array is degraded, then skipping spares is both
2219 * dangerous and fairly pointless.
2220 * Dangerous because a device that was removed from the array
2221 * might have a event_count that still looks up-to-date,
2222 * so it can be re-added without a resync.
2223 * Pointless because if there are any spares to skip,
2224 * then a recovery will happen and soon that array won't
2225 * be degraded any more and the spare can go back to sleep then.
2227 nospares = 0;
2229 sync_req = mddev->in_sync;
2231 /* If this is just a dirty<->clean transition, and the array is clean
2232 * and 'events' is odd, we can roll back to the previous clean state */
2233 if (nospares
2234 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2235 && mddev->can_decrease_events
2236 && mddev->events != 1) {
2237 mddev->events--;
2238 mddev->can_decrease_events = 0;
2239 } else {
2240 /* otherwise we have to go forward and ... */
2241 mddev->events ++;
2242 mddev->can_decrease_events = nospares;
2245 if (!mddev->events) {
2247 * oops, this 64-bit counter should never wrap.
2248 * Either we are in around ~1 trillion A.C., assuming
2249 * 1 reboot per second, or we have a bug:
2251 MD_BUG();
2252 mddev->events --;
2254 sync_sbs(mddev, nospares);
2255 spin_unlock_irq(&mddev->write_lock);
2257 dprintk(KERN_INFO
2258 "md: updating %s RAID superblock on device (in sync %d)\n",
2259 mdname(mddev),mddev->in_sync);
2261 bitmap_update_sb(mddev->bitmap);
2262 list_for_each_entry(rdev, &mddev->disks, same_set) {
2263 char b[BDEVNAME_SIZE];
2264 dprintk(KERN_INFO "md: ");
2265 if (rdev->sb_loaded != 1)
2266 continue; /* no noise on spare devices */
2267 if (test_bit(Faulty, &rdev->flags))
2268 dprintk("(skipping faulty ");
2270 dprintk("%s ", bdevname(rdev->bdev,b));
2271 if (!test_bit(Faulty, &rdev->flags)) {
2272 md_super_write(mddev,rdev,
2273 rdev->sb_start, rdev->sb_size,
2274 rdev->sb_page);
2275 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2276 bdevname(rdev->bdev,b),
2277 (unsigned long long)rdev->sb_start);
2278 rdev->sb_events = mddev->events;
2280 } else
2281 dprintk(")\n");
2282 if (mddev->level == LEVEL_MULTIPATH)
2283 /* only need to write one superblock... */
2284 break;
2286 md_super_wait(mddev);
2287 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2289 spin_lock_irq(&mddev->write_lock);
2290 if (mddev->in_sync != sync_req ||
2291 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2292 /* have to write it out again */
2293 spin_unlock_irq(&mddev->write_lock);
2294 goto repeat;
2296 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2297 spin_unlock_irq(&mddev->write_lock);
2298 wake_up(&mddev->sb_wait);
2299 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2300 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2304 /* words written to sysfs files may, or may not, be \n terminated.
2305 * We want to accept with case. For this we use cmd_match.
2307 static int cmd_match(const char *cmd, const char *str)
2309 /* See if cmd, written into a sysfs file, matches
2310 * str. They must either be the same, or cmd can
2311 * have a trailing newline
2313 while (*cmd && *str && *cmd == *str) {
2314 cmd++;
2315 str++;
2317 if (*cmd == '\n')
2318 cmd++;
2319 if (*str || *cmd)
2320 return 0;
2321 return 1;
2324 struct rdev_sysfs_entry {
2325 struct attribute attr;
2326 ssize_t (*show)(mdk_rdev_t *, char *);
2327 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2330 static ssize_t
2331 state_show(mdk_rdev_t *rdev, char *page)
2333 char *sep = "";
2334 size_t len = 0;
2336 if (test_bit(Faulty, &rdev->flags)) {
2337 len+= sprintf(page+len, "%sfaulty",sep);
2338 sep = ",";
2340 if (test_bit(In_sync, &rdev->flags)) {
2341 len += sprintf(page+len, "%sin_sync",sep);
2342 sep = ",";
2344 if (test_bit(WriteMostly, &rdev->flags)) {
2345 len += sprintf(page+len, "%swrite_mostly",sep);
2346 sep = ",";
2348 if (test_bit(Blocked, &rdev->flags)) {
2349 len += sprintf(page+len, "%sblocked", sep);
2350 sep = ",";
2352 if (!test_bit(Faulty, &rdev->flags) &&
2353 !test_bit(In_sync, &rdev->flags)) {
2354 len += sprintf(page+len, "%sspare", sep);
2355 sep = ",";
2357 return len+sprintf(page+len, "\n");
2360 static ssize_t
2361 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2363 /* can write
2364 * faulty - simulates and error
2365 * remove - disconnects the device
2366 * writemostly - sets write_mostly
2367 * -writemostly - clears write_mostly
2368 * blocked - sets the Blocked flag
2369 * -blocked - clears the Blocked flag
2370 * insync - sets Insync providing device isn't active
2372 int err = -EINVAL;
2373 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2374 md_error(rdev->mddev, rdev);
2375 err = 0;
2376 } else if (cmd_match(buf, "remove")) {
2377 if (rdev->raid_disk >= 0)
2378 err = -EBUSY;
2379 else {
2380 mddev_t *mddev = rdev->mddev;
2381 kick_rdev_from_array(rdev);
2382 if (mddev->pers)
2383 md_update_sb(mddev, 1);
2384 md_new_event(mddev);
2385 err = 0;
2387 } else if (cmd_match(buf, "writemostly")) {
2388 set_bit(WriteMostly, &rdev->flags);
2389 err = 0;
2390 } else if (cmd_match(buf, "-writemostly")) {
2391 clear_bit(WriteMostly, &rdev->flags);
2392 err = 0;
2393 } else if (cmd_match(buf, "blocked")) {
2394 set_bit(Blocked, &rdev->flags);
2395 err = 0;
2396 } else if (cmd_match(buf, "-blocked")) {
2397 clear_bit(Blocked, &rdev->flags);
2398 wake_up(&rdev->blocked_wait);
2399 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2400 md_wakeup_thread(rdev->mddev->thread);
2402 err = 0;
2403 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2404 set_bit(In_sync, &rdev->flags);
2405 err = 0;
2407 if (!err)
2408 sysfs_notify_dirent_safe(rdev->sysfs_state);
2409 return err ? err : len;
2411 static struct rdev_sysfs_entry rdev_state =
2412 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2414 static ssize_t
2415 errors_show(mdk_rdev_t *rdev, char *page)
2417 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2420 static ssize_t
2421 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2423 char *e;
2424 unsigned long n = simple_strtoul(buf, &e, 10);
2425 if (*buf && (*e == 0 || *e == '\n')) {
2426 atomic_set(&rdev->corrected_errors, n);
2427 return len;
2429 return -EINVAL;
2431 static struct rdev_sysfs_entry rdev_errors =
2432 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2434 static ssize_t
2435 slot_show(mdk_rdev_t *rdev, char *page)
2437 if (rdev->raid_disk < 0)
2438 return sprintf(page, "none\n");
2439 else
2440 return sprintf(page, "%d\n", rdev->raid_disk);
2443 static ssize_t
2444 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2446 char *e;
2447 int err;
2448 char nm[20];
2449 int slot = simple_strtoul(buf, &e, 10);
2450 if (strncmp(buf, "none", 4)==0)
2451 slot = -1;
2452 else if (e==buf || (*e && *e!= '\n'))
2453 return -EINVAL;
2454 if (rdev->mddev->pers && slot == -1) {
2455 /* Setting 'slot' on an active array requires also
2456 * updating the 'rd%d' link, and communicating
2457 * with the personality with ->hot_*_disk.
2458 * For now we only support removing
2459 * failed/spare devices. This normally happens automatically,
2460 * but not when the metadata is externally managed.
2462 if (rdev->raid_disk == -1)
2463 return -EEXIST;
2464 /* personality does all needed checks */
2465 if (rdev->mddev->pers->hot_add_disk == NULL)
2466 return -EINVAL;
2467 err = rdev->mddev->pers->
2468 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2469 if (err)
2470 return err;
2471 sprintf(nm, "rd%d", rdev->raid_disk);
2472 sysfs_remove_link(&rdev->mddev->kobj, nm);
2473 rdev->raid_disk = -1;
2474 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2475 md_wakeup_thread(rdev->mddev->thread);
2476 } else if (rdev->mddev->pers) {
2477 mdk_rdev_t *rdev2;
2478 /* Activating a spare .. or possibly reactivating
2479 * if we ever get bitmaps working here.
2482 if (rdev->raid_disk != -1)
2483 return -EBUSY;
2485 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2486 return -EBUSY;
2488 if (rdev->mddev->pers->hot_add_disk == NULL)
2489 return -EINVAL;
2491 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2492 if (rdev2->raid_disk == slot)
2493 return -EEXIST;
2495 if (slot >= rdev->mddev->raid_disks &&
2496 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2497 return -ENOSPC;
2499 rdev->raid_disk = slot;
2500 if (test_bit(In_sync, &rdev->flags))
2501 rdev->saved_raid_disk = slot;
2502 else
2503 rdev->saved_raid_disk = -1;
2504 err = rdev->mddev->pers->
2505 hot_add_disk(rdev->mddev, rdev);
2506 if (err) {
2507 rdev->raid_disk = -1;
2508 return err;
2509 } else
2510 sysfs_notify_dirent_safe(rdev->sysfs_state);
2511 sprintf(nm, "rd%d", rdev->raid_disk);
2512 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2513 /* failure here is OK */;
2514 /* don't wakeup anyone, leave that to userspace. */
2515 } else {
2516 if (slot >= rdev->mddev->raid_disks &&
2517 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2518 return -ENOSPC;
2519 rdev->raid_disk = slot;
2520 /* assume it is working */
2521 clear_bit(Faulty, &rdev->flags);
2522 clear_bit(WriteMostly, &rdev->flags);
2523 set_bit(In_sync, &rdev->flags);
2524 sysfs_notify_dirent_safe(rdev->sysfs_state);
2526 return len;
2530 static struct rdev_sysfs_entry rdev_slot =
2531 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2533 static ssize_t
2534 offset_show(mdk_rdev_t *rdev, char *page)
2536 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2539 static ssize_t
2540 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2542 char *e;
2543 unsigned long long offset = simple_strtoull(buf, &e, 10);
2544 if (e==buf || (*e && *e != '\n'))
2545 return -EINVAL;
2546 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2547 return -EBUSY;
2548 if (rdev->sectors && rdev->mddev->external)
2549 /* Must set offset before size, so overlap checks
2550 * can be sane */
2551 return -EBUSY;
2552 rdev->data_offset = offset;
2553 return len;
2556 static struct rdev_sysfs_entry rdev_offset =
2557 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2559 static ssize_t
2560 rdev_size_show(mdk_rdev_t *rdev, char *page)
2562 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2565 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2567 /* check if two start/length pairs overlap */
2568 if (s1+l1 <= s2)
2569 return 0;
2570 if (s2+l2 <= s1)
2571 return 0;
2572 return 1;
2575 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2577 unsigned long long blocks;
2578 sector_t new;
2580 if (strict_strtoull(buf, 10, &blocks) < 0)
2581 return -EINVAL;
2583 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2584 return -EINVAL; /* sector conversion overflow */
2586 new = blocks * 2;
2587 if (new != blocks * 2)
2588 return -EINVAL; /* unsigned long long to sector_t overflow */
2590 *sectors = new;
2591 return 0;
2594 static ssize_t
2595 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2597 mddev_t *my_mddev = rdev->mddev;
2598 sector_t oldsectors = rdev->sectors;
2599 sector_t sectors;
2601 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2602 return -EINVAL;
2603 if (my_mddev->pers && rdev->raid_disk >= 0) {
2604 if (my_mddev->persistent) {
2605 sectors = super_types[my_mddev->major_version].
2606 rdev_size_change(rdev, sectors);
2607 if (!sectors)
2608 return -EBUSY;
2609 } else if (!sectors)
2610 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2611 rdev->data_offset;
2613 if (sectors < my_mddev->dev_sectors)
2614 return -EINVAL; /* component must fit device */
2616 rdev->sectors = sectors;
2617 if (sectors > oldsectors && my_mddev->external) {
2618 /* need to check that all other rdevs with the same ->bdev
2619 * do not overlap. We need to unlock the mddev to avoid
2620 * a deadlock. We have already changed rdev->sectors, and if
2621 * we have to change it back, we will have the lock again.
2623 mddev_t *mddev;
2624 int overlap = 0;
2625 struct list_head *tmp;
2627 mddev_unlock(my_mddev);
2628 for_each_mddev(mddev, tmp) {
2629 mdk_rdev_t *rdev2;
2631 mddev_lock(mddev);
2632 list_for_each_entry(rdev2, &mddev->disks, same_set)
2633 if (rdev->bdev == rdev2->bdev &&
2634 rdev != rdev2 &&
2635 overlaps(rdev->data_offset, rdev->sectors,
2636 rdev2->data_offset,
2637 rdev2->sectors)) {
2638 overlap = 1;
2639 break;
2641 mddev_unlock(mddev);
2642 if (overlap) {
2643 mddev_put(mddev);
2644 break;
2647 mddev_lock(my_mddev);
2648 if (overlap) {
2649 /* Someone else could have slipped in a size
2650 * change here, but doing so is just silly.
2651 * We put oldsectors back because we *know* it is
2652 * safe, and trust userspace not to race with
2653 * itself
2655 rdev->sectors = oldsectors;
2656 return -EBUSY;
2659 return len;
2662 static struct rdev_sysfs_entry rdev_size =
2663 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2666 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2668 unsigned long long recovery_start = rdev->recovery_offset;
2670 if (test_bit(In_sync, &rdev->flags) ||
2671 recovery_start == MaxSector)
2672 return sprintf(page, "none\n");
2674 return sprintf(page, "%llu\n", recovery_start);
2677 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2679 unsigned long long recovery_start;
2681 if (cmd_match(buf, "none"))
2682 recovery_start = MaxSector;
2683 else if (strict_strtoull(buf, 10, &recovery_start))
2684 return -EINVAL;
2686 if (rdev->mddev->pers &&
2687 rdev->raid_disk >= 0)
2688 return -EBUSY;
2690 rdev->recovery_offset = recovery_start;
2691 if (recovery_start == MaxSector)
2692 set_bit(In_sync, &rdev->flags);
2693 else
2694 clear_bit(In_sync, &rdev->flags);
2695 return len;
2698 static struct rdev_sysfs_entry rdev_recovery_start =
2699 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2701 static struct attribute *rdev_default_attrs[] = {
2702 &rdev_state.attr,
2703 &rdev_errors.attr,
2704 &rdev_slot.attr,
2705 &rdev_offset.attr,
2706 &rdev_size.attr,
2707 &rdev_recovery_start.attr,
2708 NULL,
2710 static ssize_t
2711 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2713 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2714 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2715 mddev_t *mddev = rdev->mddev;
2716 ssize_t rv;
2718 if (!entry->show)
2719 return -EIO;
2721 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2722 if (!rv) {
2723 if (rdev->mddev == NULL)
2724 rv = -EBUSY;
2725 else
2726 rv = entry->show(rdev, page);
2727 mddev_unlock(mddev);
2729 return rv;
2732 static ssize_t
2733 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2734 const char *page, size_t length)
2736 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2737 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2738 ssize_t rv;
2739 mddev_t *mddev = rdev->mddev;
2741 if (!entry->store)
2742 return -EIO;
2743 if (!capable(CAP_SYS_ADMIN))
2744 return -EACCES;
2745 rv = mddev ? mddev_lock(mddev): -EBUSY;
2746 if (!rv) {
2747 if (rdev->mddev == NULL)
2748 rv = -EBUSY;
2749 else
2750 rv = entry->store(rdev, page, length);
2751 mddev_unlock(mddev);
2753 return rv;
2756 static void rdev_free(struct kobject *ko)
2758 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2759 kfree(rdev);
2761 static const struct sysfs_ops rdev_sysfs_ops = {
2762 .show = rdev_attr_show,
2763 .store = rdev_attr_store,
2765 static struct kobj_type rdev_ktype = {
2766 .release = rdev_free,
2767 .sysfs_ops = &rdev_sysfs_ops,
2768 .default_attrs = rdev_default_attrs,
2771 void md_rdev_init(mdk_rdev_t *rdev)
2773 rdev->desc_nr = -1;
2774 rdev->saved_raid_disk = -1;
2775 rdev->raid_disk = -1;
2776 rdev->flags = 0;
2777 rdev->data_offset = 0;
2778 rdev->sb_events = 0;
2779 rdev->last_read_error.tv_sec = 0;
2780 rdev->last_read_error.tv_nsec = 0;
2781 atomic_set(&rdev->nr_pending, 0);
2782 atomic_set(&rdev->read_errors, 0);
2783 atomic_set(&rdev->corrected_errors, 0);
2785 INIT_LIST_HEAD(&rdev->same_set);
2786 init_waitqueue_head(&rdev->blocked_wait);
2788 EXPORT_SYMBOL_GPL(md_rdev_init);
2790 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2792 * mark the device faulty if:
2794 * - the device is nonexistent (zero size)
2795 * - the device has no valid superblock
2797 * a faulty rdev _never_ has rdev->sb set.
2799 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2801 char b[BDEVNAME_SIZE];
2802 int err;
2803 mdk_rdev_t *rdev;
2804 sector_t size;
2806 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2807 if (!rdev) {
2808 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2809 return ERR_PTR(-ENOMEM);
2812 md_rdev_init(rdev);
2813 if ((err = alloc_disk_sb(rdev)))
2814 goto abort_free;
2816 err = lock_rdev(rdev, newdev, super_format == -2);
2817 if (err)
2818 goto abort_free;
2820 kobject_init(&rdev->kobj, &rdev_ktype);
2822 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
2823 if (!size) {
2824 printk(KERN_WARNING
2825 "md: %s has zero or unknown size, marking faulty!\n",
2826 bdevname(rdev->bdev,b));
2827 err = -EINVAL;
2828 goto abort_free;
2831 if (super_format >= 0) {
2832 err = super_types[super_format].
2833 load_super(rdev, NULL, super_minor);
2834 if (err == -EINVAL) {
2835 printk(KERN_WARNING
2836 "md: %s does not have a valid v%d.%d "
2837 "superblock, not importing!\n",
2838 bdevname(rdev->bdev,b),
2839 super_format, super_minor);
2840 goto abort_free;
2842 if (err < 0) {
2843 printk(KERN_WARNING
2844 "md: could not read %s's sb, not importing!\n",
2845 bdevname(rdev->bdev,b));
2846 goto abort_free;
2850 return rdev;
2852 abort_free:
2853 if (rdev->sb_page) {
2854 if (rdev->bdev)
2855 unlock_rdev(rdev);
2856 free_disk_sb(rdev);
2858 kfree(rdev);
2859 return ERR_PTR(err);
2863 * Check a full RAID array for plausibility
2867 static void analyze_sbs(mddev_t * mddev)
2869 int i;
2870 mdk_rdev_t *rdev, *freshest, *tmp;
2871 char b[BDEVNAME_SIZE];
2873 freshest = NULL;
2874 rdev_for_each(rdev, tmp, mddev)
2875 switch (super_types[mddev->major_version].
2876 load_super(rdev, freshest, mddev->minor_version)) {
2877 case 1:
2878 freshest = rdev;
2879 break;
2880 case 0:
2881 break;
2882 default:
2883 printk( KERN_ERR \
2884 "md: fatal superblock inconsistency in %s"
2885 " -- removing from array\n",
2886 bdevname(rdev->bdev,b));
2887 kick_rdev_from_array(rdev);
2891 super_types[mddev->major_version].
2892 validate_super(mddev, freshest);
2894 i = 0;
2895 rdev_for_each(rdev, tmp, mddev) {
2896 if (mddev->max_disks &&
2897 (rdev->desc_nr >= mddev->max_disks ||
2898 i > mddev->max_disks)) {
2899 printk(KERN_WARNING
2900 "md: %s: %s: only %d devices permitted\n",
2901 mdname(mddev), bdevname(rdev->bdev, b),
2902 mddev->max_disks);
2903 kick_rdev_from_array(rdev);
2904 continue;
2906 if (rdev != freshest)
2907 if (super_types[mddev->major_version].
2908 validate_super(mddev, rdev)) {
2909 printk(KERN_WARNING "md: kicking non-fresh %s"
2910 " from array!\n",
2911 bdevname(rdev->bdev,b));
2912 kick_rdev_from_array(rdev);
2913 continue;
2915 if (mddev->level == LEVEL_MULTIPATH) {
2916 rdev->desc_nr = i++;
2917 rdev->raid_disk = rdev->desc_nr;
2918 set_bit(In_sync, &rdev->flags);
2919 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2920 rdev->raid_disk = -1;
2921 clear_bit(In_sync, &rdev->flags);
2926 /* Read a fixed-point number.
2927 * Numbers in sysfs attributes should be in "standard" units where
2928 * possible, so time should be in seconds.
2929 * However we internally use a a much smaller unit such as
2930 * milliseconds or jiffies.
2931 * This function takes a decimal number with a possible fractional
2932 * component, and produces an integer which is the result of
2933 * multiplying that number by 10^'scale'.
2934 * all without any floating-point arithmetic.
2936 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2938 unsigned long result = 0;
2939 long decimals = -1;
2940 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2941 if (*cp == '.')
2942 decimals = 0;
2943 else if (decimals < scale) {
2944 unsigned int value;
2945 value = *cp - '0';
2946 result = result * 10 + value;
2947 if (decimals >= 0)
2948 decimals++;
2950 cp++;
2952 if (*cp == '\n')
2953 cp++;
2954 if (*cp)
2955 return -EINVAL;
2956 if (decimals < 0)
2957 decimals = 0;
2958 while (decimals < scale) {
2959 result *= 10;
2960 decimals ++;
2962 *res = result;
2963 return 0;
2967 static void md_safemode_timeout(unsigned long data);
2969 static ssize_t
2970 safe_delay_show(mddev_t *mddev, char *page)
2972 int msec = (mddev->safemode_delay*1000)/HZ;
2973 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2975 static ssize_t
2976 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2978 unsigned long msec;
2980 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2981 return -EINVAL;
2982 if (msec == 0)
2983 mddev->safemode_delay = 0;
2984 else {
2985 unsigned long old_delay = mddev->safemode_delay;
2986 mddev->safemode_delay = (msec*HZ)/1000;
2987 if (mddev->safemode_delay == 0)
2988 mddev->safemode_delay = 1;
2989 if (mddev->safemode_delay < old_delay)
2990 md_safemode_timeout((unsigned long)mddev);
2992 return len;
2994 static struct md_sysfs_entry md_safe_delay =
2995 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2997 static ssize_t
2998 level_show(mddev_t *mddev, char *page)
3000 struct mdk_personality *p = mddev->pers;
3001 if (p)
3002 return sprintf(page, "%s\n", p->name);
3003 else if (mddev->clevel[0])
3004 return sprintf(page, "%s\n", mddev->clevel);
3005 else if (mddev->level != LEVEL_NONE)
3006 return sprintf(page, "%d\n", mddev->level);
3007 else
3008 return 0;
3011 static ssize_t
3012 level_store(mddev_t *mddev, const char *buf, size_t len)
3014 char clevel[16];
3015 ssize_t rv = len;
3016 struct mdk_personality *pers;
3017 long level;
3018 void *priv;
3019 mdk_rdev_t *rdev;
3021 if (mddev->pers == NULL) {
3022 if (len == 0)
3023 return 0;
3024 if (len >= sizeof(mddev->clevel))
3025 return -ENOSPC;
3026 strncpy(mddev->clevel, buf, len);
3027 if (mddev->clevel[len-1] == '\n')
3028 len--;
3029 mddev->clevel[len] = 0;
3030 mddev->level = LEVEL_NONE;
3031 return rv;
3034 /* request to change the personality. Need to ensure:
3035 * - array is not engaged in resync/recovery/reshape
3036 * - old personality can be suspended
3037 * - new personality will access other array.
3040 if (mddev->sync_thread ||
3041 mddev->reshape_position != MaxSector ||
3042 mddev->sysfs_active)
3043 return -EBUSY;
3045 if (!mddev->pers->quiesce) {
3046 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3047 mdname(mddev), mddev->pers->name);
3048 return -EINVAL;
3051 /* Now find the new personality */
3052 if (len == 0 || len >= sizeof(clevel))
3053 return -EINVAL;
3054 strncpy(clevel, buf, len);
3055 if (clevel[len-1] == '\n')
3056 len--;
3057 clevel[len] = 0;
3058 if (strict_strtol(clevel, 10, &level))
3059 level = LEVEL_NONE;
3061 if (request_module("md-%s", clevel) != 0)
3062 request_module("md-level-%s", clevel);
3063 spin_lock(&pers_lock);
3064 pers = find_pers(level, clevel);
3065 if (!pers || !try_module_get(pers->owner)) {
3066 spin_unlock(&pers_lock);
3067 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3068 return -EINVAL;
3070 spin_unlock(&pers_lock);
3072 if (pers == mddev->pers) {
3073 /* Nothing to do! */
3074 module_put(pers->owner);
3075 return rv;
3077 if (!pers->takeover) {
3078 module_put(pers->owner);
3079 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3080 mdname(mddev), clevel);
3081 return -EINVAL;
3084 list_for_each_entry(rdev, &mddev->disks, same_set)
3085 rdev->new_raid_disk = rdev->raid_disk;
3087 /* ->takeover must set new_* and/or delta_disks
3088 * if it succeeds, and may set them when it fails.
3090 priv = pers->takeover(mddev);
3091 if (IS_ERR(priv)) {
3092 mddev->new_level = mddev->level;
3093 mddev->new_layout = mddev->layout;
3094 mddev->new_chunk_sectors = mddev->chunk_sectors;
3095 mddev->raid_disks -= mddev->delta_disks;
3096 mddev->delta_disks = 0;
3097 module_put(pers->owner);
3098 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3099 mdname(mddev), clevel);
3100 return PTR_ERR(priv);
3103 /* Looks like we have a winner */
3104 mddev_suspend(mddev);
3105 mddev->pers->stop(mddev);
3107 if (mddev->pers->sync_request == NULL &&
3108 pers->sync_request != NULL) {
3109 /* need to add the md_redundancy_group */
3110 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3111 printk(KERN_WARNING
3112 "md: cannot register extra attributes for %s\n",
3113 mdname(mddev));
3114 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3116 if (mddev->pers->sync_request != NULL &&
3117 pers->sync_request == NULL) {
3118 /* need to remove the md_redundancy_group */
3119 if (mddev->to_remove == NULL)
3120 mddev->to_remove = &md_redundancy_group;
3123 if (mddev->pers->sync_request == NULL &&
3124 mddev->external) {
3125 /* We are converting from a no-redundancy array
3126 * to a redundancy array and metadata is managed
3127 * externally so we need to be sure that writes
3128 * won't block due to a need to transition
3129 * clean->dirty
3130 * until external management is started.
3132 mddev->in_sync = 0;
3133 mddev->safemode_delay = 0;
3134 mddev->safemode = 0;
3137 list_for_each_entry(rdev, &mddev->disks, same_set) {
3138 char nm[20];
3139 if (rdev->raid_disk < 0)
3140 continue;
3141 if (rdev->new_raid_disk >= mddev->raid_disks)
3142 rdev->new_raid_disk = -1;
3143 if (rdev->new_raid_disk == rdev->raid_disk)
3144 continue;
3145 sprintf(nm, "rd%d", rdev->raid_disk);
3146 sysfs_remove_link(&mddev->kobj, nm);
3148 list_for_each_entry(rdev, &mddev->disks, same_set) {
3149 if (rdev->raid_disk < 0)
3150 continue;
3151 if (rdev->new_raid_disk == rdev->raid_disk)
3152 continue;
3153 rdev->raid_disk = rdev->new_raid_disk;
3154 if (rdev->raid_disk < 0)
3155 clear_bit(In_sync, &rdev->flags);
3156 else {
3157 char nm[20];
3158 sprintf(nm, "rd%d", rdev->raid_disk);
3159 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3160 printk("md: cannot register %s for %s after level change\n",
3161 nm, mdname(mddev));
3165 module_put(mddev->pers->owner);
3166 mddev->pers = pers;
3167 mddev->private = priv;
3168 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3169 mddev->level = mddev->new_level;
3170 mddev->layout = mddev->new_layout;
3171 mddev->chunk_sectors = mddev->new_chunk_sectors;
3172 mddev->delta_disks = 0;
3173 mddev->degraded = 0;
3174 if (mddev->pers->sync_request == NULL) {
3175 /* this is now an array without redundancy, so
3176 * it must always be in_sync
3178 mddev->in_sync = 1;
3179 del_timer_sync(&mddev->safemode_timer);
3181 pers->run(mddev);
3182 mddev_resume(mddev);
3183 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3184 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3185 md_wakeup_thread(mddev->thread);
3186 sysfs_notify(&mddev->kobj, NULL, "level");
3187 md_new_event(mddev);
3188 return rv;
3191 static struct md_sysfs_entry md_level =
3192 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3195 static ssize_t
3196 layout_show(mddev_t *mddev, char *page)
3198 /* just a number, not meaningful for all levels */
3199 if (mddev->reshape_position != MaxSector &&
3200 mddev->layout != mddev->new_layout)
3201 return sprintf(page, "%d (%d)\n",
3202 mddev->new_layout, mddev->layout);
3203 return sprintf(page, "%d\n", mddev->layout);
3206 static ssize_t
3207 layout_store(mddev_t *mddev, const char *buf, size_t len)
3209 char *e;
3210 unsigned long n = simple_strtoul(buf, &e, 10);
3212 if (!*buf || (*e && *e != '\n'))
3213 return -EINVAL;
3215 if (mddev->pers) {
3216 int err;
3217 if (mddev->pers->check_reshape == NULL)
3218 return -EBUSY;
3219 mddev->new_layout = n;
3220 err = mddev->pers->check_reshape(mddev);
3221 if (err) {
3222 mddev->new_layout = mddev->layout;
3223 return err;
3225 } else {
3226 mddev->new_layout = n;
3227 if (mddev->reshape_position == MaxSector)
3228 mddev->layout = n;
3230 return len;
3232 static struct md_sysfs_entry md_layout =
3233 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3236 static ssize_t
3237 raid_disks_show(mddev_t *mddev, char *page)
3239 if (mddev->raid_disks == 0)
3240 return 0;
3241 if (mddev->reshape_position != MaxSector &&
3242 mddev->delta_disks != 0)
3243 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3244 mddev->raid_disks - mddev->delta_disks);
3245 return sprintf(page, "%d\n", mddev->raid_disks);
3248 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3250 static ssize_t
3251 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3253 char *e;
3254 int rv = 0;
3255 unsigned long n = simple_strtoul(buf, &e, 10);
3257 if (!*buf || (*e && *e != '\n'))
3258 return -EINVAL;
3260 if (mddev->pers)
3261 rv = update_raid_disks(mddev, n);
3262 else if (mddev->reshape_position != MaxSector) {
3263 int olddisks = mddev->raid_disks - mddev->delta_disks;
3264 mddev->delta_disks = n - olddisks;
3265 mddev->raid_disks = n;
3266 } else
3267 mddev->raid_disks = n;
3268 return rv ? rv : len;
3270 static struct md_sysfs_entry md_raid_disks =
3271 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3273 static ssize_t
3274 chunk_size_show(mddev_t *mddev, char *page)
3276 if (mddev->reshape_position != MaxSector &&
3277 mddev->chunk_sectors != mddev->new_chunk_sectors)
3278 return sprintf(page, "%d (%d)\n",
3279 mddev->new_chunk_sectors << 9,
3280 mddev->chunk_sectors << 9);
3281 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3284 static ssize_t
3285 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3287 char *e;
3288 unsigned long n = simple_strtoul(buf, &e, 10);
3290 if (!*buf || (*e && *e != '\n'))
3291 return -EINVAL;
3293 if (mddev->pers) {
3294 int err;
3295 if (mddev->pers->check_reshape == NULL)
3296 return -EBUSY;
3297 mddev->new_chunk_sectors = n >> 9;
3298 err = mddev->pers->check_reshape(mddev);
3299 if (err) {
3300 mddev->new_chunk_sectors = mddev->chunk_sectors;
3301 return err;
3303 } else {
3304 mddev->new_chunk_sectors = n >> 9;
3305 if (mddev->reshape_position == MaxSector)
3306 mddev->chunk_sectors = n >> 9;
3308 return len;
3310 static struct md_sysfs_entry md_chunk_size =
3311 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3313 static ssize_t
3314 resync_start_show(mddev_t *mddev, char *page)
3316 if (mddev->recovery_cp == MaxSector)
3317 return sprintf(page, "none\n");
3318 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3321 static ssize_t
3322 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3324 char *e;
3325 unsigned long long n = simple_strtoull(buf, &e, 10);
3327 if (mddev->pers)
3328 return -EBUSY;
3329 if (cmd_match(buf, "none"))
3330 n = MaxSector;
3331 else if (!*buf || (*e && *e != '\n'))
3332 return -EINVAL;
3334 mddev->recovery_cp = n;
3335 return len;
3337 static struct md_sysfs_entry md_resync_start =
3338 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3341 * The array state can be:
3343 * clear
3344 * No devices, no size, no level
3345 * Equivalent to STOP_ARRAY ioctl
3346 * inactive
3347 * May have some settings, but array is not active
3348 * all IO results in error
3349 * When written, doesn't tear down array, but just stops it
3350 * suspended (not supported yet)
3351 * All IO requests will block. The array can be reconfigured.
3352 * Writing this, if accepted, will block until array is quiescent
3353 * readonly
3354 * no resync can happen. no superblocks get written.
3355 * write requests fail
3356 * read-auto
3357 * like readonly, but behaves like 'clean' on a write request.
3359 * clean - no pending writes, but otherwise active.
3360 * When written to inactive array, starts without resync
3361 * If a write request arrives then
3362 * if metadata is known, mark 'dirty' and switch to 'active'.
3363 * if not known, block and switch to write-pending
3364 * If written to an active array that has pending writes, then fails.
3365 * active
3366 * fully active: IO and resync can be happening.
3367 * When written to inactive array, starts with resync
3369 * write-pending
3370 * clean, but writes are blocked waiting for 'active' to be written.
3372 * active-idle
3373 * like active, but no writes have been seen for a while (100msec).
3376 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3377 write_pending, active_idle, bad_word};
3378 static char *array_states[] = {
3379 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3380 "write-pending", "active-idle", NULL };
3382 static int match_word(const char *word, char **list)
3384 int n;
3385 for (n=0; list[n]; n++)
3386 if (cmd_match(word, list[n]))
3387 break;
3388 return n;
3391 static ssize_t
3392 array_state_show(mddev_t *mddev, char *page)
3394 enum array_state st = inactive;
3396 if (mddev->pers)
3397 switch(mddev->ro) {
3398 case 1:
3399 st = readonly;
3400 break;
3401 case 2:
3402 st = read_auto;
3403 break;
3404 case 0:
3405 if (mddev->in_sync)
3406 st = clean;
3407 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3408 st = write_pending;
3409 else if (mddev->safemode)
3410 st = active_idle;
3411 else
3412 st = active;
3414 else {
3415 if (list_empty(&mddev->disks) &&
3416 mddev->raid_disks == 0 &&
3417 mddev->dev_sectors == 0)
3418 st = clear;
3419 else
3420 st = inactive;
3422 return sprintf(page, "%s\n", array_states[st]);
3425 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3426 static int md_set_readonly(mddev_t * mddev, int is_open);
3427 static int do_md_run(mddev_t * mddev);
3428 static int restart_array(mddev_t *mddev);
3430 static ssize_t
3431 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3433 int err = -EINVAL;
3434 enum array_state st = match_word(buf, array_states);
3435 switch(st) {
3436 case bad_word:
3437 break;
3438 case clear:
3439 /* stopping an active array */
3440 if (atomic_read(&mddev->openers) > 0)
3441 return -EBUSY;
3442 err = do_md_stop(mddev, 0, 0);
3443 break;
3444 case inactive:
3445 /* stopping an active array */
3446 if (mddev->pers) {
3447 if (atomic_read(&mddev->openers) > 0)
3448 return -EBUSY;
3449 err = do_md_stop(mddev, 2, 0);
3450 } else
3451 err = 0; /* already inactive */
3452 break;
3453 case suspended:
3454 break; /* not supported yet */
3455 case readonly:
3456 if (mddev->pers)
3457 err = md_set_readonly(mddev, 0);
3458 else {
3459 mddev->ro = 1;
3460 set_disk_ro(mddev->gendisk, 1);
3461 err = do_md_run(mddev);
3463 break;
3464 case read_auto:
3465 if (mddev->pers) {
3466 if (mddev->ro == 0)
3467 err = md_set_readonly(mddev, 0);
3468 else if (mddev->ro == 1)
3469 err = restart_array(mddev);
3470 if (err == 0) {
3471 mddev->ro = 2;
3472 set_disk_ro(mddev->gendisk, 0);
3474 } else {
3475 mddev->ro = 2;
3476 err = do_md_run(mddev);
3478 break;
3479 case clean:
3480 if (mddev->pers) {
3481 restart_array(mddev);
3482 spin_lock_irq(&mddev->write_lock);
3483 if (atomic_read(&mddev->writes_pending) == 0) {
3484 if (mddev->in_sync == 0) {
3485 mddev->in_sync = 1;
3486 if (mddev->safemode == 1)
3487 mddev->safemode = 0;
3488 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3490 err = 0;
3491 } else
3492 err = -EBUSY;
3493 spin_unlock_irq(&mddev->write_lock);
3494 } else
3495 err = -EINVAL;
3496 break;
3497 case active:
3498 if (mddev->pers) {
3499 restart_array(mddev);
3500 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3501 wake_up(&mddev->sb_wait);
3502 err = 0;
3503 } else {
3504 mddev->ro = 0;
3505 set_disk_ro(mddev->gendisk, 0);
3506 err = do_md_run(mddev);
3508 break;
3509 case write_pending:
3510 case active_idle:
3511 /* these cannot be set */
3512 break;
3514 if (err)
3515 return err;
3516 else {
3517 sysfs_notify_dirent_safe(mddev->sysfs_state);
3518 return len;
3521 static struct md_sysfs_entry md_array_state =
3522 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3524 static ssize_t
3525 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3526 return sprintf(page, "%d\n",
3527 atomic_read(&mddev->max_corr_read_errors));
3530 static ssize_t
3531 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3533 char *e;
3534 unsigned long n = simple_strtoul(buf, &e, 10);
3536 if (*buf && (*e == 0 || *e == '\n')) {
3537 atomic_set(&mddev->max_corr_read_errors, n);
3538 return len;
3540 return -EINVAL;
3543 static struct md_sysfs_entry max_corr_read_errors =
3544 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3545 max_corrected_read_errors_store);
3547 static ssize_t
3548 null_show(mddev_t *mddev, char *page)
3550 return -EINVAL;
3553 static ssize_t
3554 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3556 /* buf must be %d:%d\n? giving major and minor numbers */
3557 /* The new device is added to the array.
3558 * If the array has a persistent superblock, we read the
3559 * superblock to initialise info and check validity.
3560 * Otherwise, only checking done is that in bind_rdev_to_array,
3561 * which mainly checks size.
3563 char *e;
3564 int major = simple_strtoul(buf, &e, 10);
3565 int minor;
3566 dev_t dev;
3567 mdk_rdev_t *rdev;
3568 int err;
3570 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3571 return -EINVAL;
3572 minor = simple_strtoul(e+1, &e, 10);
3573 if (*e && *e != '\n')
3574 return -EINVAL;
3575 dev = MKDEV(major, minor);
3576 if (major != MAJOR(dev) ||
3577 minor != MINOR(dev))
3578 return -EOVERFLOW;
3581 if (mddev->persistent) {
3582 rdev = md_import_device(dev, mddev->major_version,
3583 mddev->minor_version);
3584 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3585 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3586 mdk_rdev_t, same_set);
3587 err = super_types[mddev->major_version]
3588 .load_super(rdev, rdev0, mddev->minor_version);
3589 if (err < 0)
3590 goto out;
3592 } else if (mddev->external)
3593 rdev = md_import_device(dev, -2, -1);
3594 else
3595 rdev = md_import_device(dev, -1, -1);
3597 if (IS_ERR(rdev))
3598 return PTR_ERR(rdev);
3599 err = bind_rdev_to_array(rdev, mddev);
3600 out:
3601 if (err)
3602 export_rdev(rdev);
3603 return err ? err : len;
3606 static struct md_sysfs_entry md_new_device =
3607 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3609 static ssize_t
3610 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3612 char *end;
3613 unsigned long chunk, end_chunk;
3615 if (!mddev->bitmap)
3616 goto out;
3617 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3618 while (*buf) {
3619 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3620 if (buf == end) break;
3621 if (*end == '-') { /* range */
3622 buf = end + 1;
3623 end_chunk = simple_strtoul(buf, &end, 0);
3624 if (buf == end) break;
3626 if (*end && !isspace(*end)) break;
3627 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3628 buf = skip_spaces(end);
3630 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3631 out:
3632 return len;
3635 static struct md_sysfs_entry md_bitmap =
3636 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3638 static ssize_t
3639 size_show(mddev_t *mddev, char *page)
3641 return sprintf(page, "%llu\n",
3642 (unsigned long long)mddev->dev_sectors / 2);
3645 static int update_size(mddev_t *mddev, sector_t num_sectors);
3647 static ssize_t
3648 size_store(mddev_t *mddev, const char *buf, size_t len)
3650 /* If array is inactive, we can reduce the component size, but
3651 * not increase it (except from 0).
3652 * If array is active, we can try an on-line resize
3654 sector_t sectors;
3655 int err = strict_blocks_to_sectors(buf, &sectors);
3657 if (err < 0)
3658 return err;
3659 if (mddev->pers) {
3660 err = update_size(mddev, sectors);
3661 md_update_sb(mddev, 1);
3662 } else {
3663 if (mddev->dev_sectors == 0 ||
3664 mddev->dev_sectors > sectors)
3665 mddev->dev_sectors = sectors;
3666 else
3667 err = -ENOSPC;
3669 return err ? err : len;
3672 static struct md_sysfs_entry md_size =
3673 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3676 /* Metdata version.
3677 * This is one of
3678 * 'none' for arrays with no metadata (good luck...)
3679 * 'external' for arrays with externally managed metadata,
3680 * or N.M for internally known formats
3682 static ssize_t
3683 metadata_show(mddev_t *mddev, char *page)
3685 if (mddev->persistent)
3686 return sprintf(page, "%d.%d\n",
3687 mddev->major_version, mddev->minor_version);
3688 else if (mddev->external)
3689 return sprintf(page, "external:%s\n", mddev->metadata_type);
3690 else
3691 return sprintf(page, "none\n");
3694 static ssize_t
3695 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3697 int major, minor;
3698 char *e;
3699 /* Changing the details of 'external' metadata is
3700 * always permitted. Otherwise there must be
3701 * no devices attached to the array.
3703 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3705 else if (!list_empty(&mddev->disks))
3706 return -EBUSY;
3708 if (cmd_match(buf, "none")) {
3709 mddev->persistent = 0;
3710 mddev->external = 0;
3711 mddev->major_version = 0;
3712 mddev->minor_version = 90;
3713 return len;
3715 if (strncmp(buf, "external:", 9) == 0) {
3716 size_t namelen = len-9;
3717 if (namelen >= sizeof(mddev->metadata_type))
3718 namelen = sizeof(mddev->metadata_type)-1;
3719 strncpy(mddev->metadata_type, buf+9, namelen);
3720 mddev->metadata_type[namelen] = 0;
3721 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3722 mddev->metadata_type[--namelen] = 0;
3723 mddev->persistent = 0;
3724 mddev->external = 1;
3725 mddev->major_version = 0;
3726 mddev->minor_version = 90;
3727 return len;
3729 major = simple_strtoul(buf, &e, 10);
3730 if (e==buf || *e != '.')
3731 return -EINVAL;
3732 buf = e+1;
3733 minor = simple_strtoul(buf, &e, 10);
3734 if (e==buf || (*e && *e != '\n') )
3735 return -EINVAL;
3736 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3737 return -ENOENT;
3738 mddev->major_version = major;
3739 mddev->minor_version = minor;
3740 mddev->persistent = 1;
3741 mddev->external = 0;
3742 return len;
3745 static struct md_sysfs_entry md_metadata =
3746 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3748 static ssize_t
3749 action_show(mddev_t *mddev, char *page)
3751 char *type = "idle";
3752 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3753 type = "frozen";
3754 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3755 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3756 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3757 type = "reshape";
3758 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3759 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3760 type = "resync";
3761 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3762 type = "check";
3763 else
3764 type = "repair";
3765 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3766 type = "recover";
3768 return sprintf(page, "%s\n", type);
3771 static void reap_sync_thread(mddev_t *mddev);
3773 static ssize_t
3774 action_store(mddev_t *mddev, const char *page, size_t len)
3776 if (!mddev->pers || !mddev->pers->sync_request)
3777 return -EINVAL;
3779 if (cmd_match(page, "frozen"))
3780 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3781 else
3782 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3784 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3785 if (mddev->sync_thread) {
3786 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3787 reap_sync_thread(mddev);
3789 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3790 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3791 return -EBUSY;
3792 else if (cmd_match(page, "resync"))
3793 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3794 else if (cmd_match(page, "recover")) {
3795 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3796 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3797 } else if (cmd_match(page, "reshape")) {
3798 int err;
3799 if (mddev->pers->start_reshape == NULL)
3800 return -EINVAL;
3801 err = mddev->pers->start_reshape(mddev);
3802 if (err)
3803 return err;
3804 sysfs_notify(&mddev->kobj, NULL, "degraded");
3805 } else {
3806 if (cmd_match(page, "check"))
3807 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3808 else if (!cmd_match(page, "repair"))
3809 return -EINVAL;
3810 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3811 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3813 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3814 md_wakeup_thread(mddev->thread);
3815 sysfs_notify_dirent_safe(mddev->sysfs_action);
3816 return len;
3819 static ssize_t
3820 mismatch_cnt_show(mddev_t *mddev, char *page)
3822 return sprintf(page, "%llu\n",
3823 (unsigned long long) mddev->resync_mismatches);
3826 static struct md_sysfs_entry md_scan_mode =
3827 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3830 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3832 static ssize_t
3833 sync_min_show(mddev_t *mddev, char *page)
3835 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3836 mddev->sync_speed_min ? "local": "system");
3839 static ssize_t
3840 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3842 int min;
3843 char *e;
3844 if (strncmp(buf, "system", 6)==0) {
3845 mddev->sync_speed_min = 0;
3846 return len;
3848 min = simple_strtoul(buf, &e, 10);
3849 if (buf == e || (*e && *e != '\n') || min <= 0)
3850 return -EINVAL;
3851 mddev->sync_speed_min = min;
3852 return len;
3855 static struct md_sysfs_entry md_sync_min =
3856 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3858 static ssize_t
3859 sync_max_show(mddev_t *mddev, char *page)
3861 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3862 mddev->sync_speed_max ? "local": "system");
3865 static ssize_t
3866 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3868 int max;
3869 char *e;
3870 if (strncmp(buf, "system", 6)==0) {
3871 mddev->sync_speed_max = 0;
3872 return len;
3874 max = simple_strtoul(buf, &e, 10);
3875 if (buf == e || (*e && *e != '\n') || max <= 0)
3876 return -EINVAL;
3877 mddev->sync_speed_max = max;
3878 return len;
3881 static struct md_sysfs_entry md_sync_max =
3882 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3884 static ssize_t
3885 degraded_show(mddev_t *mddev, char *page)
3887 return sprintf(page, "%d\n", mddev->degraded);
3889 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3891 static ssize_t
3892 sync_force_parallel_show(mddev_t *mddev, char *page)
3894 return sprintf(page, "%d\n", mddev->parallel_resync);
3897 static ssize_t
3898 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3900 long n;
3902 if (strict_strtol(buf, 10, &n))
3903 return -EINVAL;
3905 if (n != 0 && n != 1)
3906 return -EINVAL;
3908 mddev->parallel_resync = n;
3910 if (mddev->sync_thread)
3911 wake_up(&resync_wait);
3913 return len;
3916 /* force parallel resync, even with shared block devices */
3917 static struct md_sysfs_entry md_sync_force_parallel =
3918 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3919 sync_force_parallel_show, sync_force_parallel_store);
3921 static ssize_t
3922 sync_speed_show(mddev_t *mddev, char *page)
3924 unsigned long resync, dt, db;
3925 if (mddev->curr_resync == 0)
3926 return sprintf(page, "none\n");
3927 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3928 dt = (jiffies - mddev->resync_mark) / HZ;
3929 if (!dt) dt++;
3930 db = resync - mddev->resync_mark_cnt;
3931 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3934 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3936 static ssize_t
3937 sync_completed_show(mddev_t *mddev, char *page)
3939 unsigned long long max_sectors, resync;
3941 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3942 return sprintf(page, "none\n");
3944 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3945 max_sectors = mddev->resync_max_sectors;
3946 else
3947 max_sectors = mddev->dev_sectors;
3949 resync = mddev->curr_resync_completed;
3950 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
3953 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3955 static ssize_t
3956 min_sync_show(mddev_t *mddev, char *page)
3958 return sprintf(page, "%llu\n",
3959 (unsigned long long)mddev->resync_min);
3961 static ssize_t
3962 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3964 unsigned long long min;
3965 if (strict_strtoull(buf, 10, &min))
3966 return -EINVAL;
3967 if (min > mddev->resync_max)
3968 return -EINVAL;
3969 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3970 return -EBUSY;
3972 /* Must be a multiple of chunk_size */
3973 if (mddev->chunk_sectors) {
3974 sector_t temp = min;
3975 if (sector_div(temp, mddev->chunk_sectors))
3976 return -EINVAL;
3978 mddev->resync_min = min;
3980 return len;
3983 static struct md_sysfs_entry md_min_sync =
3984 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3986 static ssize_t
3987 max_sync_show(mddev_t *mddev, char *page)
3989 if (mddev->resync_max == MaxSector)
3990 return sprintf(page, "max\n");
3991 else
3992 return sprintf(page, "%llu\n",
3993 (unsigned long long)mddev->resync_max);
3995 static ssize_t
3996 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3998 if (strncmp(buf, "max", 3) == 0)
3999 mddev->resync_max = MaxSector;
4000 else {
4001 unsigned long long max;
4002 if (strict_strtoull(buf, 10, &max))
4003 return -EINVAL;
4004 if (max < mddev->resync_min)
4005 return -EINVAL;
4006 if (max < mddev->resync_max &&
4007 mddev->ro == 0 &&
4008 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4009 return -EBUSY;
4011 /* Must be a multiple of chunk_size */
4012 if (mddev->chunk_sectors) {
4013 sector_t temp = max;
4014 if (sector_div(temp, mddev->chunk_sectors))
4015 return -EINVAL;
4017 mddev->resync_max = max;
4019 wake_up(&mddev->recovery_wait);
4020 return len;
4023 static struct md_sysfs_entry md_max_sync =
4024 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4026 static ssize_t
4027 suspend_lo_show(mddev_t *mddev, char *page)
4029 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4032 static ssize_t
4033 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4035 char *e;
4036 unsigned long long new = simple_strtoull(buf, &e, 10);
4037 unsigned long long old = mddev->suspend_lo;
4039 if (mddev->pers == NULL ||
4040 mddev->pers->quiesce == NULL)
4041 return -EINVAL;
4042 if (buf == e || (*e && *e != '\n'))
4043 return -EINVAL;
4045 mddev->suspend_lo = new;
4046 if (new >= old)
4047 /* Shrinking suspended region */
4048 mddev->pers->quiesce(mddev, 2);
4049 else {
4050 /* Expanding suspended region - need to wait */
4051 mddev->pers->quiesce(mddev, 1);
4052 mddev->pers->quiesce(mddev, 0);
4054 return len;
4056 static struct md_sysfs_entry md_suspend_lo =
4057 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4060 static ssize_t
4061 suspend_hi_show(mddev_t *mddev, char *page)
4063 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4066 static ssize_t
4067 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4069 char *e;
4070 unsigned long long new = simple_strtoull(buf, &e, 10);
4071 unsigned long long old = mddev->suspend_hi;
4073 if (mddev->pers == NULL ||
4074 mddev->pers->quiesce == NULL)
4075 return -EINVAL;
4076 if (buf == e || (*e && *e != '\n'))
4077 return -EINVAL;
4079 mddev->suspend_hi = new;
4080 if (new <= old)
4081 /* Shrinking suspended region */
4082 mddev->pers->quiesce(mddev, 2);
4083 else {
4084 /* Expanding suspended region - need to wait */
4085 mddev->pers->quiesce(mddev, 1);
4086 mddev->pers->quiesce(mddev, 0);
4088 return len;
4090 static struct md_sysfs_entry md_suspend_hi =
4091 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4093 static ssize_t
4094 reshape_position_show(mddev_t *mddev, char *page)
4096 if (mddev->reshape_position != MaxSector)
4097 return sprintf(page, "%llu\n",
4098 (unsigned long long)mddev->reshape_position);
4099 strcpy(page, "none\n");
4100 return 5;
4103 static ssize_t
4104 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4106 char *e;
4107 unsigned long long new = simple_strtoull(buf, &e, 10);
4108 if (mddev->pers)
4109 return -EBUSY;
4110 if (buf == e || (*e && *e != '\n'))
4111 return -EINVAL;
4112 mddev->reshape_position = new;
4113 mddev->delta_disks = 0;
4114 mddev->new_level = mddev->level;
4115 mddev->new_layout = mddev->layout;
4116 mddev->new_chunk_sectors = mddev->chunk_sectors;
4117 return len;
4120 static struct md_sysfs_entry md_reshape_position =
4121 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4122 reshape_position_store);
4124 static ssize_t
4125 array_size_show(mddev_t *mddev, char *page)
4127 if (mddev->external_size)
4128 return sprintf(page, "%llu\n",
4129 (unsigned long long)mddev->array_sectors/2);
4130 else
4131 return sprintf(page, "default\n");
4134 static ssize_t
4135 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4137 sector_t sectors;
4139 if (strncmp(buf, "default", 7) == 0) {
4140 if (mddev->pers)
4141 sectors = mddev->pers->size(mddev, 0, 0);
4142 else
4143 sectors = mddev->array_sectors;
4145 mddev->external_size = 0;
4146 } else {
4147 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4148 return -EINVAL;
4149 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4150 return -E2BIG;
4152 mddev->external_size = 1;
4155 mddev->array_sectors = sectors;
4156 if (mddev->pers) {
4157 set_capacity(mddev->gendisk, mddev->array_sectors);
4158 revalidate_disk(mddev->gendisk);
4160 return len;
4163 static struct md_sysfs_entry md_array_size =
4164 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4165 array_size_store);
4167 static struct attribute *md_default_attrs[] = {
4168 &md_level.attr,
4169 &md_layout.attr,
4170 &md_raid_disks.attr,
4171 &md_chunk_size.attr,
4172 &md_size.attr,
4173 &md_resync_start.attr,
4174 &md_metadata.attr,
4175 &md_new_device.attr,
4176 &md_safe_delay.attr,
4177 &md_array_state.attr,
4178 &md_reshape_position.attr,
4179 &md_array_size.attr,
4180 &max_corr_read_errors.attr,
4181 NULL,
4184 static struct attribute *md_redundancy_attrs[] = {
4185 &md_scan_mode.attr,
4186 &md_mismatches.attr,
4187 &md_sync_min.attr,
4188 &md_sync_max.attr,
4189 &md_sync_speed.attr,
4190 &md_sync_force_parallel.attr,
4191 &md_sync_completed.attr,
4192 &md_min_sync.attr,
4193 &md_max_sync.attr,
4194 &md_suspend_lo.attr,
4195 &md_suspend_hi.attr,
4196 &md_bitmap.attr,
4197 &md_degraded.attr,
4198 NULL,
4200 static struct attribute_group md_redundancy_group = {
4201 .name = NULL,
4202 .attrs = md_redundancy_attrs,
4206 static ssize_t
4207 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4209 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4210 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4211 ssize_t rv;
4213 if (!entry->show)
4214 return -EIO;
4215 rv = mddev_lock(mddev);
4216 if (!rv) {
4217 rv = entry->show(mddev, page);
4218 mddev_unlock(mddev);
4220 return rv;
4223 static ssize_t
4224 md_attr_store(struct kobject *kobj, struct attribute *attr,
4225 const char *page, size_t length)
4227 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4228 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4229 ssize_t rv;
4231 if (!entry->store)
4232 return -EIO;
4233 if (!capable(CAP_SYS_ADMIN))
4234 return -EACCES;
4235 rv = mddev_lock(mddev);
4236 if (mddev->hold_active == UNTIL_IOCTL)
4237 mddev->hold_active = 0;
4238 if (!rv) {
4239 rv = entry->store(mddev, page, length);
4240 mddev_unlock(mddev);
4242 return rv;
4245 static void md_free(struct kobject *ko)
4247 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4249 if (mddev->sysfs_state)
4250 sysfs_put(mddev->sysfs_state);
4252 if (mddev->gendisk) {
4253 del_gendisk(mddev->gendisk);
4254 put_disk(mddev->gendisk);
4256 if (mddev->queue)
4257 blk_cleanup_queue(mddev->queue);
4259 kfree(mddev);
4262 static const struct sysfs_ops md_sysfs_ops = {
4263 .show = md_attr_show,
4264 .store = md_attr_store,
4266 static struct kobj_type md_ktype = {
4267 .release = md_free,
4268 .sysfs_ops = &md_sysfs_ops,
4269 .default_attrs = md_default_attrs,
4272 int mdp_major = 0;
4274 static void mddev_delayed_delete(struct work_struct *ws)
4276 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4278 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4279 kobject_del(&mddev->kobj);
4280 kobject_put(&mddev->kobj);
4283 static int md_alloc(dev_t dev, char *name)
4285 static DEFINE_MUTEX(disks_mutex);
4286 mddev_t *mddev = mddev_find(dev);
4287 struct gendisk *disk;
4288 int partitioned;
4289 int shift;
4290 int unit;
4291 int error;
4293 if (!mddev)
4294 return -ENODEV;
4296 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4297 shift = partitioned ? MdpMinorShift : 0;
4298 unit = MINOR(mddev->unit) >> shift;
4300 /* wait for any previous instance of this device to be
4301 * completely removed (mddev_delayed_delete).
4303 flush_workqueue(md_misc_wq);
4305 mutex_lock(&disks_mutex);
4306 error = -EEXIST;
4307 if (mddev->gendisk)
4308 goto abort;
4310 if (name) {
4311 /* Need to ensure that 'name' is not a duplicate.
4313 mddev_t *mddev2;
4314 spin_lock(&all_mddevs_lock);
4316 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4317 if (mddev2->gendisk &&
4318 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4319 spin_unlock(&all_mddevs_lock);
4320 goto abort;
4322 spin_unlock(&all_mddevs_lock);
4325 error = -ENOMEM;
4326 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4327 if (!mddev->queue)
4328 goto abort;
4329 mddev->queue->queuedata = mddev;
4331 blk_queue_make_request(mddev->queue, md_make_request);
4333 disk = alloc_disk(1 << shift);
4334 if (!disk) {
4335 blk_cleanup_queue(mddev->queue);
4336 mddev->queue = NULL;
4337 goto abort;
4339 disk->major = MAJOR(mddev->unit);
4340 disk->first_minor = unit << shift;
4341 if (name)
4342 strcpy(disk->disk_name, name);
4343 else if (partitioned)
4344 sprintf(disk->disk_name, "md_d%d", unit);
4345 else
4346 sprintf(disk->disk_name, "md%d", unit);
4347 disk->fops = &md_fops;
4348 disk->private_data = mddev;
4349 disk->queue = mddev->queue;
4350 /* Allow extended partitions. This makes the
4351 * 'mdp' device redundant, but we can't really
4352 * remove it now.
4354 disk->flags |= GENHD_FL_EXT_DEVT;
4355 add_disk(disk);
4356 mddev->gendisk = disk;
4357 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4358 &disk_to_dev(disk)->kobj, "%s", "md");
4359 if (error) {
4360 /* This isn't possible, but as kobject_init_and_add is marked
4361 * __must_check, we must do something with the result
4363 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4364 disk->disk_name);
4365 error = 0;
4367 if (mddev->kobj.sd &&
4368 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4369 printk(KERN_DEBUG "pointless warning\n");
4371 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4372 abort:
4373 mutex_unlock(&disks_mutex);
4374 if (!error && mddev->kobj.sd) {
4375 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4376 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4378 mddev_put(mddev);
4379 return error;
4382 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4384 md_alloc(dev, NULL);
4385 return NULL;
4388 static int add_named_array(const char *val, struct kernel_param *kp)
4390 /* val must be "md_*" where * is not all digits.
4391 * We allocate an array with a large free minor number, and
4392 * set the name to val. val must not already be an active name.
4394 int len = strlen(val);
4395 char buf[DISK_NAME_LEN];
4397 while (len && val[len-1] == '\n')
4398 len--;
4399 if (len >= DISK_NAME_LEN)
4400 return -E2BIG;
4401 strlcpy(buf, val, len+1);
4402 if (strncmp(buf, "md_", 3) != 0)
4403 return -EINVAL;
4404 return md_alloc(0, buf);
4407 static void md_safemode_timeout(unsigned long data)
4409 mddev_t *mddev = (mddev_t *) data;
4411 if (!atomic_read(&mddev->writes_pending)) {
4412 mddev->safemode = 1;
4413 if (mddev->external)
4414 sysfs_notify_dirent_safe(mddev->sysfs_state);
4416 md_wakeup_thread(mddev->thread);
4419 static int start_dirty_degraded;
4421 int md_run(mddev_t *mddev)
4423 int err;
4424 mdk_rdev_t *rdev;
4425 struct mdk_personality *pers;
4427 if (list_empty(&mddev->disks))
4428 /* cannot run an array with no devices.. */
4429 return -EINVAL;
4431 if (mddev->pers)
4432 return -EBUSY;
4433 /* Cannot run until previous stop completes properly */
4434 if (mddev->sysfs_active)
4435 return -EBUSY;
4438 * Analyze all RAID superblock(s)
4440 if (!mddev->raid_disks) {
4441 if (!mddev->persistent)
4442 return -EINVAL;
4443 analyze_sbs(mddev);
4446 if (mddev->level != LEVEL_NONE)
4447 request_module("md-level-%d", mddev->level);
4448 else if (mddev->clevel[0])
4449 request_module("md-%s", mddev->clevel);
4452 * Drop all container device buffers, from now on
4453 * the only valid external interface is through the md
4454 * device.
4456 list_for_each_entry(rdev, &mddev->disks, same_set) {
4457 if (test_bit(Faulty, &rdev->flags))
4458 continue;
4459 sync_blockdev(rdev->bdev);
4460 invalidate_bdev(rdev->bdev);
4462 /* perform some consistency tests on the device.
4463 * We don't want the data to overlap the metadata,
4464 * Internal Bitmap issues have been handled elsewhere.
4466 if (rdev->meta_bdev) {
4467 /* Nothing to check */;
4468 } else if (rdev->data_offset < rdev->sb_start) {
4469 if (mddev->dev_sectors &&
4470 rdev->data_offset + mddev->dev_sectors
4471 > rdev->sb_start) {
4472 printk("md: %s: data overlaps metadata\n",
4473 mdname(mddev));
4474 return -EINVAL;
4476 } else {
4477 if (rdev->sb_start + rdev->sb_size/512
4478 > rdev->data_offset) {
4479 printk("md: %s: metadata overlaps data\n",
4480 mdname(mddev));
4481 return -EINVAL;
4484 sysfs_notify_dirent_safe(rdev->sysfs_state);
4487 if (mddev->bio_set == NULL)
4488 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4490 spin_lock(&pers_lock);
4491 pers = find_pers(mddev->level, mddev->clevel);
4492 if (!pers || !try_module_get(pers->owner)) {
4493 spin_unlock(&pers_lock);
4494 if (mddev->level != LEVEL_NONE)
4495 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4496 mddev->level);
4497 else
4498 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4499 mddev->clevel);
4500 return -EINVAL;
4502 mddev->pers = pers;
4503 spin_unlock(&pers_lock);
4504 if (mddev->level != pers->level) {
4505 mddev->level = pers->level;
4506 mddev->new_level = pers->level;
4508 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4510 if (mddev->reshape_position != MaxSector &&
4511 pers->start_reshape == NULL) {
4512 /* This personality cannot handle reshaping... */
4513 mddev->pers = NULL;
4514 module_put(pers->owner);
4515 return -EINVAL;
4518 if (pers->sync_request) {
4519 /* Warn if this is a potentially silly
4520 * configuration.
4522 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4523 mdk_rdev_t *rdev2;
4524 int warned = 0;
4526 list_for_each_entry(rdev, &mddev->disks, same_set)
4527 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4528 if (rdev < rdev2 &&
4529 rdev->bdev->bd_contains ==
4530 rdev2->bdev->bd_contains) {
4531 printk(KERN_WARNING
4532 "%s: WARNING: %s appears to be"
4533 " on the same physical disk as"
4534 " %s.\n",
4535 mdname(mddev),
4536 bdevname(rdev->bdev,b),
4537 bdevname(rdev2->bdev,b2));
4538 warned = 1;
4542 if (warned)
4543 printk(KERN_WARNING
4544 "True protection against single-disk"
4545 " failure might be compromised.\n");
4548 mddev->recovery = 0;
4549 /* may be over-ridden by personality */
4550 mddev->resync_max_sectors = mddev->dev_sectors;
4552 mddev->ok_start_degraded = start_dirty_degraded;
4554 if (start_readonly && mddev->ro == 0)
4555 mddev->ro = 2; /* read-only, but switch on first write */
4557 err = mddev->pers->run(mddev);
4558 if (err)
4559 printk(KERN_ERR "md: pers->run() failed ...\n");
4560 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4561 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4562 " but 'external_size' not in effect?\n", __func__);
4563 printk(KERN_ERR
4564 "md: invalid array_size %llu > default size %llu\n",
4565 (unsigned long long)mddev->array_sectors / 2,
4566 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4567 err = -EINVAL;
4568 mddev->pers->stop(mddev);
4570 if (err == 0 && mddev->pers->sync_request) {
4571 err = bitmap_create(mddev);
4572 if (err) {
4573 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4574 mdname(mddev), err);
4575 mddev->pers->stop(mddev);
4578 if (err) {
4579 module_put(mddev->pers->owner);
4580 mddev->pers = NULL;
4581 bitmap_destroy(mddev);
4582 return err;
4584 if (mddev->pers->sync_request) {
4585 if (mddev->kobj.sd &&
4586 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4587 printk(KERN_WARNING
4588 "md: cannot register extra attributes for %s\n",
4589 mdname(mddev));
4590 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4591 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4592 mddev->ro = 0;
4594 atomic_set(&mddev->writes_pending,0);
4595 atomic_set(&mddev->max_corr_read_errors,
4596 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4597 mddev->safemode = 0;
4598 mddev->safemode_timer.function = md_safemode_timeout;
4599 mddev->safemode_timer.data = (unsigned long) mddev;
4600 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4601 mddev->in_sync = 1;
4602 smp_wmb();
4603 mddev->ready = 1;
4604 list_for_each_entry(rdev, &mddev->disks, same_set)
4605 if (rdev->raid_disk >= 0) {
4606 char nm[20];
4607 sprintf(nm, "rd%d", rdev->raid_disk);
4608 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4609 /* failure here is OK */;
4612 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4614 if (mddev->flags)
4615 md_update_sb(mddev, 0);
4617 md_wakeup_thread(mddev->thread);
4618 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4620 md_new_event(mddev);
4621 sysfs_notify_dirent_safe(mddev->sysfs_state);
4622 sysfs_notify_dirent_safe(mddev->sysfs_action);
4623 sysfs_notify(&mddev->kobj, NULL, "degraded");
4624 return 0;
4626 EXPORT_SYMBOL_GPL(md_run);
4628 static int do_md_run(mddev_t *mddev)
4630 int err;
4632 err = md_run(mddev);
4633 if (err)
4634 goto out;
4635 err = bitmap_load(mddev);
4636 if (err) {
4637 bitmap_destroy(mddev);
4638 goto out;
4640 set_capacity(mddev->gendisk, mddev->array_sectors);
4641 revalidate_disk(mddev->gendisk);
4642 mddev->changed = 1;
4643 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4644 out:
4645 return err;
4648 static int restart_array(mddev_t *mddev)
4650 struct gendisk *disk = mddev->gendisk;
4652 /* Complain if it has no devices */
4653 if (list_empty(&mddev->disks))
4654 return -ENXIO;
4655 if (!mddev->pers)
4656 return -EINVAL;
4657 if (!mddev->ro)
4658 return -EBUSY;
4659 mddev->safemode = 0;
4660 mddev->ro = 0;
4661 set_disk_ro(disk, 0);
4662 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4663 mdname(mddev));
4664 /* Kick recovery or resync if necessary */
4665 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4666 md_wakeup_thread(mddev->thread);
4667 md_wakeup_thread(mddev->sync_thread);
4668 sysfs_notify_dirent_safe(mddev->sysfs_state);
4669 return 0;
4672 /* similar to deny_write_access, but accounts for our holding a reference
4673 * to the file ourselves */
4674 static int deny_bitmap_write_access(struct file * file)
4676 struct inode *inode = file->f_mapping->host;
4678 spin_lock(&inode->i_lock);
4679 if (atomic_read(&inode->i_writecount) > 1) {
4680 spin_unlock(&inode->i_lock);
4681 return -ETXTBSY;
4683 atomic_set(&inode->i_writecount, -1);
4684 spin_unlock(&inode->i_lock);
4686 return 0;
4689 void restore_bitmap_write_access(struct file *file)
4691 struct inode *inode = file->f_mapping->host;
4693 spin_lock(&inode->i_lock);
4694 atomic_set(&inode->i_writecount, 1);
4695 spin_unlock(&inode->i_lock);
4698 static void md_clean(mddev_t *mddev)
4700 mddev->array_sectors = 0;
4701 mddev->external_size = 0;
4702 mddev->dev_sectors = 0;
4703 mddev->raid_disks = 0;
4704 mddev->recovery_cp = 0;
4705 mddev->resync_min = 0;
4706 mddev->resync_max = MaxSector;
4707 mddev->reshape_position = MaxSector;
4708 mddev->external = 0;
4709 mddev->persistent = 0;
4710 mddev->level = LEVEL_NONE;
4711 mddev->clevel[0] = 0;
4712 mddev->flags = 0;
4713 mddev->ro = 0;
4714 mddev->metadata_type[0] = 0;
4715 mddev->chunk_sectors = 0;
4716 mddev->ctime = mddev->utime = 0;
4717 mddev->layout = 0;
4718 mddev->max_disks = 0;
4719 mddev->events = 0;
4720 mddev->can_decrease_events = 0;
4721 mddev->delta_disks = 0;
4722 mddev->new_level = LEVEL_NONE;
4723 mddev->new_layout = 0;
4724 mddev->new_chunk_sectors = 0;
4725 mddev->curr_resync = 0;
4726 mddev->resync_mismatches = 0;
4727 mddev->suspend_lo = mddev->suspend_hi = 0;
4728 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4729 mddev->recovery = 0;
4730 mddev->in_sync = 0;
4731 mddev->changed = 0;
4732 mddev->degraded = 0;
4733 mddev->safemode = 0;
4734 mddev->bitmap_info.offset = 0;
4735 mddev->bitmap_info.default_offset = 0;
4736 mddev->bitmap_info.chunksize = 0;
4737 mddev->bitmap_info.daemon_sleep = 0;
4738 mddev->bitmap_info.max_write_behind = 0;
4741 static void __md_stop_writes(mddev_t *mddev)
4743 if (mddev->sync_thread) {
4744 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4745 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4746 reap_sync_thread(mddev);
4749 del_timer_sync(&mddev->safemode_timer);
4751 bitmap_flush(mddev);
4752 md_super_wait(mddev);
4754 if (!mddev->in_sync || mddev->flags) {
4755 /* mark array as shutdown cleanly */
4756 mddev->in_sync = 1;
4757 md_update_sb(mddev, 1);
4761 void md_stop_writes(mddev_t *mddev)
4763 mddev_lock(mddev);
4764 __md_stop_writes(mddev);
4765 mddev_unlock(mddev);
4767 EXPORT_SYMBOL_GPL(md_stop_writes);
4769 void md_stop(mddev_t *mddev)
4771 mddev->ready = 0;
4772 mddev->pers->stop(mddev);
4773 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4774 mddev->to_remove = &md_redundancy_group;
4775 module_put(mddev->pers->owner);
4776 mddev->pers = NULL;
4777 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4779 EXPORT_SYMBOL_GPL(md_stop);
4781 static int md_set_readonly(mddev_t *mddev, int is_open)
4783 int err = 0;
4784 mutex_lock(&mddev->open_mutex);
4785 if (atomic_read(&mddev->openers) > is_open) {
4786 printk("md: %s still in use.\n",mdname(mddev));
4787 err = -EBUSY;
4788 goto out;
4790 if (mddev->pers) {
4791 __md_stop_writes(mddev);
4793 err = -ENXIO;
4794 if (mddev->ro==1)
4795 goto out;
4796 mddev->ro = 1;
4797 set_disk_ro(mddev->gendisk, 1);
4798 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4799 sysfs_notify_dirent_safe(mddev->sysfs_state);
4800 err = 0;
4802 out:
4803 mutex_unlock(&mddev->open_mutex);
4804 return err;
4807 /* mode:
4808 * 0 - completely stop and dis-assemble array
4809 * 2 - stop but do not disassemble array
4811 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4813 struct gendisk *disk = mddev->gendisk;
4814 mdk_rdev_t *rdev;
4816 mutex_lock(&mddev->open_mutex);
4817 if (atomic_read(&mddev->openers) > is_open ||
4818 mddev->sysfs_active) {
4819 printk("md: %s still in use.\n",mdname(mddev));
4820 mutex_unlock(&mddev->open_mutex);
4821 return -EBUSY;
4824 if (mddev->pers) {
4825 if (mddev->ro)
4826 set_disk_ro(disk, 0);
4828 __md_stop_writes(mddev);
4829 md_stop(mddev);
4830 mddev->queue->merge_bvec_fn = NULL;
4831 mddev->queue->backing_dev_info.congested_fn = NULL;
4833 /* tell userspace to handle 'inactive' */
4834 sysfs_notify_dirent_safe(mddev->sysfs_state);
4836 list_for_each_entry(rdev, &mddev->disks, same_set)
4837 if (rdev->raid_disk >= 0) {
4838 char nm[20];
4839 sprintf(nm, "rd%d", rdev->raid_disk);
4840 sysfs_remove_link(&mddev->kobj, nm);
4843 set_capacity(disk, 0);
4844 mutex_unlock(&mddev->open_mutex);
4845 mddev->changed = 1;
4846 revalidate_disk(disk);
4848 if (mddev->ro)
4849 mddev->ro = 0;
4850 } else
4851 mutex_unlock(&mddev->open_mutex);
4853 * Free resources if final stop
4855 if (mode == 0) {
4856 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4858 bitmap_destroy(mddev);
4859 if (mddev->bitmap_info.file) {
4860 restore_bitmap_write_access(mddev->bitmap_info.file);
4861 fput(mddev->bitmap_info.file);
4862 mddev->bitmap_info.file = NULL;
4864 mddev->bitmap_info.offset = 0;
4866 export_array(mddev);
4868 md_clean(mddev);
4869 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4870 if (mddev->hold_active == UNTIL_STOP)
4871 mddev->hold_active = 0;
4873 blk_integrity_unregister(disk);
4874 md_new_event(mddev);
4875 sysfs_notify_dirent_safe(mddev->sysfs_state);
4876 return 0;
4879 #ifndef MODULE
4880 static void autorun_array(mddev_t *mddev)
4882 mdk_rdev_t *rdev;
4883 int err;
4885 if (list_empty(&mddev->disks))
4886 return;
4888 printk(KERN_INFO "md: running: ");
4890 list_for_each_entry(rdev, &mddev->disks, same_set) {
4891 char b[BDEVNAME_SIZE];
4892 printk("<%s>", bdevname(rdev->bdev,b));
4894 printk("\n");
4896 err = do_md_run(mddev);
4897 if (err) {
4898 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4899 do_md_stop(mddev, 0, 0);
4904 * lets try to run arrays based on all disks that have arrived
4905 * until now. (those are in pending_raid_disks)
4907 * the method: pick the first pending disk, collect all disks with
4908 * the same UUID, remove all from the pending list and put them into
4909 * the 'same_array' list. Then order this list based on superblock
4910 * update time (freshest comes first), kick out 'old' disks and
4911 * compare superblocks. If everything's fine then run it.
4913 * If "unit" is allocated, then bump its reference count
4915 static void autorun_devices(int part)
4917 mdk_rdev_t *rdev0, *rdev, *tmp;
4918 mddev_t *mddev;
4919 char b[BDEVNAME_SIZE];
4921 printk(KERN_INFO "md: autorun ...\n");
4922 while (!list_empty(&pending_raid_disks)) {
4923 int unit;
4924 dev_t dev;
4925 LIST_HEAD(candidates);
4926 rdev0 = list_entry(pending_raid_disks.next,
4927 mdk_rdev_t, same_set);
4929 printk(KERN_INFO "md: considering %s ...\n",
4930 bdevname(rdev0->bdev,b));
4931 INIT_LIST_HEAD(&candidates);
4932 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4933 if (super_90_load(rdev, rdev0, 0) >= 0) {
4934 printk(KERN_INFO "md: adding %s ...\n",
4935 bdevname(rdev->bdev,b));
4936 list_move(&rdev->same_set, &candidates);
4939 * now we have a set of devices, with all of them having
4940 * mostly sane superblocks. It's time to allocate the
4941 * mddev.
4943 if (part) {
4944 dev = MKDEV(mdp_major,
4945 rdev0->preferred_minor << MdpMinorShift);
4946 unit = MINOR(dev) >> MdpMinorShift;
4947 } else {
4948 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4949 unit = MINOR(dev);
4951 if (rdev0->preferred_minor != unit) {
4952 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4953 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4954 break;
4957 md_probe(dev, NULL, NULL);
4958 mddev = mddev_find(dev);
4959 if (!mddev || !mddev->gendisk) {
4960 if (mddev)
4961 mddev_put(mddev);
4962 printk(KERN_ERR
4963 "md: cannot allocate memory for md drive.\n");
4964 break;
4966 if (mddev_lock(mddev))
4967 printk(KERN_WARNING "md: %s locked, cannot run\n",
4968 mdname(mddev));
4969 else if (mddev->raid_disks || mddev->major_version
4970 || !list_empty(&mddev->disks)) {
4971 printk(KERN_WARNING
4972 "md: %s already running, cannot run %s\n",
4973 mdname(mddev), bdevname(rdev0->bdev,b));
4974 mddev_unlock(mddev);
4975 } else {
4976 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4977 mddev->persistent = 1;
4978 rdev_for_each_list(rdev, tmp, &candidates) {
4979 list_del_init(&rdev->same_set);
4980 if (bind_rdev_to_array(rdev, mddev))
4981 export_rdev(rdev);
4983 autorun_array(mddev);
4984 mddev_unlock(mddev);
4986 /* on success, candidates will be empty, on error
4987 * it won't...
4989 rdev_for_each_list(rdev, tmp, &candidates) {
4990 list_del_init(&rdev->same_set);
4991 export_rdev(rdev);
4993 mddev_put(mddev);
4995 printk(KERN_INFO "md: ... autorun DONE.\n");
4997 #endif /* !MODULE */
4999 static int get_version(void __user * arg)
5001 mdu_version_t ver;
5003 ver.major = MD_MAJOR_VERSION;
5004 ver.minor = MD_MINOR_VERSION;
5005 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5007 if (copy_to_user(arg, &ver, sizeof(ver)))
5008 return -EFAULT;
5010 return 0;
5013 static int get_array_info(mddev_t * mddev, void __user * arg)
5015 mdu_array_info_t info;
5016 int nr,working,insync,failed,spare;
5017 mdk_rdev_t *rdev;
5019 nr=working=insync=failed=spare=0;
5020 list_for_each_entry(rdev, &mddev->disks, same_set) {
5021 nr++;
5022 if (test_bit(Faulty, &rdev->flags))
5023 failed++;
5024 else {
5025 working++;
5026 if (test_bit(In_sync, &rdev->flags))
5027 insync++;
5028 else
5029 spare++;
5033 info.major_version = mddev->major_version;
5034 info.minor_version = mddev->minor_version;
5035 info.patch_version = MD_PATCHLEVEL_VERSION;
5036 info.ctime = mddev->ctime;
5037 info.level = mddev->level;
5038 info.size = mddev->dev_sectors / 2;
5039 if (info.size != mddev->dev_sectors / 2) /* overflow */
5040 info.size = -1;
5041 info.nr_disks = nr;
5042 info.raid_disks = mddev->raid_disks;
5043 info.md_minor = mddev->md_minor;
5044 info.not_persistent= !mddev->persistent;
5046 info.utime = mddev->utime;
5047 info.state = 0;
5048 if (mddev->in_sync)
5049 info.state = (1<<MD_SB_CLEAN);
5050 if (mddev->bitmap && mddev->bitmap_info.offset)
5051 info.state = (1<<MD_SB_BITMAP_PRESENT);
5052 info.active_disks = insync;
5053 info.working_disks = working;
5054 info.failed_disks = failed;
5055 info.spare_disks = spare;
5057 info.layout = mddev->layout;
5058 info.chunk_size = mddev->chunk_sectors << 9;
5060 if (copy_to_user(arg, &info, sizeof(info)))
5061 return -EFAULT;
5063 return 0;
5066 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5068 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5069 char *ptr, *buf = NULL;
5070 int err = -ENOMEM;
5072 if (md_allow_write(mddev))
5073 file = kmalloc(sizeof(*file), GFP_NOIO);
5074 else
5075 file = kmalloc(sizeof(*file), GFP_KERNEL);
5077 if (!file)
5078 goto out;
5080 /* bitmap disabled, zero the first byte and copy out */
5081 if (!mddev->bitmap || !mddev->bitmap->file) {
5082 file->pathname[0] = '\0';
5083 goto copy_out;
5086 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5087 if (!buf)
5088 goto out;
5090 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5091 if (IS_ERR(ptr))
5092 goto out;
5094 strcpy(file->pathname, ptr);
5096 copy_out:
5097 err = 0;
5098 if (copy_to_user(arg, file, sizeof(*file)))
5099 err = -EFAULT;
5100 out:
5101 kfree(buf);
5102 kfree(file);
5103 return err;
5106 static int get_disk_info(mddev_t * mddev, void __user * arg)
5108 mdu_disk_info_t info;
5109 mdk_rdev_t *rdev;
5111 if (copy_from_user(&info, arg, sizeof(info)))
5112 return -EFAULT;
5114 rdev = find_rdev_nr(mddev, info.number);
5115 if (rdev) {
5116 info.major = MAJOR(rdev->bdev->bd_dev);
5117 info.minor = MINOR(rdev->bdev->bd_dev);
5118 info.raid_disk = rdev->raid_disk;
5119 info.state = 0;
5120 if (test_bit(Faulty, &rdev->flags))
5121 info.state |= (1<<MD_DISK_FAULTY);
5122 else if (test_bit(In_sync, &rdev->flags)) {
5123 info.state |= (1<<MD_DISK_ACTIVE);
5124 info.state |= (1<<MD_DISK_SYNC);
5126 if (test_bit(WriteMostly, &rdev->flags))
5127 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5128 } else {
5129 info.major = info.minor = 0;
5130 info.raid_disk = -1;
5131 info.state = (1<<MD_DISK_REMOVED);
5134 if (copy_to_user(arg, &info, sizeof(info)))
5135 return -EFAULT;
5137 return 0;
5140 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5142 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5143 mdk_rdev_t *rdev;
5144 dev_t dev = MKDEV(info->major,info->minor);
5146 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5147 return -EOVERFLOW;
5149 if (!mddev->raid_disks) {
5150 int err;
5151 /* expecting a device which has a superblock */
5152 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5153 if (IS_ERR(rdev)) {
5154 printk(KERN_WARNING
5155 "md: md_import_device returned %ld\n",
5156 PTR_ERR(rdev));
5157 return PTR_ERR(rdev);
5159 if (!list_empty(&mddev->disks)) {
5160 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5161 mdk_rdev_t, same_set);
5162 err = super_types[mddev->major_version]
5163 .load_super(rdev, rdev0, mddev->minor_version);
5164 if (err < 0) {
5165 printk(KERN_WARNING
5166 "md: %s has different UUID to %s\n",
5167 bdevname(rdev->bdev,b),
5168 bdevname(rdev0->bdev,b2));
5169 export_rdev(rdev);
5170 return -EINVAL;
5173 err = bind_rdev_to_array(rdev, mddev);
5174 if (err)
5175 export_rdev(rdev);
5176 return err;
5180 * add_new_disk can be used once the array is assembled
5181 * to add "hot spares". They must already have a superblock
5182 * written
5184 if (mddev->pers) {
5185 int err;
5186 if (!mddev->pers->hot_add_disk) {
5187 printk(KERN_WARNING
5188 "%s: personality does not support diskops!\n",
5189 mdname(mddev));
5190 return -EINVAL;
5192 if (mddev->persistent)
5193 rdev = md_import_device(dev, mddev->major_version,
5194 mddev->minor_version);
5195 else
5196 rdev = md_import_device(dev, -1, -1);
5197 if (IS_ERR(rdev)) {
5198 printk(KERN_WARNING
5199 "md: md_import_device returned %ld\n",
5200 PTR_ERR(rdev));
5201 return PTR_ERR(rdev);
5203 /* set saved_raid_disk if appropriate */
5204 if (!mddev->persistent) {
5205 if (info->state & (1<<MD_DISK_SYNC) &&
5206 info->raid_disk < mddev->raid_disks) {
5207 rdev->raid_disk = info->raid_disk;
5208 set_bit(In_sync, &rdev->flags);
5209 } else
5210 rdev->raid_disk = -1;
5211 } else
5212 super_types[mddev->major_version].
5213 validate_super(mddev, rdev);
5214 if (test_bit(In_sync, &rdev->flags))
5215 rdev->saved_raid_disk = rdev->raid_disk;
5216 else
5217 rdev->saved_raid_disk = -1;
5219 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5220 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5221 set_bit(WriteMostly, &rdev->flags);
5222 else
5223 clear_bit(WriteMostly, &rdev->flags);
5225 rdev->raid_disk = -1;
5226 err = bind_rdev_to_array(rdev, mddev);
5227 if (!err && !mddev->pers->hot_remove_disk) {
5228 /* If there is hot_add_disk but no hot_remove_disk
5229 * then added disks for geometry changes,
5230 * and should be added immediately.
5232 super_types[mddev->major_version].
5233 validate_super(mddev, rdev);
5234 err = mddev->pers->hot_add_disk(mddev, rdev);
5235 if (err)
5236 unbind_rdev_from_array(rdev);
5238 if (err)
5239 export_rdev(rdev);
5240 else
5241 sysfs_notify_dirent_safe(rdev->sysfs_state);
5243 md_update_sb(mddev, 1);
5244 if (mddev->degraded)
5245 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5246 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5247 md_wakeup_thread(mddev->thread);
5248 return err;
5251 /* otherwise, add_new_disk is only allowed
5252 * for major_version==0 superblocks
5254 if (mddev->major_version != 0) {
5255 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5256 mdname(mddev));
5257 return -EINVAL;
5260 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5261 int err;
5262 rdev = md_import_device(dev, -1, 0);
5263 if (IS_ERR(rdev)) {
5264 printk(KERN_WARNING
5265 "md: error, md_import_device() returned %ld\n",
5266 PTR_ERR(rdev));
5267 return PTR_ERR(rdev);
5269 rdev->desc_nr = info->number;
5270 if (info->raid_disk < mddev->raid_disks)
5271 rdev->raid_disk = info->raid_disk;
5272 else
5273 rdev->raid_disk = -1;
5275 if (rdev->raid_disk < mddev->raid_disks)
5276 if (info->state & (1<<MD_DISK_SYNC))
5277 set_bit(In_sync, &rdev->flags);
5279 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5280 set_bit(WriteMostly, &rdev->flags);
5282 if (!mddev->persistent) {
5283 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5284 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5285 } else
5286 rdev->sb_start = calc_dev_sboffset(rdev);
5287 rdev->sectors = rdev->sb_start;
5289 err = bind_rdev_to_array(rdev, mddev);
5290 if (err) {
5291 export_rdev(rdev);
5292 return err;
5296 return 0;
5299 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5301 char b[BDEVNAME_SIZE];
5302 mdk_rdev_t *rdev;
5304 rdev = find_rdev(mddev, dev);
5305 if (!rdev)
5306 return -ENXIO;
5308 if (rdev->raid_disk >= 0)
5309 goto busy;
5311 kick_rdev_from_array(rdev);
5312 md_update_sb(mddev, 1);
5313 md_new_event(mddev);
5315 return 0;
5316 busy:
5317 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5318 bdevname(rdev->bdev,b), mdname(mddev));
5319 return -EBUSY;
5322 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5324 char b[BDEVNAME_SIZE];
5325 int err;
5326 mdk_rdev_t *rdev;
5328 if (!mddev->pers)
5329 return -ENODEV;
5331 if (mddev->major_version != 0) {
5332 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5333 " version-0 superblocks.\n",
5334 mdname(mddev));
5335 return -EINVAL;
5337 if (!mddev->pers->hot_add_disk) {
5338 printk(KERN_WARNING
5339 "%s: personality does not support diskops!\n",
5340 mdname(mddev));
5341 return -EINVAL;
5344 rdev = md_import_device(dev, -1, 0);
5345 if (IS_ERR(rdev)) {
5346 printk(KERN_WARNING
5347 "md: error, md_import_device() returned %ld\n",
5348 PTR_ERR(rdev));
5349 return -EINVAL;
5352 if (mddev->persistent)
5353 rdev->sb_start = calc_dev_sboffset(rdev);
5354 else
5355 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5357 rdev->sectors = rdev->sb_start;
5359 if (test_bit(Faulty, &rdev->flags)) {
5360 printk(KERN_WARNING
5361 "md: can not hot-add faulty %s disk to %s!\n",
5362 bdevname(rdev->bdev,b), mdname(mddev));
5363 err = -EINVAL;
5364 goto abort_export;
5366 clear_bit(In_sync, &rdev->flags);
5367 rdev->desc_nr = -1;
5368 rdev->saved_raid_disk = -1;
5369 err = bind_rdev_to_array(rdev, mddev);
5370 if (err)
5371 goto abort_export;
5374 * The rest should better be atomic, we can have disk failures
5375 * noticed in interrupt contexts ...
5378 rdev->raid_disk = -1;
5380 md_update_sb(mddev, 1);
5383 * Kick recovery, maybe this spare has to be added to the
5384 * array immediately.
5386 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5387 md_wakeup_thread(mddev->thread);
5388 md_new_event(mddev);
5389 return 0;
5391 abort_export:
5392 export_rdev(rdev);
5393 return err;
5396 static int set_bitmap_file(mddev_t *mddev, int fd)
5398 int err;
5400 if (mddev->pers) {
5401 if (!mddev->pers->quiesce)
5402 return -EBUSY;
5403 if (mddev->recovery || mddev->sync_thread)
5404 return -EBUSY;
5405 /* we should be able to change the bitmap.. */
5409 if (fd >= 0) {
5410 if (mddev->bitmap)
5411 return -EEXIST; /* cannot add when bitmap is present */
5412 mddev->bitmap_info.file = fget(fd);
5414 if (mddev->bitmap_info.file == NULL) {
5415 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5416 mdname(mddev));
5417 return -EBADF;
5420 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5421 if (err) {
5422 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5423 mdname(mddev));
5424 fput(mddev->bitmap_info.file);
5425 mddev->bitmap_info.file = NULL;
5426 return err;
5428 mddev->bitmap_info.offset = 0; /* file overrides offset */
5429 } else if (mddev->bitmap == NULL)
5430 return -ENOENT; /* cannot remove what isn't there */
5431 err = 0;
5432 if (mddev->pers) {
5433 mddev->pers->quiesce(mddev, 1);
5434 if (fd >= 0) {
5435 err = bitmap_create(mddev);
5436 if (!err)
5437 err = bitmap_load(mddev);
5439 if (fd < 0 || err) {
5440 bitmap_destroy(mddev);
5441 fd = -1; /* make sure to put the file */
5443 mddev->pers->quiesce(mddev, 0);
5445 if (fd < 0) {
5446 if (mddev->bitmap_info.file) {
5447 restore_bitmap_write_access(mddev->bitmap_info.file);
5448 fput(mddev->bitmap_info.file);
5450 mddev->bitmap_info.file = NULL;
5453 return err;
5457 * set_array_info is used two different ways
5458 * The original usage is when creating a new array.
5459 * In this usage, raid_disks is > 0 and it together with
5460 * level, size, not_persistent,layout,chunksize determine the
5461 * shape of the array.
5462 * This will always create an array with a type-0.90.0 superblock.
5463 * The newer usage is when assembling an array.
5464 * In this case raid_disks will be 0, and the major_version field is
5465 * use to determine which style super-blocks are to be found on the devices.
5466 * The minor and patch _version numbers are also kept incase the
5467 * super_block handler wishes to interpret them.
5469 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5472 if (info->raid_disks == 0) {
5473 /* just setting version number for superblock loading */
5474 if (info->major_version < 0 ||
5475 info->major_version >= ARRAY_SIZE(super_types) ||
5476 super_types[info->major_version].name == NULL) {
5477 /* maybe try to auto-load a module? */
5478 printk(KERN_INFO
5479 "md: superblock version %d not known\n",
5480 info->major_version);
5481 return -EINVAL;
5483 mddev->major_version = info->major_version;
5484 mddev->minor_version = info->minor_version;
5485 mddev->patch_version = info->patch_version;
5486 mddev->persistent = !info->not_persistent;
5487 /* ensure mddev_put doesn't delete this now that there
5488 * is some minimal configuration.
5490 mddev->ctime = get_seconds();
5491 return 0;
5493 mddev->major_version = MD_MAJOR_VERSION;
5494 mddev->minor_version = MD_MINOR_VERSION;
5495 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5496 mddev->ctime = get_seconds();
5498 mddev->level = info->level;
5499 mddev->clevel[0] = 0;
5500 mddev->dev_sectors = 2 * (sector_t)info->size;
5501 mddev->raid_disks = info->raid_disks;
5502 /* don't set md_minor, it is determined by which /dev/md* was
5503 * openned
5505 if (info->state & (1<<MD_SB_CLEAN))
5506 mddev->recovery_cp = MaxSector;
5507 else
5508 mddev->recovery_cp = 0;
5509 mddev->persistent = ! info->not_persistent;
5510 mddev->external = 0;
5512 mddev->layout = info->layout;
5513 mddev->chunk_sectors = info->chunk_size >> 9;
5515 mddev->max_disks = MD_SB_DISKS;
5517 if (mddev->persistent)
5518 mddev->flags = 0;
5519 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5521 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5522 mddev->bitmap_info.offset = 0;
5524 mddev->reshape_position = MaxSector;
5527 * Generate a 128 bit UUID
5529 get_random_bytes(mddev->uuid, 16);
5531 mddev->new_level = mddev->level;
5532 mddev->new_chunk_sectors = mddev->chunk_sectors;
5533 mddev->new_layout = mddev->layout;
5534 mddev->delta_disks = 0;
5536 return 0;
5539 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5541 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5543 if (mddev->external_size)
5544 return;
5546 mddev->array_sectors = array_sectors;
5548 EXPORT_SYMBOL(md_set_array_sectors);
5550 static int update_size(mddev_t *mddev, sector_t num_sectors)
5552 mdk_rdev_t *rdev;
5553 int rv;
5554 int fit = (num_sectors == 0);
5556 if (mddev->pers->resize == NULL)
5557 return -EINVAL;
5558 /* The "num_sectors" is the number of sectors of each device that
5559 * is used. This can only make sense for arrays with redundancy.
5560 * linear and raid0 always use whatever space is available. We can only
5561 * consider changing this number if no resync or reconstruction is
5562 * happening, and if the new size is acceptable. It must fit before the
5563 * sb_start or, if that is <data_offset, it must fit before the size
5564 * of each device. If num_sectors is zero, we find the largest size
5565 * that fits.
5567 if (mddev->sync_thread)
5568 return -EBUSY;
5569 if (mddev->bitmap)
5570 /* Sorry, cannot grow a bitmap yet, just remove it,
5571 * grow, and re-add.
5573 return -EBUSY;
5574 list_for_each_entry(rdev, &mddev->disks, same_set) {
5575 sector_t avail = rdev->sectors;
5577 if (fit && (num_sectors == 0 || num_sectors > avail))
5578 num_sectors = avail;
5579 if (avail < num_sectors)
5580 return -ENOSPC;
5582 rv = mddev->pers->resize(mddev, num_sectors);
5583 if (!rv)
5584 revalidate_disk(mddev->gendisk);
5585 return rv;
5588 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5590 int rv;
5591 /* change the number of raid disks */
5592 if (mddev->pers->check_reshape == NULL)
5593 return -EINVAL;
5594 if (raid_disks <= 0 ||
5595 (mddev->max_disks && raid_disks >= mddev->max_disks))
5596 return -EINVAL;
5597 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5598 return -EBUSY;
5599 mddev->delta_disks = raid_disks - mddev->raid_disks;
5601 rv = mddev->pers->check_reshape(mddev);
5602 if (rv < 0)
5603 mddev->delta_disks = 0;
5604 return rv;
5609 * update_array_info is used to change the configuration of an
5610 * on-line array.
5611 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5612 * fields in the info are checked against the array.
5613 * Any differences that cannot be handled will cause an error.
5614 * Normally, only one change can be managed at a time.
5616 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5618 int rv = 0;
5619 int cnt = 0;
5620 int state = 0;
5622 /* calculate expected state,ignoring low bits */
5623 if (mddev->bitmap && mddev->bitmap_info.offset)
5624 state |= (1 << MD_SB_BITMAP_PRESENT);
5626 if (mddev->major_version != info->major_version ||
5627 mddev->minor_version != info->minor_version ||
5628 /* mddev->patch_version != info->patch_version || */
5629 mddev->ctime != info->ctime ||
5630 mddev->level != info->level ||
5631 /* mddev->layout != info->layout || */
5632 !mddev->persistent != info->not_persistent||
5633 mddev->chunk_sectors != info->chunk_size >> 9 ||
5634 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5635 ((state^info->state) & 0xfffffe00)
5637 return -EINVAL;
5638 /* Check there is only one change */
5639 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5640 cnt++;
5641 if (mddev->raid_disks != info->raid_disks)
5642 cnt++;
5643 if (mddev->layout != info->layout)
5644 cnt++;
5645 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5646 cnt++;
5647 if (cnt == 0)
5648 return 0;
5649 if (cnt > 1)
5650 return -EINVAL;
5652 if (mddev->layout != info->layout) {
5653 /* Change layout
5654 * we don't need to do anything at the md level, the
5655 * personality will take care of it all.
5657 if (mddev->pers->check_reshape == NULL)
5658 return -EINVAL;
5659 else {
5660 mddev->new_layout = info->layout;
5661 rv = mddev->pers->check_reshape(mddev);
5662 if (rv)
5663 mddev->new_layout = mddev->layout;
5664 return rv;
5667 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5668 rv = update_size(mddev, (sector_t)info->size * 2);
5670 if (mddev->raid_disks != info->raid_disks)
5671 rv = update_raid_disks(mddev, info->raid_disks);
5673 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5674 if (mddev->pers->quiesce == NULL)
5675 return -EINVAL;
5676 if (mddev->recovery || mddev->sync_thread)
5677 return -EBUSY;
5678 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5679 /* add the bitmap */
5680 if (mddev->bitmap)
5681 return -EEXIST;
5682 if (mddev->bitmap_info.default_offset == 0)
5683 return -EINVAL;
5684 mddev->bitmap_info.offset =
5685 mddev->bitmap_info.default_offset;
5686 mddev->pers->quiesce(mddev, 1);
5687 rv = bitmap_create(mddev);
5688 if (!rv)
5689 rv = bitmap_load(mddev);
5690 if (rv)
5691 bitmap_destroy(mddev);
5692 mddev->pers->quiesce(mddev, 0);
5693 } else {
5694 /* remove the bitmap */
5695 if (!mddev->bitmap)
5696 return -ENOENT;
5697 if (mddev->bitmap->file)
5698 return -EINVAL;
5699 mddev->pers->quiesce(mddev, 1);
5700 bitmap_destroy(mddev);
5701 mddev->pers->quiesce(mddev, 0);
5702 mddev->bitmap_info.offset = 0;
5705 md_update_sb(mddev, 1);
5706 return rv;
5709 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5711 mdk_rdev_t *rdev;
5713 if (mddev->pers == NULL)
5714 return -ENODEV;
5716 rdev = find_rdev(mddev, dev);
5717 if (!rdev)
5718 return -ENODEV;
5720 md_error(mddev, rdev);
5721 return 0;
5725 * We have a problem here : there is no easy way to give a CHS
5726 * virtual geometry. We currently pretend that we have a 2 heads
5727 * 4 sectors (with a BIG number of cylinders...). This drives
5728 * dosfs just mad... ;-)
5730 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5732 mddev_t *mddev = bdev->bd_disk->private_data;
5734 geo->heads = 2;
5735 geo->sectors = 4;
5736 geo->cylinders = mddev->array_sectors / 8;
5737 return 0;
5740 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5741 unsigned int cmd, unsigned long arg)
5743 int err = 0;
5744 void __user *argp = (void __user *)arg;
5745 mddev_t *mddev = NULL;
5746 int ro;
5748 if (!capable(CAP_SYS_ADMIN))
5749 return -EACCES;
5752 * Commands dealing with the RAID driver but not any
5753 * particular array:
5755 switch (cmd)
5757 case RAID_VERSION:
5758 err = get_version(argp);
5759 goto done;
5761 case PRINT_RAID_DEBUG:
5762 err = 0;
5763 md_print_devices();
5764 goto done;
5766 #ifndef MODULE
5767 case RAID_AUTORUN:
5768 err = 0;
5769 autostart_arrays(arg);
5770 goto done;
5771 #endif
5772 default:;
5776 * Commands creating/starting a new array:
5779 mddev = bdev->bd_disk->private_data;
5781 if (!mddev) {
5782 BUG();
5783 goto abort;
5786 err = mddev_lock(mddev);
5787 if (err) {
5788 printk(KERN_INFO
5789 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5790 err, cmd);
5791 goto abort;
5794 switch (cmd)
5796 case SET_ARRAY_INFO:
5798 mdu_array_info_t info;
5799 if (!arg)
5800 memset(&info, 0, sizeof(info));
5801 else if (copy_from_user(&info, argp, sizeof(info))) {
5802 err = -EFAULT;
5803 goto abort_unlock;
5805 if (mddev->pers) {
5806 err = update_array_info(mddev, &info);
5807 if (err) {
5808 printk(KERN_WARNING "md: couldn't update"
5809 " array info. %d\n", err);
5810 goto abort_unlock;
5812 goto done_unlock;
5814 if (!list_empty(&mddev->disks)) {
5815 printk(KERN_WARNING
5816 "md: array %s already has disks!\n",
5817 mdname(mddev));
5818 err = -EBUSY;
5819 goto abort_unlock;
5821 if (mddev->raid_disks) {
5822 printk(KERN_WARNING
5823 "md: array %s already initialised!\n",
5824 mdname(mddev));
5825 err = -EBUSY;
5826 goto abort_unlock;
5828 err = set_array_info(mddev, &info);
5829 if (err) {
5830 printk(KERN_WARNING "md: couldn't set"
5831 " array info. %d\n", err);
5832 goto abort_unlock;
5835 goto done_unlock;
5837 default:;
5841 * Commands querying/configuring an existing array:
5843 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5844 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5845 if ((!mddev->raid_disks && !mddev->external)
5846 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5847 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5848 && cmd != GET_BITMAP_FILE) {
5849 err = -ENODEV;
5850 goto abort_unlock;
5854 * Commands even a read-only array can execute:
5856 switch (cmd)
5858 case GET_ARRAY_INFO:
5859 err = get_array_info(mddev, argp);
5860 goto done_unlock;
5862 case GET_BITMAP_FILE:
5863 err = get_bitmap_file(mddev, argp);
5864 goto done_unlock;
5866 case GET_DISK_INFO:
5867 err = get_disk_info(mddev, argp);
5868 goto done_unlock;
5870 case RESTART_ARRAY_RW:
5871 err = restart_array(mddev);
5872 goto done_unlock;
5874 case STOP_ARRAY:
5875 err = do_md_stop(mddev, 0, 1);
5876 goto done_unlock;
5878 case STOP_ARRAY_RO:
5879 err = md_set_readonly(mddev, 1);
5880 goto done_unlock;
5882 case BLKROSET:
5883 if (get_user(ro, (int __user *)(arg))) {
5884 err = -EFAULT;
5885 goto done_unlock;
5887 err = -EINVAL;
5889 /* if the bdev is going readonly the value of mddev->ro
5890 * does not matter, no writes are coming
5892 if (ro)
5893 goto done_unlock;
5895 /* are we are already prepared for writes? */
5896 if (mddev->ro != 1)
5897 goto done_unlock;
5899 /* transitioning to readauto need only happen for
5900 * arrays that call md_write_start
5902 if (mddev->pers) {
5903 err = restart_array(mddev);
5904 if (err == 0) {
5905 mddev->ro = 2;
5906 set_disk_ro(mddev->gendisk, 0);
5909 goto done_unlock;
5913 * The remaining ioctls are changing the state of the
5914 * superblock, so we do not allow them on read-only arrays.
5915 * However non-MD ioctls (e.g. get-size) will still come through
5916 * here and hit the 'default' below, so only disallow
5917 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5919 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5920 if (mddev->ro == 2) {
5921 mddev->ro = 0;
5922 sysfs_notify_dirent_safe(mddev->sysfs_state);
5923 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5924 md_wakeup_thread(mddev->thread);
5925 } else {
5926 err = -EROFS;
5927 goto abort_unlock;
5931 switch (cmd)
5933 case ADD_NEW_DISK:
5935 mdu_disk_info_t info;
5936 if (copy_from_user(&info, argp, sizeof(info)))
5937 err = -EFAULT;
5938 else
5939 err = add_new_disk(mddev, &info);
5940 goto done_unlock;
5943 case HOT_REMOVE_DISK:
5944 err = hot_remove_disk(mddev, new_decode_dev(arg));
5945 goto done_unlock;
5947 case HOT_ADD_DISK:
5948 err = hot_add_disk(mddev, new_decode_dev(arg));
5949 goto done_unlock;
5951 case SET_DISK_FAULTY:
5952 err = set_disk_faulty(mddev, new_decode_dev(arg));
5953 goto done_unlock;
5955 case RUN_ARRAY:
5956 err = do_md_run(mddev);
5957 goto done_unlock;
5959 case SET_BITMAP_FILE:
5960 err = set_bitmap_file(mddev, (int)arg);
5961 goto done_unlock;
5963 default:
5964 err = -EINVAL;
5965 goto abort_unlock;
5968 done_unlock:
5969 abort_unlock:
5970 if (mddev->hold_active == UNTIL_IOCTL &&
5971 err != -EINVAL)
5972 mddev->hold_active = 0;
5973 mddev_unlock(mddev);
5975 return err;
5976 done:
5977 if (err)
5978 MD_BUG();
5979 abort:
5980 return err;
5982 #ifdef CONFIG_COMPAT
5983 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5984 unsigned int cmd, unsigned long arg)
5986 switch (cmd) {
5987 case HOT_REMOVE_DISK:
5988 case HOT_ADD_DISK:
5989 case SET_DISK_FAULTY:
5990 case SET_BITMAP_FILE:
5991 /* These take in integer arg, do not convert */
5992 break;
5993 default:
5994 arg = (unsigned long)compat_ptr(arg);
5995 break;
5998 return md_ioctl(bdev, mode, cmd, arg);
6000 #endif /* CONFIG_COMPAT */
6002 static int md_open(struct block_device *bdev, fmode_t mode)
6005 * Succeed if we can lock the mddev, which confirms that
6006 * it isn't being stopped right now.
6008 mddev_t *mddev = mddev_find(bdev->bd_dev);
6009 int err;
6011 if (mddev->gendisk != bdev->bd_disk) {
6012 /* we are racing with mddev_put which is discarding this
6013 * bd_disk.
6015 mddev_put(mddev);
6016 /* Wait until bdev->bd_disk is definitely gone */
6017 flush_workqueue(md_misc_wq);
6018 /* Then retry the open from the top */
6019 return -ERESTARTSYS;
6021 BUG_ON(mddev != bdev->bd_disk->private_data);
6023 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6024 goto out;
6026 err = 0;
6027 atomic_inc(&mddev->openers);
6028 mutex_unlock(&mddev->open_mutex);
6030 check_disk_change(bdev);
6031 out:
6032 return err;
6035 static int md_release(struct gendisk *disk, fmode_t mode)
6037 mddev_t *mddev = disk->private_data;
6039 BUG_ON(!mddev);
6040 atomic_dec(&mddev->openers);
6041 mddev_put(mddev);
6043 return 0;
6046 static int md_media_changed(struct gendisk *disk)
6048 mddev_t *mddev = disk->private_data;
6050 return mddev->changed;
6053 static int md_revalidate(struct gendisk *disk)
6055 mddev_t *mddev = disk->private_data;
6057 mddev->changed = 0;
6058 return 0;
6060 static const struct block_device_operations md_fops =
6062 .owner = THIS_MODULE,
6063 .open = md_open,
6064 .release = md_release,
6065 .ioctl = md_ioctl,
6066 #ifdef CONFIG_COMPAT
6067 .compat_ioctl = md_compat_ioctl,
6068 #endif
6069 .getgeo = md_getgeo,
6070 .media_changed = md_media_changed,
6071 .revalidate_disk= md_revalidate,
6074 static int md_thread(void * arg)
6076 mdk_thread_t *thread = arg;
6079 * md_thread is a 'system-thread', it's priority should be very
6080 * high. We avoid resource deadlocks individually in each
6081 * raid personality. (RAID5 does preallocation) We also use RR and
6082 * the very same RT priority as kswapd, thus we will never get
6083 * into a priority inversion deadlock.
6085 * we definitely have to have equal or higher priority than
6086 * bdflush, otherwise bdflush will deadlock if there are too
6087 * many dirty RAID5 blocks.
6090 allow_signal(SIGKILL);
6091 while (!kthread_should_stop()) {
6093 /* We need to wait INTERRUPTIBLE so that
6094 * we don't add to the load-average.
6095 * That means we need to be sure no signals are
6096 * pending
6098 if (signal_pending(current))
6099 flush_signals(current);
6101 wait_event_interruptible_timeout
6102 (thread->wqueue,
6103 test_bit(THREAD_WAKEUP, &thread->flags)
6104 || kthread_should_stop(),
6105 thread->timeout);
6107 clear_bit(THREAD_WAKEUP, &thread->flags);
6108 if (!kthread_should_stop())
6109 thread->run(thread->mddev);
6112 return 0;
6115 void md_wakeup_thread(mdk_thread_t *thread)
6117 if (thread) {
6118 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6119 set_bit(THREAD_WAKEUP, &thread->flags);
6120 wake_up(&thread->wqueue);
6124 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6125 const char *name)
6127 mdk_thread_t *thread;
6129 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6130 if (!thread)
6131 return NULL;
6133 init_waitqueue_head(&thread->wqueue);
6135 thread->run = run;
6136 thread->mddev = mddev;
6137 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6138 thread->tsk = kthread_run(md_thread, thread,
6139 "%s_%s",
6140 mdname(thread->mddev),
6141 name ?: mddev->pers->name);
6142 if (IS_ERR(thread->tsk)) {
6143 kfree(thread);
6144 return NULL;
6146 return thread;
6149 void md_unregister_thread(mdk_thread_t *thread)
6151 if (!thread)
6152 return;
6153 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6155 kthread_stop(thread->tsk);
6156 kfree(thread);
6159 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6161 if (!mddev) {
6162 MD_BUG();
6163 return;
6166 if (!rdev || test_bit(Faulty, &rdev->flags))
6167 return;
6169 if (mddev->external)
6170 set_bit(Blocked, &rdev->flags);
6172 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6173 mdname(mddev),
6174 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6175 __builtin_return_address(0),__builtin_return_address(1),
6176 __builtin_return_address(2),__builtin_return_address(3));
6178 if (!mddev->pers)
6179 return;
6180 if (!mddev->pers->error_handler)
6181 return;
6182 mddev->pers->error_handler(mddev,rdev);
6183 if (mddev->degraded)
6184 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6185 sysfs_notify_dirent_safe(rdev->sysfs_state);
6186 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6187 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6188 md_wakeup_thread(mddev->thread);
6189 if (mddev->event_work.func)
6190 queue_work(md_misc_wq, &mddev->event_work);
6191 md_new_event_inintr(mddev);
6194 /* seq_file implementation /proc/mdstat */
6196 static void status_unused(struct seq_file *seq)
6198 int i = 0;
6199 mdk_rdev_t *rdev;
6201 seq_printf(seq, "unused devices: ");
6203 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6204 char b[BDEVNAME_SIZE];
6205 i++;
6206 seq_printf(seq, "%s ",
6207 bdevname(rdev->bdev,b));
6209 if (!i)
6210 seq_printf(seq, "<none>");
6212 seq_printf(seq, "\n");
6216 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6218 sector_t max_sectors, resync, res;
6219 unsigned long dt, db;
6220 sector_t rt;
6221 int scale;
6222 unsigned int per_milli;
6224 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6226 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6227 max_sectors = mddev->resync_max_sectors;
6228 else
6229 max_sectors = mddev->dev_sectors;
6232 * Should not happen.
6234 if (!max_sectors) {
6235 MD_BUG();
6236 return;
6238 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6239 * in a sector_t, and (max_sectors>>scale) will fit in a
6240 * u32, as those are the requirements for sector_div.
6241 * Thus 'scale' must be at least 10
6243 scale = 10;
6244 if (sizeof(sector_t) > sizeof(unsigned long)) {
6245 while ( max_sectors/2 > (1ULL<<(scale+32)))
6246 scale++;
6248 res = (resync>>scale)*1000;
6249 sector_div(res, (u32)((max_sectors>>scale)+1));
6251 per_milli = res;
6253 int i, x = per_milli/50, y = 20-x;
6254 seq_printf(seq, "[");
6255 for (i = 0; i < x; i++)
6256 seq_printf(seq, "=");
6257 seq_printf(seq, ">");
6258 for (i = 0; i < y; i++)
6259 seq_printf(seq, ".");
6260 seq_printf(seq, "] ");
6262 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6263 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6264 "reshape" :
6265 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6266 "check" :
6267 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6268 "resync" : "recovery"))),
6269 per_milli/10, per_milli % 10,
6270 (unsigned long long) resync/2,
6271 (unsigned long long) max_sectors/2);
6274 * dt: time from mark until now
6275 * db: blocks written from mark until now
6276 * rt: remaining time
6278 * rt is a sector_t, so could be 32bit or 64bit.
6279 * So we divide before multiply in case it is 32bit and close
6280 * to the limit.
6281 * We scale the divisor (db) by 32 to avoid losing precision
6282 * near the end of resync when the number of remaining sectors
6283 * is close to 'db'.
6284 * We then divide rt by 32 after multiplying by db to compensate.
6285 * The '+1' avoids division by zero if db is very small.
6287 dt = ((jiffies - mddev->resync_mark) / HZ);
6288 if (!dt) dt++;
6289 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6290 - mddev->resync_mark_cnt;
6292 rt = max_sectors - resync; /* number of remaining sectors */
6293 sector_div(rt, db/32+1);
6294 rt *= dt;
6295 rt >>= 5;
6297 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6298 ((unsigned long)rt % 60)/6);
6300 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6303 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6305 struct list_head *tmp;
6306 loff_t l = *pos;
6307 mddev_t *mddev;
6309 if (l >= 0x10000)
6310 return NULL;
6311 if (!l--)
6312 /* header */
6313 return (void*)1;
6315 spin_lock(&all_mddevs_lock);
6316 list_for_each(tmp,&all_mddevs)
6317 if (!l--) {
6318 mddev = list_entry(tmp, mddev_t, all_mddevs);
6319 mddev_get(mddev);
6320 spin_unlock(&all_mddevs_lock);
6321 return mddev;
6323 spin_unlock(&all_mddevs_lock);
6324 if (!l--)
6325 return (void*)2;/* tail */
6326 return NULL;
6329 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6331 struct list_head *tmp;
6332 mddev_t *next_mddev, *mddev = v;
6334 ++*pos;
6335 if (v == (void*)2)
6336 return NULL;
6338 spin_lock(&all_mddevs_lock);
6339 if (v == (void*)1)
6340 tmp = all_mddevs.next;
6341 else
6342 tmp = mddev->all_mddevs.next;
6343 if (tmp != &all_mddevs)
6344 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6345 else {
6346 next_mddev = (void*)2;
6347 *pos = 0x10000;
6349 spin_unlock(&all_mddevs_lock);
6351 if (v != (void*)1)
6352 mddev_put(mddev);
6353 return next_mddev;
6357 static void md_seq_stop(struct seq_file *seq, void *v)
6359 mddev_t *mddev = v;
6361 if (mddev && v != (void*)1 && v != (void*)2)
6362 mddev_put(mddev);
6365 struct mdstat_info {
6366 int event;
6369 static int md_seq_show(struct seq_file *seq, void *v)
6371 mddev_t *mddev = v;
6372 sector_t sectors;
6373 mdk_rdev_t *rdev;
6374 struct mdstat_info *mi = seq->private;
6375 struct bitmap *bitmap;
6377 if (v == (void*)1) {
6378 struct mdk_personality *pers;
6379 seq_printf(seq, "Personalities : ");
6380 spin_lock(&pers_lock);
6381 list_for_each_entry(pers, &pers_list, list)
6382 seq_printf(seq, "[%s] ", pers->name);
6384 spin_unlock(&pers_lock);
6385 seq_printf(seq, "\n");
6386 mi->event = atomic_read(&md_event_count);
6387 return 0;
6389 if (v == (void*)2) {
6390 status_unused(seq);
6391 return 0;
6394 if (mddev_lock(mddev) < 0)
6395 return -EINTR;
6397 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6398 seq_printf(seq, "%s : %sactive", mdname(mddev),
6399 mddev->pers ? "" : "in");
6400 if (mddev->pers) {
6401 if (mddev->ro==1)
6402 seq_printf(seq, " (read-only)");
6403 if (mddev->ro==2)
6404 seq_printf(seq, " (auto-read-only)");
6405 seq_printf(seq, " %s", mddev->pers->name);
6408 sectors = 0;
6409 list_for_each_entry(rdev, &mddev->disks, same_set) {
6410 char b[BDEVNAME_SIZE];
6411 seq_printf(seq, " %s[%d]",
6412 bdevname(rdev->bdev,b), rdev->desc_nr);
6413 if (test_bit(WriteMostly, &rdev->flags))
6414 seq_printf(seq, "(W)");
6415 if (test_bit(Faulty, &rdev->flags)) {
6416 seq_printf(seq, "(F)");
6417 continue;
6418 } else if (rdev->raid_disk < 0)
6419 seq_printf(seq, "(S)"); /* spare */
6420 sectors += rdev->sectors;
6423 if (!list_empty(&mddev->disks)) {
6424 if (mddev->pers)
6425 seq_printf(seq, "\n %llu blocks",
6426 (unsigned long long)
6427 mddev->array_sectors / 2);
6428 else
6429 seq_printf(seq, "\n %llu blocks",
6430 (unsigned long long)sectors / 2);
6432 if (mddev->persistent) {
6433 if (mddev->major_version != 0 ||
6434 mddev->minor_version != 90) {
6435 seq_printf(seq," super %d.%d",
6436 mddev->major_version,
6437 mddev->minor_version);
6439 } else if (mddev->external)
6440 seq_printf(seq, " super external:%s",
6441 mddev->metadata_type);
6442 else
6443 seq_printf(seq, " super non-persistent");
6445 if (mddev->pers) {
6446 mddev->pers->status(seq, mddev);
6447 seq_printf(seq, "\n ");
6448 if (mddev->pers->sync_request) {
6449 if (mddev->curr_resync > 2) {
6450 status_resync(seq, mddev);
6451 seq_printf(seq, "\n ");
6452 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6453 seq_printf(seq, "\tresync=DELAYED\n ");
6454 else if (mddev->recovery_cp < MaxSector)
6455 seq_printf(seq, "\tresync=PENDING\n ");
6457 } else
6458 seq_printf(seq, "\n ");
6460 if ((bitmap = mddev->bitmap)) {
6461 unsigned long chunk_kb;
6462 unsigned long flags;
6463 spin_lock_irqsave(&bitmap->lock, flags);
6464 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6465 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6466 "%lu%s chunk",
6467 bitmap->pages - bitmap->missing_pages,
6468 bitmap->pages,
6469 (bitmap->pages - bitmap->missing_pages)
6470 << (PAGE_SHIFT - 10),
6471 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6472 chunk_kb ? "KB" : "B");
6473 if (bitmap->file) {
6474 seq_printf(seq, ", file: ");
6475 seq_path(seq, &bitmap->file->f_path, " \t\n");
6478 seq_printf(seq, "\n");
6479 spin_unlock_irqrestore(&bitmap->lock, flags);
6482 seq_printf(seq, "\n");
6484 mddev_unlock(mddev);
6486 return 0;
6489 static const struct seq_operations md_seq_ops = {
6490 .start = md_seq_start,
6491 .next = md_seq_next,
6492 .stop = md_seq_stop,
6493 .show = md_seq_show,
6496 static int md_seq_open(struct inode *inode, struct file *file)
6498 int error;
6499 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6500 if (mi == NULL)
6501 return -ENOMEM;
6503 error = seq_open(file, &md_seq_ops);
6504 if (error)
6505 kfree(mi);
6506 else {
6507 struct seq_file *p = file->private_data;
6508 p->private = mi;
6509 mi->event = atomic_read(&md_event_count);
6511 return error;
6514 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6516 struct seq_file *m = filp->private_data;
6517 struct mdstat_info *mi = m->private;
6518 int mask;
6520 poll_wait(filp, &md_event_waiters, wait);
6522 /* always allow read */
6523 mask = POLLIN | POLLRDNORM;
6525 if (mi->event != atomic_read(&md_event_count))
6526 mask |= POLLERR | POLLPRI;
6527 return mask;
6530 static const struct file_operations md_seq_fops = {
6531 .owner = THIS_MODULE,
6532 .open = md_seq_open,
6533 .read = seq_read,
6534 .llseek = seq_lseek,
6535 .release = seq_release_private,
6536 .poll = mdstat_poll,
6539 int register_md_personality(struct mdk_personality *p)
6541 spin_lock(&pers_lock);
6542 list_add_tail(&p->list, &pers_list);
6543 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6544 spin_unlock(&pers_lock);
6545 return 0;
6548 int unregister_md_personality(struct mdk_personality *p)
6550 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6551 spin_lock(&pers_lock);
6552 list_del_init(&p->list);
6553 spin_unlock(&pers_lock);
6554 return 0;
6557 static int is_mddev_idle(mddev_t *mddev, int init)
6559 mdk_rdev_t * rdev;
6560 int idle;
6561 int curr_events;
6563 idle = 1;
6564 rcu_read_lock();
6565 rdev_for_each_rcu(rdev, mddev) {
6566 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6567 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6568 (int)part_stat_read(&disk->part0, sectors[1]) -
6569 atomic_read(&disk->sync_io);
6570 /* sync IO will cause sync_io to increase before the disk_stats
6571 * as sync_io is counted when a request starts, and
6572 * disk_stats is counted when it completes.
6573 * So resync activity will cause curr_events to be smaller than
6574 * when there was no such activity.
6575 * non-sync IO will cause disk_stat to increase without
6576 * increasing sync_io so curr_events will (eventually)
6577 * be larger than it was before. Once it becomes
6578 * substantially larger, the test below will cause
6579 * the array to appear non-idle, and resync will slow
6580 * down.
6581 * If there is a lot of outstanding resync activity when
6582 * we set last_event to curr_events, then all that activity
6583 * completing might cause the array to appear non-idle
6584 * and resync will be slowed down even though there might
6585 * not have been non-resync activity. This will only
6586 * happen once though. 'last_events' will soon reflect
6587 * the state where there is little or no outstanding
6588 * resync requests, and further resync activity will
6589 * always make curr_events less than last_events.
6592 if (init || curr_events - rdev->last_events > 64) {
6593 rdev->last_events = curr_events;
6594 idle = 0;
6597 rcu_read_unlock();
6598 return idle;
6601 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6603 /* another "blocks" (512byte) blocks have been synced */
6604 atomic_sub(blocks, &mddev->recovery_active);
6605 wake_up(&mddev->recovery_wait);
6606 if (!ok) {
6607 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6608 md_wakeup_thread(mddev->thread);
6609 // stop recovery, signal do_sync ....
6614 /* md_write_start(mddev, bi)
6615 * If we need to update some array metadata (e.g. 'active' flag
6616 * in superblock) before writing, schedule a superblock update
6617 * and wait for it to complete.
6619 void md_write_start(mddev_t *mddev, struct bio *bi)
6621 int did_change = 0;
6622 if (bio_data_dir(bi) != WRITE)
6623 return;
6625 BUG_ON(mddev->ro == 1);
6626 if (mddev->ro == 2) {
6627 /* need to switch to read/write */
6628 mddev->ro = 0;
6629 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6630 md_wakeup_thread(mddev->thread);
6631 md_wakeup_thread(mddev->sync_thread);
6632 did_change = 1;
6634 atomic_inc(&mddev->writes_pending);
6635 if (mddev->safemode == 1)
6636 mddev->safemode = 0;
6637 if (mddev->in_sync) {
6638 spin_lock_irq(&mddev->write_lock);
6639 if (mddev->in_sync) {
6640 mddev->in_sync = 0;
6641 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6642 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6643 md_wakeup_thread(mddev->thread);
6644 did_change = 1;
6646 spin_unlock_irq(&mddev->write_lock);
6648 if (did_change)
6649 sysfs_notify_dirent_safe(mddev->sysfs_state);
6650 wait_event(mddev->sb_wait,
6651 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6654 void md_write_end(mddev_t *mddev)
6656 if (atomic_dec_and_test(&mddev->writes_pending)) {
6657 if (mddev->safemode == 2)
6658 md_wakeup_thread(mddev->thread);
6659 else if (mddev->safemode_delay)
6660 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6664 /* md_allow_write(mddev)
6665 * Calling this ensures that the array is marked 'active' so that writes
6666 * may proceed without blocking. It is important to call this before
6667 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6668 * Must be called with mddev_lock held.
6670 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6671 * is dropped, so return -EAGAIN after notifying userspace.
6673 int md_allow_write(mddev_t *mddev)
6675 if (!mddev->pers)
6676 return 0;
6677 if (mddev->ro)
6678 return 0;
6679 if (!mddev->pers->sync_request)
6680 return 0;
6682 spin_lock_irq(&mddev->write_lock);
6683 if (mddev->in_sync) {
6684 mddev->in_sync = 0;
6685 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6686 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6687 if (mddev->safemode_delay &&
6688 mddev->safemode == 0)
6689 mddev->safemode = 1;
6690 spin_unlock_irq(&mddev->write_lock);
6691 md_update_sb(mddev, 0);
6692 sysfs_notify_dirent_safe(mddev->sysfs_state);
6693 } else
6694 spin_unlock_irq(&mddev->write_lock);
6696 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6697 return -EAGAIN;
6698 else
6699 return 0;
6701 EXPORT_SYMBOL_GPL(md_allow_write);
6703 #define SYNC_MARKS 10
6704 #define SYNC_MARK_STEP (3*HZ)
6705 void md_do_sync(mddev_t *mddev)
6707 mddev_t *mddev2;
6708 unsigned int currspeed = 0,
6709 window;
6710 sector_t max_sectors,j, io_sectors;
6711 unsigned long mark[SYNC_MARKS];
6712 sector_t mark_cnt[SYNC_MARKS];
6713 int last_mark,m;
6714 struct list_head *tmp;
6715 sector_t last_check;
6716 int skipped = 0;
6717 mdk_rdev_t *rdev;
6718 char *desc;
6720 /* just incase thread restarts... */
6721 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6722 return;
6723 if (mddev->ro) /* never try to sync a read-only array */
6724 return;
6726 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6727 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6728 desc = "data-check";
6729 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6730 desc = "requested-resync";
6731 else
6732 desc = "resync";
6733 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6734 desc = "reshape";
6735 else
6736 desc = "recovery";
6738 /* we overload curr_resync somewhat here.
6739 * 0 == not engaged in resync at all
6740 * 2 == checking that there is no conflict with another sync
6741 * 1 == like 2, but have yielded to allow conflicting resync to
6742 * commense
6743 * other == active in resync - this many blocks
6745 * Before starting a resync we must have set curr_resync to
6746 * 2, and then checked that every "conflicting" array has curr_resync
6747 * less than ours. When we find one that is the same or higher
6748 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6749 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6750 * This will mean we have to start checking from the beginning again.
6754 do {
6755 mddev->curr_resync = 2;
6757 try_again:
6758 if (kthread_should_stop())
6759 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6761 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6762 goto skip;
6763 for_each_mddev(mddev2, tmp) {
6764 if (mddev2 == mddev)
6765 continue;
6766 if (!mddev->parallel_resync
6767 && mddev2->curr_resync
6768 && match_mddev_units(mddev, mddev2)) {
6769 DEFINE_WAIT(wq);
6770 if (mddev < mddev2 && mddev->curr_resync == 2) {
6771 /* arbitrarily yield */
6772 mddev->curr_resync = 1;
6773 wake_up(&resync_wait);
6775 if (mddev > mddev2 && mddev->curr_resync == 1)
6776 /* no need to wait here, we can wait the next
6777 * time 'round when curr_resync == 2
6779 continue;
6780 /* We need to wait 'interruptible' so as not to
6781 * contribute to the load average, and not to
6782 * be caught by 'softlockup'
6784 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6785 if (!kthread_should_stop() &&
6786 mddev2->curr_resync >= mddev->curr_resync) {
6787 printk(KERN_INFO "md: delaying %s of %s"
6788 " until %s has finished (they"
6789 " share one or more physical units)\n",
6790 desc, mdname(mddev), mdname(mddev2));
6791 mddev_put(mddev2);
6792 if (signal_pending(current))
6793 flush_signals(current);
6794 schedule();
6795 finish_wait(&resync_wait, &wq);
6796 goto try_again;
6798 finish_wait(&resync_wait, &wq);
6801 } while (mddev->curr_resync < 2);
6803 j = 0;
6804 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6805 /* resync follows the size requested by the personality,
6806 * which defaults to physical size, but can be virtual size
6808 max_sectors = mddev->resync_max_sectors;
6809 mddev->resync_mismatches = 0;
6810 /* we don't use the checkpoint if there's a bitmap */
6811 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6812 j = mddev->resync_min;
6813 else if (!mddev->bitmap)
6814 j = mddev->recovery_cp;
6816 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6817 max_sectors = mddev->dev_sectors;
6818 else {
6819 /* recovery follows the physical size of devices */
6820 max_sectors = mddev->dev_sectors;
6821 j = MaxSector;
6822 rcu_read_lock();
6823 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6824 if (rdev->raid_disk >= 0 &&
6825 !test_bit(Faulty, &rdev->flags) &&
6826 !test_bit(In_sync, &rdev->flags) &&
6827 rdev->recovery_offset < j)
6828 j = rdev->recovery_offset;
6829 rcu_read_unlock();
6832 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6833 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6834 " %d KB/sec/disk.\n", speed_min(mddev));
6835 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6836 "(but not more than %d KB/sec) for %s.\n",
6837 speed_max(mddev), desc);
6839 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6841 io_sectors = 0;
6842 for (m = 0; m < SYNC_MARKS; m++) {
6843 mark[m] = jiffies;
6844 mark_cnt[m] = io_sectors;
6846 last_mark = 0;
6847 mddev->resync_mark = mark[last_mark];
6848 mddev->resync_mark_cnt = mark_cnt[last_mark];
6851 * Tune reconstruction:
6853 window = 32*(PAGE_SIZE/512);
6854 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6855 window/2,(unsigned long long) max_sectors/2);
6857 atomic_set(&mddev->recovery_active, 0);
6858 last_check = 0;
6860 if (j>2) {
6861 printk(KERN_INFO
6862 "md: resuming %s of %s from checkpoint.\n",
6863 desc, mdname(mddev));
6864 mddev->curr_resync = j;
6866 mddev->curr_resync_completed = j;
6868 while (j < max_sectors) {
6869 sector_t sectors;
6871 skipped = 0;
6873 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6874 ((mddev->curr_resync > mddev->curr_resync_completed &&
6875 (mddev->curr_resync - mddev->curr_resync_completed)
6876 > (max_sectors >> 4)) ||
6877 (j - mddev->curr_resync_completed)*2
6878 >= mddev->resync_max - mddev->curr_resync_completed
6879 )) {
6880 /* time to update curr_resync_completed */
6881 wait_event(mddev->recovery_wait,
6882 atomic_read(&mddev->recovery_active) == 0);
6883 mddev->curr_resync_completed = j;
6884 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6885 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6888 while (j >= mddev->resync_max && !kthread_should_stop()) {
6889 /* As this condition is controlled by user-space,
6890 * we can block indefinitely, so use '_interruptible'
6891 * to avoid triggering warnings.
6893 flush_signals(current); /* just in case */
6894 wait_event_interruptible(mddev->recovery_wait,
6895 mddev->resync_max > j
6896 || kthread_should_stop());
6899 if (kthread_should_stop())
6900 goto interrupted;
6902 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6903 currspeed < speed_min(mddev));
6904 if (sectors == 0) {
6905 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6906 goto out;
6909 if (!skipped) { /* actual IO requested */
6910 io_sectors += sectors;
6911 atomic_add(sectors, &mddev->recovery_active);
6914 j += sectors;
6915 if (j>1) mddev->curr_resync = j;
6916 mddev->curr_mark_cnt = io_sectors;
6917 if (last_check == 0)
6918 /* this is the earliers that rebuilt will be
6919 * visible in /proc/mdstat
6921 md_new_event(mddev);
6923 if (last_check + window > io_sectors || j == max_sectors)
6924 continue;
6926 last_check = io_sectors;
6928 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6929 break;
6931 repeat:
6932 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6933 /* step marks */
6934 int next = (last_mark+1) % SYNC_MARKS;
6936 mddev->resync_mark = mark[next];
6937 mddev->resync_mark_cnt = mark_cnt[next];
6938 mark[next] = jiffies;
6939 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6940 last_mark = next;
6944 if (kthread_should_stop())
6945 goto interrupted;
6949 * this loop exits only if either when we are slower than
6950 * the 'hard' speed limit, or the system was IO-idle for
6951 * a jiffy.
6952 * the system might be non-idle CPU-wise, but we only care
6953 * about not overloading the IO subsystem. (things like an
6954 * e2fsck being done on the RAID array should execute fast)
6956 cond_resched();
6958 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6959 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6961 if (currspeed > speed_min(mddev)) {
6962 if ((currspeed > speed_max(mddev)) ||
6963 !is_mddev_idle(mddev, 0)) {
6964 msleep(500);
6965 goto repeat;
6969 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6971 * this also signals 'finished resyncing' to md_stop
6973 out:
6974 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6976 /* tell personality that we are finished */
6977 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6979 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6980 mddev->curr_resync > 2) {
6981 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6982 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6983 if (mddev->curr_resync >= mddev->recovery_cp) {
6984 printk(KERN_INFO
6985 "md: checkpointing %s of %s.\n",
6986 desc, mdname(mddev));
6987 mddev->recovery_cp = mddev->curr_resync;
6989 } else
6990 mddev->recovery_cp = MaxSector;
6991 } else {
6992 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6993 mddev->curr_resync = MaxSector;
6994 rcu_read_lock();
6995 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6996 if (rdev->raid_disk >= 0 &&
6997 mddev->delta_disks >= 0 &&
6998 !test_bit(Faulty, &rdev->flags) &&
6999 !test_bit(In_sync, &rdev->flags) &&
7000 rdev->recovery_offset < mddev->curr_resync)
7001 rdev->recovery_offset = mddev->curr_resync;
7002 rcu_read_unlock();
7005 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7007 skip:
7008 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7009 /* We completed so min/max setting can be forgotten if used. */
7010 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7011 mddev->resync_min = 0;
7012 mddev->resync_max = MaxSector;
7013 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7014 mddev->resync_min = mddev->curr_resync_completed;
7015 mddev->curr_resync = 0;
7016 wake_up(&resync_wait);
7017 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7018 md_wakeup_thread(mddev->thread);
7019 return;
7021 interrupted:
7023 * got a signal, exit.
7025 printk(KERN_INFO
7026 "md: md_do_sync() got signal ... exiting\n");
7027 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7028 goto out;
7031 EXPORT_SYMBOL_GPL(md_do_sync);
7034 static int remove_and_add_spares(mddev_t *mddev)
7036 mdk_rdev_t *rdev;
7037 int spares = 0;
7039 mddev->curr_resync_completed = 0;
7041 list_for_each_entry(rdev, &mddev->disks, same_set)
7042 if (rdev->raid_disk >= 0 &&
7043 !test_bit(Blocked, &rdev->flags) &&
7044 (test_bit(Faulty, &rdev->flags) ||
7045 ! test_bit(In_sync, &rdev->flags)) &&
7046 atomic_read(&rdev->nr_pending)==0) {
7047 if (mddev->pers->hot_remove_disk(
7048 mddev, rdev->raid_disk)==0) {
7049 char nm[20];
7050 sprintf(nm,"rd%d", rdev->raid_disk);
7051 sysfs_remove_link(&mddev->kobj, nm);
7052 rdev->raid_disk = -1;
7056 if (mddev->degraded && !mddev->recovery_disabled) {
7057 list_for_each_entry(rdev, &mddev->disks, same_set) {
7058 if (rdev->raid_disk >= 0 &&
7059 !test_bit(In_sync, &rdev->flags) &&
7060 !test_bit(Blocked, &rdev->flags))
7061 spares++;
7062 if (rdev->raid_disk < 0
7063 && !test_bit(Faulty, &rdev->flags)) {
7064 rdev->recovery_offset = 0;
7065 if (mddev->pers->
7066 hot_add_disk(mddev, rdev) == 0) {
7067 char nm[20];
7068 sprintf(nm, "rd%d", rdev->raid_disk);
7069 if (sysfs_create_link(&mddev->kobj,
7070 &rdev->kobj, nm))
7071 /* failure here is OK */;
7072 spares++;
7073 md_new_event(mddev);
7074 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7075 } else
7076 break;
7080 return spares;
7083 static void reap_sync_thread(mddev_t *mddev)
7085 mdk_rdev_t *rdev;
7087 /* resync has finished, collect result */
7088 md_unregister_thread(mddev->sync_thread);
7089 mddev->sync_thread = NULL;
7090 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7091 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7092 /* success...*/
7093 /* activate any spares */
7094 if (mddev->pers->spare_active(mddev))
7095 sysfs_notify(&mddev->kobj, NULL,
7096 "degraded");
7098 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7099 mddev->pers->finish_reshape)
7100 mddev->pers->finish_reshape(mddev);
7101 md_update_sb(mddev, 1);
7103 /* if array is no-longer degraded, then any saved_raid_disk
7104 * information must be scrapped
7106 if (!mddev->degraded)
7107 list_for_each_entry(rdev, &mddev->disks, same_set)
7108 rdev->saved_raid_disk = -1;
7110 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7111 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7112 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7113 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7114 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7115 /* flag recovery needed just to double check */
7116 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7117 sysfs_notify_dirent_safe(mddev->sysfs_action);
7118 md_new_event(mddev);
7122 * This routine is regularly called by all per-raid-array threads to
7123 * deal with generic issues like resync and super-block update.
7124 * Raid personalities that don't have a thread (linear/raid0) do not
7125 * need this as they never do any recovery or update the superblock.
7127 * It does not do any resync itself, but rather "forks" off other threads
7128 * to do that as needed.
7129 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7130 * "->recovery" and create a thread at ->sync_thread.
7131 * When the thread finishes it sets MD_RECOVERY_DONE
7132 * and wakeups up this thread which will reap the thread and finish up.
7133 * This thread also removes any faulty devices (with nr_pending == 0).
7135 * The overall approach is:
7136 * 1/ if the superblock needs updating, update it.
7137 * 2/ If a recovery thread is running, don't do anything else.
7138 * 3/ If recovery has finished, clean up, possibly marking spares active.
7139 * 4/ If there are any faulty devices, remove them.
7140 * 5/ If array is degraded, try to add spares devices
7141 * 6/ If array has spares or is not in-sync, start a resync thread.
7143 void md_check_recovery(mddev_t *mddev)
7145 if (mddev->bitmap)
7146 bitmap_daemon_work(mddev);
7148 if (mddev->ro)
7149 return;
7151 if (signal_pending(current)) {
7152 if (mddev->pers->sync_request && !mddev->external) {
7153 printk(KERN_INFO "md: %s in immediate safe mode\n",
7154 mdname(mddev));
7155 mddev->safemode = 2;
7157 flush_signals(current);
7160 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7161 return;
7162 if ( ! (
7163 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7164 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7165 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7166 (mddev->external == 0 && mddev->safemode == 1) ||
7167 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7168 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7170 return;
7172 if (mddev_trylock(mddev)) {
7173 int spares = 0;
7175 if (mddev->ro) {
7176 /* Only thing we do on a ro array is remove
7177 * failed devices.
7179 mdk_rdev_t *rdev;
7180 list_for_each_entry(rdev, &mddev->disks, same_set)
7181 if (rdev->raid_disk >= 0 &&
7182 !test_bit(Blocked, &rdev->flags) &&
7183 test_bit(Faulty, &rdev->flags) &&
7184 atomic_read(&rdev->nr_pending)==0) {
7185 if (mddev->pers->hot_remove_disk(
7186 mddev, rdev->raid_disk)==0) {
7187 char nm[20];
7188 sprintf(nm,"rd%d", rdev->raid_disk);
7189 sysfs_remove_link(&mddev->kobj, nm);
7190 rdev->raid_disk = -1;
7193 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7194 goto unlock;
7197 if (!mddev->external) {
7198 int did_change = 0;
7199 spin_lock_irq(&mddev->write_lock);
7200 if (mddev->safemode &&
7201 !atomic_read(&mddev->writes_pending) &&
7202 !mddev->in_sync &&
7203 mddev->recovery_cp == MaxSector) {
7204 mddev->in_sync = 1;
7205 did_change = 1;
7206 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7208 if (mddev->safemode == 1)
7209 mddev->safemode = 0;
7210 spin_unlock_irq(&mddev->write_lock);
7211 if (did_change)
7212 sysfs_notify_dirent_safe(mddev->sysfs_state);
7215 if (mddev->flags)
7216 md_update_sb(mddev, 0);
7218 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7219 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7220 /* resync/recovery still happening */
7221 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7222 goto unlock;
7224 if (mddev->sync_thread) {
7225 reap_sync_thread(mddev);
7226 goto unlock;
7228 /* Set RUNNING before clearing NEEDED to avoid
7229 * any transients in the value of "sync_action".
7231 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7232 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7233 /* Clear some bits that don't mean anything, but
7234 * might be left set
7236 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7237 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7239 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7240 goto unlock;
7241 /* no recovery is running.
7242 * remove any failed drives, then
7243 * add spares if possible.
7244 * Spare are also removed and re-added, to allow
7245 * the personality to fail the re-add.
7248 if (mddev->reshape_position != MaxSector) {
7249 if (mddev->pers->check_reshape == NULL ||
7250 mddev->pers->check_reshape(mddev) != 0)
7251 /* Cannot proceed */
7252 goto unlock;
7253 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7254 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7255 } else if ((spares = remove_and_add_spares(mddev))) {
7256 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7257 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7258 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7259 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7260 } else if (mddev->recovery_cp < MaxSector) {
7261 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7262 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7263 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7264 /* nothing to be done ... */
7265 goto unlock;
7267 if (mddev->pers->sync_request) {
7268 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7269 /* We are adding a device or devices to an array
7270 * which has the bitmap stored on all devices.
7271 * So make sure all bitmap pages get written
7273 bitmap_write_all(mddev->bitmap);
7275 mddev->sync_thread = md_register_thread(md_do_sync,
7276 mddev,
7277 "resync");
7278 if (!mddev->sync_thread) {
7279 printk(KERN_ERR "%s: could not start resync"
7280 " thread...\n",
7281 mdname(mddev));
7282 /* leave the spares where they are, it shouldn't hurt */
7283 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7284 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7285 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7286 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7287 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7288 } else
7289 md_wakeup_thread(mddev->sync_thread);
7290 sysfs_notify_dirent_safe(mddev->sysfs_action);
7291 md_new_event(mddev);
7293 unlock:
7294 if (!mddev->sync_thread) {
7295 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7296 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7297 &mddev->recovery))
7298 if (mddev->sysfs_action)
7299 sysfs_notify_dirent_safe(mddev->sysfs_action);
7301 mddev_unlock(mddev);
7305 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7307 sysfs_notify_dirent_safe(rdev->sysfs_state);
7308 wait_event_timeout(rdev->blocked_wait,
7309 !test_bit(Blocked, &rdev->flags),
7310 msecs_to_jiffies(5000));
7311 rdev_dec_pending(rdev, mddev);
7313 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7315 static int md_notify_reboot(struct notifier_block *this,
7316 unsigned long code, void *x)
7318 struct list_head *tmp;
7319 mddev_t *mddev;
7321 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7323 printk(KERN_INFO "md: stopping all md devices.\n");
7325 for_each_mddev(mddev, tmp)
7326 if (mddev_trylock(mddev)) {
7327 /* Force a switch to readonly even array
7328 * appears to still be in use. Hence
7329 * the '100'.
7331 md_set_readonly(mddev, 100);
7332 mddev_unlock(mddev);
7335 * certain more exotic SCSI devices are known to be
7336 * volatile wrt too early system reboots. While the
7337 * right place to handle this issue is the given
7338 * driver, we do want to have a safe RAID driver ...
7340 mdelay(1000*1);
7342 return NOTIFY_DONE;
7345 static struct notifier_block md_notifier = {
7346 .notifier_call = md_notify_reboot,
7347 .next = NULL,
7348 .priority = INT_MAX, /* before any real devices */
7351 static void md_geninit(void)
7353 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7355 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7358 static int __init md_init(void)
7360 int ret = -ENOMEM;
7362 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
7363 if (!md_wq)
7364 goto err_wq;
7366 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7367 if (!md_misc_wq)
7368 goto err_misc_wq;
7370 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7371 goto err_md;
7373 if ((ret = register_blkdev(0, "mdp")) < 0)
7374 goto err_mdp;
7375 mdp_major = ret;
7377 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7378 md_probe, NULL, NULL);
7379 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7380 md_probe, NULL, NULL);
7382 register_reboot_notifier(&md_notifier);
7383 raid_table_header = register_sysctl_table(raid_root_table);
7385 md_geninit();
7386 return 0;
7388 err_mdp:
7389 unregister_blkdev(MD_MAJOR, "md");
7390 err_md:
7391 destroy_workqueue(md_misc_wq);
7392 err_misc_wq:
7393 destroy_workqueue(md_wq);
7394 err_wq:
7395 return ret;
7398 #ifndef MODULE
7401 * Searches all registered partitions for autorun RAID arrays
7402 * at boot time.
7405 static LIST_HEAD(all_detected_devices);
7406 struct detected_devices_node {
7407 struct list_head list;
7408 dev_t dev;
7411 void md_autodetect_dev(dev_t dev)
7413 struct detected_devices_node *node_detected_dev;
7415 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7416 if (node_detected_dev) {
7417 node_detected_dev->dev = dev;
7418 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7419 } else {
7420 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7421 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7426 static void autostart_arrays(int part)
7428 mdk_rdev_t *rdev;
7429 struct detected_devices_node *node_detected_dev;
7430 dev_t dev;
7431 int i_scanned, i_passed;
7433 i_scanned = 0;
7434 i_passed = 0;
7436 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7438 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7439 i_scanned++;
7440 node_detected_dev = list_entry(all_detected_devices.next,
7441 struct detected_devices_node, list);
7442 list_del(&node_detected_dev->list);
7443 dev = node_detected_dev->dev;
7444 kfree(node_detected_dev);
7445 rdev = md_import_device(dev,0, 90);
7446 if (IS_ERR(rdev))
7447 continue;
7449 if (test_bit(Faulty, &rdev->flags)) {
7450 MD_BUG();
7451 continue;
7453 set_bit(AutoDetected, &rdev->flags);
7454 list_add(&rdev->same_set, &pending_raid_disks);
7455 i_passed++;
7458 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7459 i_scanned, i_passed);
7461 autorun_devices(part);
7464 #endif /* !MODULE */
7466 static __exit void md_exit(void)
7468 mddev_t *mddev;
7469 struct list_head *tmp;
7471 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7472 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7474 unregister_blkdev(MD_MAJOR,"md");
7475 unregister_blkdev(mdp_major, "mdp");
7476 unregister_reboot_notifier(&md_notifier);
7477 unregister_sysctl_table(raid_table_header);
7478 remove_proc_entry("mdstat", NULL);
7479 for_each_mddev(mddev, tmp) {
7480 export_array(mddev);
7481 mddev->hold_active = 0;
7483 destroy_workqueue(md_misc_wq);
7484 destroy_workqueue(md_wq);
7487 subsys_initcall(md_init);
7488 module_exit(md_exit)
7490 static int get_ro(char *buffer, struct kernel_param *kp)
7492 return sprintf(buffer, "%d", start_readonly);
7494 static int set_ro(const char *val, struct kernel_param *kp)
7496 char *e;
7497 int num = simple_strtoul(val, &e, 10);
7498 if (*val && (*e == '\0' || *e == '\n')) {
7499 start_readonly = num;
7500 return 0;
7502 return -EINVAL;
7505 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7506 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7508 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7510 EXPORT_SYMBOL(register_md_personality);
7511 EXPORT_SYMBOL(unregister_md_personality);
7512 EXPORT_SYMBOL(md_error);
7513 EXPORT_SYMBOL(md_done_sync);
7514 EXPORT_SYMBOL(md_write_start);
7515 EXPORT_SYMBOL(md_write_end);
7516 EXPORT_SYMBOL(md_register_thread);
7517 EXPORT_SYMBOL(md_unregister_thread);
7518 EXPORT_SYMBOL(md_wakeup_thread);
7519 EXPORT_SYMBOL(md_check_recovery);
7520 MODULE_LICENSE("GPL");
7521 MODULE_DESCRIPTION("MD RAID framework");
7522 MODULE_ALIAS("md");
7523 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);