?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
md: check ->hot_remove_disk when removing disk
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blob010cbdc9506c361df070c3ff78d8392d8fb9c03c
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
52 #include <linux/slab.h>
53 #include "md.h"
54 #include "bitmap.h"
55
56 #define DEBUG 0
57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58
59
60 #ifndef MODULE
61 static void autostart_arrays(int part);
62 #endif
63
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
66
67 static void md_print_devices(void);
68
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70
71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
72
73 /*
74 * Default number of read corrections we'll attempt on an rdev
75 * before ejecting it from the array. We divide the read error
76 * count by 2 for every hour elapsed between read errors.
77 */
78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
79 /*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
90 */
91
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
95 {
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98 }
99
100 static inline int speed_max(mddev_t *mddev)
101 {
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104 }
105
106 static struct ctl_table_header *raid_table_header;
107
108 static ctl_table raid_table[] = {
109 {
110 .procname = "speed_limit_min",
111 .data = &sysctl_speed_limit_min,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = proc_dointvec,
115 },
116 {
117 .procname = "speed_limit_max",
118 .data = &sysctl_speed_limit_max,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
122 },
123 { }
124 };
125
126 static ctl_table raid_dir_table[] = {
127 {
128 .procname = "raid",
129 .maxlen = 0,
130 .mode = S_IRUGO|S_IXUGO,
131 .child = raid_table,
132 },
133 { }
134 };
135
136 static ctl_table raid_root_table[] = {
137 {
138 .procname = "dev",
139 .maxlen = 0,
140 .mode = 0555,
141 .child = raid_dir_table,
142 },
143 { }
144 };
145
146 static const struct block_device_operations md_fops;
147
148 static int start_readonly;
149
150 /*
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
154 * count increases.
155 *
156 * Events are:
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
159 */
160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
161 static atomic_t md_event_count;
162 void md_new_event(mddev_t *mddev)
163 {
164 atomic_inc(&md_event_count);
165 wake_up(&md_event_waiters);
166 }
167 EXPORT_SYMBOL_GPL(md_new_event);
168
169 /* Alternate version that can be called from interrupts
170 * when calling sysfs_notify isn't needed.
171 */
172 static void md_new_event_inintr(mddev_t *mddev)
173 {
174 atomic_inc(&md_event_count);
175 wake_up(&md_event_waiters);
176 }
177
178 /*
179 * Enables to iterate over all existing md arrays
180 * all_mddevs_lock protects this list.
181 */
182 static LIST_HEAD(all_mddevs);
183 static DEFINE_SPINLOCK(all_mddevs_lock);
184
185
186 /*
187 * iterates through all used mddevs in the system.
188 * We take care to grab the all_mddevs_lock whenever navigating
189 * the list, and to always hold a refcount when unlocked.
190 * Any code which breaks out of this loop while own
191 * a reference to the current mddev and must mddev_put it.
192 */
193 #define for_each_mddev(mddev,tmp) \
194 \
195 for (({ spin_lock(&all_mddevs_lock); \
196 tmp = all_mddevs.next; \
197 mddev = NULL;}); \
198 ({ if (tmp != &all_mddevs) \
199 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
200 spin_unlock(&all_mddevs_lock); \
201 if (mddev) mddev_put(mddev); \
202 mddev = list_entry(tmp, mddev_t, all_mddevs); \
203 tmp != &all_mddevs;}); \
204 ({ spin_lock(&all_mddevs_lock); \
205 tmp = tmp->next;}) \
206 )
207
208
209 /* Rather than calling directly into the personality make_request function,
210 * IO requests come here first so that we can check if the device is
211 * being suspended pending a reconfiguration.
212 * We hold a refcount over the call to ->make_request. By the time that
213 * call has finished, the bio has been linked into some internal structure
214 * and so is visible to ->quiesce(), so we don't need the refcount any more.
215 */
216 static int md_make_request(struct request_queue *q, struct bio *bio)
217 {
218 const int rw = bio_data_dir(bio);
219 mddev_t *mddev = q->queuedata;
220 int rv;
221 int cpu;
222
223 if (mddev == NULL || mddev->pers == NULL) {
224 bio_io_error(bio);
225 return 0;
226 }
227 rcu_read_lock();
228 if (mddev->suspended || mddev->barrier) {
229 DEFINE_WAIT(__wait);
230 for (;;) {
231 prepare_to_wait(&mddev->sb_wait, &__wait,
232 TASK_UNINTERRUPTIBLE);
233 if (!mddev->suspended && !mddev->barrier)
234 break;
235 rcu_read_unlock();
236 schedule();
237 rcu_read_lock();
238 }
239 finish_wait(&mddev->sb_wait, &__wait);
240 }
241 atomic_inc(&mddev->active_io);
242 rcu_read_unlock();
243
244 rv = mddev->pers->make_request(mddev, bio);
245
246 cpu = part_stat_lock();
247 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
248 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
249 bio_sectors(bio));
250 part_stat_unlock();
251
252 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
253 wake_up(&mddev->sb_wait);
254
255 return rv;
256 }
257
258 /* mddev_suspend makes sure no new requests are submitted
259 * to the device, and that any requests that have been submitted
260 * are completely handled.
261 * Once ->stop is called and completes, the module will be completely
262 * unused.
263 */
264 static void mddev_suspend(mddev_t *mddev)
265 {
266 BUG_ON(mddev->suspended);
267 mddev->suspended = 1;
268 synchronize_rcu();
269 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
270 mddev->pers->quiesce(mddev, 1);
271 }
272
273 static void mddev_resume(mddev_t *mddev)
274 {
275 mddev->suspended = 0;
276 wake_up(&mddev->sb_wait);
277 mddev->pers->quiesce(mddev, 0);
278 }
279
280 int mddev_congested(mddev_t *mddev, int bits)
281 {
282 if (mddev->barrier)
283 return 1;
284 return mddev->suspended;
285 }
286 EXPORT_SYMBOL(mddev_congested);
287
288 /*
289 * Generic barrier handling for md
290 */
291
292 #define POST_REQUEST_BARRIER ((void*)1)
293
294 static void md_end_barrier(struct bio *bio, int err)
295 {
296 mdk_rdev_t *rdev = bio->bi_private;
297 mddev_t *mddev = rdev->mddev;
298 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
299 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
300
301 rdev_dec_pending(rdev, mddev);
302
303 if (atomic_dec_and_test(&mddev->flush_pending)) {
304 if (mddev->barrier == POST_REQUEST_BARRIER) {
305 /* This was a post-request barrier */
306 mddev->barrier = NULL;
307 wake_up(&mddev->sb_wait);
308 } else
309 /* The pre-request barrier has finished */
310 schedule_work(&mddev->barrier_work);
311 }
312 bio_put(bio);
313 }
314
315 static void submit_barriers(mddev_t *mddev)
316 {
317 mdk_rdev_t *rdev;
318
319 rcu_read_lock();
320 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
321 if (rdev->raid_disk >= 0 &&
322 !test_bit(Faulty, &rdev->flags)) {
323 /* Take two references, one is dropped
324 * when request finishes, one after
325 * we reclaim rcu_read_lock
326 */
327 struct bio *bi;
328 atomic_inc(&rdev->nr_pending);
329 atomic_inc(&rdev->nr_pending);
330 rcu_read_unlock();
331 bi = bio_alloc(GFP_KERNEL, 0);
332 bi->bi_end_io = md_end_barrier;
333 bi->bi_private = rdev;
334 bi->bi_bdev = rdev->bdev;
335 atomic_inc(&mddev->flush_pending);
336 submit_bio(WRITE_BARRIER, bi);
337 rcu_read_lock();
338 rdev_dec_pending(rdev, mddev);
339 }
340 rcu_read_unlock();
341 }
342
343 static void md_submit_barrier(struct work_struct *ws)
344 {
345 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
346 struct bio *bio = mddev->barrier;
347
348 atomic_set(&mddev->flush_pending, 1);
349
350 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
351 bio_endio(bio, -EOPNOTSUPP);
352 else if (bio->bi_size == 0)
353 /* an empty barrier - all done */
354 bio_endio(bio, 0);
355 else {
356 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
357 if (mddev->pers->make_request(mddev, bio))
358 generic_make_request(bio);
359 mddev->barrier = POST_REQUEST_BARRIER;
360 submit_barriers(mddev);
361 }
362 if (atomic_dec_and_test(&mddev->flush_pending)) {
363 mddev->barrier = NULL;
364 wake_up(&mddev->sb_wait);
365 }
366 }
367
368 void md_barrier_request(mddev_t *mddev, struct bio *bio)
369 {
370 spin_lock_irq(&mddev->write_lock);
371 wait_event_lock_irq(mddev->sb_wait,
372 !mddev->barrier,
373 mddev->write_lock, /*nothing*/);
374 mddev->barrier = bio;
375 spin_unlock_irq(&mddev->write_lock);
376
377 atomic_set(&mddev->flush_pending, 1);
378 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
379
380 submit_barriers(mddev);
381
382 if (atomic_dec_and_test(&mddev->flush_pending))
383 schedule_work(&mddev->barrier_work);
384 }
385 EXPORT_SYMBOL(md_barrier_request);
386
387 static inline mddev_t *mddev_get(mddev_t *mddev)
388 {
389 atomic_inc(&mddev->active);
390 return mddev;
391 }
392
393 static void mddev_delayed_delete(struct work_struct *ws);
394
395 static void mddev_put(mddev_t *mddev)
396 {
397 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
398 return;
399 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
400 mddev->ctime == 0 && !mddev->hold_active) {
401 /* Array is not configured at all, and not held active,
402 * so destroy it */
403 list_del(&mddev->all_mddevs);
404 if (mddev->gendisk) {
405 /* we did a probe so need to clean up.
406 * Call schedule_work inside the spinlock
407 * so that flush_scheduled_work() after
408 * mddev_find will succeed in waiting for the
409 * work to be done.
410 */
411 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
412 schedule_work(&mddev->del_work);
413 } else
414 kfree(mddev);
415 }
416 spin_unlock(&all_mddevs_lock);
417 }
418
419 static void mddev_init(mddev_t *mddev)
420 {
421 mutex_init(&mddev->open_mutex);
422 mutex_init(&mddev->reconfig_mutex);
423 mutex_init(&mddev->bitmap_info.mutex);
424 INIT_LIST_HEAD(&mddev->disks);
425 INIT_LIST_HEAD(&mddev->all_mddevs);
426 init_timer(&mddev->safemode_timer);
427 atomic_set(&mddev->active, 1);
428 atomic_set(&mddev->openers, 0);
429 atomic_set(&mddev->active_io, 0);
430 spin_lock_init(&mddev->write_lock);
431 atomic_set(&mddev->flush_pending, 0);
432 init_waitqueue_head(&mddev->sb_wait);
433 init_waitqueue_head(&mddev->recovery_wait);
434 mddev->reshape_position = MaxSector;
435 mddev->resync_min = 0;
436 mddev->resync_max = MaxSector;
437 mddev->level = LEVEL_NONE;
438 }
439
440 static mddev_t * mddev_find(dev_t unit)
441 {
442 mddev_t *mddev, *new = NULL;
443
444 if (unit && MAJOR(unit) != MD_MAJOR)
445 unit &= ~((1<<MdpMinorShift)-1);
446
447 retry:
448 spin_lock(&all_mddevs_lock);
449
450 if (unit) {
451 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
452 if (mddev->unit == unit) {
453 mddev_get(mddev);
454 spin_unlock(&all_mddevs_lock);
455 kfree(new);
456 return mddev;
457 }
458
459 if (new) {
460 list_add(&new->all_mddevs, &all_mddevs);
461 spin_unlock(&all_mddevs_lock);
462 new->hold_active = UNTIL_IOCTL;
463 return new;
464 }
465 } else if (new) {
466 /* find an unused unit number */
467 static int next_minor = 512;
468 int start = next_minor;
469 int is_free = 0;
470 int dev = 0;
471 while (!is_free) {
472 dev = MKDEV(MD_MAJOR, next_minor);
473 next_minor++;
474 if (next_minor > MINORMASK)
475 next_minor = 0;
476 if (next_minor == start) {
477 /* Oh dear, all in use. */
478 spin_unlock(&all_mddevs_lock);
479 kfree(new);
480 return NULL;
481 }
482
483 is_free = 1;
484 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
485 if (mddev->unit == dev) {
486 is_free = 0;
487 break;
488 }
489 }
490 new->unit = dev;
491 new->md_minor = MINOR(dev);
492 new->hold_active = UNTIL_STOP;
493 list_add(&new->all_mddevs, &all_mddevs);
494 spin_unlock(&all_mddevs_lock);
495 return new;
496 }
497 spin_unlock(&all_mddevs_lock);
498
499 new = kzalloc(sizeof(*new), GFP_KERNEL);
500 if (!new)
501 return NULL;
502
503 new->unit = unit;
504 if (MAJOR(unit) == MD_MAJOR)
505 new->md_minor = MINOR(unit);
506 else
507 new->md_minor = MINOR(unit) >> MdpMinorShift;
508
509 mddev_init(new);
510
511 goto retry;
512 }
513
514 static inline int mddev_lock(mddev_t * mddev)
515 {
516 return mutex_lock_interruptible(&mddev->reconfig_mutex);
517 }
518
519 static inline int mddev_is_locked(mddev_t *mddev)
520 {
521 return mutex_is_locked(&mddev->reconfig_mutex);
522 }
523
524 static inline int mddev_trylock(mddev_t * mddev)
525 {
526 return mutex_trylock(&mddev->reconfig_mutex);
527 }
528
529 static struct attribute_group md_redundancy_group;
530
531 static void mddev_unlock(mddev_t * mddev)
532 {
533 if (mddev->to_remove) {
534 /* These cannot be removed under reconfig_mutex as
535 * an access to the files will try to take reconfig_mutex
536 * while holding the file unremovable, which leads to
537 * a deadlock.
538 * So hold set sysfs_active while the remove in happeing,
539 * and anything else which might set ->to_remove or my
540 * otherwise change the sysfs namespace will fail with
541 * -EBUSY if sysfs_active is still set.
542 * We set sysfs_active under reconfig_mutex and elsewhere
543 * test it under the same mutex to ensure its correct value
544 * is seen.
545 */
546 struct attribute_group *to_remove = mddev->to_remove;
547 mddev->to_remove = NULL;
548 mddev->sysfs_active = 1;
549 mutex_unlock(&mddev->reconfig_mutex);
550
551 if (to_remove != &md_redundancy_group)
552 sysfs_remove_group(&mddev->kobj, to_remove);
553 if (mddev->pers == NULL ||
554 mddev->pers->sync_request == NULL) {
555 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
556 if (mddev->sysfs_action)
557 sysfs_put(mddev->sysfs_action);
558 mddev->sysfs_action = NULL;
559 }
560 mddev->sysfs_active = 0;
561 } else
562 mutex_unlock(&mddev->reconfig_mutex);
563
564 md_wakeup_thread(mddev->thread);
565 }
566
567 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
568 {
569 mdk_rdev_t *rdev;
570
571 list_for_each_entry(rdev, &mddev->disks, same_set)
572 if (rdev->desc_nr == nr)
573 return rdev;
574
575 return NULL;
576 }
577
578 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
579 {
580 mdk_rdev_t *rdev;
581
582 list_for_each_entry(rdev, &mddev->disks, same_set)
583 if (rdev->bdev->bd_dev == dev)
584 return rdev;
585
586 return NULL;
587 }
588
589 static struct mdk_personality *find_pers(int level, char *clevel)
590 {
591 struct mdk_personality *pers;
592 list_for_each_entry(pers, &pers_list, list) {
593 if (level != LEVEL_NONE && pers->level == level)
594 return pers;
595 if (strcmp(pers->name, clevel)==0)
596 return pers;
597 }
598 return NULL;
599 }
600
601 /* return the offset of the super block in 512byte sectors */
602 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
603 {
604 sector_t num_sectors = bdev->bd_inode->i_size / 512;
605 return MD_NEW_SIZE_SECTORS(num_sectors);
606 }
607
608 static int alloc_disk_sb(mdk_rdev_t * rdev)
609 {
610 if (rdev->sb_page)
611 MD_BUG();
612
613 rdev->sb_page = alloc_page(GFP_KERNEL);
614 if (!rdev->sb_page) {
615 printk(KERN_ALERT "md: out of memory.\n");
616 return -ENOMEM;
617 }
618
619 return 0;
620 }
621
622 static void free_disk_sb(mdk_rdev_t * rdev)
623 {
624 if (rdev->sb_page) {
625 put_page(rdev->sb_page);
626 rdev->sb_loaded = 0;
627 rdev->sb_page = NULL;
628 rdev->sb_start = 0;
629 rdev->sectors = 0;
630 }
631 }
632
633
634 static void super_written(struct bio *bio, int error)
635 {
636 mdk_rdev_t *rdev = bio->bi_private;
637 mddev_t *mddev = rdev->mddev;
638
639 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
640 printk("md: super_written gets error=%d, uptodate=%d\n",
641 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
642 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
643 md_error(mddev, rdev);
644 }
645
646 if (atomic_dec_and_test(&mddev->pending_writes))
647 wake_up(&mddev->sb_wait);
648 bio_put(bio);
649 }
650
651 static void super_written_barrier(struct bio *bio, int error)
652 {
653 struct bio *bio2 = bio->bi_private;
654 mdk_rdev_t *rdev = bio2->bi_private;
655 mddev_t *mddev = rdev->mddev;
656
657 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
658 error == -EOPNOTSUPP) {
659 unsigned long flags;
660 /* barriers don't appear to be supported :-( */
661 set_bit(BarriersNotsupp, &rdev->flags);
662 mddev->barriers_work = 0;
663 spin_lock_irqsave(&mddev->write_lock, flags);
664 bio2->bi_next = mddev->biolist;
665 mddev->biolist = bio2;
666 spin_unlock_irqrestore(&mddev->write_lock, flags);
667 wake_up(&mddev->sb_wait);
668 bio_put(bio);
669 } else {
670 bio_put(bio2);
671 bio->bi_private = rdev;
672 super_written(bio, error);
673 }
674 }
675
676 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
677 sector_t sector, int size, struct page *page)
678 {
679 /* write first size bytes of page to sector of rdev
680 * Increment mddev->pending_writes before returning
681 * and decrement it on completion, waking up sb_wait
682 * if zero is reached.
683 * If an error occurred, call md_error
684 *
685 * As we might need to resubmit the request if BIO_RW_BARRIER
686 * causes ENOTSUPP, we allocate a spare bio...
687 */
688 struct bio *bio = bio_alloc(GFP_NOIO, 1);
689 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
690
691 bio->bi_bdev = rdev->bdev;
692 bio->bi_sector = sector;
693 bio_add_page(bio, page, size, 0);
694 bio->bi_private = rdev;
695 bio->bi_end_io = super_written;
696 bio->bi_rw = rw;
697
698 atomic_inc(&mddev->pending_writes);
699 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
700 struct bio *rbio;
701 rw |= (1<<BIO_RW_BARRIER);
702 rbio = bio_clone(bio, GFP_NOIO);
703 rbio->bi_private = bio;
704 rbio->bi_end_io = super_written_barrier;
705 submit_bio(rw, rbio);
706 } else
707 submit_bio(rw, bio);
708 }
709
710 void md_super_wait(mddev_t *mddev)
711 {
712 /* wait for all superblock writes that were scheduled to complete.
713 * if any had to be retried (due to BARRIER problems), retry them
714 */
715 DEFINE_WAIT(wq);
716 for(;;) {
717 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
718 if (atomic_read(&mddev->pending_writes)==0)
719 break;
720 while (mddev->biolist) {
721 struct bio *bio;
722 spin_lock_irq(&mddev->write_lock);
723 bio = mddev->biolist;
724 mddev->biolist = bio->bi_next ;
725 bio->bi_next = NULL;
726 spin_unlock_irq(&mddev->write_lock);
727 submit_bio(bio->bi_rw, bio);
728 }
729 schedule();
730 }
731 finish_wait(&mddev->sb_wait, &wq);
732 }
733
734 static void bi_complete(struct bio *bio, int error)
735 {
736 complete((struct completion*)bio->bi_private);
737 }
738
739 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
740 struct page *page, int rw)
741 {
742 struct bio *bio = bio_alloc(GFP_NOIO, 1);
743 struct completion event;
744 int ret;
745
746 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
747
748 bio->bi_bdev = bdev;
749 bio->bi_sector = sector;
750 bio_add_page(bio, page, size, 0);
751 init_completion(&event);
752 bio->bi_private = &event;
753 bio->bi_end_io = bi_complete;
754 submit_bio(rw, bio);
755 wait_for_completion(&event);
756
757 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
758 bio_put(bio);
759 return ret;
760 }
761 EXPORT_SYMBOL_GPL(sync_page_io);
762
763 static int read_disk_sb(mdk_rdev_t * rdev, int size)
764 {
765 char b[BDEVNAME_SIZE];
766 if (!rdev->sb_page) {
767 MD_BUG();
768 return -EINVAL;
769 }
770 if (rdev->sb_loaded)
771 return 0;
772
773
774 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
775 goto fail;
776 rdev->sb_loaded = 1;
777 return 0;
778
779 fail:
780 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
781 bdevname(rdev->bdev,b));
782 return -EINVAL;
783 }
784
785 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
786 {
787 return sb1->set_uuid0 == sb2->set_uuid0 &&
788 sb1->set_uuid1 == sb2->set_uuid1 &&
789 sb1->set_uuid2 == sb2->set_uuid2 &&
790 sb1->set_uuid3 == sb2->set_uuid3;
791 }
792
793 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
794 {
795 int ret;
796 mdp_super_t *tmp1, *tmp2;
797
798 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
799 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
800
801 if (!tmp1 || !tmp2) {
802 ret = 0;
803 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
804 goto abort;
805 }
806
807 *tmp1 = *sb1;
808 *tmp2 = *sb2;
809
810 /*
811 * nr_disks is not constant
812 */
813 tmp1->nr_disks = 0;
814 tmp2->nr_disks = 0;
815
816 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
817 abort:
818 kfree(tmp1);
819 kfree(tmp2);
820 return ret;
821 }
822
823
824 static u32 md_csum_fold(u32 csum)
825 {
826 csum = (csum & 0xffff) + (csum >> 16);
827 return (csum & 0xffff) + (csum >> 16);
828 }
829
830 static unsigned int calc_sb_csum(mdp_super_t * sb)
831 {
832 u64 newcsum = 0;
833 u32 *sb32 = (u32*)sb;
834 int i;
835 unsigned int disk_csum, csum;
836
837 disk_csum = sb->sb_csum;
838 sb->sb_csum = 0;
839
840 for (i = 0; i < MD_SB_BYTES/4 ; i++)
841 newcsum += sb32[i];
842 csum = (newcsum & 0xffffffff) + (newcsum>>32);
843
844
845 #ifdef CONFIG_ALPHA
846 /* This used to use csum_partial, which was wrong for several
847 * reasons including that different results are returned on
848 * different architectures. It isn't critical that we get exactly
849 * the same return value as before (we always csum_fold before
850 * testing, and that removes any differences). However as we
851 * know that csum_partial always returned a 16bit value on
852 * alphas, do a fold to maximise conformity to previous behaviour.
853 */
854 sb->sb_csum = md_csum_fold(disk_csum);
855 #else
856 sb->sb_csum = disk_csum;
857 #endif
858 return csum;
859 }
860
861
862 /*
863 * Handle superblock details.
864 * We want to be able to handle multiple superblock formats
865 * so we have a common interface to them all, and an array of
866 * different handlers.
867 * We rely on user-space to write the initial superblock, and support
868 * reading and updating of superblocks.
869 * Interface methods are:
870 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
871 * loads and validates a superblock on dev.
872 * if refdev != NULL, compare superblocks on both devices
873 * Return:
874 * 0 - dev has a superblock that is compatible with refdev
875 * 1 - dev has a superblock that is compatible and newer than refdev
876 * so dev should be used as the refdev in future
877 * -EINVAL superblock incompatible or invalid
878 * -othererror e.g. -EIO
879 *
880 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
881 * Verify that dev is acceptable into mddev.
882 * The first time, mddev->raid_disks will be 0, and data from
883 * dev should be merged in. Subsequent calls check that dev
884 * is new enough. Return 0 or -EINVAL
885 *
886 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
887 * Update the superblock for rdev with data in mddev
888 * This does not write to disc.
889 *
890 */
891
892 struct super_type {
893 char *name;
894 struct module *owner;
895 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
896 int minor_version);
897 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
898 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
899 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
900 sector_t num_sectors);
901 };
902
903 /*
904 * Check that the given mddev has no bitmap.
905 *
906 * This function is called from the run method of all personalities that do not
907 * support bitmaps. It prints an error message and returns non-zero if mddev
908 * has a bitmap. Otherwise, it returns 0.
909 *
910 */
911 int md_check_no_bitmap(mddev_t *mddev)
912 {
913 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
914 return 0;
915 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
916 mdname(mddev), mddev->pers->name);
917 return 1;
918 }
919 EXPORT_SYMBOL(md_check_no_bitmap);
920
921 /*
922 * load_super for 0.90.0
923 */
924 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
925 {
926 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
927 mdp_super_t *sb;
928 int ret;
929
930 /*
931 * Calculate the position of the superblock (512byte sectors),
932 * it's at the end of the disk.
933 *
934 * It also happens to be a multiple of 4Kb.
935 */
936 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
937
938 ret = read_disk_sb(rdev, MD_SB_BYTES);
939 if (ret) return ret;
940
941 ret = -EINVAL;
942
943 bdevname(rdev->bdev, b);
944 sb = (mdp_super_t*)page_address(rdev->sb_page);
945
946 if (sb->md_magic != MD_SB_MAGIC) {
947 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
948 b);
949 goto abort;
950 }
951
952 if (sb->major_version != 0 ||
953 sb->minor_version < 90 ||
954 sb->minor_version > 91) {
955 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
956 sb->major_version, sb->minor_version,
957 b);
958 goto abort;
959 }
960
961 if (sb->raid_disks <= 0)
962 goto abort;
963
964 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
965 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
966 b);
967 goto abort;
968 }
969
970 rdev->preferred_minor = sb->md_minor;
971 rdev->data_offset = 0;
972 rdev->sb_size = MD_SB_BYTES;
973
974 if (sb->level == LEVEL_MULTIPATH)
975 rdev->desc_nr = -1;
976 else
977 rdev->desc_nr = sb->this_disk.number;
978
979 if (!refdev) {
980 ret = 1;
981 } else {
982 __u64 ev1, ev2;
983 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
984 if (!uuid_equal(refsb, sb)) {
985 printk(KERN_WARNING "md: %s has different UUID to %s\n",
986 b, bdevname(refdev->bdev,b2));
987 goto abort;
988 }
989 if (!sb_equal(refsb, sb)) {
990 printk(KERN_WARNING "md: %s has same UUID"
991 " but different superblock to %s\n",
992 b, bdevname(refdev->bdev, b2));
993 goto abort;
994 }
995 ev1 = md_event(sb);
996 ev2 = md_event(refsb);
997 if (ev1 > ev2)
998 ret = 1;
999 else
1000 ret = 0;
1001 }
1002 rdev->sectors = rdev->sb_start;
1003
1004 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1005 /* "this cannot possibly happen" ... */
1006 ret = -EINVAL;
1007
1008 abort:
1009 return ret;
1010 }
1011
1012 /*
1013 * validate_super for 0.90.0
1014 */
1015 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1016 {
1017 mdp_disk_t *desc;
1018 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1019 __u64 ev1 = md_event(sb);
1020
1021 rdev->raid_disk = -1;
1022 clear_bit(Faulty, &rdev->flags);
1023 clear_bit(In_sync, &rdev->flags);
1024 clear_bit(WriteMostly, &rdev->flags);
1025 clear_bit(BarriersNotsupp, &rdev->flags);
1026
1027 if (mddev->raid_disks == 0) {
1028 mddev->major_version = 0;
1029 mddev->minor_version = sb->minor_version;
1030 mddev->patch_version = sb->patch_version;
1031 mddev->external = 0;
1032 mddev->chunk_sectors = sb->chunk_size >> 9;
1033 mddev->ctime = sb->ctime;
1034 mddev->utime = sb->utime;
1035 mddev->level = sb->level;
1036 mddev->clevel[0] = 0;
1037 mddev->layout = sb->layout;
1038 mddev->raid_disks = sb->raid_disks;
1039 mddev->dev_sectors = sb->size * 2;
1040 mddev->events = ev1;
1041 mddev->bitmap_info.offset = 0;
1042 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1043
1044 if (mddev->minor_version >= 91) {
1045 mddev->reshape_position = sb->reshape_position;
1046 mddev->delta_disks = sb->delta_disks;
1047 mddev->new_level = sb->new_level;
1048 mddev->new_layout = sb->new_layout;
1049 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1050 } else {
1051 mddev->reshape_position = MaxSector;
1052 mddev->delta_disks = 0;
1053 mddev->new_level = mddev->level;
1054 mddev->new_layout = mddev->layout;
1055 mddev->new_chunk_sectors = mddev->chunk_sectors;
1056 }
1057
1058 if (sb->state & (1<<MD_SB_CLEAN))
1059 mddev->recovery_cp = MaxSector;
1060 else {
1061 if (sb->events_hi == sb->cp_events_hi &&
1062 sb->events_lo == sb->cp_events_lo) {
1063 mddev->recovery_cp = sb->recovery_cp;
1064 } else
1065 mddev->recovery_cp = 0;
1066 }
1067
1068 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1069 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1070 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1071 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1072
1073 mddev->max_disks = MD_SB_DISKS;
1074
1075 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1076 mddev->bitmap_info.file == NULL)
1077 mddev->bitmap_info.offset =
1078 mddev->bitmap_info.default_offset;
1079
1080 } else if (mddev->pers == NULL) {
1081 /* Insist on good event counter while assembling, except
1082 * for spares (which don't need an event count) */
1083 ++ev1;
1084 if (sb->disks[rdev->desc_nr].state & (
1085 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1086 if (ev1 < mddev->events)
1087 return -EINVAL;
1088 } else if (mddev->bitmap) {
1089 /* if adding to array with a bitmap, then we can accept an
1090 * older device ... but not too old.
1091 */
1092 if (ev1 < mddev->bitmap->events_cleared)
1093 return 0;
1094 } else {
1095 if (ev1 < mddev->events)
1096 /* just a hot-add of a new device, leave raid_disk at -1 */
1097 return 0;
1098 }
1099
1100 if (mddev->level != LEVEL_MULTIPATH) {
1101 desc = sb->disks + rdev->desc_nr;
1102
1103 if (desc->state & (1<<MD_DISK_FAULTY))
1104 set_bit(Faulty, &rdev->flags);
1105 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1106 desc->raid_disk < mddev->raid_disks */) {
1107 set_bit(In_sync, &rdev->flags);
1108 rdev->raid_disk = desc->raid_disk;
1109 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1110 /* active but not in sync implies recovery up to
1111 * reshape position. We don't know exactly where
1112 * that is, so set to zero for now */
1113 if (mddev->minor_version >= 91) {
1114 rdev->recovery_offset = 0;
1115 rdev->raid_disk = desc->raid_disk;
1116 }
1117 }
1118 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1119 set_bit(WriteMostly, &rdev->flags);
1120 } else /* MULTIPATH are always insync */
1121 set_bit(In_sync, &rdev->flags);
1122 return 0;
1123 }
1124
1125 /*
1126 * sync_super for 0.90.0
1127 */
1128 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1129 {
1130 mdp_super_t *sb;
1131 mdk_rdev_t *rdev2;
1132 int next_spare = mddev->raid_disks;
1133
1134
1135 /* make rdev->sb match mddev data..
1136 *
1137 * 1/ zero out disks
1138 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1139 * 3/ any empty disks < next_spare become removed
1140 *
1141 * disks[0] gets initialised to REMOVED because
1142 * we cannot be sure from other fields if it has
1143 * been initialised or not.
1144 */
1145 int i;
1146 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1147
1148 rdev->sb_size = MD_SB_BYTES;
1149
1150 sb = (mdp_super_t*)page_address(rdev->sb_page);
1151
1152 memset(sb, 0, sizeof(*sb));
1153
1154 sb->md_magic = MD_SB_MAGIC;
1155 sb->major_version = mddev->major_version;
1156 sb->patch_version = mddev->patch_version;
1157 sb->gvalid_words = 0; /* ignored */
1158 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1159 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1160 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1161 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1162
1163 sb->ctime = mddev->ctime;
1164 sb->level = mddev->level;
1165 sb->size = mddev->dev_sectors / 2;
1166 sb->raid_disks = mddev->raid_disks;
1167 sb->md_minor = mddev->md_minor;
1168 sb->not_persistent = 0;
1169 sb->utime = mddev->utime;
1170 sb->state = 0;
1171 sb->events_hi = (mddev->events>>32);
1172 sb->events_lo = (u32)mddev->events;
1173
1174 if (mddev->reshape_position == MaxSector)
1175 sb->minor_version = 90;
1176 else {
1177 sb->minor_version = 91;
1178 sb->reshape_position = mddev->reshape_position;
1179 sb->new_level = mddev->new_level;
1180 sb->delta_disks = mddev->delta_disks;
1181 sb->new_layout = mddev->new_layout;
1182 sb->new_chunk = mddev->new_chunk_sectors << 9;
1183 }
1184 mddev->minor_version = sb->minor_version;
1185 if (mddev->in_sync)
1186 {
1187 sb->recovery_cp = mddev->recovery_cp;
1188 sb->cp_events_hi = (mddev->events>>32);
1189 sb->cp_events_lo = (u32)mddev->events;
1190 if (mddev->recovery_cp == MaxSector)
1191 sb->state = (1<< MD_SB_CLEAN);
1192 } else
1193 sb->recovery_cp = 0;
1194
1195 sb->layout = mddev->layout;
1196 sb->chunk_size = mddev->chunk_sectors << 9;
1197
1198 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1199 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1200
1201 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1202 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1203 mdp_disk_t *d;
1204 int desc_nr;
1205 int is_active = test_bit(In_sync, &rdev2->flags);
1206
1207 if (rdev2->raid_disk >= 0 &&
1208 sb->minor_version >= 91)
1209 /* we have nowhere to store the recovery_offset,
1210 * but if it is not below the reshape_position,
1211 * we can piggy-back on that.
1212 */
1213 is_active = 1;
1214 if (rdev2->raid_disk < 0 ||
1215 test_bit(Faulty, &rdev2->flags))
1216 is_active = 0;
1217 if (is_active)
1218 desc_nr = rdev2->raid_disk;
1219 else
1220 desc_nr = next_spare++;
1221 rdev2->desc_nr = desc_nr;
1222 d = &sb->disks[rdev2->desc_nr];
1223 nr_disks++;
1224 d->number = rdev2->desc_nr;
1225 d->major = MAJOR(rdev2->bdev->bd_dev);
1226 d->minor = MINOR(rdev2->bdev->bd_dev);
1227 if (is_active)
1228 d->raid_disk = rdev2->raid_disk;
1229 else
1230 d->raid_disk = rdev2->desc_nr; /* compatibility */
1231 if (test_bit(Faulty, &rdev2->flags))
1232 d->state = (1<<MD_DISK_FAULTY);
1233 else if (is_active) {
1234 d->state = (1<<MD_DISK_ACTIVE);
1235 if (test_bit(In_sync, &rdev2->flags))
1236 d->state |= (1<<MD_DISK_SYNC);
1237 active++;
1238 working++;
1239 } else {
1240 d->state = 0;
1241 spare++;
1242 working++;
1243 }
1244 if (test_bit(WriteMostly, &rdev2->flags))
1245 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1246 }
1247 /* now set the "removed" and "faulty" bits on any missing devices */
1248 for (i=0 ; i < mddev->raid_disks ; i++) {
1249 mdp_disk_t *d = &sb->disks[i];
1250 if (d->state == 0 && d->number == 0) {
1251 d->number = i;
1252 d->raid_disk = i;
1253 d->state = (1<<MD_DISK_REMOVED);
1254 d->state |= (1<<MD_DISK_FAULTY);
1255 failed++;
1256 }
1257 }
1258 sb->nr_disks = nr_disks;
1259 sb->active_disks = active;
1260 sb->working_disks = working;
1261 sb->failed_disks = failed;
1262 sb->spare_disks = spare;
1263
1264 sb->this_disk = sb->disks[rdev->desc_nr];
1265 sb->sb_csum = calc_sb_csum(sb);
1266 }
1267
1268 /*
1269 * rdev_size_change for 0.90.0
1270 */
1271 static unsigned long long
1272 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1273 {
1274 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1275 return 0; /* component must fit device */
1276 if (rdev->mddev->bitmap_info.offset)
1277 return 0; /* can't move bitmap */
1278 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1279 if (!num_sectors || num_sectors > rdev->sb_start)
1280 num_sectors = rdev->sb_start;
1281 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1282 rdev->sb_page);
1283 md_super_wait(rdev->mddev);
1284 return num_sectors;
1285 }
1286
1287
1288 /*
1289 * version 1 superblock
1290 */
1291
1292 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1293 {
1294 __le32 disk_csum;
1295 u32 csum;
1296 unsigned long long newcsum;
1297 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1298 __le32 *isuper = (__le32*)sb;
1299 int i;
1300
1301 disk_csum = sb->sb_csum;
1302 sb->sb_csum = 0;
1303 newcsum = 0;
1304 for (i=0; size>=4; size -= 4 )
1305 newcsum += le32_to_cpu(*isuper++);
1306
1307 if (size == 2)
1308 newcsum += le16_to_cpu(*(__le16*) isuper);
1309
1310 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1311 sb->sb_csum = disk_csum;
1312 return cpu_to_le32(csum);
1313 }
1314
1315 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1316 {
1317 struct mdp_superblock_1 *sb;
1318 int ret;
1319 sector_t sb_start;
1320 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1321 int bmask;
1322
1323 /*
1324 * Calculate the position of the superblock in 512byte sectors.
1325 * It is always aligned to a 4K boundary and
1326 * depeding on minor_version, it can be:
1327 * 0: At least 8K, but less than 12K, from end of device
1328 * 1: At start of device
1329 * 2: 4K from start of device.
1330 */
1331 switch(minor_version) {
1332 case 0:
1333 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1334 sb_start -= 8*2;
1335 sb_start &= ~(sector_t)(4*2-1);
1336 break;
1337 case 1:
1338 sb_start = 0;
1339 break;
1340 case 2:
1341 sb_start = 8;
1342 break;
1343 default:
1344 return -EINVAL;
1345 }
1346 rdev->sb_start = sb_start;
1347
1348 /* superblock is rarely larger than 1K, but it can be larger,
1349 * and it is safe to read 4k, so we do that
1350 */
1351 ret = read_disk_sb(rdev, 4096);
1352 if (ret) return ret;
1353
1354
1355 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1356
1357 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1358 sb->major_version != cpu_to_le32(1) ||
1359 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1360 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1361 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1362 return -EINVAL;
1363
1364 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1365 printk("md: invalid superblock checksum on %s\n",
1366 bdevname(rdev->bdev,b));
1367 return -EINVAL;
1368 }
1369 if (le64_to_cpu(sb->data_size) < 10) {
1370 printk("md: data_size too small on %s\n",
1371 bdevname(rdev->bdev,b));
1372 return -EINVAL;
1373 }
1374
1375 rdev->preferred_minor = 0xffff;
1376 rdev->data_offset = le64_to_cpu(sb->data_offset);
1377 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1378
1379 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1380 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1381 if (rdev->sb_size & bmask)
1382 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1383
1384 if (minor_version
1385 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1386 return -EINVAL;
1387
1388 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1389 rdev->desc_nr = -1;
1390 else
1391 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1392
1393 if (!refdev) {
1394 ret = 1;
1395 } else {
1396 __u64 ev1, ev2;
1397 struct mdp_superblock_1 *refsb =
1398 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1399
1400 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1401 sb->level != refsb->level ||
1402 sb->layout != refsb->layout ||
1403 sb->chunksize != refsb->chunksize) {
1404 printk(KERN_WARNING "md: %s has strangely different"
1405 " superblock to %s\n",
1406 bdevname(rdev->bdev,b),
1407 bdevname(refdev->bdev,b2));
1408 return -EINVAL;
1409 }
1410 ev1 = le64_to_cpu(sb->events);
1411 ev2 = le64_to_cpu(refsb->events);
1412
1413 if (ev1 > ev2)
1414 ret = 1;
1415 else
1416 ret = 0;
1417 }
1418 if (minor_version)
1419 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1420 le64_to_cpu(sb->data_offset);
1421 else
1422 rdev->sectors = rdev->sb_start;
1423 if (rdev->sectors < le64_to_cpu(sb->data_size))
1424 return -EINVAL;
1425 rdev->sectors = le64_to_cpu(sb->data_size);
1426 if (le64_to_cpu(sb->size) > rdev->sectors)
1427 return -EINVAL;
1428 return ret;
1429 }
1430
1431 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1432 {
1433 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1434 __u64 ev1 = le64_to_cpu(sb->events);
1435
1436 rdev->raid_disk = -1;
1437 clear_bit(Faulty, &rdev->flags);
1438 clear_bit(In_sync, &rdev->flags);
1439 clear_bit(WriteMostly, &rdev->flags);
1440 clear_bit(BarriersNotsupp, &rdev->flags);
1441
1442 if (mddev->raid_disks == 0) {
1443 mddev->major_version = 1;
1444 mddev->patch_version = 0;
1445 mddev->external = 0;
1446 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1447 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1448 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1449 mddev->level = le32_to_cpu(sb->level);
1450 mddev->clevel[0] = 0;
1451 mddev->layout = le32_to_cpu(sb->layout);
1452 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1453 mddev->dev_sectors = le64_to_cpu(sb->size);
1454 mddev->events = ev1;
1455 mddev->bitmap_info.offset = 0;
1456 mddev->bitmap_info.default_offset = 1024 >> 9;
1457
1458 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1459 memcpy(mddev->uuid, sb->set_uuid, 16);
1460
1461 mddev->max_disks = (4096-256)/2;
1462
1463 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1464 mddev->bitmap_info.file == NULL )
1465 mddev->bitmap_info.offset =
1466 (__s32)le32_to_cpu(sb->bitmap_offset);
1467
1468 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1469 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1470 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1471 mddev->new_level = le32_to_cpu(sb->new_level);
1472 mddev->new_layout = le32_to_cpu(sb->new_layout);
1473 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1474 } else {
1475 mddev->reshape_position = MaxSector;
1476 mddev->delta_disks = 0;
1477 mddev->new_level = mddev->level;
1478 mddev->new_layout = mddev->layout;
1479 mddev->new_chunk_sectors = mddev->chunk_sectors;
1480 }
1481
1482 } else if (mddev->pers == NULL) {
1483 /* Insist of good event counter while assembling, except for
1484 * spares (which don't need an event count) */
1485 ++ev1;
1486 if (rdev->desc_nr >= 0 &&
1487 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1488 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1489 if (ev1 < mddev->events)
1490 return -EINVAL;
1491 } else if (mddev->bitmap) {
1492 /* If adding to array with a bitmap, then we can accept an
1493 * older device, but not too old.
1494 */
1495 if (ev1 < mddev->bitmap->events_cleared)
1496 return 0;
1497 } else {
1498 if (ev1 < mddev->events)
1499 /* just a hot-add of a new device, leave raid_disk at -1 */
1500 return 0;
1501 }
1502 if (mddev->level != LEVEL_MULTIPATH) {
1503 int role;
1504 if (rdev->desc_nr < 0 ||
1505 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1506 role = 0xffff;
1507 rdev->desc_nr = -1;
1508 } else
1509 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1510 switch(role) {
1511 case 0xffff: /* spare */
1512 break;
1513 case 0xfffe: /* faulty */
1514 set_bit(Faulty, &rdev->flags);
1515 break;
1516 default:
1517 if ((le32_to_cpu(sb->feature_map) &
1518 MD_FEATURE_RECOVERY_OFFSET))
1519 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1520 else
1521 set_bit(In_sync, &rdev->flags);
1522 rdev->raid_disk = role;
1523 break;
1524 }
1525 if (sb->devflags & WriteMostly1)
1526 set_bit(WriteMostly, &rdev->flags);
1527 } else /* MULTIPATH are always insync */
1528 set_bit(In_sync, &rdev->flags);
1529
1530 return 0;
1531 }
1532
1533 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1534 {
1535 struct mdp_superblock_1 *sb;
1536 mdk_rdev_t *rdev2;
1537 int max_dev, i;
1538 /* make rdev->sb match mddev and rdev data. */
1539
1540 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1541
1542 sb->feature_map = 0;
1543 sb->pad0 = 0;
1544 sb->recovery_offset = cpu_to_le64(0);
1545 memset(sb->pad1, 0, sizeof(sb->pad1));
1546 memset(sb->pad2, 0, sizeof(sb->pad2));
1547 memset(sb->pad3, 0, sizeof(sb->pad3));
1548
1549 sb->utime = cpu_to_le64((__u64)mddev->utime);
1550 sb->events = cpu_to_le64(mddev->events);
1551 if (mddev->in_sync)
1552 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1553 else
1554 sb->resync_offset = cpu_to_le64(0);
1555
1556 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1557
1558 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1559 sb->size = cpu_to_le64(mddev->dev_sectors);
1560 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1561 sb->level = cpu_to_le32(mddev->level);
1562 sb->layout = cpu_to_le32(mddev->layout);
1563
1564 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1565 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1566 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1567 }
1568
1569 if (rdev->raid_disk >= 0 &&
1570 !test_bit(In_sync, &rdev->flags)) {
1571 sb->feature_map |=
1572 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1573 sb->recovery_offset =
1574 cpu_to_le64(rdev->recovery_offset);
1575 }
1576
1577 if (mddev->reshape_position != MaxSector) {
1578 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1579 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1580 sb->new_layout = cpu_to_le32(mddev->new_layout);
1581 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1582 sb->new_level = cpu_to_le32(mddev->new_level);
1583 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1584 }
1585
1586 max_dev = 0;
1587 list_for_each_entry(rdev2, &mddev->disks, same_set)
1588 if (rdev2->desc_nr+1 > max_dev)
1589 max_dev = rdev2->desc_nr+1;
1590
1591 if (max_dev > le32_to_cpu(sb->max_dev)) {
1592 int bmask;
1593 sb->max_dev = cpu_to_le32(max_dev);
1594 rdev->sb_size = max_dev * 2 + 256;
1595 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1596 if (rdev->sb_size & bmask)
1597 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1598 }
1599 for (i=0; i<max_dev;i++)
1600 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1601
1602 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1603 i = rdev2->desc_nr;
1604 if (test_bit(Faulty, &rdev2->flags))
1605 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1606 else if (test_bit(In_sync, &rdev2->flags))
1607 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1608 else if (rdev2->raid_disk >= 0)
1609 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1610 else
1611 sb->dev_roles[i] = cpu_to_le16(0xffff);
1612 }
1613
1614 sb->sb_csum = calc_sb_1_csum(sb);
1615 }
1616
1617 static unsigned long long
1618 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1619 {
1620 struct mdp_superblock_1 *sb;
1621 sector_t max_sectors;
1622 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1623 return 0; /* component must fit device */
1624 if (rdev->sb_start < rdev->data_offset) {
1625 /* minor versions 1 and 2; superblock before data */
1626 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1627 max_sectors -= rdev->data_offset;
1628 if (!num_sectors || num_sectors > max_sectors)
1629 num_sectors = max_sectors;
1630 } else if (rdev->mddev->bitmap_info.offset) {
1631 /* minor version 0 with bitmap we can't move */
1632 return 0;
1633 } else {
1634 /* minor version 0; superblock after data */
1635 sector_t sb_start;
1636 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1637 sb_start &= ~(sector_t)(4*2 - 1);
1638 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1639 if (!num_sectors || num_sectors > max_sectors)
1640 num_sectors = max_sectors;
1641 rdev->sb_start = sb_start;
1642 }
1643 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1644 sb->data_size = cpu_to_le64(num_sectors);
1645 sb->super_offset = rdev->sb_start;
1646 sb->sb_csum = calc_sb_1_csum(sb);
1647 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1648 rdev->sb_page);
1649 md_super_wait(rdev->mddev);
1650 return num_sectors;
1651 }
1652
1653 static struct super_type super_types[] = {
1654 [0] = {
1655 .name = "0.90.0",
1656 .owner = THIS_MODULE,
1657 .load_super = super_90_load,
1658 .validate_super = super_90_validate,
1659 .sync_super = super_90_sync,
1660 .rdev_size_change = super_90_rdev_size_change,
1661 },
1662 [1] = {
1663 .name = "md-1",
1664 .owner = THIS_MODULE,
1665 .load_super = super_1_load,
1666 .validate_super = super_1_validate,
1667 .sync_super = super_1_sync,
1668 .rdev_size_change = super_1_rdev_size_change,
1669 },
1670 };
1671
1672 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1673 {
1674 mdk_rdev_t *rdev, *rdev2;
1675
1676 rcu_read_lock();
1677 rdev_for_each_rcu(rdev, mddev1)
1678 rdev_for_each_rcu(rdev2, mddev2)
1679 if (rdev->bdev->bd_contains ==
1680 rdev2->bdev->bd_contains) {
1681 rcu_read_unlock();
1682 return 1;
1683 }
1684 rcu_read_unlock();
1685 return 0;
1686 }
1687
1688 static LIST_HEAD(pending_raid_disks);
1689
1690 /*
1691 * Try to register data integrity profile for an mddev
1692 *
1693 * This is called when an array is started and after a disk has been kicked
1694 * from the array. It only succeeds if all working and active component devices
1695 * are integrity capable with matching profiles.
1696 */
1697 int md_integrity_register(mddev_t *mddev)
1698 {
1699 mdk_rdev_t *rdev, *reference = NULL;
1700
1701 if (list_empty(&mddev->disks))
1702 return 0; /* nothing to do */
1703 if (blk_get_integrity(mddev->gendisk))
1704 return 0; /* already registered */
1705 list_for_each_entry(rdev, &mddev->disks, same_set) {
1706 /* skip spares and non-functional disks */
1707 if (test_bit(Faulty, &rdev->flags))
1708 continue;
1709 if (rdev->raid_disk < 0)
1710 continue;
1711 /*
1712 * If at least one rdev is not integrity capable, we can not
1713 * enable data integrity for the md device.
1714 */
1715 if (!bdev_get_integrity(rdev->bdev))
1716 return -EINVAL;
1717 if (!reference) {
1718 /* Use the first rdev as the reference */
1719 reference = rdev;
1720 continue;
1721 }
1722 /* does this rdev's profile match the reference profile? */
1723 if (blk_integrity_compare(reference->bdev->bd_disk,
1724 rdev->bdev->bd_disk) < 0)
1725 return -EINVAL;
1726 }
1727 /*
1728 * All component devices are integrity capable and have matching
1729 * profiles, register the common profile for the md device.
1730 */
1731 if (blk_integrity_register(mddev->gendisk,
1732 bdev_get_integrity(reference->bdev)) != 0) {
1733 printk(KERN_ERR "md: failed to register integrity for %s\n",
1734 mdname(mddev));
1735 return -EINVAL;
1736 }
1737 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1738 mdname(mddev));
1739 return 0;
1740 }
1741 EXPORT_SYMBOL(md_integrity_register);
1742
1743 /* Disable data integrity if non-capable/non-matching disk is being added */
1744 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1745 {
1746 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1747 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1748
1749 if (!bi_mddev) /* nothing to do */
1750 return;
1751 if (rdev->raid_disk < 0) /* skip spares */
1752 return;
1753 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1754 rdev->bdev->bd_disk) >= 0)
1755 return;
1756 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1757 blk_integrity_unregister(mddev->gendisk);
1758 }
1759 EXPORT_SYMBOL(md_integrity_add_rdev);
1760
1761 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1762 {
1763 char b[BDEVNAME_SIZE];
1764 struct kobject *ko;
1765 char *s;
1766 int err;
1767
1768 if (rdev->mddev) {
1769 MD_BUG();
1770 return -EINVAL;
1771 }
1772
1773 /* prevent duplicates */
1774 if (find_rdev(mddev, rdev->bdev->bd_dev))
1775 return -EEXIST;
1776
1777 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1778 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1779 rdev->sectors < mddev->dev_sectors)) {
1780 if (mddev->pers) {
1781 /* Cannot change size, so fail
1782 * If mddev->level <= 0, then we don't care
1783 * about aligning sizes (e.g. linear)
1784 */
1785 if (mddev->level > 0)
1786 return -ENOSPC;
1787 } else
1788 mddev->dev_sectors = rdev->sectors;
1789 }
1790
1791 /* Verify rdev->desc_nr is unique.
1792 * If it is -1, assign a free number, else
1793 * check number is not in use
1794 */
1795 if (rdev->desc_nr < 0) {
1796 int choice = 0;
1797 if (mddev->pers) choice = mddev->raid_disks;
1798 while (find_rdev_nr(mddev, choice))
1799 choice++;
1800 rdev->desc_nr = choice;
1801 } else {
1802 if (find_rdev_nr(mddev, rdev->desc_nr))
1803 return -EBUSY;
1804 }
1805 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1806 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1807 mdname(mddev), mddev->max_disks);
1808 return -EBUSY;
1809 }
1810 bdevname(rdev->bdev,b);
1811 while ( (s=strchr(b, '/')) != NULL)
1812 *s = '!';
1813
1814 rdev->mddev = mddev;
1815 printk(KERN_INFO "md: bind<%s>\n", b);
1816
1817 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1818 goto fail;
1819
1820 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1821 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1822 kobject_del(&rdev->kobj);
1823 goto fail;
1824 }
1825 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, NULL, "state");
1826
1827 list_add_rcu(&rdev->same_set, &mddev->disks);
1828 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1829
1830 /* May as well allow recovery to be retried once */
1831 mddev->recovery_disabled = 0;
1832
1833 return 0;
1834
1835 fail:
1836 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1837 b, mdname(mddev));
1838 return err;
1839 }
1840
1841 static void md_delayed_delete(struct work_struct *ws)
1842 {
1843 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1844 kobject_del(&rdev->kobj);
1845 kobject_put(&rdev->kobj);
1846 }
1847
1848 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1849 {
1850 char b[BDEVNAME_SIZE];
1851 if (!rdev->mddev) {
1852 MD_BUG();
1853 return;
1854 }
1855 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1856 list_del_rcu(&rdev->same_set);
1857 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1858 rdev->mddev = NULL;
1859 sysfs_remove_link(&rdev->kobj, "block");
1860 sysfs_put(rdev->sysfs_state);
1861 rdev->sysfs_state = NULL;
1862 /* We need to delay this, otherwise we can deadlock when
1863 * writing to 'remove' to "dev/state". We also need
1864 * to delay it due to rcu usage.
1865 */
1866 synchronize_rcu();
1867 INIT_WORK(&rdev->del_work, md_delayed_delete);
1868 kobject_get(&rdev->kobj);
1869 schedule_work(&rdev->del_work);
1870 }
1871
1872 /*
1873 * prevent the device from being mounted, repartitioned or
1874 * otherwise reused by a RAID array (or any other kernel
1875 * subsystem), by bd_claiming the device.
1876 */
1877 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1878 {
1879 int err = 0;
1880 struct block_device *bdev;
1881 char b[BDEVNAME_SIZE];
1882
1883 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1884 if (IS_ERR(bdev)) {
1885 printk(KERN_ERR "md: could not open %s.\n",
1886 __bdevname(dev, b));
1887 return PTR_ERR(bdev);
1888 }
1889 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1890 if (err) {
1891 printk(KERN_ERR "md: could not bd_claim %s.\n",
1892 bdevname(bdev, b));
1893 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1894 return err;
1895 }
1896 if (!shared)
1897 set_bit(AllReserved, &rdev->flags);
1898 rdev->bdev = bdev;
1899 return err;
1900 }
1901
1902 static void unlock_rdev(mdk_rdev_t *rdev)
1903 {
1904 struct block_device *bdev = rdev->bdev;
1905 rdev->bdev = NULL;
1906 if (!bdev)
1907 MD_BUG();
1908 bd_release(bdev);
1909 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1910 }
1911
1912 void md_autodetect_dev(dev_t dev);
1913
1914 static void export_rdev(mdk_rdev_t * rdev)
1915 {
1916 char b[BDEVNAME_SIZE];
1917 printk(KERN_INFO "md: export_rdev(%s)\n",
1918 bdevname(rdev->bdev,b));
1919 if (rdev->mddev)
1920 MD_BUG();
1921 free_disk_sb(rdev);
1922 #ifndef MODULE
1923 if (test_bit(AutoDetected, &rdev->flags))
1924 md_autodetect_dev(rdev->bdev->bd_dev);
1925 #endif
1926 unlock_rdev(rdev);
1927 kobject_put(&rdev->kobj);
1928 }
1929
1930 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1931 {
1932 unbind_rdev_from_array(rdev);
1933 export_rdev(rdev);
1934 }
1935
1936 static void export_array(mddev_t *mddev)
1937 {
1938 mdk_rdev_t *rdev, *tmp;
1939
1940 rdev_for_each(rdev, tmp, mddev) {
1941 if (!rdev->mddev) {
1942 MD_BUG();
1943 continue;
1944 }
1945 kick_rdev_from_array(rdev);
1946 }
1947 if (!list_empty(&mddev->disks))
1948 MD_BUG();
1949 mddev->raid_disks = 0;
1950 mddev->major_version = 0;
1951 }
1952
1953 static void print_desc(mdp_disk_t *desc)
1954 {
1955 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1956 desc->major,desc->minor,desc->raid_disk,desc->state);
1957 }
1958
1959 static void print_sb_90(mdp_super_t *sb)
1960 {
1961 int i;
1962
1963 printk(KERN_INFO
1964 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1965 sb->major_version, sb->minor_version, sb->patch_version,
1966 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1967 sb->ctime);
1968 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1969 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1970 sb->md_minor, sb->layout, sb->chunk_size);
1971 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1972 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1973 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1974 sb->failed_disks, sb->spare_disks,
1975 sb->sb_csum, (unsigned long)sb->events_lo);
1976
1977 printk(KERN_INFO);
1978 for (i = 0; i < MD_SB_DISKS; i++) {
1979 mdp_disk_t *desc;
1980
1981 desc = sb->disks + i;
1982 if (desc->number || desc->major || desc->minor ||
1983 desc->raid_disk || (desc->state && (desc->state != 4))) {
1984 printk(" D %2d: ", i);
1985 print_desc(desc);
1986 }
1987 }
1988 printk(KERN_INFO "md: THIS: ");
1989 print_desc(&sb->this_disk);
1990 }
1991
1992 static void print_sb_1(struct mdp_superblock_1 *sb)
1993 {
1994 __u8 *uuid;
1995
1996 uuid = sb->set_uuid;
1997 printk(KERN_INFO
1998 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1999 "md: Name: \"%s\" CT:%llu\n",
2000 le32_to_cpu(sb->major_version),
2001 le32_to_cpu(sb->feature_map),
2002 uuid,
2003 sb->set_name,
2004 (unsigned long long)le64_to_cpu(sb->ctime)
2005 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2006
2007 uuid = sb->device_uuid;
2008 printk(KERN_INFO
2009 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2010 " RO:%llu\n"
2011 "md: Dev:%08x UUID: %pU\n"
2012 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2013 "md: (MaxDev:%u) \n",
2014 le32_to_cpu(sb->level),
2015 (unsigned long long)le64_to_cpu(sb->size),
2016 le32_to_cpu(sb->raid_disks),
2017 le32_to_cpu(sb->layout),
2018 le32_to_cpu(sb->chunksize),
2019 (unsigned long long)le64_to_cpu(sb->data_offset),
2020 (unsigned long long)le64_to_cpu(sb->data_size),
2021 (unsigned long long)le64_to_cpu(sb->super_offset),
2022 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2023 le32_to_cpu(sb->dev_number),
2024 uuid,
2025 sb->devflags,
2026 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2027 (unsigned long long)le64_to_cpu(sb->events),
2028 (unsigned long long)le64_to_cpu(sb->resync_offset),
2029 le32_to_cpu(sb->sb_csum),
2030 le32_to_cpu(sb->max_dev)
2031 );
2032 }
2033
2034 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2035 {
2036 char b[BDEVNAME_SIZE];
2037 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2038 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2039 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2040 rdev->desc_nr);
2041 if (rdev->sb_loaded) {
2042 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2043 switch (major_version) {
2044 case 0:
2045 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2046 break;
2047 case 1:
2048 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2049 break;
2050 }
2051 } else
2052 printk(KERN_INFO "md: no rdev superblock!\n");
2053 }
2054
2055 static void md_print_devices(void)
2056 {
2057 struct list_head *tmp;
2058 mdk_rdev_t *rdev;
2059 mddev_t *mddev;
2060 char b[BDEVNAME_SIZE];
2061
2062 printk("\n");
2063 printk("md: **********************************\n");
2064 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2065 printk("md: **********************************\n");
2066 for_each_mddev(mddev, tmp) {
2067
2068 if (mddev->bitmap)
2069 bitmap_print_sb(mddev->bitmap);
2070 else
2071 printk("%s: ", mdname(mddev));
2072 list_for_each_entry(rdev, &mddev->disks, same_set)
2073 printk("<%s>", bdevname(rdev->bdev,b));
2074 printk("\n");
2075
2076 list_for_each_entry(rdev, &mddev->disks, same_set)
2077 print_rdev(rdev, mddev->major_version);
2078 }
2079 printk("md: **********************************\n");
2080 printk("\n");
2081 }
2082
2083
2084 static void sync_sbs(mddev_t * mddev, int nospares)
2085 {
2086 /* Update each superblock (in-memory image), but
2087 * if we are allowed to, skip spares which already
2088 * have the right event counter, or have one earlier
2089 * (which would mean they aren't being marked as dirty
2090 * with the rest of the array)
2091 */
2092 mdk_rdev_t *rdev;
2093
2094 /* First make sure individual recovery_offsets are correct */
2095 list_for_each_entry(rdev, &mddev->disks, same_set) {
2096 if (rdev->raid_disk >= 0 &&
2097 mddev->delta_disks >= 0 &&
2098 !test_bit(In_sync, &rdev->flags) &&
2099 mddev->curr_resync_completed > rdev->recovery_offset)
2100 rdev->recovery_offset = mddev->curr_resync_completed;
2101
2102 }
2103 list_for_each_entry(rdev, &mddev->disks, same_set) {
2104 if (rdev->sb_events == mddev->events ||
2105 (nospares &&
2106 rdev->raid_disk < 0 &&
2107 rdev->sb_events+1 == mddev->events)) {
2108 /* Don't update this superblock */
2109 rdev->sb_loaded = 2;
2110 } else {
2111 super_types[mddev->major_version].
2112 sync_super(mddev, rdev);
2113 rdev->sb_loaded = 1;
2114 }
2115 }
2116 }
2117
2118 static void md_update_sb(mddev_t * mddev, int force_change)
2119 {
2120 mdk_rdev_t *rdev;
2121 int sync_req;
2122 int nospares = 0;
2123
2124 mddev->utime = get_seconds();
2125 if (mddev->external)
2126 return;
2127 repeat:
2128 spin_lock_irq(&mddev->write_lock);
2129
2130 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2131 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2132 force_change = 1;
2133 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2134 /* just a clean<-> dirty transition, possibly leave spares alone,
2135 * though if events isn't the right even/odd, we will have to do
2136 * spares after all
2137 */
2138 nospares = 1;
2139 if (force_change)
2140 nospares = 0;
2141 if (mddev->degraded)
2142 /* If the array is degraded, then skipping spares is both
2143 * dangerous and fairly pointless.
2144 * Dangerous because a device that was removed from the array
2145 * might have a event_count that still looks up-to-date,
2146 * so it can be re-added without a resync.
2147 * Pointless because if there are any spares to skip,
2148 * then a recovery will happen and soon that array won't
2149 * be degraded any more and the spare can go back to sleep then.
2150 */
2151 nospares = 0;
2152
2153 sync_req = mddev->in_sync;
2154
2155 /* If this is just a dirty<->clean transition, and the array is clean
2156 * and 'events' is odd, we can roll back to the previous clean state */
2157 if (nospares
2158 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2159 && mddev->can_decrease_events
2160 && mddev->events != 1) {
2161 mddev->events--;
2162 mddev->can_decrease_events = 0;
2163 } else {
2164 /* otherwise we have to go forward and ... */
2165 mddev->events ++;
2166 mddev->can_decrease_events = nospares;
2167 }
2168
2169 if (!mddev->events) {
2170 /*
2171 * oops, this 64-bit counter should never wrap.
2172 * Either we are in around ~1 trillion A.C., assuming
2173 * 1 reboot per second, or we have a bug:
2174 */
2175 MD_BUG();
2176 mddev->events --;
2177 }
2178
2179 /*
2180 * do not write anything to disk if using
2181 * nonpersistent superblocks
2182 */
2183 if (!mddev->persistent) {
2184 if (!mddev->external)
2185 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2186
2187 spin_unlock_irq(&mddev->write_lock);
2188 wake_up(&mddev->sb_wait);
2189 return;
2190 }
2191 sync_sbs(mddev, nospares);
2192 spin_unlock_irq(&mddev->write_lock);
2193
2194 dprintk(KERN_INFO
2195 "md: updating %s RAID superblock on device (in sync %d)\n",
2196 mdname(mddev),mddev->in_sync);
2197
2198 bitmap_update_sb(mddev->bitmap);
2199 list_for_each_entry(rdev, &mddev->disks, same_set) {
2200 char b[BDEVNAME_SIZE];
2201 dprintk(KERN_INFO "md: ");
2202 if (rdev->sb_loaded != 1)
2203 continue; /* no noise on spare devices */
2204 if (test_bit(Faulty, &rdev->flags))
2205 dprintk("(skipping faulty ");
2206
2207 dprintk("%s ", bdevname(rdev->bdev,b));
2208 if (!test_bit(Faulty, &rdev->flags)) {
2209 md_super_write(mddev,rdev,
2210 rdev->sb_start, rdev->sb_size,
2211 rdev->sb_page);
2212 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2213 bdevname(rdev->bdev,b),
2214 (unsigned long long)rdev->sb_start);
2215 rdev->sb_events = mddev->events;
2216
2217 } else
2218 dprintk(")\n");
2219 if (mddev->level == LEVEL_MULTIPATH)
2220 /* only need to write one superblock... */
2221 break;
2222 }
2223 md_super_wait(mddev);
2224 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2225
2226 spin_lock_irq(&mddev->write_lock);
2227 if (mddev->in_sync != sync_req ||
2228 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2229 /* have to write it out again */
2230 spin_unlock_irq(&mddev->write_lock);
2231 goto repeat;
2232 }
2233 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2234 spin_unlock_irq(&mddev->write_lock);
2235 wake_up(&mddev->sb_wait);
2236 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2237 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2238
2239 }
2240
2241 /* words written to sysfs files may, or may not, be \n terminated.
2242 * We want to accept with case. For this we use cmd_match.
2243 */
2244 static int cmd_match(const char *cmd, const char *str)
2245 {
2246 /* See if cmd, written into a sysfs file, matches
2247 * str. They must either be the same, or cmd can
2248 * have a trailing newline
2249 */
2250 while (*cmd && *str && *cmd == *str) {
2251 cmd++;
2252 str++;
2253 }
2254 if (*cmd == '\n')
2255 cmd++;
2256 if (*str || *cmd)
2257 return 0;
2258 return 1;
2259 }
2260
2261 struct rdev_sysfs_entry {
2262 struct attribute attr;
2263 ssize_t (*show)(mdk_rdev_t *, char *);
2264 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2265 };
2266
2267 static ssize_t
2268 state_show(mdk_rdev_t *rdev, char *page)
2269 {
2270 char *sep = "";
2271 size_t len = 0;
2272
2273 if (test_bit(Faulty, &rdev->flags)) {
2274 len+= sprintf(page+len, "%sfaulty",sep);
2275 sep = ",";
2276 }
2277 if (test_bit(In_sync, &rdev->flags)) {
2278 len += sprintf(page+len, "%sin_sync",sep);
2279 sep = ",";
2280 }
2281 if (test_bit(WriteMostly, &rdev->flags)) {
2282 len += sprintf(page+len, "%swrite_mostly",sep);
2283 sep = ",";
2284 }
2285 if (test_bit(Blocked, &rdev->flags)) {
2286 len += sprintf(page+len, "%sblocked", sep);
2287 sep = ",";
2288 }
2289 if (!test_bit(Faulty, &rdev->flags) &&
2290 !test_bit(In_sync, &rdev->flags)) {
2291 len += sprintf(page+len, "%sspare", sep);
2292 sep = ",";
2293 }
2294 return len+sprintf(page+len, "\n");
2295 }
2296
2297 static ssize_t
2298 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2299 {
2300 /* can write
2301 * faulty - simulates and error
2302 * remove - disconnects the device
2303 * writemostly - sets write_mostly
2304 * -writemostly - clears write_mostly
2305 * blocked - sets the Blocked flag
2306 * -blocked - clears the Blocked flag
2307 * insync - sets Insync providing device isn't active
2308 */
2309 int err = -EINVAL;
2310 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2311 md_error(rdev->mddev, rdev);
2312 err = 0;
2313 } else if (cmd_match(buf, "remove")) {
2314 if (rdev->raid_disk >= 0)
2315 err = -EBUSY;
2316 else {
2317 mddev_t *mddev = rdev->mddev;
2318 kick_rdev_from_array(rdev);
2319 if (mddev->pers)
2320 md_update_sb(mddev, 1);
2321 md_new_event(mddev);
2322 err = 0;
2323 }
2324 } else if (cmd_match(buf, "writemostly")) {
2325 set_bit(WriteMostly, &rdev->flags);
2326 err = 0;
2327 } else if (cmd_match(buf, "-writemostly")) {
2328 clear_bit(WriteMostly, &rdev->flags);
2329 err = 0;
2330 } else if (cmd_match(buf, "blocked")) {
2331 set_bit(Blocked, &rdev->flags);
2332 err = 0;
2333 } else if (cmd_match(buf, "-blocked")) {
2334 clear_bit(Blocked, &rdev->flags);
2335 wake_up(&rdev->blocked_wait);
2336 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2337 md_wakeup_thread(rdev->mddev->thread);
2338
2339 err = 0;
2340 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2341 set_bit(In_sync, &rdev->flags);
2342 err = 0;
2343 }
2344 if (!err && rdev->sysfs_state)
2345 sysfs_notify_dirent(rdev->sysfs_state);
2346 return err ? err : len;
2347 }
2348 static struct rdev_sysfs_entry rdev_state =
2349 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2350
2351 static ssize_t
2352 errors_show(mdk_rdev_t *rdev, char *page)
2353 {
2354 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2355 }
2356
2357 static ssize_t
2358 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2359 {
2360 char *e;
2361 unsigned long n = simple_strtoul(buf, &e, 10);
2362 if (*buf && (*e == 0 || *e == '\n')) {
2363 atomic_set(&rdev->corrected_errors, n);
2364 return len;
2365 }
2366 return -EINVAL;
2367 }
2368 static struct rdev_sysfs_entry rdev_errors =
2369 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2370
2371 static ssize_t
2372 slot_show(mdk_rdev_t *rdev, char *page)
2373 {
2374 if (rdev->raid_disk < 0)
2375 return sprintf(page, "none\n");
2376 else
2377 return sprintf(page, "%d\n", rdev->raid_disk);
2378 }
2379
2380 static ssize_t
2381 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2382 {
2383 char *e;
2384 int err;
2385 char nm[20];
2386 int slot = simple_strtoul(buf, &e, 10);
2387 if (strncmp(buf, "none", 4)==0)
2388 slot = -1;
2389 else if (e==buf || (*e && *e!= '\n'))
2390 return -EINVAL;
2391 if (rdev->mddev->pers && slot == -1) {
2392 /* Setting 'slot' on an active array requires also
2393 * updating the 'rd%d' link, and communicating
2394 * with the personality with ->hot_*_disk.
2395 * For now we only support removing
2396 * failed/spare devices. This normally happens automatically,
2397 * but not when the metadata is externally managed.
2398 */
2399 if (rdev->raid_disk == -1)
2400 return -EEXIST;
2401 /* personality does all needed checks */
2402 if (rdev->mddev->pers->hot_remove_disk == NULL)
2403 return -EINVAL;
2404 err = rdev->mddev->pers->
2405 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2406 if (err)
2407 return err;
2408 sprintf(nm, "rd%d", rdev->raid_disk);
2409 sysfs_remove_link(&rdev->mddev->kobj, nm);
2410 rdev->raid_disk = -1;
2411 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2412 md_wakeup_thread(rdev->mddev->thread);
2413 } else if (rdev->mddev->pers) {
2414 mdk_rdev_t *rdev2;
2415 /* Activating a spare .. or possibly reactivating
2416 * if we ever get bitmaps working here.
2417 */
2418
2419 if (rdev->raid_disk != -1)
2420 return -EBUSY;
2421
2422 if (rdev->mddev->pers->hot_add_disk == NULL)
2423 return -EINVAL;
2424
2425 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2426 if (rdev2->raid_disk == slot)
2427 return -EEXIST;
2428
2429 rdev->raid_disk = slot;
2430 if (test_bit(In_sync, &rdev->flags))
2431 rdev->saved_raid_disk = slot;
2432 else
2433 rdev->saved_raid_disk = -1;
2434 err = rdev->mddev->pers->
2435 hot_add_disk(rdev->mddev, rdev);
2436 if (err) {
2437 rdev->raid_disk = -1;
2438 return err;
2439 } else
2440 sysfs_notify_dirent(rdev->sysfs_state);
2441 sprintf(nm, "rd%d", rdev->raid_disk);
2442 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2443 printk(KERN_WARNING
2444 "md: cannot register "
2445 "%s for %s\n",
2446 nm, mdname(rdev->mddev));
2447
2448 /* don't wakeup anyone, leave that to userspace. */
2449 } else {
2450 if (slot >= rdev->mddev->raid_disks)
2451 return -ENOSPC;
2452 rdev->raid_disk = slot;
2453 /* assume it is working */
2454 clear_bit(Faulty, &rdev->flags);
2455 clear_bit(WriteMostly, &rdev->flags);
2456 set_bit(In_sync, &rdev->flags);
2457 sysfs_notify_dirent(rdev->sysfs_state);
2458 }
2459 return len;
2460 }
2461
2462
2463 static struct rdev_sysfs_entry rdev_slot =
2464 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2465
2466 static ssize_t
2467 offset_show(mdk_rdev_t *rdev, char *page)
2468 {
2469 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2470 }
2471
2472 static ssize_t
2473 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2474 {
2475 char *e;
2476 unsigned long long offset = simple_strtoull(buf, &e, 10);
2477 if (e==buf || (*e && *e != '\n'))
2478 return -EINVAL;
2479 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2480 return -EBUSY;
2481 if (rdev->sectors && rdev->mddev->external)
2482 /* Must set offset before size, so overlap checks
2483 * can be sane */
2484 return -EBUSY;
2485 rdev->data_offset = offset;
2486 return len;
2487 }
2488
2489 static struct rdev_sysfs_entry rdev_offset =
2490 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2491
2492 static ssize_t
2493 rdev_size_show(mdk_rdev_t *rdev, char *page)
2494 {
2495 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2496 }
2497
2498 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2499 {
2500 /* check if two start/length pairs overlap */
2501 if (s1+l1 <= s2)
2502 return 0;
2503 if (s2+l2 <= s1)
2504 return 0;
2505 return 1;
2506 }
2507
2508 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2509 {
2510 unsigned long long blocks;
2511 sector_t new;
2512
2513 if (strict_strtoull(buf, 10, &blocks) < 0)
2514 return -EINVAL;
2515
2516 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2517 return -EINVAL; /* sector conversion overflow */
2518
2519 new = blocks * 2;
2520 if (new != blocks * 2)
2521 return -EINVAL; /* unsigned long long to sector_t overflow */
2522
2523 *sectors = new;
2524 return 0;
2525 }
2526
2527 static ssize_t
2528 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2529 {
2530 mddev_t *my_mddev = rdev->mddev;
2531 sector_t oldsectors = rdev->sectors;
2532 sector_t sectors;
2533
2534 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2535 return -EINVAL;
2536 if (my_mddev->pers && rdev->raid_disk >= 0) {
2537 if (my_mddev->persistent) {
2538 sectors = super_types[my_mddev->major_version].
2539 rdev_size_change(rdev, sectors);
2540 if (!sectors)
2541 return -EBUSY;
2542 } else if (!sectors)
2543 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2544 rdev->data_offset;
2545 }
2546 if (sectors < my_mddev->dev_sectors)
2547 return -EINVAL; /* component must fit device */
2548
2549 rdev->sectors = sectors;
2550 if (sectors > oldsectors && my_mddev->external) {
2551 /* need to check that all other rdevs with the same ->bdev
2552 * do not overlap. We need to unlock the mddev to avoid
2553 * a deadlock. We have already changed rdev->sectors, and if
2554 * we have to change it back, we will have the lock again.
2555 */
2556 mddev_t *mddev;
2557 int overlap = 0;
2558 struct list_head *tmp;
2559
2560 mddev_unlock(my_mddev);
2561 for_each_mddev(mddev, tmp) {
2562 mdk_rdev_t *rdev2;
2563
2564 mddev_lock(mddev);
2565 list_for_each_entry(rdev2, &mddev->disks, same_set)
2566 if (test_bit(AllReserved, &rdev2->flags) ||
2567 (rdev->bdev == rdev2->bdev &&
2568 rdev != rdev2 &&
2569 overlaps(rdev->data_offset, rdev->sectors,
2570 rdev2->data_offset,
2571 rdev2->sectors))) {
2572 overlap = 1;
2573 break;
2574 }
2575 mddev_unlock(mddev);
2576 if (overlap) {
2577 mddev_put(mddev);
2578 break;
2579 }
2580 }
2581 mddev_lock(my_mddev);
2582 if (overlap) {
2583 /* Someone else could have slipped in a size
2584 * change here, but doing so is just silly.
2585 * We put oldsectors back because we *know* it is
2586 * safe, and trust userspace not to race with
2587 * itself
2588 */
2589 rdev->sectors = oldsectors;
2590 return -EBUSY;
2591 }
2592 }
2593 return len;
2594 }
2595
2596 static struct rdev_sysfs_entry rdev_size =
2597 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2598
2599
2600 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2601 {
2602 unsigned long long recovery_start = rdev->recovery_offset;
2603
2604 if (test_bit(In_sync, &rdev->flags) ||
2605 recovery_start == MaxSector)
2606 return sprintf(page, "none\n");
2607
2608 return sprintf(page, "%llu\n", recovery_start);
2609 }
2610
2611 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2612 {
2613 unsigned long long recovery_start;
2614
2615 if (cmd_match(buf, "none"))
2616 recovery_start = MaxSector;
2617 else if (strict_strtoull(buf, 10, &recovery_start))
2618 return -EINVAL;
2619
2620 if (rdev->mddev->pers &&
2621 rdev->raid_disk >= 0)
2622 return -EBUSY;
2623
2624 rdev->recovery_offset = recovery_start;
2625 if (recovery_start == MaxSector)
2626 set_bit(In_sync, &rdev->flags);
2627 else
2628 clear_bit(In_sync, &rdev->flags);
2629 return len;
2630 }
2631
2632 static struct rdev_sysfs_entry rdev_recovery_start =
2633 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2634
2635 static struct attribute *rdev_default_attrs[] = {
2636 &rdev_state.attr,
2637 &rdev_errors.attr,
2638 &rdev_slot.attr,
2639 &rdev_offset.attr,
2640 &rdev_size.attr,
2641 &rdev_recovery_start.attr,
2642 NULL,
2643 };
2644 static ssize_t
2645 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2646 {
2647 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2648 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2649 mddev_t *mddev = rdev->mddev;
2650 ssize_t rv;
2651
2652 if (!entry->show)
2653 return -EIO;
2654
2655 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2656 if (!rv) {
2657 if (rdev->mddev == NULL)
2658 rv = -EBUSY;
2659 else
2660 rv = entry->show(rdev, page);
2661 mddev_unlock(mddev);
2662 }
2663 return rv;
2664 }
2665
2666 static ssize_t
2667 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2668 const char *page, size_t length)
2669 {
2670 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2671 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2672 ssize_t rv;
2673 mddev_t *mddev = rdev->mddev;
2674
2675 if (!entry->store)
2676 return -EIO;
2677 if (!capable(CAP_SYS_ADMIN))
2678 return -EACCES;
2679 rv = mddev ? mddev_lock(mddev): -EBUSY;
2680 if (!rv) {
2681 if (rdev->mddev == NULL)
2682 rv = -EBUSY;
2683 else
2684 rv = entry->store(rdev, page, length);
2685 mddev_unlock(mddev);
2686 }
2687 return rv;
2688 }
2689
2690 static void rdev_free(struct kobject *ko)
2691 {
2692 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2693 kfree(rdev);
2694 }
2695 static const struct sysfs_ops rdev_sysfs_ops = {
2696 .show = rdev_attr_show,
2697 .store = rdev_attr_store,
2698 };
2699 static struct kobj_type rdev_ktype = {
2700 .release = rdev_free,
2701 .sysfs_ops = &rdev_sysfs_ops,
2702 .default_attrs = rdev_default_attrs,
2703 };
2704
2705 /*
2706 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2707 *
2708 * mark the device faulty if:
2709 *
2710 * - the device is nonexistent (zero size)
2711 * - the device has no valid superblock
2712 *
2713 * a faulty rdev _never_ has rdev->sb set.
2714 */
2715 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2716 {
2717 char b[BDEVNAME_SIZE];
2718 int err;
2719 mdk_rdev_t *rdev;
2720 sector_t size;
2721
2722 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2723 if (!rdev) {
2724 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2725 return ERR_PTR(-ENOMEM);
2726 }
2727
2728 if ((err = alloc_disk_sb(rdev)))
2729 goto abort_free;
2730
2731 err = lock_rdev(rdev, newdev, super_format == -2);
2732 if (err)
2733 goto abort_free;
2734
2735 kobject_init(&rdev->kobj, &rdev_ktype);
2736
2737 rdev->desc_nr = -1;
2738 rdev->saved_raid_disk = -1;
2739 rdev->raid_disk = -1;
2740 rdev->flags = 0;
2741 rdev->data_offset = 0;
2742 rdev->sb_events = 0;
2743 rdev->last_read_error.tv_sec = 0;
2744 rdev->last_read_error.tv_nsec = 0;
2745 atomic_set(&rdev->nr_pending, 0);
2746 atomic_set(&rdev->read_errors, 0);
2747 atomic_set(&rdev->corrected_errors, 0);
2748
2749 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2750 if (!size) {
2751 printk(KERN_WARNING
2752 "md: %s has zero or unknown size, marking faulty!\n",
2753 bdevname(rdev->bdev,b));
2754 err = -EINVAL;
2755 goto abort_free;
2756 }
2757
2758 if (super_format >= 0) {
2759 err = super_types[super_format].
2760 load_super(rdev, NULL, super_minor);
2761 if (err == -EINVAL) {
2762 printk(KERN_WARNING
2763 "md: %s does not have a valid v%d.%d "
2764 "superblock, not importing!\n",
2765 bdevname(rdev->bdev,b),
2766 super_format, super_minor);
2767 goto abort_free;
2768 }
2769 if (err < 0) {
2770 printk(KERN_WARNING
2771 "md: could not read %s's sb, not importing!\n",
2772 bdevname(rdev->bdev,b));
2773 goto abort_free;
2774 }
2775 }
2776
2777 INIT_LIST_HEAD(&rdev->same_set);
2778 init_waitqueue_head(&rdev->blocked_wait);
2779
2780 return rdev;
2781
2782 abort_free:
2783 if (rdev->sb_page) {
2784 if (rdev->bdev)
2785 unlock_rdev(rdev);
2786 free_disk_sb(rdev);
2787 }
2788 kfree(rdev);
2789 return ERR_PTR(err);
2790 }
2791
2792 /*
2793 * Check a full RAID array for plausibility
2794 */
2795
2796
2797 static void analyze_sbs(mddev_t * mddev)
2798 {
2799 int i;
2800 mdk_rdev_t *rdev, *freshest, *tmp;
2801 char b[BDEVNAME_SIZE];
2802
2803 freshest = NULL;
2804 rdev_for_each(rdev, tmp, mddev)
2805 switch (super_types[mddev->major_version].
2806 load_super(rdev, freshest, mddev->minor_version)) {
2807 case 1:
2808 freshest = rdev;
2809 break;
2810 case 0:
2811 break;
2812 default:
2813 printk( KERN_ERR \
2814 "md: fatal superblock inconsistency in %s"
2815 " -- removing from array\n",
2816 bdevname(rdev->bdev,b));
2817 kick_rdev_from_array(rdev);
2818 }
2819
2820
2821 super_types[mddev->major_version].
2822 validate_super(mddev, freshest);
2823
2824 i = 0;
2825 rdev_for_each(rdev, tmp, mddev) {
2826 if (mddev->max_disks &&
2827 (rdev->desc_nr >= mddev->max_disks ||
2828 i > mddev->max_disks)) {
2829 printk(KERN_WARNING
2830 "md: %s: %s: only %d devices permitted\n",
2831 mdname(mddev), bdevname(rdev->bdev, b),
2832 mddev->max_disks);
2833 kick_rdev_from_array(rdev);
2834 continue;
2835 }
2836 if (rdev != freshest)
2837 if (super_types[mddev->major_version].
2838 validate_super(mddev, rdev)) {
2839 printk(KERN_WARNING "md: kicking non-fresh %s"
2840 " from array!\n",
2841 bdevname(rdev->bdev,b));
2842 kick_rdev_from_array(rdev);
2843 continue;
2844 }
2845 if (mddev->level == LEVEL_MULTIPATH) {
2846 rdev->desc_nr = i++;
2847 rdev->raid_disk = rdev->desc_nr;
2848 set_bit(In_sync, &rdev->flags);
2849 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2850 rdev->raid_disk = -1;
2851 clear_bit(In_sync, &rdev->flags);
2852 }
2853 }
2854 }
2855
2856 /* Read a fixed-point number.
2857 * Numbers in sysfs attributes should be in "standard" units where
2858 * possible, so time should be in seconds.
2859 * However we internally use a a much smaller unit such as
2860 * milliseconds or jiffies.
2861 * This function takes a decimal number with a possible fractional
2862 * component, and produces an integer which is the result of
2863 * multiplying that number by 10^'scale'.
2864 * all without any floating-point arithmetic.
2865 */
2866 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2867 {
2868 unsigned long result = 0;
2869 long decimals = -1;
2870 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2871 if (*cp == '.')
2872 decimals = 0;
2873 else if (decimals < scale) {
2874 unsigned int value;
2875 value = *cp - '0';
2876 result = result * 10 + value;
2877 if (decimals >= 0)
2878 decimals++;
2879 }
2880 cp++;
2881 }
2882 if (*cp == '\n')
2883 cp++;
2884 if (*cp)
2885 return -EINVAL;
2886 if (decimals < 0)
2887 decimals = 0;
2888 while (decimals < scale) {
2889 result *= 10;
2890 decimals ++;
2891 }
2892 *res = result;
2893 return 0;
2894 }
2895
2896
2897 static void md_safemode_timeout(unsigned long data);
2898
2899 static ssize_t
2900 safe_delay_show(mddev_t *mddev, char *page)
2901 {
2902 int msec = (mddev->safemode_delay*1000)/HZ;
2903 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2904 }
2905 static ssize_t
2906 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2907 {
2908 unsigned long msec;
2909
2910 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2911 return -EINVAL;
2912 if (msec == 0)
2913 mddev->safemode_delay = 0;
2914 else {
2915 unsigned long old_delay = mddev->safemode_delay;
2916 mddev->safemode_delay = (msec*HZ)/1000;
2917 if (mddev->safemode_delay == 0)
2918 mddev->safemode_delay = 1;
2919 if (mddev->safemode_delay < old_delay)
2920 md_safemode_timeout((unsigned long)mddev);
2921 }
2922 return len;
2923 }
2924 static struct md_sysfs_entry md_safe_delay =
2925 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2926
2927 static ssize_t
2928 level_show(mddev_t *mddev, char *page)
2929 {
2930 struct mdk_personality *p = mddev->pers;
2931 if (p)
2932 return sprintf(page, "%s\n", p->name);
2933 else if (mddev->clevel[0])
2934 return sprintf(page, "%s\n", mddev->clevel);
2935 else if (mddev->level != LEVEL_NONE)
2936 return sprintf(page, "%d\n", mddev->level);
2937 else
2938 return 0;
2939 }
2940
2941 static ssize_t
2942 level_store(mddev_t *mddev, const char *buf, size_t len)
2943 {
2944 char clevel[16];
2945 ssize_t rv = len;
2946 struct mdk_personality *pers;
2947 long level;
2948 void *priv;
2949 mdk_rdev_t *rdev;
2950
2951 if (mddev->pers == NULL) {
2952 if (len == 0)
2953 return 0;
2954 if (len >= sizeof(mddev->clevel))
2955 return -ENOSPC;
2956 strncpy(mddev->clevel, buf, len);
2957 if (mddev->clevel[len-1] == '\n')
2958 len--;
2959 mddev->clevel[len] = 0;
2960 mddev->level = LEVEL_NONE;
2961 return rv;
2962 }
2963
2964 /* request to change the personality. Need to ensure:
2965 * - array is not engaged in resync/recovery/reshape
2966 * - old personality can be suspended
2967 * - new personality will access other array.
2968 */
2969
2970 if (mddev->sync_thread ||
2971 mddev->reshape_position != MaxSector ||
2972 mddev->sysfs_active)
2973 return -EBUSY;
2974
2975 if (!mddev->pers->quiesce) {
2976 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2977 mdname(mddev), mddev->pers->name);
2978 return -EINVAL;
2979 }
2980
2981 /* Now find the new personality */
2982 if (len == 0 || len >= sizeof(clevel))
2983 return -EINVAL;
2984 strncpy(clevel, buf, len);
2985 if (clevel[len-1] == '\n')
2986 len--;
2987 clevel[len] = 0;
2988 if (strict_strtol(clevel, 10, &level))
2989 level = LEVEL_NONE;
2990
2991 if (request_module("md-%s", clevel) != 0)
2992 request_module("md-level-%s", clevel);
2993 spin_lock(&pers_lock);
2994 pers = find_pers(level, clevel);
2995 if (!pers || !try_module_get(pers->owner)) {
2996 spin_unlock(&pers_lock);
2997 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2998 return -EINVAL;
2999 }
3000 spin_unlock(&pers_lock);
3001
3002 if (pers == mddev->pers) {
3003 /* Nothing to do! */
3004 module_put(pers->owner);
3005 return rv;
3006 }
3007 if (!pers->takeover) {
3008 module_put(pers->owner);
3009 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3010 mdname(mddev), clevel);
3011 return -EINVAL;
3012 }
3013
3014 list_for_each_entry(rdev, &mddev->disks, same_set)
3015 rdev->new_raid_disk = rdev->raid_disk;
3016
3017 /* ->takeover must set new_* and/or delta_disks
3018 * if it succeeds, and may set them when it fails.
3019 */
3020 priv = pers->takeover(mddev);
3021 if (IS_ERR(priv)) {
3022 mddev->new_level = mddev->level;
3023 mddev->new_layout = mddev->layout;
3024 mddev->new_chunk_sectors = mddev->chunk_sectors;
3025 mddev->raid_disks -= mddev->delta_disks;
3026 mddev->delta_disks = 0;
3027 module_put(pers->owner);
3028 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3029 mdname(mddev), clevel);
3030 return PTR_ERR(priv);
3031 }
3032
3033 /* Looks like we have a winner */
3034 mddev_suspend(mddev);
3035 mddev->pers->stop(mddev);
3036
3037 if (mddev->pers->sync_request == NULL &&
3038 pers->sync_request != NULL) {
3039 /* need to add the md_redundancy_group */
3040 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3041 printk(KERN_WARNING
3042 "md: cannot register extra attributes for %s\n",
3043 mdname(mddev));
3044 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3045 }
3046 if (mddev->pers->sync_request != NULL &&
3047 pers->sync_request == NULL) {
3048 /* need to remove the md_redundancy_group */
3049 if (mddev->to_remove == NULL)
3050 mddev->to_remove = &md_redundancy_group;
3051 }
3052
3053 if (mddev->pers->sync_request == NULL &&
3054 mddev->external) {
3055 /* We are converting from a no-redundancy array
3056 * to a redundancy array and metadata is managed
3057 * externally so we need to be sure that writes
3058 * won't block due to a need to transition
3059 * clean->dirty
3060 * until external management is started.
3061 */
3062 mddev->in_sync = 0;
3063 mddev->safemode_delay = 0;
3064 mddev->safemode = 0;
3065 }
3066
3067 list_for_each_entry(rdev, &mddev->disks, same_set) {
3068 char nm[20];
3069 if (rdev->raid_disk < 0)
3070 continue;
3071 if (rdev->new_raid_disk > mddev->raid_disks)
3072 rdev->new_raid_disk = -1;
3073 if (rdev->new_raid_disk == rdev->raid_disk)
3074 continue;
3075 sprintf(nm, "rd%d", rdev->raid_disk);
3076 sysfs_remove_link(&mddev->kobj, nm);
3077 }
3078 list_for_each_entry(rdev, &mddev->disks, same_set) {
3079 if (rdev->raid_disk < 0)
3080 continue;
3081 if (rdev->new_raid_disk == rdev->raid_disk)
3082 continue;
3083 rdev->raid_disk = rdev->new_raid_disk;
3084 if (rdev->raid_disk < 0)
3085 clear_bit(In_sync, &rdev->flags);
3086 else {
3087 char nm[20];
3088 sprintf(nm, "rd%d", rdev->raid_disk);
3089 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3090 printk("md: cannot register %s for %s after level change\n",
3091 nm, mdname(mddev));
3092 }
3093 }
3094
3095 module_put(mddev->pers->owner);
3096 mddev->pers = pers;
3097 mddev->private = priv;
3098 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3099 mddev->level = mddev->new_level;
3100 mddev->layout = mddev->new_layout;
3101 mddev->chunk_sectors = mddev->new_chunk_sectors;
3102 mddev->delta_disks = 0;
3103 if (mddev->pers->sync_request == NULL) {
3104 /* this is now an array without redundancy, so
3105 * it must always be in_sync
3106 */
3107 mddev->in_sync = 1;
3108 del_timer_sync(&mddev->safemode_timer);
3109 }
3110 pers->run(mddev);
3111 mddev_resume(mddev);
3112 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3113 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3114 md_wakeup_thread(mddev->thread);
3115 sysfs_notify(&mddev->kobj, NULL, "level");
3116 md_new_event(mddev);
3117 return rv;
3118 }
3119
3120 static struct md_sysfs_entry md_level =
3121 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3122
3123
3124 static ssize_t
3125 layout_show(mddev_t *mddev, char *page)
3126 {
3127 /* just a number, not meaningful for all levels */
3128 if (mddev->reshape_position != MaxSector &&
3129 mddev->layout != mddev->new_layout)
3130 return sprintf(page, "%d (%d)\n",
3131 mddev->new_layout, mddev->layout);
3132 return sprintf(page, "%d\n", mddev->layout);
3133 }
3134
3135 static ssize_t
3136 layout_store(mddev_t *mddev, const char *buf, size_t len)
3137 {
3138 char *e;
3139 unsigned long n = simple_strtoul(buf, &e, 10);
3140
3141 if (!*buf || (*e && *e != '\n'))
3142 return -EINVAL;
3143
3144 if (mddev->pers) {
3145 int err;
3146 if (mddev->pers->check_reshape == NULL)
3147 return -EBUSY;
3148 mddev->new_layout = n;
3149 err = mddev->pers->check_reshape(mddev);
3150 if (err) {
3151 mddev->new_layout = mddev->layout;
3152 return err;
3153 }
3154 } else {
3155 mddev->new_layout = n;
3156 if (mddev->reshape_position == MaxSector)
3157 mddev->layout = n;
3158 }
3159 return len;
3160 }
3161 static struct md_sysfs_entry md_layout =
3162 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3163
3164
3165 static ssize_t
3166 raid_disks_show(mddev_t *mddev, char *page)
3167 {
3168 if (mddev->raid_disks == 0)
3169 return 0;
3170 if (mddev->reshape_position != MaxSector &&
3171 mddev->delta_disks != 0)
3172 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3173 mddev->raid_disks - mddev->delta_disks);
3174 return sprintf(page, "%d\n", mddev->raid_disks);
3175 }
3176
3177 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3178
3179 static ssize_t
3180 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3181 {
3182 char *e;
3183 int rv = 0;
3184 unsigned long n = simple_strtoul(buf, &e, 10);
3185
3186 if (!*buf || (*e && *e != '\n'))
3187 return -EINVAL;
3188
3189 if (mddev->pers)
3190 rv = update_raid_disks(mddev, n);
3191 else if (mddev->reshape_position != MaxSector) {
3192 int olddisks = mddev->raid_disks - mddev->delta_disks;
3193 mddev->delta_disks = n - olddisks;
3194 mddev->raid_disks = n;
3195 } else
3196 mddev->raid_disks = n;
3197 return rv ? rv : len;
3198 }
3199 static struct md_sysfs_entry md_raid_disks =
3200 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3201
3202 static ssize_t
3203 chunk_size_show(mddev_t *mddev, char *page)
3204 {
3205 if (mddev->reshape_position != MaxSector &&
3206 mddev->chunk_sectors != mddev->new_chunk_sectors)
3207 return sprintf(page, "%d (%d)\n",
3208 mddev->new_chunk_sectors << 9,
3209 mddev->chunk_sectors << 9);
3210 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3211 }
3212
3213 static ssize_t
3214 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3215 {
3216 char *e;
3217 unsigned long n = simple_strtoul(buf, &e, 10);
3218
3219 if (!*buf || (*e && *e != '\n'))
3220 return -EINVAL;
3221
3222 if (mddev->pers) {
3223 int err;
3224 if (mddev->pers->check_reshape == NULL)
3225 return -EBUSY;
3226 mddev->new_chunk_sectors = n >> 9;
3227 err = mddev->pers->check_reshape(mddev);
3228 if (err) {
3229 mddev->new_chunk_sectors = mddev->chunk_sectors;
3230 return err;
3231 }
3232 } else {
3233 mddev->new_chunk_sectors = n >> 9;
3234 if (mddev->reshape_position == MaxSector)
3235 mddev->chunk_sectors = n >> 9;
3236 }
3237 return len;
3238 }
3239 static struct md_sysfs_entry md_chunk_size =
3240 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3241
3242 static ssize_t
3243 resync_start_show(mddev_t *mddev, char *page)
3244 {
3245 if (mddev->recovery_cp == MaxSector)
3246 return sprintf(page, "none\n");
3247 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3248 }
3249
3250 static ssize_t
3251 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3252 {
3253 char *e;
3254 unsigned long long n = simple_strtoull(buf, &e, 10);
3255
3256 if (mddev->pers)
3257 return -EBUSY;
3258 if (cmd_match(buf, "none"))
3259 n = MaxSector;
3260 else if (!*buf || (*e && *e != '\n'))
3261 return -EINVAL;
3262
3263 mddev->recovery_cp = n;
3264 return len;
3265 }
3266 static struct md_sysfs_entry md_resync_start =
3267 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3268
3269 /*
3270 * The array state can be:
3271 *
3272 * clear
3273 * No devices, no size, no level
3274 * Equivalent to STOP_ARRAY ioctl
3275 * inactive
3276 * May have some settings, but array is not active
3277 * all IO results in error
3278 * When written, doesn't tear down array, but just stops it
3279 * suspended (not supported yet)
3280 * All IO requests will block. The array can be reconfigured.
3281 * Writing this, if accepted, will block until array is quiescent
3282 * readonly
3283 * no resync can happen. no superblocks get written.
3284 * write requests fail
3285 * read-auto
3286 * like readonly, but behaves like 'clean' on a write request.
3287 *
3288 * clean - no pending writes, but otherwise active.
3289 * When written to inactive array, starts without resync
3290 * If a write request arrives then
3291 * if metadata is known, mark 'dirty' and switch to 'active'.
3292 * if not known, block and switch to write-pending
3293 * If written to an active array that has pending writes, then fails.
3294 * active
3295 * fully active: IO and resync can be happening.
3296 * When written to inactive array, starts with resync
3297 *
3298 * write-pending
3299 * clean, but writes are blocked waiting for 'active' to be written.
3300 *
3301 * active-idle
3302 * like active, but no writes have been seen for a while (100msec).
3303 *
3304 */
3305 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3306 write_pending, active_idle, bad_word};
3307 static char *array_states[] = {
3308 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3309 "write-pending", "active-idle", NULL };
3310
3311 static int match_word(const char *word, char **list)
3312 {
3313 int n;
3314 for (n=0; list[n]; n++)
3315 if (cmd_match(word, list[n]))
3316 break;
3317 return n;
3318 }
3319
3320 static ssize_t
3321 array_state_show(mddev_t *mddev, char *page)
3322 {
3323 enum array_state st = inactive;
3324
3325 if (mddev->pers)
3326 switch(mddev->ro) {
3327 case 1:
3328 st = readonly;
3329 break;
3330 case 2:
3331 st = read_auto;
3332 break;
3333 case 0:
3334 if (mddev->in_sync)
3335 st = clean;
3336 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3337 st = write_pending;
3338 else if (mddev->safemode)
3339 st = active_idle;
3340 else
3341 st = active;
3342 }
3343 else {
3344 if (list_empty(&mddev->disks) &&
3345 mddev->raid_disks == 0 &&
3346 mddev->dev_sectors == 0)
3347 st = clear;
3348 else
3349 st = inactive;
3350 }
3351 return sprintf(page, "%s\n", array_states[st]);
3352 }
3353
3354 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3355 static int md_set_readonly(mddev_t * mddev, int is_open);
3356 static int do_md_run(mddev_t * mddev);
3357 static int restart_array(mddev_t *mddev);
3358
3359 static ssize_t
3360 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3361 {
3362 int err = -EINVAL;
3363 enum array_state st = match_word(buf, array_states);
3364 switch(st) {
3365 case bad_word:
3366 break;
3367 case clear:
3368 /* stopping an active array */
3369 if (atomic_read(&mddev->openers) > 0)
3370 return -EBUSY;
3371 err = do_md_stop(mddev, 0, 0);
3372 break;
3373 case inactive:
3374 /* stopping an active array */
3375 if (mddev->pers) {
3376 if (atomic_read(&mddev->openers) > 0)
3377 return -EBUSY;
3378 err = do_md_stop(mddev, 2, 0);
3379 } else
3380 err = 0; /* already inactive */
3381 break;
3382 case suspended:
3383 break; /* not supported yet */
3384 case readonly:
3385 if (mddev->pers)
3386 err = md_set_readonly(mddev, 0);
3387 else {
3388 mddev->ro = 1;
3389 set_disk_ro(mddev->gendisk, 1);
3390 err = do_md_run(mddev);
3391 }
3392 break;
3393 case read_auto:
3394 if (mddev->pers) {
3395 if (mddev->ro == 0)
3396 err = md_set_readonly(mddev, 0);
3397 else if (mddev->ro == 1)
3398 err = restart_array(mddev);
3399 if (err == 0) {
3400 mddev->ro = 2;
3401 set_disk_ro(mddev->gendisk, 0);
3402 }
3403 } else {
3404 mddev->ro = 2;
3405 err = do_md_run(mddev);
3406 }
3407 break;
3408 case clean:
3409 if (mddev->pers) {
3410 restart_array(mddev);
3411 spin_lock_irq(&mddev->write_lock);
3412 if (atomic_read(&mddev->writes_pending) == 0) {
3413 if (mddev->in_sync == 0) {
3414 mddev->in_sync = 1;
3415 if (mddev->safemode == 1)
3416 mddev->safemode = 0;
3417 if (mddev->persistent)
3418 set_bit(MD_CHANGE_CLEAN,
3419 &mddev->flags);
3420 }
3421 err = 0;
3422 } else
3423 err = -EBUSY;
3424 spin_unlock_irq(&mddev->write_lock);
3425 } else
3426 err = -EINVAL;
3427 break;
3428 case active:
3429 if (mddev->pers) {
3430 restart_array(mddev);
3431 if (mddev->external)
3432 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3433 wake_up(&mddev->sb_wait);
3434 err = 0;
3435 } else {
3436 mddev->ro = 0;
3437 set_disk_ro(mddev->gendisk, 0);
3438 err = do_md_run(mddev);
3439 }
3440 break;
3441 case write_pending:
3442 case active_idle:
3443 /* these cannot be set */
3444 break;
3445 }
3446 if (err)
3447 return err;
3448 else {
3449 sysfs_notify_dirent(mddev->sysfs_state);
3450 return len;
3451 }
3452 }
3453 static struct md_sysfs_entry md_array_state =
3454 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3455
3456 static ssize_t
3457 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3458 return sprintf(page, "%d\n",
3459 atomic_read(&mddev->max_corr_read_errors));
3460 }
3461
3462 static ssize_t
3463 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3464 {
3465 char *e;
3466 unsigned long n = simple_strtoul(buf, &e, 10);
3467
3468 if (*buf && (*e == 0 || *e == '\n')) {
3469 atomic_set(&mddev->max_corr_read_errors, n);
3470 return len;
3471 }
3472 return -EINVAL;
3473 }
3474
3475 static struct md_sysfs_entry max_corr_read_errors =
3476 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3477 max_corrected_read_errors_store);
3478
3479 static ssize_t
3480 null_show(mddev_t *mddev, char *page)
3481 {
3482 return -EINVAL;
3483 }
3484
3485 static ssize_t
3486 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3487 {
3488 /* buf must be %d:%d\n? giving major and minor numbers */
3489 /* The new device is added to the array.
3490 * If the array has a persistent superblock, we read the
3491 * superblock to initialise info and check validity.
3492 * Otherwise, only checking done is that in bind_rdev_to_array,
3493 * which mainly checks size.
3494 */
3495 char *e;
3496 int major = simple_strtoul(buf, &e, 10);
3497 int minor;
3498 dev_t dev;
3499 mdk_rdev_t *rdev;
3500 int err;
3501
3502 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3503 return -EINVAL;
3504 minor = simple_strtoul(e+1, &e, 10);
3505 if (*e && *e != '\n')
3506 return -EINVAL;
3507 dev = MKDEV(major, minor);
3508 if (major != MAJOR(dev) ||
3509 minor != MINOR(dev))
3510 return -EOVERFLOW;
3511
3512
3513 if (mddev->persistent) {
3514 rdev = md_import_device(dev, mddev->major_version,
3515 mddev->minor_version);
3516 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3517 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3518 mdk_rdev_t, same_set);
3519 err = super_types[mddev->major_version]
3520 .load_super(rdev, rdev0, mddev->minor_version);
3521 if (err < 0)
3522 goto out;
3523 }
3524 } else if (mddev->external)
3525 rdev = md_import_device(dev, -2, -1);
3526 else
3527 rdev = md_import_device(dev, -1, -1);
3528
3529 if (IS_ERR(rdev))
3530 return PTR_ERR(rdev);
3531 err = bind_rdev_to_array(rdev, mddev);
3532 out:
3533 if (err)
3534 export_rdev(rdev);
3535 return err ? err : len;
3536 }
3537
3538 static struct md_sysfs_entry md_new_device =
3539 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3540
3541 static ssize_t
3542 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3543 {
3544 char *end;
3545 unsigned long chunk, end_chunk;
3546
3547 if (!mddev->bitmap)
3548 goto out;
3549 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3550 while (*buf) {
3551 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3552 if (buf == end) break;
3553 if (*end == '-') { /* range */
3554 buf = end + 1;
3555 end_chunk = simple_strtoul(buf, &end, 0);
3556 if (buf == end) break;
3557 }
3558 if (*end && !isspace(*end)) break;
3559 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3560 buf = skip_spaces(end);
3561 }
3562 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3563 out:
3564 return len;
3565 }
3566
3567 static struct md_sysfs_entry md_bitmap =
3568 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3569
3570 static ssize_t
3571 size_show(mddev_t *mddev, char *page)
3572 {
3573 return sprintf(page, "%llu\n",
3574 (unsigned long long)mddev->dev_sectors / 2);
3575 }
3576
3577 static int update_size(mddev_t *mddev, sector_t num_sectors);
3578
3579 static ssize_t
3580 size_store(mddev_t *mddev, const char *buf, size_t len)
3581 {
3582 /* If array is inactive, we can reduce the component size, but
3583 * not increase it (except from 0).
3584 * If array is active, we can try an on-line resize
3585 */
3586 sector_t sectors;
3587 int err = strict_blocks_to_sectors(buf, &sectors);
3588
3589 if (err < 0)
3590 return err;
3591 if (mddev->pers) {
3592 err = update_size(mddev, sectors);
3593 md_update_sb(mddev, 1);
3594 } else {
3595 if (mddev->dev_sectors == 0 ||
3596 mddev->dev_sectors > sectors)
3597 mddev->dev_sectors = sectors;
3598 else
3599 err = -ENOSPC;
3600 }
3601 return err ? err : len;
3602 }
3603
3604 static struct md_sysfs_entry md_size =
3605 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3606
3607
3608 /* Metdata version.
3609 * This is one of
3610 * 'none' for arrays with no metadata (good luck...)
3611 * 'external' for arrays with externally managed metadata,
3612 * or N.M for internally known formats
3613 */
3614 static ssize_t
3615 metadata_show(mddev_t *mddev, char *page)
3616 {
3617 if (mddev->persistent)
3618 return sprintf(page, "%d.%d\n",
3619 mddev->major_version, mddev->minor_version);
3620 else if (mddev->external)
3621 return sprintf(page, "external:%s\n", mddev->metadata_type);
3622 else
3623 return sprintf(page, "none\n");
3624 }
3625
3626 static ssize_t
3627 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3628 {
3629 int major, minor;
3630 char *e;
3631 /* Changing the details of 'external' metadata is
3632 * always permitted. Otherwise there must be
3633 * no devices attached to the array.
3634 */
3635 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3636 ;
3637 else if (!list_empty(&mddev->disks))
3638 return -EBUSY;
3639
3640 if (cmd_match(buf, "none")) {
3641 mddev->persistent = 0;
3642 mddev->external = 0;
3643 mddev->major_version = 0;
3644 mddev->minor_version = 90;
3645 return len;
3646 }
3647 if (strncmp(buf, "external:", 9) == 0) {
3648 size_t namelen = len-9;
3649 if (namelen >= sizeof(mddev->metadata_type))
3650 namelen = sizeof(mddev->metadata_type)-1;
3651 strncpy(mddev->metadata_type, buf+9, namelen);
3652 mddev->metadata_type[namelen] = 0;
3653 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3654 mddev->metadata_type[--namelen] = 0;
3655 mddev->persistent = 0;
3656 mddev->external = 1;
3657 mddev->major_version = 0;
3658 mddev->minor_version = 90;
3659 return len;
3660 }
3661 major = simple_strtoul(buf, &e, 10);
3662 if (e==buf || *e != '.')
3663 return -EINVAL;
3664 buf = e+1;
3665 minor = simple_strtoul(buf, &e, 10);
3666 if (e==buf || (*e && *e != '\n') )
3667 return -EINVAL;
3668 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3669 return -ENOENT;
3670 mddev->major_version = major;
3671 mddev->minor_version = minor;
3672 mddev->persistent = 1;
3673 mddev->external = 0;
3674 return len;
3675 }
3676
3677 static struct md_sysfs_entry md_metadata =
3678 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3679
3680 static ssize_t
3681 action_show(mddev_t *mddev, char *page)
3682 {
3683 char *type = "idle";
3684 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3685 type = "frozen";
3686 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3687 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3688 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3689 type = "reshape";
3690 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3691 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3692 type = "resync";
3693 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3694 type = "check";
3695 else
3696 type = "repair";
3697 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3698 type = "recover";
3699 }
3700 return sprintf(page, "%s\n", type);
3701 }
3702
3703 static ssize_t
3704 action_store(mddev_t *mddev, const char *page, size_t len)
3705 {
3706 if (!mddev->pers || !mddev->pers->sync_request)
3707 return -EINVAL;
3708
3709 if (cmd_match(page, "frozen"))
3710 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3711 else
3712 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3713
3714 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3715 if (mddev->sync_thread) {
3716 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3717 md_unregister_thread(mddev->sync_thread);
3718 mddev->sync_thread = NULL;
3719 mddev->recovery = 0;
3720 }
3721 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3722 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3723 return -EBUSY;
3724 else if (cmd_match(page, "resync"))
3725 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3726 else if (cmd_match(page, "recover")) {
3727 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3728 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3729 } else if (cmd_match(page, "reshape")) {
3730 int err;
3731 if (mddev->pers->start_reshape == NULL)
3732 return -EINVAL;
3733 err = mddev->pers->start_reshape(mddev);
3734 if (err)
3735 return err;
3736 sysfs_notify(&mddev->kobj, NULL, "degraded");
3737 } else {
3738 if (cmd_match(page, "check"))
3739 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3740 else if (!cmd_match(page, "repair"))
3741 return -EINVAL;
3742 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3743 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3744 }
3745 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3746 md_wakeup_thread(mddev->thread);
3747 sysfs_notify_dirent(mddev->sysfs_action);
3748 return len;
3749 }
3750
3751 static ssize_t
3752 mismatch_cnt_show(mddev_t *mddev, char *page)
3753 {
3754 return sprintf(page, "%llu\n",
3755 (unsigned long long) mddev->resync_mismatches);
3756 }
3757
3758 static struct md_sysfs_entry md_scan_mode =
3759 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3760
3761
3762 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3763
3764 static ssize_t
3765 sync_min_show(mddev_t *mddev, char *page)
3766 {
3767 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3768 mddev->sync_speed_min ? "local": "system");
3769 }
3770
3771 static ssize_t
3772 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3773 {
3774 int min;
3775 char *e;
3776 if (strncmp(buf, "system", 6)==0) {
3777 mddev->sync_speed_min = 0;
3778 return len;
3779 }
3780 min = simple_strtoul(buf, &e, 10);
3781 if (buf == e || (*e && *e != '\n') || min <= 0)
3782 return -EINVAL;
3783 mddev->sync_speed_min = min;
3784 return len;
3785 }
3786
3787 static struct md_sysfs_entry md_sync_min =
3788 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3789
3790 static ssize_t
3791 sync_max_show(mddev_t *mddev, char *page)
3792 {
3793 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3794 mddev->sync_speed_max ? "local": "system");
3795 }
3796
3797 static ssize_t
3798 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3799 {
3800 int max;
3801 char *e;
3802 if (strncmp(buf, "system", 6)==0) {
3803 mddev->sync_speed_max = 0;
3804 return len;
3805 }
3806 max = simple_strtoul(buf, &e, 10);
3807 if (buf == e || (*e && *e != '\n') || max <= 0)
3808 return -EINVAL;
3809 mddev->sync_speed_max = max;
3810 return len;
3811 }
3812
3813 static struct md_sysfs_entry md_sync_max =
3814 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3815
3816 static ssize_t
3817 degraded_show(mddev_t *mddev, char *page)
3818 {
3819 return sprintf(page, "%d\n", mddev->degraded);
3820 }
3821 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3822
3823 static ssize_t
3824 sync_force_parallel_show(mddev_t *mddev, char *page)
3825 {
3826 return sprintf(page, "%d\n", mddev->parallel_resync);
3827 }
3828
3829 static ssize_t
3830 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3831 {
3832 long n;
3833
3834 if (strict_strtol(buf, 10, &n))
3835 return -EINVAL;
3836
3837 if (n != 0 && n != 1)
3838 return -EINVAL;
3839
3840 mddev->parallel_resync = n;
3841
3842 if (mddev->sync_thread)
3843 wake_up(&resync_wait);
3844
3845 return len;
3846 }
3847
3848 /* force parallel resync, even with shared block devices */
3849 static struct md_sysfs_entry md_sync_force_parallel =
3850 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3851 sync_force_parallel_show, sync_force_parallel_store);
3852
3853 static ssize_t
3854 sync_speed_show(mddev_t *mddev, char *page)
3855 {
3856 unsigned long resync, dt, db;
3857 if (mddev->curr_resync == 0)
3858 return sprintf(page, "none\n");
3859 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3860 dt = (jiffies - mddev->resync_mark) / HZ;
3861 if (!dt) dt++;
3862 db = resync - mddev->resync_mark_cnt;
3863 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3864 }
3865
3866 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3867
3868 static ssize_t
3869 sync_completed_show(mddev_t *mddev, char *page)
3870 {
3871 unsigned long max_sectors, resync;
3872
3873 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3874 return sprintf(page, "none\n");
3875
3876 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3877 max_sectors = mddev->resync_max_sectors;
3878 else
3879 max_sectors = mddev->dev_sectors;
3880
3881 resync = mddev->curr_resync_completed;
3882 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3883 }
3884
3885 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3886
3887 static ssize_t
3888 min_sync_show(mddev_t *mddev, char *page)
3889 {
3890 return sprintf(page, "%llu\n",
3891 (unsigned long long)mddev->resync_min);
3892 }
3893 static ssize_t
3894 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3895 {
3896 unsigned long long min;
3897 if (strict_strtoull(buf, 10, &min))
3898 return -EINVAL;
3899 if (min > mddev->resync_max)
3900 return -EINVAL;
3901 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3902 return -EBUSY;
3903
3904 /* Must be a multiple of chunk_size */
3905 if (mddev->chunk_sectors) {
3906 sector_t temp = min;
3907 if (sector_div(temp, mddev->chunk_sectors))
3908 return -EINVAL;
3909 }
3910 mddev->resync_min = min;
3911
3912 return len;
3913 }
3914
3915 static struct md_sysfs_entry md_min_sync =
3916 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3917
3918 static ssize_t
3919 max_sync_show(mddev_t *mddev, char *page)
3920 {
3921 if (mddev->resync_max == MaxSector)
3922 return sprintf(page, "max\n");
3923 else
3924 return sprintf(page, "%llu\n",
3925 (unsigned long long)mddev->resync_max);
3926 }
3927 static ssize_t
3928 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3929 {
3930 if (strncmp(buf, "max", 3) == 0)
3931 mddev->resync_max = MaxSector;
3932 else {
3933 unsigned long long max;
3934 if (strict_strtoull(buf, 10, &max))
3935 return -EINVAL;
3936 if (max < mddev->resync_min)
3937 return -EINVAL;
3938 if (max < mddev->resync_max &&
3939 mddev->ro == 0 &&
3940 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3941 return -EBUSY;
3942
3943 /* Must be a multiple of chunk_size */
3944 if (mddev->chunk_sectors) {
3945 sector_t temp = max;
3946 if (sector_div(temp, mddev->chunk_sectors))
3947 return -EINVAL;
3948 }
3949 mddev->resync_max = max;
3950 }
3951 wake_up(&mddev->recovery_wait);
3952 return len;
3953 }
3954
3955 static struct md_sysfs_entry md_max_sync =
3956 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3957
3958 static ssize_t
3959 suspend_lo_show(mddev_t *mddev, char *page)
3960 {
3961 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3962 }
3963
3964 static ssize_t
3965 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3966 {
3967 char *e;
3968 unsigned long long new = simple_strtoull(buf, &e, 10);
3969
3970 if (mddev->pers == NULL ||
3971 mddev->pers->quiesce == NULL)
3972 return -EINVAL;
3973 if (buf == e || (*e && *e != '\n'))
3974 return -EINVAL;
3975 if (new >= mddev->suspend_hi ||
3976 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3977 mddev->suspend_lo = new;
3978 mddev->pers->quiesce(mddev, 2);
3979 return len;
3980 } else
3981 return -EINVAL;
3982 }
3983 static struct md_sysfs_entry md_suspend_lo =
3984 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3985
3986
3987 static ssize_t
3988 suspend_hi_show(mddev_t *mddev, char *page)
3989 {
3990 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3991 }
3992
3993 static ssize_t
3994 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3995 {
3996 char *e;
3997 unsigned long long new = simple_strtoull(buf, &e, 10);
3998
3999 if (mddev->pers == NULL ||
4000 mddev->pers->quiesce == NULL)
4001 return -EINVAL;
4002 if (buf == e || (*e && *e != '\n'))
4003 return -EINVAL;
4004 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4005 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4006 mddev->suspend_hi = new;
4007 mddev->pers->quiesce(mddev, 1);
4008 mddev->pers->quiesce(mddev, 0);
4009 return len;
4010 } else
4011 return -EINVAL;
4012 }
4013 static struct md_sysfs_entry md_suspend_hi =
4014 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4015
4016 static ssize_t
4017 reshape_position_show(mddev_t *mddev, char *page)
4018 {
4019 if (mddev->reshape_position != MaxSector)
4020 return sprintf(page, "%llu\n",
4021 (unsigned long long)mddev->reshape_position);
4022 strcpy(page, "none\n");
4023 return 5;
4024 }
4025
4026 static ssize_t
4027 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4028 {
4029 char *e;
4030 unsigned long long new = simple_strtoull(buf, &e, 10);
4031 if (mddev->pers)
4032 return -EBUSY;
4033 if (buf == e || (*e && *e != '\n'))
4034 return -EINVAL;
4035 mddev->reshape_position = new;
4036 mddev->delta_disks = 0;
4037 mddev->new_level = mddev->level;
4038 mddev->new_layout = mddev->layout;
4039 mddev->new_chunk_sectors = mddev->chunk_sectors;
4040 return len;
4041 }
4042
4043 static struct md_sysfs_entry md_reshape_position =
4044 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4045 reshape_position_store);
4046
4047 static ssize_t
4048 array_size_show(mddev_t *mddev, char *page)
4049 {
4050 if (mddev->external_size)
4051 return sprintf(page, "%llu\n",
4052 (unsigned long long)mddev->array_sectors/2);
4053 else
4054 return sprintf(page, "default\n");
4055 }
4056
4057 static ssize_t
4058 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4059 {
4060 sector_t sectors;
4061
4062 if (strncmp(buf, "default", 7) == 0) {
4063 if (mddev->pers)
4064 sectors = mddev->pers->size(mddev, 0, 0);
4065 else
4066 sectors = mddev->array_sectors;
4067
4068 mddev->external_size = 0;
4069 } else {
4070 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4071 return -EINVAL;
4072 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4073 return -E2BIG;
4074
4075 mddev->external_size = 1;
4076 }
4077
4078 mddev->array_sectors = sectors;
4079 set_capacity(mddev->gendisk, mddev->array_sectors);
4080 if (mddev->pers)
4081 revalidate_disk(mddev->gendisk);
4082
4083 return len;
4084 }
4085
4086 static struct md_sysfs_entry md_array_size =
4087 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4088 array_size_store);
4089
4090 static struct attribute *md_default_attrs[] = {
4091 &md_level.attr,
4092 &md_layout.attr,
4093 &md_raid_disks.attr,
4094 &md_chunk_size.attr,
4095 &md_size.attr,
4096 &md_resync_start.attr,
4097 &md_metadata.attr,
4098 &md_new_device.attr,
4099 &md_safe_delay.attr,
4100 &md_array_state.attr,
4101 &md_reshape_position.attr,
4102 &md_array_size.attr,
4103 &max_corr_read_errors.attr,
4104 NULL,
4105 };
4106
4107 static struct attribute *md_redundancy_attrs[] = {
4108 &md_scan_mode.attr,
4109 &md_mismatches.attr,
4110 &md_sync_min.attr,
4111 &md_sync_max.attr,
4112 &md_sync_speed.attr,
4113 &md_sync_force_parallel.attr,
4114 &md_sync_completed.attr,
4115 &md_min_sync.attr,
4116 &md_max_sync.attr,
4117 &md_suspend_lo.attr,
4118 &md_suspend_hi.attr,
4119 &md_bitmap.attr,
4120 &md_degraded.attr,
4121 NULL,
4122 };
4123 static struct attribute_group md_redundancy_group = {
4124 .name = NULL,
4125 .attrs = md_redundancy_attrs,
4126 };
4127
4128
4129 static ssize_t
4130 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4131 {
4132 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4133 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4134 ssize_t rv;
4135
4136 if (!entry->show)
4137 return -EIO;
4138 rv = mddev_lock(mddev);
4139 if (!rv) {
4140 rv = entry->show(mddev, page);
4141 mddev_unlock(mddev);
4142 }
4143 return rv;
4144 }
4145
4146 static ssize_t
4147 md_attr_store(struct kobject *kobj, struct attribute *attr,
4148 const char *page, size_t length)
4149 {
4150 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4151 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4152 ssize_t rv;
4153
4154 if (!entry->store)
4155 return -EIO;
4156 if (!capable(CAP_SYS_ADMIN))
4157 return -EACCES;
4158 rv = mddev_lock(mddev);
4159 if (mddev->hold_active == UNTIL_IOCTL)
4160 mddev->hold_active = 0;
4161 if (!rv) {
4162 rv = entry->store(mddev, page, length);
4163 mddev_unlock(mddev);
4164 }
4165 return rv;
4166 }
4167
4168 static void md_free(struct kobject *ko)
4169 {
4170 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4171
4172 if (mddev->sysfs_state)
4173 sysfs_put(mddev->sysfs_state);
4174
4175 if (mddev->gendisk) {
4176 del_gendisk(mddev->gendisk);
4177 put_disk(mddev->gendisk);
4178 }
4179 if (mddev->queue)
4180 blk_cleanup_queue(mddev->queue);
4181
4182 kfree(mddev);
4183 }
4184
4185 static const struct sysfs_ops md_sysfs_ops = {
4186 .show = md_attr_show,
4187 .store = md_attr_store,
4188 };
4189 static struct kobj_type md_ktype = {
4190 .release = md_free,
4191 .sysfs_ops = &md_sysfs_ops,
4192 .default_attrs = md_default_attrs,
4193 };
4194
4195 int mdp_major = 0;
4196
4197 static void mddev_delayed_delete(struct work_struct *ws)
4198 {
4199 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4200
4201 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4202 kobject_del(&mddev->kobj);
4203 kobject_put(&mddev->kobj);
4204 }
4205
4206 static int md_alloc(dev_t dev, char *name)
4207 {
4208 static DEFINE_MUTEX(disks_mutex);
4209 mddev_t *mddev = mddev_find(dev);
4210 struct gendisk *disk;
4211 int partitioned;
4212 int shift;
4213 int unit;
4214 int error;
4215
4216 if (!mddev)
4217 return -ENODEV;
4218
4219 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4220 shift = partitioned ? MdpMinorShift : 0;
4221 unit = MINOR(mddev->unit) >> shift;
4222
4223 /* wait for any previous instance if this device
4224 * to be completed removed (mddev_delayed_delete).
4225 */
4226 flush_scheduled_work();
4227
4228 mutex_lock(&disks_mutex);
4229 error = -EEXIST;
4230 if (mddev->gendisk)
4231 goto abort;
4232
4233 if (name) {
4234 /* Need to ensure that 'name' is not a duplicate.
4235 */
4236 mddev_t *mddev2;
4237 spin_lock(&all_mddevs_lock);
4238
4239 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4240 if (mddev2->gendisk &&
4241 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4242 spin_unlock(&all_mddevs_lock);
4243 goto abort;
4244 }
4245 spin_unlock(&all_mddevs_lock);
4246 }
4247
4248 error = -ENOMEM;
4249 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4250 if (!mddev->queue)
4251 goto abort;
4252 mddev->queue->queuedata = mddev;
4253
4254 blk_queue_make_request(mddev->queue, md_make_request);
4255
4256 disk = alloc_disk(1 << shift);
4257 if (!disk) {
4258 blk_cleanup_queue(mddev->queue);
4259 mddev->queue = NULL;
4260 goto abort;
4261 }
4262 disk->major = MAJOR(mddev->unit);
4263 disk->first_minor = unit << shift;
4264 if (name)
4265 strcpy(disk->disk_name, name);
4266 else if (partitioned)
4267 sprintf(disk->disk_name, "md_d%d", unit);
4268 else
4269 sprintf(disk->disk_name, "md%d", unit);
4270 disk->fops = &md_fops;
4271 disk->private_data = mddev;
4272 disk->queue = mddev->queue;
4273 /* Allow extended partitions. This makes the
4274 * 'mdp' device redundant, but we can't really
4275 * remove it now.
4276 */
4277 disk->flags |= GENHD_FL_EXT_DEVT;
4278 add_disk(disk);
4279 mddev->gendisk = disk;
4280 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4281 &disk_to_dev(disk)->kobj, "%s", "md");
4282 if (error) {
4283 /* This isn't possible, but as kobject_init_and_add is marked
4284 * __must_check, we must do something with the result
4285 */
4286 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4287 disk->disk_name);
4288 error = 0;
4289 }
4290 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4291 printk(KERN_DEBUG "pointless warning\n");
4292 abort:
4293 mutex_unlock(&disks_mutex);
4294 if (!error) {
4295 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4296 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, NULL, "array_state");
4297 }
4298 mddev_put(mddev);
4299 return error;
4300 }
4301
4302 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4303 {
4304 md_alloc(dev, NULL);
4305 return NULL;
4306 }
4307
4308 static int add_named_array(const char *val, struct kernel_param *kp)
4309 {
4310 /* val must be "md_*" where * is not all digits.
4311 * We allocate an array with a large free minor number, and
4312 * set the name to val. val must not already be an active name.
4313 */
4314 int len = strlen(val);
4315 char buf[DISK_NAME_LEN];
4316
4317 while (len && val[len-1] == '\n')
4318 len--;
4319 if (len >= DISK_NAME_LEN)
4320 return -E2BIG;
4321 strlcpy(buf, val, len+1);
4322 if (strncmp(buf, "md_", 3) != 0)
4323 return -EINVAL;
4324 return md_alloc(0, buf);
4325 }
4326
4327 static void md_safemode_timeout(unsigned long data)
4328 {
4329 mddev_t *mddev = (mddev_t *) data;
4330
4331 if (!atomic_read(&mddev->writes_pending)) {
4332 mddev->safemode = 1;
4333 if (mddev->external)
4334 sysfs_notify_dirent(mddev->sysfs_state);
4335 }
4336 md_wakeup_thread(mddev->thread);
4337 }
4338
4339 static int start_dirty_degraded;
4340
4341 static int md_run(mddev_t *mddev)
4342 {
4343 int err;
4344 mdk_rdev_t *rdev;
4345 struct mdk_personality *pers;
4346
4347 if (list_empty(&mddev->disks))
4348 /* cannot run an array with no devices.. */
4349 return -EINVAL;
4350
4351 if (mddev->pers)
4352 return -EBUSY;
4353 /* Cannot run until previous stop completes properly */
4354 if (mddev->sysfs_active)
4355 return -EBUSY;
4356
4357 /*
4358 * Analyze all RAID superblock(s)
4359 */
4360 if (!mddev->raid_disks) {
4361 if (!mddev->persistent)
4362 return -EINVAL;
4363 analyze_sbs(mddev);
4364 }
4365
4366 if (mddev->level != LEVEL_NONE)
4367 request_module("md-level-%d", mddev->level);
4368 else if (mddev->clevel[0])
4369 request_module("md-%s", mddev->clevel);
4370
4371 /*
4372 * Drop all container device buffers, from now on
4373 * the only valid external interface is through the md
4374 * device.
4375 */
4376 list_for_each_entry(rdev, &mddev->disks, same_set) {
4377 if (test_bit(Faulty, &rdev->flags))
4378 continue;
4379 sync_blockdev(rdev->bdev);
4380 invalidate_bdev(rdev->bdev);
4381
4382 /* perform some consistency tests on the device.
4383 * We don't want the data to overlap the metadata,
4384 * Internal Bitmap issues have been handled elsewhere.
4385 */
4386 if (rdev->data_offset < rdev->sb_start) {
4387 if (mddev->dev_sectors &&
4388 rdev->data_offset + mddev->dev_sectors
4389 > rdev->sb_start) {
4390 printk("md: %s: data overlaps metadata\n",
4391 mdname(mddev));
4392 return -EINVAL;
4393 }
4394 } else {
4395 if (rdev->sb_start + rdev->sb_size/512
4396 > rdev->data_offset) {
4397 printk("md: %s: metadata overlaps data\n",
4398 mdname(mddev));
4399 return -EINVAL;
4400 }
4401 }
4402 sysfs_notify_dirent(rdev->sysfs_state);
4403 }
4404
4405 spin_lock(&pers_lock);
4406 pers = find_pers(mddev->level, mddev->clevel);
4407 if (!pers || !try_module_get(pers->owner)) {
4408 spin_unlock(&pers_lock);
4409 if (mddev->level != LEVEL_NONE)
4410 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4411 mddev->level);
4412 else
4413 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4414 mddev->clevel);
4415 return -EINVAL;
4416 }
4417 mddev->pers = pers;
4418 spin_unlock(&pers_lock);
4419 if (mddev->level != pers->level) {
4420 mddev->level = pers->level;
4421 mddev->new_level = pers->level;
4422 }
4423 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4424
4425 if (mddev->reshape_position != MaxSector &&
4426 pers->start_reshape == NULL) {
4427 /* This personality cannot handle reshaping... */
4428 mddev->pers = NULL;
4429 module_put(pers->owner);
4430 return -EINVAL;
4431 }
4432
4433 if (pers->sync_request) {
4434 /* Warn if this is a potentially silly
4435 * configuration.
4436 */
4437 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4438 mdk_rdev_t *rdev2;
4439 int warned = 0;
4440
4441 list_for_each_entry(rdev, &mddev->disks, same_set)
4442 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4443 if (rdev < rdev2 &&
4444 rdev->bdev->bd_contains ==
4445 rdev2->bdev->bd_contains) {
4446 printk(KERN_WARNING
4447 "%s: WARNING: %s appears to be"
4448 " on the same physical disk as"
4449 " %s.\n",
4450 mdname(mddev),
4451 bdevname(rdev->bdev,b),
4452 bdevname(rdev2->bdev,b2));
4453 warned = 1;
4454 }
4455 }
4456
4457 if (warned)
4458 printk(KERN_WARNING
4459 "True protection against single-disk"
4460 " failure might be compromised.\n");
4461 }
4462
4463 mddev->recovery = 0;
4464 /* may be over-ridden by personality */
4465 mddev->resync_max_sectors = mddev->dev_sectors;
4466
4467 mddev->barriers_work = 1;
4468 mddev->ok_start_degraded = start_dirty_degraded;
4469
4470 if (start_readonly && mddev->ro == 0)
4471 mddev->ro = 2; /* read-only, but switch on first write */
4472
4473 err = mddev->pers->run(mddev);
4474 if (err)
4475 printk(KERN_ERR "md: pers->run() failed ...\n");
4476 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4477 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4478 " but 'external_size' not in effect?\n", __func__);
4479 printk(KERN_ERR
4480 "md: invalid array_size %llu > default size %llu\n",
4481 (unsigned long long)mddev->array_sectors / 2,
4482 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4483 err = -EINVAL;
4484 mddev->pers->stop(mddev);
4485 }
4486 if (err == 0 && mddev->pers->sync_request) {
4487 err = bitmap_create(mddev);
4488 if (err) {
4489 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4490 mdname(mddev), err);
4491 mddev->pers->stop(mddev);
4492 }
4493 }
4494 if (err) {
4495 module_put(mddev->pers->owner);
4496 mddev->pers = NULL;
4497 bitmap_destroy(mddev);
4498 return err;
4499 }
4500 if (mddev->pers->sync_request) {
4501 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4502 printk(KERN_WARNING
4503 "md: cannot register extra attributes for %s\n",
4504 mdname(mddev));
4505 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
4506 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4507 mddev->ro = 0;
4508
4509 atomic_set(&mddev->writes_pending,0);
4510 atomic_set(&mddev->max_corr_read_errors,
4511 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4512 mddev->safemode = 0;
4513 mddev->safemode_timer.function = md_safemode_timeout;
4514 mddev->safemode_timer.data = (unsigned long) mddev;
4515 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4516 mddev->in_sync = 1;
4517
4518 list_for_each_entry(rdev, &mddev->disks, same_set)
4519 if (rdev->raid_disk >= 0) {
4520 char nm[20];
4521 sprintf(nm, "rd%d", rdev->raid_disk);
4522 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4523 printk("md: cannot register %s for %s\n",
4524 nm, mdname(mddev));
4525 }
4526
4527 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4528
4529 if (mddev->flags)
4530 md_update_sb(mddev, 0);
4531
4532 md_wakeup_thread(mddev->thread);
4533 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4534
4535 md_new_event(mddev);
4536 sysfs_notify_dirent(mddev->sysfs_state);
4537 if (mddev->sysfs_action)
4538 sysfs_notify_dirent(mddev->sysfs_action);
4539 sysfs_notify(&mddev->kobj, NULL, "degraded");
4540 return 0;
4541 }
4542
4543 static int do_md_run(mddev_t *mddev)
4544 {
4545 int err;
4546
4547 err = md_run(mddev);
4548 if (err)
4549 goto out;
4550
4551 set_capacity(mddev->gendisk, mddev->array_sectors);
4552 revalidate_disk(mddev->gendisk);
4553 mddev->changed = 1;
4554 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4555 out:
4556 return err;
4557 }
4558
4559 static int restart_array(mddev_t *mddev)
4560 {
4561 struct gendisk *disk = mddev->gendisk;
4562
4563 /* Complain if it has no devices */
4564 if (list_empty(&mddev->disks))
4565 return -ENXIO;
4566 if (!mddev->pers)
4567 return -EINVAL;
4568 if (!mddev->ro)
4569 return -EBUSY;
4570 mddev->safemode = 0;
4571 mddev->ro = 0;
4572 set_disk_ro(disk, 0);
4573 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4574 mdname(mddev));
4575 /* Kick recovery or resync if necessary */
4576 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4577 md_wakeup_thread(mddev->thread);
4578 md_wakeup_thread(mddev->sync_thread);
4579 sysfs_notify_dirent(mddev->sysfs_state);
4580 return 0;
4581 }
4582
4583 /* similar to deny_write_access, but accounts for our holding a reference
4584 * to the file ourselves */
4585 static int deny_bitmap_write_access(struct file * file)
4586 {
4587 struct inode *inode = file->f_mapping->host;
4588
4589 spin_lock(&inode->i_lock);
4590 if (atomic_read(&inode->i_writecount) > 1) {
4591 spin_unlock(&inode->i_lock);
4592 return -ETXTBSY;
4593 }
4594 atomic_set(&inode->i_writecount, -1);
4595 spin_unlock(&inode->i_lock);
4596
4597 return 0;
4598 }
4599
4600 void restore_bitmap_write_access(struct file *file)
4601 {
4602 struct inode *inode = file->f_mapping->host;
4603
4604 spin_lock(&inode->i_lock);
4605 atomic_set(&inode->i_writecount, 1);
4606 spin_unlock(&inode->i_lock);
4607 }
4608
4609 static void md_clean(mddev_t *mddev)
4610 {
4611 mddev->array_sectors = 0;
4612 mddev->external_size = 0;
4613 mddev->dev_sectors = 0;
4614 mddev->raid_disks = 0;
4615 mddev->recovery_cp = 0;
4616 mddev->resync_min = 0;
4617 mddev->resync_max = MaxSector;
4618 mddev->reshape_position = MaxSector;
4619 mddev->external = 0;
4620 mddev->persistent = 0;
4621 mddev->level = LEVEL_NONE;
4622 mddev->clevel[0] = 0;
4623 mddev->flags = 0;
4624 mddev->ro = 0;
4625 mddev->metadata_type[0] = 0;
4626 mddev->chunk_sectors = 0;
4627 mddev->ctime = mddev->utime = 0;
4628 mddev->layout = 0;
4629 mddev->max_disks = 0;
4630 mddev->events = 0;
4631 mddev->can_decrease_events = 0;
4632 mddev->delta_disks = 0;
4633 mddev->new_level = LEVEL_NONE;
4634 mddev->new_layout = 0;
4635 mddev->new_chunk_sectors = 0;
4636 mddev->curr_resync = 0;
4637 mddev->resync_mismatches = 0;
4638 mddev->suspend_lo = mddev->suspend_hi = 0;
4639 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4640 mddev->recovery = 0;
4641 mddev->in_sync = 0;
4642 mddev->changed = 0;
4643 mddev->degraded = 0;
4644 mddev->barriers_work = 0;
4645 mddev->safemode = 0;
4646 mddev->bitmap_info.offset = 0;
4647 mddev->bitmap_info.default_offset = 0;
4648 mddev->bitmap_info.chunksize = 0;
4649 mddev->bitmap_info.daemon_sleep = 0;
4650 mddev->bitmap_info.max_write_behind = 0;
4651 }
4652
4653 static void md_stop_writes(mddev_t *mddev)
4654 {
4655 if (mddev->sync_thread) {
4656 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4657 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4658 md_unregister_thread(mddev->sync_thread);
4659 mddev->sync_thread = NULL;
4660 }
4661
4662 del_timer_sync(&mddev->safemode_timer);
4663
4664 bitmap_flush(mddev);
4665 md_super_wait(mddev);
4666
4667 if (!mddev->in_sync || mddev->flags) {
4668 /* mark array as shutdown cleanly */
4669 mddev->in_sync = 1;
4670 md_update_sb(mddev, 1);
4671 }
4672 }
4673
4674 static void md_stop(mddev_t *mddev)
4675 {
4676 md_stop_writes(mddev);
4677
4678 mddev->pers->stop(mddev);
4679 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4680 mddev->to_remove = &md_redundancy_group;
4681 module_put(mddev->pers->owner);
4682 mddev->pers = NULL;
4683 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4684 }
4685
4686 static int md_set_readonly(mddev_t *mddev, int is_open)
4687 {
4688 int err = 0;
4689 mutex_lock(&mddev->open_mutex);
4690 if (atomic_read(&mddev->openers) > is_open) {
4691 printk("md: %s still in use.\n",mdname(mddev));
4692 err = -EBUSY;
4693 goto out;
4694 }
4695 if (mddev->pers) {
4696 md_stop_writes(mddev);
4697
4698 err = -ENXIO;
4699 if (mddev->ro==1)
4700 goto out;
4701 mddev->ro = 1;
4702 set_disk_ro(mddev->gendisk, 1);
4703 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4704 sysfs_notify_dirent(mddev->sysfs_state);
4705 err = 0;
4706 }
4707 out:
4708 mutex_unlock(&mddev->open_mutex);
4709 return err;
4710 }
4711
4712 /* mode:
4713 * 0 - completely stop and dis-assemble array
4714 * 2 - stop but do not disassemble array
4715 */
4716 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4717 {
4718 int err = 0, revalidate = 0;
4719 struct gendisk *disk = mddev->gendisk;
4720 mdk_rdev_t *rdev;
4721
4722 mutex_lock(&mddev->open_mutex);
4723 if (atomic_read(&mddev->openers) > is_open ||
4724 mddev->sysfs_active) {
4725 printk("md: %s still in use.\n",mdname(mddev));
4726 err = -EBUSY;
4727 } else if (mddev->pers) {
4728
4729 if (mddev->ro)
4730 set_disk_ro(disk, 0);
4731
4732 md_stop(mddev);
4733 mddev->queue->merge_bvec_fn = NULL;
4734 mddev->queue->unplug_fn = NULL;
4735 mddev->queue->backing_dev_info.congested_fn = NULL;
4736
4737 /* tell userspace to handle 'inactive' */
4738 sysfs_notify_dirent(mddev->sysfs_state);
4739
4740 list_for_each_entry(rdev, &mddev->disks, same_set)
4741 if (rdev->raid_disk >= 0) {
4742 char nm[20];
4743 sprintf(nm, "rd%d", rdev->raid_disk);
4744 sysfs_remove_link(&mddev->kobj, nm);
4745 }
4746
4747 set_capacity(disk, 0);
4748 revalidate = 1;
4749 mddev->changed = 1;
4750
4751 if (mddev->ro)
4752 mddev->ro = 0;
4753
4754 err = 0;
4755 }
4756 mutex_unlock(&mddev->open_mutex);
4757 if (revalidate)
4758 revalidate_disk(disk);
4759 if (err)
4760 return err;
4761 /*
4762 * Free resources if final stop
4763 */
4764 if (mode == 0) {
4765
4766 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4767
4768 bitmap_destroy(mddev);
4769 if (mddev->bitmap_info.file) {
4770 restore_bitmap_write_access(mddev->bitmap_info.file);
4771 fput(mddev->bitmap_info.file);
4772 mddev->bitmap_info.file = NULL;
4773 }
4774 mddev->bitmap_info.offset = 0;
4775
4776 export_array(mddev);
4777
4778 md_clean(mddev);
4779 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4780 if (mddev->hold_active == UNTIL_STOP)
4781 mddev->hold_active = 0;
4782
4783 }
4784 err = 0;
4785 blk_integrity_unregister(disk);
4786 md_new_event(mddev);
4787 sysfs_notify_dirent(mddev->sysfs_state);
4788 return err;
4789 }
4790
4791 #ifndef MODULE
4792 static void autorun_array(mddev_t *mddev)
4793 {
4794 mdk_rdev_t *rdev;
4795 int err;
4796
4797 if (list_empty(&mddev->disks))
4798 return;
4799
4800 printk(KERN_INFO "md: running: ");
4801
4802 list_for_each_entry(rdev, &mddev->disks, same_set) {
4803 char b[BDEVNAME_SIZE];
4804 printk("<%s>", bdevname(rdev->bdev,b));
4805 }
4806 printk("\n");
4807
4808 err = do_md_run(mddev);
4809 if (err) {
4810 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4811 do_md_stop(mddev, 0, 0);
4812 }
4813 }
4814
4815 /*
4816 * lets try to run arrays based on all disks that have arrived
4817 * until now. (those are in pending_raid_disks)
4818 *
4819 * the method: pick the first pending disk, collect all disks with
4820 * the same UUID, remove all from the pending list and put them into
4821 * the 'same_array' list. Then order this list based on superblock
4822 * update time (freshest comes first), kick out 'old' disks and
4823 * compare superblocks. If everything's fine then run it.
4824 *
4825 * If "unit" is allocated, then bump its reference count
4826 */
4827 static void autorun_devices(int part)
4828 {
4829 mdk_rdev_t *rdev0, *rdev, *tmp;
4830 mddev_t *mddev;
4831 char b[BDEVNAME_SIZE];
4832
4833 printk(KERN_INFO "md: autorun ...\n");
4834 while (!list_empty(&pending_raid_disks)) {
4835 int unit;
4836 dev_t dev;
4837 LIST_HEAD(candidates);
4838 rdev0 = list_entry(pending_raid_disks.next,
4839 mdk_rdev_t, same_set);
4840
4841 printk(KERN_INFO "md: considering %s ...\n",
4842 bdevname(rdev0->bdev,b));
4843 INIT_LIST_HEAD(&candidates);
4844 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4845 if (super_90_load(rdev, rdev0, 0) >= 0) {
4846 printk(KERN_INFO "md: adding %s ...\n",
4847 bdevname(rdev->bdev,b));
4848 list_move(&rdev->same_set, &candidates);
4849 }
4850 /*
4851 * now we have a set of devices, with all of them having
4852 * mostly sane superblocks. It's time to allocate the
4853 * mddev.
4854 */
4855 if (part) {
4856 dev = MKDEV(mdp_major,
4857 rdev0->preferred_minor << MdpMinorShift);
4858 unit = MINOR(dev) >> MdpMinorShift;
4859 } else {
4860 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4861 unit = MINOR(dev);
4862 }
4863 if (rdev0->preferred_minor != unit) {
4864 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4865 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4866 break;
4867 }
4868
4869 md_probe(dev, NULL, NULL);
4870 mddev = mddev_find(dev);
4871 if (!mddev || !mddev->gendisk) {
4872 if (mddev)
4873 mddev_put(mddev);
4874 printk(KERN_ERR
4875 "md: cannot allocate memory for md drive.\n");
4876 break;
4877 }
4878 if (mddev_lock(mddev))
4879 printk(KERN_WARNING "md: %s locked, cannot run\n",
4880 mdname(mddev));
4881 else if (mddev->raid_disks || mddev->major_version
4882 || !list_empty(&mddev->disks)) {
4883 printk(KERN_WARNING
4884 "md: %s already running, cannot run %s\n",
4885 mdname(mddev), bdevname(rdev0->bdev,b));
4886 mddev_unlock(mddev);
4887 } else {
4888 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4889 mddev->persistent = 1;
4890 rdev_for_each_list(rdev, tmp, &candidates) {
4891 list_del_init(&rdev->same_set);
4892 if (bind_rdev_to_array(rdev, mddev))
4893 export_rdev(rdev);
4894 }
4895 autorun_array(mddev);
4896 mddev_unlock(mddev);
4897 }
4898 /* on success, candidates will be empty, on error
4899 * it won't...
4900 */
4901 rdev_for_each_list(rdev, tmp, &candidates) {
4902 list_del_init(&rdev->same_set);
4903 export_rdev(rdev);
4904 }
4905 mddev_put(mddev);
4906 }
4907 printk(KERN_INFO "md: ... autorun DONE.\n");
4908 }
4909 #endif /* !MODULE */
4910
4911 static int get_version(void __user * arg)
4912 {
4913 mdu_version_t ver;
4914
4915 ver.major = MD_MAJOR_VERSION;
4916 ver.minor = MD_MINOR_VERSION;
4917 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4918
4919 if (copy_to_user(arg, &ver, sizeof(ver)))
4920 return -EFAULT;
4921
4922 return 0;
4923 }
4924
4925 static int get_array_info(mddev_t * mddev, void __user * arg)
4926 {
4927 mdu_array_info_t info;
4928 int nr,working,insync,failed,spare;
4929 mdk_rdev_t *rdev;
4930
4931 nr=working=insync=failed=spare=0;
4932 list_for_each_entry(rdev, &mddev->disks, same_set) {
4933 nr++;
4934 if (test_bit(Faulty, &rdev->flags))
4935 failed++;
4936 else {
4937 working++;
4938 if (test_bit(In_sync, &rdev->flags))
4939 insync++;
4940 else
4941 spare++;
4942 }
4943 }
4944
4945 info.major_version = mddev->major_version;
4946 info.minor_version = mddev->minor_version;
4947 info.patch_version = MD_PATCHLEVEL_VERSION;
4948 info.ctime = mddev->ctime;
4949 info.level = mddev->level;
4950 info.size = mddev->dev_sectors / 2;
4951 if (info.size != mddev->dev_sectors / 2) /* overflow */
4952 info.size = -1;
4953 info.nr_disks = nr;
4954 info.raid_disks = mddev->raid_disks;
4955 info.md_minor = mddev->md_minor;
4956 info.not_persistent= !mddev->persistent;
4957
4958 info.utime = mddev->utime;
4959 info.state = 0;
4960 if (mddev->in_sync)
4961 info.state = (1<<MD_SB_CLEAN);
4962 if (mddev->bitmap && mddev->bitmap_info.offset)
4963 info.state = (1<<MD_SB_BITMAP_PRESENT);
4964 info.active_disks = insync;
4965 info.working_disks = working;
4966 info.failed_disks = failed;
4967 info.spare_disks = spare;
4968
4969 info.layout = mddev->layout;
4970 info.chunk_size = mddev->chunk_sectors << 9;
4971
4972 if (copy_to_user(arg, &info, sizeof(info)))
4973 return -EFAULT;
4974
4975 return 0;
4976 }
4977
4978 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4979 {
4980 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4981 char *ptr, *buf = NULL;
4982 int err = -ENOMEM;
4983
4984 if (md_allow_write(mddev))
4985 file = kmalloc(sizeof(*file), GFP_NOIO);
4986 else
4987 file = kmalloc(sizeof(*file), GFP_KERNEL);
4988
4989 if (!file)
4990 goto out;
4991
4992 /* bitmap disabled, zero the first byte and copy out */
4993 if (!mddev->bitmap || !mddev->bitmap->file) {
4994 file->pathname[0] = '\0';
4995 goto copy_out;
4996 }
4997
4998 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4999 if (!buf)
5000 goto out;
5001
5002 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5003 if (IS_ERR(ptr))
5004 goto out;
5005
5006 strcpy(file->pathname, ptr);
5007
5008 copy_out:
5009 err = 0;
5010 if (copy_to_user(arg, file, sizeof(*file)))
5011 err = -EFAULT;
5012 out:
5013 kfree(buf);
5014 kfree(file);
5015 return err;
5016 }
5017
5018 static int get_disk_info(mddev_t * mddev, void __user * arg)
5019 {
5020 mdu_disk_info_t info;
5021 mdk_rdev_t *rdev;
5022
5023 if (copy_from_user(&info, arg, sizeof(info)))
5024 return -EFAULT;
5025
5026 rdev = find_rdev_nr(mddev, info.number);
5027 if (rdev) {
5028 info.major = MAJOR(rdev->bdev->bd_dev);
5029 info.minor = MINOR(rdev->bdev->bd_dev);
5030 info.raid_disk = rdev->raid_disk;
5031 info.state = 0;
5032 if (test_bit(Faulty, &rdev->flags))
5033 info.state |= (1<<MD_DISK_FAULTY);
5034 else if (test_bit(In_sync, &rdev->flags)) {
5035 info.state |= (1<<MD_DISK_ACTIVE);
5036 info.state |= (1<<MD_DISK_SYNC);
5037 }
5038 if (test_bit(WriteMostly, &rdev->flags))
5039 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5040 } else {
5041 info.major = info.minor = 0;
5042 info.raid_disk = -1;
5043 info.state = (1<<MD_DISK_REMOVED);
5044 }
5045
5046 if (copy_to_user(arg, &info, sizeof(info)))
5047 return -EFAULT;
5048
5049 return 0;
5050 }
5051
5052 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5053 {
5054 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5055 mdk_rdev_t *rdev;
5056 dev_t dev = MKDEV(info->major,info->minor);
5057
5058 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5059 return -EOVERFLOW;
5060
5061 if (!mddev->raid_disks) {
5062 int err;
5063 /* expecting a device which has a superblock */
5064 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5065 if (IS_ERR(rdev)) {
5066 printk(KERN_WARNING
5067 "md: md_import_device returned %ld\n",
5068 PTR_ERR(rdev));
5069 return PTR_ERR(rdev);
5070 }
5071 if (!list_empty(&mddev->disks)) {
5072 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5073 mdk_rdev_t, same_set);
5074 err = super_types[mddev->major_version]
5075 .load_super(rdev, rdev0, mddev->minor_version);
5076 if (err < 0) {
5077 printk(KERN_WARNING
5078 "md: %s has different UUID to %s\n",
5079 bdevname(rdev->bdev,b),
5080 bdevname(rdev0->bdev,b2));
5081 export_rdev(rdev);
5082 return -EINVAL;
5083 }
5084 }
5085 err = bind_rdev_to_array(rdev, mddev);
5086 if (err)
5087 export_rdev(rdev);
5088 return err;
5089 }
5090
5091 /*
5092 * add_new_disk can be used once the array is assembled
5093 * to add "hot spares". They must already have a superblock
5094 * written
5095 */
5096 if (mddev->pers) {
5097 int err;
5098 if (!mddev->pers->hot_add_disk) {
5099 printk(KERN_WARNING
5100 "%s: personality does not support diskops!\n",
5101 mdname(mddev));
5102 return -EINVAL;
5103 }
5104 if (mddev->persistent)
5105 rdev = md_import_device(dev, mddev->major_version,
5106 mddev->minor_version);
5107 else
5108 rdev = md_import_device(dev, -1, -1);
5109 if (IS_ERR(rdev)) {
5110 printk(KERN_WARNING
5111 "md: md_import_device returned %ld\n",
5112 PTR_ERR(rdev));
5113 return PTR_ERR(rdev);
5114 }
5115 /* set saved_raid_disk if appropriate */
5116 if (!mddev->persistent) {
5117 if (info->state & (1<<MD_DISK_SYNC) &&
5118 info->raid_disk < mddev->raid_disks) {
5119 rdev->raid_disk = info->raid_disk;
5120 set_bit(In_sync, &rdev->flags);
5121 } else
5122 rdev->raid_disk = -1;
5123 } else
5124 super_types[mddev->major_version].
5125 validate_super(mddev, rdev);
5126 if (test_bit(In_sync, &rdev->flags))
5127 rdev->saved_raid_disk = rdev->raid_disk;
5128 else
5129 rdev->saved_raid_disk = -1;
5130
5131 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5132 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5133 set_bit(WriteMostly, &rdev->flags);
5134 else
5135 clear_bit(WriteMostly, &rdev->flags);
5136
5137 rdev->raid_disk = -1;
5138 err = bind_rdev_to_array(rdev, mddev);
5139 if (!err && !mddev->pers->hot_remove_disk) {
5140 /* If there is hot_add_disk but no hot_remove_disk
5141 * then added disks for geometry changes,
5142 * and should be added immediately.
5143 */
5144 super_types[mddev->major_version].
5145 validate_super(mddev, rdev);
5146 err = mddev->pers->hot_add_disk(mddev, rdev);
5147 if (err)
5148 unbind_rdev_from_array(rdev);
5149 }
5150 if (err)
5151 export_rdev(rdev);
5152 else
5153 sysfs_notify_dirent(rdev->sysfs_state);
5154
5155 md_update_sb(mddev, 1);
5156 if (mddev->degraded)
5157 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5158 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5159 md_wakeup_thread(mddev->thread);
5160 return err;
5161 }
5162
5163 /* otherwise, add_new_disk is only allowed
5164 * for major_version==0 superblocks
5165 */
5166 if (mddev->major_version != 0) {
5167 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5168 mdname(mddev));
5169 return -EINVAL;
5170 }
5171
5172 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5173 int err;
5174 rdev = md_import_device(dev, -1, 0);
5175 if (IS_ERR(rdev)) {
5176 printk(KERN_WARNING
5177 "md: error, md_import_device() returned %ld\n",
5178 PTR_ERR(rdev));
5179 return PTR_ERR(rdev);
5180 }
5181 rdev->desc_nr = info->number;
5182 if (info->raid_disk < mddev->raid_disks)
5183 rdev->raid_disk = info->raid_disk;
5184 else
5185 rdev->raid_disk = -1;
5186
5187 if (rdev->raid_disk < mddev->raid_disks)
5188 if (info->state & (1<<MD_DISK_SYNC))
5189 set_bit(In_sync, &rdev->flags);
5190
5191 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5192 set_bit(WriteMostly, &rdev->flags);
5193
5194 if (!mddev->persistent) {
5195 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5196 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5197 } else
5198 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5199 rdev->sectors = rdev->sb_start;
5200
5201 err = bind_rdev_to_array(rdev, mddev);
5202 if (err) {
5203 export_rdev(rdev);
5204 return err;
5205 }
5206 }
5207
5208 return 0;
5209 }
5210
5211 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5212 {
5213 char b[BDEVNAME_SIZE];
5214 mdk_rdev_t *rdev;
5215
5216 rdev = find_rdev(mddev, dev);
5217 if (!rdev)
5218 return -ENXIO;
5219
5220 if (rdev->raid_disk >= 0)
5221 goto busy;
5222
5223 kick_rdev_from_array(rdev);
5224 md_update_sb(mddev, 1);
5225 md_new_event(mddev);
5226
5227 return 0;
5228 busy:
5229 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5230 bdevname(rdev->bdev,b), mdname(mddev));
5231 return -EBUSY;
5232 }
5233
5234 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5235 {
5236 char b[BDEVNAME_SIZE];
5237 int err;
5238 mdk_rdev_t *rdev;
5239
5240 if (!mddev->pers)
5241 return -ENODEV;
5242
5243 if (mddev->major_version != 0) {
5244 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5245 " version-0 superblocks.\n",
5246 mdname(mddev));
5247 return -EINVAL;
5248 }
5249 if (!mddev->pers->hot_add_disk) {
5250 printk(KERN_WARNING
5251 "%s: personality does not support diskops!\n",
5252 mdname(mddev));
5253 return -EINVAL;
5254 }
5255
5256 rdev = md_import_device(dev, -1, 0);
5257 if (IS_ERR(rdev)) {
5258 printk(KERN_WARNING
5259 "md: error, md_import_device() returned %ld\n",
5260 PTR_ERR(rdev));
5261 return -EINVAL;
5262 }
5263
5264 if (mddev->persistent)
5265 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5266 else
5267 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5268
5269 rdev->sectors = rdev->sb_start;
5270
5271 if (test_bit(Faulty, &rdev->flags)) {
5272 printk(KERN_WARNING
5273 "md: can not hot-add faulty %s disk to %s!\n",
5274 bdevname(rdev->bdev,b), mdname(mddev));
5275 err = -EINVAL;
5276 goto abort_export;
5277 }
5278 clear_bit(In_sync, &rdev->flags);
5279 rdev->desc_nr = -1;
5280 rdev->saved_raid_disk = -1;
5281 err = bind_rdev_to_array(rdev, mddev);
5282 if (err)
5283 goto abort_export;
5284
5285 /*
5286 * The rest should better be atomic, we can have disk failures
5287 * noticed in interrupt contexts ...
5288 */
5289
5290 rdev->raid_disk = -1;
5291
5292 md_update_sb(mddev, 1);
5293
5294 /*
5295 * Kick recovery, maybe this spare has to be added to the
5296 * array immediately.
5297 */
5298 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5299 md_wakeup_thread(mddev->thread);
5300 md_new_event(mddev);
5301 return 0;
5302
5303 abort_export:
5304 export_rdev(rdev);
5305 return err;
5306 }
5307
5308 static int set_bitmap_file(mddev_t *mddev, int fd)
5309 {
5310 int err;
5311
5312 if (mddev->pers) {
5313 if (!mddev->pers->quiesce)
5314 return -EBUSY;
5315 if (mddev->recovery || mddev->sync_thread)
5316 return -EBUSY;
5317 /* we should be able to change the bitmap.. */
5318 }
5319
5320
5321 if (fd >= 0) {
5322 if (mddev->bitmap)
5323 return -EEXIST; /* cannot add when bitmap is present */
5324 mddev->bitmap_info.file = fget(fd);
5325
5326 if (mddev->bitmap_info.file == NULL) {
5327 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5328 mdname(mddev));
5329 return -EBADF;
5330 }
5331
5332 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5333 if (err) {
5334 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5335 mdname(mddev));
5336 fput(mddev->bitmap_info.file);
5337 mddev->bitmap_info.file = NULL;
5338 return err;
5339 }
5340 mddev->bitmap_info.offset = 0; /* file overrides offset */
5341 } else if (mddev->bitmap == NULL)
5342 return -ENOENT; /* cannot remove what isn't there */
5343 err = 0;
5344 if (mddev->pers) {
5345 mddev->pers->quiesce(mddev, 1);
5346 if (fd >= 0)
5347 err = bitmap_create(mddev);
5348 if (fd < 0 || err) {
5349 bitmap_destroy(mddev);
5350 fd = -1; /* make sure to put the file */
5351 }
5352 mddev->pers->quiesce(mddev, 0);
5353 }
5354 if (fd < 0) {
5355 if (mddev->bitmap_info.file) {
5356 restore_bitmap_write_access(mddev->bitmap_info.file);
5357 fput(mddev->bitmap_info.file);
5358 }
5359 mddev->bitmap_info.file = NULL;
5360 }
5361
5362 return err;
5363 }
5364
5365 /*
5366 * set_array_info is used two different ways
5367 * The original usage is when creating a new array.
5368 * In this usage, raid_disks is > 0 and it together with
5369 * level, size, not_persistent,layout,chunksize determine the
5370 * shape of the array.
5371 * This will always create an array with a type-0.90.0 superblock.
5372 * The newer usage is when assembling an array.
5373 * In this case raid_disks will be 0, and the major_version field is
5374 * use to determine which style super-blocks are to be found on the devices.
5375 * The minor and patch _version numbers are also kept incase the
5376 * super_block handler wishes to interpret them.
5377 */
5378 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5379 {
5380
5381 if (info->raid_disks == 0) {
5382 /* just setting version number for superblock loading */
5383 if (info->major_version < 0 ||
5384 info->major_version >= ARRAY_SIZE(super_types) ||
5385 super_types[info->major_version].name == NULL) {
5386 /* maybe try to auto-load a module? */
5387 printk(KERN_INFO
5388 "md: superblock version %d not known\n",
5389 info->major_version);
5390 return -EINVAL;
5391 }
5392 mddev->major_version = info->major_version;
5393 mddev->minor_version = info->minor_version;
5394 mddev->patch_version = info->patch_version;
5395 mddev->persistent = !info->not_persistent;
5396 /* ensure mddev_put doesn't delete this now that there
5397 * is some minimal configuration.
5398 */
5399 mddev->ctime = get_seconds();
5400 return 0;
5401 }
5402 mddev->major_version = MD_MAJOR_VERSION;
5403 mddev->minor_version = MD_MINOR_VERSION;
5404 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5405 mddev->ctime = get_seconds();
5406
5407 mddev->level = info->level;
5408 mddev->clevel[0] = 0;
5409 mddev->dev_sectors = 2 * (sector_t)info->size;
5410 mddev->raid_disks = info->raid_disks;
5411 /* don't set md_minor, it is determined by which /dev/md* was
5412 * openned
5413 */
5414 if (info->state & (1<<MD_SB_CLEAN))
5415 mddev->recovery_cp = MaxSector;
5416 else
5417 mddev->recovery_cp = 0;
5418 mddev->persistent = ! info->not_persistent;
5419 mddev->external = 0;
5420
5421 mddev->layout = info->layout;
5422 mddev->chunk_sectors = info->chunk_size >> 9;
5423
5424 mddev->max_disks = MD_SB_DISKS;
5425
5426 if (mddev->persistent)
5427 mddev->flags = 0;
5428 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5429
5430 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5431 mddev->bitmap_info.offset = 0;
5432
5433 mddev->reshape_position = MaxSector;
5434
5435 /*
5436 * Generate a 128 bit UUID
5437 */
5438 get_random_bytes(mddev->uuid, 16);
5439
5440 mddev->new_level = mddev->level;
5441 mddev->new_chunk_sectors = mddev->chunk_sectors;
5442 mddev->new_layout = mddev->layout;
5443 mddev->delta_disks = 0;
5444
5445 return 0;
5446 }
5447
5448 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5449 {
5450 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5451
5452 if (mddev->external_size)
5453 return;
5454
5455 mddev->array_sectors = array_sectors;
5456 }
5457 EXPORT_SYMBOL(md_set_array_sectors);
5458
5459 static int update_size(mddev_t *mddev, sector_t num_sectors)
5460 {
5461 mdk_rdev_t *rdev;
5462 int rv;
5463 int fit = (num_sectors == 0);
5464
5465 if (mddev->pers->resize == NULL)
5466 return -EINVAL;
5467 /* The "num_sectors" is the number of sectors of each device that
5468 * is used. This can only make sense for arrays with redundancy.
5469 * linear and raid0 always use whatever space is available. We can only
5470 * consider changing this number if no resync or reconstruction is
5471 * happening, and if the new size is acceptable. It must fit before the
5472 * sb_start or, if that is <data_offset, it must fit before the size
5473 * of each device. If num_sectors is zero, we find the largest size
5474 * that fits.
5475
5476 */
5477 if (mddev->sync_thread)
5478 return -EBUSY;
5479 if (mddev->bitmap)
5480 /* Sorry, cannot grow a bitmap yet, just remove it,
5481 * grow, and re-add.
5482 */
5483 return -EBUSY;
5484 list_for_each_entry(rdev, &mddev->disks, same_set) {
5485 sector_t avail = rdev->sectors;
5486
5487 if (fit && (num_sectors == 0 || num_sectors > avail))
5488 num_sectors = avail;
5489 if (avail < num_sectors)
5490 return -ENOSPC;
5491 }
5492 rv = mddev->pers->resize(mddev, num_sectors);
5493 if (!rv)
5494 revalidate_disk(mddev->gendisk);
5495 return rv;
5496 }
5497
5498 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5499 {
5500 int rv;
5501 /* change the number of raid disks */
5502 if (mddev->pers->check_reshape == NULL)
5503 return -EINVAL;
5504 if (raid_disks <= 0 ||
5505 (mddev->max_disks && raid_disks >= mddev->max_disks))
5506 return -EINVAL;
5507 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5508 return -EBUSY;
5509 mddev->delta_disks = raid_disks - mddev->raid_disks;
5510
5511 rv = mddev->pers->check_reshape(mddev);
5512 return rv;
5513 }
5514
5515
5516 /*
5517 * update_array_info is used to change the configuration of an
5518 * on-line array.
5519 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5520 * fields in the info are checked against the array.
5521 * Any differences that cannot be handled will cause an error.
5522 * Normally, only one change can be managed at a time.
5523 */
5524 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5525 {
5526 int rv = 0;
5527 int cnt = 0;
5528 int state = 0;
5529
5530 /* calculate expected state,ignoring low bits */
5531 if (mddev->bitmap && mddev->bitmap_info.offset)
5532 state |= (1 << MD_SB_BITMAP_PRESENT);
5533
5534 if (mddev->major_version != info->major_version ||
5535 mddev->minor_version != info->minor_version ||
5536 /* mddev->patch_version != info->patch_version || */
5537 mddev->ctime != info->ctime ||
5538 mddev->level != info->level ||
5539 /* mddev->layout != info->layout || */
5540 !mddev->persistent != info->not_persistent||
5541 mddev->chunk_sectors != info->chunk_size >> 9 ||
5542 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5543 ((state^info->state) & 0xfffffe00)
5544 )
5545 return -EINVAL;
5546 /* Check there is only one change */
5547 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5548 cnt++;
5549 if (mddev->raid_disks != info->raid_disks)
5550 cnt++;
5551 if (mddev->layout != info->layout)
5552 cnt++;
5553 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5554 cnt++;
5555 if (cnt == 0)
5556 return 0;
5557 if (cnt > 1)
5558 return -EINVAL;
5559
5560 if (mddev->layout != info->layout) {
5561 /* Change layout
5562 * we don't need to do anything at the md level, the
5563 * personality will take care of it all.
5564 */
5565 if (mddev->pers->check_reshape == NULL)
5566 return -EINVAL;
5567 else {
5568 mddev->new_layout = info->layout;
5569 rv = mddev->pers->check_reshape(mddev);
5570 if (rv)
5571 mddev->new_layout = mddev->layout;
5572 return rv;
5573 }
5574 }
5575 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5576 rv = update_size(mddev, (sector_t)info->size * 2);
5577
5578 if (mddev->raid_disks != info->raid_disks)
5579 rv = update_raid_disks(mddev, info->raid_disks);
5580
5581 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5582 if (mddev->pers->quiesce == NULL)
5583 return -EINVAL;
5584 if (mddev->recovery || mddev->sync_thread)
5585 return -EBUSY;
5586 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5587 /* add the bitmap */
5588 if (mddev->bitmap)
5589 return -EEXIST;
5590 if (mddev->bitmap_info.default_offset == 0)
5591 return -EINVAL;
5592 mddev->bitmap_info.offset =
5593 mddev->bitmap_info.default_offset;
5594 mddev->pers->quiesce(mddev, 1);
5595 rv = bitmap_create(mddev);
5596 if (rv)
5597 bitmap_destroy(mddev);
5598 mddev->pers->quiesce(mddev, 0);
5599 } else {
5600 /* remove the bitmap */
5601 if (!mddev->bitmap)
5602 return -ENOENT;
5603 if (mddev->bitmap->file)
5604 return -EINVAL;
5605 mddev->pers->quiesce(mddev, 1);
5606 bitmap_destroy(mddev);
5607 mddev->pers->quiesce(mddev, 0);
5608 mddev->bitmap_info.offset = 0;
5609 }
5610 }
5611 md_update_sb(mddev, 1);
5612 return rv;
5613 }
5614
5615 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5616 {
5617 mdk_rdev_t *rdev;
5618
5619 if (mddev->pers == NULL)
5620 return -ENODEV;
5621
5622 rdev = find_rdev(mddev, dev);
5623 if (!rdev)
5624 return -ENODEV;
5625
5626 md_error(mddev, rdev);
5627 return 0;
5628 }
5629
5630 /*
5631 * We have a problem here : there is no easy way to give a CHS
5632 * virtual geometry. We currently pretend that we have a 2 heads
5633 * 4 sectors (with a BIG number of cylinders...). This drives
5634 * dosfs just mad... ;-)
5635 */
5636 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5637 {
5638 mddev_t *mddev = bdev->bd_disk->private_data;
5639
5640 geo->heads = 2;
5641 geo->sectors = 4;
5642 geo->cylinders = mddev->array_sectors / 8;
5643 return 0;
5644 }
5645
5646 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5647 unsigned int cmd, unsigned long arg)
5648 {
5649 int err = 0;
5650 void __user *argp = (void __user *)arg;
5651 mddev_t *mddev = NULL;
5652 int ro;
5653
5654 if (!capable(CAP_SYS_ADMIN))
5655 return -EACCES;
5656
5657 /*
5658 * Commands dealing with the RAID driver but not any
5659 * particular array:
5660 */
5661 switch (cmd)
5662 {
5663 case RAID_VERSION:
5664 err = get_version(argp);
5665 goto done;
5666
5667 case PRINT_RAID_DEBUG:
5668 err = 0;
5669 md_print_devices();
5670 goto done;
5671
5672 #ifndef MODULE
5673 case RAID_AUTORUN:
5674 err = 0;
5675 autostart_arrays(arg);
5676 goto done;
5677 #endif
5678 default:;
5679 }
5680
5681 /*
5682 * Commands creating/starting a new array:
5683 */
5684
5685 mddev = bdev->bd_disk->private_data;
5686
5687 if (!mddev) {
5688 BUG();
5689 goto abort;
5690 }
5691
5692 err = mddev_lock(mddev);
5693 if (err) {
5694 printk(KERN_INFO
5695 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5696 err, cmd);
5697 goto abort;
5698 }
5699
5700 switch (cmd)
5701 {
5702 case SET_ARRAY_INFO:
5703 {
5704 mdu_array_info_t info;
5705 if (!arg)
5706 memset(&info, 0, sizeof(info));
5707 else if (copy_from_user(&info, argp, sizeof(info))) {
5708 err = -EFAULT;
5709 goto abort_unlock;
5710 }
5711 if (mddev->pers) {
5712 err = update_array_info(mddev, &info);
5713 if (err) {
5714 printk(KERN_WARNING "md: couldn't update"
5715 " array info. %d\n", err);
5716 goto abort_unlock;
5717 }
5718 goto done_unlock;
5719 }
5720 if (!list_empty(&mddev->disks)) {
5721 printk(KERN_WARNING
5722 "md: array %s already has disks!\n",
5723 mdname(mddev));
5724 err = -EBUSY;
5725 goto abort_unlock;
5726 }
5727 if (mddev->raid_disks) {
5728 printk(KERN_WARNING
5729 "md: array %s already initialised!\n",
5730 mdname(mddev));
5731 err = -EBUSY;
5732 goto abort_unlock;
5733 }
5734 err = set_array_info(mddev, &info);
5735 if (err) {
5736 printk(KERN_WARNING "md: couldn't set"
5737 " array info. %d\n", err);
5738 goto abort_unlock;
5739 }
5740 }
5741 goto done_unlock;
5742
5743 default:;
5744 }
5745
5746 /*
5747 * Commands querying/configuring an existing array:
5748 */
5749 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5750 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5751 if ((!mddev->raid_disks && !mddev->external)
5752 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5753 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5754 && cmd != GET_BITMAP_FILE) {
5755 err = -ENODEV;
5756 goto abort_unlock;
5757 }
5758
5759 /*
5760 * Commands even a read-only array can execute:
5761 */
5762 switch (cmd)
5763 {
5764 case GET_ARRAY_INFO:
5765 err = get_array_info(mddev, argp);
5766 goto done_unlock;
5767
5768 case GET_BITMAP_FILE:
5769 err = get_bitmap_file(mddev, argp);
5770 goto done_unlock;
5771
5772 case GET_DISK_INFO:
5773 err = get_disk_info(mddev, argp);
5774 goto done_unlock;
5775
5776 case RESTART_ARRAY_RW:
5777 err = restart_array(mddev);
5778 goto done_unlock;
5779
5780 case STOP_ARRAY:
5781 err = do_md_stop(mddev, 0, 1);
5782 goto done_unlock;
5783
5784 case STOP_ARRAY_RO:
5785 err = md_set_readonly(mddev, 1);
5786 goto done_unlock;
5787
5788 case BLKROSET:
5789 if (get_user(ro, (int __user *)(arg))) {
5790 err = -EFAULT;
5791 goto done_unlock;
5792 }
5793 err = -EINVAL;
5794
5795 /* if the bdev is going readonly the value of mddev->ro
5796 * does not matter, no writes are coming
5797 */
5798 if (ro)
5799 goto done_unlock;
5800
5801 /* are we are already prepared for writes? */
5802 if (mddev->ro != 1)
5803 goto done_unlock;
5804
5805 /* transitioning to readauto need only happen for
5806 * arrays that call md_write_start
5807 */
5808 if (mddev->pers) {
5809 err = restart_array(mddev);
5810 if (err == 0) {
5811 mddev->ro = 2;
5812 set_disk_ro(mddev->gendisk, 0);
5813 }
5814 }
5815 goto done_unlock;
5816 }
5817
5818 /*
5819 * The remaining ioctls are changing the state of the
5820 * superblock, so we do not allow them on read-only arrays.
5821 * However non-MD ioctls (e.g. get-size) will still come through
5822 * here and hit the 'default' below, so only disallow
5823 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5824 */
5825 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5826 if (mddev->ro == 2) {
5827 mddev->ro = 0;
5828 sysfs_notify_dirent(mddev->sysfs_state);
5829 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5830 md_wakeup_thread(mddev->thread);
5831 } else {
5832 err = -EROFS;
5833 goto abort_unlock;
5834 }
5835 }
5836
5837 switch (cmd)
5838 {
5839 case ADD_NEW_DISK:
5840 {
5841 mdu_disk_info_t info;
5842 if (copy_from_user(&info, argp, sizeof(info)))
5843 err = -EFAULT;
5844 else
5845 err = add_new_disk(mddev, &info);
5846 goto done_unlock;
5847 }
5848
5849 case HOT_REMOVE_DISK:
5850 err = hot_remove_disk(mddev, new_decode_dev(arg));
5851 goto done_unlock;
5852
5853 case HOT_ADD_DISK:
5854 err = hot_add_disk(mddev, new_decode_dev(arg));
5855 goto done_unlock;
5856
5857 case SET_DISK_FAULTY:
5858 err = set_disk_faulty(mddev, new_decode_dev(arg));
5859 goto done_unlock;
5860
5861 case RUN_ARRAY:
5862 err = do_md_run(mddev);
5863 goto done_unlock;
5864
5865 case SET_BITMAP_FILE:
5866 err = set_bitmap_file(mddev, (int)arg);
5867 goto done_unlock;
5868
5869 default:
5870 err = -EINVAL;
5871 goto abort_unlock;
5872 }
5873
5874 done_unlock:
5875 abort_unlock:
5876 if (mddev->hold_active == UNTIL_IOCTL &&
5877 err != -EINVAL)
5878 mddev->hold_active = 0;
5879 mddev_unlock(mddev);
5880
5881 return err;
5882 done:
5883 if (err)
5884 MD_BUG();
5885 abort:
5886 return err;
5887 }
5888 #ifdef CONFIG_COMPAT
5889 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5890 unsigned int cmd, unsigned long arg)
5891 {
5892 switch (cmd) {
5893 case HOT_REMOVE_DISK:
5894 case HOT_ADD_DISK:
5895 case SET_DISK_FAULTY:
5896 case SET_BITMAP_FILE:
5897 /* These take in integer arg, do not convert */
5898 break;
5899 default:
5900 arg = (unsigned long)compat_ptr(arg);
5901 break;
5902 }
5903
5904 return md_ioctl(bdev, mode, cmd, arg);
5905 }
5906 #endif /* CONFIG_COMPAT */
5907
5908 static int md_open(struct block_device *bdev, fmode_t mode)
5909 {
5910 /*
5911 * Succeed if we can lock the mddev, which confirms that
5912 * it isn't being stopped right now.
5913 */
5914 mddev_t *mddev = mddev_find(bdev->bd_dev);
5915 int err;
5916
5917 if (mddev->gendisk != bdev->bd_disk) {
5918 /* we are racing with mddev_put which is discarding this
5919 * bd_disk.
5920 */
5921 mddev_put(mddev);
5922 /* Wait until bdev->bd_disk is definitely gone */
5923 flush_scheduled_work();
5924 /* Then retry the open from the top */
5925 return -ERESTARTSYS;
5926 }
5927 BUG_ON(mddev != bdev->bd_disk->private_data);
5928
5929 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5930 goto out;
5931
5932 err = 0;
5933 atomic_inc(&mddev->openers);
5934 mutex_unlock(&mddev->open_mutex);
5935
5936 check_disk_change(bdev);
5937 out:
5938 return err;
5939 }
5940
5941 static int md_release(struct gendisk *disk, fmode_t mode)
5942 {
5943 mddev_t *mddev = disk->private_data;
5944
5945 BUG_ON(!mddev);
5946 atomic_dec(&mddev->openers);
5947 mddev_put(mddev);
5948
5949 return 0;
5950 }
5951
5952 static int md_media_changed(struct gendisk *disk)
5953 {
5954 mddev_t *mddev = disk->private_data;
5955
5956 return mddev->changed;
5957 }
5958
5959 static int md_revalidate(struct gendisk *disk)
5960 {
5961 mddev_t *mddev = disk->private_data;
5962
5963 mddev->changed = 0;
5964 return 0;
5965 }
5966 static const struct block_device_operations md_fops =
5967 {
5968 .owner = THIS_MODULE,
5969 .open = md_open,
5970 .release = md_release,
5971 .ioctl = md_ioctl,
5972 #ifdef CONFIG_COMPAT
5973 .compat_ioctl = md_compat_ioctl,
5974 #endif
5975 .getgeo = md_getgeo,
5976 .media_changed = md_media_changed,
5977 .revalidate_disk= md_revalidate,
5978 };
5979
5980 static int md_thread(void * arg)
5981 {
5982 mdk_thread_t *thread = arg;
5983
5984 /*
5985 * md_thread is a 'system-thread', it's priority should be very
5986 * high. We avoid resource deadlocks individually in each
5987 * raid personality. (RAID5 does preallocation) We also use RR and
5988 * the very same RT priority as kswapd, thus we will never get
5989 * into a priority inversion deadlock.
5990 *
5991 * we definitely have to have equal or higher priority than
5992 * bdflush, otherwise bdflush will deadlock if there are too
5993 * many dirty RAID5 blocks.
5994 */
5995
5996 allow_signal(SIGKILL);
5997 while (!kthread_should_stop()) {
5998
5999 /* We need to wait INTERRUPTIBLE so that
6000 * we don't add to the load-average.
6001 * That means we need to be sure no signals are
6002 * pending
6003 */
6004 if (signal_pending(current))
6005 flush_signals(current);
6006
6007 wait_event_interruptible_timeout
6008 (thread->wqueue,
6009 test_bit(THREAD_WAKEUP, &thread->flags)
6010 || kthread_should_stop(),
6011 thread->timeout);
6012
6013 if (test_and_clear_bit(THREAD_WAKEUP, &thread->flags))
6014 thread->run(thread->mddev);
6015 }
6016
6017 return 0;
6018 }
6019
6020 void md_wakeup_thread(mdk_thread_t *thread)
6021 {
6022 if (thread) {
6023 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6024 set_bit(THREAD_WAKEUP, &thread->flags);
6025 wake_up(&thread->wqueue);
6026 }
6027 }
6028
6029 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6030 const char *name)
6031 {
6032 mdk_thread_t *thread;
6033
6034 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6035 if (!thread)
6036 return NULL;
6037
6038 init_waitqueue_head(&thread->wqueue);
6039
6040 thread->run = run;
6041 thread->mddev = mddev;
6042 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6043 thread->tsk = kthread_run(md_thread, thread,
6044 "%s_%s",
6045 mdname(thread->mddev),
6046 name ?: mddev->pers->name);
6047 if (IS_ERR(thread->tsk)) {
6048 kfree(thread);
6049 return NULL;
6050 }
6051 return thread;
6052 }
6053
6054 void md_unregister_thread(mdk_thread_t *thread)
6055 {
6056 if (!thread)
6057 return;
6058 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6059
6060 kthread_stop(thread->tsk);
6061 kfree(thread);
6062 }
6063
6064 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6065 {
6066 if (!mddev) {
6067 MD_BUG();
6068 return;
6069 }
6070
6071 if (!rdev || test_bit(Faulty, &rdev->flags))
6072 return;
6073
6074 if (mddev->external)
6075 set_bit(Blocked, &rdev->flags);
6076 /*
6077 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6078 mdname(mddev),
6079 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6080 __builtin_return_address(0),__builtin_return_address(1),
6081 __builtin_return_address(2),__builtin_return_address(3));
6082 */
6083 if (!mddev->pers)
6084 return;
6085 if (!mddev->pers->error_handler)
6086 return;
6087 mddev->pers->error_handler(mddev,rdev);
6088 if (mddev->degraded)
6089 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6090 sysfs_notify_dirent(rdev->sysfs_state);
6091 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6092 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6093 md_wakeup_thread(mddev->thread);
6094 md_new_event_inintr(mddev);
6095 }
6096
6097 /* seq_file implementation /proc/mdstat */
6098
6099 static void status_unused(struct seq_file *seq)
6100 {
6101 int i = 0;
6102 mdk_rdev_t *rdev;
6103
6104 seq_printf(seq, "unused devices: ");
6105
6106 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6107 char b[BDEVNAME_SIZE];
6108 i++;
6109 seq_printf(seq, "%s ",
6110 bdevname(rdev->bdev,b));
6111 }
6112 if (!i)
6113 seq_printf(seq, "<none>");
6114
6115 seq_printf(seq, "\n");
6116 }
6117
6118
6119 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6120 {
6121 sector_t max_sectors, resync, res;
6122 unsigned long dt, db;
6123 sector_t rt;
6124 int scale;
6125 unsigned int per_milli;
6126
6127 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6128
6129 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6130 max_sectors = mddev->resync_max_sectors;
6131 else
6132 max_sectors = mddev->dev_sectors;
6133
6134 /*
6135 * Should not happen.
6136 */
6137 if (!max_sectors) {
6138 MD_BUG();
6139 return;
6140 }
6141 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6142 * in a sector_t, and (max_sectors>>scale) will fit in a
6143 * u32, as those are the requirements for sector_div.
6144 * Thus 'scale' must be at least 10
6145 */
6146 scale = 10;
6147 if (sizeof(sector_t) > sizeof(unsigned long)) {
6148 while ( max_sectors/2 > (1ULL<<(scale+32)))
6149 scale++;
6150 }
6151 res = (resync>>scale)*1000;
6152 sector_div(res, (u32)((max_sectors>>scale)+1));
6153
6154 per_milli = res;
6155 {
6156 int i, x = per_milli/50, y = 20-x;
6157 seq_printf(seq, "[");
6158 for (i = 0; i < x; i++)
6159 seq_printf(seq, "=");
6160 seq_printf(seq, ">");
6161 for (i = 0; i < y; i++)
6162 seq_printf(seq, ".");
6163 seq_printf(seq, "] ");
6164 }
6165 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6166 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6167 "reshape" :
6168 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6169 "check" :
6170 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6171 "resync" : "recovery"))),
6172 per_milli/10, per_milli % 10,
6173 (unsigned long long) resync/2,
6174 (unsigned long long) max_sectors/2);
6175
6176 /*
6177 * dt: time from mark until now
6178 * db: blocks written from mark until now
6179 * rt: remaining time
6180 *
6181 * rt is a sector_t, so could be 32bit or 64bit.
6182 * So we divide before multiply in case it is 32bit and close
6183 * to the limit.
6184 * We scale the divisor (db) by 32 to avoid loosing precision
6185 * near the end of resync when the number of remaining sectors
6186 * is close to 'db'.
6187 * We then divide rt by 32 after multiplying by db to compensate.
6188 * The '+1' avoids division by zero if db is very small.
6189 */
6190 dt = ((jiffies - mddev->resync_mark) / HZ);
6191 if (!dt) dt++;
6192 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6193 - mddev->resync_mark_cnt;
6194
6195 rt = max_sectors - resync; /* number of remaining sectors */
6196 sector_div(rt, db/32+1);
6197 rt *= dt;
6198 rt >>= 5;
6199
6200 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6201 ((unsigned long)rt % 60)/6);
6202
6203 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6204 }
6205
6206 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6207 {
6208 struct list_head *tmp;
6209 loff_t l = *pos;
6210 mddev_t *mddev;
6211
6212 if (l >= 0x10000)
6213 return NULL;
6214 if (!l--)
6215 /* header */
6216 return (void*)1;
6217
6218 spin_lock(&all_mddevs_lock);
6219 list_for_each(tmp,&all_mddevs)
6220 if (!l--) {
6221 mddev = list_entry(tmp, mddev_t, all_mddevs);
6222 mddev_get(mddev);
6223 spin_unlock(&all_mddevs_lock);
6224 return mddev;
6225 }
6226 spin_unlock(&all_mddevs_lock);
6227 if (!l--)
6228 return (void*)2;/* tail */
6229 return NULL;
6230 }
6231
6232 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6233 {
6234 struct list_head *tmp;
6235 mddev_t *next_mddev, *mddev = v;
6236
6237 ++*pos;
6238 if (v == (void*)2)
6239 return NULL;
6240
6241 spin_lock(&all_mddevs_lock);
6242 if (v == (void*)1)
6243 tmp = all_mddevs.next;
6244 else
6245 tmp = mddev->all_mddevs.next;
6246 if (tmp != &all_mddevs)
6247 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6248 else {
6249 next_mddev = (void*)2;
6250 *pos = 0x10000;
6251 }
6252 spin_unlock(&all_mddevs_lock);
6253
6254 if (v != (void*)1)
6255 mddev_put(mddev);
6256 return next_mddev;
6257
6258 }
6259
6260 static void md_seq_stop(struct seq_file *seq, void *v)
6261 {
6262 mddev_t *mddev = v;
6263
6264 if (mddev && v != (void*)1 && v != (void*)2)
6265 mddev_put(mddev);
6266 }
6267
6268 struct mdstat_info {
6269 int event;
6270 };
6271
6272 static int md_seq_show(struct seq_file *seq, void *v)
6273 {
6274 mddev_t *mddev = v;
6275 sector_t sectors;
6276 mdk_rdev_t *rdev;
6277 struct mdstat_info *mi = seq->private;
6278 struct bitmap *bitmap;
6279
6280 if (v == (void*)1) {
6281 struct mdk_personality *pers;
6282 seq_printf(seq, "Personalities : ");
6283 spin_lock(&pers_lock);
6284 list_for_each_entry(pers, &pers_list, list)
6285 seq_printf(seq, "[%s] ", pers->name);
6286
6287 spin_unlock(&pers_lock);
6288 seq_printf(seq, "\n");
6289 mi->event = atomic_read(&md_event_count);
6290 return 0;
6291 }
6292 if (v == (void*)2) {
6293 status_unused(seq);
6294 return 0;
6295 }
6296
6297 if (mddev_lock(mddev) < 0)
6298 return -EINTR;
6299
6300 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6301 seq_printf(seq, "%s : %sactive", mdname(mddev),
6302 mddev->pers ? "" : "in");
6303 if (mddev->pers) {
6304 if (mddev->ro==1)
6305 seq_printf(seq, " (read-only)");
6306 if (mddev->ro==2)
6307 seq_printf(seq, " (auto-read-only)");
6308 seq_printf(seq, " %s", mddev->pers->name);
6309 }
6310
6311 sectors = 0;
6312 list_for_each_entry(rdev, &mddev->disks, same_set) {
6313 char b[BDEVNAME_SIZE];
6314 seq_printf(seq, " %s[%d]",
6315 bdevname(rdev->bdev,b), rdev->desc_nr);
6316 if (test_bit(WriteMostly, &rdev->flags))
6317 seq_printf(seq, "(W)");
6318 if (test_bit(Faulty, &rdev->flags)) {
6319 seq_printf(seq, "(F)");
6320 continue;
6321 } else if (rdev->raid_disk < 0)
6322 seq_printf(seq, "(S)"); /* spare */
6323 sectors += rdev->sectors;
6324 }
6325
6326 if (!list_empty(&mddev->disks)) {
6327 if (mddev->pers)
6328 seq_printf(seq, "\n %llu blocks",
6329 (unsigned long long)
6330 mddev->array_sectors / 2);
6331 else
6332 seq_printf(seq, "\n %llu blocks",
6333 (unsigned long long)sectors / 2);
6334 }
6335 if (mddev->persistent) {
6336 if (mddev->major_version != 0 ||
6337 mddev->minor_version != 90) {
6338 seq_printf(seq," super %d.%d",
6339 mddev->major_version,
6340 mddev->minor_version);
6341 }
6342 } else if (mddev->external)
6343 seq_printf(seq, " super external:%s",
6344 mddev->metadata_type);
6345 else
6346 seq_printf(seq, " super non-persistent");
6347
6348 if (mddev->pers) {
6349 mddev->pers->status(seq, mddev);
6350 seq_printf(seq, "\n ");
6351 if (mddev->pers->sync_request) {
6352 if (mddev->curr_resync > 2) {
6353 status_resync(seq, mddev);
6354 seq_printf(seq, "\n ");
6355 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6356 seq_printf(seq, "\tresync=DELAYED\n ");
6357 else if (mddev->recovery_cp < MaxSector)
6358 seq_printf(seq, "\tresync=PENDING\n ");
6359 }
6360 } else
6361 seq_printf(seq, "\n ");
6362
6363 if ((bitmap = mddev->bitmap)) {
6364 unsigned long chunk_kb;
6365 unsigned long flags;
6366 spin_lock_irqsave(&bitmap->lock, flags);
6367 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6368 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6369 "%lu%s chunk",
6370 bitmap->pages - bitmap->missing_pages,
6371 bitmap->pages,
6372 (bitmap->pages - bitmap->missing_pages)
6373 << (PAGE_SHIFT - 10),
6374 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6375 chunk_kb ? "KB" : "B");
6376 if (bitmap->file) {
6377 seq_printf(seq, ", file: ");
6378 seq_path(seq, &bitmap->file->f_path, " \t\n");
6379 }
6380
6381 seq_printf(seq, "\n");
6382 spin_unlock_irqrestore(&bitmap->lock, flags);
6383 }
6384
6385 seq_printf(seq, "\n");
6386 }
6387 mddev_unlock(mddev);
6388
6389 return 0;
6390 }
6391
6392 static const struct seq_operations md_seq_ops = {
6393 .start = md_seq_start,
6394 .next = md_seq_next,
6395 .stop = md_seq_stop,
6396 .show = md_seq_show,
6397 };
6398
6399 static int md_seq_open(struct inode *inode, struct file *file)
6400 {
6401 int error;
6402 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6403 if (mi == NULL)
6404 return -ENOMEM;
6405
6406 error = seq_open(file, &md_seq_ops);
6407 if (error)
6408 kfree(mi);
6409 else {
6410 struct seq_file *p = file->private_data;
6411 p->private = mi;
6412 mi->event = atomic_read(&md_event_count);
6413 }
6414 return error;
6415 }
6416
6417 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6418 {
6419 struct seq_file *m = filp->private_data;
6420 struct mdstat_info *mi = m->private;
6421 int mask;
6422
6423 poll_wait(filp, &md_event_waiters, wait);
6424
6425 /* always allow read */
6426 mask = POLLIN | POLLRDNORM;
6427
6428 if (mi->event != atomic_read(&md_event_count))
6429 mask |= POLLERR | POLLPRI;
6430 return mask;
6431 }
6432
6433 static const struct file_operations md_seq_fops = {
6434 .owner = THIS_MODULE,
6435 .open = md_seq_open,
6436 .read = seq_read,
6437 .llseek = seq_lseek,
6438 .release = seq_release_private,
6439 .poll = mdstat_poll,
6440 };
6441
6442 int register_md_personality(struct mdk_personality *p)
6443 {
6444 spin_lock(&pers_lock);
6445 list_add_tail(&p->list, &pers_list);
6446 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6447 spin_unlock(&pers_lock);
6448 return 0;
6449 }
6450
6451 int unregister_md_personality(struct mdk_personality *p)
6452 {
6453 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6454 spin_lock(&pers_lock);
6455 list_del_init(&p->list);
6456 spin_unlock(&pers_lock);
6457 return 0;
6458 }
6459
6460 static int is_mddev_idle(mddev_t *mddev, int init)
6461 {
6462 mdk_rdev_t * rdev;
6463 int idle;
6464 int curr_events;
6465
6466 idle = 1;
6467 rcu_read_lock();
6468 rdev_for_each_rcu(rdev, mddev) {
6469 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6470 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6471 (int)part_stat_read(&disk->part0, sectors[1]) -
6472 atomic_read(&disk->sync_io);
6473 /* sync IO will cause sync_io to increase before the disk_stats
6474 * as sync_io is counted when a request starts, and
6475 * disk_stats is counted when it completes.
6476 * So resync activity will cause curr_events to be smaller than
6477 * when there was no such activity.
6478 * non-sync IO will cause disk_stat to increase without
6479 * increasing sync_io so curr_events will (eventually)
6480 * be larger than it was before. Once it becomes
6481 * substantially larger, the test below will cause
6482 * the array to appear non-idle, and resync will slow
6483 * down.
6484 * If there is a lot of outstanding resync activity when
6485 * we set last_event to curr_events, then all that activity
6486 * completing might cause the array to appear non-idle
6487 * and resync will be slowed down even though there might
6488 * not have been non-resync activity. This will only
6489 * happen once though. 'last_events' will soon reflect
6490 * the state where there is little or no outstanding
6491 * resync requests, and further resync activity will
6492 * always make curr_events less than last_events.
6493 *
6494 */
6495 if (init || curr_events - rdev->last_events > 64) {
6496 rdev->last_events = curr_events;
6497 idle = 0;
6498 }
6499 }
6500 rcu_read_unlock();
6501 return idle;
6502 }
6503
6504 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6505 {
6506 /* another "blocks" (512byte) blocks have been synced */
6507 atomic_sub(blocks, &mddev->recovery_active);
6508 wake_up(&mddev->recovery_wait);
6509 if (!ok) {
6510 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6511 md_wakeup_thread(mddev->thread);
6512 // stop recovery, signal do_sync ....
6513 }
6514 }
6515
6516
6517 /* md_write_start(mddev, bi)
6518 * If we need to update some array metadata (e.g. 'active' flag
6519 * in superblock) before writing, schedule a superblock update
6520 * and wait for it to complete.
6521 */
6522 void md_write_start(mddev_t *mddev, struct bio *bi)
6523 {
6524 int did_change = 0;
6525 if (bio_data_dir(bi) != WRITE)
6526 return;
6527
6528 BUG_ON(mddev->ro == 1);
6529 if (mddev->ro == 2) {
6530 /* need to switch to read/write */
6531 mddev->ro = 0;
6532 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6533 md_wakeup_thread(mddev->thread);
6534 md_wakeup_thread(mddev->sync_thread);
6535 did_change = 1;
6536 }
6537 atomic_inc(&mddev->writes_pending);
6538 if (mddev->safemode == 1)
6539 mddev->safemode = 0;
6540 if (mddev->in_sync) {
6541 spin_lock_irq(&mddev->write_lock);
6542 if (mddev->in_sync) {
6543 mddev->in_sync = 0;
6544 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6545 md_wakeup_thread(mddev->thread);
6546 did_change = 1;
6547 }
6548 spin_unlock_irq(&mddev->write_lock);
6549 }
6550 if (did_change)
6551 sysfs_notify_dirent(mddev->sysfs_state);
6552 wait_event(mddev->sb_wait,
6553 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6554 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6555 }
6556
6557 void md_write_end(mddev_t *mddev)
6558 {
6559 if (atomic_dec_and_test(&mddev->writes_pending)) {
6560 if (mddev->safemode == 2)
6561 md_wakeup_thread(mddev->thread);
6562 else if (mddev->safemode_delay)
6563 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6564 }
6565 }
6566
6567 /* md_allow_write(mddev)
6568 * Calling this ensures that the array is marked 'active' so that writes
6569 * may proceed without blocking. It is important to call this before
6570 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6571 * Must be called with mddev_lock held.
6572 *
6573 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6574 * is dropped, so return -EAGAIN after notifying userspace.
6575 */
6576 int md_allow_write(mddev_t *mddev)
6577 {
6578 if (!mddev->pers)
6579 return 0;
6580 if (mddev->ro)
6581 return 0;
6582 if (!mddev->pers->sync_request)
6583 return 0;
6584
6585 spin_lock_irq(&mddev->write_lock);
6586 if (mddev->in_sync) {
6587 mddev->in_sync = 0;
6588 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6589 if (mddev->safemode_delay &&
6590 mddev->safemode == 0)
6591 mddev->safemode = 1;
6592 spin_unlock_irq(&mddev->write_lock);
6593 md_update_sb(mddev, 0);
6594 sysfs_notify_dirent(mddev->sysfs_state);
6595 } else
6596 spin_unlock_irq(&mddev->write_lock);
6597
6598 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6599 return -EAGAIN;
6600 else
6601 return 0;
6602 }
6603 EXPORT_SYMBOL_GPL(md_allow_write);
6604
6605 #define SYNC_MARKS 10
6606 #define SYNC_MARK_STEP (3*HZ)
6607 void md_do_sync(mddev_t *mddev)
6608 {
6609 mddev_t *mddev2;
6610 unsigned int currspeed = 0,
6611 window;
6612 sector_t max_sectors,j, io_sectors;
6613 unsigned long mark[SYNC_MARKS];
6614 sector_t mark_cnt[SYNC_MARKS];
6615 int last_mark,m;
6616 struct list_head *tmp;
6617 sector_t last_check;
6618 int skipped = 0;
6619 mdk_rdev_t *rdev;
6620 char *desc;
6621
6622 /* just incase thread restarts... */
6623 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6624 return;
6625 if (mddev->ro) /* never try to sync a read-only array */
6626 return;
6627
6628 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6629 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6630 desc = "data-check";
6631 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6632 desc = "requested-resync";
6633 else
6634 desc = "resync";
6635 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6636 desc = "reshape";
6637 else
6638 desc = "recovery";
6639
6640 /* we overload curr_resync somewhat here.
6641 * 0 == not engaged in resync at all
6642 * 2 == checking that there is no conflict with another sync
6643 * 1 == like 2, but have yielded to allow conflicting resync to
6644 * commense
6645 * other == active in resync - this many blocks
6646 *
6647 * Before starting a resync we must have set curr_resync to
6648 * 2, and then checked that every "conflicting" array has curr_resync
6649 * less than ours. When we find one that is the same or higher
6650 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6651 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6652 * This will mean we have to start checking from the beginning again.
6653 *
6654 */
6655
6656 do {
6657 mddev->curr_resync = 2;
6658
6659 try_again:
6660 if (kthread_should_stop())
6661 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6662
6663 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6664 goto skip;
6665 for_each_mddev(mddev2, tmp) {
6666 if (mddev2 == mddev)
6667 continue;
6668 if (!mddev->parallel_resync
6669 && mddev2->curr_resync
6670 && match_mddev_units(mddev, mddev2)) {
6671 DEFINE_WAIT(wq);
6672 if (mddev < mddev2 && mddev->curr_resync == 2) {
6673 /* arbitrarily yield */
6674 mddev->curr_resync = 1;
6675 wake_up(&resync_wait);
6676 }
6677 if (mddev > mddev2 && mddev->curr_resync == 1)
6678 /* no need to wait here, we can wait the next
6679 * time 'round when curr_resync == 2
6680 */
6681 continue;
6682 /* We need to wait 'interruptible' so as not to
6683 * contribute to the load average, and not to
6684 * be caught by 'softlockup'
6685 */
6686 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6687 if (!kthread_should_stop() &&
6688 mddev2->curr_resync >= mddev->curr_resync) {
6689 printk(KERN_INFO "md: delaying %s of %s"
6690 " until %s has finished (they"
6691 " share one or more physical units)\n",
6692 desc, mdname(mddev), mdname(mddev2));
6693 mddev_put(mddev2);
6694 if (signal_pending(current))
6695 flush_signals(current);
6696 schedule();
6697 finish_wait(&resync_wait, &wq);
6698 goto try_again;
6699 }
6700 finish_wait(&resync_wait, &wq);
6701 }
6702 }
6703 } while (mddev->curr_resync < 2);
6704
6705 j = 0;
6706 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6707 /* resync follows the size requested by the personality,
6708 * which defaults to physical size, but can be virtual size
6709 */
6710 max_sectors = mddev->resync_max_sectors;
6711 mddev->resync_mismatches = 0;
6712 /* we don't use the checkpoint if there's a bitmap */
6713 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6714 j = mddev->resync_min;
6715 else if (!mddev->bitmap)
6716 j = mddev->recovery_cp;
6717
6718 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6719 max_sectors = mddev->dev_sectors;
6720 else {
6721 /* recovery follows the physical size of devices */
6722 max_sectors = mddev->dev_sectors;
6723 j = MaxSector;
6724 rcu_read_lock();
6725 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6726 if (rdev->raid_disk >= 0 &&
6727 !test_bit(Faulty, &rdev->flags) &&
6728 !test_bit(In_sync, &rdev->flags) &&
6729 rdev->recovery_offset < j)
6730 j = rdev->recovery_offset;
6731 rcu_read_unlock();
6732 }
6733
6734 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6735 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6736 " %d KB/sec/disk.\n", speed_min(mddev));
6737 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6738 "(but not more than %d KB/sec) for %s.\n",
6739 speed_max(mddev), desc);
6740
6741 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6742
6743 io_sectors = 0;
6744 for (m = 0; m < SYNC_MARKS; m++) {
6745 mark[m] = jiffies;
6746 mark_cnt[m] = io_sectors;
6747 }
6748 last_mark = 0;
6749 mddev->resync_mark = mark[last_mark];
6750 mddev->resync_mark_cnt = mark_cnt[last_mark];
6751
6752 /*
6753 * Tune reconstruction:
6754 */
6755 window = 32*(PAGE_SIZE/512);
6756 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6757 window/2,(unsigned long long) max_sectors/2);
6758
6759 atomic_set(&mddev->recovery_active, 0);
6760 last_check = 0;
6761
6762 if (j>2) {
6763 printk(KERN_INFO
6764 "md: resuming %s of %s from checkpoint.\n",
6765 desc, mdname(mddev));
6766 mddev->curr_resync = j;
6767 }
6768 mddev->curr_resync_completed = mddev->curr_resync;
6769
6770 while (j < max_sectors) {
6771 sector_t sectors;
6772
6773 skipped = 0;
6774
6775 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6776 ((mddev->curr_resync > mddev->curr_resync_completed &&
6777 (mddev->curr_resync - mddev->curr_resync_completed)
6778 > (max_sectors >> 4)) ||
6779 (j - mddev->curr_resync_completed)*2
6780 >= mddev->resync_max - mddev->curr_resync_completed
6781 )) {
6782 /* time to update curr_resync_completed */
6783 blk_unplug(mddev->queue);
6784 wait_event(mddev->recovery_wait,
6785 atomic_read(&mddev->recovery_active) == 0);
6786 mddev->curr_resync_completed =
6787 mddev->curr_resync;
6788 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6789 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6790 }
6791
6792 while (j >= mddev->resync_max && !kthread_should_stop()) {
6793 /* As this condition is controlled by user-space,
6794 * we can block indefinitely, so use '_interruptible'
6795 * to avoid triggering warnings.
6796 */
6797 flush_signals(current); /* just in case */
6798 wait_event_interruptible(mddev->recovery_wait,
6799 mddev->resync_max > j
6800 || kthread_should_stop());
6801 }
6802
6803 if (kthread_should_stop())
6804 goto interrupted;
6805
6806 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6807 currspeed < speed_min(mddev));
6808 if (sectors == 0) {
6809 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6810 goto out;
6811 }
6812
6813 if (!skipped) { /* actual IO requested */
6814 io_sectors += sectors;
6815 atomic_add(sectors, &mddev->recovery_active);
6816 }
6817
6818 j += sectors;
6819 if (j>1) mddev->curr_resync = j;
6820 mddev->curr_mark_cnt = io_sectors;
6821 if (last_check == 0)
6822 /* this is the earliers that rebuilt will be
6823 * visible in /proc/mdstat
6824 */
6825 md_new_event(mddev);
6826
6827 if (last_check + window > io_sectors || j == max_sectors)
6828 continue;
6829
6830 last_check = io_sectors;
6831
6832 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6833 break;
6834
6835 repeat:
6836 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6837 /* step marks */
6838 int next = (last_mark+1) % SYNC_MARKS;
6839
6840 mddev->resync_mark = mark[next];
6841 mddev->resync_mark_cnt = mark_cnt[next];
6842 mark[next] = jiffies;
6843 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6844 last_mark = next;
6845 }
6846
6847
6848 if (kthread_should_stop())
6849 goto interrupted;
6850
6851
6852 /*
6853 * this loop exits only if either when we are slower than
6854 * the 'hard' speed limit, or the system was IO-idle for
6855 * a jiffy.
6856 * the system might be non-idle CPU-wise, but we only care
6857 * about not overloading the IO subsystem. (things like an
6858 * e2fsck being done on the RAID array should execute fast)
6859 */
6860 blk_unplug(mddev->queue);
6861 cond_resched();
6862
6863 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6864 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6865
6866 if (currspeed > speed_min(mddev)) {
6867 if ((currspeed > speed_max(mddev)) ||
6868 !is_mddev_idle(mddev, 0)) {
6869 msleep(500);
6870 goto repeat;
6871 }
6872 }
6873 }
6874 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6875 /*
6876 * this also signals 'finished resyncing' to md_stop
6877 */
6878 out:
6879 blk_unplug(mddev->queue);
6880
6881 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6882
6883 /* tell personality that we are finished */
6884 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6885
6886 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6887 mddev->curr_resync > 2) {
6888 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6889 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6890 if (mddev->curr_resync >= mddev->recovery_cp) {
6891 printk(KERN_INFO
6892 "md: checkpointing %s of %s.\n",
6893 desc, mdname(mddev));
6894 mddev->recovery_cp = mddev->curr_resync;
6895 }
6896 } else
6897 mddev->recovery_cp = MaxSector;
6898 } else {
6899 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6900 mddev->curr_resync = MaxSector;
6901 rcu_read_lock();
6902 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6903 if (rdev->raid_disk >= 0 &&
6904 mddev->delta_disks >= 0 &&
6905 !test_bit(Faulty, &rdev->flags) &&
6906 !test_bit(In_sync, &rdev->flags) &&
6907 rdev->recovery_offset < mddev->curr_resync)
6908 rdev->recovery_offset = mddev->curr_resync;
6909 rcu_read_unlock();
6910 }
6911 }
6912 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6913
6914 skip:
6915 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6916 /* We completed so min/max setting can be forgotten if used. */
6917 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6918 mddev->resync_min = 0;
6919 mddev->resync_max = MaxSector;
6920 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6921 mddev->resync_min = mddev->curr_resync_completed;
6922 mddev->curr_resync = 0;
6923 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6924 mddev->curr_resync_completed = 0;
6925 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6926 wake_up(&resync_wait);
6927 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6928 md_wakeup_thread(mddev->thread);
6929 return;
6930
6931 interrupted:
6932 /*
6933 * got a signal, exit.
6934 */
6935 printk(KERN_INFO
6936 "md: md_do_sync() got signal ... exiting\n");
6937 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6938 goto out;
6939
6940 }
6941 EXPORT_SYMBOL_GPL(md_do_sync);
6942
6943
6944 static int remove_and_add_spares(mddev_t *mddev)
6945 {
6946 mdk_rdev_t *rdev;
6947 int spares = 0;
6948
6949 mddev->curr_resync_completed = 0;
6950
6951 list_for_each_entry(rdev, &mddev->disks, same_set)
6952 if (rdev->raid_disk >= 0 &&
6953 !test_bit(Blocked, &rdev->flags) &&
6954 (test_bit(Faulty, &rdev->flags) ||
6955 ! test_bit(In_sync, &rdev->flags)) &&
6956 atomic_read(&rdev->nr_pending)==0) {
6957 if (mddev->pers->hot_remove_disk(
6958 mddev, rdev->raid_disk)==0) {
6959 char nm[20];
6960 sprintf(nm,"rd%d", rdev->raid_disk);
6961 sysfs_remove_link(&mddev->kobj, nm);
6962 rdev->raid_disk = -1;
6963 }
6964 }
6965
6966 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6967 list_for_each_entry(rdev, &mddev->disks, same_set) {
6968 if (rdev->raid_disk >= 0 &&
6969 !test_bit(In_sync, &rdev->flags) &&
6970 !test_bit(Blocked, &rdev->flags))
6971 spares++;
6972 if (rdev->raid_disk < 0
6973 && !test_bit(Faulty, &rdev->flags)) {
6974 rdev->recovery_offset = 0;
6975 if (mddev->pers->
6976 hot_add_disk(mddev, rdev) == 0) {
6977 char nm[20];
6978 sprintf(nm, "rd%d", rdev->raid_disk);
6979 if (sysfs_create_link(&mddev->kobj,
6980 &rdev->kobj, nm))
6981 printk(KERN_WARNING
6982 "md: cannot register "
6983 "%s for %s\n",
6984 nm, mdname(mddev));
6985 spares++;
6986 md_new_event(mddev);
6987 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6988 } else
6989 break;
6990 }
6991 }
6992 }
6993 return spares;
6994 }
6995 /*
6996 * This routine is regularly called by all per-raid-array threads to
6997 * deal with generic issues like resync and super-block update.
6998 * Raid personalities that don't have a thread (linear/raid0) do not
6999 * need this as they never do any recovery or update the superblock.
7000 *
7001 * It does not do any resync itself, but rather "forks" off other threads
7002 * to do that as needed.
7003 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7004 * "->recovery" and create a thread at ->sync_thread.
7005 * When the thread finishes it sets MD_RECOVERY_DONE
7006 * and wakeups up this thread which will reap the thread and finish up.
7007 * This thread also removes any faulty devices (with nr_pending == 0).
7008 *
7009 * The overall approach is:
7010 * 1/ if the superblock needs updating, update it.
7011 * 2/ If a recovery thread is running, don't do anything else.
7012 * 3/ If recovery has finished, clean up, possibly marking spares active.
7013 * 4/ If there are any faulty devices, remove them.
7014 * 5/ If array is degraded, try to add spares devices
7015 * 6/ If array has spares or is not in-sync, start a resync thread.
7016 */
7017 void md_check_recovery(mddev_t *mddev)
7018 {
7019 mdk_rdev_t *rdev;
7020
7021
7022 if (mddev->bitmap)
7023 bitmap_daemon_work(mddev);
7024
7025 if (mddev->ro)
7026 return;
7027
7028 if (signal_pending(current)) {
7029 if (mddev->pers->sync_request && !mddev->external) {
7030 printk(KERN_INFO "md: %s in immediate safe mode\n",
7031 mdname(mddev));
7032 mddev->safemode = 2;
7033 }
7034 flush_signals(current);
7035 }
7036
7037 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7038 return;
7039 if ( ! (
7040 (mddev->flags && !mddev->external) ||
7041 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7042 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7043 (mddev->external == 0 && mddev->safemode == 1) ||
7044 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7045 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7046 ))
7047 return;
7048
7049 if (mddev_trylock(mddev)) {
7050 int spares = 0;
7051
7052 if (mddev->ro) {
7053 /* Only thing we do on a ro array is remove
7054 * failed devices.
7055 */
7056 remove_and_add_spares(mddev);
7057 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7058 goto unlock;
7059 }
7060
7061 if (!mddev->external) {
7062 int did_change = 0;
7063 spin_lock_irq(&mddev->write_lock);
7064 if (mddev->safemode &&
7065 !atomic_read(&mddev->writes_pending) &&
7066 !mddev->in_sync &&
7067 mddev->recovery_cp == MaxSector) {
7068 mddev->in_sync = 1;
7069 did_change = 1;
7070 if (mddev->persistent)
7071 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7072 }
7073 if (mddev->safemode == 1)
7074 mddev->safemode = 0;
7075 spin_unlock_irq(&mddev->write_lock);
7076 if (did_change)
7077 sysfs_notify_dirent(mddev->sysfs_state);
7078 }
7079
7080 if (mddev->flags)
7081 md_update_sb(mddev, 0);
7082
7083 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7084 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7085 /* resync/recovery still happening */
7086 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7087 goto unlock;
7088 }
7089 if (mddev->sync_thread) {
7090 /* resync has finished, collect result */
7091 md_unregister_thread(mddev->sync_thread);
7092 mddev->sync_thread = NULL;
7093 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7094 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7095 /* success...*/
7096 /* activate any spares */
7097 if (mddev->pers->spare_active(mddev))
7098 sysfs_notify(&mddev->kobj, NULL,
7099 "degraded");
7100 }
7101 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7102 mddev->pers->finish_reshape)
7103 mddev->pers->finish_reshape(mddev);
7104 md_update_sb(mddev, 1);
7105
7106 /* if array is no-longer degraded, then any saved_raid_disk
7107 * information must be scrapped
7108 */
7109 if (!mddev->degraded)
7110 list_for_each_entry(rdev, &mddev->disks, same_set)
7111 rdev->saved_raid_disk = -1;
7112
7113 mddev->recovery = 0;
7114 /* flag recovery needed just to double check */
7115 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7116 sysfs_notify_dirent(mddev->sysfs_action);
7117 md_new_event(mddev);
7118 goto unlock;
7119 }
7120 /* Set RUNNING before clearing NEEDED to avoid
7121 * any transients in the value of "sync_action".
7122 */
7123 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7124 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7125 /* Clear some bits that don't mean anything, but
7126 * might be left set
7127 */
7128 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7129 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7130
7131 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7132 goto unlock;
7133 /* no recovery is running.
7134 * remove any failed drives, then
7135 * add spares if possible.
7136 * Spare are also removed and re-added, to allow
7137 * the personality to fail the re-add.
7138 */
7139
7140 if (mddev->reshape_position != MaxSector) {
7141 if (mddev->pers->check_reshape == NULL ||
7142 mddev->pers->check_reshape(mddev) != 0)
7143 /* Cannot proceed */
7144 goto unlock;
7145 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7146 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7147 } else if ((spares = remove_and_add_spares(mddev))) {
7148 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7149 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7150 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7151 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7152 } else if (mddev->recovery_cp < MaxSector) {
7153 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7154 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7155 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7156 /* nothing to be done ... */
7157 goto unlock;
7158
7159 if (mddev->pers->sync_request) {
7160 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7161 /* We are adding a device or devices to an array
7162 * which has the bitmap stored on all devices.
7163 * So make sure all bitmap pages get written
7164 */
7165 bitmap_write_all(mddev->bitmap);
7166 }
7167 mddev->sync_thread = md_register_thread(md_do_sync,
7168 mddev,
7169 "resync");
7170 if (!mddev->sync_thread) {
7171 printk(KERN_ERR "%s: could not start resync"
7172 " thread...\n",
7173 mdname(mddev));
7174 /* leave the spares where they are, it shouldn't hurt */
7175 mddev->recovery = 0;
7176 } else
7177 md_wakeup_thread(mddev->sync_thread);
7178 sysfs_notify_dirent(mddev->sysfs_action);
7179 md_new_event(mddev);
7180 }
7181 unlock:
7182 if (!mddev->sync_thread) {
7183 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7184 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7185 &mddev->recovery))
7186 if (mddev->sysfs_action)
7187 sysfs_notify_dirent(mddev->sysfs_action);
7188 }
7189 mddev_unlock(mddev);
7190 }
7191 }
7192
7193 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7194 {
7195 sysfs_notify_dirent(rdev->sysfs_state);
7196 wait_event_timeout(rdev->blocked_wait,
7197 !test_bit(Blocked, &rdev->flags),
7198 msecs_to_jiffies(5000));
7199 rdev_dec_pending(rdev, mddev);
7200 }
7201 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7202
7203 static int md_notify_reboot(struct notifier_block *this,
7204 unsigned long code, void *x)
7205 {
7206 struct list_head *tmp;
7207 mddev_t *mddev;
7208
7209 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7210
7211 printk(KERN_INFO "md: stopping all md devices.\n");
7212
7213 for_each_mddev(mddev, tmp)
7214 if (mddev_trylock(mddev)) {
7215 /* Force a switch to readonly even array
7216 * appears to still be in use. Hence
7217 * the '100'.
7218 */
7219 md_set_readonly(mddev, 100);
7220 mddev_unlock(mddev);
7221 }
7222 /*
7223 * certain more exotic SCSI devices are known to be
7224 * volatile wrt too early system reboots. While the
7225 * right place to handle this issue is the given
7226 * driver, we do want to have a safe RAID driver ...
7227 */
7228 mdelay(1000*1);
7229 }
7230 return NOTIFY_DONE;
7231 }
7232
7233 static struct notifier_block md_notifier = {
7234 .notifier_call = md_notify_reboot,
7235 .next = NULL,
7236 .priority = INT_MAX, /* before any real devices */
7237 };
7238
7239 static void md_geninit(void)
7240 {
7241 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7242
7243 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7244 }
7245
7246 static int __init md_init(void)
7247 {
7248 if (register_blkdev(MD_MAJOR, "md"))
7249 return -1;
7250 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7251 unregister_blkdev(MD_MAJOR, "md");
7252 return -1;
7253 }
7254 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7255 md_probe, NULL, NULL);
7256 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7257 md_probe, NULL, NULL);
7258
7259 register_reboot_notifier(&md_notifier);
7260 raid_table_header = register_sysctl_table(raid_root_table);
7261
7262 md_geninit();
7263 return 0;
7264 }
7265
7266
7267 #ifndef MODULE
7268
7269 /*
7270 * Searches all registered partitions for autorun RAID arrays
7271 * at boot time.
7272 */
7273
7274 static LIST_HEAD(all_detected_devices);
7275 struct detected_devices_node {
7276 struct list_head list;
7277 dev_t dev;
7278 };
7279
7280 void md_autodetect_dev(dev_t dev)
7281 {
7282 struct detected_devices_node *node_detected_dev;
7283
7284 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7285 if (node_detected_dev) {
7286 node_detected_dev->dev = dev;
7287 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7288 } else {
7289 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7290 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7291 }
7292 }
7293
7294
7295 static void autostart_arrays(int part)
7296 {
7297 mdk_rdev_t *rdev;
7298 struct detected_devices_node *node_detected_dev;
7299 dev_t dev;
7300 int i_scanned, i_passed;
7301
7302 i_scanned = 0;
7303 i_passed = 0;
7304
7305 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7306
7307 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7308 i_scanned++;
7309 node_detected_dev = list_entry(all_detected_devices.next,
7310 struct detected_devices_node, list);
7311 list_del(&node_detected_dev->list);
7312 dev = node_detected_dev->dev;
7313 kfree(node_detected_dev);
7314 rdev = md_import_device(dev,0, 90);
7315 if (IS_ERR(rdev))
7316 continue;
7317
7318 if (test_bit(Faulty, &rdev->flags)) {
7319 MD_BUG();
7320 continue;
7321 }
7322 set_bit(AutoDetected, &rdev->flags);
7323 list_add(&rdev->same_set, &pending_raid_disks);
7324 i_passed++;
7325 }
7326
7327 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7328 i_scanned, i_passed);
7329
7330 autorun_devices(part);
7331 }
7332
7333 #endif /* !MODULE */
7334
7335 static __exit void md_exit(void)
7336 {
7337 mddev_t *mddev;
7338 struct list_head *tmp;
7339
7340 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7341 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7342
7343 unregister_blkdev(MD_MAJOR,"md");
7344 unregister_blkdev(mdp_major, "mdp");
7345 unregister_reboot_notifier(&md_notifier);
7346 unregister_sysctl_table(raid_table_header);
7347 remove_proc_entry("mdstat", NULL);
7348 for_each_mddev(mddev, tmp) {
7349 export_array(mddev);
7350 mddev->hold_active = 0;
7351 }
7352 }
7353
7354 subsys_initcall(md_init);
7355 module_exit(md_exit)
7356
7357 static int get_ro(char *buffer, struct kernel_param *kp)
7358 {
7359 return sprintf(buffer, "%d", start_readonly);
7360 }
7361 static int set_ro(const char *val, struct kernel_param *kp)
7362 {
7363 char *e;
7364 int num = simple_strtoul(val, &e, 10);
7365 if (*val && (*e == '\0' || *e == '\n')) {
7366 start_readonly = num;
7367 return 0;
7368 }
7369 return -EINVAL;
7370 }
7371
7372 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7373 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7374
7375 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7376
7377 EXPORT_SYMBOL(register_md_personality);
7378 EXPORT_SYMBOL(unregister_md_personality);
7379 EXPORT_SYMBOL(md_error);
7380 EXPORT_SYMBOL(md_done_sync);
7381 EXPORT_SYMBOL(md_write_start);
7382 EXPORT_SYMBOL(md_write_end);
7383 EXPORT_SYMBOL(md_register_thread);
7384 EXPORT_SYMBOL(md_unregister_thread);
7385 EXPORT_SYMBOL(md_wakeup_thread);
7386 EXPORT_SYMBOL(md_check_recovery);
7387 MODULE_LICENSE("GPL");
7388 MODULE_DESCRIPTION("MD RAID framework");
7389 MODULE_ALIAS("md");
7390 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree