?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ocfs2: Add metaecc for ocfs2_refcount_block.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blobd4351ff0849fb455974208b881348ce9baf2a801
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/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
77 * idle IO detection.
78 *
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
81 */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
107 },
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
115 },
116 { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120 {
121 .ctl_name = DEV_RAID,
122 .procname = "raid",
123 .maxlen = 0,
124 .mode = S_IRUGO|S_IXUGO,
125 .child = raid_table,
126 },
127 { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131 {
132 .ctl_name = CTL_DEV,
133 .procname = "dev",
134 .maxlen = 0,
135 .mode = 0555,
136 .child = raid_dir_table,
137 },
138 { .ctl_name = 0 }
139 };
140
141 static struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
149 * count increases.
150 *
151 * Events are:
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
154 */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
166 */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171 }
172
173 /*
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
176 */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
187 */
188 #define for_each_mddev(mddev,tmp) \
189 \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
192 mddev = NULL;}); \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
200 tmp = tmp->next;}) \
201 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
210 */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213 mddev_t *mddev = q->queuedata;
214 int rv;
215 if (mddev == NULL || mddev->pers == NULL) {
216 bio_io_error(bio);
217 return 0;
218 }
219 rcu_read_lock();
220 if (mddev->suspended) {
221 DEFINE_WAIT(__wait);
222 for (;;) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
226 break;
227 rcu_read_unlock();
228 schedule();
229 rcu_read_lock();
230 }
231 finish_wait(&mddev->sb_wait, &__wait);
232 }
233 atomic_inc(&mddev->active_io);
234 rcu_read_unlock();
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
238
239 return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
246 synchronize_rcu();
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
254 * unloaded
255 */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
263 }
264
265
266 static inline mddev_t *mddev_get(mddev_t *mddev)
267 {
268 atomic_inc(&mddev->active);
269 return mddev;
270 }
271
272 static void mddev_delayed_delete(struct work_struct *ws);
273
274 static void mddev_put(mddev_t *mddev)
275 {
276 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277 return;
278 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279 !mddev->hold_active) {
280 list_del(&mddev->all_mddevs);
281 if (mddev->gendisk) {
282 /* we did a probe so need to clean up.
283 * Call schedule_work inside the spinlock
284 * so that flush_scheduled_work() after
285 * mddev_find will succeed in waiting for the
286 * work to be done.
287 */
288 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289 schedule_work(&mddev->del_work);
290 } else
291 kfree(mddev);
292 }
293 spin_unlock(&all_mddevs_lock);
294 }
295
296 static mddev_t * mddev_find(dev_t unit)
297 {
298 mddev_t *mddev, *new = NULL;
299
300 retry:
301 spin_lock(&all_mddevs_lock);
302
303 if (unit) {
304 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305 if (mddev->unit == unit) {
306 mddev_get(mddev);
307 spin_unlock(&all_mddevs_lock);
308 kfree(new);
309 return mddev;
310 }
311
312 if (new) {
313 list_add(&new->all_mddevs, &all_mddevs);
314 spin_unlock(&all_mddevs_lock);
315 new->hold_active = UNTIL_IOCTL;
316 return new;
317 }
318 } else if (new) {
319 /* find an unused unit number */
320 static int next_minor = 512;
321 int start = next_minor;
322 int is_free = 0;
323 int dev = 0;
324 while (!is_free) {
325 dev = MKDEV(MD_MAJOR, next_minor);
326 next_minor++;
327 if (next_minor > MINORMASK)
328 next_minor = 0;
329 if (next_minor == start) {
330 /* Oh dear, all in use. */
331 spin_unlock(&all_mddevs_lock);
332 kfree(new);
333 return NULL;
334 }
335
336 is_free = 1;
337 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338 if (mddev->unit == dev) {
339 is_free = 0;
340 break;
341 }
342 }
343 new->unit = dev;
344 new->md_minor = MINOR(dev);
345 new->hold_active = UNTIL_STOP;
346 list_add(&new->all_mddevs, &all_mddevs);
347 spin_unlock(&all_mddevs_lock);
348 return new;
349 }
350 spin_unlock(&all_mddevs_lock);
351
352 new = kzalloc(sizeof(*new), GFP_KERNEL);
353 if (!new)
354 return NULL;
355
356 new->unit = unit;
357 if (MAJOR(unit) == MD_MAJOR)
358 new->md_minor = MINOR(unit);
359 else
360 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
362 mutex_init(&new->reconfig_mutex);
363 INIT_LIST_HEAD(&new->disks);
364 INIT_LIST_HEAD(&new->all_mddevs);
365 init_timer(&new->safemode_timer);
366 atomic_set(&new->active, 1);
367 atomic_set(&new->openers, 0);
368 atomic_set(&new->active_io, 0);
369 spin_lock_init(&new->write_lock);
370 init_waitqueue_head(&new->sb_wait);
371 init_waitqueue_head(&new->recovery_wait);
372 new->reshape_position = MaxSector;
373 new->resync_min = 0;
374 new->resync_max = MaxSector;
375 new->level = LEVEL_NONE;
376
377 goto retry;
378 }
379
380 static inline int mddev_lock(mddev_t * mddev)
381 {
382 return mutex_lock_interruptible(&mddev->reconfig_mutex);
383 }
384
385 static inline int mddev_is_locked(mddev_t *mddev)
386 {
387 return mutex_is_locked(&mddev->reconfig_mutex);
388 }
389
390 static inline int mddev_trylock(mddev_t * mddev)
391 {
392 return mutex_trylock(&mddev->reconfig_mutex);
393 }
394
395 static inline void mddev_unlock(mddev_t * mddev)
396 {
397 mutex_unlock(&mddev->reconfig_mutex);
398
399 md_wakeup_thread(mddev->thread);
400 }
401
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
403 {
404 mdk_rdev_t *rdev;
405
406 list_for_each_entry(rdev, &mddev->disks, same_set)
407 if (rdev->desc_nr == nr)
408 return rdev;
409
410 return NULL;
411 }
412
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414 {
415 mdk_rdev_t *rdev;
416
417 list_for_each_entry(rdev, &mddev->disks, same_set)
418 if (rdev->bdev->bd_dev == dev)
419 return rdev;
420
421 return NULL;
422 }
423
424 static struct mdk_personality *find_pers(int level, char *clevel)
425 {
426 struct mdk_personality *pers;
427 list_for_each_entry(pers, &pers_list, list) {
428 if (level != LEVEL_NONE && pers->level == level)
429 return pers;
430 if (strcmp(pers->name, clevel)==0)
431 return pers;
432 }
433 return NULL;
434 }
435
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
438 {
439 sector_t num_sectors = bdev->bd_inode->i_size / 512;
440 return MD_NEW_SIZE_SECTORS(num_sectors);
441 }
442
443 static int alloc_disk_sb(mdk_rdev_t * rdev)
444 {
445 if (rdev->sb_page)
446 MD_BUG();
447
448 rdev->sb_page = alloc_page(GFP_KERNEL);
449 if (!rdev->sb_page) {
450 printk(KERN_ALERT "md: out of memory.\n");
451 return -ENOMEM;
452 }
453
454 return 0;
455 }
456
457 static void free_disk_sb(mdk_rdev_t * rdev)
458 {
459 if (rdev->sb_page) {
460 put_page(rdev->sb_page);
461 rdev->sb_loaded = 0;
462 rdev->sb_page = NULL;
463 rdev->sb_start = 0;
464 rdev->sectors = 0;
465 }
466 }
467
468
469 static void super_written(struct bio *bio, int error)
470 {
471 mdk_rdev_t *rdev = bio->bi_private;
472 mddev_t *mddev = rdev->mddev;
473
474 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
475 printk("md: super_written gets error=%d, uptodate=%d\n",
476 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
477 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
478 md_error(mddev, rdev);
479 }
480
481 if (atomic_dec_and_test(&mddev->pending_writes))
482 wake_up(&mddev->sb_wait);
483 bio_put(bio);
484 }
485
486 static void super_written_barrier(struct bio *bio, int error)
487 {
488 struct bio *bio2 = bio->bi_private;
489 mdk_rdev_t *rdev = bio2->bi_private;
490 mddev_t *mddev = rdev->mddev;
491
492 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
493 error == -EOPNOTSUPP) {
494 unsigned long flags;
495 /* barriers don't appear to be supported :-( */
496 set_bit(BarriersNotsupp, &rdev->flags);
497 mddev->barriers_work = 0;
498 spin_lock_irqsave(&mddev->write_lock, flags);
499 bio2->bi_next = mddev->biolist;
500 mddev->biolist = bio2;
501 spin_unlock_irqrestore(&mddev->write_lock, flags);
502 wake_up(&mddev->sb_wait);
503 bio_put(bio);
504 } else {
505 bio_put(bio2);
506 bio->bi_private = rdev;
507 super_written(bio, error);
508 }
509 }
510
511 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
512 sector_t sector, int size, struct page *page)
513 {
514 /* write first size bytes of page to sector of rdev
515 * Increment mddev->pending_writes before returning
516 * and decrement it on completion, waking up sb_wait
517 * if zero is reached.
518 * If an error occurred, call md_error
519 *
520 * As we might need to resubmit the request if BIO_RW_BARRIER
521 * causes ENOTSUPP, we allocate a spare bio...
522 */
523 struct bio *bio = bio_alloc(GFP_NOIO, 1);
524 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
525
526 bio->bi_bdev = rdev->bdev;
527 bio->bi_sector = sector;
528 bio_add_page(bio, page, size, 0);
529 bio->bi_private = rdev;
530 bio->bi_end_io = super_written;
531 bio->bi_rw = rw;
532
533 atomic_inc(&mddev->pending_writes);
534 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
535 struct bio *rbio;
536 rw |= (1<<BIO_RW_BARRIER);
537 rbio = bio_clone(bio, GFP_NOIO);
538 rbio->bi_private = bio;
539 rbio->bi_end_io = super_written_barrier;
540 submit_bio(rw, rbio);
541 } else
542 submit_bio(rw, bio);
543 }
544
545 void md_super_wait(mddev_t *mddev)
546 {
547 /* wait for all superblock writes that were scheduled to complete.
548 * if any had to be retried (due to BARRIER problems), retry them
549 */
550 DEFINE_WAIT(wq);
551 for(;;) {
552 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
553 if (atomic_read(&mddev->pending_writes)==0)
554 break;
555 while (mddev->biolist) {
556 struct bio *bio;
557 spin_lock_irq(&mddev->write_lock);
558 bio = mddev->biolist;
559 mddev->biolist = bio->bi_next ;
560 bio->bi_next = NULL;
561 spin_unlock_irq(&mddev->write_lock);
562 submit_bio(bio->bi_rw, bio);
563 }
564 schedule();
565 }
566 finish_wait(&mddev->sb_wait, &wq);
567 }
568
569 static void bi_complete(struct bio *bio, int error)
570 {
571 complete((struct completion*)bio->bi_private);
572 }
573
574 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
575 struct page *page, int rw)
576 {
577 struct bio *bio = bio_alloc(GFP_NOIO, 1);
578 struct completion event;
579 int ret;
580
581 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
582
583 bio->bi_bdev = bdev;
584 bio->bi_sector = sector;
585 bio_add_page(bio, page, size, 0);
586 init_completion(&event);
587 bio->bi_private = &event;
588 bio->bi_end_io = bi_complete;
589 submit_bio(rw, bio);
590 wait_for_completion(&event);
591
592 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
593 bio_put(bio);
594 return ret;
595 }
596 EXPORT_SYMBOL_GPL(sync_page_io);
597
598 static int read_disk_sb(mdk_rdev_t * rdev, int size)
599 {
600 char b[BDEVNAME_SIZE];
601 if (!rdev->sb_page) {
602 MD_BUG();
603 return -EINVAL;
604 }
605 if (rdev->sb_loaded)
606 return 0;
607
608
609 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
610 goto fail;
611 rdev->sb_loaded = 1;
612 return 0;
613
614 fail:
615 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
616 bdevname(rdev->bdev,b));
617 return -EINVAL;
618 }
619
620 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
621 {
622 return sb1->set_uuid0 == sb2->set_uuid0 &&
623 sb1->set_uuid1 == sb2->set_uuid1 &&
624 sb1->set_uuid2 == sb2->set_uuid2 &&
625 sb1->set_uuid3 == sb2->set_uuid3;
626 }
627
628 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
629 {
630 int ret;
631 mdp_super_t *tmp1, *tmp2;
632
633 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
634 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
635
636 if (!tmp1 || !tmp2) {
637 ret = 0;
638 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
639 goto abort;
640 }
641
642 *tmp1 = *sb1;
643 *tmp2 = *sb2;
644
645 /*
646 * nr_disks is not constant
647 */
648 tmp1->nr_disks = 0;
649 tmp2->nr_disks = 0;
650
651 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
652 abort:
653 kfree(tmp1);
654 kfree(tmp2);
655 return ret;
656 }
657
658
659 static u32 md_csum_fold(u32 csum)
660 {
661 csum = (csum & 0xffff) + (csum >> 16);
662 return (csum & 0xffff) + (csum >> 16);
663 }
664
665 static unsigned int calc_sb_csum(mdp_super_t * sb)
666 {
667 u64 newcsum = 0;
668 u32 *sb32 = (u32*)sb;
669 int i;
670 unsigned int disk_csum, csum;
671
672 disk_csum = sb->sb_csum;
673 sb->sb_csum = 0;
674
675 for (i = 0; i < MD_SB_BYTES/4 ; i++)
676 newcsum += sb32[i];
677 csum = (newcsum & 0xffffffff) + (newcsum>>32);
678
679
680 #ifdef CONFIG_ALPHA
681 /* This used to use csum_partial, which was wrong for several
682 * reasons including that different results are returned on
683 * different architectures. It isn't critical that we get exactly
684 * the same return value as before (we always csum_fold before
685 * testing, and that removes any differences). However as we
686 * know that csum_partial always returned a 16bit value on
687 * alphas, do a fold to maximise conformity to previous behaviour.
688 */
689 sb->sb_csum = md_csum_fold(disk_csum);
690 #else
691 sb->sb_csum = disk_csum;
692 #endif
693 return csum;
694 }
695
696
697 /*
698 * Handle superblock details.
699 * We want to be able to handle multiple superblock formats
700 * so we have a common interface to them all, and an array of
701 * different handlers.
702 * We rely on user-space to write the initial superblock, and support
703 * reading and updating of superblocks.
704 * Interface methods are:
705 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
706 * loads and validates a superblock on dev.
707 * if refdev != NULL, compare superblocks on both devices
708 * Return:
709 * 0 - dev has a superblock that is compatible with refdev
710 * 1 - dev has a superblock that is compatible and newer than refdev
711 * so dev should be used as the refdev in future
712 * -EINVAL superblock incompatible or invalid
713 * -othererror e.g. -EIO
714 *
715 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
716 * Verify that dev is acceptable into mddev.
717 * The first time, mddev->raid_disks will be 0, and data from
718 * dev should be merged in. Subsequent calls check that dev
719 * is new enough. Return 0 or -EINVAL
720 *
721 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
722 * Update the superblock for rdev with data in mddev
723 * This does not write to disc.
724 *
725 */
726
727 struct super_type {
728 char *name;
729 struct module *owner;
730 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
731 int minor_version);
732 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
733 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
734 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
735 sector_t num_sectors);
736 };
737
738 /*
739 * Check that the given mddev has no bitmap.
740 *
741 * This function is called from the run method of all personalities that do not
742 * support bitmaps. It prints an error message and returns non-zero if mddev
743 * has a bitmap. Otherwise, it returns 0.
744 *
745 */
746 int md_check_no_bitmap(mddev_t *mddev)
747 {
748 if (!mddev->bitmap_file && !mddev->bitmap_offset)
749 return 0;
750 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
751 mdname(mddev), mddev->pers->name);
752 return 1;
753 }
754 EXPORT_SYMBOL(md_check_no_bitmap);
755
756 /*
757 * load_super for 0.90.0
758 */
759 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
760 {
761 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
762 mdp_super_t *sb;
763 int ret;
764
765 /*
766 * Calculate the position of the superblock (512byte sectors),
767 * it's at the end of the disk.
768 *
769 * It also happens to be a multiple of 4Kb.
770 */
771 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
772
773 ret = read_disk_sb(rdev, MD_SB_BYTES);
774 if (ret) return ret;
775
776 ret = -EINVAL;
777
778 bdevname(rdev->bdev, b);
779 sb = (mdp_super_t*)page_address(rdev->sb_page);
780
781 if (sb->md_magic != MD_SB_MAGIC) {
782 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
783 b);
784 goto abort;
785 }
786
787 if (sb->major_version != 0 ||
788 sb->minor_version < 90 ||
789 sb->minor_version > 91) {
790 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
791 sb->major_version, sb->minor_version,
792 b);
793 goto abort;
794 }
795
796 if (sb->raid_disks <= 0)
797 goto abort;
798
799 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
800 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
801 b);
802 goto abort;
803 }
804
805 rdev->preferred_minor = sb->md_minor;
806 rdev->data_offset = 0;
807 rdev->sb_size = MD_SB_BYTES;
808
809 if (sb->level == LEVEL_MULTIPATH)
810 rdev->desc_nr = -1;
811 else
812 rdev->desc_nr = sb->this_disk.number;
813
814 if (!refdev) {
815 ret = 1;
816 } else {
817 __u64 ev1, ev2;
818 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
819 if (!uuid_equal(refsb, sb)) {
820 printk(KERN_WARNING "md: %s has different UUID to %s\n",
821 b, bdevname(refdev->bdev,b2));
822 goto abort;
823 }
824 if (!sb_equal(refsb, sb)) {
825 printk(KERN_WARNING "md: %s has same UUID"
826 " but different superblock to %s\n",
827 b, bdevname(refdev->bdev, b2));
828 goto abort;
829 }
830 ev1 = md_event(sb);
831 ev2 = md_event(refsb);
832 if (ev1 > ev2)
833 ret = 1;
834 else
835 ret = 0;
836 }
837 rdev->sectors = rdev->sb_start;
838
839 if (rdev->sectors < sb->size * 2 && sb->level > 1)
840 /* "this cannot possibly happen" ... */
841 ret = -EINVAL;
842
843 abort:
844 return ret;
845 }
846
847 /*
848 * validate_super for 0.90.0
849 */
850 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
851 {
852 mdp_disk_t *desc;
853 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
854 __u64 ev1 = md_event(sb);
855
856 rdev->raid_disk = -1;
857 clear_bit(Faulty, &rdev->flags);
858 clear_bit(In_sync, &rdev->flags);
859 clear_bit(WriteMostly, &rdev->flags);
860 clear_bit(BarriersNotsupp, &rdev->flags);
861
862 if (mddev->raid_disks == 0) {
863 mddev->major_version = 0;
864 mddev->minor_version = sb->minor_version;
865 mddev->patch_version = sb->patch_version;
866 mddev->external = 0;
867 mddev->chunk_sectors = sb->chunk_size >> 9;
868 mddev->ctime = sb->ctime;
869 mddev->utime = sb->utime;
870 mddev->level = sb->level;
871 mddev->clevel[0] = 0;
872 mddev->layout = sb->layout;
873 mddev->raid_disks = sb->raid_disks;
874 mddev->dev_sectors = sb->size * 2;
875 mddev->events = ev1;
876 mddev->bitmap_offset = 0;
877 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
878
879 if (mddev->minor_version >= 91) {
880 mddev->reshape_position = sb->reshape_position;
881 mddev->delta_disks = sb->delta_disks;
882 mddev->new_level = sb->new_level;
883 mddev->new_layout = sb->new_layout;
884 mddev->new_chunk_sectors = sb->new_chunk >> 9;
885 } else {
886 mddev->reshape_position = MaxSector;
887 mddev->delta_disks = 0;
888 mddev->new_level = mddev->level;
889 mddev->new_layout = mddev->layout;
890 mddev->new_chunk_sectors = mddev->chunk_sectors;
891 }
892
893 if (sb->state & (1<<MD_SB_CLEAN))
894 mddev->recovery_cp = MaxSector;
895 else {
896 if (sb->events_hi == sb->cp_events_hi &&
897 sb->events_lo == sb->cp_events_lo) {
898 mddev->recovery_cp = sb->recovery_cp;
899 } else
900 mddev->recovery_cp = 0;
901 }
902
903 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
904 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
905 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
906 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
907
908 mddev->max_disks = MD_SB_DISKS;
909
910 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
911 mddev->bitmap_file == NULL)
912 mddev->bitmap_offset = mddev->default_bitmap_offset;
913
914 } else if (mddev->pers == NULL) {
915 /* Insist on good event counter while assembling */
916 ++ev1;
917 if (ev1 < mddev->events)
918 return -EINVAL;
919 } else if (mddev->bitmap) {
920 /* if adding to array with a bitmap, then we can accept an
921 * older device ... but not too old.
922 */
923 if (ev1 < mddev->bitmap->events_cleared)
924 return 0;
925 } else {
926 if (ev1 < mddev->events)
927 /* just a hot-add of a new device, leave raid_disk at -1 */
928 return 0;
929 }
930
931 if (mddev->level != LEVEL_MULTIPATH) {
932 desc = sb->disks + rdev->desc_nr;
933
934 if (desc->state & (1<<MD_DISK_FAULTY))
935 set_bit(Faulty, &rdev->flags);
936 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
937 desc->raid_disk < mddev->raid_disks */) {
938 set_bit(In_sync, &rdev->flags);
939 rdev->raid_disk = desc->raid_disk;
940 }
941 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
942 set_bit(WriteMostly, &rdev->flags);
943 } else /* MULTIPATH are always insync */
944 set_bit(In_sync, &rdev->flags);
945 return 0;
946 }
947
948 /*
949 * sync_super for 0.90.0
950 */
951 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
952 {
953 mdp_super_t *sb;
954 mdk_rdev_t *rdev2;
955 int next_spare = mddev->raid_disks;
956
957
958 /* make rdev->sb match mddev data..
959 *
960 * 1/ zero out disks
961 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
962 * 3/ any empty disks < next_spare become removed
963 *
964 * disks[0] gets initialised to REMOVED because
965 * we cannot be sure from other fields if it has
966 * been initialised or not.
967 */
968 int i;
969 int active=0, working=0,failed=0,spare=0,nr_disks=0;
970
971 rdev->sb_size = MD_SB_BYTES;
972
973 sb = (mdp_super_t*)page_address(rdev->sb_page);
974
975 memset(sb, 0, sizeof(*sb));
976
977 sb->md_magic = MD_SB_MAGIC;
978 sb->major_version = mddev->major_version;
979 sb->patch_version = mddev->patch_version;
980 sb->gvalid_words = 0; /* ignored */
981 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
982 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
983 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
984 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
985
986 sb->ctime = mddev->ctime;
987 sb->level = mddev->level;
988 sb->size = mddev->dev_sectors / 2;
989 sb->raid_disks = mddev->raid_disks;
990 sb->md_minor = mddev->md_minor;
991 sb->not_persistent = 0;
992 sb->utime = mddev->utime;
993 sb->state = 0;
994 sb->events_hi = (mddev->events>>32);
995 sb->events_lo = (u32)mddev->events;
996
997 if (mddev->reshape_position == MaxSector)
998 sb->minor_version = 90;
999 else {
1000 sb->minor_version = 91;
1001 sb->reshape_position = mddev->reshape_position;
1002 sb->new_level = mddev->new_level;
1003 sb->delta_disks = mddev->delta_disks;
1004 sb->new_layout = mddev->new_layout;
1005 sb->new_chunk = mddev->new_chunk_sectors << 9;
1006 }
1007 mddev->minor_version = sb->minor_version;
1008 if (mddev->in_sync)
1009 {
1010 sb->recovery_cp = mddev->recovery_cp;
1011 sb->cp_events_hi = (mddev->events>>32);
1012 sb->cp_events_lo = (u32)mddev->events;
1013 if (mddev->recovery_cp == MaxSector)
1014 sb->state = (1<< MD_SB_CLEAN);
1015 } else
1016 sb->recovery_cp = 0;
1017
1018 sb->layout = mddev->layout;
1019 sb->chunk_size = mddev->chunk_sectors << 9;
1020
1021 if (mddev->bitmap && mddev->bitmap_file == NULL)
1022 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1023
1024 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1025 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1026 mdp_disk_t *d;
1027 int desc_nr;
1028 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1029 && !test_bit(Faulty, &rdev2->flags))
1030 desc_nr = rdev2->raid_disk;
1031 else
1032 desc_nr = next_spare++;
1033 rdev2->desc_nr = desc_nr;
1034 d = &sb->disks[rdev2->desc_nr];
1035 nr_disks++;
1036 d->number = rdev2->desc_nr;
1037 d->major = MAJOR(rdev2->bdev->bd_dev);
1038 d->minor = MINOR(rdev2->bdev->bd_dev);
1039 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1040 && !test_bit(Faulty, &rdev2->flags))
1041 d->raid_disk = rdev2->raid_disk;
1042 else
1043 d->raid_disk = rdev2->desc_nr; /* compatibility */
1044 if (test_bit(Faulty, &rdev2->flags))
1045 d->state = (1<<MD_DISK_FAULTY);
1046 else if (test_bit(In_sync, &rdev2->flags)) {
1047 d->state = (1<<MD_DISK_ACTIVE);
1048 d->state |= (1<<MD_DISK_SYNC);
1049 active++;
1050 working++;
1051 } else {
1052 d->state = 0;
1053 spare++;
1054 working++;
1055 }
1056 if (test_bit(WriteMostly, &rdev2->flags))
1057 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1058 }
1059 /* now set the "removed" and "faulty" bits on any missing devices */
1060 for (i=0 ; i < mddev->raid_disks ; i++) {
1061 mdp_disk_t *d = &sb->disks[i];
1062 if (d->state == 0 && d->number == 0) {
1063 d->number = i;
1064 d->raid_disk = i;
1065 d->state = (1<<MD_DISK_REMOVED);
1066 d->state |= (1<<MD_DISK_FAULTY);
1067 failed++;
1068 }
1069 }
1070 sb->nr_disks = nr_disks;
1071 sb->active_disks = active;
1072 sb->working_disks = working;
1073 sb->failed_disks = failed;
1074 sb->spare_disks = spare;
1075
1076 sb->this_disk = sb->disks[rdev->desc_nr];
1077 sb->sb_csum = calc_sb_csum(sb);
1078 }
1079
1080 /*
1081 * rdev_size_change for 0.90.0
1082 */
1083 static unsigned long long
1084 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1085 {
1086 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1087 return 0; /* component must fit device */
1088 if (rdev->mddev->bitmap_offset)
1089 return 0; /* can't move bitmap */
1090 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1091 if (!num_sectors || num_sectors > rdev->sb_start)
1092 num_sectors = rdev->sb_start;
1093 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1094 rdev->sb_page);
1095 md_super_wait(rdev->mddev);
1096 return num_sectors / 2; /* kB for sysfs */
1097 }
1098
1099
1100 /*
1101 * version 1 superblock
1102 */
1103
1104 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1105 {
1106 __le32 disk_csum;
1107 u32 csum;
1108 unsigned long long newcsum;
1109 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1110 __le32 *isuper = (__le32*)sb;
1111 int i;
1112
1113 disk_csum = sb->sb_csum;
1114 sb->sb_csum = 0;
1115 newcsum = 0;
1116 for (i=0; size>=4; size -= 4 )
1117 newcsum += le32_to_cpu(*isuper++);
1118
1119 if (size == 2)
1120 newcsum += le16_to_cpu(*(__le16*) isuper);
1121
1122 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1123 sb->sb_csum = disk_csum;
1124 return cpu_to_le32(csum);
1125 }
1126
1127 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1128 {
1129 struct mdp_superblock_1 *sb;
1130 int ret;
1131 sector_t sb_start;
1132 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1133 int bmask;
1134
1135 /*
1136 * Calculate the position of the superblock in 512byte sectors.
1137 * It is always aligned to a 4K boundary and
1138 * depeding on minor_version, it can be:
1139 * 0: At least 8K, but less than 12K, from end of device
1140 * 1: At start of device
1141 * 2: 4K from start of device.
1142 */
1143 switch(minor_version) {
1144 case 0:
1145 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1146 sb_start -= 8*2;
1147 sb_start &= ~(sector_t)(4*2-1);
1148 break;
1149 case 1:
1150 sb_start = 0;
1151 break;
1152 case 2:
1153 sb_start = 8;
1154 break;
1155 default:
1156 return -EINVAL;
1157 }
1158 rdev->sb_start = sb_start;
1159
1160 /* superblock is rarely larger than 1K, but it can be larger,
1161 * and it is safe to read 4k, so we do that
1162 */
1163 ret = read_disk_sb(rdev, 4096);
1164 if (ret) return ret;
1165
1166
1167 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1168
1169 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1170 sb->major_version != cpu_to_le32(1) ||
1171 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1172 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1173 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1174 return -EINVAL;
1175
1176 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1177 printk("md: invalid superblock checksum on %s\n",
1178 bdevname(rdev->bdev,b));
1179 return -EINVAL;
1180 }
1181 if (le64_to_cpu(sb->data_size) < 10) {
1182 printk("md: data_size too small on %s\n",
1183 bdevname(rdev->bdev,b));
1184 return -EINVAL;
1185 }
1186
1187 rdev->preferred_minor = 0xffff;
1188 rdev->data_offset = le64_to_cpu(sb->data_offset);
1189 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1190
1191 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1192 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1193 if (rdev->sb_size & bmask)
1194 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1195
1196 if (minor_version
1197 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1198 return -EINVAL;
1199
1200 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1201 rdev->desc_nr = -1;
1202 else
1203 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1204
1205 if (!refdev) {
1206 ret = 1;
1207 } else {
1208 __u64 ev1, ev2;
1209 struct mdp_superblock_1 *refsb =
1210 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1211
1212 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1213 sb->level != refsb->level ||
1214 sb->layout != refsb->layout ||
1215 sb->chunksize != refsb->chunksize) {
1216 printk(KERN_WARNING "md: %s has strangely different"
1217 " superblock to %s\n",
1218 bdevname(rdev->bdev,b),
1219 bdevname(refdev->bdev,b2));
1220 return -EINVAL;
1221 }
1222 ev1 = le64_to_cpu(sb->events);
1223 ev2 = le64_to_cpu(refsb->events);
1224
1225 if (ev1 > ev2)
1226 ret = 1;
1227 else
1228 ret = 0;
1229 }
1230 if (minor_version)
1231 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1232 le64_to_cpu(sb->data_offset);
1233 else
1234 rdev->sectors = rdev->sb_start;
1235 if (rdev->sectors < le64_to_cpu(sb->data_size))
1236 return -EINVAL;
1237 rdev->sectors = le64_to_cpu(sb->data_size);
1238 if (le64_to_cpu(sb->size) > rdev->sectors)
1239 return -EINVAL;
1240 return ret;
1241 }
1242
1243 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1244 {
1245 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1246 __u64 ev1 = le64_to_cpu(sb->events);
1247
1248 rdev->raid_disk = -1;
1249 clear_bit(Faulty, &rdev->flags);
1250 clear_bit(In_sync, &rdev->flags);
1251 clear_bit(WriteMostly, &rdev->flags);
1252 clear_bit(BarriersNotsupp, &rdev->flags);
1253
1254 if (mddev->raid_disks == 0) {
1255 mddev->major_version = 1;
1256 mddev->patch_version = 0;
1257 mddev->external = 0;
1258 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1259 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1260 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1261 mddev->level = le32_to_cpu(sb->level);
1262 mddev->clevel[0] = 0;
1263 mddev->layout = le32_to_cpu(sb->layout);
1264 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1265 mddev->dev_sectors = le64_to_cpu(sb->size);
1266 mddev->events = ev1;
1267 mddev->bitmap_offset = 0;
1268 mddev->default_bitmap_offset = 1024 >> 9;
1269
1270 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1271 memcpy(mddev->uuid, sb->set_uuid, 16);
1272
1273 mddev->max_disks = (4096-256)/2;
1274
1275 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1276 mddev->bitmap_file == NULL )
1277 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1278
1279 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1280 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1281 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1282 mddev->new_level = le32_to_cpu(sb->new_level);
1283 mddev->new_layout = le32_to_cpu(sb->new_layout);
1284 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1285 } else {
1286 mddev->reshape_position = MaxSector;
1287 mddev->delta_disks = 0;
1288 mddev->new_level = mddev->level;
1289 mddev->new_layout = mddev->layout;
1290 mddev->new_chunk_sectors = mddev->chunk_sectors;
1291 }
1292
1293 } else if (mddev->pers == NULL) {
1294 /* Insist of good event counter while assembling */
1295 ++ev1;
1296 if (ev1 < mddev->events)
1297 return -EINVAL;
1298 } else if (mddev->bitmap) {
1299 /* If adding to array with a bitmap, then we can accept an
1300 * older device, but not too old.
1301 */
1302 if (ev1 < mddev->bitmap->events_cleared)
1303 return 0;
1304 } else {
1305 if (ev1 < mddev->events)
1306 /* just a hot-add of a new device, leave raid_disk at -1 */
1307 return 0;
1308 }
1309 if (mddev->level != LEVEL_MULTIPATH) {
1310 int role;
1311 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1312 switch(role) {
1313 case 0xffff: /* spare */
1314 break;
1315 case 0xfffe: /* faulty */
1316 set_bit(Faulty, &rdev->flags);
1317 break;
1318 default:
1319 if ((le32_to_cpu(sb->feature_map) &
1320 MD_FEATURE_RECOVERY_OFFSET))
1321 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1322 else
1323 set_bit(In_sync, &rdev->flags);
1324 rdev->raid_disk = role;
1325 break;
1326 }
1327 if (sb->devflags & WriteMostly1)
1328 set_bit(WriteMostly, &rdev->flags);
1329 } else /* MULTIPATH are always insync */
1330 set_bit(In_sync, &rdev->flags);
1331
1332 return 0;
1333 }
1334
1335 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1336 {
1337 struct mdp_superblock_1 *sb;
1338 mdk_rdev_t *rdev2;
1339 int max_dev, i;
1340 /* make rdev->sb match mddev and rdev data. */
1341
1342 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1343
1344 sb->feature_map = 0;
1345 sb->pad0 = 0;
1346 sb->recovery_offset = cpu_to_le64(0);
1347 memset(sb->pad1, 0, sizeof(sb->pad1));
1348 memset(sb->pad2, 0, sizeof(sb->pad2));
1349 memset(sb->pad3, 0, sizeof(sb->pad3));
1350
1351 sb->utime = cpu_to_le64((__u64)mddev->utime);
1352 sb->events = cpu_to_le64(mddev->events);
1353 if (mddev->in_sync)
1354 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1355 else
1356 sb->resync_offset = cpu_to_le64(0);
1357
1358 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1359
1360 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1361 sb->size = cpu_to_le64(mddev->dev_sectors);
1362 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1363 sb->level = cpu_to_le32(mddev->level);
1364 sb->layout = cpu_to_le32(mddev->layout);
1365
1366 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1367 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1368 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1369 }
1370
1371 if (rdev->raid_disk >= 0 &&
1372 !test_bit(In_sync, &rdev->flags)) {
1373 if (mddev->curr_resync_completed > rdev->recovery_offset)
1374 rdev->recovery_offset = mddev->curr_resync_completed;
1375 if (rdev->recovery_offset > 0) {
1376 sb->feature_map |=
1377 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1378 sb->recovery_offset =
1379 cpu_to_le64(rdev->recovery_offset);
1380 }
1381 }
1382
1383 if (mddev->reshape_position != MaxSector) {
1384 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1385 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1386 sb->new_layout = cpu_to_le32(mddev->new_layout);
1387 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1388 sb->new_level = cpu_to_le32(mddev->new_level);
1389 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1390 }
1391
1392 max_dev = 0;
1393 list_for_each_entry(rdev2, &mddev->disks, same_set)
1394 if (rdev2->desc_nr+1 > max_dev)
1395 max_dev = rdev2->desc_nr+1;
1396
1397 if (max_dev > le32_to_cpu(sb->max_dev))
1398 sb->max_dev = cpu_to_le32(max_dev);
1399 for (i=0; i<max_dev;i++)
1400 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1401
1402 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1403 i = rdev2->desc_nr;
1404 if (test_bit(Faulty, &rdev2->flags))
1405 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1406 else if (test_bit(In_sync, &rdev2->flags))
1407 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1408 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1409 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1410 else
1411 sb->dev_roles[i] = cpu_to_le16(0xffff);
1412 }
1413
1414 sb->sb_csum = calc_sb_1_csum(sb);
1415 }
1416
1417 static unsigned long long
1418 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1419 {
1420 struct mdp_superblock_1 *sb;
1421 sector_t max_sectors;
1422 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1423 return 0; /* component must fit device */
1424 if (rdev->sb_start < rdev->data_offset) {
1425 /* minor versions 1 and 2; superblock before data */
1426 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1427 max_sectors -= rdev->data_offset;
1428 if (!num_sectors || num_sectors > max_sectors)
1429 num_sectors = max_sectors;
1430 } else if (rdev->mddev->bitmap_offset) {
1431 /* minor version 0 with bitmap we can't move */
1432 return 0;
1433 } else {
1434 /* minor version 0; superblock after data */
1435 sector_t sb_start;
1436 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1437 sb_start &= ~(sector_t)(4*2 - 1);
1438 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1439 if (!num_sectors || num_sectors > max_sectors)
1440 num_sectors = max_sectors;
1441 rdev->sb_start = sb_start;
1442 }
1443 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1444 sb->data_size = cpu_to_le64(num_sectors);
1445 sb->super_offset = rdev->sb_start;
1446 sb->sb_csum = calc_sb_1_csum(sb);
1447 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1448 rdev->sb_page);
1449 md_super_wait(rdev->mddev);
1450 return num_sectors / 2; /* kB for sysfs */
1451 }
1452
1453 static struct super_type super_types[] = {
1454 [0] = {
1455 .name = "0.90.0",
1456 .owner = THIS_MODULE,
1457 .load_super = super_90_load,
1458 .validate_super = super_90_validate,
1459 .sync_super = super_90_sync,
1460 .rdev_size_change = super_90_rdev_size_change,
1461 },
1462 [1] = {
1463 .name = "md-1",
1464 .owner = THIS_MODULE,
1465 .load_super = super_1_load,
1466 .validate_super = super_1_validate,
1467 .sync_super = super_1_sync,
1468 .rdev_size_change = super_1_rdev_size_change,
1469 },
1470 };
1471
1472 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1473 {
1474 mdk_rdev_t *rdev, *rdev2;
1475
1476 rcu_read_lock();
1477 rdev_for_each_rcu(rdev, mddev1)
1478 rdev_for_each_rcu(rdev2, mddev2)
1479 if (rdev->bdev->bd_contains ==
1480 rdev2->bdev->bd_contains) {
1481 rcu_read_unlock();
1482 return 1;
1483 }
1484 rcu_read_unlock();
1485 return 0;
1486 }
1487
1488 static LIST_HEAD(pending_raid_disks);
1489
1490 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1491 {
1492 struct mdk_personality *pers = mddev->pers;
1493 struct gendisk *disk = mddev->gendisk;
1494 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1495 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1496
1497 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1498 if (pers && pers->level >= 4 && pers->level <= 6)
1499 return;
1500
1501 /* If rdev is integrity capable, register profile for mddev */
1502 if (!bi_mddev && bi_rdev) {
1503 if (blk_integrity_register(disk, bi_rdev))
1504 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1505 __func__, disk->disk_name);
1506 else
1507 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1508 disk->disk_name);
1509 return;
1510 }
1511
1512 /* Check that mddev and rdev have matching profiles */
1513 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1514 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1515 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1516 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1517 disk->disk_name);
1518 blk_integrity_unregister(disk);
1519 }
1520 }
1521
1522 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1523 {
1524 char b[BDEVNAME_SIZE];
1525 struct kobject *ko;
1526 char *s;
1527 int err;
1528
1529 if (rdev->mddev) {
1530 MD_BUG();
1531 return -EINVAL;
1532 }
1533
1534 /* prevent duplicates */
1535 if (find_rdev(mddev, rdev->bdev->bd_dev))
1536 return -EEXIST;
1537
1538 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1539 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1540 rdev->sectors < mddev->dev_sectors)) {
1541 if (mddev->pers) {
1542 /* Cannot change size, so fail
1543 * If mddev->level <= 0, then we don't care
1544 * about aligning sizes (e.g. linear)
1545 */
1546 if (mddev->level > 0)
1547 return -ENOSPC;
1548 } else
1549 mddev->dev_sectors = rdev->sectors;
1550 }
1551
1552 /* Verify rdev->desc_nr is unique.
1553 * If it is -1, assign a free number, else
1554 * check number is not in use
1555 */
1556 if (rdev->desc_nr < 0) {
1557 int choice = 0;
1558 if (mddev->pers) choice = mddev->raid_disks;
1559 while (find_rdev_nr(mddev, choice))
1560 choice++;
1561 rdev->desc_nr = choice;
1562 } else {
1563 if (find_rdev_nr(mddev, rdev->desc_nr))
1564 return -EBUSY;
1565 }
1566 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1567 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1568 mdname(mddev), mddev->max_disks);
1569 return -EBUSY;
1570 }
1571 bdevname(rdev->bdev,b);
1572 while ( (s=strchr(b, '/')) != NULL)
1573 *s = '!';
1574
1575 rdev->mddev = mddev;
1576 printk(KERN_INFO "md: bind<%s>\n", b);
1577
1578 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1579 goto fail;
1580
1581 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1582 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1583 kobject_del(&rdev->kobj);
1584 goto fail;
1585 }
1586 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1587
1588 list_add_rcu(&rdev->same_set, &mddev->disks);
1589 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1590
1591 /* May as well allow recovery to be retried once */
1592 mddev->recovery_disabled = 0;
1593
1594 md_integrity_check(rdev, mddev);
1595 return 0;
1596
1597 fail:
1598 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1599 b, mdname(mddev));
1600 return err;
1601 }
1602
1603 static void md_delayed_delete(struct work_struct *ws)
1604 {
1605 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1606 kobject_del(&rdev->kobj);
1607 kobject_put(&rdev->kobj);
1608 }
1609
1610 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1611 {
1612 char b[BDEVNAME_SIZE];
1613 if (!rdev->mddev) {
1614 MD_BUG();
1615 return;
1616 }
1617 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1618 list_del_rcu(&rdev->same_set);
1619 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1620 rdev->mddev = NULL;
1621 sysfs_remove_link(&rdev->kobj, "block");
1622 sysfs_put(rdev->sysfs_state);
1623 rdev->sysfs_state = NULL;
1624 /* We need to delay this, otherwise we can deadlock when
1625 * writing to 'remove' to "dev/state". We also need
1626 * to delay it due to rcu usage.
1627 */
1628 synchronize_rcu();
1629 INIT_WORK(&rdev->del_work, md_delayed_delete);
1630 kobject_get(&rdev->kobj);
1631 schedule_work(&rdev->del_work);
1632 }
1633
1634 /*
1635 * prevent the device from being mounted, repartitioned or
1636 * otherwise reused by a RAID array (or any other kernel
1637 * subsystem), by bd_claiming the device.
1638 */
1639 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1640 {
1641 int err = 0;
1642 struct block_device *bdev;
1643 char b[BDEVNAME_SIZE];
1644
1645 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1646 if (IS_ERR(bdev)) {
1647 printk(KERN_ERR "md: could not open %s.\n",
1648 __bdevname(dev, b));
1649 return PTR_ERR(bdev);
1650 }
1651 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1652 if (err) {
1653 printk(KERN_ERR "md: could not bd_claim %s.\n",
1654 bdevname(bdev, b));
1655 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1656 return err;
1657 }
1658 if (!shared)
1659 set_bit(AllReserved, &rdev->flags);
1660 rdev->bdev = bdev;
1661 return err;
1662 }
1663
1664 static void unlock_rdev(mdk_rdev_t *rdev)
1665 {
1666 struct block_device *bdev = rdev->bdev;
1667 rdev->bdev = NULL;
1668 if (!bdev)
1669 MD_BUG();
1670 bd_release(bdev);
1671 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1672 }
1673
1674 void md_autodetect_dev(dev_t dev);
1675
1676 static void export_rdev(mdk_rdev_t * rdev)
1677 {
1678 char b[BDEVNAME_SIZE];
1679 printk(KERN_INFO "md: export_rdev(%s)\n",
1680 bdevname(rdev->bdev,b));
1681 if (rdev->mddev)
1682 MD_BUG();
1683 free_disk_sb(rdev);
1684 #ifndef MODULE
1685 if (test_bit(AutoDetected, &rdev->flags))
1686 md_autodetect_dev(rdev->bdev->bd_dev);
1687 #endif
1688 unlock_rdev(rdev);
1689 kobject_put(&rdev->kobj);
1690 }
1691
1692 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1693 {
1694 unbind_rdev_from_array(rdev);
1695 export_rdev(rdev);
1696 }
1697
1698 static void export_array(mddev_t *mddev)
1699 {
1700 mdk_rdev_t *rdev, *tmp;
1701
1702 rdev_for_each(rdev, tmp, mddev) {
1703 if (!rdev->mddev) {
1704 MD_BUG();
1705 continue;
1706 }
1707 kick_rdev_from_array(rdev);
1708 }
1709 if (!list_empty(&mddev->disks))
1710 MD_BUG();
1711 mddev->raid_disks = 0;
1712 mddev->major_version = 0;
1713 }
1714
1715 static void print_desc(mdp_disk_t *desc)
1716 {
1717 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1718 desc->major,desc->minor,desc->raid_disk,desc->state);
1719 }
1720
1721 static void print_sb_90(mdp_super_t *sb)
1722 {
1723 int i;
1724
1725 printk(KERN_INFO
1726 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1727 sb->major_version, sb->minor_version, sb->patch_version,
1728 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1729 sb->ctime);
1730 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1731 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1732 sb->md_minor, sb->layout, sb->chunk_size);
1733 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1734 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1735 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1736 sb->failed_disks, sb->spare_disks,
1737 sb->sb_csum, (unsigned long)sb->events_lo);
1738
1739 printk(KERN_INFO);
1740 for (i = 0; i < MD_SB_DISKS; i++) {
1741 mdp_disk_t *desc;
1742
1743 desc = sb->disks + i;
1744 if (desc->number || desc->major || desc->minor ||
1745 desc->raid_disk || (desc->state && (desc->state != 4))) {
1746 printk(" D %2d: ", i);
1747 print_desc(desc);
1748 }
1749 }
1750 printk(KERN_INFO "md: THIS: ");
1751 print_desc(&sb->this_disk);
1752 }
1753
1754 static void print_sb_1(struct mdp_superblock_1 *sb)
1755 {
1756 __u8 *uuid;
1757
1758 uuid = sb->set_uuid;
1759 printk(KERN_INFO
1760 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1761 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1762 "md: Name: \"%s\" CT:%llu\n",
1763 le32_to_cpu(sb->major_version),
1764 le32_to_cpu(sb->feature_map),
1765 uuid[0], uuid[1], uuid[2], uuid[3],
1766 uuid[4], uuid[5], uuid[6], uuid[7],
1767 uuid[8], uuid[9], uuid[10], uuid[11],
1768 uuid[12], uuid[13], uuid[14], uuid[15],
1769 sb->set_name,
1770 (unsigned long long)le64_to_cpu(sb->ctime)
1771 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1772
1773 uuid = sb->device_uuid;
1774 printk(KERN_INFO
1775 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1776 " RO:%llu\n"
1777 "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1778 ":%02x%02x%02x%02x%02x%02x\n"
1779 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1780 "md: (MaxDev:%u) \n",
1781 le32_to_cpu(sb->level),
1782 (unsigned long long)le64_to_cpu(sb->size),
1783 le32_to_cpu(sb->raid_disks),
1784 le32_to_cpu(sb->layout),
1785 le32_to_cpu(sb->chunksize),
1786 (unsigned long long)le64_to_cpu(sb->data_offset),
1787 (unsigned long long)le64_to_cpu(sb->data_size),
1788 (unsigned long long)le64_to_cpu(sb->super_offset),
1789 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1790 le32_to_cpu(sb->dev_number),
1791 uuid[0], uuid[1], uuid[2], uuid[3],
1792 uuid[4], uuid[5], uuid[6], uuid[7],
1793 uuid[8], uuid[9], uuid[10], uuid[11],
1794 uuid[12], uuid[13], uuid[14], uuid[15],
1795 sb->devflags,
1796 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1797 (unsigned long long)le64_to_cpu(sb->events),
1798 (unsigned long long)le64_to_cpu(sb->resync_offset),
1799 le32_to_cpu(sb->sb_csum),
1800 le32_to_cpu(sb->max_dev)
1801 );
1802 }
1803
1804 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1805 {
1806 char b[BDEVNAME_SIZE];
1807 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1808 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1809 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1810 rdev->desc_nr);
1811 if (rdev->sb_loaded) {
1812 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1813 switch (major_version) {
1814 case 0:
1815 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1816 break;
1817 case 1:
1818 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1819 break;
1820 }
1821 } else
1822 printk(KERN_INFO "md: no rdev superblock!\n");
1823 }
1824
1825 static void md_print_devices(void)
1826 {
1827 struct list_head *tmp;
1828 mdk_rdev_t *rdev;
1829 mddev_t *mddev;
1830 char b[BDEVNAME_SIZE];
1831
1832 printk("\n");
1833 printk("md: **********************************\n");
1834 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1835 printk("md: **********************************\n");
1836 for_each_mddev(mddev, tmp) {
1837
1838 if (mddev->bitmap)
1839 bitmap_print_sb(mddev->bitmap);
1840 else
1841 printk("%s: ", mdname(mddev));
1842 list_for_each_entry(rdev, &mddev->disks, same_set)
1843 printk("<%s>", bdevname(rdev->bdev,b));
1844 printk("\n");
1845
1846 list_for_each_entry(rdev, &mddev->disks, same_set)
1847 print_rdev(rdev, mddev->major_version);
1848 }
1849 printk("md: **********************************\n");
1850 printk("\n");
1851 }
1852
1853
1854 static void sync_sbs(mddev_t * mddev, int nospares)
1855 {
1856 /* Update each superblock (in-memory image), but
1857 * if we are allowed to, skip spares which already
1858 * have the right event counter, or have one earlier
1859 * (which would mean they aren't being marked as dirty
1860 * with the rest of the array)
1861 */
1862 mdk_rdev_t *rdev;
1863
1864 list_for_each_entry(rdev, &mddev->disks, same_set) {
1865 if (rdev->sb_events == mddev->events ||
1866 (nospares &&
1867 rdev->raid_disk < 0 &&
1868 (rdev->sb_events&1)==0 &&
1869 rdev->sb_events+1 == mddev->events)) {
1870 /* Don't update this superblock */
1871 rdev->sb_loaded = 2;
1872 } else {
1873 super_types[mddev->major_version].
1874 sync_super(mddev, rdev);
1875 rdev->sb_loaded = 1;
1876 }
1877 }
1878 }
1879
1880 static void md_update_sb(mddev_t * mddev, int force_change)
1881 {
1882 mdk_rdev_t *rdev;
1883 int sync_req;
1884 int nospares = 0;
1885
1886 mddev->utime = get_seconds();
1887 if (mddev->external)
1888 return;
1889 repeat:
1890 spin_lock_irq(&mddev->write_lock);
1891
1892 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1893 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1894 force_change = 1;
1895 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1896 /* just a clean<-> dirty transition, possibly leave spares alone,
1897 * though if events isn't the right even/odd, we will have to do
1898 * spares after all
1899 */
1900 nospares = 1;
1901 if (force_change)
1902 nospares = 0;
1903 if (mddev->degraded)
1904 /* If the array is degraded, then skipping spares is both
1905 * dangerous and fairly pointless.
1906 * Dangerous because a device that was removed from the array
1907 * might have a event_count that still looks up-to-date,
1908 * so it can be re-added without a resync.
1909 * Pointless because if there are any spares to skip,
1910 * then a recovery will happen and soon that array won't
1911 * be degraded any more and the spare can go back to sleep then.
1912 */
1913 nospares = 0;
1914
1915 sync_req = mddev->in_sync;
1916
1917 /* If this is just a dirty<->clean transition, and the array is clean
1918 * and 'events' is odd, we can roll back to the previous clean state */
1919 if (nospares
1920 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1921 && (mddev->events & 1)
1922 && mddev->events != 1)
1923 mddev->events--;
1924 else {
1925 /* otherwise we have to go forward and ... */
1926 mddev->events ++;
1927 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1928 /* .. if the array isn't clean, insist on an odd 'events' */
1929 if ((mddev->events&1)==0) {
1930 mddev->events++;
1931 nospares = 0;
1932 }
1933 } else {
1934 /* otherwise insist on an even 'events' (for clean states) */
1935 if ((mddev->events&1)) {
1936 mddev->events++;
1937 nospares = 0;
1938 }
1939 }
1940 }
1941
1942 if (!mddev->events) {
1943 /*
1944 * oops, this 64-bit counter should never wrap.
1945 * Either we are in around ~1 trillion A.C., assuming
1946 * 1 reboot per second, or we have a bug:
1947 */
1948 MD_BUG();
1949 mddev->events --;
1950 }
1951
1952 /*
1953 * do not write anything to disk if using
1954 * nonpersistent superblocks
1955 */
1956 if (!mddev->persistent) {
1957 if (!mddev->external)
1958 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1959
1960 spin_unlock_irq(&mddev->write_lock);
1961 wake_up(&mddev->sb_wait);
1962 return;
1963 }
1964 sync_sbs(mddev, nospares);
1965 spin_unlock_irq(&mddev->write_lock);
1966
1967 dprintk(KERN_INFO
1968 "md: updating %s RAID superblock on device (in sync %d)\n",
1969 mdname(mddev),mddev->in_sync);
1970
1971 bitmap_update_sb(mddev->bitmap);
1972 list_for_each_entry(rdev, &mddev->disks, same_set) {
1973 char b[BDEVNAME_SIZE];
1974 dprintk(KERN_INFO "md: ");
1975 if (rdev->sb_loaded != 1)
1976 continue; /* no noise on spare devices */
1977 if (test_bit(Faulty, &rdev->flags))
1978 dprintk("(skipping faulty ");
1979
1980 dprintk("%s ", bdevname(rdev->bdev,b));
1981 if (!test_bit(Faulty, &rdev->flags)) {
1982 md_super_write(mddev,rdev,
1983 rdev->sb_start, rdev->sb_size,
1984 rdev->sb_page);
1985 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1986 bdevname(rdev->bdev,b),
1987 (unsigned long long)rdev->sb_start);
1988 rdev->sb_events = mddev->events;
1989
1990 } else
1991 dprintk(")\n");
1992 if (mddev->level == LEVEL_MULTIPATH)
1993 /* only need to write one superblock... */
1994 break;
1995 }
1996 md_super_wait(mddev);
1997 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1998
1999 spin_lock_irq(&mddev->write_lock);
2000 if (mddev->in_sync != sync_req ||
2001 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2002 /* have to write it out again */
2003 spin_unlock_irq(&mddev->write_lock);
2004 goto repeat;
2005 }
2006 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2007 spin_unlock_irq(&mddev->write_lock);
2008 wake_up(&mddev->sb_wait);
2009 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2010 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2011
2012 }
2013
2014 /* words written to sysfs files may, or may not, be \n terminated.
2015 * We want to accept with case. For this we use cmd_match.
2016 */
2017 static int cmd_match(const char *cmd, const char *str)
2018 {
2019 /* See if cmd, written into a sysfs file, matches
2020 * str. They must either be the same, or cmd can
2021 * have a trailing newline
2022 */
2023 while (*cmd && *str && *cmd == *str) {
2024 cmd++;
2025 str++;
2026 }
2027 if (*cmd == '\n')
2028 cmd++;
2029 if (*str || *cmd)
2030 return 0;
2031 return 1;
2032 }
2033
2034 struct rdev_sysfs_entry {
2035 struct attribute attr;
2036 ssize_t (*show)(mdk_rdev_t *, char *);
2037 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2038 };
2039
2040 static ssize_t
2041 state_show(mdk_rdev_t *rdev, char *page)
2042 {
2043 char *sep = "";
2044 size_t len = 0;
2045
2046 if (test_bit(Faulty, &rdev->flags)) {
2047 len+= sprintf(page+len, "%sfaulty",sep);
2048 sep = ",";
2049 }
2050 if (test_bit(In_sync, &rdev->flags)) {
2051 len += sprintf(page+len, "%sin_sync",sep);
2052 sep = ",";
2053 }
2054 if (test_bit(WriteMostly, &rdev->flags)) {
2055 len += sprintf(page+len, "%swrite_mostly",sep);
2056 sep = ",";
2057 }
2058 if (test_bit(Blocked, &rdev->flags)) {
2059 len += sprintf(page+len, "%sblocked", sep);
2060 sep = ",";
2061 }
2062 if (!test_bit(Faulty, &rdev->flags) &&
2063 !test_bit(In_sync, &rdev->flags)) {
2064 len += sprintf(page+len, "%sspare", sep);
2065 sep = ",";
2066 }
2067 return len+sprintf(page+len, "\n");
2068 }
2069
2070 static ssize_t
2071 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2072 {
2073 /* can write
2074 * faulty - simulates and error
2075 * remove - disconnects the device
2076 * writemostly - sets write_mostly
2077 * -writemostly - clears write_mostly
2078 * blocked - sets the Blocked flag
2079 * -blocked - clears the Blocked flag
2080 * insync - sets Insync providing device isn't active
2081 */
2082 int err = -EINVAL;
2083 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2084 md_error(rdev->mddev, rdev);
2085 err = 0;
2086 } else if (cmd_match(buf, "remove")) {
2087 if (rdev->raid_disk >= 0)
2088 err = -EBUSY;
2089 else {
2090 mddev_t *mddev = rdev->mddev;
2091 kick_rdev_from_array(rdev);
2092 if (mddev->pers)
2093 md_update_sb(mddev, 1);
2094 md_new_event(mddev);
2095 err = 0;
2096 }
2097 } else if (cmd_match(buf, "writemostly")) {
2098 set_bit(WriteMostly, &rdev->flags);
2099 err = 0;
2100 } else if (cmd_match(buf, "-writemostly")) {
2101 clear_bit(WriteMostly, &rdev->flags);
2102 err = 0;
2103 } else if (cmd_match(buf, "blocked")) {
2104 set_bit(Blocked, &rdev->flags);
2105 err = 0;
2106 } else if (cmd_match(buf, "-blocked")) {
2107 clear_bit(Blocked, &rdev->flags);
2108 wake_up(&rdev->blocked_wait);
2109 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2110 md_wakeup_thread(rdev->mddev->thread);
2111
2112 err = 0;
2113 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2114 set_bit(In_sync, &rdev->flags);
2115 err = 0;
2116 }
2117 if (!err && rdev->sysfs_state)
2118 sysfs_notify_dirent(rdev->sysfs_state);
2119 return err ? err : len;
2120 }
2121 static struct rdev_sysfs_entry rdev_state =
2122 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2123
2124 static ssize_t
2125 errors_show(mdk_rdev_t *rdev, char *page)
2126 {
2127 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2128 }
2129
2130 static ssize_t
2131 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2132 {
2133 char *e;
2134 unsigned long n = simple_strtoul(buf, &e, 10);
2135 if (*buf && (*e == 0 || *e == '\n')) {
2136 atomic_set(&rdev->corrected_errors, n);
2137 return len;
2138 }
2139 return -EINVAL;
2140 }
2141 static struct rdev_sysfs_entry rdev_errors =
2142 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2143
2144 static ssize_t
2145 slot_show(mdk_rdev_t *rdev, char *page)
2146 {
2147 if (rdev->raid_disk < 0)
2148 return sprintf(page, "none\n");
2149 else
2150 return sprintf(page, "%d\n", rdev->raid_disk);
2151 }
2152
2153 static ssize_t
2154 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2155 {
2156 char *e;
2157 int err;
2158 char nm[20];
2159 int slot = simple_strtoul(buf, &e, 10);
2160 if (strncmp(buf, "none", 4)==0)
2161 slot = -1;
2162 else if (e==buf || (*e && *e!= '\n'))
2163 return -EINVAL;
2164 if (rdev->mddev->pers && slot == -1) {
2165 /* Setting 'slot' on an active array requires also
2166 * updating the 'rd%d' link, and communicating
2167 * with the personality with ->hot_*_disk.
2168 * For now we only support removing
2169 * failed/spare devices. This normally happens automatically,
2170 * but not when the metadata is externally managed.
2171 */
2172 if (rdev->raid_disk == -1)
2173 return -EEXIST;
2174 /* personality does all needed checks */
2175 if (rdev->mddev->pers->hot_add_disk == NULL)
2176 return -EINVAL;
2177 err = rdev->mddev->pers->
2178 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2179 if (err)
2180 return err;
2181 sprintf(nm, "rd%d", rdev->raid_disk);
2182 sysfs_remove_link(&rdev->mddev->kobj, nm);
2183 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2184 md_wakeup_thread(rdev->mddev->thread);
2185 } else if (rdev->mddev->pers) {
2186 mdk_rdev_t *rdev2;
2187 /* Activating a spare .. or possibly reactivating
2188 * if we ever get bitmaps working here.
2189 */
2190
2191 if (rdev->raid_disk != -1)
2192 return -EBUSY;
2193
2194 if (rdev->mddev->pers->hot_add_disk == NULL)
2195 return -EINVAL;
2196
2197 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2198 if (rdev2->raid_disk == slot)
2199 return -EEXIST;
2200
2201 rdev->raid_disk = slot;
2202 if (test_bit(In_sync, &rdev->flags))
2203 rdev->saved_raid_disk = slot;
2204 else
2205 rdev->saved_raid_disk = -1;
2206 err = rdev->mddev->pers->
2207 hot_add_disk(rdev->mddev, rdev);
2208 if (err) {
2209 rdev->raid_disk = -1;
2210 return err;
2211 } else
2212 sysfs_notify_dirent(rdev->sysfs_state);
2213 sprintf(nm, "rd%d", rdev->raid_disk);
2214 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2215 printk(KERN_WARNING
2216 "md: cannot register "
2217 "%s for %s\n",
2218 nm, mdname(rdev->mddev));
2219
2220 /* don't wakeup anyone, leave that to userspace. */
2221 } else {
2222 if (slot >= rdev->mddev->raid_disks)
2223 return -ENOSPC;
2224 rdev->raid_disk = slot;
2225 /* assume it is working */
2226 clear_bit(Faulty, &rdev->flags);
2227 clear_bit(WriteMostly, &rdev->flags);
2228 set_bit(In_sync, &rdev->flags);
2229 sysfs_notify_dirent(rdev->sysfs_state);
2230 }
2231 return len;
2232 }
2233
2234
2235 static struct rdev_sysfs_entry rdev_slot =
2236 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2237
2238 static ssize_t
2239 offset_show(mdk_rdev_t *rdev, char *page)
2240 {
2241 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2242 }
2243
2244 static ssize_t
2245 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2246 {
2247 char *e;
2248 unsigned long long offset = simple_strtoull(buf, &e, 10);
2249 if (e==buf || (*e && *e != '\n'))
2250 return -EINVAL;
2251 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2252 return -EBUSY;
2253 if (rdev->sectors && rdev->mddev->external)
2254 /* Must set offset before size, so overlap checks
2255 * can be sane */
2256 return -EBUSY;
2257 rdev->data_offset = offset;
2258 return len;
2259 }
2260
2261 static struct rdev_sysfs_entry rdev_offset =
2262 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2263
2264 static ssize_t
2265 rdev_size_show(mdk_rdev_t *rdev, char *page)
2266 {
2267 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2268 }
2269
2270 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2271 {
2272 /* check if two start/length pairs overlap */
2273 if (s1+l1 <= s2)
2274 return 0;
2275 if (s2+l2 <= s1)
2276 return 0;
2277 return 1;
2278 }
2279
2280 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2281 {
2282 unsigned long long blocks;
2283 sector_t new;
2284
2285 if (strict_strtoull(buf, 10, &blocks) < 0)
2286 return -EINVAL;
2287
2288 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2289 return -EINVAL; /* sector conversion overflow */
2290
2291 new = blocks * 2;
2292 if (new != blocks * 2)
2293 return -EINVAL; /* unsigned long long to sector_t overflow */
2294
2295 *sectors = new;
2296 return 0;
2297 }
2298
2299 static ssize_t
2300 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2301 {
2302 mddev_t *my_mddev = rdev->mddev;
2303 sector_t oldsectors = rdev->sectors;
2304 sector_t sectors;
2305
2306 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2307 return -EINVAL;
2308 if (my_mddev->pers && rdev->raid_disk >= 0) {
2309 if (my_mddev->persistent) {
2310 sectors = super_types[my_mddev->major_version].
2311 rdev_size_change(rdev, sectors);
2312 if (!sectors)
2313 return -EBUSY;
2314 } else if (!sectors)
2315 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2316 rdev->data_offset;
2317 }
2318 if (sectors < my_mddev->dev_sectors)
2319 return -EINVAL; /* component must fit device */
2320
2321 rdev->sectors = sectors;
2322 if (sectors > oldsectors && my_mddev->external) {
2323 /* need to check that all other rdevs with the same ->bdev
2324 * do not overlap. We need to unlock the mddev to avoid
2325 * a deadlock. We have already changed rdev->sectors, and if
2326 * we have to change it back, we will have the lock again.
2327 */
2328 mddev_t *mddev;
2329 int overlap = 0;
2330 struct list_head *tmp;
2331
2332 mddev_unlock(my_mddev);
2333 for_each_mddev(mddev, tmp) {
2334 mdk_rdev_t *rdev2;
2335
2336 mddev_lock(mddev);
2337 list_for_each_entry(rdev2, &mddev->disks, same_set)
2338 if (test_bit(AllReserved, &rdev2->flags) ||
2339 (rdev->bdev == rdev2->bdev &&
2340 rdev != rdev2 &&
2341 overlaps(rdev->data_offset, rdev->sectors,
2342 rdev2->data_offset,
2343 rdev2->sectors))) {
2344 overlap = 1;
2345 break;
2346 }
2347 mddev_unlock(mddev);
2348 if (overlap) {
2349 mddev_put(mddev);
2350 break;
2351 }
2352 }
2353 mddev_lock(my_mddev);
2354 if (overlap) {
2355 /* Someone else could have slipped in a size
2356 * change here, but doing so is just silly.
2357 * We put oldsectors back because we *know* it is
2358 * safe, and trust userspace not to race with
2359 * itself
2360 */
2361 rdev->sectors = oldsectors;
2362 return -EBUSY;
2363 }
2364 }
2365 return len;
2366 }
2367
2368 static struct rdev_sysfs_entry rdev_size =
2369 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2370
2371 static struct attribute *rdev_default_attrs[] = {
2372 &rdev_state.attr,
2373 &rdev_errors.attr,
2374 &rdev_slot.attr,
2375 &rdev_offset.attr,
2376 &rdev_size.attr,
2377 NULL,
2378 };
2379 static ssize_t
2380 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2381 {
2382 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2383 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2384 mddev_t *mddev = rdev->mddev;
2385 ssize_t rv;
2386
2387 if (!entry->show)
2388 return -EIO;
2389
2390 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2391 if (!rv) {
2392 if (rdev->mddev == NULL)
2393 rv = -EBUSY;
2394 else
2395 rv = entry->show(rdev, page);
2396 mddev_unlock(mddev);
2397 }
2398 return rv;
2399 }
2400
2401 static ssize_t
2402 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2403 const char *page, size_t length)
2404 {
2405 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2406 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2407 ssize_t rv;
2408 mddev_t *mddev = rdev->mddev;
2409
2410 if (!entry->store)
2411 return -EIO;
2412 if (!capable(CAP_SYS_ADMIN))
2413 return -EACCES;
2414 rv = mddev ? mddev_lock(mddev): -EBUSY;
2415 if (!rv) {
2416 if (rdev->mddev == NULL)
2417 rv = -EBUSY;
2418 else
2419 rv = entry->store(rdev, page, length);
2420 mddev_unlock(mddev);
2421 }
2422 return rv;
2423 }
2424
2425 static void rdev_free(struct kobject *ko)
2426 {
2427 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2428 kfree(rdev);
2429 }
2430 static struct sysfs_ops rdev_sysfs_ops = {
2431 .show = rdev_attr_show,
2432 .store = rdev_attr_store,
2433 };
2434 static struct kobj_type rdev_ktype = {
2435 .release = rdev_free,
2436 .sysfs_ops = &rdev_sysfs_ops,
2437 .default_attrs = rdev_default_attrs,
2438 };
2439
2440 /*
2441 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2442 *
2443 * mark the device faulty if:
2444 *
2445 * - the device is nonexistent (zero size)
2446 * - the device has no valid superblock
2447 *
2448 * a faulty rdev _never_ has rdev->sb set.
2449 */
2450 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2451 {
2452 char b[BDEVNAME_SIZE];
2453 int err;
2454 mdk_rdev_t *rdev;
2455 sector_t size;
2456
2457 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2458 if (!rdev) {
2459 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2460 return ERR_PTR(-ENOMEM);
2461 }
2462
2463 if ((err = alloc_disk_sb(rdev)))
2464 goto abort_free;
2465
2466 err = lock_rdev(rdev, newdev, super_format == -2);
2467 if (err)
2468 goto abort_free;
2469
2470 kobject_init(&rdev->kobj, &rdev_ktype);
2471
2472 rdev->desc_nr = -1;
2473 rdev->saved_raid_disk = -1;
2474 rdev->raid_disk = -1;
2475 rdev->flags = 0;
2476 rdev->data_offset = 0;
2477 rdev->sb_events = 0;
2478 atomic_set(&rdev->nr_pending, 0);
2479 atomic_set(&rdev->read_errors, 0);
2480 atomic_set(&rdev->corrected_errors, 0);
2481
2482 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2483 if (!size) {
2484 printk(KERN_WARNING
2485 "md: %s has zero or unknown size, marking faulty!\n",
2486 bdevname(rdev->bdev,b));
2487 err = -EINVAL;
2488 goto abort_free;
2489 }
2490
2491 if (super_format >= 0) {
2492 err = super_types[super_format].
2493 load_super(rdev, NULL, super_minor);
2494 if (err == -EINVAL) {
2495 printk(KERN_WARNING
2496 "md: %s does not have a valid v%d.%d "
2497 "superblock, not importing!\n",
2498 bdevname(rdev->bdev,b),
2499 super_format, super_minor);
2500 goto abort_free;
2501 }
2502 if (err < 0) {
2503 printk(KERN_WARNING
2504 "md: could not read %s's sb, not importing!\n",
2505 bdevname(rdev->bdev,b));
2506 goto abort_free;
2507 }
2508 }
2509
2510 INIT_LIST_HEAD(&rdev->same_set);
2511 init_waitqueue_head(&rdev->blocked_wait);
2512
2513 return rdev;
2514
2515 abort_free:
2516 if (rdev->sb_page) {
2517 if (rdev->bdev)
2518 unlock_rdev(rdev);
2519 free_disk_sb(rdev);
2520 }
2521 kfree(rdev);
2522 return ERR_PTR(err);
2523 }
2524
2525 /*
2526 * Check a full RAID array for plausibility
2527 */
2528
2529
2530 static void analyze_sbs(mddev_t * mddev)
2531 {
2532 int i;
2533 mdk_rdev_t *rdev, *freshest, *tmp;
2534 char b[BDEVNAME_SIZE];
2535
2536 freshest = NULL;
2537 rdev_for_each(rdev, tmp, mddev)
2538 switch (super_types[mddev->major_version].
2539 load_super(rdev, freshest, mddev->minor_version)) {
2540 case 1:
2541 freshest = rdev;
2542 break;
2543 case 0:
2544 break;
2545 default:
2546 printk( KERN_ERR \
2547 "md: fatal superblock inconsistency in %s"
2548 " -- removing from array\n",
2549 bdevname(rdev->bdev,b));
2550 kick_rdev_from_array(rdev);
2551 }
2552
2553
2554 super_types[mddev->major_version].
2555 validate_super(mddev, freshest);
2556
2557 i = 0;
2558 rdev_for_each(rdev, tmp, mddev) {
2559 if (rdev->desc_nr >= mddev->max_disks ||
2560 i > mddev->max_disks) {
2561 printk(KERN_WARNING
2562 "md: %s: %s: only %d devices permitted\n",
2563 mdname(mddev), bdevname(rdev->bdev, b),
2564 mddev->max_disks);
2565 kick_rdev_from_array(rdev);
2566 continue;
2567 }
2568 if (rdev != freshest)
2569 if (super_types[mddev->major_version].
2570 validate_super(mddev, rdev)) {
2571 printk(KERN_WARNING "md: kicking non-fresh %s"
2572 " from array!\n",
2573 bdevname(rdev->bdev,b));
2574 kick_rdev_from_array(rdev);
2575 continue;
2576 }
2577 if (mddev->level == LEVEL_MULTIPATH) {
2578 rdev->desc_nr = i++;
2579 rdev->raid_disk = rdev->desc_nr;
2580 set_bit(In_sync, &rdev->flags);
2581 } else if (rdev->raid_disk >= mddev->raid_disks) {
2582 rdev->raid_disk = -1;
2583 clear_bit(In_sync, &rdev->flags);
2584 }
2585 }
2586 }
2587
2588 static void md_safemode_timeout(unsigned long data);
2589
2590 static ssize_t
2591 safe_delay_show(mddev_t *mddev, char *page)
2592 {
2593 int msec = (mddev->safemode_delay*1000)/HZ;
2594 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2595 }
2596 static ssize_t
2597 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2598 {
2599 int scale=1;
2600 int dot=0;
2601 int i;
2602 unsigned long msec;
2603 char buf[30];
2604
2605 /* remove a period, and count digits after it */
2606 if (len >= sizeof(buf))
2607 return -EINVAL;
2608 strlcpy(buf, cbuf, sizeof(buf));
2609 for (i=0; i<len; i++) {
2610 if (dot) {
2611 if (isdigit(buf[i])) {
2612 buf[i-1] = buf[i];
2613 scale *= 10;
2614 }
2615 buf[i] = 0;
2616 } else if (buf[i] == '.') {
2617 dot=1;
2618 buf[i] = 0;
2619 }
2620 }
2621 if (strict_strtoul(buf, 10, &msec) < 0)
2622 return -EINVAL;
2623 msec = (msec * 1000) / scale;
2624 if (msec == 0)
2625 mddev->safemode_delay = 0;
2626 else {
2627 unsigned long old_delay = mddev->safemode_delay;
2628 mddev->safemode_delay = (msec*HZ)/1000;
2629 if (mddev->safemode_delay == 0)
2630 mddev->safemode_delay = 1;
2631 if (mddev->safemode_delay < old_delay)
2632 md_safemode_timeout((unsigned long)mddev);
2633 }
2634 return len;
2635 }
2636 static struct md_sysfs_entry md_safe_delay =
2637 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2638
2639 static ssize_t
2640 level_show(mddev_t *mddev, char *page)
2641 {
2642 struct mdk_personality *p = mddev->pers;
2643 if (p)
2644 return sprintf(page, "%s\n", p->name);
2645 else if (mddev->clevel[0])
2646 return sprintf(page, "%s\n", mddev->clevel);
2647 else if (mddev->level != LEVEL_NONE)
2648 return sprintf(page, "%d\n", mddev->level);
2649 else
2650 return 0;
2651 }
2652
2653 static ssize_t
2654 level_store(mddev_t *mddev, const char *buf, size_t len)
2655 {
2656 char level[16];
2657 ssize_t rv = len;
2658 struct mdk_personality *pers;
2659 void *priv;
2660
2661 if (mddev->pers == NULL) {
2662 if (len == 0)
2663 return 0;
2664 if (len >= sizeof(mddev->clevel))
2665 return -ENOSPC;
2666 strncpy(mddev->clevel, buf, len);
2667 if (mddev->clevel[len-1] == '\n')
2668 len--;
2669 mddev->clevel[len] = 0;
2670 mddev->level = LEVEL_NONE;
2671 return rv;
2672 }
2673
2674 /* request to change the personality. Need to ensure:
2675 * - array is not engaged in resync/recovery/reshape
2676 * - old personality can be suspended
2677 * - new personality will access other array.
2678 */
2679
2680 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2681 return -EBUSY;
2682
2683 if (!mddev->pers->quiesce) {
2684 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2685 mdname(mddev), mddev->pers->name);
2686 return -EINVAL;
2687 }
2688
2689 /* Now find the new personality */
2690 if (len == 0 || len >= sizeof(level))
2691 return -EINVAL;
2692 strncpy(level, buf, len);
2693 if (level[len-1] == '\n')
2694 len--;
2695 level[len] = 0;
2696
2697 request_module("md-%s", level);
2698 spin_lock(&pers_lock);
2699 pers = find_pers(LEVEL_NONE, level);
2700 if (!pers || !try_module_get(pers->owner)) {
2701 spin_unlock(&pers_lock);
2702 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2703 return -EINVAL;
2704 }
2705 spin_unlock(&pers_lock);
2706
2707 if (pers == mddev->pers) {
2708 /* Nothing to do! */
2709 module_put(pers->owner);
2710 return rv;
2711 }
2712 if (!pers->takeover) {
2713 module_put(pers->owner);
2714 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2715 mdname(mddev), level);
2716 return -EINVAL;
2717 }
2718
2719 /* ->takeover must set new_* and/or delta_disks
2720 * if it succeeds, and may set them when it fails.
2721 */
2722 priv = pers->takeover(mddev);
2723 if (IS_ERR(priv)) {
2724 mddev->new_level = mddev->level;
2725 mddev->new_layout = mddev->layout;
2726 mddev->new_chunk_sectors = mddev->chunk_sectors;
2727 mddev->raid_disks -= mddev->delta_disks;
2728 mddev->delta_disks = 0;
2729 module_put(pers->owner);
2730 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2731 mdname(mddev), level);
2732 return PTR_ERR(priv);
2733 }
2734
2735 /* Looks like we have a winner */
2736 mddev_suspend(mddev);
2737 mddev->pers->stop(mddev);
2738 module_put(mddev->pers->owner);
2739 mddev->pers = pers;
2740 mddev->private = priv;
2741 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2742 mddev->level = mddev->new_level;
2743 mddev->layout = mddev->new_layout;
2744 mddev->chunk_sectors = mddev->new_chunk_sectors;
2745 mddev->delta_disks = 0;
2746 pers->run(mddev);
2747 mddev_resume(mddev);
2748 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2749 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2750 md_wakeup_thread(mddev->thread);
2751 return rv;
2752 }
2753
2754 static struct md_sysfs_entry md_level =
2755 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2756
2757
2758 static ssize_t
2759 layout_show(mddev_t *mddev, char *page)
2760 {
2761 /* just a number, not meaningful for all levels */
2762 if (mddev->reshape_position != MaxSector &&
2763 mddev->layout != mddev->new_layout)
2764 return sprintf(page, "%d (%d)\n",
2765 mddev->new_layout, mddev->layout);
2766 return sprintf(page, "%d\n", mddev->layout);
2767 }
2768
2769 static ssize_t
2770 layout_store(mddev_t *mddev, const char *buf, size_t len)
2771 {
2772 char *e;
2773 unsigned long n = simple_strtoul(buf, &e, 10);
2774
2775 if (!*buf || (*e && *e != '\n'))
2776 return -EINVAL;
2777
2778 if (mddev->pers) {
2779 int err;
2780 if (mddev->pers->check_reshape == NULL)
2781 return -EBUSY;
2782 mddev->new_layout = n;
2783 err = mddev->pers->check_reshape(mddev);
2784 if (err) {
2785 mddev->new_layout = mddev->layout;
2786 return err;
2787 }
2788 } else {
2789 mddev->new_layout = n;
2790 if (mddev->reshape_position == MaxSector)
2791 mddev->layout = n;
2792 }
2793 return len;
2794 }
2795 static struct md_sysfs_entry md_layout =
2796 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2797
2798
2799 static ssize_t
2800 raid_disks_show(mddev_t *mddev, char *page)
2801 {
2802 if (mddev->raid_disks == 0)
2803 return 0;
2804 if (mddev->reshape_position != MaxSector &&
2805 mddev->delta_disks != 0)
2806 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2807 mddev->raid_disks - mddev->delta_disks);
2808 return sprintf(page, "%d\n", mddev->raid_disks);
2809 }
2810
2811 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2812
2813 static ssize_t
2814 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2815 {
2816 char *e;
2817 int rv = 0;
2818 unsigned long n = simple_strtoul(buf, &e, 10);
2819
2820 if (!*buf || (*e && *e != '\n'))
2821 return -EINVAL;
2822
2823 if (mddev->pers)
2824 rv = update_raid_disks(mddev, n);
2825 else if (mddev->reshape_position != MaxSector) {
2826 int olddisks = mddev->raid_disks - mddev->delta_disks;
2827 mddev->delta_disks = n - olddisks;
2828 mddev->raid_disks = n;
2829 } else
2830 mddev->raid_disks = n;
2831 return rv ? rv : len;
2832 }
2833 static struct md_sysfs_entry md_raid_disks =
2834 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2835
2836 static ssize_t
2837 chunk_size_show(mddev_t *mddev, char *page)
2838 {
2839 if (mddev->reshape_position != MaxSector &&
2840 mddev->chunk_sectors != mddev->new_chunk_sectors)
2841 return sprintf(page, "%d (%d)\n",
2842 mddev->new_chunk_sectors << 9,
2843 mddev->chunk_sectors << 9);
2844 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2845 }
2846
2847 static ssize_t
2848 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2849 {
2850 char *e;
2851 unsigned long n = simple_strtoul(buf, &e, 10);
2852
2853 if (!*buf || (*e && *e != '\n'))
2854 return -EINVAL;
2855
2856 if (mddev->pers) {
2857 int err;
2858 if (mddev->pers->check_reshape == NULL)
2859 return -EBUSY;
2860 mddev->new_chunk_sectors = n >> 9;
2861 err = mddev->pers->check_reshape(mddev);
2862 if (err) {
2863 mddev->new_chunk_sectors = mddev->chunk_sectors;
2864 return err;
2865 }
2866 } else {
2867 mddev->new_chunk_sectors = n >> 9;
2868 if (mddev->reshape_position == MaxSector)
2869 mddev->chunk_sectors = n >> 9;
2870 }
2871 return len;
2872 }
2873 static struct md_sysfs_entry md_chunk_size =
2874 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2875
2876 static ssize_t
2877 resync_start_show(mddev_t *mddev, char *page)
2878 {
2879 if (mddev->recovery_cp == MaxSector)
2880 return sprintf(page, "none\n");
2881 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2882 }
2883
2884 static ssize_t
2885 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2886 {
2887 char *e;
2888 unsigned long long n = simple_strtoull(buf, &e, 10);
2889
2890 if (mddev->pers)
2891 return -EBUSY;
2892 if (!*buf || (*e && *e != '\n'))
2893 return -EINVAL;
2894
2895 mddev->recovery_cp = n;
2896 return len;
2897 }
2898 static struct md_sysfs_entry md_resync_start =
2899 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2900
2901 /*
2902 * The array state can be:
2903 *
2904 * clear
2905 * No devices, no size, no level
2906 * Equivalent to STOP_ARRAY ioctl
2907 * inactive
2908 * May have some settings, but array is not active
2909 * all IO results in error
2910 * When written, doesn't tear down array, but just stops it
2911 * suspended (not supported yet)
2912 * All IO requests will block. The array can be reconfigured.
2913 * Writing this, if accepted, will block until array is quiescent
2914 * readonly
2915 * no resync can happen. no superblocks get written.
2916 * write requests fail
2917 * read-auto
2918 * like readonly, but behaves like 'clean' on a write request.
2919 *
2920 * clean - no pending writes, but otherwise active.
2921 * When written to inactive array, starts without resync
2922 * If a write request arrives then
2923 * if metadata is known, mark 'dirty' and switch to 'active'.
2924 * if not known, block and switch to write-pending
2925 * If written to an active array that has pending writes, then fails.
2926 * active
2927 * fully active: IO and resync can be happening.
2928 * When written to inactive array, starts with resync
2929 *
2930 * write-pending
2931 * clean, but writes are blocked waiting for 'active' to be written.
2932 *
2933 * active-idle
2934 * like active, but no writes have been seen for a while (100msec).
2935 *
2936 */
2937 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2938 write_pending, active_idle, bad_word};
2939 static char *array_states[] = {
2940 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2941 "write-pending", "active-idle", NULL };
2942
2943 static int match_word(const char *word, char **list)
2944 {
2945 int n;
2946 for (n=0; list[n]; n++)
2947 if (cmd_match(word, list[n]))
2948 break;
2949 return n;
2950 }
2951
2952 static ssize_t
2953 array_state_show(mddev_t *mddev, char *page)
2954 {
2955 enum array_state st = inactive;
2956
2957 if (mddev->pers)
2958 switch(mddev->ro) {
2959 case 1:
2960 st = readonly;
2961 break;
2962 case 2:
2963 st = read_auto;
2964 break;
2965 case 0:
2966 if (mddev->in_sync)
2967 st = clean;
2968 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2969 st = write_pending;
2970 else if (mddev->safemode)
2971 st = active_idle;
2972 else
2973 st = active;
2974 }
2975 else {
2976 if (list_empty(&mddev->disks) &&
2977 mddev->raid_disks == 0 &&
2978 mddev->dev_sectors == 0)
2979 st = clear;
2980 else
2981 st = inactive;
2982 }
2983 return sprintf(page, "%s\n", array_states[st]);
2984 }
2985
2986 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2987 static int do_md_run(mddev_t * mddev);
2988 static int restart_array(mddev_t *mddev);
2989
2990 static ssize_t
2991 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2992 {
2993 int err = -EINVAL;
2994 enum array_state st = match_word(buf, array_states);
2995 switch(st) {
2996 case bad_word:
2997 break;
2998 case clear:
2999 /* stopping an active array */
3000 if (atomic_read(&mddev->openers) > 0)
3001 return -EBUSY;
3002 err = do_md_stop(mddev, 0, 0);
3003 break;
3004 case inactive:
3005 /* stopping an active array */
3006 if (mddev->pers) {
3007 if (atomic_read(&mddev->openers) > 0)
3008 return -EBUSY;
3009 err = do_md_stop(mddev, 2, 0);
3010 } else
3011 err = 0; /* already inactive */
3012 break;
3013 case suspended:
3014 break; /* not supported yet */
3015 case readonly:
3016 if (mddev->pers)
3017 err = do_md_stop(mddev, 1, 0);
3018 else {
3019 mddev->ro = 1;
3020 set_disk_ro(mddev->gendisk, 1);
3021 err = do_md_run(mddev);
3022 }
3023 break;
3024 case read_auto:
3025 if (mddev->pers) {
3026 if (mddev->ro == 0)
3027 err = do_md_stop(mddev, 1, 0);
3028 else if (mddev->ro == 1)
3029 err = restart_array(mddev);
3030 if (err == 0) {
3031 mddev->ro = 2;
3032 set_disk_ro(mddev->gendisk, 0);
3033 }
3034 } else {
3035 mddev->ro = 2;
3036 err = do_md_run(mddev);
3037 }
3038 break;
3039 case clean:
3040 if (mddev->pers) {
3041 restart_array(mddev);
3042 spin_lock_irq(&mddev->write_lock);
3043 if (atomic_read(&mddev->writes_pending) == 0) {
3044 if (mddev->in_sync == 0) {
3045 mddev->in_sync = 1;
3046 if (mddev->safemode == 1)
3047 mddev->safemode = 0;
3048 if (mddev->persistent)
3049 set_bit(MD_CHANGE_CLEAN,
3050 &mddev->flags);
3051 }
3052 err = 0;
3053 } else
3054 err = -EBUSY;
3055 spin_unlock_irq(&mddev->write_lock);
3056 } else
3057 err = -EINVAL;
3058 break;
3059 case active:
3060 if (mddev->pers) {
3061 restart_array(mddev);
3062 if (mddev->external)
3063 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3064 wake_up(&mddev->sb_wait);
3065 err = 0;
3066 } else {
3067 mddev->ro = 0;
3068 set_disk_ro(mddev->gendisk, 0);
3069 err = do_md_run(mddev);
3070 }
3071 break;
3072 case write_pending:
3073 case active_idle:
3074 /* these cannot be set */
3075 break;
3076 }
3077 if (err)
3078 return err;
3079 else {
3080 sysfs_notify_dirent(mddev->sysfs_state);
3081 return len;
3082 }
3083 }
3084 static struct md_sysfs_entry md_array_state =
3085 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3086
3087 static ssize_t
3088 null_show(mddev_t *mddev, char *page)
3089 {
3090 return -EINVAL;
3091 }
3092
3093 static ssize_t
3094 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3095 {
3096 /* buf must be %d:%d\n? giving major and minor numbers */
3097 /* The new device is added to the array.
3098 * If the array has a persistent superblock, we read the
3099 * superblock to initialise info and check validity.
3100 * Otherwise, only checking done is that in bind_rdev_to_array,
3101 * which mainly checks size.
3102 */
3103 char *e;
3104 int major = simple_strtoul(buf, &e, 10);
3105 int minor;
3106 dev_t dev;
3107 mdk_rdev_t *rdev;
3108 int err;
3109
3110 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3111 return -EINVAL;
3112 minor = simple_strtoul(e+1, &e, 10);
3113 if (*e && *e != '\n')
3114 return -EINVAL;
3115 dev = MKDEV(major, minor);
3116 if (major != MAJOR(dev) ||
3117 minor != MINOR(dev))
3118 return -EOVERFLOW;
3119
3120
3121 if (mddev->persistent) {
3122 rdev = md_import_device(dev, mddev->major_version,
3123 mddev->minor_version);
3124 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3125 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3126 mdk_rdev_t, same_set);
3127 err = super_types[mddev->major_version]
3128 .load_super(rdev, rdev0, mddev->minor_version);
3129 if (err < 0)
3130 goto out;
3131 }
3132 } else if (mddev->external)
3133 rdev = md_import_device(dev, -2, -1);
3134 else
3135 rdev = md_import_device(dev, -1, -1);
3136
3137 if (IS_ERR(rdev))
3138 return PTR_ERR(rdev);
3139 err = bind_rdev_to_array(rdev, mddev);
3140 out:
3141 if (err)
3142 export_rdev(rdev);
3143 return err ? err : len;
3144 }
3145
3146 static struct md_sysfs_entry md_new_device =
3147 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3148
3149 static ssize_t
3150 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3151 {
3152 char *end;
3153 unsigned long chunk, end_chunk;
3154
3155 if (!mddev->bitmap)
3156 goto out;
3157 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3158 while (*buf) {
3159 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3160 if (buf == end) break;
3161 if (*end == '-') { /* range */
3162 buf = end + 1;
3163 end_chunk = simple_strtoul(buf, &end, 0);
3164 if (buf == end) break;
3165 }
3166 if (*end && !isspace(*end)) break;
3167 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3168 buf = end;
3169 while (isspace(*buf)) buf++;
3170 }
3171 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3172 out:
3173 return len;
3174 }
3175
3176 static struct md_sysfs_entry md_bitmap =
3177 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3178
3179 static ssize_t
3180 size_show(mddev_t *mddev, char *page)
3181 {
3182 return sprintf(page, "%llu\n",
3183 (unsigned long long)mddev->dev_sectors / 2);
3184 }
3185
3186 static int update_size(mddev_t *mddev, sector_t num_sectors);
3187
3188 static ssize_t
3189 size_store(mddev_t *mddev, const char *buf, size_t len)
3190 {
3191 /* If array is inactive, we can reduce the component size, but
3192 * not increase it (except from 0).
3193 * If array is active, we can try an on-line resize
3194 */
3195 sector_t sectors;
3196 int err = strict_blocks_to_sectors(buf, &sectors);
3197
3198 if (err < 0)
3199 return err;
3200 if (mddev->pers) {
3201 err = update_size(mddev, sectors);
3202 md_update_sb(mddev, 1);
3203 } else {
3204 if (mddev->dev_sectors == 0 ||
3205 mddev->dev_sectors > sectors)
3206 mddev->dev_sectors = sectors;
3207 else
3208 err = -ENOSPC;
3209 }
3210 return err ? err : len;
3211 }
3212
3213 static struct md_sysfs_entry md_size =
3214 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3215
3216
3217 /* Metdata version.
3218 * This is one of
3219 * 'none' for arrays with no metadata (good luck...)
3220 * 'external' for arrays with externally managed metadata,
3221 * or N.M for internally known formats
3222 */
3223 static ssize_t
3224 metadata_show(mddev_t *mddev, char *page)
3225 {
3226 if (mddev->persistent)
3227 return sprintf(page, "%d.%d\n",
3228 mddev->major_version, mddev->minor_version);
3229 else if (mddev->external)
3230 return sprintf(page, "external:%s\n", mddev->metadata_type);
3231 else
3232 return sprintf(page, "none\n");
3233 }
3234
3235 static ssize_t
3236 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3237 {
3238 int major, minor;
3239 char *e;
3240 /* Changing the details of 'external' metadata is
3241 * always permitted. Otherwise there must be
3242 * no devices attached to the array.
3243 */
3244 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3245 ;
3246 else if (!list_empty(&mddev->disks))
3247 return -EBUSY;
3248
3249 if (cmd_match(buf, "none")) {
3250 mddev->persistent = 0;
3251 mddev->external = 0;
3252 mddev->major_version = 0;
3253 mddev->minor_version = 90;
3254 return len;
3255 }
3256 if (strncmp(buf, "external:", 9) == 0) {
3257 size_t namelen = len-9;
3258 if (namelen >= sizeof(mddev->metadata_type))
3259 namelen = sizeof(mddev->metadata_type)-1;
3260 strncpy(mddev->metadata_type, buf+9, namelen);
3261 mddev->metadata_type[namelen] = 0;
3262 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3263 mddev->metadata_type[--namelen] = 0;
3264 mddev->persistent = 0;
3265 mddev->external = 1;
3266 mddev->major_version = 0;
3267 mddev->minor_version = 90;
3268 return len;
3269 }
3270 major = simple_strtoul(buf, &e, 10);
3271 if (e==buf || *e != '.')
3272 return -EINVAL;
3273 buf = e+1;
3274 minor = simple_strtoul(buf, &e, 10);
3275 if (e==buf || (*e && *e != '\n') )
3276 return -EINVAL;
3277 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3278 return -ENOENT;
3279 mddev->major_version = major;
3280 mddev->minor_version = minor;
3281 mddev->persistent = 1;
3282 mddev->external = 0;
3283 return len;
3284 }
3285
3286 static struct md_sysfs_entry md_metadata =
3287 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3288
3289 static ssize_t
3290 action_show(mddev_t *mddev, char *page)
3291 {
3292 char *type = "idle";
3293 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3294 type = "frozen";
3295 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3296 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3297 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3298 type = "reshape";
3299 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3300 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3301 type = "resync";
3302 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3303 type = "check";
3304 else
3305 type = "repair";
3306 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3307 type = "recover";
3308 }
3309 return sprintf(page, "%s\n", type);
3310 }
3311
3312 static ssize_t
3313 action_store(mddev_t *mddev, const char *page, size_t len)
3314 {
3315 if (!mddev->pers || !mddev->pers->sync_request)
3316 return -EINVAL;
3317
3318 if (cmd_match(page, "frozen"))
3319 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3320 else
3321 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3322
3323 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3324 if (mddev->sync_thread) {
3325 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3326 md_unregister_thread(mddev->sync_thread);
3327 mddev->sync_thread = NULL;
3328 mddev->recovery = 0;
3329 }
3330 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3331 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3332 return -EBUSY;
3333 else if (cmd_match(page, "resync"))
3334 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3335 else if (cmd_match(page, "recover")) {
3336 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3337 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3338 } else if (cmd_match(page, "reshape")) {
3339 int err;
3340 if (mddev->pers->start_reshape == NULL)
3341 return -EINVAL;
3342 err = mddev->pers->start_reshape(mddev);
3343 if (err)
3344 return err;
3345 sysfs_notify(&mddev->kobj, NULL, "degraded");
3346 } else {
3347 if (cmd_match(page, "check"))
3348 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3349 else if (!cmd_match(page, "repair"))
3350 return -EINVAL;
3351 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3352 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3353 }
3354 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3355 md_wakeup_thread(mddev->thread);
3356 sysfs_notify_dirent(mddev->sysfs_action);
3357 return len;
3358 }
3359
3360 static ssize_t
3361 mismatch_cnt_show(mddev_t *mddev, char *page)
3362 {
3363 return sprintf(page, "%llu\n",
3364 (unsigned long long) mddev->resync_mismatches);
3365 }
3366
3367 static struct md_sysfs_entry md_scan_mode =
3368 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3369
3370
3371 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3372
3373 static ssize_t
3374 sync_min_show(mddev_t *mddev, char *page)
3375 {
3376 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3377 mddev->sync_speed_min ? "local": "system");
3378 }
3379
3380 static ssize_t
3381 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3382 {
3383 int min;
3384 char *e;
3385 if (strncmp(buf, "system", 6)==0) {
3386 mddev->sync_speed_min = 0;
3387 return len;
3388 }
3389 min = simple_strtoul(buf, &e, 10);
3390 if (buf == e || (*e && *e != '\n') || min <= 0)
3391 return -EINVAL;
3392 mddev->sync_speed_min = min;
3393 return len;
3394 }
3395
3396 static struct md_sysfs_entry md_sync_min =
3397 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3398
3399 static ssize_t
3400 sync_max_show(mddev_t *mddev, char *page)
3401 {
3402 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3403 mddev->sync_speed_max ? "local": "system");
3404 }
3405
3406 static ssize_t
3407 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3408 {
3409 int max;
3410 char *e;
3411 if (strncmp(buf, "system", 6)==0) {
3412 mddev->sync_speed_max = 0;
3413 return len;
3414 }
3415 max = simple_strtoul(buf, &e, 10);
3416 if (buf == e || (*e && *e != '\n') || max <= 0)
3417 return -EINVAL;
3418 mddev->sync_speed_max = max;
3419 return len;
3420 }
3421
3422 static struct md_sysfs_entry md_sync_max =
3423 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3424
3425 static ssize_t
3426 degraded_show(mddev_t *mddev, char *page)
3427 {
3428 return sprintf(page, "%d\n", mddev->degraded);
3429 }
3430 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3431
3432 static ssize_t
3433 sync_force_parallel_show(mddev_t *mddev, char *page)
3434 {
3435 return sprintf(page, "%d\n", mddev->parallel_resync);
3436 }
3437
3438 static ssize_t
3439 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3440 {
3441 long n;
3442
3443 if (strict_strtol(buf, 10, &n))
3444 return -EINVAL;
3445
3446 if (n != 0 && n != 1)
3447 return -EINVAL;
3448
3449 mddev->parallel_resync = n;
3450
3451 if (mddev->sync_thread)
3452 wake_up(&resync_wait);
3453
3454 return len;
3455 }
3456
3457 /* force parallel resync, even with shared block devices */
3458 static struct md_sysfs_entry md_sync_force_parallel =
3459 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3460 sync_force_parallel_show, sync_force_parallel_store);
3461
3462 static ssize_t
3463 sync_speed_show(mddev_t *mddev, char *page)
3464 {
3465 unsigned long resync, dt, db;
3466 if (mddev->curr_resync == 0)
3467 return sprintf(page, "none\n");
3468 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3469 dt = (jiffies - mddev->resync_mark) / HZ;
3470 if (!dt) dt++;
3471 db = resync - mddev->resync_mark_cnt;
3472 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3473 }
3474
3475 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3476
3477 static ssize_t
3478 sync_completed_show(mddev_t *mddev, char *page)
3479 {
3480 unsigned long max_sectors, resync;
3481
3482 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3483 return sprintf(page, "none\n");
3484
3485 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3486 max_sectors = mddev->resync_max_sectors;
3487 else
3488 max_sectors = mddev->dev_sectors;
3489
3490 resync = mddev->curr_resync_completed;
3491 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3492 }
3493
3494 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3495
3496 static ssize_t
3497 min_sync_show(mddev_t *mddev, char *page)
3498 {
3499 return sprintf(page, "%llu\n",
3500 (unsigned long long)mddev->resync_min);
3501 }
3502 static ssize_t
3503 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3504 {
3505 unsigned long long min;
3506 if (strict_strtoull(buf, 10, &min))
3507 return -EINVAL;
3508 if (min > mddev->resync_max)
3509 return -EINVAL;
3510 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3511 return -EBUSY;
3512
3513 /* Must be a multiple of chunk_size */
3514 if (mddev->chunk_sectors) {
3515 sector_t temp = min;
3516 if (sector_div(temp, mddev->chunk_sectors))
3517 return -EINVAL;
3518 }
3519 mddev->resync_min = min;
3520
3521 return len;
3522 }
3523
3524 static struct md_sysfs_entry md_min_sync =
3525 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3526
3527 static ssize_t
3528 max_sync_show(mddev_t *mddev, char *page)
3529 {
3530 if (mddev->resync_max == MaxSector)
3531 return sprintf(page, "max\n");
3532 else
3533 return sprintf(page, "%llu\n",
3534 (unsigned long long)mddev->resync_max);
3535 }
3536 static ssize_t
3537 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3538 {
3539 if (strncmp(buf, "max", 3) == 0)
3540 mddev->resync_max = MaxSector;
3541 else {
3542 unsigned long long max;
3543 if (strict_strtoull(buf, 10, &max))
3544 return -EINVAL;
3545 if (max < mddev->resync_min)
3546 return -EINVAL;
3547 if (max < mddev->resync_max &&
3548 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3549 return -EBUSY;
3550
3551 /* Must be a multiple of chunk_size */
3552 if (mddev->chunk_sectors) {
3553 sector_t temp = max;
3554 if (sector_div(temp, mddev->chunk_sectors))
3555 return -EINVAL;
3556 }
3557 mddev->resync_max = max;
3558 }
3559 wake_up(&mddev->recovery_wait);
3560 return len;
3561 }
3562
3563 static struct md_sysfs_entry md_max_sync =
3564 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3565
3566 static ssize_t
3567 suspend_lo_show(mddev_t *mddev, char *page)
3568 {
3569 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3570 }
3571
3572 static ssize_t
3573 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3574 {
3575 char *e;
3576 unsigned long long new = simple_strtoull(buf, &e, 10);
3577
3578 if (mddev->pers == NULL ||
3579 mddev->pers->quiesce == NULL)
3580 return -EINVAL;
3581 if (buf == e || (*e && *e != '\n'))
3582 return -EINVAL;
3583 if (new >= mddev->suspend_hi ||
3584 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3585 mddev->suspend_lo = new;
3586 mddev->pers->quiesce(mddev, 2);
3587 return len;
3588 } else
3589 return -EINVAL;
3590 }
3591 static struct md_sysfs_entry md_suspend_lo =
3592 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3593
3594
3595 static ssize_t
3596 suspend_hi_show(mddev_t *mddev, char *page)
3597 {
3598 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3599 }
3600
3601 static ssize_t
3602 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3603 {
3604 char *e;
3605 unsigned long long new = simple_strtoull(buf, &e, 10);
3606
3607 if (mddev->pers == NULL ||
3608 mddev->pers->quiesce == NULL)
3609 return -EINVAL;
3610 if (buf == e || (*e && *e != '\n'))
3611 return -EINVAL;
3612 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3613 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3614 mddev->suspend_hi = new;
3615 mddev->pers->quiesce(mddev, 1);
3616 mddev->pers->quiesce(mddev, 0);
3617 return len;
3618 } else
3619 return -EINVAL;
3620 }
3621 static struct md_sysfs_entry md_suspend_hi =
3622 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3623
3624 static ssize_t
3625 reshape_position_show(mddev_t *mddev, char *page)
3626 {
3627 if (mddev->reshape_position != MaxSector)
3628 return sprintf(page, "%llu\n",
3629 (unsigned long long)mddev->reshape_position);
3630 strcpy(page, "none\n");
3631 return 5;
3632 }
3633
3634 static ssize_t
3635 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3636 {
3637 char *e;
3638 unsigned long long new = simple_strtoull(buf, &e, 10);
3639 if (mddev->pers)
3640 return -EBUSY;
3641 if (buf == e || (*e && *e != '\n'))
3642 return -EINVAL;
3643 mddev->reshape_position = new;
3644 mddev->delta_disks = 0;
3645 mddev->new_level = mddev->level;
3646 mddev->new_layout = mddev->layout;
3647 mddev->new_chunk_sectors = mddev->chunk_sectors;
3648 return len;
3649 }
3650
3651 static struct md_sysfs_entry md_reshape_position =
3652 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3653 reshape_position_store);
3654
3655 static ssize_t
3656 array_size_show(mddev_t *mddev, char *page)
3657 {
3658 if (mddev->external_size)
3659 return sprintf(page, "%llu\n",
3660 (unsigned long long)mddev->array_sectors/2);
3661 else
3662 return sprintf(page, "default\n");
3663 }
3664
3665 static ssize_t
3666 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3667 {
3668 sector_t sectors;
3669
3670 if (strncmp(buf, "default", 7) == 0) {
3671 if (mddev->pers)
3672 sectors = mddev->pers->size(mddev, 0, 0);
3673 else
3674 sectors = mddev->array_sectors;
3675
3676 mddev->external_size = 0;
3677 } else {
3678 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3679 return -EINVAL;
3680 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3681 return -E2BIG;
3682
3683 mddev->external_size = 1;
3684 }
3685
3686 mddev->array_sectors = sectors;
3687 set_capacity(mddev->gendisk, mddev->array_sectors);
3688 if (mddev->pers) {
3689 struct block_device *bdev = bdget_disk(mddev->gendisk, 0);
3690
3691 if (bdev) {
3692 mutex_lock(&bdev->bd_inode->i_mutex);
3693 i_size_write(bdev->bd_inode,
3694 (loff_t)mddev->array_sectors << 9);
3695 mutex_unlock(&bdev->bd_inode->i_mutex);
3696 bdput(bdev);
3697 }
3698 }
3699
3700 return len;
3701 }
3702
3703 static struct md_sysfs_entry md_array_size =
3704 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3705 array_size_store);
3706
3707 static struct attribute *md_default_attrs[] = {
3708 &md_level.attr,
3709 &md_layout.attr,
3710 &md_raid_disks.attr,
3711 &md_chunk_size.attr,
3712 &md_size.attr,
3713 &md_resync_start.attr,
3714 &md_metadata.attr,
3715 &md_new_device.attr,
3716 &md_safe_delay.attr,
3717 &md_array_state.attr,
3718 &md_reshape_position.attr,
3719 &md_array_size.attr,
3720 NULL,
3721 };
3722
3723 static struct attribute *md_redundancy_attrs[] = {
3724 &md_scan_mode.attr,
3725 &md_mismatches.attr,
3726 &md_sync_min.attr,
3727 &md_sync_max.attr,
3728 &md_sync_speed.attr,
3729 &md_sync_force_parallel.attr,
3730 &md_sync_completed.attr,
3731 &md_min_sync.attr,
3732 &md_max_sync.attr,
3733 &md_suspend_lo.attr,
3734 &md_suspend_hi.attr,
3735 &md_bitmap.attr,
3736 &md_degraded.attr,
3737 NULL,
3738 };
3739 static struct attribute_group md_redundancy_group = {
3740 .name = NULL,
3741 .attrs = md_redundancy_attrs,
3742 };
3743
3744
3745 static ssize_t
3746 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3747 {
3748 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3749 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3750 ssize_t rv;
3751
3752 if (!entry->show)
3753 return -EIO;
3754 rv = mddev_lock(mddev);
3755 if (!rv) {
3756 rv = entry->show(mddev, page);
3757 mddev_unlock(mddev);
3758 }
3759 return rv;
3760 }
3761
3762 static ssize_t
3763 md_attr_store(struct kobject *kobj, struct attribute *attr,
3764 const char *page, size_t length)
3765 {
3766 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3767 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3768 ssize_t rv;
3769
3770 if (!entry->store)
3771 return -EIO;
3772 if (!capable(CAP_SYS_ADMIN))
3773 return -EACCES;
3774 rv = mddev_lock(mddev);
3775 if (mddev->hold_active == UNTIL_IOCTL)
3776 mddev->hold_active = 0;
3777 if (!rv) {
3778 rv = entry->store(mddev, page, length);
3779 mddev_unlock(mddev);
3780 }
3781 return rv;
3782 }
3783
3784 static void md_free(struct kobject *ko)
3785 {
3786 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3787
3788 if (mddev->sysfs_state)
3789 sysfs_put(mddev->sysfs_state);
3790
3791 if (mddev->gendisk) {
3792 del_gendisk(mddev->gendisk);
3793 put_disk(mddev->gendisk);
3794 }
3795 if (mddev->queue)
3796 blk_cleanup_queue(mddev->queue);
3797
3798 kfree(mddev);
3799 }
3800
3801 static struct sysfs_ops md_sysfs_ops = {
3802 .show = md_attr_show,
3803 .store = md_attr_store,
3804 };
3805 static struct kobj_type md_ktype = {
3806 .release = md_free,
3807 .sysfs_ops = &md_sysfs_ops,
3808 .default_attrs = md_default_attrs,
3809 };
3810
3811 int mdp_major = 0;
3812
3813 static void mddev_delayed_delete(struct work_struct *ws)
3814 {
3815 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3816
3817 if (mddev->private == &md_redundancy_group) {
3818 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3819 if (mddev->sysfs_action)
3820 sysfs_put(mddev->sysfs_action);
3821 mddev->sysfs_action = NULL;
3822 mddev->private = NULL;
3823 }
3824 kobject_del(&mddev->kobj);
3825 kobject_put(&mddev->kobj);
3826 }
3827
3828 static int md_alloc(dev_t dev, char *name)
3829 {
3830 static DEFINE_MUTEX(disks_mutex);
3831 mddev_t *mddev = mddev_find(dev);
3832 struct gendisk *disk;
3833 int partitioned;
3834 int shift;
3835 int unit;
3836 int error;
3837
3838 if (!mddev)
3839 return -ENODEV;
3840
3841 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3842 shift = partitioned ? MdpMinorShift : 0;
3843 unit = MINOR(mddev->unit) >> shift;
3844
3845 /* wait for any previous instance if this device
3846 * to be completed removed (mddev_delayed_delete).
3847 */
3848 flush_scheduled_work();
3849
3850 mutex_lock(&disks_mutex);
3851 error = -EEXIST;
3852 if (mddev->gendisk)
3853 goto abort;
3854
3855 if (name) {
3856 /* Need to ensure that 'name' is not a duplicate.
3857 */
3858 mddev_t *mddev2;
3859 spin_lock(&all_mddevs_lock);
3860
3861 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3862 if (mddev2->gendisk &&
3863 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3864 spin_unlock(&all_mddevs_lock);
3865 goto abort;
3866 }
3867 spin_unlock(&all_mddevs_lock);
3868 }
3869
3870 error = -ENOMEM;
3871 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3872 if (!mddev->queue)
3873 goto abort;
3874 mddev->queue->queuedata = mddev;
3875
3876 /* Can be unlocked because the queue is new: no concurrency */
3877 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3878
3879 blk_queue_make_request(mddev->queue, md_make_request);
3880
3881 disk = alloc_disk(1 << shift);
3882 if (!disk) {
3883 blk_cleanup_queue(mddev->queue);
3884 mddev->queue = NULL;
3885 goto abort;
3886 }
3887 disk->major = MAJOR(mddev->unit);
3888 disk->first_minor = unit << shift;
3889 if (name)
3890 strcpy(disk->disk_name, name);
3891 else if (partitioned)
3892 sprintf(disk->disk_name, "md_d%d", unit);
3893 else
3894 sprintf(disk->disk_name, "md%d", unit);
3895 disk->fops = &md_fops;
3896 disk->private_data = mddev;
3897 disk->queue = mddev->queue;
3898 /* Allow extended partitions. This makes the
3899 * 'mdp' device redundant, but we can't really
3900 * remove it now.
3901 */
3902 disk->flags |= GENHD_FL_EXT_DEVT;
3903 add_disk(disk);
3904 mddev->gendisk = disk;
3905 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3906 &disk_to_dev(disk)->kobj, "%s", "md");
3907 if (error) {
3908 /* This isn't possible, but as kobject_init_and_add is marked
3909 * __must_check, we must do something with the result
3910 */
3911 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3912 disk->disk_name);
3913 error = 0;
3914 }
3915 abort:
3916 mutex_unlock(&disks_mutex);
3917 if (!error) {
3918 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3919 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3920 }
3921 mddev_put(mddev);
3922 return error;
3923 }
3924
3925 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3926 {
3927 md_alloc(dev, NULL);
3928 return NULL;
3929 }
3930
3931 static int add_named_array(const char *val, struct kernel_param *kp)
3932 {
3933 /* val must be "md_*" where * is not all digits.
3934 * We allocate an array with a large free minor number, and
3935 * set the name to val. val must not already be an active name.
3936 */
3937 int len = strlen(val);
3938 char buf[DISK_NAME_LEN];
3939
3940 while (len && val[len-1] == '\n')
3941 len--;
3942 if (len >= DISK_NAME_LEN)
3943 return -E2BIG;
3944 strlcpy(buf, val, len+1);
3945 if (strncmp(buf, "md_", 3) != 0)
3946 return -EINVAL;
3947 return md_alloc(0, buf);
3948 }
3949
3950 static void md_safemode_timeout(unsigned long data)
3951 {
3952 mddev_t *mddev = (mddev_t *) data;
3953
3954 if (!atomic_read(&mddev->writes_pending)) {
3955 mddev->safemode = 1;
3956 if (mddev->external)
3957 sysfs_notify_dirent(mddev->sysfs_state);
3958 }
3959 md_wakeup_thread(mddev->thread);
3960 }
3961
3962 static int start_dirty_degraded;
3963
3964 static int do_md_run(mddev_t * mddev)
3965 {
3966 int err;
3967 mdk_rdev_t *rdev;
3968 struct gendisk *disk;
3969 struct mdk_personality *pers;
3970
3971 if (list_empty(&mddev->disks))
3972 /* cannot run an array with no devices.. */
3973 return -EINVAL;
3974
3975 if (mddev->pers)
3976 return -EBUSY;
3977
3978 /*
3979 * Analyze all RAID superblock(s)
3980 */
3981 if (!mddev->raid_disks) {
3982 if (!mddev->persistent)
3983 return -EINVAL;
3984 analyze_sbs(mddev);
3985 }
3986
3987 if (mddev->level != LEVEL_NONE)
3988 request_module("md-level-%d", mddev->level);
3989 else if (mddev->clevel[0])
3990 request_module("md-%s", mddev->clevel);
3991
3992 /*
3993 * Drop all container device buffers, from now on
3994 * the only valid external interface is through the md
3995 * device.
3996 */
3997 list_for_each_entry(rdev, &mddev->disks, same_set) {
3998 if (test_bit(Faulty, &rdev->flags))
3999 continue;
4000 sync_blockdev(rdev->bdev);
4001 invalidate_bdev(rdev->bdev);
4002
4003 /* perform some consistency tests on the device.
4004 * We don't want the data to overlap the metadata,
4005 * Internal Bitmap issues have been handled elsewhere.
4006 */
4007 if (rdev->data_offset < rdev->sb_start) {
4008 if (mddev->dev_sectors &&
4009 rdev->data_offset + mddev->dev_sectors
4010 > rdev->sb_start) {
4011 printk("md: %s: data overlaps metadata\n",
4012 mdname(mddev));
4013 return -EINVAL;
4014 }
4015 } else {
4016 if (rdev->sb_start + rdev->sb_size/512
4017 > rdev->data_offset) {
4018 printk("md: %s: metadata overlaps data\n",
4019 mdname(mddev));
4020 return -EINVAL;
4021 }
4022 }
4023 sysfs_notify_dirent(rdev->sysfs_state);
4024 }
4025
4026 md_probe(mddev->unit, NULL, NULL);
4027 disk = mddev->gendisk;
4028 if (!disk)
4029 return -ENOMEM;
4030
4031 spin_lock(&pers_lock);
4032 pers = find_pers(mddev->level, mddev->clevel);
4033 if (!pers || !try_module_get(pers->owner)) {
4034 spin_unlock(&pers_lock);
4035 if (mddev->level != LEVEL_NONE)
4036 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4037 mddev->level);
4038 else
4039 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4040 mddev->clevel);
4041 return -EINVAL;
4042 }
4043 mddev->pers = pers;
4044 spin_unlock(&pers_lock);
4045 if (mddev->level != pers->level) {
4046 mddev->level = pers->level;
4047 mddev->new_level = pers->level;
4048 }
4049 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4050
4051 if (pers->level >= 4 && pers->level <= 6)
4052 /* Cannot support integrity (yet) */
4053 blk_integrity_unregister(mddev->gendisk);
4054
4055 if (mddev->reshape_position != MaxSector &&
4056 pers->start_reshape == NULL) {
4057 /* This personality cannot handle reshaping... */
4058 mddev->pers = NULL;
4059 module_put(pers->owner);
4060 return -EINVAL;
4061 }
4062
4063 if (pers->sync_request) {
4064 /* Warn if this is a potentially silly
4065 * configuration.
4066 */
4067 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4068 mdk_rdev_t *rdev2;
4069 int warned = 0;
4070
4071 list_for_each_entry(rdev, &mddev->disks, same_set)
4072 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4073 if (rdev < rdev2 &&
4074 rdev->bdev->bd_contains ==
4075 rdev2->bdev->bd_contains) {
4076 printk(KERN_WARNING
4077 "%s: WARNING: %s appears to be"
4078 " on the same physical disk as"
4079 " %s.\n",
4080 mdname(mddev),
4081 bdevname(rdev->bdev,b),
4082 bdevname(rdev2->bdev,b2));
4083 warned = 1;
4084 }
4085 }
4086
4087 if (warned)
4088 printk(KERN_WARNING
4089 "True protection against single-disk"
4090 " failure might be compromised.\n");
4091 }
4092
4093 mddev->recovery = 0;
4094 /* may be over-ridden by personality */
4095 mddev->resync_max_sectors = mddev->dev_sectors;
4096
4097 mddev->barriers_work = 1;
4098 mddev->ok_start_degraded = start_dirty_degraded;
4099
4100 if (start_readonly)
4101 mddev->ro = 2; /* read-only, but switch on first write */
4102
4103 err = mddev->pers->run(mddev);
4104 if (err)
4105 printk(KERN_ERR "md: pers->run() failed ...\n");
4106 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4107 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4108 " but 'external_size' not in effect?\n", __func__);
4109 printk(KERN_ERR
4110 "md: invalid array_size %llu > default size %llu\n",
4111 (unsigned long long)mddev->array_sectors / 2,
4112 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4113 err = -EINVAL;
4114 mddev->pers->stop(mddev);
4115 }
4116 if (err == 0 && mddev->pers->sync_request) {
4117 err = bitmap_create(mddev);
4118 if (err) {
4119 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4120 mdname(mddev), err);
4121 mddev->pers->stop(mddev);
4122 }
4123 }
4124 if (err) {
4125 module_put(mddev->pers->owner);
4126 mddev->pers = NULL;
4127 bitmap_destroy(mddev);
4128 return err;
4129 }
4130 if (mddev->pers->sync_request) {
4131 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4132 printk(KERN_WARNING
4133 "md: cannot register extra attributes for %s\n",
4134 mdname(mddev));
4135 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4136 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4137 mddev->ro = 0;
4138
4139 atomic_set(&mddev->writes_pending,0);
4140 mddev->safemode = 0;
4141 mddev->safemode_timer.function = md_safemode_timeout;
4142 mddev->safemode_timer.data = (unsigned long) mddev;
4143 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4144 mddev->in_sync = 1;
4145
4146 list_for_each_entry(rdev, &mddev->disks, same_set)
4147 if (rdev->raid_disk >= 0) {
4148 char nm[20];
4149 sprintf(nm, "rd%d", rdev->raid_disk);
4150 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4151 printk("md: cannot register %s for %s\n",
4152 nm, mdname(mddev));
4153 }
4154
4155 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4156
4157 if (mddev->flags)
4158 md_update_sb(mddev, 0);
4159
4160 set_capacity(disk, mddev->array_sectors);
4161
4162 /* If there is a partially-recovered drive we need to
4163 * start recovery here. If we leave it to md_check_recovery,
4164 * it will remove the drives and not do the right thing
4165 */
4166 if (mddev->degraded && !mddev->sync_thread) {
4167 int spares = 0;
4168 list_for_each_entry(rdev, &mddev->disks, same_set)
4169 if (rdev->raid_disk >= 0 &&
4170 !test_bit(In_sync, &rdev->flags) &&
4171 !test_bit(Faulty, &rdev->flags))
4172 /* complete an interrupted recovery */
4173 spares++;
4174 if (spares && mddev->pers->sync_request) {
4175 mddev->recovery = 0;
4176 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4177 mddev->sync_thread = md_register_thread(md_do_sync,
4178 mddev,
4179 "%s_resync");
4180 if (!mddev->sync_thread) {
4181 printk(KERN_ERR "%s: could not start resync"
4182 " thread...\n",
4183 mdname(mddev));
4184 /* leave the spares where they are, it shouldn't hurt */
4185 mddev->recovery = 0;
4186 }
4187 }
4188 }
4189 md_wakeup_thread(mddev->thread);
4190 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4191
4192 mddev->changed = 1;
4193 md_new_event(mddev);
4194 sysfs_notify_dirent(mddev->sysfs_state);
4195 if (mddev->sysfs_action)
4196 sysfs_notify_dirent(mddev->sysfs_action);
4197 sysfs_notify(&mddev->kobj, NULL, "degraded");
4198 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4199 return 0;
4200 }
4201
4202 static int restart_array(mddev_t *mddev)
4203 {
4204 struct gendisk *disk = mddev->gendisk;
4205
4206 /* Complain if it has no devices */
4207 if (list_empty(&mddev->disks))
4208 return -ENXIO;
4209 if (!mddev->pers)
4210 return -EINVAL;
4211 if (!mddev->ro)
4212 return -EBUSY;
4213 mddev->safemode = 0;
4214 mddev->ro = 0;
4215 set_disk_ro(disk, 0);
4216 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4217 mdname(mddev));
4218 /* Kick recovery or resync if necessary */
4219 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4220 md_wakeup_thread(mddev->thread);
4221 md_wakeup_thread(mddev->sync_thread);
4222 sysfs_notify_dirent(mddev->sysfs_state);
4223 return 0;
4224 }
4225
4226 /* similar to deny_write_access, but accounts for our holding a reference
4227 * to the file ourselves */
4228 static int deny_bitmap_write_access(struct file * file)
4229 {
4230 struct inode *inode = file->f_mapping->host;
4231
4232 spin_lock(&inode->i_lock);
4233 if (atomic_read(&inode->i_writecount) > 1) {
4234 spin_unlock(&inode->i_lock);
4235 return -ETXTBSY;
4236 }
4237 atomic_set(&inode->i_writecount, -1);
4238 spin_unlock(&inode->i_lock);
4239
4240 return 0;
4241 }
4242
4243 static void restore_bitmap_write_access(struct file *file)
4244 {
4245 struct inode *inode = file->f_mapping->host;
4246
4247 spin_lock(&inode->i_lock);
4248 atomic_set(&inode->i_writecount, 1);
4249 spin_unlock(&inode->i_lock);
4250 }
4251
4252 /* mode:
4253 * 0 - completely stop and dis-assemble array
4254 * 1 - switch to readonly
4255 * 2 - stop but do not disassemble array
4256 */
4257 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4258 {
4259 int err = 0;
4260 struct gendisk *disk = mddev->gendisk;
4261 mdk_rdev_t *rdev;
4262
4263 if (atomic_read(&mddev->openers) > is_open) {
4264 printk("md: %s still in use.\n",mdname(mddev));
4265 return -EBUSY;
4266 }
4267
4268 if (mddev->pers) {
4269
4270 if (mddev->sync_thread) {
4271 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4272 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4273 md_unregister_thread(mddev->sync_thread);
4274 mddev->sync_thread = NULL;
4275 }
4276
4277 del_timer_sync(&mddev->safemode_timer);
4278
4279 switch(mode) {
4280 case 1: /* readonly */
4281 err = -ENXIO;
4282 if (mddev->ro==1)
4283 goto out;
4284 mddev->ro = 1;
4285 break;
4286 case 0: /* disassemble */
4287 case 2: /* stop */
4288 bitmap_flush(mddev);
4289 md_super_wait(mddev);
4290 if (mddev->ro)
4291 set_disk_ro(disk, 0);
4292
4293 mddev->pers->stop(mddev);
4294 mddev->queue->merge_bvec_fn = NULL;
4295 mddev->queue->unplug_fn = NULL;
4296 mddev->queue->backing_dev_info.congested_fn = NULL;
4297 module_put(mddev->pers->owner);
4298 if (mddev->pers->sync_request)
4299 mddev->private = &md_redundancy_group;
4300 mddev->pers = NULL;
4301 /* tell userspace to handle 'inactive' */
4302 sysfs_notify_dirent(mddev->sysfs_state);
4303
4304 list_for_each_entry(rdev, &mddev->disks, same_set)
4305 if (rdev->raid_disk >= 0) {
4306 char nm[20];
4307 sprintf(nm, "rd%d", rdev->raid_disk);
4308 sysfs_remove_link(&mddev->kobj, nm);
4309 }
4310
4311 set_capacity(disk, 0);
4312 mddev->changed = 1;
4313
4314 if (mddev->ro)
4315 mddev->ro = 0;
4316 }
4317 if (!mddev->in_sync || mddev->flags) {
4318 /* mark array as shutdown cleanly */
4319 mddev->in_sync = 1;
4320 md_update_sb(mddev, 1);
4321 }
4322 if (mode == 1)
4323 set_disk_ro(disk, 1);
4324 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4325 }
4326
4327 /*
4328 * Free resources if final stop
4329 */
4330 if (mode == 0) {
4331
4332 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4333
4334 bitmap_destroy(mddev);
4335 if (mddev->bitmap_file) {
4336 restore_bitmap_write_access(mddev->bitmap_file);
4337 fput(mddev->bitmap_file);
4338 mddev->bitmap_file = NULL;
4339 }
4340 mddev->bitmap_offset = 0;
4341
4342 /* make sure all md_delayed_delete calls have finished */
4343 flush_scheduled_work();
4344
4345 export_array(mddev);
4346
4347 mddev->array_sectors = 0;
4348 mddev->external_size = 0;
4349 mddev->dev_sectors = 0;
4350 mddev->raid_disks = 0;
4351 mddev->recovery_cp = 0;
4352 mddev->resync_min = 0;
4353 mddev->resync_max = MaxSector;
4354 mddev->reshape_position = MaxSector;
4355 mddev->external = 0;
4356 mddev->persistent = 0;
4357 mddev->level = LEVEL_NONE;
4358 mddev->clevel[0] = 0;
4359 mddev->flags = 0;
4360 mddev->ro = 0;
4361 mddev->metadata_type[0] = 0;
4362 mddev->chunk_sectors = 0;
4363 mddev->ctime = mddev->utime = 0;
4364 mddev->layout = 0;
4365 mddev->max_disks = 0;
4366 mddev->events = 0;
4367 mddev->delta_disks = 0;
4368 mddev->new_level = LEVEL_NONE;
4369 mddev->new_layout = 0;
4370 mddev->new_chunk_sectors = 0;
4371 mddev->curr_resync = 0;
4372 mddev->resync_mismatches = 0;
4373 mddev->suspend_lo = mddev->suspend_hi = 0;
4374 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4375 mddev->recovery = 0;
4376 mddev->in_sync = 0;
4377 mddev->changed = 0;
4378 mddev->degraded = 0;
4379 mddev->barriers_work = 0;
4380 mddev->safemode = 0;
4381 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4382 if (mddev->hold_active == UNTIL_STOP)
4383 mddev->hold_active = 0;
4384
4385 } else if (mddev->pers)
4386 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4387 mdname(mddev));
4388 err = 0;
4389 blk_integrity_unregister(disk);
4390 md_new_event(mddev);
4391 sysfs_notify_dirent(mddev->sysfs_state);
4392 out:
4393 return err;
4394 }
4395
4396 #ifndef MODULE
4397 static void autorun_array(mddev_t *mddev)
4398 {
4399 mdk_rdev_t *rdev;
4400 int err;
4401
4402 if (list_empty(&mddev->disks))
4403 return;
4404
4405 printk(KERN_INFO "md: running: ");
4406
4407 list_for_each_entry(rdev, &mddev->disks, same_set) {
4408 char b[BDEVNAME_SIZE];
4409 printk("<%s>", bdevname(rdev->bdev,b));
4410 }
4411 printk("\n");
4412
4413 err = do_md_run(mddev);
4414 if (err) {
4415 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4416 do_md_stop(mddev, 0, 0);
4417 }
4418 }
4419
4420 /*
4421 * lets try to run arrays based on all disks that have arrived
4422 * until now. (those are in pending_raid_disks)
4423 *
4424 * the method: pick the first pending disk, collect all disks with
4425 * the same UUID, remove all from the pending list and put them into
4426 * the 'same_array' list. Then order this list based on superblock
4427 * update time (freshest comes first), kick out 'old' disks and
4428 * compare superblocks. If everything's fine then run it.
4429 *
4430 * If "unit" is allocated, then bump its reference count
4431 */
4432 static void autorun_devices(int part)
4433 {
4434 mdk_rdev_t *rdev0, *rdev, *tmp;
4435 mddev_t *mddev;
4436 char b[BDEVNAME_SIZE];
4437
4438 printk(KERN_INFO "md: autorun ...\n");
4439 while (!list_empty(&pending_raid_disks)) {
4440 int unit;
4441 dev_t dev;
4442 LIST_HEAD(candidates);
4443 rdev0 = list_entry(pending_raid_disks.next,
4444 mdk_rdev_t, same_set);
4445
4446 printk(KERN_INFO "md: considering %s ...\n",
4447 bdevname(rdev0->bdev,b));
4448 INIT_LIST_HEAD(&candidates);
4449 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4450 if (super_90_load(rdev, rdev0, 0) >= 0) {
4451 printk(KERN_INFO "md: adding %s ...\n",
4452 bdevname(rdev->bdev,b));
4453 list_move(&rdev->same_set, &candidates);
4454 }
4455 /*
4456 * now we have a set of devices, with all of them having
4457 * mostly sane superblocks. It's time to allocate the
4458 * mddev.
4459 */
4460 if (part) {
4461 dev = MKDEV(mdp_major,
4462 rdev0->preferred_minor << MdpMinorShift);
4463 unit = MINOR(dev) >> MdpMinorShift;
4464 } else {
4465 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4466 unit = MINOR(dev);
4467 }
4468 if (rdev0->preferred_minor != unit) {
4469 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4470 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4471 break;
4472 }
4473
4474 md_probe(dev, NULL, NULL);
4475 mddev = mddev_find(dev);
4476 if (!mddev || !mddev->gendisk) {
4477 if (mddev)
4478 mddev_put(mddev);
4479 printk(KERN_ERR
4480 "md: cannot allocate memory for md drive.\n");
4481 break;
4482 }
4483 if (mddev_lock(mddev))
4484 printk(KERN_WARNING "md: %s locked, cannot run\n",
4485 mdname(mddev));
4486 else if (mddev->raid_disks || mddev->major_version
4487 || !list_empty(&mddev->disks)) {
4488 printk(KERN_WARNING
4489 "md: %s already running, cannot run %s\n",
4490 mdname(mddev), bdevname(rdev0->bdev,b));
4491 mddev_unlock(mddev);
4492 } else {
4493 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4494 mddev->persistent = 1;
4495 rdev_for_each_list(rdev, tmp, &candidates) {
4496 list_del_init(&rdev->same_set);
4497 if (bind_rdev_to_array(rdev, mddev))
4498 export_rdev(rdev);
4499 }
4500 autorun_array(mddev);
4501 mddev_unlock(mddev);
4502 }
4503 /* on success, candidates will be empty, on error
4504 * it won't...
4505 */
4506 rdev_for_each_list(rdev, tmp, &candidates) {
4507 list_del_init(&rdev->same_set);
4508 export_rdev(rdev);
4509 }
4510 mddev_put(mddev);
4511 }
4512 printk(KERN_INFO "md: ... autorun DONE.\n");
4513 }
4514 #endif /* !MODULE */
4515
4516 static int get_version(void __user * arg)
4517 {
4518 mdu_version_t ver;
4519
4520 ver.major = MD_MAJOR_VERSION;
4521 ver.minor = MD_MINOR_VERSION;
4522 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4523
4524 if (copy_to_user(arg, &ver, sizeof(ver)))
4525 return -EFAULT;
4526
4527 return 0;
4528 }
4529
4530 static int get_array_info(mddev_t * mddev, void __user * arg)
4531 {
4532 mdu_array_info_t info;
4533 int nr,working,active,failed,spare;
4534 mdk_rdev_t *rdev;
4535
4536 nr=working=active=failed=spare=0;
4537 list_for_each_entry(rdev, &mddev->disks, same_set) {
4538 nr++;
4539 if (test_bit(Faulty, &rdev->flags))
4540 failed++;
4541 else {
4542 working++;
4543 if (test_bit(In_sync, &rdev->flags))
4544 active++;
4545 else
4546 spare++;
4547 }
4548 }
4549
4550 info.major_version = mddev->major_version;
4551 info.minor_version = mddev->minor_version;
4552 info.patch_version = MD_PATCHLEVEL_VERSION;
4553 info.ctime = mddev->ctime;
4554 info.level = mddev->level;
4555 info.size = mddev->dev_sectors / 2;
4556 if (info.size != mddev->dev_sectors / 2) /* overflow */
4557 info.size = -1;
4558 info.nr_disks = nr;
4559 info.raid_disks = mddev->raid_disks;
4560 info.md_minor = mddev->md_minor;
4561 info.not_persistent= !mddev->persistent;
4562
4563 info.utime = mddev->utime;
4564 info.state = 0;
4565 if (mddev->in_sync)
4566 info.state = (1<<MD_SB_CLEAN);
4567 if (mddev->bitmap && mddev->bitmap_offset)
4568 info.state = (1<<MD_SB_BITMAP_PRESENT);
4569 info.active_disks = active;
4570 info.working_disks = working;
4571 info.failed_disks = failed;
4572 info.spare_disks = spare;
4573
4574 info.layout = mddev->layout;
4575 info.chunk_size = mddev->chunk_sectors << 9;
4576
4577 if (copy_to_user(arg, &info, sizeof(info)))
4578 return -EFAULT;
4579
4580 return 0;
4581 }
4582
4583 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4584 {
4585 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4586 char *ptr, *buf = NULL;
4587 int err = -ENOMEM;
4588
4589 if (md_allow_write(mddev))
4590 file = kmalloc(sizeof(*file), GFP_NOIO);
4591 else
4592 file = kmalloc(sizeof(*file), GFP_KERNEL);
4593
4594 if (!file)
4595 goto out;
4596
4597 /* bitmap disabled, zero the first byte and copy out */
4598 if (!mddev->bitmap || !mddev->bitmap->file) {
4599 file->pathname[0] = '\0';
4600 goto copy_out;
4601 }
4602
4603 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4604 if (!buf)
4605 goto out;
4606
4607 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4608 if (IS_ERR(ptr))
4609 goto out;
4610
4611 strcpy(file->pathname, ptr);
4612
4613 copy_out:
4614 err = 0;
4615 if (copy_to_user(arg, file, sizeof(*file)))
4616 err = -EFAULT;
4617 out:
4618 kfree(buf);
4619 kfree(file);
4620 return err;
4621 }
4622
4623 static int get_disk_info(mddev_t * mddev, void __user * arg)
4624 {
4625 mdu_disk_info_t info;
4626 mdk_rdev_t *rdev;
4627
4628 if (copy_from_user(&info, arg, sizeof(info)))
4629 return -EFAULT;
4630
4631 rdev = find_rdev_nr(mddev, info.number);
4632 if (rdev) {
4633 info.major = MAJOR(rdev->bdev->bd_dev);
4634 info.minor = MINOR(rdev->bdev->bd_dev);
4635 info.raid_disk = rdev->raid_disk;
4636 info.state = 0;
4637 if (test_bit(Faulty, &rdev->flags))
4638 info.state |= (1<<MD_DISK_FAULTY);
4639 else if (test_bit(In_sync, &rdev->flags)) {
4640 info.state |= (1<<MD_DISK_ACTIVE);
4641 info.state |= (1<<MD_DISK_SYNC);
4642 }
4643 if (test_bit(WriteMostly, &rdev->flags))
4644 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4645 } else {
4646 info.major = info.minor = 0;
4647 info.raid_disk = -1;
4648 info.state = (1<<MD_DISK_REMOVED);
4649 }
4650
4651 if (copy_to_user(arg, &info, sizeof(info)))
4652 return -EFAULT;
4653
4654 return 0;
4655 }
4656
4657 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4658 {
4659 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4660 mdk_rdev_t *rdev;
4661 dev_t dev = MKDEV(info->major,info->minor);
4662
4663 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4664 return -EOVERFLOW;
4665
4666 if (!mddev->raid_disks) {
4667 int err;
4668 /* expecting a device which has a superblock */
4669 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4670 if (IS_ERR(rdev)) {
4671 printk(KERN_WARNING
4672 "md: md_import_device returned %ld\n",
4673 PTR_ERR(rdev));
4674 return PTR_ERR(rdev);
4675 }
4676 if (!list_empty(&mddev->disks)) {
4677 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4678 mdk_rdev_t, same_set);
4679 int err = super_types[mddev->major_version]
4680 .load_super(rdev, rdev0, mddev->minor_version);
4681 if (err < 0) {
4682 printk(KERN_WARNING
4683 "md: %s has different UUID to %s\n",
4684 bdevname(rdev->bdev,b),
4685 bdevname(rdev0->bdev,b2));
4686 export_rdev(rdev);
4687 return -EINVAL;
4688 }
4689 }
4690 err = bind_rdev_to_array(rdev, mddev);
4691 if (err)
4692 export_rdev(rdev);
4693 return err;
4694 }
4695
4696 /*
4697 * add_new_disk can be used once the array is assembled
4698 * to add "hot spares". They must already have a superblock
4699 * written
4700 */
4701 if (mddev->pers) {
4702 int err;
4703 if (!mddev->pers->hot_add_disk) {
4704 printk(KERN_WARNING
4705 "%s: personality does not support diskops!\n",
4706 mdname(mddev));
4707 return -EINVAL;
4708 }
4709 if (mddev->persistent)
4710 rdev = md_import_device(dev, mddev->major_version,
4711 mddev->minor_version);
4712 else
4713 rdev = md_import_device(dev, -1, -1);
4714 if (IS_ERR(rdev)) {
4715 printk(KERN_WARNING
4716 "md: md_import_device returned %ld\n",
4717 PTR_ERR(rdev));
4718 return PTR_ERR(rdev);
4719 }
4720 /* set save_raid_disk if appropriate */
4721 if (!mddev->persistent) {
4722 if (info->state & (1<<MD_DISK_SYNC) &&
4723 info->raid_disk < mddev->raid_disks)
4724 rdev->raid_disk = info->raid_disk;
4725 else
4726 rdev->raid_disk = -1;
4727 } else
4728 super_types[mddev->major_version].
4729 validate_super(mddev, rdev);
4730 rdev->saved_raid_disk = rdev->raid_disk;
4731
4732 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4733 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4734 set_bit(WriteMostly, &rdev->flags);
4735 else
4736 clear_bit(WriteMostly, &rdev->flags);
4737
4738 rdev->raid_disk = -1;
4739 err = bind_rdev_to_array(rdev, mddev);
4740 if (!err && !mddev->pers->hot_remove_disk) {
4741 /* If there is hot_add_disk but no hot_remove_disk
4742 * then added disks for geometry changes,
4743 * and should be added immediately.
4744 */
4745 super_types[mddev->major_version].
4746 validate_super(mddev, rdev);
4747 err = mddev->pers->hot_add_disk(mddev, rdev);
4748 if (err)
4749 unbind_rdev_from_array(rdev);
4750 }
4751 if (err)
4752 export_rdev(rdev);
4753 else
4754 sysfs_notify_dirent(rdev->sysfs_state);
4755
4756 md_update_sb(mddev, 1);
4757 if (mddev->degraded)
4758 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4759 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4760 md_wakeup_thread(mddev->thread);
4761 return err;
4762 }
4763
4764 /* otherwise, add_new_disk is only allowed
4765 * for major_version==0 superblocks
4766 */
4767 if (mddev->major_version != 0) {
4768 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4769 mdname(mddev));
4770 return -EINVAL;
4771 }
4772
4773 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4774 int err;
4775 rdev = md_import_device(dev, -1, 0);
4776 if (IS_ERR(rdev)) {
4777 printk(KERN_WARNING
4778 "md: error, md_import_device() returned %ld\n",
4779 PTR_ERR(rdev));
4780 return PTR_ERR(rdev);
4781 }
4782 rdev->desc_nr = info->number;
4783 if (info->raid_disk < mddev->raid_disks)
4784 rdev->raid_disk = info->raid_disk;
4785 else
4786 rdev->raid_disk = -1;
4787
4788 if (rdev->raid_disk < mddev->raid_disks)
4789 if (info->state & (1<<MD_DISK_SYNC))
4790 set_bit(In_sync, &rdev->flags);
4791
4792 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4793 set_bit(WriteMostly, &rdev->flags);
4794
4795 if (!mddev->persistent) {
4796 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4797 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4798 } else
4799 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4800 rdev->sectors = rdev->sb_start;
4801
4802 err = bind_rdev_to_array(rdev, mddev);
4803 if (err) {
4804 export_rdev(rdev);
4805 return err;
4806 }
4807 }
4808
4809 return 0;
4810 }
4811
4812 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4813 {
4814 char b[BDEVNAME_SIZE];
4815 mdk_rdev_t *rdev;
4816
4817 rdev = find_rdev(mddev, dev);
4818 if (!rdev)
4819 return -ENXIO;
4820
4821 if (rdev->raid_disk >= 0)
4822 goto busy;
4823
4824 kick_rdev_from_array(rdev);
4825 md_update_sb(mddev, 1);
4826 md_new_event(mddev);
4827
4828 return 0;
4829 busy:
4830 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4831 bdevname(rdev->bdev,b), mdname(mddev));
4832 return -EBUSY;
4833 }
4834
4835 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4836 {
4837 char b[BDEVNAME_SIZE];
4838 int err;
4839 mdk_rdev_t *rdev;
4840
4841 if (!mddev->pers)
4842 return -ENODEV;
4843
4844 if (mddev->major_version != 0) {
4845 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4846 " version-0 superblocks.\n",
4847 mdname(mddev));
4848 return -EINVAL;
4849 }
4850 if (!mddev->pers->hot_add_disk) {
4851 printk(KERN_WARNING
4852 "%s: personality does not support diskops!\n",
4853 mdname(mddev));
4854 return -EINVAL;
4855 }
4856
4857 rdev = md_import_device(dev, -1, 0);
4858 if (IS_ERR(rdev)) {
4859 printk(KERN_WARNING
4860 "md: error, md_import_device() returned %ld\n",
4861 PTR_ERR(rdev));
4862 return -EINVAL;
4863 }
4864
4865 if (mddev->persistent)
4866 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4867 else
4868 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4869
4870 rdev->sectors = rdev->sb_start;
4871
4872 if (test_bit(Faulty, &rdev->flags)) {
4873 printk(KERN_WARNING
4874 "md: can not hot-add faulty %s disk to %s!\n",
4875 bdevname(rdev->bdev,b), mdname(mddev));
4876 err = -EINVAL;
4877 goto abort_export;
4878 }
4879 clear_bit(In_sync, &rdev->flags);
4880 rdev->desc_nr = -1;
4881 rdev->saved_raid_disk = -1;
4882 err = bind_rdev_to_array(rdev, mddev);
4883 if (err)
4884 goto abort_export;
4885
4886 /*
4887 * The rest should better be atomic, we can have disk failures
4888 * noticed in interrupt contexts ...
4889 */
4890
4891 rdev->raid_disk = -1;
4892
4893 md_update_sb(mddev, 1);
4894
4895 /*
4896 * Kick recovery, maybe this spare has to be added to the
4897 * array immediately.
4898 */
4899 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4900 md_wakeup_thread(mddev->thread);
4901 md_new_event(mddev);
4902 return 0;
4903
4904 abort_export:
4905 export_rdev(rdev);
4906 return err;
4907 }
4908
4909 static int set_bitmap_file(mddev_t *mddev, int fd)
4910 {
4911 int err;
4912
4913 if (mddev->pers) {
4914 if (!mddev->pers->quiesce)
4915 return -EBUSY;
4916 if (mddev->recovery || mddev->sync_thread)
4917 return -EBUSY;
4918 /* we should be able to change the bitmap.. */
4919 }
4920
4921
4922 if (fd >= 0) {
4923 if (mddev->bitmap)
4924 return -EEXIST; /* cannot add when bitmap is present */
4925 mddev->bitmap_file = fget(fd);
4926
4927 if (mddev->bitmap_file == NULL) {
4928 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4929 mdname(mddev));
4930 return -EBADF;
4931 }
4932
4933 err = deny_bitmap_write_access(mddev->bitmap_file);
4934 if (err) {
4935 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4936 mdname(mddev));
4937 fput(mddev->bitmap_file);
4938 mddev->bitmap_file = NULL;
4939 return err;
4940 }
4941 mddev->bitmap_offset = 0; /* file overrides offset */
4942 } else if (mddev->bitmap == NULL)
4943 return -ENOENT; /* cannot remove what isn't there */
4944 err = 0;
4945 if (mddev->pers) {
4946 mddev->pers->quiesce(mddev, 1);
4947 if (fd >= 0)
4948 err = bitmap_create(mddev);
4949 if (fd < 0 || err) {
4950 bitmap_destroy(mddev);
4951 fd = -1; /* make sure to put the file */
4952 }
4953 mddev->pers->quiesce(mddev, 0);
4954 }
4955 if (fd < 0) {
4956 if (mddev->bitmap_file) {
4957 restore_bitmap_write_access(mddev->bitmap_file);
4958 fput(mddev->bitmap_file);
4959 }
4960 mddev->bitmap_file = NULL;
4961 }
4962
4963 return err;
4964 }
4965
4966 /*
4967 * set_array_info is used two different ways
4968 * The original usage is when creating a new array.
4969 * In this usage, raid_disks is > 0 and it together with
4970 * level, size, not_persistent,layout,chunksize determine the
4971 * shape of the array.
4972 * This will always create an array with a type-0.90.0 superblock.
4973 * The newer usage is when assembling an array.
4974 * In this case raid_disks will be 0, and the major_version field is
4975 * use to determine which style super-blocks are to be found on the devices.
4976 * The minor and patch _version numbers are also kept incase the
4977 * super_block handler wishes to interpret them.
4978 */
4979 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4980 {
4981
4982 if (info->raid_disks == 0) {
4983 /* just setting version number for superblock loading */
4984 if (info->major_version < 0 ||
4985 info->major_version >= ARRAY_SIZE(super_types) ||
4986 super_types[info->major_version].name == NULL) {
4987 /* maybe try to auto-load a module? */
4988 printk(KERN_INFO
4989 "md: superblock version %d not known\n",
4990 info->major_version);
4991 return -EINVAL;
4992 }
4993 mddev->major_version = info->major_version;
4994 mddev->minor_version = info->minor_version;
4995 mddev->patch_version = info->patch_version;
4996 mddev->persistent = !info->not_persistent;
4997 return 0;
4998 }
4999 mddev->major_version = MD_MAJOR_VERSION;
5000 mddev->minor_version = MD_MINOR_VERSION;
5001 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5002 mddev->ctime = get_seconds();
5003
5004 mddev->level = info->level;
5005 mddev->clevel[0] = 0;
5006 mddev->dev_sectors = 2 * (sector_t)info->size;
5007 mddev->raid_disks = info->raid_disks;
5008 /* don't set md_minor, it is determined by which /dev/md* was
5009 * openned
5010 */
5011 if (info->state & (1<<MD_SB_CLEAN))
5012 mddev->recovery_cp = MaxSector;
5013 else
5014 mddev->recovery_cp = 0;
5015 mddev->persistent = ! info->not_persistent;
5016 mddev->external = 0;
5017
5018 mddev->layout = info->layout;
5019 mddev->chunk_sectors = info->chunk_size >> 9;
5020
5021 mddev->max_disks = MD_SB_DISKS;
5022
5023 if (mddev->persistent)
5024 mddev->flags = 0;
5025 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5026
5027 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5028 mddev->bitmap_offset = 0;
5029
5030 mddev->reshape_position = MaxSector;
5031
5032 /*
5033 * Generate a 128 bit UUID
5034 */
5035 get_random_bytes(mddev->uuid, 16);
5036
5037 mddev->new_level = mddev->level;
5038 mddev->new_chunk_sectors = mddev->chunk_sectors;
5039 mddev->new_layout = mddev->layout;
5040 mddev->delta_disks = 0;
5041
5042 return 0;
5043 }
5044
5045 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5046 {
5047 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5048
5049 if (mddev->external_size)
5050 return;
5051
5052 mddev->array_sectors = array_sectors;
5053 }
5054 EXPORT_SYMBOL(md_set_array_sectors);
5055
5056 static int update_size(mddev_t *mddev, sector_t num_sectors)
5057 {
5058 mdk_rdev_t *rdev;
5059 int rv;
5060 int fit = (num_sectors == 0);
5061
5062 if (mddev->pers->resize == NULL)
5063 return -EINVAL;
5064 /* The "num_sectors" is the number of sectors of each device that
5065 * is used. This can only make sense for arrays with redundancy.
5066 * linear and raid0 always use whatever space is available. We can only
5067 * consider changing this number if no resync or reconstruction is
5068 * happening, and if the new size is acceptable. It must fit before the
5069 * sb_start or, if that is <data_offset, it must fit before the size
5070 * of each device. If num_sectors is zero, we find the largest size
5071 * that fits.
5072
5073 */
5074 if (mddev->sync_thread)
5075 return -EBUSY;
5076 if (mddev->bitmap)
5077 /* Sorry, cannot grow a bitmap yet, just remove it,
5078 * grow, and re-add.
5079 */
5080 return -EBUSY;
5081 list_for_each_entry(rdev, &mddev->disks, same_set) {
5082 sector_t avail = rdev->sectors;
5083
5084 if (fit && (num_sectors == 0 || num_sectors > avail))
5085 num_sectors = avail;
5086 if (avail < num_sectors)
5087 return -ENOSPC;
5088 }
5089 rv = mddev->pers->resize(mddev, num_sectors);
5090 if (!rv) {
5091 struct block_device *bdev;
5092
5093 bdev = bdget_disk(mddev->gendisk, 0);
5094 if (bdev) {
5095 mutex_lock(&bdev->bd_inode->i_mutex);
5096 i_size_write(bdev->bd_inode,
5097 (loff_t)mddev->array_sectors << 9);
5098 mutex_unlock(&bdev->bd_inode->i_mutex);
5099 bdput(bdev);
5100 }
5101 }
5102 return rv;
5103 }
5104
5105 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5106 {
5107 int rv;
5108 /* change the number of raid disks */
5109 if (mddev->pers->check_reshape == NULL)
5110 return -EINVAL;
5111 if (raid_disks <= 0 ||
5112 raid_disks >= mddev->max_disks)
5113 return -EINVAL;
5114 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5115 return -EBUSY;
5116 mddev->delta_disks = raid_disks - mddev->raid_disks;
5117
5118 rv = mddev->pers->check_reshape(mddev);
5119 return rv;
5120 }
5121
5122
5123 /*
5124 * update_array_info is used to change the configuration of an
5125 * on-line array.
5126 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5127 * fields in the info are checked against the array.
5128 * Any differences that cannot be handled will cause an error.
5129 * Normally, only one change can be managed at a time.
5130 */
5131 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5132 {
5133 int rv = 0;
5134 int cnt = 0;
5135 int state = 0;
5136
5137 /* calculate expected state,ignoring low bits */
5138 if (mddev->bitmap && mddev->bitmap_offset)
5139 state |= (1 << MD_SB_BITMAP_PRESENT);
5140
5141 if (mddev->major_version != info->major_version ||
5142 mddev->minor_version != info->minor_version ||
5143 /* mddev->patch_version != info->patch_version || */
5144 mddev->ctime != info->ctime ||
5145 mddev->level != info->level ||
5146 /* mddev->layout != info->layout || */
5147 !mddev->persistent != info->not_persistent||
5148 mddev->chunk_sectors != info->chunk_size >> 9 ||
5149 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5150 ((state^info->state) & 0xfffffe00)
5151 )
5152 return -EINVAL;
5153 /* Check there is only one change */
5154 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5155 cnt++;
5156 if (mddev->raid_disks != info->raid_disks)
5157 cnt++;
5158 if (mddev->layout != info->layout)
5159 cnt++;
5160 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5161 cnt++;
5162 if (cnt == 0)
5163 return 0;
5164 if (cnt > 1)
5165 return -EINVAL;
5166
5167 if (mddev->layout != info->layout) {
5168 /* Change layout
5169 * we don't need to do anything at the md level, the
5170 * personality will take care of it all.
5171 */
5172 if (mddev->pers->check_reshape == NULL)
5173 return -EINVAL;
5174 else {
5175 mddev->new_layout = info->layout;
5176 rv = mddev->pers->check_reshape(mddev);
5177 if (rv)
5178 mddev->new_layout = mddev->layout;
5179 return rv;
5180 }
5181 }
5182 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5183 rv = update_size(mddev, (sector_t)info->size * 2);
5184
5185 if (mddev->raid_disks != info->raid_disks)
5186 rv = update_raid_disks(mddev, info->raid_disks);
5187
5188 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5189 if (mddev->pers->quiesce == NULL)
5190 return -EINVAL;
5191 if (mddev->recovery || mddev->sync_thread)
5192 return -EBUSY;
5193 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5194 /* add the bitmap */
5195 if (mddev->bitmap)
5196 return -EEXIST;
5197 if (mddev->default_bitmap_offset == 0)
5198 return -EINVAL;
5199 mddev->bitmap_offset = mddev->default_bitmap_offset;
5200 mddev->pers->quiesce(mddev, 1);
5201 rv = bitmap_create(mddev);
5202 if (rv)
5203 bitmap_destroy(mddev);
5204 mddev->pers->quiesce(mddev, 0);
5205 } else {
5206 /* remove the bitmap */
5207 if (!mddev->bitmap)
5208 return -ENOENT;
5209 if (mddev->bitmap->file)
5210 return -EINVAL;
5211 mddev->pers->quiesce(mddev, 1);
5212 bitmap_destroy(mddev);
5213 mddev->pers->quiesce(mddev, 0);
5214 mddev->bitmap_offset = 0;
5215 }
5216 }
5217 md_update_sb(mddev, 1);
5218 return rv;
5219 }
5220
5221 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5222 {
5223 mdk_rdev_t *rdev;
5224
5225 if (mddev->pers == NULL)
5226 return -ENODEV;
5227
5228 rdev = find_rdev(mddev, dev);
5229 if (!rdev)
5230 return -ENODEV;
5231
5232 md_error(mddev, rdev);
5233 return 0;
5234 }
5235
5236 /*
5237 * We have a problem here : there is no easy way to give a CHS
5238 * virtual geometry. We currently pretend that we have a 2 heads
5239 * 4 sectors (with a BIG number of cylinders...). This drives
5240 * dosfs just mad... ;-)
5241 */
5242 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5243 {
5244 mddev_t *mddev = bdev->bd_disk->private_data;
5245
5246 geo->heads = 2;
5247 geo->sectors = 4;
5248 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5249 return 0;
5250 }
5251
5252 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5253 unsigned int cmd, unsigned long arg)
5254 {
5255 int err = 0;
5256 void __user *argp = (void __user *)arg;
5257 mddev_t *mddev = NULL;
5258
5259 if (!capable(CAP_SYS_ADMIN))
5260 return -EACCES;
5261
5262 /*
5263 * Commands dealing with the RAID driver but not any
5264 * particular array:
5265 */
5266 switch (cmd)
5267 {
5268 case RAID_VERSION:
5269 err = get_version(argp);
5270 goto done;
5271
5272 case PRINT_RAID_DEBUG:
5273 err = 0;
5274 md_print_devices();
5275 goto done;
5276
5277 #ifndef MODULE
5278 case RAID_AUTORUN:
5279 err = 0;
5280 autostart_arrays(arg);
5281 goto done;
5282 #endif
5283 default:;
5284 }
5285
5286 /*
5287 * Commands creating/starting a new array:
5288 */
5289
5290 mddev = bdev->bd_disk->private_data;
5291
5292 if (!mddev) {
5293 BUG();
5294 goto abort;
5295 }
5296
5297 err = mddev_lock(mddev);
5298 if (err) {
5299 printk(KERN_INFO
5300 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5301 err, cmd);
5302 goto abort;
5303 }
5304
5305 switch (cmd)
5306 {
5307 case SET_ARRAY_INFO:
5308 {
5309 mdu_array_info_t info;
5310 if (!arg)
5311 memset(&info, 0, sizeof(info));
5312 else if (copy_from_user(&info, argp, sizeof(info))) {
5313 err = -EFAULT;
5314 goto abort_unlock;
5315 }
5316 if (mddev->pers) {
5317 err = update_array_info(mddev, &info);
5318 if (err) {
5319 printk(KERN_WARNING "md: couldn't update"
5320 " array info. %d\n", err);
5321 goto abort_unlock;
5322 }
5323 goto done_unlock;
5324 }
5325 if (!list_empty(&mddev->disks)) {
5326 printk(KERN_WARNING
5327 "md: array %s already has disks!\n",
5328 mdname(mddev));
5329 err = -EBUSY;
5330 goto abort_unlock;
5331 }
5332 if (mddev->raid_disks) {
5333 printk(KERN_WARNING
5334 "md: array %s already initialised!\n",
5335 mdname(mddev));
5336 err = -EBUSY;
5337 goto abort_unlock;
5338 }
5339 err = set_array_info(mddev, &info);
5340 if (err) {
5341 printk(KERN_WARNING "md: couldn't set"
5342 " array info. %d\n", err);
5343 goto abort_unlock;
5344 }
5345 }
5346 goto done_unlock;
5347
5348 default:;
5349 }
5350
5351 /*
5352 * Commands querying/configuring an existing array:
5353 */
5354 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5355 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5356 if ((!mddev->raid_disks && !mddev->external)
5357 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5358 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5359 && cmd != GET_BITMAP_FILE) {
5360 err = -ENODEV;
5361 goto abort_unlock;
5362 }
5363
5364 /*
5365 * Commands even a read-only array can execute:
5366 */
5367 switch (cmd)
5368 {
5369 case GET_ARRAY_INFO:
5370 err = get_array_info(mddev, argp);
5371 goto done_unlock;
5372
5373 case GET_BITMAP_FILE:
5374 err = get_bitmap_file(mddev, argp);
5375 goto done_unlock;
5376
5377 case GET_DISK_INFO:
5378 err = get_disk_info(mddev, argp);
5379 goto done_unlock;
5380
5381 case RESTART_ARRAY_RW:
5382 err = restart_array(mddev);
5383 goto done_unlock;
5384
5385 case STOP_ARRAY:
5386 err = do_md_stop(mddev, 0, 1);
5387 goto done_unlock;
5388
5389 case STOP_ARRAY_RO:
5390 err = do_md_stop(mddev, 1, 1);
5391 goto done_unlock;
5392
5393 }
5394
5395 /*
5396 * The remaining ioctls are changing the state of the
5397 * superblock, so we do not allow them on read-only arrays.
5398 * However non-MD ioctls (e.g. get-size) will still come through
5399 * here and hit the 'default' below, so only disallow
5400 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5401 */
5402 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5403 if (mddev->ro == 2) {
5404 mddev->ro = 0;
5405 sysfs_notify_dirent(mddev->sysfs_state);
5406 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5407 md_wakeup_thread(mddev->thread);
5408 } else {
5409 err = -EROFS;
5410 goto abort_unlock;
5411 }
5412 }
5413
5414 switch (cmd)
5415 {
5416 case ADD_NEW_DISK:
5417 {
5418 mdu_disk_info_t info;
5419 if (copy_from_user(&info, argp, sizeof(info)))
5420 err = -EFAULT;
5421 else
5422 err = add_new_disk(mddev, &info);
5423 goto done_unlock;
5424 }
5425
5426 case HOT_REMOVE_DISK:
5427 err = hot_remove_disk(mddev, new_decode_dev(arg));
5428 goto done_unlock;
5429
5430 case HOT_ADD_DISK:
5431 err = hot_add_disk(mddev, new_decode_dev(arg));
5432 goto done_unlock;
5433
5434 case SET_DISK_FAULTY:
5435 err = set_disk_faulty(mddev, new_decode_dev(arg));
5436 goto done_unlock;
5437
5438 case RUN_ARRAY:
5439 err = do_md_run(mddev);
5440 goto done_unlock;
5441
5442 case SET_BITMAP_FILE:
5443 err = set_bitmap_file(mddev, (int)arg);
5444 goto done_unlock;
5445
5446 default:
5447 err = -EINVAL;
5448 goto abort_unlock;
5449 }
5450
5451 done_unlock:
5452 abort_unlock:
5453 if (mddev->hold_active == UNTIL_IOCTL &&
5454 err != -EINVAL)
5455 mddev->hold_active = 0;
5456 mddev_unlock(mddev);
5457
5458 return err;
5459 done:
5460 if (err)
5461 MD_BUG();
5462 abort:
5463 return err;
5464 }
5465
5466 static int md_open(struct block_device *bdev, fmode_t mode)
5467 {
5468 /*
5469 * Succeed if we can lock the mddev, which confirms that
5470 * it isn't being stopped right now.
5471 */
5472 mddev_t *mddev = mddev_find(bdev->bd_dev);
5473 int err;
5474
5475 if (mddev->gendisk != bdev->bd_disk) {
5476 /* we are racing with mddev_put which is discarding this
5477 * bd_disk.
5478 */
5479 mddev_put(mddev);
5480 /* Wait until bdev->bd_disk is definitely gone */
5481 flush_scheduled_work();
5482 /* Then retry the open from the top */
5483 return -ERESTARTSYS;
5484 }
5485 BUG_ON(mddev != bdev->bd_disk->private_data);
5486
5487 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5488 goto out;
5489
5490 err = 0;
5491 atomic_inc(&mddev->openers);
5492 mddev_unlock(mddev);
5493
5494 check_disk_change(bdev);
5495 out:
5496 return err;
5497 }
5498
5499 static int md_release(struct gendisk *disk, fmode_t mode)
5500 {
5501 mddev_t *mddev = disk->private_data;
5502
5503 BUG_ON(!mddev);
5504 atomic_dec(&mddev->openers);
5505 mddev_put(mddev);
5506
5507 return 0;
5508 }
5509
5510 static int md_media_changed(struct gendisk *disk)
5511 {
5512 mddev_t *mddev = disk->private_data;
5513
5514 return mddev->changed;
5515 }
5516
5517 static int md_revalidate(struct gendisk *disk)
5518 {
5519 mddev_t *mddev = disk->private_data;
5520
5521 mddev->changed = 0;
5522 return 0;
5523 }
5524 static struct block_device_operations md_fops =
5525 {
5526 .owner = THIS_MODULE,
5527 .open = md_open,
5528 .release = md_release,
5529 .ioctl = md_ioctl,
5530 .getgeo = md_getgeo,
5531 .media_changed = md_media_changed,
5532 .revalidate_disk= md_revalidate,
5533 };
5534
5535 static int md_thread(void * arg)
5536 {
5537 mdk_thread_t *thread = arg;
5538
5539 /*
5540 * md_thread is a 'system-thread', it's priority should be very
5541 * high. We avoid resource deadlocks individually in each
5542 * raid personality. (RAID5 does preallocation) We also use RR and
5543 * the very same RT priority as kswapd, thus we will never get
5544 * into a priority inversion deadlock.
5545 *
5546 * we definitely have to have equal or higher priority than
5547 * bdflush, otherwise bdflush will deadlock if there are too
5548 * many dirty RAID5 blocks.
5549 */
5550
5551 allow_signal(SIGKILL);
5552 while (!kthread_should_stop()) {
5553
5554 /* We need to wait INTERRUPTIBLE so that
5555 * we don't add to the load-average.
5556 * That means we need to be sure no signals are
5557 * pending
5558 */
5559 if (signal_pending(current))
5560 flush_signals(current);
5561
5562 wait_event_interruptible_timeout
5563 (thread->wqueue,
5564 test_bit(THREAD_WAKEUP, &thread->flags)
5565 || kthread_should_stop(),
5566 thread->timeout);
5567
5568 clear_bit(THREAD_WAKEUP, &thread->flags);
5569
5570 thread->run(thread->mddev);
5571 }
5572
5573 return 0;
5574 }
5575
5576 void md_wakeup_thread(mdk_thread_t *thread)
5577 {
5578 if (thread) {
5579 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5580 set_bit(THREAD_WAKEUP, &thread->flags);
5581 wake_up(&thread->wqueue);
5582 }
5583 }
5584
5585 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5586 const char *name)
5587 {
5588 mdk_thread_t *thread;
5589
5590 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5591 if (!thread)
5592 return NULL;
5593
5594 init_waitqueue_head(&thread->wqueue);
5595
5596 thread->run = run;
5597 thread->mddev = mddev;
5598 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5599 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5600 if (IS_ERR(thread->tsk)) {
5601 kfree(thread);
5602 return NULL;
5603 }
5604 return thread;
5605 }
5606
5607 void md_unregister_thread(mdk_thread_t *thread)
5608 {
5609 if (!thread)
5610 return;
5611 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5612
5613 kthread_stop(thread->tsk);
5614 kfree(thread);
5615 }
5616
5617 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5618 {
5619 if (!mddev) {
5620 MD_BUG();
5621 return;
5622 }
5623
5624 if (!rdev || test_bit(Faulty, &rdev->flags))
5625 return;
5626
5627 if (mddev->external)
5628 set_bit(Blocked, &rdev->flags);
5629 /*
5630 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5631 mdname(mddev),
5632 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5633 __builtin_return_address(0),__builtin_return_address(1),
5634 __builtin_return_address(2),__builtin_return_address(3));
5635 */
5636 if (!mddev->pers)
5637 return;
5638 if (!mddev->pers->error_handler)
5639 return;
5640 mddev->pers->error_handler(mddev,rdev);
5641 if (mddev->degraded)
5642 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5643 set_bit(StateChanged, &rdev->flags);
5644 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5645 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5646 md_wakeup_thread(mddev->thread);
5647 md_new_event_inintr(mddev);
5648 }
5649
5650 /* seq_file implementation /proc/mdstat */
5651
5652 static void status_unused(struct seq_file *seq)
5653 {
5654 int i = 0;
5655 mdk_rdev_t *rdev;
5656
5657 seq_printf(seq, "unused devices: ");
5658
5659 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5660 char b[BDEVNAME_SIZE];
5661 i++;
5662 seq_printf(seq, "%s ",
5663 bdevname(rdev->bdev,b));
5664 }
5665 if (!i)
5666 seq_printf(seq, "<none>");
5667
5668 seq_printf(seq, "\n");
5669 }
5670
5671
5672 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5673 {
5674 sector_t max_sectors, resync, res;
5675 unsigned long dt, db;
5676 sector_t rt;
5677 int scale;
5678 unsigned int per_milli;
5679
5680 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5681
5682 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5683 max_sectors = mddev->resync_max_sectors;
5684 else
5685 max_sectors = mddev->dev_sectors;
5686
5687 /*
5688 * Should not happen.
5689 */
5690 if (!max_sectors) {
5691 MD_BUG();
5692 return;
5693 }
5694 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5695 * in a sector_t, and (max_sectors>>scale) will fit in a
5696 * u32, as those are the requirements for sector_div.
5697 * Thus 'scale' must be at least 10
5698 */
5699 scale = 10;
5700 if (sizeof(sector_t) > sizeof(unsigned long)) {
5701 while ( max_sectors/2 > (1ULL<<(scale+32)))
5702 scale++;
5703 }
5704 res = (resync>>scale)*1000;
5705 sector_div(res, (u32)((max_sectors>>scale)+1));
5706
5707 per_milli = res;
5708 {
5709 int i, x = per_milli/50, y = 20-x;
5710 seq_printf(seq, "[");
5711 for (i = 0; i < x; i++)
5712 seq_printf(seq, "=");
5713 seq_printf(seq, ">");
5714 for (i = 0; i < y; i++)
5715 seq_printf(seq, ".");
5716 seq_printf(seq, "] ");
5717 }
5718 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5719 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5720 "reshape" :
5721 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5722 "check" :
5723 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5724 "resync" : "recovery"))),
5725 per_milli/10, per_milli % 10,
5726 (unsigned long long) resync/2,
5727 (unsigned long long) max_sectors/2);
5728
5729 /*
5730 * dt: time from mark until now
5731 * db: blocks written from mark until now
5732 * rt: remaining time
5733 *
5734 * rt is a sector_t, so could be 32bit or 64bit.
5735 * So we divide before multiply in case it is 32bit and close
5736 * to the limit.
5737 * We scale the divisor (db) by 32 to avoid loosing precision
5738 * near the end of resync when the number of remaining sectors
5739 * is close to 'db'.
5740 * We then divide rt by 32 after multiplying by db to compensate.
5741 * The '+1' avoids division by zero if db is very small.
5742 */
5743 dt = ((jiffies - mddev->resync_mark) / HZ);
5744 if (!dt) dt++;
5745 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5746 - mddev->resync_mark_cnt;
5747
5748 rt = max_sectors - resync; /* number of remaining sectors */
5749 sector_div(rt, db/32+1);
5750 rt *= dt;
5751 rt >>= 5;
5752
5753 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5754 ((unsigned long)rt % 60)/6);
5755
5756 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5757 }
5758
5759 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5760 {
5761 struct list_head *tmp;
5762 loff_t l = *pos;
5763 mddev_t *mddev;
5764
5765 if (l >= 0x10000)
5766 return NULL;
5767 if (!l--)
5768 /* header */
5769 return (void*)1;
5770
5771 spin_lock(&all_mddevs_lock);
5772 list_for_each(tmp,&all_mddevs)
5773 if (!l--) {
5774 mddev = list_entry(tmp, mddev_t, all_mddevs);
5775 mddev_get(mddev);
5776 spin_unlock(&all_mddevs_lock);
5777 return mddev;
5778 }
5779 spin_unlock(&all_mddevs_lock);
5780 if (!l--)
5781 return (void*)2;/* tail */
5782 return NULL;
5783 }
5784
5785 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5786 {
5787 struct list_head *tmp;
5788 mddev_t *next_mddev, *mddev = v;
5789
5790 ++*pos;
5791 if (v == (void*)2)
5792 return NULL;
5793
5794 spin_lock(&all_mddevs_lock);
5795 if (v == (void*)1)
5796 tmp = all_mddevs.next;
5797 else
5798 tmp = mddev->all_mddevs.next;
5799 if (tmp != &all_mddevs)
5800 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5801 else {
5802 next_mddev = (void*)2;
5803 *pos = 0x10000;
5804 }
5805 spin_unlock(&all_mddevs_lock);
5806
5807 if (v != (void*)1)
5808 mddev_put(mddev);
5809 return next_mddev;
5810
5811 }
5812
5813 static void md_seq_stop(struct seq_file *seq, void *v)
5814 {
5815 mddev_t *mddev = v;
5816
5817 if (mddev && v != (void*)1 && v != (void*)2)
5818 mddev_put(mddev);
5819 }
5820
5821 struct mdstat_info {
5822 int event;
5823 };
5824
5825 static int md_seq_show(struct seq_file *seq, void *v)
5826 {
5827 mddev_t *mddev = v;
5828 sector_t sectors;
5829 mdk_rdev_t *rdev;
5830 struct mdstat_info *mi = seq->private;
5831 struct bitmap *bitmap;
5832
5833 if (v == (void*)1) {
5834 struct mdk_personality *pers;
5835 seq_printf(seq, "Personalities : ");
5836 spin_lock(&pers_lock);
5837 list_for_each_entry(pers, &pers_list, list)
5838 seq_printf(seq, "[%s] ", pers->name);
5839
5840 spin_unlock(&pers_lock);
5841 seq_printf(seq, "\n");
5842 mi->event = atomic_read(&md_event_count);
5843 return 0;
5844 }
5845 if (v == (void*)2) {
5846 status_unused(seq);
5847 return 0;
5848 }
5849
5850 if (mddev_lock(mddev) < 0)
5851 return -EINTR;
5852
5853 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5854 seq_printf(seq, "%s : %sactive", mdname(mddev),
5855 mddev->pers ? "" : "in");
5856 if (mddev->pers) {
5857 if (mddev->ro==1)
5858 seq_printf(seq, " (read-only)");
5859 if (mddev->ro==2)
5860 seq_printf(seq, " (auto-read-only)");
5861 seq_printf(seq, " %s", mddev->pers->name);
5862 }
5863
5864 sectors = 0;
5865 list_for_each_entry(rdev, &mddev->disks, same_set) {
5866 char b[BDEVNAME_SIZE];
5867 seq_printf(seq, " %s[%d]",
5868 bdevname(rdev->bdev,b), rdev->desc_nr);
5869 if (test_bit(WriteMostly, &rdev->flags))
5870 seq_printf(seq, "(W)");
5871 if (test_bit(Faulty, &rdev->flags)) {
5872 seq_printf(seq, "(F)");
5873 continue;
5874 } else if (rdev->raid_disk < 0)
5875 seq_printf(seq, "(S)"); /* spare */
5876 sectors += rdev->sectors;
5877 }
5878
5879 if (!list_empty(&mddev->disks)) {
5880 if (mddev->pers)
5881 seq_printf(seq, "\n %llu blocks",
5882 (unsigned long long)
5883 mddev->array_sectors / 2);
5884 else
5885 seq_printf(seq, "\n %llu blocks",
5886 (unsigned long long)sectors / 2);
5887 }
5888 if (mddev->persistent) {
5889 if (mddev->major_version != 0 ||
5890 mddev->minor_version != 90) {
5891 seq_printf(seq," super %d.%d",
5892 mddev->major_version,
5893 mddev->minor_version);
5894 }
5895 } else if (mddev->external)
5896 seq_printf(seq, " super external:%s",
5897 mddev->metadata_type);
5898 else
5899 seq_printf(seq, " super non-persistent");
5900
5901 if (mddev->pers) {
5902 mddev->pers->status(seq, mddev);
5903 seq_printf(seq, "\n ");
5904 if (mddev->pers->sync_request) {
5905 if (mddev->curr_resync > 2) {
5906 status_resync(seq, mddev);
5907 seq_printf(seq, "\n ");
5908 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5909 seq_printf(seq, "\tresync=DELAYED\n ");
5910 else if (mddev->recovery_cp < MaxSector)
5911 seq_printf(seq, "\tresync=PENDING\n ");
5912 }
5913 } else
5914 seq_printf(seq, "\n ");
5915
5916 if ((bitmap = mddev->bitmap)) {
5917 unsigned long chunk_kb;
5918 unsigned long flags;
5919 spin_lock_irqsave(&bitmap->lock, flags);
5920 chunk_kb = bitmap->chunksize >> 10;
5921 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5922 "%lu%s chunk",
5923 bitmap->pages - bitmap->missing_pages,
5924 bitmap->pages,
5925 (bitmap->pages - bitmap->missing_pages)
5926 << (PAGE_SHIFT - 10),
5927 chunk_kb ? chunk_kb : bitmap->chunksize,
5928 chunk_kb ? "KB" : "B");
5929 if (bitmap->file) {
5930 seq_printf(seq, ", file: ");
5931 seq_path(seq, &bitmap->file->f_path, " \t\n");
5932 }
5933
5934 seq_printf(seq, "\n");
5935 spin_unlock_irqrestore(&bitmap->lock, flags);
5936 }
5937
5938 seq_printf(seq, "\n");
5939 }
5940 mddev_unlock(mddev);
5941
5942 return 0;
5943 }
5944
5945 static const struct seq_operations md_seq_ops = {
5946 .start = md_seq_start,
5947 .next = md_seq_next,
5948 .stop = md_seq_stop,
5949 .show = md_seq_show,
5950 };
5951
5952 static int md_seq_open(struct inode *inode, struct file *file)
5953 {
5954 int error;
5955 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5956 if (mi == NULL)
5957 return -ENOMEM;
5958
5959 error = seq_open(file, &md_seq_ops);
5960 if (error)
5961 kfree(mi);
5962 else {
5963 struct seq_file *p = file->private_data;
5964 p->private = mi;
5965 mi->event = atomic_read(&md_event_count);
5966 }
5967 return error;
5968 }
5969
5970 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5971 {
5972 struct seq_file *m = filp->private_data;
5973 struct mdstat_info *mi = m->private;
5974 int mask;
5975
5976 poll_wait(filp, &md_event_waiters, wait);
5977
5978 /* always allow read */
5979 mask = POLLIN | POLLRDNORM;
5980
5981 if (mi->event != atomic_read(&md_event_count))
5982 mask |= POLLERR | POLLPRI;
5983 return mask;
5984 }
5985
5986 static const struct file_operations md_seq_fops = {
5987 .owner = THIS_MODULE,
5988 .open = md_seq_open,
5989 .read = seq_read,
5990 .llseek = seq_lseek,
5991 .release = seq_release_private,
5992 .poll = mdstat_poll,
5993 };
5994
5995 int register_md_personality(struct mdk_personality *p)
5996 {
5997 spin_lock(&pers_lock);
5998 list_add_tail(&p->list, &pers_list);
5999 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6000 spin_unlock(&pers_lock);
6001 return 0;
6002 }
6003
6004 int unregister_md_personality(struct mdk_personality *p)
6005 {
6006 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6007 spin_lock(&pers_lock);
6008 list_del_init(&p->list);
6009 spin_unlock(&pers_lock);
6010 return 0;
6011 }
6012
6013 static int is_mddev_idle(mddev_t *mddev, int init)
6014 {
6015 mdk_rdev_t * rdev;
6016 int idle;
6017 int curr_events;
6018
6019 idle = 1;
6020 rcu_read_lock();
6021 rdev_for_each_rcu(rdev, mddev) {
6022 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6023 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6024 (int)part_stat_read(&disk->part0, sectors[1]) -
6025 atomic_read(&disk->sync_io);
6026 /* sync IO will cause sync_io to increase before the disk_stats
6027 * as sync_io is counted when a request starts, and
6028 * disk_stats is counted when it completes.
6029 * So resync activity will cause curr_events to be smaller than
6030 * when there was no such activity.
6031 * non-sync IO will cause disk_stat to increase without
6032 * increasing sync_io so curr_events will (eventually)
6033 * be larger than it was before. Once it becomes
6034 * substantially larger, the test below will cause
6035 * the array to appear non-idle, and resync will slow
6036 * down.
6037 * If there is a lot of outstanding resync activity when
6038 * we set last_event to curr_events, then all that activity
6039 * completing might cause the array to appear non-idle
6040 * and resync will be slowed down even though there might
6041 * not have been non-resync activity. This will only
6042 * happen once though. 'last_events' will soon reflect
6043 * the state where there is little or no outstanding
6044 * resync requests, and further resync activity will
6045 * always make curr_events less than last_events.
6046 *
6047 */
6048 if (init || curr_events - rdev->last_events > 64) {
6049 rdev->last_events = curr_events;
6050 idle = 0;
6051 }
6052 }
6053 rcu_read_unlock();
6054 return idle;
6055 }
6056
6057 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6058 {
6059 /* another "blocks" (512byte) blocks have been synced */
6060 atomic_sub(blocks, &mddev->recovery_active);
6061 wake_up(&mddev->recovery_wait);
6062 if (!ok) {
6063 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6064 md_wakeup_thread(mddev->thread);
6065 // stop recovery, signal do_sync ....
6066 }
6067 }
6068
6069
6070 /* md_write_start(mddev, bi)
6071 * If we need to update some array metadata (e.g. 'active' flag
6072 * in superblock) before writing, schedule a superblock update
6073 * and wait for it to complete.
6074 */
6075 void md_write_start(mddev_t *mddev, struct bio *bi)
6076 {
6077 int did_change = 0;
6078 if (bio_data_dir(bi) != WRITE)
6079 return;
6080
6081 BUG_ON(mddev->ro == 1);
6082 if (mddev->ro == 2) {
6083 /* need to switch to read/write */
6084 mddev->ro = 0;
6085 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6086 md_wakeup_thread(mddev->thread);
6087 md_wakeup_thread(mddev->sync_thread);
6088 did_change = 1;
6089 }
6090 atomic_inc(&mddev->writes_pending);
6091 if (mddev->safemode == 1)
6092 mddev->safemode = 0;
6093 if (mddev->in_sync) {
6094 spin_lock_irq(&mddev->write_lock);
6095 if (mddev->in_sync) {
6096 mddev->in_sync = 0;
6097 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6098 md_wakeup_thread(mddev->thread);
6099 did_change = 1;
6100 }
6101 spin_unlock_irq(&mddev->write_lock);
6102 }
6103 if (did_change)
6104 sysfs_notify_dirent(mddev->sysfs_state);
6105 wait_event(mddev->sb_wait,
6106 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6107 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6108 }
6109
6110 void md_write_end(mddev_t *mddev)
6111 {
6112 if (atomic_dec_and_test(&mddev->writes_pending)) {
6113 if (mddev->safemode == 2)
6114 md_wakeup_thread(mddev->thread);
6115 else if (mddev->safemode_delay)
6116 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6117 }
6118 }
6119
6120 /* md_allow_write(mddev)
6121 * Calling this ensures that the array is marked 'active' so that writes
6122 * may proceed without blocking. It is important to call this before
6123 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6124 * Must be called with mddev_lock held.
6125 *
6126 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6127 * is dropped, so return -EAGAIN after notifying userspace.
6128 */
6129 int md_allow_write(mddev_t *mddev)
6130 {
6131 if (!mddev->pers)
6132 return 0;
6133 if (mddev->ro)
6134 return 0;
6135 if (!mddev->pers->sync_request)
6136 return 0;
6137
6138 spin_lock_irq(&mddev->write_lock);
6139 if (mddev->in_sync) {
6140 mddev->in_sync = 0;
6141 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6142 if (mddev->safemode_delay &&
6143 mddev->safemode == 0)
6144 mddev->safemode = 1;
6145 spin_unlock_irq(&mddev->write_lock);
6146 md_update_sb(mddev, 0);
6147 sysfs_notify_dirent(mddev->sysfs_state);
6148 } else
6149 spin_unlock_irq(&mddev->write_lock);
6150
6151 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6152 return -EAGAIN;
6153 else
6154 return 0;
6155 }
6156 EXPORT_SYMBOL_GPL(md_allow_write);
6157
6158 #define SYNC_MARKS 10
6159 #define SYNC_MARK_STEP (3*HZ)
6160 void md_do_sync(mddev_t *mddev)
6161 {
6162 mddev_t *mddev2;
6163 unsigned int currspeed = 0,
6164 window;
6165 sector_t max_sectors,j, io_sectors;
6166 unsigned long mark[SYNC_MARKS];
6167 sector_t mark_cnt[SYNC_MARKS];
6168 int last_mark,m;
6169 struct list_head *tmp;
6170 sector_t last_check;
6171 int skipped = 0;
6172 mdk_rdev_t *rdev;
6173 char *desc;
6174
6175 /* just incase thread restarts... */
6176 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6177 return;
6178 if (mddev->ro) /* never try to sync a read-only array */
6179 return;
6180
6181 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6182 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6183 desc = "data-check";
6184 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6185 desc = "requested-resync";
6186 else
6187 desc = "resync";
6188 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6189 desc = "reshape";
6190 else
6191 desc = "recovery";
6192
6193 /* we overload curr_resync somewhat here.
6194 * 0 == not engaged in resync at all
6195 * 2 == checking that there is no conflict with another sync
6196 * 1 == like 2, but have yielded to allow conflicting resync to
6197 * commense
6198 * other == active in resync - this many blocks
6199 *
6200 * Before starting a resync we must have set curr_resync to
6201 * 2, and then checked that every "conflicting" array has curr_resync
6202 * less than ours. When we find one that is the same or higher
6203 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6204 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6205 * This will mean we have to start checking from the beginning again.
6206 *
6207 */
6208
6209 do {
6210 mddev->curr_resync = 2;
6211
6212 try_again:
6213 if (kthread_should_stop()) {
6214 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6215 goto skip;
6216 }
6217 for_each_mddev(mddev2, tmp) {
6218 if (mddev2 == mddev)
6219 continue;
6220 if (!mddev->parallel_resync
6221 && mddev2->curr_resync
6222 && match_mddev_units(mddev, mddev2)) {
6223 DEFINE_WAIT(wq);
6224 if (mddev < mddev2 && mddev->curr_resync == 2) {
6225 /* arbitrarily yield */
6226 mddev->curr_resync = 1;
6227 wake_up(&resync_wait);
6228 }
6229 if (mddev > mddev2 && mddev->curr_resync == 1)
6230 /* no need to wait here, we can wait the next
6231 * time 'round when curr_resync == 2
6232 */
6233 continue;
6234 /* We need to wait 'interruptible' so as not to
6235 * contribute to the load average, and not to
6236 * be caught by 'softlockup'
6237 */
6238 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6239 if (!kthread_should_stop() &&
6240 mddev2->curr_resync >= mddev->curr_resync) {
6241 printk(KERN_INFO "md: delaying %s of %s"
6242 " until %s has finished (they"
6243 " share one or more physical units)\n",
6244 desc, mdname(mddev), mdname(mddev2));
6245 mddev_put(mddev2);
6246 if (signal_pending(current))
6247 flush_signals(current);
6248 schedule();
6249 finish_wait(&resync_wait, &wq);
6250 goto try_again;
6251 }
6252 finish_wait(&resync_wait, &wq);
6253 }
6254 }
6255 } while (mddev->curr_resync < 2);
6256
6257 j = 0;
6258 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6259 /* resync follows the size requested by the personality,
6260 * which defaults to physical size, but can be virtual size
6261 */
6262 max_sectors = mddev->resync_max_sectors;
6263 mddev->resync_mismatches = 0;
6264 /* we don't use the checkpoint if there's a bitmap */
6265 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6266 j = mddev->resync_min;
6267 else if (!mddev->bitmap)
6268 j = mddev->recovery_cp;
6269
6270 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6271 max_sectors = mddev->dev_sectors;
6272 else {
6273 /* recovery follows the physical size of devices */
6274 max_sectors = mddev->dev_sectors;
6275 j = MaxSector;
6276 list_for_each_entry(rdev, &mddev->disks, same_set)
6277 if (rdev->raid_disk >= 0 &&
6278 !test_bit(Faulty, &rdev->flags) &&
6279 !test_bit(In_sync, &rdev->flags) &&
6280 rdev->recovery_offset < j)
6281 j = rdev->recovery_offset;
6282 }
6283
6284 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6285 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6286 " %d KB/sec/disk.\n", speed_min(mddev));
6287 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6288 "(but not more than %d KB/sec) for %s.\n",
6289 speed_max(mddev), desc);
6290
6291 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6292
6293 io_sectors = 0;
6294 for (m = 0; m < SYNC_MARKS; m++) {
6295 mark[m] = jiffies;
6296 mark_cnt[m] = io_sectors;
6297 }
6298 last_mark = 0;
6299 mddev->resync_mark = mark[last_mark];
6300 mddev->resync_mark_cnt = mark_cnt[last_mark];
6301
6302 /*
6303 * Tune reconstruction:
6304 */
6305 window = 32*(PAGE_SIZE/512);
6306 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6307 window/2,(unsigned long long) max_sectors/2);
6308
6309 atomic_set(&mddev->recovery_active, 0);
6310 last_check = 0;
6311
6312 if (j>2) {
6313 printk(KERN_INFO
6314 "md: resuming %s of %s from checkpoint.\n",
6315 desc, mdname(mddev));
6316 mddev->curr_resync = j;
6317 }
6318
6319 while (j < max_sectors) {
6320 sector_t sectors;
6321
6322 skipped = 0;
6323
6324 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6325 ((mddev->curr_resync > mddev->curr_resync_completed &&
6326 (mddev->curr_resync - mddev->curr_resync_completed)
6327 > (max_sectors >> 4)) ||
6328 (j - mddev->curr_resync_completed)*2
6329 >= mddev->resync_max - mddev->curr_resync_completed
6330 )) {
6331 /* time to update curr_resync_completed */
6332 blk_unplug(mddev->queue);
6333 wait_event(mddev->recovery_wait,
6334 atomic_read(&mddev->recovery_active) == 0);
6335 mddev->curr_resync_completed =
6336 mddev->curr_resync;
6337 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6338 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6339 }
6340
6341 while (j >= mddev->resync_max && !kthread_should_stop()) {
6342 /* As this condition is controlled by user-space,
6343 * we can block indefinitely, so use '_interruptible'
6344 * to avoid triggering warnings.
6345 */
6346 flush_signals(current); /* just in case */
6347 wait_event_interruptible(mddev->recovery_wait,
6348 mddev->resync_max > j
6349 || kthread_should_stop());
6350 }
6351
6352 if (kthread_should_stop())
6353 goto interrupted;
6354
6355 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6356 currspeed < speed_min(mddev));
6357 if (sectors == 0) {
6358 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6359 goto out;
6360 }
6361
6362 if (!skipped) { /* actual IO requested */
6363 io_sectors += sectors;
6364 atomic_add(sectors, &mddev->recovery_active);
6365 }
6366
6367 j += sectors;
6368 if (j>1) mddev->curr_resync = j;
6369 mddev->curr_mark_cnt = io_sectors;
6370 if (last_check == 0)
6371 /* this is the earliers that rebuilt will be
6372 * visible in /proc/mdstat
6373 */
6374 md_new_event(mddev);
6375
6376 if (last_check + window > io_sectors || j == max_sectors)
6377 continue;
6378
6379 last_check = io_sectors;
6380
6381 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6382 break;
6383
6384 repeat:
6385 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6386 /* step marks */
6387 int next = (last_mark+1) % SYNC_MARKS;
6388
6389 mddev->resync_mark = mark[next];
6390 mddev->resync_mark_cnt = mark_cnt[next];
6391 mark[next] = jiffies;
6392 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6393 last_mark = next;
6394 }
6395
6396
6397 if (kthread_should_stop())
6398 goto interrupted;
6399
6400
6401 /*
6402 * this loop exits only if either when we are slower than
6403 * the 'hard' speed limit, or the system was IO-idle for
6404 * a jiffy.
6405 * the system might be non-idle CPU-wise, but we only care
6406 * about not overloading the IO subsystem. (things like an
6407 * e2fsck being done on the RAID array should execute fast)
6408 */
6409 blk_unplug(mddev->queue);
6410 cond_resched();
6411
6412 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6413 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6414
6415 if (currspeed > speed_min(mddev)) {
6416 if ((currspeed > speed_max(mddev)) ||
6417 !is_mddev_idle(mddev, 0)) {
6418 msleep(500);
6419 goto repeat;
6420 }
6421 }
6422 }
6423 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6424 /*
6425 * this also signals 'finished resyncing' to md_stop
6426 */
6427 out:
6428 blk_unplug(mddev->queue);
6429
6430 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6431
6432 /* tell personality that we are finished */
6433 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6434
6435 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6436 mddev->curr_resync > 2) {
6437 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6438 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6439 if (mddev->curr_resync >= mddev->recovery_cp) {
6440 printk(KERN_INFO
6441 "md: checkpointing %s of %s.\n",
6442 desc, mdname(mddev));
6443 mddev->recovery_cp = mddev->curr_resync;
6444 }
6445 } else
6446 mddev->recovery_cp = MaxSector;
6447 } else {
6448 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6449 mddev->curr_resync = MaxSector;
6450 list_for_each_entry(rdev, &mddev->disks, same_set)
6451 if (rdev->raid_disk >= 0 &&
6452 !test_bit(Faulty, &rdev->flags) &&
6453 !test_bit(In_sync, &rdev->flags) &&
6454 rdev->recovery_offset < mddev->curr_resync)
6455 rdev->recovery_offset = mddev->curr_resync;
6456 }
6457 }
6458 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6459
6460 skip:
6461 mddev->curr_resync = 0;
6462 mddev->curr_resync_completed = 0;
6463 mddev->resync_min = 0;
6464 mddev->resync_max = MaxSector;
6465 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6466 wake_up(&resync_wait);
6467 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6468 md_wakeup_thread(mddev->thread);
6469 return;
6470
6471 interrupted:
6472 /*
6473 * got a signal, exit.
6474 */
6475 printk(KERN_INFO
6476 "md: md_do_sync() got signal ... exiting\n");
6477 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6478 goto out;
6479
6480 }
6481 EXPORT_SYMBOL_GPL(md_do_sync);
6482
6483
6484 static int remove_and_add_spares(mddev_t *mddev)
6485 {
6486 mdk_rdev_t *rdev;
6487 int spares = 0;
6488
6489 mddev->curr_resync_completed = 0;
6490
6491 list_for_each_entry(rdev, &mddev->disks, same_set)
6492 if (rdev->raid_disk >= 0 &&
6493 !test_bit(Blocked, &rdev->flags) &&
6494 (test_bit(Faulty, &rdev->flags) ||
6495 ! test_bit(In_sync, &rdev->flags)) &&
6496 atomic_read(&rdev->nr_pending)==0) {
6497 if (mddev->pers->hot_remove_disk(
6498 mddev, rdev->raid_disk)==0) {
6499 char nm[20];
6500 sprintf(nm,"rd%d", rdev->raid_disk);
6501 sysfs_remove_link(&mddev->kobj, nm);
6502 rdev->raid_disk = -1;
6503 }
6504 }
6505
6506 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6507 list_for_each_entry(rdev, &mddev->disks, same_set) {
6508 if (rdev->raid_disk >= 0 &&
6509 !test_bit(In_sync, &rdev->flags) &&
6510 !test_bit(Blocked, &rdev->flags))
6511 spares++;
6512 if (rdev->raid_disk < 0
6513 && !test_bit(Faulty, &rdev->flags)) {
6514 rdev->recovery_offset = 0;
6515 if (mddev->pers->
6516 hot_add_disk(mddev, rdev) == 0) {
6517 char nm[20];
6518 sprintf(nm, "rd%d", rdev->raid_disk);
6519 if (sysfs_create_link(&mddev->kobj,
6520 &rdev->kobj, nm))
6521 printk(KERN_WARNING
6522 "md: cannot register "
6523 "%s for %s\n",
6524 nm, mdname(mddev));
6525 spares++;
6526 md_new_event(mddev);
6527 } else
6528 break;
6529 }
6530 }
6531 }
6532 return spares;
6533 }
6534 /*
6535 * This routine is regularly called by all per-raid-array threads to
6536 * deal with generic issues like resync and super-block update.
6537 * Raid personalities that don't have a thread (linear/raid0) do not
6538 * need this as they never do any recovery or update the superblock.
6539 *
6540 * It does not do any resync itself, but rather "forks" off other threads
6541 * to do that as needed.
6542 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6543 * "->recovery" and create a thread at ->sync_thread.
6544 * When the thread finishes it sets MD_RECOVERY_DONE
6545 * and wakeups up this thread which will reap the thread and finish up.
6546 * This thread also removes any faulty devices (with nr_pending == 0).
6547 *
6548 * The overall approach is:
6549 * 1/ if the superblock needs updating, update it.
6550 * 2/ If a recovery thread is running, don't do anything else.
6551 * 3/ If recovery has finished, clean up, possibly marking spares active.
6552 * 4/ If there are any faulty devices, remove them.
6553 * 5/ If array is degraded, try to add spares devices
6554 * 6/ If array has spares or is not in-sync, start a resync thread.
6555 */
6556 void md_check_recovery(mddev_t *mddev)
6557 {
6558 mdk_rdev_t *rdev;
6559
6560
6561 if (mddev->bitmap)
6562 bitmap_daemon_work(mddev->bitmap);
6563
6564 if (mddev->ro)
6565 return;
6566
6567 if (signal_pending(current)) {
6568 if (mddev->pers->sync_request && !mddev->external) {
6569 printk(KERN_INFO "md: %s in immediate safe mode\n",
6570 mdname(mddev));
6571 mddev->safemode = 2;
6572 }
6573 flush_signals(current);
6574 }
6575
6576 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6577 return;
6578 if ( ! (
6579 (mddev->flags && !mddev->external) ||
6580 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6581 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6582 (mddev->external == 0 && mddev->safemode == 1) ||
6583 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6584 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6585 ))
6586 return;
6587
6588 if (mddev_trylock(mddev)) {
6589 int spares = 0;
6590
6591 if (mddev->ro) {
6592 /* Only thing we do on a ro array is remove
6593 * failed devices.
6594 */
6595 remove_and_add_spares(mddev);
6596 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6597 goto unlock;
6598 }
6599
6600 if (!mddev->external) {
6601 int did_change = 0;
6602 spin_lock_irq(&mddev->write_lock);
6603 if (mddev->safemode &&
6604 !atomic_read(&mddev->writes_pending) &&
6605 !mddev->in_sync &&
6606 mddev->recovery_cp == MaxSector) {
6607 mddev->in_sync = 1;
6608 did_change = 1;
6609 if (mddev->persistent)
6610 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6611 }
6612 if (mddev->safemode == 1)
6613 mddev->safemode = 0;
6614 spin_unlock_irq(&mddev->write_lock);
6615 if (did_change)
6616 sysfs_notify_dirent(mddev->sysfs_state);
6617 }
6618
6619 if (mddev->flags)
6620 md_update_sb(mddev, 0);
6621
6622 list_for_each_entry(rdev, &mddev->disks, same_set)
6623 if (test_and_clear_bit(StateChanged, &rdev->flags))
6624 sysfs_notify_dirent(rdev->sysfs_state);
6625
6626
6627 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6628 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6629 /* resync/recovery still happening */
6630 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6631 goto unlock;
6632 }
6633 if (mddev->sync_thread) {
6634 /* resync has finished, collect result */
6635 md_unregister_thread(mddev->sync_thread);
6636 mddev->sync_thread = NULL;
6637 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6638 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6639 /* success...*/
6640 /* activate any spares */
6641 if (mddev->pers->spare_active(mddev))
6642 sysfs_notify(&mddev->kobj, NULL,
6643 "degraded");
6644 }
6645 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6646 mddev->pers->finish_reshape)
6647 mddev->pers->finish_reshape(mddev);
6648 md_update_sb(mddev, 1);
6649
6650 /* if array is no-longer degraded, then any saved_raid_disk
6651 * information must be scrapped
6652 */
6653 if (!mddev->degraded)
6654 list_for_each_entry(rdev, &mddev->disks, same_set)
6655 rdev->saved_raid_disk = -1;
6656
6657 mddev->recovery = 0;
6658 /* flag recovery needed just to double check */
6659 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6660 sysfs_notify_dirent(mddev->sysfs_action);
6661 md_new_event(mddev);
6662 goto unlock;
6663 }
6664 /* Set RUNNING before clearing NEEDED to avoid
6665 * any transients in the value of "sync_action".
6666 */
6667 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6668 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6669 /* Clear some bits that don't mean anything, but
6670 * might be left set
6671 */
6672 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6673 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6674
6675 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6676 goto unlock;
6677 /* no recovery is running.
6678 * remove any failed drives, then
6679 * add spares if possible.
6680 * Spare are also removed and re-added, to allow
6681 * the personality to fail the re-add.
6682 */
6683
6684 if (mddev->reshape_position != MaxSector) {
6685 if (mddev->pers->check_reshape == NULL ||
6686 mddev->pers->check_reshape(mddev) != 0)
6687 /* Cannot proceed */
6688 goto unlock;
6689 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6690 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6691 } else if ((spares = remove_and_add_spares(mddev))) {
6692 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6693 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6694 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6695 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6696 } else if (mddev->recovery_cp < MaxSector) {
6697 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6698 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6699 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6700 /* nothing to be done ... */
6701 goto unlock;
6702
6703 if (mddev->pers->sync_request) {
6704 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6705 /* We are adding a device or devices to an array
6706 * which has the bitmap stored on all devices.
6707 * So make sure all bitmap pages get written
6708 */
6709 bitmap_write_all(mddev->bitmap);
6710 }
6711 mddev->sync_thread = md_register_thread(md_do_sync,
6712 mddev,
6713 "%s_resync");
6714 if (!mddev->sync_thread) {
6715 printk(KERN_ERR "%s: could not start resync"
6716 " thread...\n",
6717 mdname(mddev));
6718 /* leave the spares where they are, it shouldn't hurt */
6719 mddev->recovery = 0;
6720 } else
6721 md_wakeup_thread(mddev->sync_thread);
6722 sysfs_notify_dirent(mddev->sysfs_action);
6723 md_new_event(mddev);
6724 }
6725 unlock:
6726 if (!mddev->sync_thread) {
6727 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6728 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6729 &mddev->recovery))
6730 if (mddev->sysfs_action)
6731 sysfs_notify_dirent(mddev->sysfs_action);
6732 }
6733 mddev_unlock(mddev);
6734 }
6735 }
6736
6737 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6738 {
6739 sysfs_notify_dirent(rdev->sysfs_state);
6740 wait_event_timeout(rdev->blocked_wait,
6741 !test_bit(Blocked, &rdev->flags),
6742 msecs_to_jiffies(5000));
6743 rdev_dec_pending(rdev, mddev);
6744 }
6745 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6746
6747 static int md_notify_reboot(struct notifier_block *this,
6748 unsigned long code, void *x)
6749 {
6750 struct list_head *tmp;
6751 mddev_t *mddev;
6752
6753 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6754
6755 printk(KERN_INFO "md: stopping all md devices.\n");
6756
6757 for_each_mddev(mddev, tmp)
6758 if (mddev_trylock(mddev)) {
6759 /* Force a switch to readonly even array
6760 * appears to still be in use. Hence
6761 * the '100'.
6762 */
6763 do_md_stop(mddev, 1, 100);
6764 mddev_unlock(mddev);
6765 }
6766 /*
6767 * certain more exotic SCSI devices are known to be
6768 * volatile wrt too early system reboots. While the
6769 * right place to handle this issue is the given
6770 * driver, we do want to have a safe RAID driver ...
6771 */
6772 mdelay(1000*1);
6773 }
6774 return NOTIFY_DONE;
6775 }
6776
6777 static struct notifier_block md_notifier = {
6778 .notifier_call = md_notify_reboot,
6779 .next = NULL,
6780 .priority = INT_MAX, /* before any real devices */
6781 };
6782
6783 static void md_geninit(void)
6784 {
6785 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6786
6787 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6788 }
6789
6790 static int __init md_init(void)
6791 {
6792 if (register_blkdev(MD_MAJOR, "md"))
6793 return -1;
6794 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6795 unregister_blkdev(MD_MAJOR, "md");
6796 return -1;
6797 }
6798 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6799 md_probe, NULL, NULL);
6800 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6801 md_probe, NULL, NULL);
6802
6803 register_reboot_notifier(&md_notifier);
6804 raid_table_header = register_sysctl_table(raid_root_table);
6805
6806 md_geninit();
6807 return 0;
6808 }
6809
6810
6811 #ifndef MODULE
6812
6813 /*
6814 * Searches all registered partitions for autorun RAID arrays
6815 * at boot time.
6816 */
6817
6818 static LIST_HEAD(all_detected_devices);
6819 struct detected_devices_node {
6820 struct list_head list;
6821 dev_t dev;
6822 };
6823
6824 void md_autodetect_dev(dev_t dev)
6825 {
6826 struct detected_devices_node *node_detected_dev;
6827
6828 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6829 if (node_detected_dev) {
6830 node_detected_dev->dev = dev;
6831 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6832 } else {
6833 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6834 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6835 }
6836 }
6837
6838
6839 static void autostart_arrays(int part)
6840 {
6841 mdk_rdev_t *rdev;
6842 struct detected_devices_node *node_detected_dev;
6843 dev_t dev;
6844 int i_scanned, i_passed;
6845
6846 i_scanned = 0;
6847 i_passed = 0;
6848
6849 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6850
6851 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6852 i_scanned++;
6853 node_detected_dev = list_entry(all_detected_devices.next,
6854 struct detected_devices_node, list);
6855 list_del(&node_detected_dev->list);
6856 dev = node_detected_dev->dev;
6857 kfree(node_detected_dev);
6858 rdev = md_import_device(dev,0, 90);
6859 if (IS_ERR(rdev))
6860 continue;
6861
6862 if (test_bit(Faulty, &rdev->flags)) {
6863 MD_BUG();
6864 continue;
6865 }
6866 set_bit(AutoDetected, &rdev->flags);
6867 list_add(&rdev->same_set, &pending_raid_disks);
6868 i_passed++;
6869 }
6870
6871 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6872 i_scanned, i_passed);
6873
6874 autorun_devices(part);
6875 }
6876
6877 #endif /* !MODULE */
6878
6879 static __exit void md_exit(void)
6880 {
6881 mddev_t *mddev;
6882 struct list_head *tmp;
6883
6884 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6885 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6886
6887 unregister_blkdev(MD_MAJOR,"md");
6888 unregister_blkdev(mdp_major, "mdp");
6889 unregister_reboot_notifier(&md_notifier);
6890 unregister_sysctl_table(raid_table_header);
6891 remove_proc_entry("mdstat", NULL);
6892 for_each_mddev(mddev, tmp) {
6893 export_array(mddev);
6894 mddev->hold_active = 0;
6895 }
6896 }
6897
6898 subsys_initcall(md_init);
6899 module_exit(md_exit)
6900
6901 static int get_ro(char *buffer, struct kernel_param *kp)
6902 {
6903 return sprintf(buffer, "%d", start_readonly);
6904 }
6905 static int set_ro(const char *val, struct kernel_param *kp)
6906 {
6907 char *e;
6908 int num = simple_strtoul(val, &e, 10);
6909 if (*val && (*e == '\0' || *e == '\n')) {
6910 start_readonly = num;
6911 return 0;
6912 }
6913 return -EINVAL;
6914 }
6915
6916 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6917 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6918
6919 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6920
6921 EXPORT_SYMBOL(register_md_personality);
6922 EXPORT_SYMBOL(unregister_md_personality);
6923 EXPORT_SYMBOL(md_error);
6924 EXPORT_SYMBOL(md_done_sync);
6925 EXPORT_SYMBOL(md_write_start);
6926 EXPORT_SYMBOL(md_write_end);
6927 EXPORT_SYMBOL(md_register_thread);
6928 EXPORT_SYMBOL(md_unregister_thread);
6929 EXPORT_SYMBOL(md_wakeup_thread);
6930 EXPORT_SYMBOL(md_check_recovery);
6931 MODULE_LICENSE("GPL");
6932 MODULE_ALIAS("md");
6933 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree