]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/md/md.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] md: unify usage of symbolic names for perms
[net-next-2.6.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 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
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 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/module.h>
a6fb0934 36#include <linux/kthread.h>
1da177e4
LT		 37#include <linux/linkage.h>
38#include <linux/raid/md.h>
32a7627c 39#include <linux/raid/bitmap.h>
1da177e4 40#include <linux/sysctl.h>
1da177e4
LT		 41#include <linux/buffer_head.h> /* for invalidate_bdev */
42#include <linux/suspend.h>
d7603b7e 43#include <linux/poll.h>
48c9c27b 44#include <linux/mutex.h>
16f17b39 45#include <linux/ctype.h>
1da177e4
LT		 46
47#include <linux/init.h>
48
32a7627c
N		 49#include <linux/file.h>
50
1da177e4
LT		 51#ifdef CONFIG_KMOD
52#include <linux/kmod.h>
53#endif
54
55#include <asm/unaligned.h>
56
57#define MAJOR_NR MD_MAJOR
58#define MD_DRIVER
59
60/* 63 partitions with the alternate major number (mdp) */
61#define MdpMinorShift 6
62
63#define DEBUG 0
64#define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67#ifndef MODULE
68static void autostart_arrays (int part);
69#endif
70
2604b703 71static LIST_HEAD(pers_list);
1da177e4
LT		 72static DEFINE_SPINLOCK(pers_lock);
73
5e56341d
AB		 74static void md_print_devices(void);
75
76#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
1da177e4
LT		 78/*
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 82 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
85 * idle IO detection.
86 *
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 88 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 89 */
90
91static int sysctl_speed_limit_min = 1000;
92static int sysctl_speed_limit_max = 200000;
88202a0c
N		 93static inline int speed_min(mddev_t *mddev)
94{
95 return mddev->sync_speed_min ?
96 mddev->sync_speed_min : sysctl_speed_limit_min;
97}
98
99static inline int speed_max(mddev_t *mddev)
100{
101 return mddev->sync_speed_max ?
102 mddev->sync_speed_max : sysctl_speed_limit_max;
103}
1da177e4
LT		 104
105static struct ctl_table_header *raid_table_header;
106
107static ctl_table raid_table[] = {
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
110 .procname = "speed_limit_min",
111 .data = &sysctl_speed_limit_min,
112 .maxlen = sizeof(int),
80ca3a44 113 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 114 .proc_handler = &proc_dointvec,
115 },
116 {
117 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
118 .procname = "speed_limit_max",
119 .data = &sysctl_speed_limit_max,
120 .maxlen = sizeof(int),
80ca3a44 121 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 122 .proc_handler = &proc_dointvec,
123 },
124 { .ctl_name = 0 }
125};
126
127static ctl_table raid_dir_table[] = {
128 {
129 .ctl_name = DEV_RAID,
130 .procname = "raid",
131 .maxlen = 0,
80ca3a44 132 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 133 .child = raid_table,
134 },
135 { .ctl_name = 0 }
136};
137
138static ctl_table raid_root_table[] = {
139 {
140 .ctl_name = CTL_DEV,
141 .procname = "dev",
142 .maxlen = 0,
143 .mode = 0555,
144 .child = raid_dir_table,
145 },
146 { .ctl_name = 0 }
147};
148
149static struct block_device_operations md_fops;
150
f91de92e
N		 151static int start_readonly;
152
d7603b7e
N		 153/*
154 * We have a system wide 'event count' that is incremented
155 * on any 'interesting' event, and readers of /proc/mdstat
156 * can use 'poll' or 'select' to find out when the event
157 * count increases.
158 *
159 * Events are:
160 * start array, stop array, error, add device, remove device,
161 * start build, activate spare
162 */
2989ddbd 163static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 164static atomic_t md_event_count;
29269553 165void md_new_event(mddev_t *mddev)
d7603b7e
N		 166{
167 atomic_inc(&md_event_count);
168 wake_up(&md_event_waiters);
4508a7a7 169 sysfs_notify(&mddev->kobj, NULL, "sync_action");
d7603b7e 170}
29269553 171EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 172
c331eb04
N		 173/* Alternate version that can be called from interrupts
174 * when calling sysfs_notify isn't needed.
175 */
05381954 176static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 177{
178 atomic_inc(&md_event_count);
179 wake_up(&md_event_waiters);
180}
181
1da177e4
LT		 182/*
183 * Enables to iterate over all existing md arrays
184 * all_mddevs_lock protects this list.
185 */
186static LIST_HEAD(all_mddevs);
187static DEFINE_SPINLOCK(all_mddevs_lock);
188
189
190/*
191 * iterates through all used mddevs in the system.
192 * We take care to grab the all_mddevs_lock whenever navigating
193 * the list, and to always hold a refcount when unlocked.
194 * Any code which breaks out of this loop while own
195 * a reference to the current mddev and must mddev_put it.
196 */
197#define ITERATE_MDDEV(mddev,tmp) \
198 \
199 for (({ spin_lock(&all_mddevs_lock); \
200 tmp = all_mddevs.next; \
201 mddev = NULL;}); \
202 ({ if (tmp != &all_mddevs) \
203 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
204 spin_unlock(&all_mddevs_lock); \
205 if (mddev) mddev_put(mddev); \
206 mddev = list_entry(tmp, mddev_t, all_mddevs); \
207 tmp != &all_mddevs;}); \
208 ({ spin_lock(&all_mddevs_lock); \
209 tmp = tmp->next;}) \
210 )
211
212
213static int md_fail_request (request_queue_t *q, struct bio *bio)
214{
215 bio_io_error(bio, bio->bi_size);
216 return 0;
217}
218
219static inline mddev_t *mddev_get(mddev_t *mddev)
220{
221 atomic_inc(&mddev->active);
222 return mddev;
223}
224
225static void mddev_put(mddev_t *mddev)
226{
227 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
228 return;
229 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
230 list_del(&mddev->all_mddevs);
926ce2d8 231 spin_unlock(&all_mddevs_lock);
1312f40e 232 blk_cleanup_queue(mddev->queue);
eae1701f 233 kobject_unregister(&mddev->kobj);
926ce2d8
N		 234 } else
235 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 236}
237
238static mddev_t * mddev_find(dev_t unit)
239{
240 mddev_t *mddev, *new = NULL;
241
242 retry:
243 spin_lock(&all_mddevs_lock);
244 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
245 if (mddev->unit == unit) {
246 mddev_get(mddev);
247 spin_unlock(&all_mddevs_lock);
990a8baf 248 kfree(new);
1da177e4
LT		 249 return mddev;
250 }
251
252 if (new) {
253 list_add(&new->all_mddevs, &all_mddevs);
254 spin_unlock(&all_mddevs_lock);
255 return new;
256 }
257 spin_unlock(&all_mddevs_lock);
258
9ffae0cf 259 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 260 if (!new)
261 return NULL;
262
1da177e4
LT		 263 new->unit = unit;
264 if (MAJOR(unit) == MD_MAJOR)
265 new->md_minor = MINOR(unit);
266 else
267 new->md_minor = MINOR(unit) >> MdpMinorShift;
268
df5b89b3 269 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 270 INIT_LIST_HEAD(&new->disks);
271 INIT_LIST_HEAD(&new->all_mddevs);
272 init_timer(&new->safemode_timer);
273 atomic_set(&new->active, 1);
06d91a5f 274 spin_lock_init(&new->write_lock);
3d310eb7 275 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 276
277 new->queue = blk_alloc_queue(GFP_KERNEL);
278 if (!new->queue) {
279 kfree(new);
280 return NULL;
281 }
89e5c8b5 282 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
1da177e4
LT		 283
284 blk_queue_make_request(new->queue, md_fail_request);
285
286 goto retry;
287}
288
289static inline int mddev_lock(mddev_t * mddev)
290{
df5b89b3 291 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 292}
293
1da177e4
LT		 294static inline int mddev_trylock(mddev_t * mddev)
295{
df5b89b3 296 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 297}
298
299static inline void mddev_unlock(mddev_t * mddev)
300{
df5b89b3 301 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 302
005eca5e 303 md_wakeup_thread(mddev->thread);
1da177e4
LT		 304}
305
2989ddbd 306static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 307{
308 mdk_rdev_t * rdev;
309 struct list_head *tmp;
310
311 ITERATE_RDEV(mddev,rdev,tmp) {
312 if (rdev->desc_nr == nr)
313 return rdev;
314 }
315 return NULL;
316}
317
318static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
319{
320 struct list_head *tmp;
321 mdk_rdev_t *rdev;
322
323 ITERATE_RDEV(mddev,rdev,tmp) {
324 if (rdev->bdev->bd_dev == dev)
325 return rdev;
326 }
327 return NULL;
328}
329
d9d166c2 330static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 331{
332 struct mdk_personality *pers;
d9d166c2
N		 333 list_for_each_entry(pers, &pers_list, list) {
334 if (level != LEVEL_NONE && pers->level == level)
2604b703 335 return pers;
d9d166c2
N		 336 if (strcmp(pers->name, clevel)==0)
337 return pers;
338 }
2604b703
N		 339 return NULL;
340}
341
77933d72 342static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 343{
344 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
345 return MD_NEW_SIZE_BLOCKS(size);
346}
347
348static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
349{
350 sector_t size;
351
352 size = rdev->sb_offset;
353
354 if (chunk_size)
355 size &= ~((sector_t)chunk_size/1024 - 1);
356 return size;
357}
358
359static int alloc_disk_sb(mdk_rdev_t * rdev)
360{
361 if (rdev->sb_page)
362 MD_BUG();
363
364 rdev->sb_page = alloc_page(GFP_KERNEL);
365 if (!rdev->sb_page) {
366 printk(KERN_ALERT "md: out of memory.\n");
367 return -EINVAL;
368 }
369
370 return 0;
371}
372
373static void free_disk_sb(mdk_rdev_t * rdev)
374{
375 if (rdev->sb_page) {
2d1f3b5d 376 put_page(rdev->sb_page);
1da177e4
LT		 377 rdev->sb_loaded = 0;
378 rdev->sb_page = NULL;
379 rdev->sb_offset = 0;
380 rdev->size = 0;
381 }
382}
383
384
7bfa19f2
N		 385static int super_written(struct bio *bio, unsigned int bytes_done, int error)
386{
387 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 388 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 389 if (bio->bi_size)
390 return 1;
391
392 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
a9701a30 393 md_error(mddev, rdev);
7bfa19f2 394
a9701a30
N		 395 if (atomic_dec_and_test(&mddev->pending_writes))
396 wake_up(&mddev->sb_wait);
f8b58edf 397 bio_put(bio);
7bfa19f2
N		 398 return 0;
399}
400
a9701a30
N		 401static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
402{
403 struct bio *bio2 = bio->bi_private;
404 mdk_rdev_t *rdev = bio2->bi_private;
405 mddev_t *mddev = rdev->mddev;
406 if (bio->bi_size)
407 return 1;
408
409 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
410 error == -EOPNOTSUPP) {
411 unsigned long flags;
412 /* barriers don't appear to be supported :-( */
413 set_bit(BarriersNotsupp, &rdev->flags);
414 mddev->barriers_work = 0;
415 spin_lock_irqsave(&mddev->write_lock, flags);
416 bio2->bi_next = mddev->biolist;
417 mddev->biolist = bio2;
418 spin_unlock_irqrestore(&mddev->write_lock, flags);
419 wake_up(&mddev->sb_wait);
420 bio_put(bio);
421 return 0;
422 }
423 bio_put(bio2);
424 bio->bi_private = rdev;
425 return super_written(bio, bytes_done, error);
426}
427
7bfa19f2
N		 428void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
429 sector_t sector, int size, struct page *page)
430{
431 /* write first size bytes of page to sector of rdev
432 * Increment mddev->pending_writes before returning
433 * and decrement it on completion, waking up sb_wait
434 * if zero is reached.
435 * If an error occurred, call md_error
a9701a30
N		 436 *
437 * As we might need to resubmit the request if BIO_RW_BARRIER
438 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 439 */
440 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 441 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 442
443 bio->bi_bdev = rdev->bdev;
444 bio->bi_sector = sector;
445 bio_add_page(bio, page, size, 0);
446 bio->bi_private = rdev;
447 bio->bi_end_io = super_written;
a9701a30
N		 448 bio->bi_rw = rw;
449
7bfa19f2 450 atomic_inc(&mddev->pending_writes);
a9701a30
N		 451 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
452 struct bio *rbio;
453 rw |= (1<<BIO_RW_BARRIER);
454 rbio = bio_clone(bio, GFP_NOIO);
455 rbio->bi_private = bio;
456 rbio->bi_end_io = super_written_barrier;
457 submit_bio(rw, rbio);
458 } else
459 submit_bio(rw, bio);
460}
461
462void md_super_wait(mddev_t *mddev)
463{
464 /* wait for all superblock writes that were scheduled to complete.
465 * if any had to be retried (due to BARRIER problems), retry them
466 */
467 DEFINE_WAIT(wq);
468 for(;;) {
469 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
470 if (atomic_read(&mddev->pending_writes)==0)
471 break;
472 while (mddev->biolist) {
473 struct bio *bio;
474 spin_lock_irq(&mddev->write_lock);
475 bio = mddev->biolist;
476 mddev->biolist = bio->bi_next ;
477 bio->bi_next = NULL;
478 spin_unlock_irq(&mddev->write_lock);
479 submit_bio(bio->bi_rw, bio);
480 }
481 schedule();
482 }
483 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 484}
485
1da177e4
LT		 486static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
487{
488 if (bio->bi_size)
489 return 1;
490
491 complete((struct completion*)bio->bi_private);
492 return 0;
493}
494
a654b9d8 495int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 496 struct page *page, int rw)
497{
baaa2c51 498 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 499 struct completion event;
500 int ret;
501
502 rw |= (1 << BIO_RW_SYNC);
503
504 bio->bi_bdev = bdev;
505 bio->bi_sector = sector;
506 bio_add_page(bio, page, size, 0);
507 init_completion(&event);
508 bio->bi_private = &event;
509 bio->bi_end_io = bi_complete;
510 submit_bio(rw, bio);
511 wait_for_completion(&event);
512
513 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
514 bio_put(bio);
515 return ret;
516}
a8745db2 517EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 518
0002b271 519static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 520{
521 char b[BDEVNAME_SIZE];
522 if (!rdev->sb_page) {
523 MD_BUG();
524 return -EINVAL;
525 }
526 if (rdev->sb_loaded)
527 return 0;
528
529
0002b271 530 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 531 goto fail;
532 rdev->sb_loaded = 1;
533 return 0;
534
535fail:
536 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
537 bdevname(rdev->bdev,b));
538 return -EINVAL;
539}
540
541static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
542{
543 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
544 (sb1->set_uuid1 == sb2->set_uuid1) &&
545 (sb1->set_uuid2 == sb2->set_uuid2) &&
546 (sb1->set_uuid3 == sb2->set_uuid3))
547
548 return 1;
549
550 return 0;
551}
552
553
554static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
555{
556 int ret;
557 mdp_super_t *tmp1, *tmp2;
558
559 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
560 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
561
562 if (!tmp1 || !tmp2) {
563 ret = 0;
564 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
565 goto abort;
566 }
567
568 *tmp1 = *sb1;
569 *tmp2 = *sb2;
570
571 /*
572 * nr_disks is not constant
573 */
574 tmp1->nr_disks = 0;
575 tmp2->nr_disks = 0;
576
577 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
578 ret = 0;
579 else
580 ret = 1;
581
582abort:
990a8baf
JJ		 583 kfree(tmp1);
584 kfree(tmp2);
1da177e4
LT		 585 return ret;
586}
587
588static unsigned int calc_sb_csum(mdp_super_t * sb)
589{
590 unsigned int disk_csum, csum;
591
592 disk_csum = sb->sb_csum;
593 sb->sb_csum = 0;
594 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
595 sb->sb_csum = disk_csum;
596 return csum;
597}
598
599
600/*
601 * Handle superblock details.
602 * We want to be able to handle multiple superblock formats
603 * so we have a common interface to them all, and an array of
604 * different handlers.
605 * We rely on user-space to write the initial superblock, and support
606 * reading and updating of superblocks.
607 * Interface methods are:
608 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
609 * loads and validates a superblock on dev.
610 * if refdev != NULL, compare superblocks on both devices
611 * Return:
612 * 0 - dev has a superblock that is compatible with refdev
613 * 1 - dev has a superblock that is compatible and newer than refdev
614 * so dev should be used as the refdev in future
615 * -EINVAL superblock incompatible or invalid
616 * -othererror e.g. -EIO
617 *
618 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
619 * Verify that dev is acceptable into mddev.
620 * The first time, mddev->raid_disks will be 0, and data from
621 * dev should be merged in. Subsequent calls check that dev
622 * is new enough. Return 0 or -EINVAL
623 *
624 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
625 * Update the superblock for rdev with data in mddev
626 * This does not write to disc.
627 *
628 */
629
630struct super_type {
631 char *name;
632 struct module *owner;
633 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
634 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
635 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
636};
637
638/*
639 * load_super for 0.90.0
640 */
641static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
642{
643 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
644 mdp_super_t *sb;
645 int ret;
646 sector_t sb_offset;
647
648 /*
649 * Calculate the position of the superblock,
650 * it's at the end of the disk.
651 *
652 * It also happens to be a multiple of 4Kb.
653 */
654 sb_offset = calc_dev_sboffset(rdev->bdev);
655 rdev->sb_offset = sb_offset;
656
0002b271 657 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 658 if (ret) return ret;
659
660 ret = -EINVAL;
661
662 bdevname(rdev->bdev, b);
663 sb = (mdp_super_t*)page_address(rdev->sb_page);
664
665 if (sb->md_magic != MD_SB_MAGIC) {
666 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
667 b);
668 goto abort;
669 }
670
671 if (sb->major_version != 0 ||
f6705578
N		 672 sb->minor_version < 90 ||
673 sb->minor_version > 91) {
1da177e4
LT		 674 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
675 sb->major_version, sb->minor_version,
676 b);
677 goto abort;
678 }
679
680 if (sb->raid_disks <= 0)
681 goto abort;
682
683 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
684 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
685 b);
686 goto abort;
687 }
688
689 rdev->preferred_minor = sb->md_minor;
690 rdev->data_offset = 0;
0002b271 691 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 692
693 if (sb->level == LEVEL_MULTIPATH)
694 rdev->desc_nr = -1;
695 else
696 rdev->desc_nr = sb->this_disk.number;
697
698 if (refdev == 0)
699 ret = 1;
700 else {
701 __u64 ev1, ev2;
702 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
703 if (!uuid_equal(refsb, sb)) {
704 printk(KERN_WARNING "md: %s has different UUID to %s\n",
705 b, bdevname(refdev->bdev,b2));
706 goto abort;
707 }
708 if (!sb_equal(refsb, sb)) {
709 printk(KERN_WARNING "md: %s has same UUID"
710 " but different superblock to %s\n",
711 b, bdevname(refdev->bdev, b2));
712 goto abort;
713 }
714 ev1 = md_event(sb);
715 ev2 = md_event(refsb);
716 if (ev1 > ev2)
717 ret = 1;
718 else
719 ret = 0;
720 }
721 rdev->size = calc_dev_size(rdev, sb->chunk_size);
722
2bf071bf
N		 723 if (rdev->size < sb->size && sb->level > 1)
724 /* "this cannot possibly happen" ... */
725 ret = -EINVAL;
726
1da177e4
LT		 727 abort:
728 return ret;
729}
730
731/*
732 * validate_super for 0.90.0
733 */
734static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
735{
736 mdp_disk_t *desc;
737 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 738 __u64 ev1 = md_event(sb);
1da177e4 739
41158c7e 740 rdev->raid_disk = -1;
b2d444d7 741 rdev->flags = 0;
1da177e4
LT		 742 if (mddev->raid_disks == 0) {
743 mddev->major_version = 0;
744 mddev->minor_version = sb->minor_version;
745 mddev->patch_version = sb->patch_version;
746 mddev->persistent = ! sb->not_persistent;
747 mddev->chunk_size = sb->chunk_size;
748 mddev->ctime = sb->ctime;
749 mddev->utime = sb->utime;
750 mddev->level = sb->level;
d9d166c2 751 mddev->clevel[0] = 0;
1da177e4
LT		 752 mddev->layout = sb->layout;
753 mddev->raid_disks = sb->raid_disks;
754 mddev->size = sb->size;
07d84d10 755 mddev->events = ev1;
9223214e 756 mddev->bitmap_offset = 0;
36fa3063 757 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 758
f6705578
N		 759 if (mddev->minor_version >= 91) {
760 mddev->reshape_position = sb->reshape_position;
761 mddev->delta_disks = sb->delta_disks;
762 mddev->new_level = sb->new_level;
763 mddev->new_layout = sb->new_layout;
764 mddev->new_chunk = sb->new_chunk;
765 } else {
766 mddev->reshape_position = MaxSector;
767 mddev->delta_disks = 0;
768 mddev->new_level = mddev->level;
769 mddev->new_layout = mddev->layout;
770 mddev->new_chunk = mddev->chunk_size;
771 }
772
1da177e4
LT		 773 if (sb->state & (1<<MD_SB_CLEAN))
774 mddev->recovery_cp = MaxSector;
775 else {
776 if (sb->events_hi == sb->cp_events_hi &&
777 sb->events_lo == sb->cp_events_lo) {
778 mddev->recovery_cp = sb->recovery_cp;
779 } else
780 mddev->recovery_cp = 0;
781 }
782
783 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
784 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
785 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
786 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
787
788 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 789
790 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
791 mddev->bitmap_file == NULL) {
c5a10f62
N		 792 if (mddev->level != 1 && mddev->level != 4
793 && mddev->level != 5 && mddev->level != 6
6cce3b23 794 && mddev->level != 10) {
a654b9d8 795 /* FIXME use a better test */
6cce3b23 796 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 797 return -EINVAL;
798 }
36fa3063 799 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8
N		 800 }
801
41158c7e
N		 802 } else if (mddev->pers == NULL) {
803 /* Insist on good event counter while assembling */
1da177e4
LT		 804 ++ev1;
805 if (ev1 < mddev->events)
806 return -EINVAL;
41158c7e
N		 807 } else if (mddev->bitmap) {
808 /* if adding to array with a bitmap, then we can accept an
809 * older device ... but not too old.
810 */
41158c7e
N		 811 if (ev1 < mddev->bitmap->events_cleared)
812 return 0;
07d84d10
N		 813 } else {
814 if (ev1 < mddev->events)
815 /* just a hot-add of a new device, leave raid_disk at -1 */
816 return 0;
817 }
41158c7e 818
1da177e4 819 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 820 desc = sb->disks + rdev->desc_nr;
821
822 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 823 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 824 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
825 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 826 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 827 rdev->raid_disk = desc->raid_disk;
828 }
8ddf9efe
N		 829 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
830 set_bit(WriteMostly, &rdev->flags);
41158c7e 831 } else /* MULTIPATH are always insync */
b2d444d7 832 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 833 return 0;
834}
835
836/*
837 * sync_super for 0.90.0
838 */
839static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
840{
841 mdp_super_t *sb;
842 struct list_head *tmp;
843 mdk_rdev_t *rdev2;
844 int next_spare = mddev->raid_disks;
19133a42 845
1da177e4
LT		 846
847 /* make rdev->sb match mddev data..
848 *
849 * 1/ zero out disks
850 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
851 * 3/ any empty disks < next_spare become removed
852 *
853 * disks[0] gets initialised to REMOVED because
854 * we cannot be sure from other fields if it has
855 * been initialised or not.
856 */
857 int i;
858 int active=0, working=0,failed=0,spare=0,nr_disks=0;
859
61181565
N		 860 rdev->sb_size = MD_SB_BYTES;
861
1da177e4
LT		 862 sb = (mdp_super_t*)page_address(rdev->sb_page);
863
864 memset(sb, 0, sizeof(*sb));
865
866 sb->md_magic = MD_SB_MAGIC;
867 sb->major_version = mddev->major_version;
1da177e4
LT		 868 sb->patch_version = mddev->patch_version;
869 sb->gvalid_words = 0; /* ignored */
870 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
871 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
872 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
873 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
874
875 sb->ctime = mddev->ctime;
876 sb->level = mddev->level;
877 sb->size = mddev->size;
878 sb->raid_disks = mddev->raid_disks;
879 sb->md_minor = mddev->md_minor;
880 sb->not_persistent = !mddev->persistent;
881 sb->utime = mddev->utime;
882 sb->state = 0;
883 sb->events_hi = (mddev->events>>32);
884 sb->events_lo = (u32)mddev->events;
885
f6705578
N		 886 if (mddev->reshape_position == MaxSector)
887 sb->minor_version = 90;
888 else {
889 sb->minor_version = 91;
890 sb->reshape_position = mddev->reshape_position;
891 sb->new_level = mddev->new_level;
892 sb->delta_disks = mddev->delta_disks;
893 sb->new_layout = mddev->new_layout;
894 sb->new_chunk = mddev->new_chunk;
895 }
896 mddev->minor_version = sb->minor_version;
1da177e4
LT		 897 if (mddev->in_sync)
898 {
899 sb->recovery_cp = mddev->recovery_cp;
900 sb->cp_events_hi = (mddev->events>>32);
901 sb->cp_events_lo = (u32)mddev->events;
902 if (mddev->recovery_cp == MaxSector)
903 sb->state = (1<< MD_SB_CLEAN);
904 } else
905 sb->recovery_cp = 0;
906
907 sb->layout = mddev->layout;
908 sb->chunk_size = mddev->chunk_size;
909
a654b9d8
N		 910 if (mddev->bitmap && mddev->bitmap_file == NULL)
911 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
912
1da177e4
LT		 913 sb->disks[0].state = (1<<MD_DISK_REMOVED);
914 ITERATE_RDEV(mddev,rdev2,tmp) {
915 mdp_disk_t *d;
86e6ffdd 916 int desc_nr;
b2d444d7
N		 917 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
918 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 919 desc_nr = rdev2->raid_disk;
1da177e4 920 else
86e6ffdd 921 desc_nr = next_spare++;
19133a42 922 rdev2->desc_nr = desc_nr;
1da177e4
LT		 923 d = &sb->disks[rdev2->desc_nr];
924 nr_disks++;
925 d->number = rdev2->desc_nr;
926 d->major = MAJOR(rdev2->bdev->bd_dev);
927 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 928 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
929 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 930 d->raid_disk = rdev2->raid_disk;
931 else
932 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 933 if (test_bit(Faulty, &rdev2->flags))
1da177e4 934 d->state = (1<<MD_DISK_FAULTY);
1be7892f 935 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 936 d->state = (1<<MD_DISK_ACTIVE);
937 d->state |= (1<<MD_DISK_SYNC);
938 active++;
939 working++;
940 } else {
941 d->state = 0;
942 spare++;
943 working++;
944 }
8ddf9efe
N		 945 if (test_bit(WriteMostly, &rdev2->flags))
946 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 947 }
1da177e4
LT		 948 /* now set the "removed" and "faulty" bits on any missing devices */
949 for (i=0 ; i < mddev->raid_disks ; i++) {
950 mdp_disk_t *d = &sb->disks[i];
951 if (d->state == 0 && d->number == 0) {
952 d->number = i;
953 d->raid_disk = i;
954 d->state = (1<<MD_DISK_REMOVED);
955 d->state |= (1<<MD_DISK_FAULTY);
956 failed++;
957 }
958 }
959 sb->nr_disks = nr_disks;
960 sb->active_disks = active;
961 sb->working_disks = working;
962 sb->failed_disks = failed;
963 sb->spare_disks = spare;
964
965 sb->this_disk = sb->disks[rdev->desc_nr];
966 sb->sb_csum = calc_sb_csum(sb);
967}
968
969/*
970 * version 1 superblock
971 */
972
973static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
974{
975 unsigned int disk_csum, csum;
976 unsigned long long newcsum;
977 int size = 256 + le32_to_cpu(sb->max_dev)*2;
978 unsigned int *isuper = (unsigned int*)sb;
979 int i;
980
981 disk_csum = sb->sb_csum;
982 sb->sb_csum = 0;
983 newcsum = 0;
984 for (i=0; size>=4; size -= 4 )
985 newcsum += le32_to_cpu(*isuper++);
986
987 if (size == 2)
988 newcsum += le16_to_cpu(*(unsigned short*) isuper);
989
990 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
991 sb->sb_csum = disk_csum;
992 return cpu_to_le32(csum);
993}
994
995static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
996{
997 struct mdp_superblock_1 *sb;
998 int ret;
999 sector_t sb_offset;
1000 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1001 int bmask;
1da177e4
LT		 1002
1003 /*
1004 * Calculate the position of the superblock.
1005 * It is always aligned to a 4K boundary and
1006 * depeding on minor_version, it can be:
1007 * 0: At least 8K, but less than 12K, from end of device
1008 * 1: At start of device
1009 * 2: 4K from start of device.
1010 */
1011 switch(minor_version) {
1012 case 0:
1013 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1014 sb_offset -= 8*2;
39730960 1015 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 1016 /* convert from sectors to K */
1017 sb_offset /= 2;
1018 break;
1019 case 1:
1020 sb_offset = 0;
1021 break;
1022 case 2:
1023 sb_offset = 4;
1024 break;
1025 default:
1026 return -EINVAL;
1027 }
1028 rdev->sb_offset = sb_offset;
1029
0002b271
N		 1030 /* superblock is rarely larger than 1K, but it can be larger,
1031 * and it is safe to read 4k, so we do that
1032 */
1033 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1034 if (ret) return ret;
1035
1036
1037 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1038
1039 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1040 sb->major_version != cpu_to_le32(1) ||
1041 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1042 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1043 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1044 return -EINVAL;
1045
1046 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1047 printk("md: invalid superblock checksum on %s\n",
1048 bdevname(rdev->bdev,b));
1049 return -EINVAL;
1050 }
1051 if (le64_to_cpu(sb->data_size) < 10) {
1052 printk("md: data_size too small on %s\n",
1053 bdevname(rdev->bdev,b));
1054 return -EINVAL;
1055 }
1056 rdev->preferred_minor = 0xffff;
1057 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1058 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1059
0002b271 1060 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1061 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1062 if (rdev->sb_size & bmask)
1063 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1064
31b65a0d
N		 1065 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1066 rdev->desc_nr = -1;
1067 else
1068 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1069
1da177e4 1070 if (refdev == 0)
8ed75463 1071 ret = 1;
1da177e4
LT		 1072 else {
1073 __u64 ev1, ev2;
1074 struct mdp_superblock_1 *refsb =
1075 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1076
1077 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1078 sb->level != refsb->level ||
1079 sb->layout != refsb->layout ||
1080 sb->chunksize != refsb->chunksize) {
1081 printk(KERN_WARNING "md: %s has strangely different"
1082 " superblock to %s\n",
1083 bdevname(rdev->bdev,b),
1084 bdevname(refdev->bdev,b2));
1085 return -EINVAL;
1086 }
1087 ev1 = le64_to_cpu(sb->events);
1088 ev2 = le64_to_cpu(refsb->events);
1089
1090 if (ev1 > ev2)
8ed75463
N		 1091 ret = 1;
1092 else
1093 ret = 0;
1da177e4
LT		 1094 }
1095 if (minor_version)
1096 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1097 else
1098 rdev->size = rdev->sb_offset;
1099 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1100 return -EINVAL;
1101 rdev->size = le64_to_cpu(sb->data_size)/2;
1102 if (le32_to_cpu(sb->chunksize))
1103 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf
N		 1104
1105 if (le32_to_cpu(sb->size) > rdev->size*2)
1106 return -EINVAL;
8ed75463 1107 return ret;
1da177e4
LT		 1108}
1109
1110static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1111{
1112 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1113 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1114
41158c7e 1115 rdev->raid_disk = -1;
b2d444d7 1116 rdev->flags = 0;
1da177e4
LT		 1117 if (mddev->raid_disks == 0) {
1118 mddev->major_version = 1;
1119 mddev->patch_version = 0;
1120 mddev->persistent = 1;
1121 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1122 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1123 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1124 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1125 mddev->clevel[0] = 0;
1da177e4
LT		 1126 mddev->layout = le32_to_cpu(sb->layout);
1127 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1128 mddev->size = le64_to_cpu(sb->size)/2;
07d84d10 1129 mddev->events = ev1;
9223214e 1130 mddev->bitmap_offset = 0;
29fc7e3e 1131 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1132
1133 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1134 memcpy(mddev->uuid, sb->set_uuid, 16);
1135
1136 mddev->max_disks = (4096-256)/2;
a654b9d8 1137
71c0805c 1138 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
a654b9d8 1139 mddev->bitmap_file == NULL ) {
6cce3b23
N		 1140 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1141 && mddev->level != 10) {
1142 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 1143 return -EINVAL;
1144 }
1145 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1146 }
f6705578
N		 1147 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1148 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1149 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1150 mddev->new_level = le32_to_cpu(sb->new_level);
1151 mddev->new_layout = le32_to_cpu(sb->new_layout);
1152 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1153 } else {
1154 mddev->reshape_position = MaxSector;
1155 mddev->delta_disks = 0;
1156 mddev->new_level = mddev->level;
1157 mddev->new_layout = mddev->layout;
1158 mddev->new_chunk = mddev->chunk_size;
1159 }
1160
41158c7e
N		 1161 } else if (mddev->pers == NULL) {
1162 /* Insist of good event counter while assembling */
1da177e4
LT		 1163 ++ev1;
1164 if (ev1 < mddev->events)
1165 return -EINVAL;
41158c7e
N		 1166 } else if (mddev->bitmap) {
1167 /* If adding to array with a bitmap, then we can accept an
1168 * older device, but not too old.
1169 */
41158c7e
N		 1170 if (ev1 < mddev->bitmap->events_cleared)
1171 return 0;
07d84d10
N		 1172 } else {
1173 if (ev1 < mddev->events)
1174 /* just a hot-add of a new device, leave raid_disk at -1 */
1175 return 0;
1176 }
1da177e4
LT		 1177 if (mddev->level != LEVEL_MULTIPATH) {
1178 int role;
1da177e4
LT		 1179 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1180 switch(role) {
1181 case 0xffff: /* spare */
1da177e4
LT		 1182 break;
1183 case 0xfffe: /* faulty */
b2d444d7 1184 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1185 break;
1186 default:
5fd6c1dc
N		 1187 if ((le32_to_cpu(sb->feature_map) &
1188 MD_FEATURE_RECOVERY_OFFSET))
1189 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1190 else
1191 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1192 rdev->raid_disk = role;
1193 break;
1194 }
8ddf9efe
N		 1195 if (sb->devflags & WriteMostly1)
1196 set_bit(WriteMostly, &rdev->flags);
41158c7e 1197 } else /* MULTIPATH are always insync */
b2d444d7 1198 set_bit(In_sync, &rdev->flags);
41158c7e 1199
1da177e4
LT		 1200 return 0;
1201}
1202
1203static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1204{
1205 struct mdp_superblock_1 *sb;
1206 struct list_head *tmp;
1207 mdk_rdev_t *rdev2;
1208 int max_dev, i;
1209 /* make rdev->sb match mddev and rdev data. */
1210
1211 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1212
1213 sb->feature_map = 0;
1214 sb->pad0 = 0;
5fd6c1dc 1215 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1216 memset(sb->pad1, 0, sizeof(sb->pad1));
1217 memset(sb->pad2, 0, sizeof(sb->pad2));
1218 memset(sb->pad3, 0, sizeof(sb->pad3));
1219
1220 sb->utime = cpu_to_le64((__u64)mddev->utime);
1221 sb->events = cpu_to_le64(mddev->events);
1222 if (mddev->in_sync)
1223 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1224 else
1225 sb->resync_offset = cpu_to_le64(0);
1226
4dbcdc75
N		 1227 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1228
f0ca340c 1229 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1230 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1231
a654b9d8
N		 1232 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1233 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1234 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1235 }
5fd6c1dc
N		 1236
1237 if (rdev->raid_disk >= 0 &&
1238 !test_bit(In_sync, &rdev->flags) &&
1239 rdev->recovery_offset > 0) {
1240 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1241 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1242 }
1243
f6705578
N		 1244 if (mddev->reshape_position != MaxSector) {
1245 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1246 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1247 sb->new_layout = cpu_to_le32(mddev->new_layout);
1248 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1249 sb->new_level = cpu_to_le32(mddev->new_level);
1250 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1251 }
a654b9d8 1252
1da177e4
LT		 1253 max_dev = 0;
1254 ITERATE_RDEV(mddev,rdev2,tmp)
1255 if (rdev2->desc_nr+1 > max_dev)
1256 max_dev = rdev2->desc_nr+1;
1257
1258 sb->max_dev = cpu_to_le32(max_dev);
1259 for (i=0; i<max_dev;i++)
1260 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1261
1262 ITERATE_RDEV(mddev,rdev2,tmp) {
1263 i = rdev2->desc_nr;
b2d444d7 1264 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1265 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1266 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1267 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1268 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1269 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1270 else
1271 sb->dev_roles[i] = cpu_to_le16(0xffff);
1272 }
1273
1da177e4
LT		 1274 sb->sb_csum = calc_sb_1_csum(sb);
1275}
1276
1277
75c96f85 1278static struct super_type super_types[] = {
1da177e4
LT		 1279 [0] = {
1280 .name = "0.90.0",
1281 .owner = THIS_MODULE,
1282 .load_super = super_90_load,
1283 .validate_super = super_90_validate,
1284 .sync_super = super_90_sync,
1285 },
1286 [1] = {
1287 .name = "md-1",
1288 .owner = THIS_MODULE,
1289 .load_super = super_1_load,
1290 .validate_super = super_1_validate,
1291 .sync_super = super_1_sync,
1292 },
1293};
1294
1295static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1296{
1297 struct list_head *tmp;
1298 mdk_rdev_t *rdev;
1299
1300 ITERATE_RDEV(mddev,rdev,tmp)
1301 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1302 return rdev;
1303
1304 return NULL;
1305}
1306
1307static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1308{
1309 struct list_head *tmp;
1310 mdk_rdev_t *rdev;
1311
1312 ITERATE_RDEV(mddev1,rdev,tmp)
1313 if (match_dev_unit(mddev2, rdev))
1314 return 1;
1315
1316 return 0;
1317}
1318
1319static LIST_HEAD(pending_raid_disks);
1320
1321static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1322{
1323 mdk_rdev_t *same_pdev;
1324 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
f637b9f9 1325 struct kobject *ko;
1edf80d3 1326 char *s;
1da177e4
LT		 1327
1328 if (rdev->mddev) {
1329 MD_BUG();
1330 return -EINVAL;
1331 }
2bf071bf
N		 1332 /* make sure rdev->size exceeds mddev->size */
1333 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1334 if (mddev->pers)
1335 /* Cannot change size, so fail */
1336 return -ENOSPC;
1337 else
1338 mddev->size = rdev->size;
1339 }
1da177e4
LT		 1340 same_pdev = match_dev_unit(mddev, rdev);
1341 if (same_pdev)
1342 printk(KERN_WARNING
1343 "%s: WARNING: %s appears to be on the same physical"
1344 " disk as %s. True\n protection against single-disk"
1345 " failure might be compromised.\n",
1346 mdname(mddev), bdevname(rdev->bdev,b),
1347 bdevname(same_pdev->bdev,b2));
1348
1349 /* Verify rdev->desc_nr is unique.
1350 * If it is -1, assign a free number, else
1351 * check number is not in use
1352 */
1353 if (rdev->desc_nr < 0) {
1354 int choice = 0;
1355 if (mddev->pers) choice = mddev->raid_disks;
1356 while (find_rdev_nr(mddev, choice))
1357 choice++;
1358 rdev->desc_nr = choice;
1359 } else {
1360 if (find_rdev_nr(mddev, rdev->desc_nr))
1361 return -EBUSY;
1362 }
19133a42
N		 1363 bdevname(rdev->bdev,b);
1364 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1365 return -ENOMEM;
1edf80d3
NB		 1366 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1367 *s = '!';
1da177e4
LT		 1368
1369 list_add(&rdev->same_set, &mddev->disks);
1370 rdev->mddev = mddev;
19133a42 1371 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1372
9c791977 1373 rdev->kobj.parent = &mddev->kobj;
86e6ffdd
N		 1374 kobject_add(&rdev->kobj);
1375
f637b9f9
N		 1376 if (rdev->bdev->bd_part)
1377 ko = &rdev->bdev->bd_part->kobj;
1378 else
1379 ko = &rdev->bdev->bd_disk->kobj;
1380 sysfs_create_link(&rdev->kobj, ko, "block");
5463c790 1381 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1da177e4
LT		 1382 return 0;
1383}
1384
1385static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1386{
1387 char b[BDEVNAME_SIZE];
1388 if (!rdev->mddev) {
1389 MD_BUG();
1390 return;
1391 }
5463c790 1392 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1393 list_del_init(&rdev->same_set);
1394 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1395 rdev->mddev = NULL;
86e6ffdd
N		 1396 sysfs_remove_link(&rdev->kobj, "block");
1397 kobject_del(&rdev->kobj);
1da177e4
LT		 1398}
1399
1400/*
1401 * prevent the device from being mounted, repartitioned or
1402 * otherwise reused by a RAID array (or any other kernel
1403 * subsystem), by bd_claiming the device.
1404 */
1405static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1406{
1407 int err = 0;
1408 struct block_device *bdev;
1409 char b[BDEVNAME_SIZE];
1410
663d440e 1411 bdev = open_partition_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1412 if (IS_ERR(bdev)) {
1413 printk(KERN_ERR "md: could not open %s.\n",
1414 __bdevname(dev, b));
1415 return PTR_ERR(bdev);
1416 }
1417 err = bd_claim(bdev, rdev);
1418 if (err) {
1419 printk(KERN_ERR "md: could not bd_claim %s.\n",
1420 bdevname(bdev, b));
663d440e 1421 blkdev_put_partition(bdev);
1da177e4
LT		 1422 return err;
1423 }
1424 rdev->bdev = bdev;
1425 return err;
1426}
1427
1428static void unlock_rdev(mdk_rdev_t *rdev)
1429{
1430 struct block_device *bdev = rdev->bdev;
1431 rdev->bdev = NULL;
1432 if (!bdev)
1433 MD_BUG();
1434 bd_release(bdev);
663d440e 1435 blkdev_put_partition(bdev);
1da177e4
LT		 1436}
1437
1438void md_autodetect_dev(dev_t dev);
1439
1440static void export_rdev(mdk_rdev_t * rdev)
1441{
1442 char b[BDEVNAME_SIZE];
1443 printk(KERN_INFO "md: export_rdev(%s)\n",
1444 bdevname(rdev->bdev,b));
1445 if (rdev->mddev)
1446 MD_BUG();
1447 free_disk_sb(rdev);
1448 list_del_init(&rdev->same_set);
1449#ifndef MODULE
1450 md_autodetect_dev(rdev->bdev->bd_dev);
1451#endif
1452 unlock_rdev(rdev);
86e6ffdd 1453 kobject_put(&rdev->kobj);
1da177e4
LT		 1454}
1455
1456static void kick_rdev_from_array(mdk_rdev_t * rdev)
1457{
1458 unbind_rdev_from_array(rdev);
1459 export_rdev(rdev);
1460}
1461
1462static void export_array(mddev_t *mddev)
1463{
1464 struct list_head *tmp;
1465 mdk_rdev_t *rdev;
1466
1467 ITERATE_RDEV(mddev,rdev,tmp) {
1468 if (!rdev->mddev) {
1469 MD_BUG();
1470 continue;
1471 }
1472 kick_rdev_from_array(rdev);
1473 }
1474 if (!list_empty(&mddev->disks))
1475 MD_BUG();
1476 mddev->raid_disks = 0;
1477 mddev->major_version = 0;
1478}
1479
1480static void print_desc(mdp_disk_t *desc)
1481{
1482 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1483 desc->major,desc->minor,desc->raid_disk,desc->state);
1484}
1485
1486static void print_sb(mdp_super_t *sb)
1487{
1488 int i;
1489
1490 printk(KERN_INFO
1491 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1492 sb->major_version, sb->minor_version, sb->patch_version,
1493 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1494 sb->ctime);
1495 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1496 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1497 sb->md_minor, sb->layout, sb->chunk_size);
1498 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1499 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1500 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1501 sb->failed_disks, sb->spare_disks,
1502 sb->sb_csum, (unsigned long)sb->events_lo);
1503
1504 printk(KERN_INFO);
1505 for (i = 0; i < MD_SB_DISKS; i++) {
1506 mdp_disk_t *desc;
1507
1508 desc = sb->disks + i;
1509 if (desc->number || desc->major || desc->minor ||
1510 desc->raid_disk || (desc->state && (desc->state != 4))) {
1511 printk(" D %2d: ", i);
1512 print_desc(desc);
1513 }
1514 }
1515 printk(KERN_INFO "md: THIS: ");
1516 print_desc(&sb->this_disk);
1517
1518}
1519
1520static void print_rdev(mdk_rdev_t *rdev)
1521{
1522 char b[BDEVNAME_SIZE];
1523 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1524 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1525 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1526 rdev->desc_nr);
1da177e4
LT		 1527 if (rdev->sb_loaded) {
1528 printk(KERN_INFO "md: rdev superblock:\n");
1529 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1530 } else
1531 printk(KERN_INFO "md: no rdev superblock!\n");
1532}
1533
5e56341d 1534static void md_print_devices(void)
1da177e4
LT		 1535{
1536 struct list_head *tmp, *tmp2;
1537 mdk_rdev_t *rdev;
1538 mddev_t *mddev;
1539 char b[BDEVNAME_SIZE];
1540
1541 printk("\n");
1542 printk("md: **********************************\n");
1543 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1544 printk("md: **********************************\n");
1545 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1546
32a7627c
N		 1547 if (mddev->bitmap)
1548 bitmap_print_sb(mddev->bitmap);
1549 else
1550 printk("%s: ", mdname(mddev));
1da177e4
LT		 1551 ITERATE_RDEV(mddev,rdev,tmp2)
1552 printk("<%s>", bdevname(rdev->bdev,b));
1553 printk("\n");
1554
1555 ITERATE_RDEV(mddev,rdev,tmp2)
1556 print_rdev(rdev);
1557 }
1558 printk("md: **********************************\n");
1559 printk("\n");
1560}
1561
1562
42543769 1563static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1564{
42543769
N		 1565 /* Update each superblock (in-memory image), but
1566 * if we are allowed to, skip spares which already
1567 * have the right event counter, or have one earlier
1568 * (which would mean they aren't being marked as dirty
1569 * with the rest of the array)
1570 */
1da177e4
LT		 1571 mdk_rdev_t *rdev;
1572 struct list_head *tmp;
1573
1574 ITERATE_RDEV(mddev,rdev,tmp) {
42543769
N		 1575 if (rdev->sb_events == mddev->events ||
1576 (nospares &&
1577 rdev->raid_disk < 0 &&
1578 (rdev->sb_events&1)==0 &&
1579 rdev->sb_events+1 == mddev->events)) {
1580 /* Don't update this superblock */
1581 rdev->sb_loaded = 2;
1582 } else {
1583 super_types[mddev->major_version].
1584 sync_super(mddev, rdev);
1585 rdev->sb_loaded = 1;
1586 }
1da177e4
LT		 1587 }
1588}
1589
f6705578 1590void md_update_sb(mddev_t * mddev)
1da177e4 1591{
7bfa19f2 1592 int err;
1da177e4
LT		 1593 struct list_head *tmp;
1594 mdk_rdev_t *rdev;
06d91a5f 1595 int sync_req;
42543769 1596 int nospares = 0;
1da177e4 1597
1da177e4 1598repeat:
a9701a30 1599 spin_lock_irq(&mddev->write_lock);
06d91a5f 1600 sync_req = mddev->in_sync;
1da177e4 1601 mddev->utime = get_seconds();
42543769
N		 1602 if (mddev->sb_dirty == 3)
1603 /* just a clean<-> dirty transition, possibly leave spares alone,
1604 * though if events isn't the right even/odd, we will have to do
1605 * spares after all
1606 */
1607 nospares = 1;
1608
1609 /* If this is just a dirty<->clean transition, and the array is clean
1610 * and 'events' is odd, we can roll back to the previous clean state */
1611 if (mddev->sb_dirty == 3
1612 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1613 && (mddev->events & 1))
1614 mddev->events--;
1615 else {
1616 /* otherwise we have to go forward and ... */
1617 mddev->events ++;
1618 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1619 /* .. if the array isn't clean, insist on an odd 'events' */
1620 if ((mddev->events&1)==0) {
1621 mddev->events++;
1622 nospares = 0;
1623 }
1624 } else {
1625 /* otherwise insist on an even 'events' (for clean states) */
1626 if ((mddev->events&1)) {
1627 mddev->events++;
1628 nospares = 0;
1629 }
1630 }
1631 }
1da177e4
LT		 1632
1633 if (!mddev->events) {
1634 /*
1635 * oops, this 64-bit counter should never wrap.
1636 * Either we are in around ~1 trillion A.C., assuming
1637 * 1 reboot per second, or we have a bug:
1638 */
1639 MD_BUG();
1640 mddev->events --;
1641 }
7bfa19f2 1642 mddev->sb_dirty = 2;
42543769 1643 sync_sbs(mddev, nospares);
1da177e4
LT		 1644
1645 /*
1646 * do not write anything to disk if using
1647 * nonpersistent superblocks
1648 */
06d91a5f
N		 1649 if (!mddev->persistent) {
1650 mddev->sb_dirty = 0;
a9701a30 1651 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1652 wake_up(&mddev->sb_wait);
1da177e4 1653 return;
06d91a5f 1654 }
a9701a30 1655 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1656
1657 dprintk(KERN_INFO
1658 "md: updating %s RAID superblock on device (in sync %d)\n",
1659 mdname(mddev),mddev->in_sync);
1660
32a7627c 1661 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1662 ITERATE_RDEV(mddev,rdev,tmp) {
1663 char b[BDEVNAME_SIZE];
1664 dprintk(KERN_INFO "md: ");
42543769
N		 1665 if (rdev->sb_loaded != 1)
1666 continue; /* no noise on spare devices */
b2d444d7 1667 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1668 dprintk("(skipping faulty ");
1669
1670 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1671 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1672 md_super_write(mddev,rdev,
0002b271 1673 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1674 rdev->sb_page);
1675 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1676 bdevname(rdev->bdev,b),
1677 (unsigned long long)rdev->sb_offset);
42543769 1678 rdev->sb_events = mddev->events;
7bfa19f2 1679
1da177e4
LT		 1680 } else
1681 dprintk(")\n");
7bfa19f2 1682 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1683 /* only need to write one superblock... */
1684 break;
1685 }
a9701a30 1686 md_super_wait(mddev);
7bfa19f2
N		 1687 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1688
a9701a30 1689 spin_lock_irq(&mddev->write_lock);
7bfa19f2 1690 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
06d91a5f 1691 /* have to write it out again */
a9701a30 1692 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1693 goto repeat;
1694 }
1695 mddev->sb_dirty = 0;
a9701a30 1696 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1697 wake_up(&mddev->sb_wait);
06d91a5f 1698
1da177e4 1699}
f6705578 1700EXPORT_SYMBOL_GPL(md_update_sb);
1da177e4 1701
bce74dac
N		 1702/* words written to sysfs files may, or my not, be \n terminated.
1703 * We want to accept with case. For this we use cmd_match.
1704 */
1705static int cmd_match(const char *cmd, const char *str)
1706{
1707 /* See if cmd, written into a sysfs file, matches
1708 * str. They must either be the same, or cmd can
1709 * have a trailing newline
1710 */
1711 while (*cmd && *str && *cmd == *str) {
1712 cmd++;
1713 str++;
1714 }
1715 if (*cmd == '\n')
1716 cmd++;
1717 if (*str || *cmd)
1718 return 0;
1719 return 1;
1720}
1721
86e6ffdd
N		 1722struct rdev_sysfs_entry {
1723 struct attribute attr;
1724 ssize_t (*show)(mdk_rdev_t *, char *);
1725 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1726};
1727
1728static ssize_t
96de1e66 1729state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1730{
1731 char *sep = "";
1732 int len=0;
1733
b2d444d7 1734 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1735 len+= sprintf(page+len, "%sfaulty",sep);
1736 sep = ",";
1737 }
b2d444d7 1738 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1739 len += sprintf(page+len, "%sin_sync",sep);
1740 sep = ",";
1741 }
f655675b
N		 1742 if (test_bit(WriteMostly, &rdev->flags)) {
1743 len += sprintf(page+len, "%swrite_mostly",sep);
1744 sep = ",";
1745 }
b2d444d7
N		 1746 if (!test_bit(Faulty, &rdev->flags) &&
1747 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1748 len += sprintf(page+len, "%sspare", sep);
1749 sep = ",";
1750 }
1751 return len+sprintf(page+len, "\n");
1752}
1753
45dc2de1
N		 1754static ssize_t
1755state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1756{
1757 /* can write
1758 * faulty - simulates and error
1759 * remove - disconnects the device
f655675b
N		 1760 * writemostly - sets write_mostly
1761 * -writemostly - clears write_mostly
45dc2de1
N		 1762 */
1763 int err = -EINVAL;
1764 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1765 md_error(rdev->mddev, rdev);
1766 err = 0;
1767 } else if (cmd_match(buf, "remove")) {
1768 if (rdev->raid_disk >= 0)
1769 err = -EBUSY;
1770 else {
1771 mddev_t *mddev = rdev->mddev;
1772 kick_rdev_from_array(rdev);
1773 md_update_sb(mddev);
1774 md_new_event(mddev);
1775 err = 0;
1776 }
f655675b
N		 1777 } else if (cmd_match(buf, "writemostly")) {
1778 set_bit(WriteMostly, &rdev->flags);
1779 err = 0;
1780 } else if (cmd_match(buf, "-writemostly")) {
1781 clear_bit(WriteMostly, &rdev->flags);
1782 err = 0;
45dc2de1
N		 1783 }
1784 return err ? err : len;
1785}
80ca3a44
N		 1786static struct rdev_sysfs_entry rdev_state =
1787__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd
N		 1788
1789static ssize_t
96de1e66 1790super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1791{
1792 if (rdev->sb_loaded && rdev->sb_size) {
1793 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1794 return rdev->sb_size;
1795 } else
1796 return 0;
1797}
96de1e66
N		 1798static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1799
4dbcdc75
N		 1800static ssize_t
1801errors_show(mdk_rdev_t *rdev, char *page)
1802{
1803 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1804}
1805
1806static ssize_t
1807errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1808{
1809 char *e;
1810 unsigned long n = simple_strtoul(buf, &e, 10);
1811 if (*buf && (*e == 0 || *e == '\n')) {
1812 atomic_set(&rdev->corrected_errors, n);
1813 return len;
1814 }
1815 return -EINVAL;
1816}
1817static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1818__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1819
014236d2
N		 1820static ssize_t
1821slot_show(mdk_rdev_t *rdev, char *page)
1822{
1823 if (rdev->raid_disk < 0)
1824 return sprintf(page, "none\n");
1825 else
1826 return sprintf(page, "%d\n", rdev->raid_disk);
1827}
1828
1829static ssize_t
1830slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1831{
1832 char *e;
1833 int slot = simple_strtoul(buf, &e, 10);
1834 if (strncmp(buf, "none", 4)==0)
1835 slot = -1;
1836 else if (e==buf || (*e && *e!= '\n'))
1837 return -EINVAL;
1838 if (rdev->mddev->pers)
1839 /* Cannot set slot in active array (yet) */
1840 return -EBUSY;
1841 if (slot >= rdev->mddev->raid_disks)
1842 return -ENOSPC;
1843 rdev->raid_disk = slot;
1844 /* assume it is working */
1845 rdev->flags = 0;
1846 set_bit(In_sync, &rdev->flags);
1847 return len;
1848}
1849
1850
1851static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1852__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1853
93c8cad0
N		 1854static ssize_t
1855offset_show(mdk_rdev_t *rdev, char *page)
1856{
6961ece4 1857 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1858}
1859
1860static ssize_t
1861offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1862{
1863 char *e;
1864 unsigned long long offset = simple_strtoull(buf, &e, 10);
1865 if (e==buf || (*e && *e != '\n'))
1866 return -EINVAL;
1867 if (rdev->mddev->pers)
1868 return -EBUSY;
1869 rdev->data_offset = offset;
1870 return len;
1871}
1872
1873static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1874__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1875
83303b61
N		 1876static ssize_t
1877rdev_size_show(mdk_rdev_t *rdev, char *page)
1878{
1879 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1880}
1881
1882static ssize_t
1883rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1884{
1885 char *e;
1886 unsigned long long size = simple_strtoull(buf, &e, 10);
1887 if (e==buf || (*e && *e != '\n'))
1888 return -EINVAL;
1889 if (rdev->mddev->pers)
1890 return -EBUSY;
1891 rdev->size = size;
1892 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1893 rdev->mddev->size = size;
1894 return len;
1895}
1896
1897static struct rdev_sysfs_entry rdev_size =
80ca3a44 1898__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 1899
86e6ffdd
N		 1900static struct attribute *rdev_default_attrs[] = {
1901 &rdev_state.attr,
1902 &rdev_super.attr,
4dbcdc75 1903 &rdev_errors.attr,
014236d2 1904 &rdev_slot.attr,
93c8cad0 1905 &rdev_offset.attr,
83303b61 1906 &rdev_size.attr,
86e6ffdd
N		 1907 NULL,
1908};
1909static ssize_t
1910rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1911{
1912 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1913 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1914
1915 if (!entry->show)
1916 return -EIO;
1917 return entry->show(rdev, page);
1918}
1919
1920static ssize_t
1921rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1922 const char *page, size_t length)
1923{
1924 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1925 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1926
1927 if (!entry->store)
1928 return -EIO;
1929 return entry->store(rdev, page, length);
1930}
1931
1932static void rdev_free(struct kobject *ko)
1933{
1934 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1935 kfree(rdev);
1936}
1937static struct sysfs_ops rdev_sysfs_ops = {
1938 .show = rdev_attr_show,
1939 .store = rdev_attr_store,
1940};
1941static struct kobj_type rdev_ktype = {
1942 .release = rdev_free,
1943 .sysfs_ops = &rdev_sysfs_ops,
1944 .default_attrs = rdev_default_attrs,
1945};
1946
1da177e4
LT		 1947/*
1948 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1949 *
1950 * mark the device faulty if:
1951 *
1952 * - the device is nonexistent (zero size)
1953 * - the device has no valid superblock
1954 *
1955 * a faulty rdev _never_ has rdev->sb set.
1956 */
1957static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1958{
1959 char b[BDEVNAME_SIZE];
1960 int err;
1961 mdk_rdev_t *rdev;
1962 sector_t size;
1963
9ffae0cf 1964 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 1965 if (!rdev) {
1966 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1967 return ERR_PTR(-ENOMEM);
1968 }
1da177e4
LT		 1969
1970 if ((err = alloc_disk_sb(rdev)))
1971 goto abort_free;
1972
1973 err = lock_rdev(rdev, newdev);
1974 if (err)
1975 goto abort_free;
1976
86e6ffdd
N		 1977 rdev->kobj.parent = NULL;
1978 rdev->kobj.ktype = &rdev_ktype;
1979 kobject_init(&rdev->kobj);
1980
1da177e4 1981 rdev->desc_nr = -1;
b2d444d7 1982 rdev->flags = 0;
1da177e4 1983 rdev->data_offset = 0;
42543769 1984 rdev->sb_events = 0;
1da177e4 1985 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 1986 atomic_set(&rdev->read_errors, 0);
4dbcdc75 1987 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 1988
1989 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1990 if (!size) {
1991 printk(KERN_WARNING
1992 "md: %s has zero or unknown size, marking faulty!\n",
1993 bdevname(rdev->bdev,b));
1994 err = -EINVAL;
1995 goto abort_free;
1996 }
1997
1998 if (super_format >= 0) {
1999 err = super_types[super_format].
2000 load_super(rdev, NULL, super_minor);
2001 if (err == -EINVAL) {
2002 printk(KERN_WARNING
2003 "md: %s has invalid sb, not importing!\n",
2004 bdevname(rdev->bdev,b));
2005 goto abort_free;
2006 }
2007 if (err < 0) {
2008 printk(KERN_WARNING
2009 "md: could not read %s's sb, not importing!\n",
2010 bdevname(rdev->bdev,b));
2011 goto abort_free;
2012 }
2013 }
2014 INIT_LIST_HEAD(&rdev->same_set);
2015
2016 return rdev;
2017
2018abort_free:
2019 if (rdev->sb_page) {
2020 if (rdev->bdev)
2021 unlock_rdev(rdev);
2022 free_disk_sb(rdev);
2023 }
2024 kfree(rdev);
2025 return ERR_PTR(err);
2026}
2027
2028/*
2029 * Check a full RAID array for plausibility
2030 */
2031
2032
a757e64c 2033static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2034{
2035 int i;
2036 struct list_head *tmp;
2037 mdk_rdev_t *rdev, *freshest;
2038 char b[BDEVNAME_SIZE];
2039
2040 freshest = NULL;
2041 ITERATE_RDEV(mddev,rdev,tmp)
2042 switch (super_types[mddev->major_version].
2043 load_super(rdev, freshest, mddev->minor_version)) {
2044 case 1:
2045 freshest = rdev;
2046 break;
2047 case 0:
2048 break;
2049 default:
2050 printk( KERN_ERR \
2051 "md: fatal superblock inconsistency in %s"
2052 " -- removing from array\n",
2053 bdevname(rdev->bdev,b));
2054 kick_rdev_from_array(rdev);
2055 }
2056
2057
2058 super_types[mddev->major_version].
2059 validate_super(mddev, freshest);
2060
2061 i = 0;
2062 ITERATE_RDEV(mddev,rdev,tmp) {
2063 if (rdev != freshest)
2064 if (super_types[mddev->major_version].
2065 validate_super(mddev, rdev)) {
2066 printk(KERN_WARNING "md: kicking non-fresh %s"
2067 " from array!\n",
2068 bdevname(rdev->bdev,b));
2069 kick_rdev_from_array(rdev);
2070 continue;
2071 }
2072 if (mddev->level == LEVEL_MULTIPATH) {
2073 rdev->desc_nr = i++;
2074 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2075 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 2076 }
2077 }
2078
2079
2080
2081 if (mddev->recovery_cp != MaxSector &&
2082 mddev->level >= 1)
2083 printk(KERN_ERR "md: %s: raid array is not clean"
2084 " -- starting background reconstruction\n",
2085 mdname(mddev));
2086
1da177e4
LT		 2087}
2088
16f17b39
N		 2089static ssize_t
2090safe_delay_show(mddev_t *mddev, char *page)
2091{
2092 int msec = (mddev->safemode_delay*1000)/HZ;
2093 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2094}
2095static ssize_t
2096safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2097{
2098 int scale=1;
2099 int dot=0;
2100 int i;
2101 unsigned long msec;
2102 char buf[30];
2103 char *e;
2104 /* remove a period, and count digits after it */
2105 if (len >= sizeof(buf))
2106 return -EINVAL;
2107 strlcpy(buf, cbuf, len);
2108 buf[len] = 0;
2109 for (i=0; i<len; i++) {
2110 if (dot) {
2111 if (isdigit(buf[i])) {
2112 buf[i-1] = buf[i];
2113 scale *= 10;
2114 }
2115 buf[i] = 0;
2116 } else if (buf[i] == '.') {
2117 dot=1;
2118 buf[i] = 0;
2119 }
2120 }
2121 msec = simple_strtoul(buf, &e, 10);
2122 if (e == buf || (*e && *e != '\n'))
2123 return -EINVAL;
2124 msec = (msec * 1000) / scale;
2125 if (msec == 0)
2126 mddev->safemode_delay = 0;
2127 else {
2128 mddev->safemode_delay = (msec*HZ)/1000;
2129 if (mddev->safemode_delay == 0)
2130 mddev->safemode_delay = 1;
2131 }
2132 return len;
2133}
2134static struct md_sysfs_entry md_safe_delay =
80ca3a44 2135__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2136
eae1701f 2137static ssize_t
96de1e66 2138level_show(mddev_t *mddev, char *page)
eae1701f 2139{
2604b703 2140 struct mdk_personality *p = mddev->pers;
d9d166c2 2141 if (p)
eae1701f 2142 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2143 else if (mddev->clevel[0])
2144 return sprintf(page, "%s\n", mddev->clevel);
2145 else if (mddev->level != LEVEL_NONE)
2146 return sprintf(page, "%d\n", mddev->level);
2147 else
2148 return 0;
eae1701f
N		 2149}
2150
d9d166c2
N		 2151static ssize_t
2152level_store(mddev_t *mddev, const char *buf, size_t len)
2153{
2154 int rv = len;
2155 if (mddev->pers)
2156 return -EBUSY;
2157 if (len == 0)
2158 return 0;
2159 if (len >= sizeof(mddev->clevel))
2160 return -ENOSPC;
2161 strncpy(mddev->clevel, buf, len);
2162 if (mddev->clevel[len-1] == '\n')
2163 len--;
2164 mddev->clevel[len] = 0;
2165 mddev->level = LEVEL_NONE;
2166 return rv;
2167}
2168
2169static struct md_sysfs_entry md_level =
80ca3a44 2170__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2171
d4dbd025
N		 2172
2173static ssize_t
2174layout_show(mddev_t *mddev, char *page)
2175{
2176 /* just a number, not meaningful for all levels */
2177 return sprintf(page, "%d\n", mddev->layout);
2178}
2179
2180static ssize_t
2181layout_store(mddev_t *mddev, const char *buf, size_t len)
2182{
2183 char *e;
2184 unsigned long n = simple_strtoul(buf, &e, 10);
2185 if (mddev->pers)
2186 return -EBUSY;
2187
2188 if (!*buf || (*e && *e != '\n'))
2189 return -EINVAL;
2190
2191 mddev->layout = n;
2192 return len;
2193}
2194static struct md_sysfs_entry md_layout =
80ca3a44 2195__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2196
2197
eae1701f 2198static ssize_t
96de1e66 2199raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2200{
bb636547
N		 2201 if (mddev->raid_disks == 0)
2202 return 0;
eae1701f
N		 2203 return sprintf(page, "%d\n", mddev->raid_disks);
2204}
2205
da943b99
N		 2206static int update_raid_disks(mddev_t *mddev, int raid_disks);
2207
2208static ssize_t
2209raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2210{
2211 /* can only set raid_disks if array is not yet active */
2212 char *e;
2213 int rv = 0;
2214 unsigned long n = simple_strtoul(buf, &e, 10);
2215
2216 if (!*buf || (*e && *e != '\n'))
2217 return -EINVAL;
2218
2219 if (mddev->pers)
2220 rv = update_raid_disks(mddev, n);
2221 else
2222 mddev->raid_disks = n;
2223 return rv ? rv : len;
2224}
2225static struct md_sysfs_entry md_raid_disks =
80ca3a44 2226__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2227
3b34380a
N		 2228static ssize_t
2229chunk_size_show(mddev_t *mddev, char *page)
2230{
2231 return sprintf(page, "%d\n", mddev->chunk_size);
2232}
2233
2234static ssize_t
2235chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2236{
2237 /* can only set chunk_size if array is not yet active */
2238 char *e;
2239 unsigned long n = simple_strtoul(buf, &e, 10);
2240
2241 if (mddev->pers)
2242 return -EBUSY;
2243 if (!*buf || (*e && *e != '\n'))
2244 return -EINVAL;
2245
2246 mddev->chunk_size = n;
2247 return len;
2248}
2249static struct md_sysfs_entry md_chunk_size =
80ca3a44 2250__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2251
a94213b1
N		 2252static ssize_t
2253resync_start_show(mddev_t *mddev, char *page)
2254{
2255 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2256}
2257
2258static ssize_t
2259resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2260{
2261 /* can only set chunk_size if array is not yet active */
2262 char *e;
2263 unsigned long long n = simple_strtoull(buf, &e, 10);
2264
2265 if (mddev->pers)
2266 return -EBUSY;
2267 if (!*buf || (*e && *e != '\n'))
2268 return -EINVAL;
2269
2270 mddev->recovery_cp = n;
2271 return len;
2272}
2273static struct md_sysfs_entry md_resync_start =
80ca3a44 2274__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2275
9e653b63
N		 2276/*
2277 * The array state can be:
2278 *
2279 * clear
2280 * No devices, no size, no level
2281 * Equivalent to STOP_ARRAY ioctl
2282 * inactive
2283 * May have some settings, but array is not active
2284 * all IO results in error
2285 * When written, doesn't tear down array, but just stops it
2286 * suspended (not supported yet)
2287 * All IO requests will block. The array can be reconfigured.
2288 * Writing this, if accepted, will block until array is quiessent
2289 * readonly
2290 * no resync can happen. no superblocks get written.
2291 * write requests fail
2292 * read-auto
2293 * like readonly, but behaves like 'clean' on a write request.
2294 *
2295 * clean - no pending writes, but otherwise active.
2296 * When written to inactive array, starts without resync
2297 * If a write request arrives then
2298 * if metadata is known, mark 'dirty' and switch to 'active'.
2299 * if not known, block and switch to write-pending
2300 * If written to an active array that has pending writes, then fails.
2301 * active
2302 * fully active: IO and resync can be happening.
2303 * When written to inactive array, starts with resync
2304 *
2305 * write-pending
2306 * clean, but writes are blocked waiting for 'active' to be written.
2307 *
2308 * active-idle
2309 * like active, but no writes have been seen for a while (100msec).
2310 *
2311 */
2312enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2313 write_pending, active_idle, bad_word};
05381954 2314static char *array_states[] = {
9e653b63
N		 2315 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2316 "write-pending", "active-idle", NULL };
2317
2318static int match_word(const char *word, char **list)
2319{
2320 int n;
2321 for (n=0; list[n]; n++)
2322 if (cmd_match(word, list[n]))
2323 break;
2324 return n;
2325}
2326
2327static ssize_t
2328array_state_show(mddev_t *mddev, char *page)
2329{
2330 enum array_state st = inactive;
2331
2332 if (mddev->pers)
2333 switch(mddev->ro) {
2334 case 1:
2335 st = readonly;
2336 break;
2337 case 2:
2338 st = read_auto;
2339 break;
2340 case 0:
2341 if (mddev->in_sync)
2342 st = clean;
2343 else if (mddev->safemode)
2344 st = active_idle;
2345 else
2346 st = active;
2347 }
2348 else {
2349 if (list_empty(&mddev->disks) &&
2350 mddev->raid_disks == 0 &&
2351 mddev->size == 0)
2352 st = clear;
2353 else
2354 st = inactive;
2355 }
2356 return sprintf(page, "%s\n", array_states[st]);
2357}
2358
2359static int do_md_stop(mddev_t * mddev, int ro);
2360static int do_md_run(mddev_t * mddev);
2361static int restart_array(mddev_t *mddev);
2362
2363static ssize_t
2364array_state_store(mddev_t *mddev, const char *buf, size_t len)
2365{
2366 int err = -EINVAL;
2367 enum array_state st = match_word(buf, array_states);
2368 switch(st) {
2369 case bad_word:
2370 break;
2371 case clear:
2372 /* stopping an active array */
2373 if (mddev->pers) {
2374 if (atomic_read(&mddev->active) > 1)
2375 return -EBUSY;
2376 err = do_md_stop(mddev, 0);
2377 }
2378 break;
2379 case inactive:
2380 /* stopping an active array */
2381 if (mddev->pers) {
2382 if (atomic_read(&mddev->active) > 1)
2383 return -EBUSY;
2384 err = do_md_stop(mddev, 2);
2385 }
2386 break;
2387 case suspended:
2388 break; /* not supported yet */
2389 case readonly:
2390 if (mddev->pers)
2391 err = do_md_stop(mddev, 1);
2392 else {
2393 mddev->ro = 1;
2394 err = do_md_run(mddev);
2395 }
2396 break;
2397 case read_auto:
2398 /* stopping an active array */
2399 if (mddev->pers) {
2400 err = do_md_stop(mddev, 1);
2401 if (err == 0)
2402 mddev->ro = 2; /* FIXME mark devices writable */
2403 } else {
2404 mddev->ro = 2;
2405 err = do_md_run(mddev);
2406 }
2407 break;
2408 case clean:
2409 if (mddev->pers) {
2410 restart_array(mddev);
2411 spin_lock_irq(&mddev->write_lock);
2412 if (atomic_read(&mddev->writes_pending) == 0) {
2413 mddev->in_sync = 1;
2414 mddev->sb_dirty = 1;
2415 }
2416 spin_unlock_irq(&mddev->write_lock);
2417 } else {
2418 mddev->ro = 0;
2419 mddev->recovery_cp = MaxSector;
2420 err = do_md_run(mddev);
2421 }
2422 break;
2423 case active:
2424 if (mddev->pers) {
2425 restart_array(mddev);
2426 mddev->sb_dirty = 0;
2427 wake_up(&mddev->sb_wait);
2428 err = 0;
2429 } else {
2430 mddev->ro = 0;
2431 err = do_md_run(mddev);
2432 }
2433 break;
2434 case write_pending:
2435 case active_idle:
2436 /* these cannot be set */
2437 break;
2438 }
2439 if (err)
2440 return err;
2441 else
2442 return len;
2443}
80ca3a44
N		 2444static struct md_sysfs_entry md_array_state =
2445__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2446
6d7ff738
N		 2447static ssize_t
2448null_show(mddev_t *mddev, char *page)
2449{
2450 return -EINVAL;
2451}
2452
2453static ssize_t
2454new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2455{
2456 /* buf must be %d:%d\n? giving major and minor numbers */
2457 /* The new device is added to the array.
2458 * If the array has a persistent superblock, we read the
2459 * superblock to initialise info and check validity.
2460 * Otherwise, only checking done is that in bind_rdev_to_array,
2461 * which mainly checks size.
2462 */
2463 char *e;
2464 int major = simple_strtoul(buf, &e, 10);
2465 int minor;
2466 dev_t dev;
2467 mdk_rdev_t *rdev;
2468 int err;
2469
2470 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2471 return -EINVAL;
2472 minor = simple_strtoul(e+1, &e, 10);
2473 if (*e && *e != '\n')
2474 return -EINVAL;
2475 dev = MKDEV(major, minor);
2476 if (major != MAJOR(dev) ||
2477 minor != MINOR(dev))
2478 return -EOVERFLOW;
2479
2480
2481 if (mddev->persistent) {
2482 rdev = md_import_device(dev, mddev->major_version,
2483 mddev->minor_version);
2484 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2485 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2486 mdk_rdev_t, same_set);
2487 err = super_types[mddev->major_version]
2488 .load_super(rdev, rdev0, mddev->minor_version);
2489 if (err < 0)
2490 goto out;
2491 }
2492 } else
2493 rdev = md_import_device(dev, -1, -1);
2494
2495 if (IS_ERR(rdev))
2496 return PTR_ERR(rdev);
2497 err = bind_rdev_to_array(rdev, mddev);
2498 out:
2499 if (err)
2500 export_rdev(rdev);
2501 return err ? err : len;
2502}
2503
2504static struct md_sysfs_entry md_new_device =
80ca3a44 2505__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2506
a35b0d69
N		 2507static ssize_t
2508size_show(mddev_t *mddev, char *page)
2509{
2510 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2511}
2512
2513static int update_size(mddev_t *mddev, unsigned long size);
2514
2515static ssize_t
2516size_store(mddev_t *mddev, const char *buf, size_t len)
2517{
2518 /* If array is inactive, we can reduce the component size, but
2519 * not increase it (except from 0).
2520 * If array is active, we can try an on-line resize
2521 */
2522 char *e;
2523 int err = 0;
2524 unsigned long long size = simple_strtoull(buf, &e, 10);
2525 if (!*buf || *buf == '\n' ||
2526 (*e && *e != '\n'))
2527 return -EINVAL;
2528
2529 if (mddev->pers) {
2530 err = update_size(mddev, size);
2531 md_update_sb(mddev);
2532 } else {
2533 if (mddev->size == 0 ||
2534 mddev->size > size)
2535 mddev->size = size;
2536 else
2537 err = -ENOSPC;
2538 }
2539 return err ? err : len;
2540}
2541
2542static struct md_sysfs_entry md_size =
80ca3a44 2543__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2544
8bb93aac
N		 2545
2546/* Metdata version.
2547 * This is either 'none' for arrays with externally managed metadata,
2548 * or N.M for internally known formats
2549 */
2550static ssize_t
2551metadata_show(mddev_t *mddev, char *page)
2552{
2553 if (mddev->persistent)
2554 return sprintf(page, "%d.%d\n",
2555 mddev->major_version, mddev->minor_version);
2556 else
2557 return sprintf(page, "none\n");
2558}
2559
2560static ssize_t
2561metadata_store(mddev_t *mddev, const char *buf, size_t len)
2562{
2563 int major, minor;
2564 char *e;
2565 if (!list_empty(&mddev->disks))
2566 return -EBUSY;
2567
2568 if (cmd_match(buf, "none")) {
2569 mddev->persistent = 0;
2570 mddev->major_version = 0;
2571 mddev->minor_version = 90;
2572 return len;
2573 }
2574 major = simple_strtoul(buf, &e, 10);
2575 if (e==buf || *e != '.')
2576 return -EINVAL;
2577 buf = e+1;
2578 minor = simple_strtoul(buf, &e, 10);
2579 if (e==buf || *e != '\n')
2580 return -EINVAL;
2581 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2582 super_types[major].name == NULL)
2583 return -ENOENT;
2584 mddev->major_version = major;
2585 mddev->minor_version = minor;
2586 mddev->persistent = 1;
2587 return len;
2588}
2589
2590static struct md_sysfs_entry md_metadata =
80ca3a44 2591__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2592
24dd469d 2593static ssize_t
7eec314d 2594action_show(mddev_t *mddev, char *page)
24dd469d 2595{
7eec314d 2596 char *type = "idle";
31399d9e
N		 2597 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2598 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
ccfcc3c1
N		 2599 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2600 type = "reshape";
2601 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2602 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2603 type = "resync";
2604 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2605 type = "check";
2606 else
2607 type = "repair";
2608 } else
2609 type = "recover";
2610 }
2611 return sprintf(page, "%s\n", type);
2612}
2613
2614static ssize_t
7eec314d 2615action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2616{
7eec314d
N		 2617 if (!mddev->pers || !mddev->pers->sync_request)
2618 return -EINVAL;
2619
bce74dac 2620 if (cmd_match(page, "idle")) {
7eec314d
N		 2621 if (mddev->sync_thread) {
2622 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2623 md_unregister_thread(mddev->sync_thread);
2624 mddev->sync_thread = NULL;
2625 mddev->recovery = 0;
2626 }
03c902e1
N		 2627 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2628 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2629 return -EBUSY;
03c902e1 2630 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2631 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2632 else if (cmd_match(page, "reshape")) {
2633 int err;
2634 if (mddev->pers->start_reshape == NULL)
2635 return -EINVAL;
2636 err = mddev->pers->start_reshape(mddev);
2637 if (err)
2638 return err;
2639 } else {
bce74dac 2640 if (cmd_match(page, "check"))
7eec314d 2641 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2642 else if (!cmd_match(page, "repair"))
7eec314d
N		 2643 return -EINVAL;
2644 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2645 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2646 }
03c902e1 2647 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2648 md_wakeup_thread(mddev->thread);
2649 return len;
2650}
2651
9d88883e 2652static ssize_t
96de1e66 2653mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2654{
2655 return sprintf(page, "%llu\n",
2656 (unsigned long long) mddev->resync_mismatches);
2657}
2658
80ca3a44
N		 2659static struct md_sysfs_entry md_scan_mode =
2660__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2661
96de1e66 2662
80ca3a44 2663static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2664
88202a0c
N		 2665static ssize_t
2666sync_min_show(mddev_t *mddev, char *page)
2667{
2668 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2669 mddev->sync_speed_min ? "local": "system");
2670}
2671
2672static ssize_t
2673sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2674{
2675 int min;
2676 char *e;
2677 if (strncmp(buf, "system", 6)==0) {
2678 mddev->sync_speed_min = 0;
2679 return len;
2680 }
2681 min = simple_strtoul(buf, &e, 10);
2682 if (buf == e || (*e && *e != '\n') || min <= 0)
2683 return -EINVAL;
2684 mddev->sync_speed_min = min;
2685 return len;
2686}
2687
2688static struct md_sysfs_entry md_sync_min =
2689__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2690
2691static ssize_t
2692sync_max_show(mddev_t *mddev, char *page)
2693{
2694 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2695 mddev->sync_speed_max ? "local": "system");
2696}
2697
2698static ssize_t
2699sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2700{
2701 int max;
2702 char *e;
2703 if (strncmp(buf, "system", 6)==0) {
2704 mddev->sync_speed_max = 0;
2705 return len;
2706 }
2707 max = simple_strtoul(buf, &e, 10);
2708 if (buf == e || (*e && *e != '\n') || max <= 0)
2709 return -EINVAL;
2710 mddev->sync_speed_max = max;
2711 return len;
2712}
2713
2714static struct md_sysfs_entry md_sync_max =
2715__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2716
2717
2718static ssize_t
2719sync_speed_show(mddev_t *mddev, char *page)
2720{
2721 unsigned long resync, dt, db;
ff4e8d9a 2722 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2723 dt = ((jiffies - mddev->resync_mark) / HZ);
2724 if (!dt) dt++;
2725 db = resync - (mddev->resync_mark_cnt);
2726 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2727}
2728
80ca3a44 2729static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2730
2731static ssize_t
2732sync_completed_show(mddev_t *mddev, char *page)
2733{
2734 unsigned long max_blocks, resync;
2735
2736 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2737 max_blocks = mddev->resync_max_sectors;
2738 else
2739 max_blocks = mddev->size << 1;
2740
2741 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2742 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2743}
2744
80ca3a44 2745static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 2746
e464eafd
N		 2747static ssize_t
2748suspend_lo_show(mddev_t *mddev, char *page)
2749{
2750 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2751}
2752
2753static ssize_t
2754suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2755{
2756 char *e;
2757 unsigned long long new = simple_strtoull(buf, &e, 10);
2758
2759 if (mddev->pers->quiesce == NULL)
2760 return -EINVAL;
2761 if (buf == e || (*e && *e != '\n'))
2762 return -EINVAL;
2763 if (new >= mddev->suspend_hi ||
2764 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2765 mddev->suspend_lo = new;
2766 mddev->pers->quiesce(mddev, 2);
2767 return len;
2768 } else
2769 return -EINVAL;
2770}
2771static struct md_sysfs_entry md_suspend_lo =
2772__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2773
2774
2775static ssize_t
2776suspend_hi_show(mddev_t *mddev, char *page)
2777{
2778 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2779}
2780
2781static ssize_t
2782suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2783{
2784 char *e;
2785 unsigned long long new = simple_strtoull(buf, &e, 10);
2786
2787 if (mddev->pers->quiesce == NULL)
2788 return -EINVAL;
2789 if (buf == e || (*e && *e != '\n'))
2790 return -EINVAL;
2791 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2792 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2793 mddev->suspend_hi = new;
2794 mddev->pers->quiesce(mddev, 1);
2795 mddev->pers->quiesce(mddev, 0);
2796 return len;
2797 } else
2798 return -EINVAL;
2799}
2800static struct md_sysfs_entry md_suspend_hi =
2801__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2802
2803
eae1701f
N		 2804static struct attribute *md_default_attrs[] = {
2805 &md_level.attr,
d4dbd025 2806 &md_layout.attr,
eae1701f 2807 &md_raid_disks.attr,
3b34380a 2808 &md_chunk_size.attr,
a35b0d69 2809 &md_size.attr,
a94213b1 2810 &md_resync_start.attr,
8bb93aac 2811 &md_metadata.attr,
6d7ff738 2812 &md_new_device.attr,
16f17b39 2813 &md_safe_delay.attr,
9e653b63 2814 &md_array_state.attr,
411036fa
N		 2815 NULL,
2816};
2817
2818static struct attribute *md_redundancy_attrs[] = {
24dd469d 2819 &md_scan_mode.attr,
9d88883e 2820 &md_mismatches.attr,
88202a0c
N		 2821 &md_sync_min.attr,
2822 &md_sync_max.attr,
2823 &md_sync_speed.attr,
2824 &md_sync_completed.attr,
e464eafd
N		 2825 &md_suspend_lo.attr,
2826 &md_suspend_hi.attr,
eae1701f
N		 2827 NULL,
2828};
411036fa
N		 2829static struct attribute_group md_redundancy_group = {
2830 .name = NULL,
2831 .attrs = md_redundancy_attrs,
2832};
2833
eae1701f
N		 2834
2835static ssize_t
2836md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2837{
2838 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2839 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2840 ssize_t rv;
eae1701f
N		 2841
2842 if (!entry->show)
2843 return -EIO;
5dc5cf7d
IM		 2844 rv = mddev_lock(mddev);
2845 if (!rv) {
2846 rv = entry->show(mddev, page);
2847 mddev_unlock(mddev);
2848 }
96de1e66 2849 return rv;
eae1701f
N		 2850}
2851
2852static ssize_t
2853md_attr_store(struct kobject *kobj, struct attribute *attr,
2854 const char *page, size_t length)
2855{
2856 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2857 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2858 ssize_t rv;
eae1701f
N		 2859
2860 if (!entry->store)
2861 return -EIO;
5dc5cf7d
IM		 2862 rv = mddev_lock(mddev);
2863 if (!rv) {
2864 rv = entry->store(mddev, page, length);
2865 mddev_unlock(mddev);
2866 }
96de1e66 2867 return rv;
eae1701f
N		 2868}
2869
2870static void md_free(struct kobject *ko)
2871{
2872 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2873 kfree(mddev);
2874}
2875
2876static struct sysfs_ops md_sysfs_ops = {
2877 .show = md_attr_show,
2878 .store = md_attr_store,
2879};
2880static struct kobj_type md_ktype = {
2881 .release = md_free,
2882 .sysfs_ops = &md_sysfs_ops,
2883 .default_attrs = md_default_attrs,
2884};
2885
1da177e4
LT		 2886int mdp_major = 0;
2887
2888static struct kobject *md_probe(dev_t dev, int *part, void *data)
2889{
48c9c27b 2890 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 2891 mddev_t *mddev = mddev_find(dev);
2892 struct gendisk *disk;
2893 int partitioned = (MAJOR(dev) != MD_MAJOR);
2894 int shift = partitioned ? MdpMinorShift : 0;
2895 int unit = MINOR(dev) >> shift;
2896
2897 if (!mddev)
2898 return NULL;
2899
48c9c27b 2900 mutex_lock(&disks_mutex);
1da177e4 2901 if (mddev->gendisk) {
48c9c27b 2902 mutex_unlock(&disks_mutex);
1da177e4
LT		 2903 mddev_put(mddev);
2904 return NULL;
2905 }
2906 disk = alloc_disk(1 << shift);
2907 if (!disk) {
48c9c27b 2908 mutex_unlock(&disks_mutex);
1da177e4
LT		 2909 mddev_put(mddev);
2910 return NULL;
2911 }
2912 disk->major = MAJOR(dev);
2913 disk->first_minor = unit << shift;
ce7b0f46 2914 if (partitioned)
1da177e4 2915 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 2916 else
1da177e4 2917 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 2918 disk->fops = &md_fops;
2919 disk->private_data = mddev;
2920 disk->queue = mddev->queue;
2921 add_disk(disk);
2922 mddev->gendisk = disk;
48c9c27b 2923 mutex_unlock(&disks_mutex);
9c791977 2924 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 2925 mddev->kobj.k_name = NULL;
2926 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2927 mddev->kobj.ktype = &md_ktype;
2928 kobject_register(&mddev->kobj);
1da177e4
LT		 2929 return NULL;
2930}
2931
1da177e4
LT		 2932static void md_safemode_timeout(unsigned long data)
2933{
2934 mddev_t *mddev = (mddev_t *) data;
2935
2936 mddev->safemode = 1;
2937 md_wakeup_thread(mddev->thread);
2938}
2939
6ff8d8ec 2940static int start_dirty_degraded;
1da177e4
LT		 2941
2942static int do_md_run(mddev_t * mddev)
2943{
2604b703 2944 int err;
1da177e4
LT		 2945 int chunk_size;
2946 struct list_head *tmp;
2947 mdk_rdev_t *rdev;
2948 struct gendisk *disk;
2604b703 2949 struct mdk_personality *pers;
1da177e4
LT		 2950 char b[BDEVNAME_SIZE];
2951
a757e64c
N		 2952 if (list_empty(&mddev->disks))
2953 /* cannot run an array with no devices.. */
1da177e4 2954 return -EINVAL;
1da177e4
LT		 2955
2956 if (mddev->pers)
2957 return -EBUSY;
2958
2959 /*
2960 * Analyze all RAID superblock(s)
2961 */
a757e64c
N		 2962 if (!mddev->raid_disks)
2963 analyze_sbs(mddev);
1da177e4
LT		 2964
2965 chunk_size = mddev->chunk_size;
2604b703
N		 2966
2967 if (chunk_size) {
1da177e4
LT		 2968 if (chunk_size > MAX_CHUNK_SIZE) {
2969 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2970 chunk_size, MAX_CHUNK_SIZE);
2971 return -EINVAL;
2972 }
2973 /*
2974 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2975 */
2976 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 2977 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 2978 return -EINVAL;
2979 }
2980 if (chunk_size < PAGE_SIZE) {
2981 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2982 chunk_size, PAGE_SIZE);
2983 return -EINVAL;
2984 }
2985
2986 /* devices must have minimum size of one chunk */
2987 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2988 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2989 continue;
2990 if (rdev->size < chunk_size / 1024) {
2991 printk(KERN_WARNING
2992 "md: Dev %s smaller than chunk_size:"
2993 " %lluk < %dk\n",
2994 bdevname(rdev->bdev,b),
2995 (unsigned long long)rdev->size,
2996 chunk_size / 1024);
2997 return -EINVAL;
2998 }
2999 }
3000 }
3001
1da177e4 3002#ifdef CONFIG_KMOD
d9d166c2
N		 3003 if (mddev->level != LEVEL_NONE)
3004 request_module("md-level-%d", mddev->level);
3005 else if (mddev->clevel[0])
3006 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3007#endif
3008
3009 /*
3010 * Drop all container device buffers, from now on
3011 * the only valid external interface is through the md
3012 * device.
3013 * Also find largest hardsector size
3014 */
3015 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3016 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3017 continue;
3018 sync_blockdev(rdev->bdev);
3019 invalidate_bdev(rdev->bdev, 0);
3020 }
3021
3022 md_probe(mddev->unit, NULL, NULL);
3023 disk = mddev->gendisk;
3024 if (!disk)
3025 return -ENOMEM;
3026
3027 spin_lock(&pers_lock);
d9d166c2 3028 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3029 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3030 spin_unlock(&pers_lock);
d9d166c2
N		 3031 if (mddev->level != LEVEL_NONE)
3032 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3033 mddev->level);
3034 else
3035 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3036 mddev->clevel);
1da177e4
LT		 3037 return -EINVAL;
3038 }
2604b703 3039 mddev->pers = pers;
1da177e4 3040 spin_unlock(&pers_lock);
d9d166c2
N		 3041 mddev->level = pers->level;
3042 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3043
f6705578 3044 if (mddev->reshape_position != MaxSector &&
63c70c4f 3045 pers->start_reshape == NULL) {
f6705578
N		 3046 /* This personality cannot handle reshaping... */
3047 mddev->pers = NULL;
3048 module_put(pers->owner);
3049 return -EINVAL;
3050 }
3051
657390d2 3052 mddev->recovery = 0;
1da177e4 3053 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3054 mddev->barriers_work = 1;
6ff8d8ec 3055 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3056
f91de92e
N		 3057 if (start_readonly)
3058 mddev->ro = 2; /* read-only, but switch on first write */
3059
b15c2e57
N		 3060 err = mddev->pers->run(mddev);
3061 if (!err && mddev->pers->sync_request) {
3062 err = bitmap_create(mddev);
3063 if (err) {
3064 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3065 mdname(mddev), err);
3066 mddev->pers->stop(mddev);
3067 }
3068 }
1da177e4
LT		 3069 if (err) {
3070 printk(KERN_ERR "md: pers->run() failed ...\n");
3071 module_put(mddev->pers->owner);
3072 mddev->pers = NULL;
32a7627c
N		 3073 bitmap_destroy(mddev);
3074 return err;
1da177e4 3075 }
411036fa
N		 3076 if (mddev->pers->sync_request)
3077 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
fd9d49ca
N		 3078 else if (mddev->ro == 2) /* auto-readonly not meaningful */
3079 mddev->ro = 0;
3080
1da177e4
LT		 3081 atomic_set(&mddev->writes_pending,0);
3082 mddev->safemode = 0;
3083 mddev->safemode_timer.function = md_safemode_timeout;
3084 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3085 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3086 mddev->in_sync = 1;
86e6ffdd
N		 3087
3088 ITERATE_RDEV(mddev,rdev,tmp)
3089 if (rdev->raid_disk >= 0) {
3090 char nm[20];
3091 sprintf(nm, "rd%d", rdev->raid_disk);
3092 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3093 }
1da177e4
LT		 3094
3095 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3096
3097 if (mddev->sb_dirty)
3098 md_update_sb(mddev);
3099
3100 set_capacity(disk, mddev->array_size<<1);
3101
3102 /* If we call blk_queue_make_request here, it will
3103 * re-initialise max_sectors etc which may have been
3104 * refined inside -> run. So just set the bits we need to set.
3105 * Most initialisation happended when we called
3106 * blk_queue_make_request(..., md_fail_request)
3107 * earlier.
3108 */
3109 mddev->queue->queuedata = mddev;
3110 mddev->queue->make_request_fn = mddev->pers->make_request;
3111
5fd6c1dc
N		 3112 /* If there is a partially-recovered drive we need to
3113 * start recovery here. If we leave it to md_check_recovery,
3114 * it will remove the drives and not do the right thing
3115 */
0b8c9de0 3116 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3117 struct list_head *rtmp;
3118 int spares = 0;
3119 ITERATE_RDEV(mddev,rdev,rtmp)
3120 if (rdev->raid_disk >= 0 &&
3121 !test_bit(In_sync, &rdev->flags) &&
3122 !test_bit(Faulty, &rdev->flags))
3123 /* complete an interrupted recovery */
3124 spares++;
3125 if (spares && mddev->pers->sync_request) {
3126 mddev->recovery = 0;
3127 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3128 mddev->sync_thread = md_register_thread(md_do_sync,
3129 mddev,
3130 "%s_resync");
3131 if (!mddev->sync_thread) {
3132 printk(KERN_ERR "%s: could not start resync"
3133 " thread...\n",
3134 mdname(mddev));
3135 /* leave the spares where they are, it shouldn't hurt */
3136 mddev->recovery = 0;
0b8c9de0 3137 }
5fd6c1dc
N		 3138 }
3139 }
0b8c9de0
N		 3140 md_wakeup_thread(mddev->thread);
3141 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3142
1da177e4 3143 mddev->changed = 1;
d7603b7e 3144 md_new_event(mddev);
1da177e4
LT		 3145 return 0;
3146}
3147
3148static int restart_array(mddev_t *mddev)
3149{
3150 struct gendisk *disk = mddev->gendisk;
3151 int err;
3152
3153 /*
3154 * Complain if it has no devices
3155 */
3156 err = -ENXIO;
3157 if (list_empty(&mddev->disks))
3158 goto out;
3159
3160 if (mddev->pers) {
3161 err = -EBUSY;
3162 if (!mddev->ro)
3163 goto out;
3164
3165 mddev->safemode = 0;
3166 mddev->ro = 0;
3167 set_disk_ro(disk, 0);
3168
3169 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3170 mdname(mddev));
3171 /*
3172 * Kick recovery or resync if necessary
3173 */
3174 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3175 md_wakeup_thread(mddev->thread);
5fd6c1dc 3176 md_wakeup_thread(mddev->sync_thread);
1da177e4 3177 err = 0;
9e653b63 3178 } else
1da177e4 3179 err = -EINVAL;
1da177e4
LT		 3180
3181out:
3182 return err;
3183}
3184
acc55e22
N		 3185/* similar to deny_write_access, but accounts for our holding a reference
3186 * to the file ourselves */
3187static int deny_bitmap_write_access(struct file * file)
3188{
3189 struct inode *inode = file->f_mapping->host;
3190
3191 spin_lock(&inode->i_lock);
3192 if (atomic_read(&inode->i_writecount) > 1) {
3193 spin_unlock(&inode->i_lock);
3194 return -ETXTBSY;
3195 }
3196 atomic_set(&inode->i_writecount, -1);
3197 spin_unlock(&inode->i_lock);
3198
3199 return 0;
3200}
3201
3202static void restore_bitmap_write_access(struct file *file)
3203{
3204 struct inode *inode = file->f_mapping->host;
3205
3206 spin_lock(&inode->i_lock);
3207 atomic_set(&inode->i_writecount, 1);
3208 spin_unlock(&inode->i_lock);
3209}
3210
9e653b63
N		 3211/* mode:
3212 * 0 - completely stop and dis-assemble array
3213 * 1 - switch to readonly
3214 * 2 - stop but do not disassemble array
3215 */
3216static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3217{
3218 int err = 0;
3219 struct gendisk *disk = mddev->gendisk;
3220
3221 if (mddev->pers) {
3222 if (atomic_read(&mddev->active)>2) {
3223 printk("md: %s still in use.\n",mdname(mddev));
3224 return -EBUSY;
3225 }
3226
3227 if (mddev->sync_thread) {
5fd6c1dc 3228 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3229 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3230 md_unregister_thread(mddev->sync_thread);
3231 mddev->sync_thread = NULL;
3232 }
3233
3234 del_timer_sync(&mddev->safemode_timer);
3235
3236 invalidate_partition(disk, 0);
3237
9e653b63
N		 3238 switch(mode) {
3239 case 1: /* readonly */
1da177e4 3240 err = -ENXIO;
f91de92e 3241 if (mddev->ro==1)
1da177e4
LT		 3242 goto out;
3243 mddev->ro = 1;
9e653b63
N		 3244 break;
3245 case 0: /* disassemble */
3246 case 2: /* stop */
6b8b3e8a 3247 bitmap_flush(mddev);
a9701a30 3248 md_super_wait(mddev);
1da177e4
LT		 3249 if (mddev->ro)
3250 set_disk_ro(disk, 0);
3251 blk_queue_make_request(mddev->queue, md_fail_request);
3252 mddev->pers->stop(mddev);
411036fa
N		 3253 if (mddev->pers->sync_request)
3254 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3255
1da177e4
LT		 3256 module_put(mddev->pers->owner);
3257 mddev->pers = NULL;
3258 if (mddev->ro)
3259 mddev->ro = 0;
3260 }
5fd6c1dc 3261 if (!mddev->in_sync || mddev->sb_dirty) {
1da177e4
LT		 3262 /* mark array as shutdown cleanly */
3263 mddev->in_sync = 1;
3264 md_update_sb(mddev);
3265 }
9e653b63 3266 if (mode == 1)
1da177e4 3267 set_disk_ro(disk, 1);
5fd6c1dc 3268 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3269 }
32a7627c 3270
1da177e4
LT		 3271 /*
3272 * Free resources if final stop
3273 */
9e653b63 3274 if (mode == 0) {
86e6ffdd
N		 3275 mdk_rdev_t *rdev;
3276 struct list_head *tmp;
1da177e4
LT		 3277 struct gendisk *disk;
3278 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3279
978f946b
N		 3280 bitmap_destroy(mddev);
3281 if (mddev->bitmap_file) {
acc55e22 3282 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3283 fput(mddev->bitmap_file);
3284 mddev->bitmap_file = NULL;
3285 }
3286 mddev->bitmap_offset = 0;
3287
86e6ffdd
N		 3288 ITERATE_RDEV(mddev,rdev,tmp)
3289 if (rdev->raid_disk >= 0) {
3290 char nm[20];
3291 sprintf(nm, "rd%d", rdev->raid_disk);
3292 sysfs_remove_link(&mddev->kobj, nm);
3293 }
3294
1da177e4
LT		 3295 export_array(mddev);
3296
3297 mddev->array_size = 0;
9e653b63
N		 3298 mddev->size = 0;
3299 mddev->raid_disks = 0;
a94213b1 3300 mddev->recovery_cp = 0;
9e653b63 3301
1da177e4
LT		 3302 disk = mddev->gendisk;
3303 if (disk)
3304 set_capacity(disk, 0);
3305 mddev->changed = 1;
a8a55c38 3306 } else if (mddev->pers)
1da177e4
LT		 3307 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3308 mdname(mddev));
3309 err = 0;
d7603b7e 3310 md_new_event(mddev);
1da177e4
LT		 3311out:
3312 return err;
3313}
3314
3315static void autorun_array(mddev_t *mddev)
3316{
3317 mdk_rdev_t *rdev;
3318 struct list_head *tmp;
3319 int err;
3320
a757e64c 3321 if (list_empty(&mddev->disks))
1da177e4 3322 return;
1da177e4
LT		 3323
3324 printk(KERN_INFO "md: running: ");
3325
3326 ITERATE_RDEV(mddev,rdev,tmp) {
3327 char b[BDEVNAME_SIZE];
3328 printk("<%s>", bdevname(rdev->bdev,b));
3329 }
3330 printk("\n");
3331
3332 err = do_md_run (mddev);
3333 if (err) {
3334 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3335 do_md_stop (mddev, 0);
3336 }
3337}
3338
3339/*
3340 * lets try to run arrays based on all disks that have arrived
3341 * until now. (those are in pending_raid_disks)
3342 *
3343 * the method: pick the first pending disk, collect all disks with
3344 * the same UUID, remove all from the pending list and put them into
3345 * the 'same_array' list. Then order this list based on superblock
3346 * update time (freshest comes first), kick out 'old' disks and
3347 * compare superblocks. If everything's fine then run it.
3348 *
3349 * If "unit" is allocated, then bump its reference count
3350 */
3351static void autorun_devices(int part)
3352{
1da177e4
LT		 3353 struct list_head *tmp;
3354 mdk_rdev_t *rdev0, *rdev;
3355 mddev_t *mddev;
3356 char b[BDEVNAME_SIZE];
3357
3358 printk(KERN_INFO "md: autorun ...\n");
3359 while (!list_empty(&pending_raid_disks)) {
3360 dev_t dev;
ad01c9e3 3361 LIST_HEAD(candidates);
1da177e4
LT		 3362 rdev0 = list_entry(pending_raid_disks.next,
3363 mdk_rdev_t, same_set);
3364
3365 printk(KERN_INFO "md: considering %s ...\n",
3366 bdevname(rdev0->bdev,b));
3367 INIT_LIST_HEAD(&candidates);
3368 ITERATE_RDEV_PENDING(rdev,tmp)
3369 if (super_90_load(rdev, rdev0, 0) >= 0) {
3370 printk(KERN_INFO "md: adding %s ...\n",
3371 bdevname(rdev->bdev,b));
3372 list_move(&rdev->same_set, &candidates);
3373 }
3374 /*
3375 * now we have a set of devices, with all of them having
3376 * mostly sane superblocks. It's time to allocate the
3377 * mddev.
3378 */
3379 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
3380 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3381 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3382 break;
3383 }
3384 if (part)
3385 dev = MKDEV(mdp_major,
3386 rdev0->preferred_minor << MdpMinorShift);
3387 else
3388 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3389
3390 md_probe(dev, NULL, NULL);
3391 mddev = mddev_find(dev);
3392 if (!mddev) {
3393 printk(KERN_ERR
3394 "md: cannot allocate memory for md drive.\n");
3395 break;
3396 }
3397 if (mddev_lock(mddev))
3398 printk(KERN_WARNING "md: %s locked, cannot run\n",
3399 mdname(mddev));
3400 else if (mddev->raid_disks || mddev->major_version
3401 || !list_empty(&mddev->disks)) {
3402 printk(KERN_WARNING
3403 "md: %s already running, cannot run %s\n",
3404 mdname(mddev), bdevname(rdev0->bdev,b));
3405 mddev_unlock(mddev);
3406 } else {
3407 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3408 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3409 list_del_init(&rdev->same_set);
3410 if (bind_rdev_to_array(rdev, mddev))
3411 export_rdev(rdev);
3412 }
3413 autorun_array(mddev);
3414 mddev_unlock(mddev);
3415 }
3416 /* on success, candidates will be empty, on error
3417 * it won't...
3418 */
3419 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3420 export_rdev(rdev);
3421 mddev_put(mddev);
3422 }
3423 printk(KERN_INFO "md: ... autorun DONE.\n");
3424}
3425
3426/*
3427 * import RAID devices based on one partition
3428 * if possible, the array gets run as well.
3429 */
3430
3431static int autostart_array(dev_t startdev)
3432{
3433 char b[BDEVNAME_SIZE];
3434 int err = -EINVAL, i;
3435 mdp_super_t *sb = NULL;
3436 mdk_rdev_t *start_rdev = NULL, *rdev;
3437
3438 start_rdev = md_import_device(startdev, 0, 0);
3439 if (IS_ERR(start_rdev))
3440 return err;
3441
3442
3443 /* NOTE: this can only work for 0.90.0 superblocks */
3444 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
3445 if (sb->major_version != 0 ||
3446 sb->minor_version != 90 ) {
3447 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
3448 export_rdev(start_rdev);
3449 return err;
3450 }
3451
b2d444d7 3452 if (test_bit(Faulty, &start_rdev->flags)) {
1da177e4
LT		 3453 printk(KERN_WARNING
3454 "md: can not autostart based on faulty %s!\n",
3455 bdevname(start_rdev->bdev,b));
3456 export_rdev(start_rdev);
3457 return err;
3458 }
3459 list_add(&start_rdev->same_set, &pending_raid_disks);
3460
3461 for (i = 0; i < MD_SB_DISKS; i++) {
3462 mdp_disk_t *desc = sb->disks + i;
3463 dev_t dev = MKDEV(desc->major, desc->minor);
3464
3465 if (!dev)
3466 continue;
3467 if (dev == startdev)
3468 continue;
3469 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3470 continue;
3471 rdev = md_import_device(dev, 0, 0);
3472 if (IS_ERR(rdev))
3473 continue;
3474
3475 list_add(&rdev->same_set, &pending_raid_disks);
3476 }
3477
3478 /*
3479 * possibly return codes
3480 */
3481 autorun_devices(0);
3482 return 0;
3483
3484}
3485
3486
3487static int get_version(void __user * arg)
3488{
3489 mdu_version_t ver;
3490
3491 ver.major = MD_MAJOR_VERSION;
3492 ver.minor = MD_MINOR_VERSION;
3493 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3494
3495 if (copy_to_user(arg, &ver, sizeof(ver)))
3496 return -EFAULT;
3497
3498 return 0;
3499}
3500
3501static int get_array_info(mddev_t * mddev, void __user * arg)
3502{
3503 mdu_array_info_t info;
3504 int nr,working,active,failed,spare;
3505 mdk_rdev_t *rdev;
3506 struct list_head *tmp;
3507
3508 nr=working=active=failed=spare=0;
3509 ITERATE_RDEV(mddev,rdev,tmp) {
3510 nr++;
b2d444d7 3511 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3512 failed++;
3513 else {
3514 working++;
b2d444d7 3515 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3516 active++;
3517 else
3518 spare++;
3519 }
3520 }
3521
3522 info.major_version = mddev->major_version;
3523 info.minor_version = mddev->minor_version;
3524 info.patch_version = MD_PATCHLEVEL_VERSION;
3525 info.ctime = mddev->ctime;
3526 info.level = mddev->level;
3527 info.size = mddev->size;
284ae7ca
N		 3528 if (info.size != mddev->size) /* overflow */
3529 info.size = -1;
1da177e4
LT		 3530 info.nr_disks = nr;
3531 info.raid_disks = mddev->raid_disks;
3532 info.md_minor = mddev->md_minor;
3533 info.not_persistent= !mddev->persistent;
3534
3535 info.utime = mddev->utime;
3536 info.state = 0;
3537 if (mddev->in_sync)
3538 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3539 if (mddev->bitmap && mddev->bitmap_offset)
3540 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3541 info.active_disks = active;
3542 info.working_disks = working;
3543 info.failed_disks = failed;
3544 info.spare_disks = spare;
3545
3546 info.layout = mddev->layout;
3547 info.chunk_size = mddev->chunk_size;
3548
3549 if (copy_to_user(arg, &info, sizeof(info)))
3550 return -EFAULT;
3551
3552 return 0;
3553}
3554
87162a28 3555static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3556{
3557 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3558 char *ptr, *buf = NULL;
3559 int err = -ENOMEM;
3560
3561 file = kmalloc(sizeof(*file), GFP_KERNEL);
3562 if (!file)
3563 goto out;
3564
3565 /* bitmap disabled, zero the first byte and copy out */
3566 if (!mddev->bitmap || !mddev->bitmap->file) {
3567 file->pathname[0] = '\0';
3568 goto copy_out;
3569 }
3570
3571 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3572 if (!buf)
3573 goto out;
3574
3575 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3576 if (!ptr)
3577 goto out;
3578
3579 strcpy(file->pathname, ptr);
3580
3581copy_out:
3582 err = 0;
3583 if (copy_to_user(arg, file, sizeof(*file)))
3584 err = -EFAULT;
3585out:
3586 kfree(buf);
3587 kfree(file);
3588 return err;
3589}
3590
1da177e4
LT		 3591static int get_disk_info(mddev_t * mddev, void __user * arg)
3592{
3593 mdu_disk_info_t info;
3594 unsigned int nr;
3595 mdk_rdev_t *rdev;
3596
3597 if (copy_from_user(&info, arg, sizeof(info)))
3598 return -EFAULT;
3599
3600 nr = info.number;
3601
3602 rdev = find_rdev_nr(mddev, nr);
3603 if (rdev) {
3604 info.major = MAJOR(rdev->bdev->bd_dev);
3605 info.minor = MINOR(rdev->bdev->bd_dev);
3606 info.raid_disk = rdev->raid_disk;
3607 info.state = 0;
b2d444d7 3608 if (test_bit(Faulty, &rdev->flags))
1da177e4 3609 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3610 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3611 info.state |= (1<<MD_DISK_ACTIVE);
3612 info.state |= (1<<MD_DISK_SYNC);
3613 }
8ddf9efe
N		 3614 if (test_bit(WriteMostly, &rdev->flags))
3615 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3616 } else {
3617 info.major = info.minor = 0;
3618 info.raid_disk = -1;
3619 info.state = (1<<MD_DISK_REMOVED);
3620 }
3621
3622 if (copy_to_user(arg, &info, sizeof(info)))
3623 return -EFAULT;
3624
3625 return 0;
3626}
3627
3628static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3629{
3630 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3631 mdk_rdev_t *rdev;
3632 dev_t dev = MKDEV(info->major,info->minor);
3633
3634 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3635 return -EOVERFLOW;
3636
3637 if (!mddev->raid_disks) {
3638 int err;
3639 /* expecting a device which has a superblock */
3640 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3641 if (IS_ERR(rdev)) {
3642 printk(KERN_WARNING
3643 "md: md_import_device returned %ld\n",
3644 PTR_ERR(rdev));
3645 return PTR_ERR(rdev);
3646 }
3647 if (!list_empty(&mddev->disks)) {
3648 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3649 mdk_rdev_t, same_set);
3650 int err = super_types[mddev->major_version]
3651 .load_super(rdev, rdev0, mddev->minor_version);
3652 if (err < 0) {
3653 printk(KERN_WARNING
3654 "md: %s has different UUID to %s\n",
3655 bdevname(rdev->bdev,b),
3656 bdevname(rdev0->bdev,b2));
3657 export_rdev(rdev);
3658 return -EINVAL;
3659 }
3660 }
3661 err = bind_rdev_to_array(rdev, mddev);
3662 if (err)
3663 export_rdev(rdev);
3664 return err;
3665 }
3666
3667 /*
3668 * add_new_disk can be used once the array is assembled
3669 * to add "hot spares". They must already have a superblock
3670 * written
3671 */
3672 if (mddev->pers) {
3673 int err;
3674 if (!mddev->pers->hot_add_disk) {
3675 printk(KERN_WARNING
3676 "%s: personality does not support diskops!\n",
3677 mdname(mddev));
3678 return -EINVAL;
3679 }
7b1e35f6
N		 3680 if (mddev->persistent)
3681 rdev = md_import_device(dev, mddev->major_version,
3682 mddev->minor_version);
3683 else
3684 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 3685 if (IS_ERR(rdev)) {
3686 printk(KERN_WARNING
3687 "md: md_import_device returned %ld\n",
3688 PTR_ERR(rdev));
3689 return PTR_ERR(rdev);
3690 }
41158c7e
N		 3691 /* set save_raid_disk if appropriate */
3692 if (!mddev->persistent) {
3693 if (info->state & (1<<MD_DISK_SYNC) &&
3694 info->raid_disk < mddev->raid_disks)
3695 rdev->raid_disk = info->raid_disk;
3696 else
3697 rdev->raid_disk = -1;
3698 } else
3699 super_types[mddev->major_version].
3700 validate_super(mddev, rdev);
3701 rdev->saved_raid_disk = rdev->raid_disk;
3702
b2d444d7 3703 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 3704 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3705 set_bit(WriteMostly, &rdev->flags);
3706
1da177e4
LT		 3707 rdev->raid_disk = -1;
3708 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 3709 if (!err && !mddev->pers->hot_remove_disk) {
3710 /* If there is hot_add_disk but no hot_remove_disk
3711 * then added disks for geometry changes,
3712 * and should be added immediately.
3713 */
3714 super_types[mddev->major_version].
3715 validate_super(mddev, rdev);
3716 err = mddev->pers->hot_add_disk(mddev, rdev);
3717 if (err)
3718 unbind_rdev_from_array(rdev);
3719 }
1da177e4
LT		 3720 if (err)
3721 export_rdev(rdev);
c361777f
N		 3722
3723 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 3724 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3725 return err;
3726 }
3727
3728 /* otherwise, add_new_disk is only allowed
3729 * for major_version==0 superblocks
3730 */
3731 if (mddev->major_version != 0) {
3732 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3733 mdname(mddev));
3734 return -EINVAL;
3735 }
3736
3737 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3738 int err;
3739 rdev = md_import_device (dev, -1, 0);
3740 if (IS_ERR(rdev)) {
3741 printk(KERN_WARNING
3742 "md: error, md_import_device() returned %ld\n",
3743 PTR_ERR(rdev));
3744 return PTR_ERR(rdev);
3745 }
3746 rdev->desc_nr = info->number;
3747 if (info->raid_disk < mddev->raid_disks)
3748 rdev->raid_disk = info->raid_disk;
3749 else
3750 rdev->raid_disk = -1;
3751
b2d444d7
N		 3752 rdev->flags = 0;
3753
1da177e4 3754 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 3755 if (info->state & (1<<MD_DISK_SYNC))
3756 set_bit(In_sync, &rdev->flags);
1da177e4 3757
8ddf9efe
N		 3758 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3759 set_bit(WriteMostly, &rdev->flags);
3760
1da177e4
LT		 3761 if (!mddev->persistent) {
3762 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3763 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3764 } else
3765 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3766 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3767
2bf071bf
N		 3768 err = bind_rdev_to_array(rdev, mddev);
3769 if (err) {
3770 export_rdev(rdev);
3771 return err;
3772 }
1da177e4
LT		 3773 }
3774
3775 return 0;
3776}
3777
3778static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3779{
3780 char b[BDEVNAME_SIZE];
3781 mdk_rdev_t *rdev;
3782
3783 if (!mddev->pers)
3784 return -ENODEV;
3785
3786 rdev = find_rdev(mddev, dev);
3787 if (!rdev)
3788 return -ENXIO;
3789
3790 if (rdev->raid_disk >= 0)
3791 goto busy;
3792
3793 kick_rdev_from_array(rdev);
3794 md_update_sb(mddev);
d7603b7e 3795 md_new_event(mddev);
1da177e4
LT		 3796
3797 return 0;
3798busy:
3799 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3800 bdevname(rdev->bdev,b), mdname(mddev));
3801 return -EBUSY;
3802}
3803
3804static int hot_add_disk(mddev_t * mddev, dev_t dev)
3805{
3806 char b[BDEVNAME_SIZE];
3807 int err;
3808 unsigned int size;
3809 mdk_rdev_t *rdev;
3810
3811 if (!mddev->pers)
3812 return -ENODEV;
3813
3814 if (mddev->major_version != 0) {
3815 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3816 " version-0 superblocks.\n",
3817 mdname(mddev));
3818 return -EINVAL;
3819 }
3820 if (!mddev->pers->hot_add_disk) {
3821 printk(KERN_WARNING
3822 "%s: personality does not support diskops!\n",
3823 mdname(mddev));
3824 return -EINVAL;
3825 }
3826
3827 rdev = md_import_device (dev, -1, 0);
3828 if (IS_ERR(rdev)) {
3829 printk(KERN_WARNING
3830 "md: error, md_import_device() returned %ld\n",
3831 PTR_ERR(rdev));
3832 return -EINVAL;
3833 }
3834
3835 if (mddev->persistent)
3836 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3837 else
3838 rdev->sb_offset =
3839 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3840
3841 size = calc_dev_size(rdev, mddev->chunk_size);
3842 rdev->size = size;
3843
b2d444d7 3844 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3845 printk(KERN_WARNING
3846 "md: can not hot-add faulty %s disk to %s!\n",
3847 bdevname(rdev->bdev,b), mdname(mddev));
3848 err = -EINVAL;
3849 goto abort_export;
3850 }
b2d444d7 3851 clear_bit(In_sync, &rdev->flags);
1da177e4 3852 rdev->desc_nr = -1;
2bf071bf
N		 3853 err = bind_rdev_to_array(rdev, mddev);
3854 if (err)
3855 goto abort_export;
1da177e4
LT		 3856
3857 /*
3858 * The rest should better be atomic, we can have disk failures
3859 * noticed in interrupt contexts ...
3860 */
3861
3862 if (rdev->desc_nr == mddev->max_disks) {
3863 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3864 mdname(mddev));
3865 err = -EBUSY;
3866 goto abort_unbind_export;
3867 }
3868
3869 rdev->raid_disk = -1;
3870
3871 md_update_sb(mddev);
3872
3873 /*
3874 * Kick recovery, maybe this spare has to be added to the
3875 * array immediately.
3876 */
3877 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3878 md_wakeup_thread(mddev->thread);
d7603b7e 3879 md_new_event(mddev);
1da177e4
LT		 3880 return 0;
3881
3882abort_unbind_export:
3883 unbind_rdev_from_array(rdev);
3884
3885abort_export:
3886 export_rdev(rdev);
3887 return err;
3888}
3889
32a7627c
N		 3890static int set_bitmap_file(mddev_t *mddev, int fd)
3891{
3892 int err;
3893
36fa3063
N		 3894 if (mddev->pers) {
3895 if (!mddev->pers->quiesce)
3896 return -EBUSY;
3897 if (mddev->recovery || mddev->sync_thread)
3898 return -EBUSY;
3899 /* we should be able to change the bitmap.. */
3900 }
32a7627c 3901
32a7627c 3902
36fa3063
N		 3903 if (fd >= 0) {
3904 if (mddev->bitmap)
3905 return -EEXIST; /* cannot add when bitmap is present */
3906 mddev->bitmap_file = fget(fd);
32a7627c 3907
36fa3063
N		 3908 if (mddev->bitmap_file == NULL) {
3909 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3910 mdname(mddev));
3911 return -EBADF;
3912 }
3913
3914 err = deny_bitmap_write_access(mddev->bitmap_file);
3915 if (err) {
3916 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3917 mdname(mddev));
3918 fput(mddev->bitmap_file);
3919 mddev->bitmap_file = NULL;
3920 return err;
3921 }
a654b9d8 3922 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 3923 } else if (mddev->bitmap == NULL)
3924 return -ENOENT; /* cannot remove what isn't there */
3925 err = 0;
3926 if (mddev->pers) {
3927 mddev->pers->quiesce(mddev, 1);
3928 if (fd >= 0)
3929 err = bitmap_create(mddev);
d7375ab3 3930 if (fd < 0 || err) {
36fa3063 3931 bitmap_destroy(mddev);
d7375ab3
N		 3932 fd = -1; /* make sure to put the file */
3933 }
36fa3063 3934 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 3935 }
3936 if (fd < 0) {
acc55e22
N		 3937 if (mddev->bitmap_file) {
3938 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 3939 fput(mddev->bitmap_file);
acc55e22 3940 }
36fa3063
N		 3941 mddev->bitmap_file = NULL;
3942 }
3943
32a7627c
N		 3944 return err;
3945}
3946
1da177e4
LT		 3947/*
3948 * set_array_info is used two different ways
3949 * The original usage is when creating a new array.
3950 * In this usage, raid_disks is > 0 and it together with
3951 * level, size, not_persistent,layout,chunksize determine the
3952 * shape of the array.
3953 * This will always create an array with a type-0.90.0 superblock.
3954 * The newer usage is when assembling an array.
3955 * In this case raid_disks will be 0, and the major_version field is
3956 * use to determine which style super-blocks are to be found on the devices.
3957 * The minor and patch _version numbers are also kept incase the
3958 * super_block handler wishes to interpret them.
3959 */
3960static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3961{
3962
3963 if (info->raid_disks == 0) {
3964 /* just setting version number for superblock loading */
3965 if (info->major_version < 0 ||
3966 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3967 super_types[info->major_version].name == NULL) {
3968 /* maybe try to auto-load a module? */
3969 printk(KERN_INFO
3970 "md: superblock version %d not known\n",
3971 info->major_version);
3972 return -EINVAL;
3973 }
3974 mddev->major_version = info->major_version;
3975 mddev->minor_version = info->minor_version;
3976 mddev->patch_version = info->patch_version;
3977 return 0;
3978 }
3979 mddev->major_version = MD_MAJOR_VERSION;
3980 mddev->minor_version = MD_MINOR_VERSION;
3981 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3982 mddev->ctime = get_seconds();
3983
3984 mddev->level = info->level;
17115e03 3985 mddev->clevel[0] = 0;
1da177e4
LT		 3986 mddev->size = info->size;
3987 mddev->raid_disks = info->raid_disks;
3988 /* don't set md_minor, it is determined by which /dev/md* was
3989 * openned
3990 */
3991 if (info->state & (1<<MD_SB_CLEAN))
3992 mddev->recovery_cp = MaxSector;
3993 else
3994 mddev->recovery_cp = 0;
3995 mddev->persistent = ! info->not_persistent;
3996
3997 mddev->layout = info->layout;
3998 mddev->chunk_size = info->chunk_size;
3999
4000 mddev->max_disks = MD_SB_DISKS;
4001
4002 mddev->sb_dirty = 1;
4003
b2a2703c
N		 4004 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4005 mddev->bitmap_offset = 0;
4006
f6705578
N		 4007 mddev->reshape_position = MaxSector;
4008
1da177e4
LT		 4009 /*
4010 * Generate a 128 bit UUID
4011 */
4012 get_random_bytes(mddev->uuid, 16);
4013
f6705578
N		 4014 mddev->new_level = mddev->level;
4015 mddev->new_chunk = mddev->chunk_size;
4016 mddev->new_layout = mddev->layout;
4017 mddev->delta_disks = 0;
4018
1da177e4
LT		 4019 return 0;
4020}
4021
a35b0d69
N		 4022static int update_size(mddev_t *mddev, unsigned long size)
4023{
4024 mdk_rdev_t * rdev;
4025 int rv;
4026 struct list_head *tmp;
8ddeeae5 4027 int fit = (size == 0);
a35b0d69
N		 4028
4029 if (mddev->pers->resize == NULL)
4030 return -EINVAL;
4031 /* The "size" is the amount of each device that is used.
4032 * This can only make sense for arrays with redundancy.
4033 * linear and raid0 always use whatever space is available
4034 * We can only consider changing the size if no resync
4035 * or reconstruction is happening, and if the new size
4036 * is acceptable. It must fit before the sb_offset or,
4037 * if that is <data_offset, it must fit before the
4038 * size of each device.
4039 * If size is zero, we find the largest size that fits.
4040 */
4041 if (mddev->sync_thread)
4042 return -EBUSY;
4043 ITERATE_RDEV(mddev,rdev,tmp) {
4044 sector_t avail;
a35b0d69
N		 4045 if (rdev->sb_offset > rdev->data_offset)
4046 avail = (rdev->sb_offset*2) - rdev->data_offset;
4047 else
4048 avail = get_capacity(rdev->bdev->bd_disk)
4049 - rdev->data_offset;
4050 if (fit && (size == 0 || size > avail/2))
4051 size = avail/2;
4052 if (avail < ((sector_t)size << 1))
4053 return -ENOSPC;
4054 }
4055 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4056 if (!rv) {
4057 struct block_device *bdev;
4058
4059 bdev = bdget_disk(mddev->gendisk, 0);
4060 if (bdev) {
1b1dcc1b 4061 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4062 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4063 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4064 bdput(bdev);
4065 }
4066 }
4067 return rv;
4068}
4069
da943b99
N		 4070static int update_raid_disks(mddev_t *mddev, int raid_disks)
4071{
4072 int rv;
4073 /* change the number of raid disks */
63c70c4f 4074 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4075 return -EINVAL;
4076 if (raid_disks <= 0 ||
4077 raid_disks >= mddev->max_disks)
4078 return -EINVAL;
63c70c4f 4079 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4080 return -EBUSY;
63c70c4f
N		 4081 mddev->delta_disks = raid_disks - mddev->raid_disks;
4082
4083 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4084 return rv;
4085}
4086
4087
1da177e4
LT		 4088/*
4089 * update_array_info is used to change the configuration of an
4090 * on-line array.
4091 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4092 * fields in the info are checked against the array.
4093 * Any differences that cannot be handled will cause an error.
4094 * Normally, only one change can be managed at a time.
4095 */
4096static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4097{
4098 int rv = 0;
4099 int cnt = 0;
36fa3063
N		 4100 int state = 0;
4101
4102 /* calculate expected state,ignoring low bits */
4103 if (mddev->bitmap && mddev->bitmap_offset)
4104 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4105
4106 if (mddev->major_version != info->major_version ||
4107 mddev->minor_version != info->minor_version ||
4108/* mddev->patch_version != info->patch_version || */
4109 mddev->ctime != info->ctime ||
4110 mddev->level != info->level ||
4111/* mddev->layout != info->layout || */
4112 !mddev->persistent != info->not_persistent||
36fa3063
N		 4113 mddev->chunk_size != info->chunk_size ||
4114 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4115 ((state^info->state) & 0xfffffe00)
4116 )
1da177e4
LT		 4117 return -EINVAL;
4118 /* Check there is only one change */
284ae7ca 4119 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4120 if (mddev->raid_disks != info->raid_disks) cnt++;
4121 if (mddev->layout != info->layout) cnt++;
36fa3063 4122 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4123 if (cnt == 0) return 0;
4124 if (cnt > 1) return -EINVAL;
4125
4126 if (mddev->layout != info->layout) {
4127 /* Change layout
4128 * we don't need to do anything at the md level, the
4129 * personality will take care of it all.
4130 */
4131 if (mddev->pers->reconfig == NULL)
4132 return -EINVAL;
4133 else
4134 return mddev->pers->reconfig(mddev, info->layout, -1);
4135 }
284ae7ca 4136 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4137 rv = update_size(mddev, info->size);
4138
da943b99
N		 4139 if (mddev->raid_disks != info->raid_disks)
4140 rv = update_raid_disks(mddev, info->raid_disks);
4141
36fa3063
N		 4142 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4143 if (mddev->pers->quiesce == NULL)
4144 return -EINVAL;
4145 if (mddev->recovery || mddev->sync_thread)
4146 return -EBUSY;
4147 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4148 /* add the bitmap */
4149 if (mddev->bitmap)
4150 return -EEXIST;
4151 if (mddev->default_bitmap_offset == 0)
4152 return -EINVAL;
4153 mddev->bitmap_offset = mddev->default_bitmap_offset;
4154 mddev->pers->quiesce(mddev, 1);
4155 rv = bitmap_create(mddev);
4156 if (rv)
4157 bitmap_destroy(mddev);
4158 mddev->pers->quiesce(mddev, 0);
4159 } else {
4160 /* remove the bitmap */
4161 if (!mddev->bitmap)
4162 return -ENOENT;
4163 if (mddev->bitmap->file)
4164 return -EINVAL;
4165 mddev->pers->quiesce(mddev, 1);
4166 bitmap_destroy(mddev);
4167 mddev->pers->quiesce(mddev, 0);
4168 mddev->bitmap_offset = 0;
4169 }
4170 }
1da177e4
LT		 4171 md_update_sb(mddev);
4172 return rv;
4173}
4174
4175static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4176{
4177 mdk_rdev_t *rdev;
4178
4179 if (mddev->pers == NULL)
4180 return -ENODEV;
4181
4182 rdev = find_rdev(mddev, dev);
4183 if (!rdev)
4184 return -ENODEV;
4185
4186 md_error(mddev, rdev);
4187 return 0;
4188}
4189
a885c8c4
CH		 4190static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4191{
4192 mddev_t *mddev = bdev->bd_disk->private_data;
4193
4194 geo->heads = 2;
4195 geo->sectors = 4;
4196 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4197 return 0;
4198}
4199
1da177e4
LT		 4200static int md_ioctl(struct inode *inode, struct file *file,
4201 unsigned int cmd, unsigned long arg)
4202{
4203 int err = 0;
4204 void __user *argp = (void __user *)arg;
1da177e4
LT		 4205 mddev_t *mddev = NULL;
4206
4207 if (!capable(CAP_SYS_ADMIN))
4208 return -EACCES;
4209
4210 /*
4211 * Commands dealing with the RAID driver but not any
4212 * particular array:
4213 */
4214 switch (cmd)
4215 {
4216 case RAID_VERSION:
4217 err = get_version(argp);
4218 goto done;
4219
4220 case PRINT_RAID_DEBUG:
4221 err = 0;
4222 md_print_devices();
4223 goto done;
4224
4225#ifndef MODULE
4226 case RAID_AUTORUN:
4227 err = 0;
4228 autostart_arrays(arg);
4229 goto done;
4230#endif
4231 default:;
4232 }
4233
4234 /*
4235 * Commands creating/starting a new array:
4236 */
4237
4238 mddev = inode->i_bdev->bd_disk->private_data;
4239
4240 if (!mddev) {
4241 BUG();
4242 goto abort;
4243 }
4244
4245
4246 if (cmd == START_ARRAY) {
4247 /* START_ARRAY doesn't need to lock the array as autostart_array
4248 * does the locking, and it could even be a different array
4249 */
4250 static int cnt = 3;
4251 if (cnt > 0 ) {
4252 printk(KERN_WARNING
4253 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
e8a00334 4254 "This will not be supported beyond July 2006\n",
1da177e4
LT		 4255 current->comm, current->pid);
4256 cnt--;
4257 }
4258 err = autostart_array(new_decode_dev(arg));
4259 if (err) {
4260 printk(KERN_WARNING "md: autostart failed!\n");
4261 goto abort;
4262 }
4263 goto done;
4264 }
4265
4266 err = mddev_lock(mddev);
4267 if (err) {
4268 printk(KERN_INFO
4269 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4270 err, cmd);
4271 goto abort;
4272 }
4273
4274 switch (cmd)
4275 {
4276 case SET_ARRAY_INFO:
4277 {
4278 mdu_array_info_t info;
4279 if (!arg)
4280 memset(&info, 0, sizeof(info));
4281 else if (copy_from_user(&info, argp, sizeof(info))) {
4282 err = -EFAULT;
4283 goto abort_unlock;
4284 }
4285 if (mddev->pers) {
4286 err = update_array_info(mddev, &info);
4287 if (err) {
4288 printk(KERN_WARNING "md: couldn't update"
4289 " array info. %d\n", err);
4290 goto abort_unlock;
4291 }
4292 goto done_unlock;
4293 }
4294 if (!list_empty(&mddev->disks)) {
4295 printk(KERN_WARNING
4296 "md: array %s already has disks!\n",
4297 mdname(mddev));
4298 err = -EBUSY;
4299 goto abort_unlock;
4300 }
4301 if (mddev->raid_disks) {
4302 printk(KERN_WARNING
4303 "md: array %s already initialised!\n",
4304 mdname(mddev));
4305 err = -EBUSY;
4306 goto abort_unlock;
4307 }
4308 err = set_array_info(mddev, &info);
4309 if (err) {
4310 printk(KERN_WARNING "md: couldn't set"
4311 " array info. %d\n", err);
4312 goto abort_unlock;
4313 }
4314 }
4315 goto done_unlock;
4316
4317 default:;
4318 }
4319
4320 /*
4321 * Commands querying/configuring an existing array:
4322 */
32a7627c
N		 4323 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4324 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4325 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4326 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 4327 err = -ENODEV;
4328 goto abort_unlock;
4329 }
4330
4331 /*
4332 * Commands even a read-only array can execute:
4333 */
4334 switch (cmd)
4335 {
4336 case GET_ARRAY_INFO:
4337 err = get_array_info(mddev, argp);
4338 goto done_unlock;
4339
32a7627c 4340 case GET_BITMAP_FILE:
87162a28 4341 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4342 goto done_unlock;
4343
1da177e4
LT		 4344 case GET_DISK_INFO:
4345 err = get_disk_info(mddev, argp);
4346 goto done_unlock;
4347
4348 case RESTART_ARRAY_RW:
4349 err = restart_array(mddev);
4350 goto done_unlock;
4351
4352 case STOP_ARRAY:
4353 err = do_md_stop (mddev, 0);
4354 goto done_unlock;
4355
4356 case STOP_ARRAY_RO:
4357 err = do_md_stop (mddev, 1);
4358 goto done_unlock;
4359
4360 /*
4361 * We have a problem here : there is no easy way to give a CHS
4362 * virtual geometry. We currently pretend that we have a 2 heads
4363 * 4 sectors (with a BIG number of cylinders...). This drives
4364 * dosfs just mad... ;-)
4365 */
1da177e4
LT		 4366 }
4367
4368 /*
4369 * The remaining ioctls are changing the state of the
f91de92e
N		 4370 * superblock, so we do not allow them on read-only arrays.
4371 * However non-MD ioctls (e.g. get-size) will still come through
4372 * here and hit the 'default' below, so only disallow
4373 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4374 */
f91de92e
N		 4375 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4376 mddev->ro && mddev->pers) {
4377 if (mddev->ro == 2) {
4378 mddev->ro = 0;
4379 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4380 md_wakeup_thread(mddev->thread);
4381
4382 } else {
4383 err = -EROFS;
4384 goto abort_unlock;
4385 }
1da177e4
LT		 4386 }
4387
4388 switch (cmd)
4389 {
4390 case ADD_NEW_DISK:
4391 {
4392 mdu_disk_info_t info;
4393 if (copy_from_user(&info, argp, sizeof(info)))
4394 err = -EFAULT;
4395 else
4396 err = add_new_disk(mddev, &info);
4397 goto done_unlock;
4398 }
4399
4400 case HOT_REMOVE_DISK:
4401 err = hot_remove_disk(mddev, new_decode_dev(arg));
4402 goto done_unlock;
4403
4404 case HOT_ADD_DISK:
4405 err = hot_add_disk(mddev, new_decode_dev(arg));
4406 goto done_unlock;
4407
4408 case SET_DISK_FAULTY:
4409 err = set_disk_faulty(mddev, new_decode_dev(arg));
4410 goto done_unlock;
4411
4412 case RUN_ARRAY:
4413 err = do_md_run (mddev);
4414 goto done_unlock;
4415
32a7627c
N		 4416 case SET_BITMAP_FILE:
4417 err = set_bitmap_file(mddev, (int)arg);
4418 goto done_unlock;
4419
1da177e4 4420 default:
1da177e4
LT		 4421 err = -EINVAL;
4422 goto abort_unlock;
4423 }
4424
4425done_unlock:
4426abort_unlock:
4427 mddev_unlock(mddev);
4428
4429 return err;
4430done:
4431 if (err)
4432 MD_BUG();
4433abort:
4434 return err;
4435}
4436
4437static int md_open(struct inode *inode, struct file *file)
4438{
4439 /*
4440 * Succeed if we can lock the mddev, which confirms that
4441 * it isn't being stopped right now.
4442 */
4443 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4444 int err;
4445
4446 if ((err = mddev_lock(mddev)))
4447 goto out;
4448
4449 err = 0;
4450 mddev_get(mddev);
4451 mddev_unlock(mddev);
4452
4453 check_disk_change(inode->i_bdev);
4454 out:
4455 return err;
4456}
4457
4458static int md_release(struct inode *inode, struct file * file)
4459{
4460 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4461
4462 if (!mddev)
4463 BUG();
4464 mddev_put(mddev);
4465
4466 return 0;
4467}
4468
4469static int md_media_changed(struct gendisk *disk)
4470{
4471 mddev_t *mddev = disk->private_data;
4472
4473 return mddev->changed;
4474}
4475
4476static int md_revalidate(struct gendisk *disk)
4477{
4478 mddev_t *mddev = disk->private_data;
4479
4480 mddev->changed = 0;
4481 return 0;
4482}
4483static struct block_device_operations md_fops =
4484{
4485 .owner = THIS_MODULE,
4486 .open = md_open,
4487 .release = md_release,
4488 .ioctl = md_ioctl,
a885c8c4 4489 .getgeo = md_getgeo,
1da177e4
LT		 4490 .media_changed = md_media_changed,
4491 .revalidate_disk= md_revalidate,
4492};
4493
75c96f85 4494static int md_thread(void * arg)
1da177e4
LT		 4495{
4496 mdk_thread_t *thread = arg;
4497
1da177e4
LT		 4498 /*
4499 * md_thread is a 'system-thread', it's priority should be very
4500 * high. We avoid resource deadlocks individually in each
4501 * raid personality. (RAID5 does preallocation) We also use RR and
4502 * the very same RT priority as kswapd, thus we will never get
4503 * into a priority inversion deadlock.
4504 *
4505 * we definitely have to have equal or higher priority than
4506 * bdflush, otherwise bdflush will deadlock if there are too
4507 * many dirty RAID5 blocks.
4508 */
1da177e4 4509
6985c43f 4510 allow_signal(SIGKILL);
a6fb0934 4511 while (!kthread_should_stop()) {
1da177e4 4512
93588e22
N		 4513 /* We need to wait INTERRUPTIBLE so that
4514 * we don't add to the load-average.
4515 * That means we need to be sure no signals are
4516 * pending
4517 */
4518 if (signal_pending(current))
4519 flush_signals(current);
4520
4521 wait_event_interruptible_timeout
4522 (thread->wqueue,
4523 test_bit(THREAD_WAKEUP, &thread->flags)
4524 || kthread_should_stop(),
4525 thread->timeout);
3e1d1d28 4526 try_to_freeze();
1da177e4
LT		 4527
4528 clear_bit(THREAD_WAKEUP, &thread->flags);
4529
787453c2 4530 thread->run(thread->mddev);
1da177e4 4531 }
a6fb0934 4532
1da177e4
LT		 4533 return 0;
4534}
4535
4536void md_wakeup_thread(mdk_thread_t *thread)
4537{
4538 if (thread) {
4539 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4540 set_bit(THREAD_WAKEUP, &thread->flags);
4541 wake_up(&thread->wqueue);
4542 }
4543}
4544
4545mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4546 const char *name)
4547{
4548 mdk_thread_t *thread;
1da177e4 4549
9ffae0cf 4550 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4551 if (!thread)
4552 return NULL;
4553
1da177e4
LT		 4554 init_waitqueue_head(&thread->wqueue);
4555
1da177e4
LT		 4556 thread->run = run;
4557 thread->mddev = mddev;
32a7627c 4558 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4559 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4560 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4561 kfree(thread);
4562 return NULL;
4563 }
1da177e4
LT		 4564 return thread;
4565}
4566
1da177e4
LT		 4567void md_unregister_thread(mdk_thread_t *thread)
4568{
d28446fe 4569 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 4570
4571 kthread_stop(thread->tsk);
1da177e4
LT		 4572 kfree(thread);
4573}
4574
4575void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4576{
4577 if (!mddev) {
4578 MD_BUG();
4579 return;
4580 }
4581
b2d444d7 4582 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4583 return;
32a7627c 4584/*
1da177e4
LT		 4585 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4586 mdname(mddev),
4587 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4588 __builtin_return_address(0),__builtin_return_address(1),
4589 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4590*/
1da177e4
LT		 4591 if (!mddev->pers->error_handler)
4592 return;
4593 mddev->pers->error_handler(mddev,rdev);
4594 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4595 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4596 md_wakeup_thread(mddev->thread);
c331eb04 4597 md_new_event_inintr(mddev);
1da177e4
LT		 4598}
4599
4600/* seq_file implementation /proc/mdstat */
4601
4602static void status_unused(struct seq_file *seq)
4603{
4604 int i = 0;
4605 mdk_rdev_t *rdev;
4606 struct list_head *tmp;
4607
4608 seq_printf(seq, "unused devices: ");
4609
4610 ITERATE_RDEV_PENDING(rdev,tmp) {
4611 char b[BDEVNAME_SIZE];
4612 i++;
4613 seq_printf(seq, "%s ",
4614 bdevname(rdev->bdev,b));
4615 }
4616 if (!i)
4617 seq_printf(seq, "<none>");
4618
4619 seq_printf(seq, "\n");
4620}
4621
4622
4623static void status_resync(struct seq_file *seq, mddev_t * mddev)
4624{
4588b42e
N		 4625 sector_t max_blocks, resync, res;
4626 unsigned long dt, db, rt;
4627 int scale;
4628 unsigned int per_milli;
1da177e4
LT		 4629
4630 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4631
4632 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4633 max_blocks = mddev->resync_max_sectors >> 1;
4634 else
4635 max_blocks = mddev->size;
4636
4637 /*
4638 * Should not happen.
4639 */
4640 if (!max_blocks) {
4641 MD_BUG();
4642 return;
4643 }
4588b42e
N		 4644 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4645 * in a sector_t, and (max_blocks>>scale) will fit in a
4646 * u32, as those are the requirements for sector_div.
4647 * Thus 'scale' must be at least 10
4648 */
4649 scale = 10;
4650 if (sizeof(sector_t) > sizeof(unsigned long)) {
4651 while ( max_blocks/2 > (1ULL<<(scale+32)))
4652 scale++;
4653 }
4654 res = (resync>>scale)*1000;
4655 sector_div(res, (u32)((max_blocks>>scale)+1));
4656
4657 per_milli = res;
1da177e4 4658 {
4588b42e 4659 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4660 seq_printf(seq, "[");
4661 for (i = 0; i < x; i++)
4662 seq_printf(seq, "=");
4663 seq_printf(seq, ">");
4664 for (i = 0; i < y; i++)
4665 seq_printf(seq, ".");
4666 seq_printf(seq, "] ");
4667 }
4588b42e 4668 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 4669 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4670 "reshape" :
1da177e4 4671 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
ccfcc3c1 4672 "resync" : "recovery")),
4588b42e
N		 4673 per_milli/10, per_milli % 10,
4674 (unsigned long long) resync,
4675 (unsigned long long) max_blocks);
1da177e4
LT		 4676
4677 /*
4678 * We do not want to overflow, so the order of operands and
4679 * the * 100 / 100 trick are important. We do a +1 to be
4680 * safe against division by zero. We only estimate anyway.
4681 *
4682 * dt: time from mark until now
4683 * db: blocks written from mark until now
4684 * rt: remaining time
4685 */
4686 dt = ((jiffies - mddev->resync_mark) / HZ);
4687 if (!dt) dt++;
ff4e8d9a
N		 4688 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4689 - mddev->resync_mark_cnt;
4690 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 4691
4692 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4693
ff4e8d9a 4694 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 4695}
4696
4697static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4698{
4699 struct list_head *tmp;
4700 loff_t l = *pos;
4701 mddev_t *mddev;
4702
4703 if (l >= 0x10000)
4704 return NULL;
4705 if (!l--)
4706 /* header */
4707 return (void*)1;
4708
4709 spin_lock(&all_mddevs_lock);
4710 list_for_each(tmp,&all_mddevs)
4711 if (!l--) {
4712 mddev = list_entry(tmp, mddev_t, all_mddevs);
4713 mddev_get(mddev);
4714 spin_unlock(&all_mddevs_lock);
4715 return mddev;
4716 }
4717 spin_unlock(&all_mddevs_lock);
4718 if (!l--)
4719 return (void*)2;/* tail */
4720 return NULL;
4721}
4722
4723static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4724{
4725 struct list_head *tmp;
4726 mddev_t *next_mddev, *mddev = v;
4727
4728 ++*pos;
4729 if (v == (void*)2)
4730 return NULL;
4731
4732 spin_lock(&all_mddevs_lock);
4733 if (v == (void*)1)
4734 tmp = all_mddevs.next;
4735 else
4736 tmp = mddev->all_mddevs.next;
4737 if (tmp != &all_mddevs)
4738 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4739 else {
4740 next_mddev = (void*)2;
4741 *pos = 0x10000;
4742 }
4743 spin_unlock(&all_mddevs_lock);
4744
4745 if (v != (void*)1)
4746 mddev_put(mddev);
4747 return next_mddev;
4748
4749}
4750
4751static void md_seq_stop(struct seq_file *seq, void *v)
4752{
4753 mddev_t *mddev = v;
4754
4755 if (mddev && v != (void*)1 && v != (void*)2)
4756 mddev_put(mddev);
4757}
4758
d7603b7e
N		 4759struct mdstat_info {
4760 int event;
4761};
4762
1da177e4
LT		 4763static int md_seq_show(struct seq_file *seq, void *v)
4764{
4765 mddev_t *mddev = v;
4766 sector_t size;
4767 struct list_head *tmp2;
4768 mdk_rdev_t *rdev;
d7603b7e 4769 struct mdstat_info *mi = seq->private;
32a7627c 4770 struct bitmap *bitmap;
1da177e4
LT		 4771
4772 if (v == (void*)1) {
2604b703 4773 struct mdk_personality *pers;
1da177e4
LT		 4774 seq_printf(seq, "Personalities : ");
4775 spin_lock(&pers_lock);
2604b703
N		 4776 list_for_each_entry(pers, &pers_list, list)
4777 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 4778
4779 spin_unlock(&pers_lock);
4780 seq_printf(seq, "\n");
d7603b7e 4781 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 4782 return 0;
4783 }
4784 if (v == (void*)2) {
4785 status_unused(seq);
4786 return 0;
4787 }
4788
5dc5cf7d 4789 if (mddev_lock(mddev) < 0)
1da177e4 4790 return -EINTR;
5dc5cf7d 4791
1da177e4
LT		 4792 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4793 seq_printf(seq, "%s : %sactive", mdname(mddev),
4794 mddev->pers ? "" : "in");
4795 if (mddev->pers) {
f91de92e 4796 if (mddev->ro==1)
1da177e4 4797 seq_printf(seq, " (read-only)");
f91de92e
N		 4798 if (mddev->ro==2)
4799 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 4800 seq_printf(seq, " %s", mddev->pers->name);
4801 }
4802
4803 size = 0;
4804 ITERATE_RDEV(mddev,rdev,tmp2) {
4805 char b[BDEVNAME_SIZE];
4806 seq_printf(seq, " %s[%d]",
4807 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 4808 if (test_bit(WriteMostly, &rdev->flags))
4809 seq_printf(seq, "(W)");
b2d444d7 4810 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4811 seq_printf(seq, "(F)");
4812 continue;
b325a32e
N		 4813 } else if (rdev->raid_disk < 0)
4814 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 4815 size += rdev->size;
4816 }
4817
4818 if (!list_empty(&mddev->disks)) {
4819 if (mddev->pers)
4820 seq_printf(seq, "\n %llu blocks",
4821 (unsigned long long)mddev->array_size);
4822 else
4823 seq_printf(seq, "\n %llu blocks",
4824 (unsigned long long)size);
4825 }
1cd6bf19
N		 4826 if (mddev->persistent) {
4827 if (mddev->major_version != 0 ||
4828 mddev->minor_version != 90) {
4829 seq_printf(seq," super %d.%d",
4830 mddev->major_version,
4831 mddev->minor_version);
4832 }
4833 } else
4834 seq_printf(seq, " super non-persistent");
1da177e4
LT		 4835
4836 if (mddev->pers) {
4837 mddev->pers->status (seq, mddev);
4838 seq_printf(seq, "\n ");
8e1b39d6
N		 4839 if (mddev->pers->sync_request) {
4840 if (mddev->curr_resync > 2) {
4841 status_resync (seq, mddev);
4842 seq_printf(seq, "\n ");
4843 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4844 seq_printf(seq, "\tresync=DELAYED\n ");
4845 else if (mddev->recovery_cp < MaxSector)
4846 seq_printf(seq, "\tresync=PENDING\n ");
4847 }
32a7627c
N		 4848 } else
4849 seq_printf(seq, "\n ");
4850
4851 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 4852 unsigned long chunk_kb;
4853 unsigned long flags;
32a7627c
N		 4854 spin_lock_irqsave(&bitmap->lock, flags);
4855 chunk_kb = bitmap->chunksize >> 10;
4856 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4857 "%lu%s chunk",
4858 bitmap->pages - bitmap->missing_pages,
4859 bitmap->pages,
4860 (bitmap->pages - bitmap->missing_pages)
4861 << (PAGE_SHIFT - 10),
4862 chunk_kb ? chunk_kb : bitmap->chunksize,
4863 chunk_kb ? "KB" : "B");
78d742d8
N		 4864 if (bitmap->file) {
4865 seq_printf(seq, ", file: ");
4866 seq_path(seq, bitmap->file->f_vfsmnt,
4867 bitmap->file->f_dentry," \t\n");
32a7627c 4868 }
78d742d8 4869
32a7627c
N		 4870 seq_printf(seq, "\n");
4871 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 4872 }
4873
4874 seq_printf(seq, "\n");
4875 }
4876 mddev_unlock(mddev);
4877
4878 return 0;
4879}
4880
4881static struct seq_operations md_seq_ops = {
4882 .start = md_seq_start,
4883 .next = md_seq_next,
4884 .stop = md_seq_stop,
4885 .show = md_seq_show,
4886};
4887
4888static int md_seq_open(struct inode *inode, struct file *file)
4889{
4890 int error;
d7603b7e
N		 4891 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4892 if (mi == NULL)
4893 return -ENOMEM;
1da177e4
LT		 4894
4895 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 4896 if (error)
4897 kfree(mi);
4898 else {
4899 struct seq_file *p = file->private_data;
4900 p->private = mi;
4901 mi->event = atomic_read(&md_event_count);
4902 }
1da177e4
LT		 4903 return error;
4904}
4905
d7603b7e
N		 4906static int md_seq_release(struct inode *inode, struct file *file)
4907{
4908 struct seq_file *m = file->private_data;
4909 struct mdstat_info *mi = m->private;
4910 m->private = NULL;
4911 kfree(mi);
4912 return seq_release(inode, file);
4913}
4914
4915static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4916{
4917 struct seq_file *m = filp->private_data;
4918 struct mdstat_info *mi = m->private;
4919 int mask;
4920
4921 poll_wait(filp, &md_event_waiters, wait);
4922
4923 /* always allow read */
4924 mask = POLLIN | POLLRDNORM;
4925
4926 if (mi->event != atomic_read(&md_event_count))
4927 mask |= POLLERR | POLLPRI;
4928 return mask;
4929}
4930
1da177e4
LT		 4931static struct file_operations md_seq_fops = {
4932 .open = md_seq_open,
4933 .read = seq_read,
4934 .llseek = seq_lseek,
d7603b7e
N		 4935 .release = md_seq_release,
4936 .poll = mdstat_poll,
1da177e4
LT		 4937};
4938
2604b703 4939int register_md_personality(struct mdk_personality *p)
1da177e4 4940{
1da177e4 4941 spin_lock(&pers_lock);
2604b703
N		 4942 list_add_tail(&p->list, &pers_list);
4943 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 4944 spin_unlock(&pers_lock);
4945 return 0;
4946}
4947
2604b703 4948int unregister_md_personality(struct mdk_personality *p)
1da177e4 4949{
2604b703 4950 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 4951 spin_lock(&pers_lock);
2604b703 4952 list_del_init(&p->list);
1da177e4
LT		 4953 spin_unlock(&pers_lock);
4954 return 0;
4955}
4956
4957static int is_mddev_idle(mddev_t *mddev)
4958{
4959 mdk_rdev_t * rdev;
4960 struct list_head *tmp;
4961 int idle;
4962 unsigned long curr_events;
4963
4964 idle = 1;
4965 ITERATE_RDEV(mddev,rdev,tmp) {
4966 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 4967 curr_events = disk_stat_read(disk, sectors[0]) +
4968 disk_stat_read(disk, sectors[1]) -
1da177e4 4969 atomic_read(&disk->sync_io);
c0e48521
N		 4970 /* The difference between curr_events and last_events
4971 * will be affected by any new non-sync IO (making
4972 * curr_events bigger) and any difference in the amount of
4973 * in-flight syncio (making current_events bigger or smaller)
4974 * The amount in-flight is currently limited to
4975 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4976 * which is at most 4096 sectors.
4977 * These numbers are fairly fragile and should be made
4978 * more robust, probably by enforcing the
4979 * 'window size' that md_do_sync sort-of uses.
4980 *
1da177e4
LT		 4981 * Note: the following is an unsigned comparison.
4982 */
c0e48521 4983 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 4984 rdev->last_events = curr_events;
4985 idle = 0;
4986 }
4987 }
4988 return idle;
4989}
4990
4991void md_done_sync(mddev_t *mddev, int blocks, int ok)
4992{
4993 /* another "blocks" (512byte) blocks have been synced */
4994 atomic_sub(blocks, &mddev->recovery_active);
4995 wake_up(&mddev->recovery_wait);
4996 if (!ok) {
4997 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4998 md_wakeup_thread(mddev->thread);
4999 // stop recovery, signal do_sync ....
5000 }
5001}
5002
5003
06d91a5f
N		 5004/* md_write_start(mddev, bi)
5005 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 5006 * in superblock) before writing, schedule a superblock update
5007 * and wait for it to complete.
06d91a5f 5008 */
3d310eb7 5009void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 5010{
06d91a5f 5011 if (bio_data_dir(bi) != WRITE)
3d310eb7 5012 return;
06d91a5f 5013
f91de92e
N		 5014 BUG_ON(mddev->ro == 1);
5015 if (mddev->ro == 2) {
5016 /* need to switch to read/write */
5017 mddev->ro = 0;
5018 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5019 md_wakeup_thread(mddev->thread);
5020 }
06d91a5f 5021 atomic_inc(&mddev->writes_pending);
06d91a5f 5022 if (mddev->in_sync) {
a9701a30 5023 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5024 if (mddev->in_sync) {
5025 mddev->in_sync = 0;
42543769 5026 mddev->sb_dirty = 3;
3d310eb7
N		 5027 md_wakeup_thread(mddev->thread);
5028 }
a9701a30 5029 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5030 }
3d310eb7 5031 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
1da177e4
LT		 5032}
5033
5034void md_write_end(mddev_t *mddev)
5035{
5036 if (atomic_dec_and_test(&mddev->writes_pending)) {
5037 if (mddev->safemode == 2)
5038 md_wakeup_thread(mddev->thread);
16f17b39 5039 else if (mddev->safemode_delay)
1da177e4
LT		 5040 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5041 }
5042}
5043
75c96f85 5044static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5045
5046#define SYNC_MARKS 10
5047#define SYNC_MARK_STEP (3*HZ)
29269553 5048void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5049{
5050 mddev_t *mddev2;
5051 unsigned int currspeed = 0,
5052 window;
57afd89f 5053 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5054 unsigned long mark[SYNC_MARKS];
5055 sector_t mark_cnt[SYNC_MARKS];
5056 int last_mark,m;
5057 struct list_head *tmp;
5058 sector_t last_check;
57afd89f 5059 int skipped = 0;
5fd6c1dc
N		 5060 struct list_head *rtmp;
5061 mdk_rdev_t *rdev;
1da177e4
LT		 5062
5063 /* just incase thread restarts... */
5064 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5065 return;
5fd6c1dc
N		 5066 if (mddev->ro) /* never try to sync a read-only array */
5067 return;
1da177e4
LT		 5068
5069 /* we overload curr_resync somewhat here.
5070 * 0 == not engaged in resync at all
5071 * 2 == checking that there is no conflict with another sync
5072 * 1 == like 2, but have yielded to allow conflicting resync to
5073 * commense
5074 * other == active in resync - this many blocks
5075 *
5076 * Before starting a resync we must have set curr_resync to
5077 * 2, and then checked that every "conflicting" array has curr_resync
5078 * less than ours. When we find one that is the same or higher
5079 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5080 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5081 * This will mean we have to start checking from the beginning again.
5082 *
5083 */
5084
5085 do {
5086 mddev->curr_resync = 2;
5087
5088 try_again:
787453c2 5089 if (kthread_should_stop()) {
6985c43f 5090 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5091 goto skip;
5092 }
5093 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 5094 if (mddev2 == mddev)
5095 continue;
5096 if (mddev2->curr_resync &&
5097 match_mddev_units(mddev,mddev2)) {
5098 DEFINE_WAIT(wq);
5099 if (mddev < mddev2 && mddev->curr_resync == 2) {
5100 /* arbitrarily yield */
5101 mddev->curr_resync = 1;
5102 wake_up(&resync_wait);
5103 }
5104 if (mddev > mddev2 && mddev->curr_resync == 1)
5105 /* no need to wait here, we can wait the next
5106 * time 'round when curr_resync == 2
5107 */
5108 continue;
787453c2
N		 5109 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5110 if (!kthread_should_stop() &&
8712e553 5111 mddev2->curr_resync >= mddev->curr_resync) {
1da177e4
LT		 5112 printk(KERN_INFO "md: delaying resync of %s"
5113 " until %s has finished resync (they"
5114 " share one or more physical units)\n",
5115 mdname(mddev), mdname(mddev2));
5116 mddev_put(mddev2);
5117 schedule();
5118 finish_wait(&resync_wait, &wq);
5119 goto try_again;
5120 }
5121 finish_wait(&resync_wait, &wq);
5122 }
5123 }
5124 } while (mddev->curr_resync < 2);
5125
5fd6c1dc 5126 j = 0;
9d88883e 5127 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5128 /* resync follows the size requested by the personality,
57afd89f 5129 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5130 */
5131 max_sectors = mddev->resync_max_sectors;
9d88883e 5132 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5133 /* we don't use the checkpoint if there's a bitmap */
5134 if (!mddev->bitmap &&
5135 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5136 j = mddev->recovery_cp;
ccfcc3c1
N		 5137 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5138 max_sectors = mddev->size << 1;
5fd6c1dc 5139 else {
1da177e4
LT		 5140 /* recovery follows the physical size of devices */
5141 max_sectors = mddev->size << 1;
5fd6c1dc
N		 5142 j = MaxSector;
5143 ITERATE_RDEV(mddev,rdev,rtmp)
5144 if (rdev->raid_disk >= 0 &&
5145 !test_bit(Faulty, &rdev->flags) &&
5146 !test_bit(In_sync, &rdev->flags) &&
5147 rdev->recovery_offset < j)
5148 j = rdev->recovery_offset;
5149 }
1da177e4
LT		 5150
5151 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
5152 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
88202a0c 5153 " %d KB/sec/disc.\n", speed_min(mddev));
338cec32 5154 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
1da177e4 5155 "(but not more than %d KB/sec) for reconstruction.\n",
88202a0c 5156 speed_max(mddev));
1da177e4
LT		 5157
5158 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5159
57afd89f 5160 io_sectors = 0;
1da177e4
LT		 5161 for (m = 0; m < SYNC_MARKS; m++) {
5162 mark[m] = jiffies;
57afd89f 5163 mark_cnt[m] = io_sectors;
1da177e4
LT		 5164 }
5165 last_mark = 0;
5166 mddev->resync_mark = mark[last_mark];
5167 mddev->resync_mark_cnt = mark_cnt[last_mark];
5168
5169 /*
5170 * Tune reconstruction:
5171 */
5172 window = 32*(PAGE_SIZE/512);
5173 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5174 window/2,(unsigned long long) max_sectors/2);
5175
5176 atomic_set(&mddev->recovery_active, 0);
5177 init_waitqueue_head(&mddev->recovery_wait);
5178 last_check = 0;
5179
5180 if (j>2) {
5181 printk(KERN_INFO
5182 "md: resuming recovery of %s from checkpoint.\n",
5183 mdname(mddev));
5184 mddev->curr_resync = j;
5185 }
5186
5187 while (j < max_sectors) {
57afd89f 5188 sector_t sectors;
1da177e4 5189
57afd89f
N		 5190 skipped = 0;
5191 sectors = mddev->pers->sync_request(mddev, j, &skipped,
88202a0c 5192 currspeed < speed_min(mddev));
57afd89f 5193 if (sectors == 0) {
1da177e4
LT		 5194 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5195 goto out;
5196 }
57afd89f
N		 5197
5198 if (!skipped) { /* actual IO requested */
5199 io_sectors += sectors;
5200 atomic_add(sectors, &mddev->recovery_active);
5201 }
5202
1da177e4
LT		 5203 j += sectors;
5204 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5205 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5206 if (last_check == 0)
5207 /* this is the earliers that rebuilt will be
5208 * visible in /proc/mdstat
5209 */
5210 md_new_event(mddev);
57afd89f
N		 5211
5212 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5213 continue;
5214
57afd89f 5215 last_check = io_sectors;
1da177e4
LT		 5216
5217 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5218 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5219 break;
5220
5221 repeat:
5222 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5223 /* step marks */
5224 int next = (last_mark+1) % SYNC_MARKS;
5225
5226 mddev->resync_mark = mark[next];
5227 mddev->resync_mark_cnt = mark_cnt[next];
5228 mark[next] = jiffies;
57afd89f 5229 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5230 last_mark = next;
5231 }
5232
5233
787453c2 5234 if (kthread_should_stop()) {
1da177e4
LT		 5235 /*
5236 * got a signal, exit.
5237 */
5238 printk(KERN_INFO
5239 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 5240 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5241 goto out;
5242 }
5243
5244 /*
5245 * this loop exits only if either when we are slower than
5246 * the 'hard' speed limit, or the system was IO-idle for
5247 * a jiffy.
5248 * the system might be non-idle CPU-wise, but we only care
5249 * about not overloading the IO subsystem. (things like an
5250 * e2fsck being done on the RAID array should execute fast)
5251 */
5252 mddev->queue->unplug_fn(mddev->queue);
5253 cond_resched();
5254
57afd89f
N		 5255 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5256 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5257
88202a0c
N		 5258 if (currspeed > speed_min(mddev)) {
5259 if ((currspeed > speed_max(mddev)) ||
1da177e4 5260 !is_mddev_idle(mddev)) {
c0e48521 5261 msleep(500);
1da177e4
LT		 5262 goto repeat;
5263 }
5264 }
5265 }
5266 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
5267 /*
5268 * this also signals 'finished resyncing' to md_stop
5269 */
5270 out:
5271 mddev->queue->unplug_fn(mddev->queue);
5272
5273 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5274
5275 /* tell personality that we are finished */
57afd89f 5276 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5277
5278 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1
N		 5279 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
5280 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5281 mddev->curr_resync > 2) {
5282 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5283 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5284 if (mddev->curr_resync >= mddev->recovery_cp) {
5285 printk(KERN_INFO
5286 "md: checkpointing recovery of %s.\n",
5287 mdname(mddev));
5288 mddev->recovery_cp = mddev->curr_resync;
5289 }
5290 } else
5291 mddev->recovery_cp = MaxSector;
5292 } else {
5293 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5294 mddev->curr_resync = MaxSector;
5295 ITERATE_RDEV(mddev,rdev,rtmp)
5296 if (rdev->raid_disk >= 0 &&
5297 !test_bit(Faulty, &rdev->flags) &&
5298 !test_bit(In_sync, &rdev->flags) &&
5299 rdev->recovery_offset < mddev->curr_resync)
5300 rdev->recovery_offset = mddev->curr_resync;
5301 mddev->sb_dirty = 1;
5302 }
1da177e4
LT		 5303 }
5304
1da177e4
LT		 5305 skip:
5306 mddev->curr_resync = 0;
5307 wake_up(&resync_wait);
5308 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5309 md_wakeup_thread(mddev->thread);
5310}
29269553 5311EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5312
5313
5314/*
5315 * This routine is regularly called by all per-raid-array threads to
5316 * deal with generic issues like resync and super-block update.
5317 * Raid personalities that don't have a thread (linear/raid0) do not
5318 * need this as they never do any recovery or update the superblock.
5319 *
5320 * It does not do any resync itself, but rather "forks" off other threads
5321 * to do that as needed.
5322 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5323 * "->recovery" and create a thread at ->sync_thread.
5324 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5325 * and wakeups up this thread which will reap the thread and finish up.
5326 * This thread also removes any faulty devices (with nr_pending == 0).
5327 *
5328 * The overall approach is:
5329 * 1/ if the superblock needs updating, update it.
5330 * 2/ If a recovery thread is running, don't do anything else.
5331 * 3/ If recovery has finished, clean up, possibly marking spares active.
5332 * 4/ If there are any faulty devices, remove them.
5333 * 5/ If array is degraded, try to add spares devices
5334 * 6/ If array has spares or is not in-sync, start a resync thread.
5335 */
5336void md_check_recovery(mddev_t *mddev)
5337{
5338 mdk_rdev_t *rdev;
5339 struct list_head *rtmp;
5340
5341
5f40402d
N		 5342 if (mddev->bitmap)
5343 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5344
5345 if (mddev->ro)
5346 return;
fca4d848
N		 5347
5348 if (signal_pending(current)) {
5349 if (mddev->pers->sync_request) {
5350 printk(KERN_INFO "md: %s in immediate safe mode\n",
5351 mdname(mddev));
5352 mddev->safemode = 2;
5353 }
5354 flush_signals(current);
5355 }
5356
1da177e4
LT		 5357 if ( ! (
5358 mddev->sb_dirty ||
5359 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5360 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5361 (mddev->safemode == 1) ||
5362 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5363 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5364 ))
5365 return;
fca4d848 5366
df5b89b3 5367 if (mddev_trylock(mddev)) {
1da177e4 5368 int spares =0;
fca4d848 5369
a9701a30 5370 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5371 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5372 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5373 mddev->in_sync = 1;
42543769 5374 mddev->sb_dirty = 3;
fca4d848
N		 5375 }
5376 if (mddev->safemode == 1)
5377 mddev->safemode = 0;
a9701a30 5378 spin_unlock_irq(&mddev->write_lock);
fca4d848 5379
1da177e4
LT		 5380 if (mddev->sb_dirty)
5381 md_update_sb(mddev);
06d91a5f 5382
06d91a5f 5383
1da177e4
LT		 5384 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5385 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5386 /* resync/recovery still happening */
5387 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5388 goto unlock;
5389 }
5390 if (mddev->sync_thread) {
5391 /* resync has finished, collect result */
5392 md_unregister_thread(mddev->sync_thread);
5393 mddev->sync_thread = NULL;
5394 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5395 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5396 /* success...*/
5397 /* activate any spares */
5398 mddev->pers->spare_active(mddev);
5399 }
5400 md_update_sb(mddev);
41158c7e
N		 5401
5402 /* if array is no-longer degraded, then any saved_raid_disk
5403 * information must be scrapped
5404 */
5405 if (!mddev->degraded)
5406 ITERATE_RDEV(mddev,rdev,rtmp)
5407 rdev->saved_raid_disk = -1;
5408
1da177e4
LT		 5409 mddev->recovery = 0;
5410 /* flag recovery needed just to double check */
5411 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5412 md_new_event(mddev);
1da177e4
LT		 5413 goto unlock;
5414 }
24dd469d
N		 5415 /* Clear some bits that don't mean anything, but
5416 * might be left set
5417 */
5418 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5419 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5420 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5421 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5422
5fd6c1dc
N		 5423 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5424 goto unlock;
1da177e4
LT		 5425 /* no recovery is running.
5426 * remove any failed drives, then
5427 * add spares if possible.
5428 * Spare are also removed and re-added, to allow
5429 * the personality to fail the re-add.
5430 */
5431 ITERATE_RDEV(mddev,rdev,rtmp)
5432 if (rdev->raid_disk >= 0 &&
b2d444d7 5433 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 5434 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 5435 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5436 char nm[20];
5437 sprintf(nm,"rd%d", rdev->raid_disk);
5438 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 5439 rdev->raid_disk = -1;
86e6ffdd 5440 }
1da177e4
LT		 5441 }
5442
5443 if (mddev->degraded) {
5444 ITERATE_RDEV(mddev,rdev,rtmp)
5445 if (rdev->raid_disk < 0
b2d444d7 5446 && !test_bit(Faulty, &rdev->flags)) {
5fd6c1dc 5447 rdev->recovery_offset = 0;
86e6ffdd
N		 5448 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5449 char nm[20];
5450 sprintf(nm, "rd%d", rdev->raid_disk);
5451 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 5452 spares++;
d7603b7e 5453 md_new_event(mddev);
86e6ffdd 5454 } else
1da177e4
LT		 5455 break;
5456 }
5457 }
5458
24dd469d
N		 5459 if (spares) {
5460 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5461 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5462 } else if (mddev->recovery_cp < MaxSector) {
5463 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5464 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5465 /* nothing to be done ... */
1da177e4 5466 goto unlock;
24dd469d 5467
1da177e4
LT		 5468 if (mddev->pers->sync_request) {
5469 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5470 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5471 /* We are adding a device or devices to an array
5472 * which has the bitmap stored on all devices.
5473 * So make sure all bitmap pages get written
5474 */
5475 bitmap_write_all(mddev->bitmap);
5476 }
1da177e4
LT		 5477 mddev->sync_thread = md_register_thread(md_do_sync,
5478 mddev,
5479 "%s_resync");
5480 if (!mddev->sync_thread) {
5481 printk(KERN_ERR "%s: could not start resync"
5482 " thread...\n",
5483 mdname(mddev));
5484 /* leave the spares where they are, it shouldn't hurt */
5485 mddev->recovery = 0;
d7603b7e 5486 } else
1da177e4 5487 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5488 md_new_event(mddev);
1da177e4
LT		 5489 }
5490 unlock:
5491 mddev_unlock(mddev);
5492 }
5493}
5494
75c96f85
AB		 5495static int md_notify_reboot(struct notifier_block *this,
5496 unsigned long code, void *x)
1da177e4
LT		 5497{
5498 struct list_head *tmp;
5499 mddev_t *mddev;
5500
5501 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5502
5503 printk(KERN_INFO "md: stopping all md devices.\n");
5504
5505 ITERATE_MDDEV(mddev,tmp)
c71d4887 5506 if (mddev_trylock(mddev)) {
1da177e4 5507 do_md_stop (mddev, 1);
c71d4887
NB		 5508 mddev_unlock(mddev);
5509 }
1da177e4
LT		 5510 /*
5511 * certain more exotic SCSI devices are known to be
5512 * volatile wrt too early system reboots. While the
5513 * right place to handle this issue is the given
5514 * driver, we do want to have a safe RAID driver ...
5515 */
5516 mdelay(1000*1);
5517 }
5518 return NOTIFY_DONE;
5519}
5520
75c96f85 5521static struct notifier_block md_notifier = {
1da177e4
LT		 5522 .notifier_call = md_notify_reboot,
5523 .next = NULL,
5524 .priority = INT_MAX, /* before any real devices */
5525};
5526
5527static void md_geninit(void)
5528{
5529 struct proc_dir_entry *p;
5530
5531 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5532
5533 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5534 if (p)
5535 p->proc_fops = &md_seq_fops;
5536}
5537
75c96f85 5538static int __init md_init(void)
1da177e4 5539{
1da177e4
LT		 5540 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5541 " MD_SB_DISKS=%d\n",
5542 MD_MAJOR_VERSION, MD_MINOR_VERSION,
5543 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
bd926c63 5544 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
32a7627c 5545 BITMAP_MINOR);
1da177e4
LT		 5546
5547 if (register_blkdev(MAJOR_NR, "md"))
5548 return -1;
5549 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5550 unregister_blkdev(MAJOR_NR, "md");
5551 return -1;
5552 }
1da177e4
LT		 5553 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5554 md_probe, NULL, NULL);
5555 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5556 md_probe, NULL, NULL);
5557
1da177e4
LT		 5558 register_reboot_notifier(&md_notifier);
5559 raid_table_header = register_sysctl_table(raid_root_table, 1);
5560
5561 md_geninit();
5562 return (0);
5563}
5564
5565
5566#ifndef MODULE
5567
5568/*
5569 * Searches all registered partitions for autorun RAID arrays
5570 * at boot time.
5571 */
5572static dev_t detected_devices[128];
5573static int dev_cnt;
5574
5575void md_autodetect_dev(dev_t dev)
5576{
5577 if (dev_cnt >= 0 && dev_cnt < 127)
5578 detected_devices[dev_cnt++] = dev;
5579}
5580
5581
5582static void autostart_arrays(int part)
5583{
5584 mdk_rdev_t *rdev;
5585 int i;
5586
5587 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5588
5589 for (i = 0; i < dev_cnt; i++) {
5590 dev_t dev = detected_devices[i];
5591
5592 rdev = md_import_device(dev,0, 0);
5593 if (IS_ERR(rdev))
5594 continue;
5595
b2d444d7 5596 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5597 MD_BUG();
5598 continue;
5599 }
5600 list_add(&rdev->same_set, &pending_raid_disks);
5601 }
5602 dev_cnt = 0;
5603
5604 autorun_devices(part);
5605}
5606
5607#endif
5608
5609static __exit void md_exit(void)
5610{
5611 mddev_t *mddev;
5612 struct list_head *tmp;
8ab5e4c1 5613
1da177e4
LT		 5614 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5615 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
1da177e4
LT		 5616
5617 unregister_blkdev(MAJOR_NR,"md");
5618 unregister_blkdev(mdp_major, "mdp");
5619 unregister_reboot_notifier(&md_notifier);
5620 unregister_sysctl_table(raid_table_header);
5621 remove_proc_entry("mdstat", NULL);
5622 ITERATE_MDDEV(mddev,tmp) {
5623 struct gendisk *disk = mddev->gendisk;
5624 if (!disk)
5625 continue;
5626 export_array(mddev);
5627 del_gendisk(disk);
5628 put_disk(disk);
5629 mddev->gendisk = NULL;
5630 mddev_put(mddev);
5631 }
5632}
5633
5634module_init(md_init)
5635module_exit(md_exit)
5636
f91de92e
N		 5637static int get_ro(char *buffer, struct kernel_param *kp)
5638{
5639 return sprintf(buffer, "%d", start_readonly);
5640}
5641static int set_ro(const char *val, struct kernel_param *kp)
5642{
5643 char *e;
5644 int num = simple_strtoul(val, &e, 10);
5645 if (*val && (*e == '\0' || *e == '\n')) {
5646 start_readonly = num;
4dbcdc75 5647 return 0;
f91de92e
N		 5648 }
5649 return -EINVAL;
5650}
5651
80ca3a44
N		 5652module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5653module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 5654
f91de92e 5655
1da177e4
LT		 5656EXPORT_SYMBOL(register_md_personality);
5657EXPORT_SYMBOL(unregister_md_personality);
5658EXPORT_SYMBOL(md_error);
5659EXPORT_SYMBOL(md_done_sync);
5660EXPORT_SYMBOL(md_write_start);
5661EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 5662EXPORT_SYMBOL(md_register_thread);
5663EXPORT_SYMBOL(md_unregister_thread);
5664EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 5665EXPORT_SYMBOL(md_check_recovery);
5666MODULE_LICENSE("GPL");
aa1595e9 5667MODULE_ALIAS("md");
72008652 5668MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Clone of davem's net-next-2.6 tree