]> 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: remove MAX_MD_DEVS which is an arbitrary limit
[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
850b2b42 1590static void md_update_sb(mddev_t * mddev, int force_change)
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);
84692195 1600
850b2b42
N		 1601 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1602 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1603 force_change = 1;
1604 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1605 /* just a clean<-> dirty transition, possibly leave spares alone,
1606 * though if events isn't the right even/odd, we will have to do
1607 * spares after all
1608 */
1609 nospares = 1;
1610 if (force_change)
1611 nospares = 0;
1612 if (mddev->degraded)
84692195
N		 1613 /* If the array is degraded, then skipping spares is both
1614 * dangerous and fairly pointless.
1615 * Dangerous because a device that was removed from the array
1616 * might have a event_count that still looks up-to-date,
1617 * so it can be re-added without a resync.
1618 * Pointless because if there are any spares to skip,
1619 * then a recovery will happen and soon that array won't
1620 * be degraded any more and the spare can go back to sleep then.
1621 */
850b2b42 1622 nospares = 0;
84692195 1623
06d91a5f 1624 sync_req = mddev->in_sync;
1da177e4 1625 mddev->utime = get_seconds();
42543769
N		 1626
1627 /* If this is just a dirty<->clean transition, and the array is clean
1628 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1629 if (nospares
42543769
N		 1630 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1631 && (mddev->events & 1))
1632 mddev->events--;
1633 else {
1634 /* otherwise we have to go forward and ... */
1635 mddev->events ++;
1636 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1637 /* .. if the array isn't clean, insist on an odd 'events' */
1638 if ((mddev->events&1)==0) {
1639 mddev->events++;
1640 nospares = 0;
1641 }
1642 } else {
1643 /* otherwise insist on an even 'events' (for clean states) */
1644 if ((mddev->events&1)) {
1645 mddev->events++;
1646 nospares = 0;
1647 }
1648 }
1649 }
1da177e4
LT		 1650
1651 if (!mddev->events) {
1652 /*
1653 * oops, this 64-bit counter should never wrap.
1654 * Either we are in around ~1 trillion A.C., assuming
1655 * 1 reboot per second, or we have a bug:
1656 */
1657 MD_BUG();
1658 mddev->events --;
1659 }
42543769 1660 sync_sbs(mddev, nospares);
1da177e4
LT		 1661
1662 /*
1663 * do not write anything to disk if using
1664 * nonpersistent superblocks
1665 */
06d91a5f 1666 if (!mddev->persistent) {
850b2b42 1667 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1668 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1669 wake_up(&mddev->sb_wait);
1da177e4 1670 return;
06d91a5f 1671 }
a9701a30 1672 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1673
1674 dprintk(KERN_INFO
1675 "md: updating %s RAID superblock on device (in sync %d)\n",
1676 mdname(mddev),mddev->in_sync);
1677
32a7627c 1678 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1679 ITERATE_RDEV(mddev,rdev,tmp) {
1680 char b[BDEVNAME_SIZE];
1681 dprintk(KERN_INFO "md: ");
42543769
N		 1682 if (rdev->sb_loaded != 1)
1683 continue; /* no noise on spare devices */
b2d444d7 1684 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1685 dprintk("(skipping faulty ");
1686
1687 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1688 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1689 md_super_write(mddev,rdev,
0002b271 1690 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1691 rdev->sb_page);
1692 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1693 bdevname(rdev->bdev,b),
1694 (unsigned long long)rdev->sb_offset);
42543769 1695 rdev->sb_events = mddev->events;
7bfa19f2 1696
1da177e4
LT		 1697 } else
1698 dprintk(")\n");
7bfa19f2 1699 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1700 /* only need to write one superblock... */
1701 break;
1702 }
a9701a30 1703 md_super_wait(mddev);
850b2b42 1704 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 1705
a9701a30 1706 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 1707 if (mddev->in_sync != sync_req ||
1708 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 1709 /* have to write it out again */
a9701a30 1710 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1711 goto repeat;
1712 }
850b2b42 1713 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1714 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1715 wake_up(&mddev->sb_wait);
06d91a5f 1716
1da177e4
LT		 1717}
1718
bce74dac
N		 1719/* words written to sysfs files may, or my not, be \n terminated.
1720 * We want to accept with case. For this we use cmd_match.
1721 */
1722static int cmd_match(const char *cmd, const char *str)
1723{
1724 /* See if cmd, written into a sysfs file, matches
1725 * str. They must either be the same, or cmd can
1726 * have a trailing newline
1727 */
1728 while (*cmd && *str && *cmd == *str) {
1729 cmd++;
1730 str++;
1731 }
1732 if (*cmd == '\n')
1733 cmd++;
1734 if (*str || *cmd)
1735 return 0;
1736 return 1;
1737}
1738
86e6ffdd
N		 1739struct rdev_sysfs_entry {
1740 struct attribute attr;
1741 ssize_t (*show)(mdk_rdev_t *, char *);
1742 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1743};
1744
1745static ssize_t
96de1e66 1746state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1747{
1748 char *sep = "";
1749 int len=0;
1750
b2d444d7 1751 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1752 len+= sprintf(page+len, "%sfaulty",sep);
1753 sep = ",";
1754 }
b2d444d7 1755 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1756 len += sprintf(page+len, "%sin_sync",sep);
1757 sep = ",";
1758 }
f655675b
N		 1759 if (test_bit(WriteMostly, &rdev->flags)) {
1760 len += sprintf(page+len, "%swrite_mostly",sep);
1761 sep = ",";
1762 }
b2d444d7
N		 1763 if (!test_bit(Faulty, &rdev->flags) &&
1764 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1765 len += sprintf(page+len, "%sspare", sep);
1766 sep = ",";
1767 }
1768 return len+sprintf(page+len, "\n");
1769}
1770
45dc2de1
N		 1771static ssize_t
1772state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1773{
1774 /* can write
1775 * faulty - simulates and error
1776 * remove - disconnects the device
f655675b
N		 1777 * writemostly - sets write_mostly
1778 * -writemostly - clears write_mostly
45dc2de1
N		 1779 */
1780 int err = -EINVAL;
1781 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1782 md_error(rdev->mddev, rdev);
1783 err = 0;
1784 } else if (cmd_match(buf, "remove")) {
1785 if (rdev->raid_disk >= 0)
1786 err = -EBUSY;
1787 else {
1788 mddev_t *mddev = rdev->mddev;
1789 kick_rdev_from_array(rdev);
850b2b42 1790 md_update_sb(mddev, 1);
45dc2de1
N		 1791 md_new_event(mddev);
1792 err = 0;
1793 }
f655675b
N		 1794 } else if (cmd_match(buf, "writemostly")) {
1795 set_bit(WriteMostly, &rdev->flags);
1796 err = 0;
1797 } else if (cmd_match(buf, "-writemostly")) {
1798 clear_bit(WriteMostly, &rdev->flags);
1799 err = 0;
45dc2de1
N		 1800 }
1801 return err ? err : len;
1802}
80ca3a44
N		 1803static struct rdev_sysfs_entry rdev_state =
1804__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd
N		 1805
1806static ssize_t
96de1e66 1807super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1808{
1809 if (rdev->sb_loaded && rdev->sb_size) {
1810 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1811 return rdev->sb_size;
1812 } else
1813 return 0;
1814}
96de1e66
N		 1815static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1816
4dbcdc75
N		 1817static ssize_t
1818errors_show(mdk_rdev_t *rdev, char *page)
1819{
1820 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1821}
1822
1823static ssize_t
1824errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1825{
1826 char *e;
1827 unsigned long n = simple_strtoul(buf, &e, 10);
1828 if (*buf && (*e == 0 || *e == '\n')) {
1829 atomic_set(&rdev->corrected_errors, n);
1830 return len;
1831 }
1832 return -EINVAL;
1833}
1834static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1835__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1836
014236d2
N		 1837static ssize_t
1838slot_show(mdk_rdev_t *rdev, char *page)
1839{
1840 if (rdev->raid_disk < 0)
1841 return sprintf(page, "none\n");
1842 else
1843 return sprintf(page, "%d\n", rdev->raid_disk);
1844}
1845
1846static ssize_t
1847slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1848{
1849 char *e;
1850 int slot = simple_strtoul(buf, &e, 10);
1851 if (strncmp(buf, "none", 4)==0)
1852 slot = -1;
1853 else if (e==buf || (*e && *e!= '\n'))
1854 return -EINVAL;
1855 if (rdev->mddev->pers)
1856 /* Cannot set slot in active array (yet) */
1857 return -EBUSY;
1858 if (slot >= rdev->mddev->raid_disks)
1859 return -ENOSPC;
1860 rdev->raid_disk = slot;
1861 /* assume it is working */
1862 rdev->flags = 0;
1863 set_bit(In_sync, &rdev->flags);
1864 return len;
1865}
1866
1867
1868static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1869__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1870
93c8cad0
N		 1871static ssize_t
1872offset_show(mdk_rdev_t *rdev, char *page)
1873{
6961ece4 1874 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1875}
1876
1877static ssize_t
1878offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1879{
1880 char *e;
1881 unsigned long long offset = simple_strtoull(buf, &e, 10);
1882 if (e==buf || (*e && *e != '\n'))
1883 return -EINVAL;
1884 if (rdev->mddev->pers)
1885 return -EBUSY;
1886 rdev->data_offset = offset;
1887 return len;
1888}
1889
1890static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1891__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1892
83303b61
N		 1893static ssize_t
1894rdev_size_show(mdk_rdev_t *rdev, char *page)
1895{
1896 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1897}
1898
1899static ssize_t
1900rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1901{
1902 char *e;
1903 unsigned long long size = simple_strtoull(buf, &e, 10);
1904 if (e==buf || (*e && *e != '\n'))
1905 return -EINVAL;
1906 if (rdev->mddev->pers)
1907 return -EBUSY;
1908 rdev->size = size;
1909 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1910 rdev->mddev->size = size;
1911 return len;
1912}
1913
1914static struct rdev_sysfs_entry rdev_size =
80ca3a44 1915__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 1916
86e6ffdd
N		 1917static struct attribute *rdev_default_attrs[] = {
1918 &rdev_state.attr,
1919 &rdev_super.attr,
4dbcdc75 1920 &rdev_errors.attr,
014236d2 1921 &rdev_slot.attr,
93c8cad0 1922 &rdev_offset.attr,
83303b61 1923 &rdev_size.attr,
86e6ffdd
N		 1924 NULL,
1925};
1926static ssize_t
1927rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1928{
1929 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1930 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1931
1932 if (!entry->show)
1933 return -EIO;
1934 return entry->show(rdev, page);
1935}
1936
1937static ssize_t
1938rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1939 const char *page, size_t length)
1940{
1941 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1942 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1943
1944 if (!entry->store)
1945 return -EIO;
67463acb
N		 1946 if (!capable(CAP_SYS_ADMIN))
1947 return -EACCES;
86e6ffdd
N		 1948 return entry->store(rdev, page, length);
1949}
1950
1951static void rdev_free(struct kobject *ko)
1952{
1953 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1954 kfree(rdev);
1955}
1956static struct sysfs_ops rdev_sysfs_ops = {
1957 .show = rdev_attr_show,
1958 .store = rdev_attr_store,
1959};
1960static struct kobj_type rdev_ktype = {
1961 .release = rdev_free,
1962 .sysfs_ops = &rdev_sysfs_ops,
1963 .default_attrs = rdev_default_attrs,
1964};
1965
1da177e4
LT		 1966/*
1967 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1968 *
1969 * mark the device faulty if:
1970 *
1971 * - the device is nonexistent (zero size)
1972 * - the device has no valid superblock
1973 *
1974 * a faulty rdev _never_ has rdev->sb set.
1975 */
1976static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1977{
1978 char b[BDEVNAME_SIZE];
1979 int err;
1980 mdk_rdev_t *rdev;
1981 sector_t size;
1982
9ffae0cf 1983 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 1984 if (!rdev) {
1985 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1986 return ERR_PTR(-ENOMEM);
1987 }
1da177e4
LT		 1988
1989 if ((err = alloc_disk_sb(rdev)))
1990 goto abort_free;
1991
1992 err = lock_rdev(rdev, newdev);
1993 if (err)
1994 goto abort_free;
1995
86e6ffdd
N		 1996 rdev->kobj.parent = NULL;
1997 rdev->kobj.ktype = &rdev_ktype;
1998 kobject_init(&rdev->kobj);
1999
1da177e4 2000 rdev->desc_nr = -1;
b2d444d7 2001 rdev->flags = 0;
1da177e4 2002 rdev->data_offset = 0;
42543769 2003 rdev->sb_events = 0;
1da177e4 2004 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2005 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2006 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2007
2008 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2009 if (!size) {
2010 printk(KERN_WARNING
2011 "md: %s has zero or unknown size, marking faulty!\n",
2012 bdevname(rdev->bdev,b));
2013 err = -EINVAL;
2014 goto abort_free;
2015 }
2016
2017 if (super_format >= 0) {
2018 err = super_types[super_format].
2019 load_super(rdev, NULL, super_minor);
2020 if (err == -EINVAL) {
2021 printk(KERN_WARNING
2022 "md: %s has invalid sb, not importing!\n",
2023 bdevname(rdev->bdev,b));
2024 goto abort_free;
2025 }
2026 if (err < 0) {
2027 printk(KERN_WARNING
2028 "md: could not read %s's sb, not importing!\n",
2029 bdevname(rdev->bdev,b));
2030 goto abort_free;
2031 }
2032 }
2033 INIT_LIST_HEAD(&rdev->same_set);
2034
2035 return rdev;
2036
2037abort_free:
2038 if (rdev->sb_page) {
2039 if (rdev->bdev)
2040 unlock_rdev(rdev);
2041 free_disk_sb(rdev);
2042 }
2043 kfree(rdev);
2044 return ERR_PTR(err);
2045}
2046
2047/*
2048 * Check a full RAID array for plausibility
2049 */
2050
2051
a757e64c 2052static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2053{
2054 int i;
2055 struct list_head *tmp;
2056 mdk_rdev_t *rdev, *freshest;
2057 char b[BDEVNAME_SIZE];
2058
2059 freshest = NULL;
2060 ITERATE_RDEV(mddev,rdev,tmp)
2061 switch (super_types[mddev->major_version].
2062 load_super(rdev, freshest, mddev->minor_version)) {
2063 case 1:
2064 freshest = rdev;
2065 break;
2066 case 0:
2067 break;
2068 default:
2069 printk( KERN_ERR \
2070 "md: fatal superblock inconsistency in %s"
2071 " -- removing from array\n",
2072 bdevname(rdev->bdev,b));
2073 kick_rdev_from_array(rdev);
2074 }
2075
2076
2077 super_types[mddev->major_version].
2078 validate_super(mddev, freshest);
2079
2080 i = 0;
2081 ITERATE_RDEV(mddev,rdev,tmp) {
2082 if (rdev != freshest)
2083 if (super_types[mddev->major_version].
2084 validate_super(mddev, rdev)) {
2085 printk(KERN_WARNING "md: kicking non-fresh %s"
2086 " from array!\n",
2087 bdevname(rdev->bdev,b));
2088 kick_rdev_from_array(rdev);
2089 continue;
2090 }
2091 if (mddev->level == LEVEL_MULTIPATH) {
2092 rdev->desc_nr = i++;
2093 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2094 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 2095 }
2096 }
2097
2098
2099
2100 if (mddev->recovery_cp != MaxSector &&
2101 mddev->level >= 1)
2102 printk(KERN_ERR "md: %s: raid array is not clean"
2103 " -- starting background reconstruction\n",
2104 mdname(mddev));
2105
1da177e4
LT		 2106}
2107
16f17b39
N		 2108static ssize_t
2109safe_delay_show(mddev_t *mddev, char *page)
2110{
2111 int msec = (mddev->safemode_delay*1000)/HZ;
2112 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2113}
2114static ssize_t
2115safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2116{
2117 int scale=1;
2118 int dot=0;
2119 int i;
2120 unsigned long msec;
2121 char buf[30];
2122 char *e;
2123 /* remove a period, and count digits after it */
2124 if (len >= sizeof(buf))
2125 return -EINVAL;
2126 strlcpy(buf, cbuf, len);
2127 buf[len] = 0;
2128 for (i=0; i<len; i++) {
2129 if (dot) {
2130 if (isdigit(buf[i])) {
2131 buf[i-1] = buf[i];
2132 scale *= 10;
2133 }
2134 buf[i] = 0;
2135 } else if (buf[i] == '.') {
2136 dot=1;
2137 buf[i] = 0;
2138 }
2139 }
2140 msec = simple_strtoul(buf, &e, 10);
2141 if (e == buf || (*e && *e != '\n'))
2142 return -EINVAL;
2143 msec = (msec * 1000) / scale;
2144 if (msec == 0)
2145 mddev->safemode_delay = 0;
2146 else {
2147 mddev->safemode_delay = (msec*HZ)/1000;
2148 if (mddev->safemode_delay == 0)
2149 mddev->safemode_delay = 1;
2150 }
2151 return len;
2152}
2153static struct md_sysfs_entry md_safe_delay =
80ca3a44 2154__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2155
eae1701f 2156static ssize_t
96de1e66 2157level_show(mddev_t *mddev, char *page)
eae1701f 2158{
2604b703 2159 struct mdk_personality *p = mddev->pers;
d9d166c2 2160 if (p)
eae1701f 2161 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2162 else if (mddev->clevel[0])
2163 return sprintf(page, "%s\n", mddev->clevel);
2164 else if (mddev->level != LEVEL_NONE)
2165 return sprintf(page, "%d\n", mddev->level);
2166 else
2167 return 0;
eae1701f
N		 2168}
2169
d9d166c2
N		 2170static ssize_t
2171level_store(mddev_t *mddev, const char *buf, size_t len)
2172{
2173 int rv = len;
2174 if (mddev->pers)
2175 return -EBUSY;
2176 if (len == 0)
2177 return 0;
2178 if (len >= sizeof(mddev->clevel))
2179 return -ENOSPC;
2180 strncpy(mddev->clevel, buf, len);
2181 if (mddev->clevel[len-1] == '\n')
2182 len--;
2183 mddev->clevel[len] = 0;
2184 mddev->level = LEVEL_NONE;
2185 return rv;
2186}
2187
2188static struct md_sysfs_entry md_level =
80ca3a44 2189__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2190
d4dbd025
N		 2191
2192static ssize_t
2193layout_show(mddev_t *mddev, char *page)
2194{
2195 /* just a number, not meaningful for all levels */
2196 return sprintf(page, "%d\n", mddev->layout);
2197}
2198
2199static ssize_t
2200layout_store(mddev_t *mddev, const char *buf, size_t len)
2201{
2202 char *e;
2203 unsigned long n = simple_strtoul(buf, &e, 10);
2204 if (mddev->pers)
2205 return -EBUSY;
2206
2207 if (!*buf || (*e && *e != '\n'))
2208 return -EINVAL;
2209
2210 mddev->layout = n;
2211 return len;
2212}
2213static struct md_sysfs_entry md_layout =
80ca3a44 2214__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2215
2216
eae1701f 2217static ssize_t
96de1e66 2218raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2219{
bb636547
N		 2220 if (mddev->raid_disks == 0)
2221 return 0;
eae1701f
N		 2222 return sprintf(page, "%d\n", mddev->raid_disks);
2223}
2224
da943b99
N		 2225static int update_raid_disks(mddev_t *mddev, int raid_disks);
2226
2227static ssize_t
2228raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2229{
2230 /* can only set raid_disks if array is not yet active */
2231 char *e;
2232 int rv = 0;
2233 unsigned long n = simple_strtoul(buf, &e, 10);
2234
2235 if (!*buf || (*e && *e != '\n'))
2236 return -EINVAL;
2237
2238 if (mddev->pers)
2239 rv = update_raid_disks(mddev, n);
2240 else
2241 mddev->raid_disks = n;
2242 return rv ? rv : len;
2243}
2244static struct md_sysfs_entry md_raid_disks =
80ca3a44 2245__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2246
3b34380a
N		 2247static ssize_t
2248chunk_size_show(mddev_t *mddev, char *page)
2249{
2250 return sprintf(page, "%d\n", mddev->chunk_size);
2251}
2252
2253static ssize_t
2254chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2255{
2256 /* can only set chunk_size if array is not yet active */
2257 char *e;
2258 unsigned long n = simple_strtoul(buf, &e, 10);
2259
2260 if (mddev->pers)
2261 return -EBUSY;
2262 if (!*buf || (*e && *e != '\n'))
2263 return -EINVAL;
2264
2265 mddev->chunk_size = n;
2266 return len;
2267}
2268static struct md_sysfs_entry md_chunk_size =
80ca3a44 2269__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2270
a94213b1
N		 2271static ssize_t
2272resync_start_show(mddev_t *mddev, char *page)
2273{
2274 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2275}
2276
2277static ssize_t
2278resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2279{
2280 /* can only set chunk_size if array is not yet active */
2281 char *e;
2282 unsigned long long n = simple_strtoull(buf, &e, 10);
2283
2284 if (mddev->pers)
2285 return -EBUSY;
2286 if (!*buf || (*e && *e != '\n'))
2287 return -EINVAL;
2288
2289 mddev->recovery_cp = n;
2290 return len;
2291}
2292static struct md_sysfs_entry md_resync_start =
80ca3a44 2293__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2294
9e653b63
N		 2295/*
2296 * The array state can be:
2297 *
2298 * clear
2299 * No devices, no size, no level
2300 * Equivalent to STOP_ARRAY ioctl
2301 * inactive
2302 * May have some settings, but array is not active
2303 * all IO results in error
2304 * When written, doesn't tear down array, but just stops it
2305 * suspended (not supported yet)
2306 * All IO requests will block. The array can be reconfigured.
2307 * Writing this, if accepted, will block until array is quiessent
2308 * readonly
2309 * no resync can happen. no superblocks get written.
2310 * write requests fail
2311 * read-auto
2312 * like readonly, but behaves like 'clean' on a write request.
2313 *
2314 * clean - no pending writes, but otherwise active.
2315 * When written to inactive array, starts without resync
2316 * If a write request arrives then
2317 * if metadata is known, mark 'dirty' and switch to 'active'.
2318 * if not known, block and switch to write-pending
2319 * If written to an active array that has pending writes, then fails.
2320 * active
2321 * fully active: IO and resync can be happening.
2322 * When written to inactive array, starts with resync
2323 *
2324 * write-pending
2325 * clean, but writes are blocked waiting for 'active' to be written.
2326 *
2327 * active-idle
2328 * like active, but no writes have been seen for a while (100msec).
2329 *
2330 */
2331enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2332 write_pending, active_idle, bad_word};
05381954 2333static char *array_states[] = {
9e653b63
N		 2334 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2335 "write-pending", "active-idle", NULL };
2336
2337static int match_word(const char *word, char **list)
2338{
2339 int n;
2340 for (n=0; list[n]; n++)
2341 if (cmd_match(word, list[n]))
2342 break;
2343 return n;
2344}
2345
2346static ssize_t
2347array_state_show(mddev_t *mddev, char *page)
2348{
2349 enum array_state st = inactive;
2350
2351 if (mddev->pers)
2352 switch(mddev->ro) {
2353 case 1:
2354 st = readonly;
2355 break;
2356 case 2:
2357 st = read_auto;
2358 break;
2359 case 0:
2360 if (mddev->in_sync)
2361 st = clean;
2362 else if (mddev->safemode)
2363 st = active_idle;
2364 else
2365 st = active;
2366 }
2367 else {
2368 if (list_empty(&mddev->disks) &&
2369 mddev->raid_disks == 0 &&
2370 mddev->size == 0)
2371 st = clear;
2372 else
2373 st = inactive;
2374 }
2375 return sprintf(page, "%s\n", array_states[st]);
2376}
2377
2378static int do_md_stop(mddev_t * mddev, int ro);
2379static int do_md_run(mddev_t * mddev);
2380static int restart_array(mddev_t *mddev);
2381
2382static ssize_t
2383array_state_store(mddev_t *mddev, const char *buf, size_t len)
2384{
2385 int err = -EINVAL;
2386 enum array_state st = match_word(buf, array_states);
2387 switch(st) {
2388 case bad_word:
2389 break;
2390 case clear:
2391 /* stopping an active array */
2392 if (mddev->pers) {
2393 if (atomic_read(&mddev->active) > 1)
2394 return -EBUSY;
2395 err = do_md_stop(mddev, 0);
2396 }
2397 break;
2398 case inactive:
2399 /* stopping an active array */
2400 if (mddev->pers) {
2401 if (atomic_read(&mddev->active) > 1)
2402 return -EBUSY;
2403 err = do_md_stop(mddev, 2);
2404 }
2405 break;
2406 case suspended:
2407 break; /* not supported yet */
2408 case readonly:
2409 if (mddev->pers)
2410 err = do_md_stop(mddev, 1);
2411 else {
2412 mddev->ro = 1;
2413 err = do_md_run(mddev);
2414 }
2415 break;
2416 case read_auto:
2417 /* stopping an active array */
2418 if (mddev->pers) {
2419 err = do_md_stop(mddev, 1);
2420 if (err == 0)
2421 mddev->ro = 2; /* FIXME mark devices writable */
2422 } else {
2423 mddev->ro = 2;
2424 err = do_md_run(mddev);
2425 }
2426 break;
2427 case clean:
2428 if (mddev->pers) {
2429 restart_array(mddev);
2430 spin_lock_irq(&mddev->write_lock);
2431 if (atomic_read(&mddev->writes_pending) == 0) {
2432 mddev->in_sync = 1;
850b2b42 2433 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2434 }
2435 spin_unlock_irq(&mddev->write_lock);
2436 } else {
2437 mddev->ro = 0;
2438 mddev->recovery_cp = MaxSector;
2439 err = do_md_run(mddev);
2440 }
2441 break;
2442 case active:
2443 if (mddev->pers) {
2444 restart_array(mddev);
850b2b42 2445 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2446 wake_up(&mddev->sb_wait);
2447 err = 0;
2448 } else {
2449 mddev->ro = 0;
2450 err = do_md_run(mddev);
2451 }
2452 break;
2453 case write_pending:
2454 case active_idle:
2455 /* these cannot be set */
2456 break;
2457 }
2458 if (err)
2459 return err;
2460 else
2461 return len;
2462}
80ca3a44
N		 2463static struct md_sysfs_entry md_array_state =
2464__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2465
6d7ff738
N		 2466static ssize_t
2467null_show(mddev_t *mddev, char *page)
2468{
2469 return -EINVAL;
2470}
2471
2472static ssize_t
2473new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2474{
2475 /* buf must be %d:%d\n? giving major and minor numbers */
2476 /* The new device is added to the array.
2477 * If the array has a persistent superblock, we read the
2478 * superblock to initialise info and check validity.
2479 * Otherwise, only checking done is that in bind_rdev_to_array,
2480 * which mainly checks size.
2481 */
2482 char *e;
2483 int major = simple_strtoul(buf, &e, 10);
2484 int minor;
2485 dev_t dev;
2486 mdk_rdev_t *rdev;
2487 int err;
2488
2489 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2490 return -EINVAL;
2491 minor = simple_strtoul(e+1, &e, 10);
2492 if (*e && *e != '\n')
2493 return -EINVAL;
2494 dev = MKDEV(major, minor);
2495 if (major != MAJOR(dev) ||
2496 minor != MINOR(dev))
2497 return -EOVERFLOW;
2498
2499
2500 if (mddev->persistent) {
2501 rdev = md_import_device(dev, mddev->major_version,
2502 mddev->minor_version);
2503 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2504 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2505 mdk_rdev_t, same_set);
2506 err = super_types[mddev->major_version]
2507 .load_super(rdev, rdev0, mddev->minor_version);
2508 if (err < 0)
2509 goto out;
2510 }
2511 } else
2512 rdev = md_import_device(dev, -1, -1);
2513
2514 if (IS_ERR(rdev))
2515 return PTR_ERR(rdev);
2516 err = bind_rdev_to_array(rdev, mddev);
2517 out:
2518 if (err)
2519 export_rdev(rdev);
2520 return err ? err : len;
2521}
2522
2523static struct md_sysfs_entry md_new_device =
80ca3a44 2524__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2525
9b1d1dac
PC		 2526static ssize_t
2527bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2528{
2529 char *end;
2530 unsigned long chunk, end_chunk;
2531
2532 if (!mddev->bitmap)
2533 goto out;
2534 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2535 while (*buf) {
2536 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2537 if (buf == end) break;
2538 if (*end == '-') { /* range */
2539 buf = end + 1;
2540 end_chunk = simple_strtoul(buf, &end, 0);
2541 if (buf == end) break;
2542 }
2543 if (*end && !isspace(*end)) break;
2544 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2545 buf = end;
2546 while (isspace(*buf)) buf++;
2547 }
2548 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2549out:
2550 return len;
2551}
2552
2553static struct md_sysfs_entry md_bitmap =
2554__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2555
a35b0d69
N		 2556static ssize_t
2557size_show(mddev_t *mddev, char *page)
2558{
2559 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2560}
2561
2562static int update_size(mddev_t *mddev, unsigned long size);
2563
2564static ssize_t
2565size_store(mddev_t *mddev, const char *buf, size_t len)
2566{
2567 /* If array is inactive, we can reduce the component size, but
2568 * not increase it (except from 0).
2569 * If array is active, we can try an on-line resize
2570 */
2571 char *e;
2572 int err = 0;
2573 unsigned long long size = simple_strtoull(buf, &e, 10);
2574 if (!*buf || *buf == '\n' ||
2575 (*e && *e != '\n'))
2576 return -EINVAL;
2577
2578 if (mddev->pers) {
2579 err = update_size(mddev, size);
850b2b42 2580 md_update_sb(mddev, 1);
a35b0d69
N		 2581 } else {
2582 if (mddev->size == 0 ||
2583 mddev->size > size)
2584 mddev->size = size;
2585 else
2586 err = -ENOSPC;
2587 }
2588 return err ? err : len;
2589}
2590
2591static struct md_sysfs_entry md_size =
80ca3a44 2592__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2593
8bb93aac
N		 2594
2595/* Metdata version.
2596 * This is either 'none' for arrays with externally managed metadata,
2597 * or N.M for internally known formats
2598 */
2599static ssize_t
2600metadata_show(mddev_t *mddev, char *page)
2601{
2602 if (mddev->persistent)
2603 return sprintf(page, "%d.%d\n",
2604 mddev->major_version, mddev->minor_version);
2605 else
2606 return sprintf(page, "none\n");
2607}
2608
2609static ssize_t
2610metadata_store(mddev_t *mddev, const char *buf, size_t len)
2611{
2612 int major, minor;
2613 char *e;
2614 if (!list_empty(&mddev->disks))
2615 return -EBUSY;
2616
2617 if (cmd_match(buf, "none")) {
2618 mddev->persistent = 0;
2619 mddev->major_version = 0;
2620 mddev->minor_version = 90;
2621 return len;
2622 }
2623 major = simple_strtoul(buf, &e, 10);
2624 if (e==buf || *e != '.')
2625 return -EINVAL;
2626 buf = e+1;
2627 minor = simple_strtoul(buf, &e, 10);
2628 if (e==buf || *e != '\n')
2629 return -EINVAL;
2630 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2631 super_types[major].name == NULL)
2632 return -ENOENT;
2633 mddev->major_version = major;
2634 mddev->minor_version = minor;
2635 mddev->persistent = 1;
2636 return len;
2637}
2638
2639static struct md_sysfs_entry md_metadata =
80ca3a44 2640__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2641
24dd469d 2642static ssize_t
7eec314d 2643action_show(mddev_t *mddev, char *page)
24dd469d 2644{
7eec314d 2645 char *type = "idle";
31399d9e
N		 2646 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2647 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
ccfcc3c1
N		 2648 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2649 type = "reshape";
2650 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2651 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2652 type = "resync";
2653 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2654 type = "check";
2655 else
2656 type = "repair";
2657 } else
2658 type = "recover";
2659 }
2660 return sprintf(page, "%s\n", type);
2661}
2662
2663static ssize_t
7eec314d 2664action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2665{
7eec314d
N		 2666 if (!mddev->pers || !mddev->pers->sync_request)
2667 return -EINVAL;
2668
bce74dac 2669 if (cmd_match(page, "idle")) {
7eec314d
N		 2670 if (mddev->sync_thread) {
2671 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2672 md_unregister_thread(mddev->sync_thread);
2673 mddev->sync_thread = NULL;
2674 mddev->recovery = 0;
2675 }
03c902e1
N		 2676 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2677 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2678 return -EBUSY;
03c902e1 2679 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2680 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2681 else if (cmd_match(page, "reshape")) {
2682 int err;
2683 if (mddev->pers->start_reshape == NULL)
2684 return -EINVAL;
2685 err = mddev->pers->start_reshape(mddev);
2686 if (err)
2687 return err;
2688 } else {
bce74dac 2689 if (cmd_match(page, "check"))
7eec314d 2690 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2691 else if (!cmd_match(page, "repair"))
7eec314d
N		 2692 return -EINVAL;
2693 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2694 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2695 }
03c902e1 2696 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2697 md_wakeup_thread(mddev->thread);
2698 return len;
2699}
2700
9d88883e 2701static ssize_t
96de1e66 2702mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2703{
2704 return sprintf(page, "%llu\n",
2705 (unsigned long long) mddev->resync_mismatches);
2706}
2707
80ca3a44
N		 2708static struct md_sysfs_entry md_scan_mode =
2709__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2710
96de1e66 2711
80ca3a44 2712static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2713
88202a0c
N		 2714static ssize_t
2715sync_min_show(mddev_t *mddev, char *page)
2716{
2717 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2718 mddev->sync_speed_min ? "local": "system");
2719}
2720
2721static ssize_t
2722sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2723{
2724 int min;
2725 char *e;
2726 if (strncmp(buf, "system", 6)==0) {
2727 mddev->sync_speed_min = 0;
2728 return len;
2729 }
2730 min = simple_strtoul(buf, &e, 10);
2731 if (buf == e || (*e && *e != '\n') || min <= 0)
2732 return -EINVAL;
2733 mddev->sync_speed_min = min;
2734 return len;
2735}
2736
2737static struct md_sysfs_entry md_sync_min =
2738__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2739
2740static ssize_t
2741sync_max_show(mddev_t *mddev, char *page)
2742{
2743 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2744 mddev->sync_speed_max ? "local": "system");
2745}
2746
2747static ssize_t
2748sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2749{
2750 int max;
2751 char *e;
2752 if (strncmp(buf, "system", 6)==0) {
2753 mddev->sync_speed_max = 0;
2754 return len;
2755 }
2756 max = simple_strtoul(buf, &e, 10);
2757 if (buf == e || (*e && *e != '\n') || max <= 0)
2758 return -EINVAL;
2759 mddev->sync_speed_max = max;
2760 return len;
2761}
2762
2763static struct md_sysfs_entry md_sync_max =
2764__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2765
2766
2767static ssize_t
2768sync_speed_show(mddev_t *mddev, char *page)
2769{
2770 unsigned long resync, dt, db;
ff4e8d9a 2771 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2772 dt = ((jiffies - mddev->resync_mark) / HZ);
2773 if (!dt) dt++;
2774 db = resync - (mddev->resync_mark_cnt);
2775 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2776}
2777
80ca3a44 2778static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2779
2780static ssize_t
2781sync_completed_show(mddev_t *mddev, char *page)
2782{
2783 unsigned long max_blocks, resync;
2784
2785 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2786 max_blocks = mddev->resync_max_sectors;
2787 else
2788 max_blocks = mddev->size << 1;
2789
2790 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2791 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2792}
2793
80ca3a44 2794static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 2795
e464eafd
N		 2796static ssize_t
2797suspend_lo_show(mddev_t *mddev, char *page)
2798{
2799 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2800}
2801
2802static ssize_t
2803suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2804{
2805 char *e;
2806 unsigned long long new = simple_strtoull(buf, &e, 10);
2807
2808 if (mddev->pers->quiesce == NULL)
2809 return -EINVAL;
2810 if (buf == e || (*e && *e != '\n'))
2811 return -EINVAL;
2812 if (new >= mddev->suspend_hi ||
2813 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2814 mddev->suspend_lo = new;
2815 mddev->pers->quiesce(mddev, 2);
2816 return len;
2817 } else
2818 return -EINVAL;
2819}
2820static struct md_sysfs_entry md_suspend_lo =
2821__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2822
2823
2824static ssize_t
2825suspend_hi_show(mddev_t *mddev, char *page)
2826{
2827 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2828}
2829
2830static ssize_t
2831suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2832{
2833 char *e;
2834 unsigned long long new = simple_strtoull(buf, &e, 10);
2835
2836 if (mddev->pers->quiesce == NULL)
2837 return -EINVAL;
2838 if (buf == e || (*e && *e != '\n'))
2839 return -EINVAL;
2840 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2841 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2842 mddev->suspend_hi = new;
2843 mddev->pers->quiesce(mddev, 1);
2844 mddev->pers->quiesce(mddev, 0);
2845 return len;
2846 } else
2847 return -EINVAL;
2848}
2849static struct md_sysfs_entry md_suspend_hi =
2850__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2851
2852
eae1701f
N		 2853static struct attribute *md_default_attrs[] = {
2854 &md_level.attr,
d4dbd025 2855 &md_layout.attr,
eae1701f 2856 &md_raid_disks.attr,
3b34380a 2857 &md_chunk_size.attr,
a35b0d69 2858 &md_size.attr,
a94213b1 2859 &md_resync_start.attr,
8bb93aac 2860 &md_metadata.attr,
6d7ff738 2861 &md_new_device.attr,
16f17b39 2862 &md_safe_delay.attr,
9e653b63 2863 &md_array_state.attr,
411036fa
N		 2864 NULL,
2865};
2866
2867static struct attribute *md_redundancy_attrs[] = {
24dd469d 2868 &md_scan_mode.attr,
9d88883e 2869 &md_mismatches.attr,
88202a0c
N		 2870 &md_sync_min.attr,
2871 &md_sync_max.attr,
2872 &md_sync_speed.attr,
2873 &md_sync_completed.attr,
e464eafd
N		 2874 &md_suspend_lo.attr,
2875 &md_suspend_hi.attr,
9b1d1dac 2876 &md_bitmap.attr,
eae1701f
N		 2877 NULL,
2878};
411036fa
N		 2879static struct attribute_group md_redundancy_group = {
2880 .name = NULL,
2881 .attrs = md_redundancy_attrs,
2882};
2883
eae1701f
N		 2884
2885static ssize_t
2886md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2887{
2888 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2889 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2890 ssize_t rv;
eae1701f
N		 2891
2892 if (!entry->show)
2893 return -EIO;
5dc5cf7d
IM		 2894 rv = mddev_lock(mddev);
2895 if (!rv) {
2896 rv = entry->show(mddev, page);
2897 mddev_unlock(mddev);
2898 }
96de1e66 2899 return rv;
eae1701f
N		 2900}
2901
2902static ssize_t
2903md_attr_store(struct kobject *kobj, struct attribute *attr,
2904 const char *page, size_t length)
2905{
2906 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2907 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2908 ssize_t rv;
eae1701f
N		 2909
2910 if (!entry->store)
2911 return -EIO;
67463acb
N		 2912 if (!capable(CAP_SYS_ADMIN))
2913 return -EACCES;
5dc5cf7d
IM		 2914 rv = mddev_lock(mddev);
2915 if (!rv) {
2916 rv = entry->store(mddev, page, length);
2917 mddev_unlock(mddev);
2918 }
96de1e66 2919 return rv;
eae1701f
N		 2920}
2921
2922static void md_free(struct kobject *ko)
2923{
2924 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2925 kfree(mddev);
2926}
2927
2928static struct sysfs_ops md_sysfs_ops = {
2929 .show = md_attr_show,
2930 .store = md_attr_store,
2931};
2932static struct kobj_type md_ktype = {
2933 .release = md_free,
2934 .sysfs_ops = &md_sysfs_ops,
2935 .default_attrs = md_default_attrs,
2936};
2937
1da177e4
LT		 2938int mdp_major = 0;
2939
2940static struct kobject *md_probe(dev_t dev, int *part, void *data)
2941{
48c9c27b 2942 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 2943 mddev_t *mddev = mddev_find(dev);
2944 struct gendisk *disk;
2945 int partitioned = (MAJOR(dev) != MD_MAJOR);
2946 int shift = partitioned ? MdpMinorShift : 0;
2947 int unit = MINOR(dev) >> shift;
2948
2949 if (!mddev)
2950 return NULL;
2951
48c9c27b 2952 mutex_lock(&disks_mutex);
1da177e4 2953 if (mddev->gendisk) {
48c9c27b 2954 mutex_unlock(&disks_mutex);
1da177e4
LT		 2955 mddev_put(mddev);
2956 return NULL;
2957 }
2958 disk = alloc_disk(1 << shift);
2959 if (!disk) {
48c9c27b 2960 mutex_unlock(&disks_mutex);
1da177e4
LT		 2961 mddev_put(mddev);
2962 return NULL;
2963 }
2964 disk->major = MAJOR(dev);
2965 disk->first_minor = unit << shift;
ce7b0f46 2966 if (partitioned)
1da177e4 2967 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 2968 else
1da177e4 2969 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 2970 disk->fops = &md_fops;
2971 disk->private_data = mddev;
2972 disk->queue = mddev->queue;
2973 add_disk(disk);
2974 mddev->gendisk = disk;
48c9c27b 2975 mutex_unlock(&disks_mutex);
9c791977 2976 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 2977 mddev->kobj.k_name = NULL;
2978 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2979 mddev->kobj.ktype = &md_ktype;
2980 kobject_register(&mddev->kobj);
1da177e4
LT		 2981 return NULL;
2982}
2983
1da177e4
LT		 2984static void md_safemode_timeout(unsigned long data)
2985{
2986 mddev_t *mddev = (mddev_t *) data;
2987
2988 mddev->safemode = 1;
2989 md_wakeup_thread(mddev->thread);
2990}
2991
6ff8d8ec 2992static int start_dirty_degraded;
1da177e4
LT		 2993
2994static int do_md_run(mddev_t * mddev)
2995{
2604b703 2996 int err;
1da177e4
LT		 2997 int chunk_size;
2998 struct list_head *tmp;
2999 mdk_rdev_t *rdev;
3000 struct gendisk *disk;
2604b703 3001 struct mdk_personality *pers;
1da177e4
LT		 3002 char b[BDEVNAME_SIZE];
3003
a757e64c
N		 3004 if (list_empty(&mddev->disks))
3005 /* cannot run an array with no devices.. */
1da177e4 3006 return -EINVAL;
1da177e4
LT		 3007
3008 if (mddev->pers)
3009 return -EBUSY;
3010
3011 /*
3012 * Analyze all RAID superblock(s)
3013 */
a757e64c
N		 3014 if (!mddev->raid_disks)
3015 analyze_sbs(mddev);
1da177e4
LT		 3016
3017 chunk_size = mddev->chunk_size;
2604b703
N		 3018
3019 if (chunk_size) {
1da177e4
LT		 3020 if (chunk_size > MAX_CHUNK_SIZE) {
3021 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3022 chunk_size, MAX_CHUNK_SIZE);
3023 return -EINVAL;
3024 }
3025 /*
3026 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3027 */
3028 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 3029 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 3030 return -EINVAL;
3031 }
3032 if (chunk_size < PAGE_SIZE) {
3033 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3034 chunk_size, PAGE_SIZE);
3035 return -EINVAL;
3036 }
3037
3038 /* devices must have minimum size of one chunk */
3039 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3040 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3041 continue;
3042 if (rdev->size < chunk_size / 1024) {
3043 printk(KERN_WARNING
3044 "md: Dev %s smaller than chunk_size:"
3045 " %lluk < %dk\n",
3046 bdevname(rdev->bdev,b),
3047 (unsigned long long)rdev->size,
3048 chunk_size / 1024);
3049 return -EINVAL;
3050 }
3051 }
3052 }
3053
1da177e4 3054#ifdef CONFIG_KMOD
d9d166c2
N		 3055 if (mddev->level != LEVEL_NONE)
3056 request_module("md-level-%d", mddev->level);
3057 else if (mddev->clevel[0])
3058 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3059#endif
3060
3061 /*
3062 * Drop all container device buffers, from now on
3063 * the only valid external interface is through the md
3064 * device.
3065 * Also find largest hardsector size
3066 */
3067 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3068 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3069 continue;
3070 sync_blockdev(rdev->bdev);
3071 invalidate_bdev(rdev->bdev, 0);
3072 }
3073
3074 md_probe(mddev->unit, NULL, NULL);
3075 disk = mddev->gendisk;
3076 if (!disk)
3077 return -ENOMEM;
3078
3079 spin_lock(&pers_lock);
d9d166c2 3080 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3081 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3082 spin_unlock(&pers_lock);
d9d166c2
N		 3083 if (mddev->level != LEVEL_NONE)
3084 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3085 mddev->level);
3086 else
3087 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3088 mddev->clevel);
1da177e4
LT		 3089 return -EINVAL;
3090 }
2604b703 3091 mddev->pers = pers;
1da177e4 3092 spin_unlock(&pers_lock);
d9d166c2
N		 3093 mddev->level = pers->level;
3094 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3095
f6705578 3096 if (mddev->reshape_position != MaxSector &&
63c70c4f 3097 pers->start_reshape == NULL) {
f6705578
N		 3098 /* This personality cannot handle reshaping... */
3099 mddev->pers = NULL;
3100 module_put(pers->owner);
3101 return -EINVAL;
3102 }
3103
657390d2 3104 mddev->recovery = 0;
1da177e4 3105 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3106 mddev->barriers_work = 1;
6ff8d8ec 3107 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3108
f91de92e
N		 3109 if (start_readonly)
3110 mddev->ro = 2; /* read-only, but switch on first write */
3111
b15c2e57
N		 3112 err = mddev->pers->run(mddev);
3113 if (!err && mddev->pers->sync_request) {
3114 err = bitmap_create(mddev);
3115 if (err) {
3116 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3117 mdname(mddev), err);
3118 mddev->pers->stop(mddev);
3119 }
3120 }
1da177e4
LT		 3121 if (err) {
3122 printk(KERN_ERR "md: pers->run() failed ...\n");
3123 module_put(mddev->pers->owner);
3124 mddev->pers = NULL;
32a7627c
N		 3125 bitmap_destroy(mddev);
3126 return err;
1da177e4 3127 }
411036fa
N		 3128 if (mddev->pers->sync_request)
3129 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
fd9d49ca
N		 3130 else if (mddev->ro == 2) /* auto-readonly not meaningful */
3131 mddev->ro = 0;
3132
1da177e4
LT		 3133 atomic_set(&mddev->writes_pending,0);
3134 mddev->safemode = 0;
3135 mddev->safemode_timer.function = md_safemode_timeout;
3136 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3137 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3138 mddev->in_sync = 1;
86e6ffdd
N		 3139
3140 ITERATE_RDEV(mddev,rdev,tmp)
3141 if (rdev->raid_disk >= 0) {
3142 char nm[20];
3143 sprintf(nm, "rd%d", rdev->raid_disk);
3144 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3145 }
1da177e4
LT		 3146
3147 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3148
850b2b42
N		 3149 if (mddev->flags)
3150 md_update_sb(mddev, 0);
1da177e4
LT		 3151
3152 set_capacity(disk, mddev->array_size<<1);
3153
3154 /* If we call blk_queue_make_request here, it will
3155 * re-initialise max_sectors etc which may have been
3156 * refined inside -> run. So just set the bits we need to set.
3157 * Most initialisation happended when we called
3158 * blk_queue_make_request(..., md_fail_request)
3159 * earlier.
3160 */
3161 mddev->queue->queuedata = mddev;
3162 mddev->queue->make_request_fn = mddev->pers->make_request;
3163
5fd6c1dc
N		 3164 /* If there is a partially-recovered drive we need to
3165 * start recovery here. If we leave it to md_check_recovery,
3166 * it will remove the drives and not do the right thing
3167 */
0b8c9de0 3168 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3169 struct list_head *rtmp;
3170 int spares = 0;
3171 ITERATE_RDEV(mddev,rdev,rtmp)
3172 if (rdev->raid_disk >= 0 &&
3173 !test_bit(In_sync, &rdev->flags) &&
3174 !test_bit(Faulty, &rdev->flags))
3175 /* complete an interrupted recovery */
3176 spares++;
3177 if (spares && mddev->pers->sync_request) {
3178 mddev->recovery = 0;
3179 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3180 mddev->sync_thread = md_register_thread(md_do_sync,
3181 mddev,
3182 "%s_resync");
3183 if (!mddev->sync_thread) {
3184 printk(KERN_ERR "%s: could not start resync"
3185 " thread...\n",
3186 mdname(mddev));
3187 /* leave the spares where they are, it shouldn't hurt */
3188 mddev->recovery = 0;
0b8c9de0 3189 }
5fd6c1dc
N		 3190 }
3191 }
0b8c9de0
N		 3192 md_wakeup_thread(mddev->thread);
3193 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3194
1da177e4 3195 mddev->changed = 1;
d7603b7e 3196 md_new_event(mddev);
1da177e4
LT		 3197 return 0;
3198}
3199
3200static int restart_array(mddev_t *mddev)
3201{
3202 struct gendisk *disk = mddev->gendisk;
3203 int err;
3204
3205 /*
3206 * Complain if it has no devices
3207 */
3208 err = -ENXIO;
3209 if (list_empty(&mddev->disks))
3210 goto out;
3211
3212 if (mddev->pers) {
3213 err = -EBUSY;
3214 if (!mddev->ro)
3215 goto out;
3216
3217 mddev->safemode = 0;
3218 mddev->ro = 0;
3219 set_disk_ro(disk, 0);
3220
3221 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3222 mdname(mddev));
3223 /*
3224 * Kick recovery or resync if necessary
3225 */
3226 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3227 md_wakeup_thread(mddev->thread);
5fd6c1dc 3228 md_wakeup_thread(mddev->sync_thread);
1da177e4 3229 err = 0;
9e653b63 3230 } else
1da177e4 3231 err = -EINVAL;
1da177e4
LT		 3232
3233out:
3234 return err;
3235}
3236
acc55e22
N		 3237/* similar to deny_write_access, but accounts for our holding a reference
3238 * to the file ourselves */
3239static int deny_bitmap_write_access(struct file * file)
3240{
3241 struct inode *inode = file->f_mapping->host;
3242
3243 spin_lock(&inode->i_lock);
3244 if (atomic_read(&inode->i_writecount) > 1) {
3245 spin_unlock(&inode->i_lock);
3246 return -ETXTBSY;
3247 }
3248 atomic_set(&inode->i_writecount, -1);
3249 spin_unlock(&inode->i_lock);
3250
3251 return 0;
3252}
3253
3254static void restore_bitmap_write_access(struct file *file)
3255{
3256 struct inode *inode = file->f_mapping->host;
3257
3258 spin_lock(&inode->i_lock);
3259 atomic_set(&inode->i_writecount, 1);
3260 spin_unlock(&inode->i_lock);
3261}
3262
9e653b63
N		 3263/* mode:
3264 * 0 - completely stop and dis-assemble array
3265 * 1 - switch to readonly
3266 * 2 - stop but do not disassemble array
3267 */
3268static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3269{
3270 int err = 0;
3271 struct gendisk *disk = mddev->gendisk;
3272
3273 if (mddev->pers) {
3274 if (atomic_read(&mddev->active)>2) {
3275 printk("md: %s still in use.\n",mdname(mddev));
3276 return -EBUSY;
3277 }
3278
3279 if (mddev->sync_thread) {
5fd6c1dc 3280 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3281 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3282 md_unregister_thread(mddev->sync_thread);
3283 mddev->sync_thread = NULL;
3284 }
3285
3286 del_timer_sync(&mddev->safemode_timer);
3287
3288 invalidate_partition(disk, 0);
3289
9e653b63
N		 3290 switch(mode) {
3291 case 1: /* readonly */
1da177e4 3292 err = -ENXIO;
f91de92e 3293 if (mddev->ro==1)
1da177e4
LT		 3294 goto out;
3295 mddev->ro = 1;
9e653b63
N		 3296 break;
3297 case 0: /* disassemble */
3298 case 2: /* stop */
6b8b3e8a 3299 bitmap_flush(mddev);
a9701a30 3300 md_super_wait(mddev);
1da177e4
LT		 3301 if (mddev->ro)
3302 set_disk_ro(disk, 0);
3303 blk_queue_make_request(mddev->queue, md_fail_request);
3304 mddev->pers->stop(mddev);
411036fa
N		 3305 if (mddev->pers->sync_request)
3306 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3307
1da177e4
LT		 3308 module_put(mddev->pers->owner);
3309 mddev->pers = NULL;
3310 if (mddev->ro)
3311 mddev->ro = 0;
3312 }
850b2b42 3313 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 3314 /* mark array as shutdown cleanly */
3315 mddev->in_sync = 1;
850b2b42 3316 md_update_sb(mddev, 1);
1da177e4 3317 }
9e653b63 3318 if (mode == 1)
1da177e4 3319 set_disk_ro(disk, 1);
5fd6c1dc 3320 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3321 }
32a7627c 3322
1da177e4
LT		 3323 /*
3324 * Free resources if final stop
3325 */
9e653b63 3326 if (mode == 0) {
86e6ffdd
N		 3327 mdk_rdev_t *rdev;
3328 struct list_head *tmp;
1da177e4
LT		 3329 struct gendisk *disk;
3330 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3331
978f946b
N		 3332 bitmap_destroy(mddev);
3333 if (mddev->bitmap_file) {
acc55e22 3334 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3335 fput(mddev->bitmap_file);
3336 mddev->bitmap_file = NULL;
3337 }
3338 mddev->bitmap_offset = 0;
3339
86e6ffdd
N		 3340 ITERATE_RDEV(mddev,rdev,tmp)
3341 if (rdev->raid_disk >= 0) {
3342 char nm[20];
3343 sprintf(nm, "rd%d", rdev->raid_disk);
3344 sysfs_remove_link(&mddev->kobj, nm);
3345 }
3346
1da177e4
LT		 3347 export_array(mddev);
3348
3349 mddev->array_size = 0;
9e653b63
N		 3350 mddev->size = 0;
3351 mddev->raid_disks = 0;
a94213b1 3352 mddev->recovery_cp = 0;
9e653b63 3353
1da177e4
LT		 3354 disk = mddev->gendisk;
3355 if (disk)
3356 set_capacity(disk, 0);
3357 mddev->changed = 1;
a8a55c38 3358 } else if (mddev->pers)
1da177e4
LT		 3359 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3360 mdname(mddev));
3361 err = 0;
d7603b7e 3362 md_new_event(mddev);
1da177e4
LT		 3363out:
3364 return err;
3365}
3366
3367static void autorun_array(mddev_t *mddev)
3368{
3369 mdk_rdev_t *rdev;
3370 struct list_head *tmp;
3371 int err;
3372
a757e64c 3373 if (list_empty(&mddev->disks))
1da177e4 3374 return;
1da177e4
LT		 3375
3376 printk(KERN_INFO "md: running: ");
3377
3378 ITERATE_RDEV(mddev,rdev,tmp) {
3379 char b[BDEVNAME_SIZE];
3380 printk("<%s>", bdevname(rdev->bdev,b));
3381 }
3382 printk("\n");
3383
3384 err = do_md_run (mddev);
3385 if (err) {
3386 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3387 do_md_stop (mddev, 0);
3388 }
3389}
3390
3391/*
3392 * lets try to run arrays based on all disks that have arrived
3393 * until now. (those are in pending_raid_disks)
3394 *
3395 * the method: pick the first pending disk, collect all disks with
3396 * the same UUID, remove all from the pending list and put them into
3397 * the 'same_array' list. Then order this list based on superblock
3398 * update time (freshest comes first), kick out 'old' disks and
3399 * compare superblocks. If everything's fine then run it.
3400 *
3401 * If "unit" is allocated, then bump its reference count
3402 */
3403static void autorun_devices(int part)
3404{
1da177e4
LT		 3405 struct list_head *tmp;
3406 mdk_rdev_t *rdev0, *rdev;
3407 mddev_t *mddev;
3408 char b[BDEVNAME_SIZE];
3409
3410 printk(KERN_INFO "md: autorun ...\n");
3411 while (!list_empty(&pending_raid_disks)) {
e8703fe1 3412 int unit;
1da177e4 3413 dev_t dev;
ad01c9e3 3414 LIST_HEAD(candidates);
1da177e4
LT		 3415 rdev0 = list_entry(pending_raid_disks.next,
3416 mdk_rdev_t, same_set);
3417
3418 printk(KERN_INFO "md: considering %s ...\n",
3419 bdevname(rdev0->bdev,b));
3420 INIT_LIST_HEAD(&candidates);
3421 ITERATE_RDEV_PENDING(rdev,tmp)
3422 if (super_90_load(rdev, rdev0, 0) >= 0) {
3423 printk(KERN_INFO "md: adding %s ...\n",
3424 bdevname(rdev->bdev,b));
3425 list_move(&rdev->same_set, &candidates);
3426 }
3427 /*
3428 * now we have a set of devices, with all of them having
3429 * mostly sane superblocks. It's time to allocate the
3430 * mddev.
3431 */
e8703fe1
N		 3432 if (part) {
3433 dev = MKDEV(mdp_major,
3434 rdev0->preferred_minor << MdpMinorShift);
3435 unit = MINOR(dev) >> MdpMinorShift;
3436 } else {
3437 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3438 unit = MINOR(dev);
3439 }
3440 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 3441 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3442 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3443 break;
3444 }
1da177e4
LT		 3445
3446 md_probe(dev, NULL, NULL);
3447 mddev = mddev_find(dev);
3448 if (!mddev) {
3449 printk(KERN_ERR
3450 "md: cannot allocate memory for md drive.\n");
3451 break;
3452 }
3453 if (mddev_lock(mddev))
3454 printk(KERN_WARNING "md: %s locked, cannot run\n",
3455 mdname(mddev));
3456 else if (mddev->raid_disks || mddev->major_version
3457 || !list_empty(&mddev->disks)) {
3458 printk(KERN_WARNING
3459 "md: %s already running, cannot run %s\n",
3460 mdname(mddev), bdevname(rdev0->bdev,b));
3461 mddev_unlock(mddev);
3462 } else {
3463 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3464 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3465 list_del_init(&rdev->same_set);
3466 if (bind_rdev_to_array(rdev, mddev))
3467 export_rdev(rdev);
3468 }
3469 autorun_array(mddev);
3470 mddev_unlock(mddev);
3471 }
3472 /* on success, candidates will be empty, on error
3473 * it won't...
3474 */
3475 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3476 export_rdev(rdev);
3477 mddev_put(mddev);
3478 }
3479 printk(KERN_INFO "md: ... autorun DONE.\n");
3480}
3481
1da177e4
LT		 3482static int get_version(void __user * arg)
3483{
3484 mdu_version_t ver;
3485
3486 ver.major = MD_MAJOR_VERSION;
3487 ver.minor = MD_MINOR_VERSION;
3488 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3489
3490 if (copy_to_user(arg, &ver, sizeof(ver)))
3491 return -EFAULT;
3492
3493 return 0;
3494}
3495
3496static int get_array_info(mddev_t * mddev, void __user * arg)
3497{
3498 mdu_array_info_t info;
3499 int nr,working,active,failed,spare;
3500 mdk_rdev_t *rdev;
3501 struct list_head *tmp;
3502
3503 nr=working=active=failed=spare=0;
3504 ITERATE_RDEV(mddev,rdev,tmp) {
3505 nr++;
b2d444d7 3506 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3507 failed++;
3508 else {
3509 working++;
b2d444d7 3510 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3511 active++;
3512 else
3513 spare++;
3514 }
3515 }
3516
3517 info.major_version = mddev->major_version;
3518 info.minor_version = mddev->minor_version;
3519 info.patch_version = MD_PATCHLEVEL_VERSION;
3520 info.ctime = mddev->ctime;
3521 info.level = mddev->level;
3522 info.size = mddev->size;
284ae7ca
N		 3523 if (info.size != mddev->size) /* overflow */
3524 info.size = -1;
1da177e4
LT		 3525 info.nr_disks = nr;
3526 info.raid_disks = mddev->raid_disks;
3527 info.md_minor = mddev->md_minor;
3528 info.not_persistent= !mddev->persistent;
3529
3530 info.utime = mddev->utime;
3531 info.state = 0;
3532 if (mddev->in_sync)
3533 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3534 if (mddev->bitmap && mddev->bitmap_offset)
3535 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3536 info.active_disks = active;
3537 info.working_disks = working;
3538 info.failed_disks = failed;
3539 info.spare_disks = spare;
3540
3541 info.layout = mddev->layout;
3542 info.chunk_size = mddev->chunk_size;
3543
3544 if (copy_to_user(arg, &info, sizeof(info)))
3545 return -EFAULT;
3546
3547 return 0;
3548}
3549
87162a28 3550static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3551{
3552 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3553 char *ptr, *buf = NULL;
3554 int err = -ENOMEM;
3555
3556 file = kmalloc(sizeof(*file), GFP_KERNEL);
3557 if (!file)
3558 goto out;
3559
3560 /* bitmap disabled, zero the first byte and copy out */
3561 if (!mddev->bitmap || !mddev->bitmap->file) {
3562 file->pathname[0] = '\0';
3563 goto copy_out;
3564 }
3565
3566 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3567 if (!buf)
3568 goto out;
3569
3570 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3571 if (!ptr)
3572 goto out;
3573
3574 strcpy(file->pathname, ptr);
3575
3576copy_out:
3577 err = 0;
3578 if (copy_to_user(arg, file, sizeof(*file)))
3579 err = -EFAULT;
3580out:
3581 kfree(buf);
3582 kfree(file);
3583 return err;
3584}
3585
1da177e4
LT		 3586static int get_disk_info(mddev_t * mddev, void __user * arg)
3587{
3588 mdu_disk_info_t info;
3589 unsigned int nr;
3590 mdk_rdev_t *rdev;
3591
3592 if (copy_from_user(&info, arg, sizeof(info)))
3593 return -EFAULT;
3594
3595 nr = info.number;
3596
3597 rdev = find_rdev_nr(mddev, nr);
3598 if (rdev) {
3599 info.major = MAJOR(rdev->bdev->bd_dev);
3600 info.minor = MINOR(rdev->bdev->bd_dev);
3601 info.raid_disk = rdev->raid_disk;
3602 info.state = 0;
b2d444d7 3603 if (test_bit(Faulty, &rdev->flags))
1da177e4 3604 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3605 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3606 info.state |= (1<<MD_DISK_ACTIVE);
3607 info.state |= (1<<MD_DISK_SYNC);
3608 }
8ddf9efe
N		 3609 if (test_bit(WriteMostly, &rdev->flags))
3610 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3611 } else {
3612 info.major = info.minor = 0;
3613 info.raid_disk = -1;
3614 info.state = (1<<MD_DISK_REMOVED);
3615 }
3616
3617 if (copy_to_user(arg, &info, sizeof(info)))
3618 return -EFAULT;
3619
3620 return 0;
3621}
3622
3623static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3624{
3625 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3626 mdk_rdev_t *rdev;
3627 dev_t dev = MKDEV(info->major,info->minor);
3628
3629 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3630 return -EOVERFLOW;
3631
3632 if (!mddev->raid_disks) {
3633 int err;
3634 /* expecting a device which has a superblock */
3635 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3636 if (IS_ERR(rdev)) {
3637 printk(KERN_WARNING
3638 "md: md_import_device returned %ld\n",
3639 PTR_ERR(rdev));
3640 return PTR_ERR(rdev);
3641 }
3642 if (!list_empty(&mddev->disks)) {
3643 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3644 mdk_rdev_t, same_set);
3645 int err = super_types[mddev->major_version]
3646 .load_super(rdev, rdev0, mddev->minor_version);
3647 if (err < 0) {
3648 printk(KERN_WARNING
3649 "md: %s has different UUID to %s\n",
3650 bdevname(rdev->bdev,b),
3651 bdevname(rdev0->bdev,b2));
3652 export_rdev(rdev);
3653 return -EINVAL;
3654 }
3655 }
3656 err = bind_rdev_to_array(rdev, mddev);
3657 if (err)
3658 export_rdev(rdev);
3659 return err;
3660 }
3661
3662 /*
3663 * add_new_disk can be used once the array is assembled
3664 * to add "hot spares". They must already have a superblock
3665 * written
3666 */
3667 if (mddev->pers) {
3668 int err;
3669 if (!mddev->pers->hot_add_disk) {
3670 printk(KERN_WARNING
3671 "%s: personality does not support diskops!\n",
3672 mdname(mddev));
3673 return -EINVAL;
3674 }
7b1e35f6
N		 3675 if (mddev->persistent)
3676 rdev = md_import_device(dev, mddev->major_version,
3677 mddev->minor_version);
3678 else
3679 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 3680 if (IS_ERR(rdev)) {
3681 printk(KERN_WARNING
3682 "md: md_import_device returned %ld\n",
3683 PTR_ERR(rdev));
3684 return PTR_ERR(rdev);
3685 }
41158c7e
N		 3686 /* set save_raid_disk if appropriate */
3687 if (!mddev->persistent) {
3688 if (info->state & (1<<MD_DISK_SYNC) &&
3689 info->raid_disk < mddev->raid_disks)
3690 rdev->raid_disk = info->raid_disk;
3691 else
3692 rdev->raid_disk = -1;
3693 } else
3694 super_types[mddev->major_version].
3695 validate_super(mddev, rdev);
3696 rdev->saved_raid_disk = rdev->raid_disk;
3697
b2d444d7 3698 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 3699 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3700 set_bit(WriteMostly, &rdev->flags);
3701
1da177e4
LT		 3702 rdev->raid_disk = -1;
3703 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 3704 if (!err && !mddev->pers->hot_remove_disk) {
3705 /* If there is hot_add_disk but no hot_remove_disk
3706 * then added disks for geometry changes,
3707 * and should be added immediately.
3708 */
3709 super_types[mddev->major_version].
3710 validate_super(mddev, rdev);
3711 err = mddev->pers->hot_add_disk(mddev, rdev);
3712 if (err)
3713 unbind_rdev_from_array(rdev);
3714 }
1da177e4
LT		 3715 if (err)
3716 export_rdev(rdev);
c361777f
N		 3717
3718 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 3719 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3720 return err;
3721 }
3722
3723 /* otherwise, add_new_disk is only allowed
3724 * for major_version==0 superblocks
3725 */
3726 if (mddev->major_version != 0) {
3727 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3728 mdname(mddev));
3729 return -EINVAL;
3730 }
3731
3732 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3733 int err;
3734 rdev = md_import_device (dev, -1, 0);
3735 if (IS_ERR(rdev)) {
3736 printk(KERN_WARNING
3737 "md: error, md_import_device() returned %ld\n",
3738 PTR_ERR(rdev));
3739 return PTR_ERR(rdev);
3740 }
3741 rdev->desc_nr = info->number;
3742 if (info->raid_disk < mddev->raid_disks)
3743 rdev->raid_disk = info->raid_disk;
3744 else
3745 rdev->raid_disk = -1;
3746
b2d444d7
N		 3747 rdev->flags = 0;
3748
1da177e4 3749 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 3750 if (info->state & (1<<MD_DISK_SYNC))
3751 set_bit(In_sync, &rdev->flags);
1da177e4 3752
8ddf9efe
N		 3753 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3754 set_bit(WriteMostly, &rdev->flags);
3755
1da177e4
LT		 3756 if (!mddev->persistent) {
3757 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3758 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3759 } else
3760 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3761 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3762
2bf071bf
N		 3763 err = bind_rdev_to_array(rdev, mddev);
3764 if (err) {
3765 export_rdev(rdev);
3766 return err;
3767 }
1da177e4
LT		 3768 }
3769
3770 return 0;
3771}
3772
3773static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3774{
3775 char b[BDEVNAME_SIZE];
3776 mdk_rdev_t *rdev;
3777
3778 if (!mddev->pers)
3779 return -ENODEV;
3780
3781 rdev = find_rdev(mddev, dev);
3782 if (!rdev)
3783 return -ENXIO;
3784
3785 if (rdev->raid_disk >= 0)
3786 goto busy;
3787
3788 kick_rdev_from_array(rdev);
850b2b42 3789 md_update_sb(mddev, 1);
d7603b7e 3790 md_new_event(mddev);
1da177e4
LT		 3791
3792 return 0;
3793busy:
3794 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3795 bdevname(rdev->bdev,b), mdname(mddev));
3796 return -EBUSY;
3797}
3798
3799static int hot_add_disk(mddev_t * mddev, dev_t dev)
3800{
3801 char b[BDEVNAME_SIZE];
3802 int err;
3803 unsigned int size;
3804 mdk_rdev_t *rdev;
3805
3806 if (!mddev->pers)
3807 return -ENODEV;
3808
3809 if (mddev->major_version != 0) {
3810 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3811 " version-0 superblocks.\n",
3812 mdname(mddev));
3813 return -EINVAL;
3814 }
3815 if (!mddev->pers->hot_add_disk) {
3816 printk(KERN_WARNING
3817 "%s: personality does not support diskops!\n",
3818 mdname(mddev));
3819 return -EINVAL;
3820 }
3821
3822 rdev = md_import_device (dev, -1, 0);
3823 if (IS_ERR(rdev)) {
3824 printk(KERN_WARNING
3825 "md: error, md_import_device() returned %ld\n",
3826 PTR_ERR(rdev));
3827 return -EINVAL;
3828 }
3829
3830 if (mddev->persistent)
3831 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3832 else
3833 rdev->sb_offset =
3834 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3835
3836 size = calc_dev_size(rdev, mddev->chunk_size);
3837 rdev->size = size;
3838
b2d444d7 3839 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3840 printk(KERN_WARNING
3841 "md: can not hot-add faulty %s disk to %s!\n",
3842 bdevname(rdev->bdev,b), mdname(mddev));
3843 err = -EINVAL;
3844 goto abort_export;
3845 }
b2d444d7 3846 clear_bit(In_sync, &rdev->flags);
1da177e4 3847 rdev->desc_nr = -1;
2bf071bf
N		 3848 err = bind_rdev_to_array(rdev, mddev);
3849 if (err)
3850 goto abort_export;
1da177e4
LT		 3851
3852 /*
3853 * The rest should better be atomic, we can have disk failures
3854 * noticed in interrupt contexts ...
3855 */
3856
3857 if (rdev->desc_nr == mddev->max_disks) {
3858 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3859 mdname(mddev));
3860 err = -EBUSY;
3861 goto abort_unbind_export;
3862 }
3863
3864 rdev->raid_disk = -1;
3865
850b2b42 3866 md_update_sb(mddev, 1);
1da177e4
LT		 3867
3868 /*
3869 * Kick recovery, maybe this spare has to be added to the
3870 * array immediately.
3871 */
3872 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3873 md_wakeup_thread(mddev->thread);
d7603b7e 3874 md_new_event(mddev);
1da177e4
LT		 3875 return 0;
3876
3877abort_unbind_export:
3878 unbind_rdev_from_array(rdev);
3879
3880abort_export:
3881 export_rdev(rdev);
3882 return err;
3883}
3884
32a7627c
N		 3885static int set_bitmap_file(mddev_t *mddev, int fd)
3886{
3887 int err;
3888
36fa3063
N		 3889 if (mddev->pers) {
3890 if (!mddev->pers->quiesce)
3891 return -EBUSY;
3892 if (mddev->recovery || mddev->sync_thread)
3893 return -EBUSY;
3894 /* we should be able to change the bitmap.. */
3895 }
32a7627c 3896
32a7627c 3897
36fa3063
N		 3898 if (fd >= 0) {
3899 if (mddev->bitmap)
3900 return -EEXIST; /* cannot add when bitmap is present */
3901 mddev->bitmap_file = fget(fd);
32a7627c 3902
36fa3063
N		 3903 if (mddev->bitmap_file == NULL) {
3904 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3905 mdname(mddev));
3906 return -EBADF;
3907 }
3908
3909 err = deny_bitmap_write_access(mddev->bitmap_file);
3910 if (err) {
3911 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3912 mdname(mddev));
3913 fput(mddev->bitmap_file);
3914 mddev->bitmap_file = NULL;
3915 return err;
3916 }
a654b9d8 3917 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 3918 } else if (mddev->bitmap == NULL)
3919 return -ENOENT; /* cannot remove what isn't there */
3920 err = 0;
3921 if (mddev->pers) {
3922 mddev->pers->quiesce(mddev, 1);
3923 if (fd >= 0)
3924 err = bitmap_create(mddev);
d7375ab3 3925 if (fd < 0 || err) {
36fa3063 3926 bitmap_destroy(mddev);
d7375ab3
N		 3927 fd = -1; /* make sure to put the file */
3928 }
36fa3063 3929 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 3930 }
3931 if (fd < 0) {
acc55e22
N		 3932 if (mddev->bitmap_file) {
3933 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 3934 fput(mddev->bitmap_file);
acc55e22 3935 }
36fa3063
N		 3936 mddev->bitmap_file = NULL;
3937 }
3938
32a7627c
N		 3939 return err;
3940}
3941
1da177e4
LT		 3942/*
3943 * set_array_info is used two different ways
3944 * The original usage is when creating a new array.
3945 * In this usage, raid_disks is > 0 and it together with
3946 * level, size, not_persistent,layout,chunksize determine the
3947 * shape of the array.
3948 * This will always create an array with a type-0.90.0 superblock.
3949 * The newer usage is when assembling an array.
3950 * In this case raid_disks will be 0, and the major_version field is
3951 * use to determine which style super-blocks are to be found on the devices.
3952 * The minor and patch _version numbers are also kept incase the
3953 * super_block handler wishes to interpret them.
3954 */
3955static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3956{
3957
3958 if (info->raid_disks == 0) {
3959 /* just setting version number for superblock loading */
3960 if (info->major_version < 0 ||
3961 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3962 super_types[info->major_version].name == NULL) {
3963 /* maybe try to auto-load a module? */
3964 printk(KERN_INFO
3965 "md: superblock version %d not known\n",
3966 info->major_version);
3967 return -EINVAL;
3968 }
3969 mddev->major_version = info->major_version;
3970 mddev->minor_version = info->minor_version;
3971 mddev->patch_version = info->patch_version;
3972 return 0;
3973 }
3974 mddev->major_version = MD_MAJOR_VERSION;
3975 mddev->minor_version = MD_MINOR_VERSION;
3976 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3977 mddev->ctime = get_seconds();
3978
3979 mddev->level = info->level;
17115e03 3980 mddev->clevel[0] = 0;
1da177e4
LT		 3981 mddev->size = info->size;
3982 mddev->raid_disks = info->raid_disks;
3983 /* don't set md_minor, it is determined by which /dev/md* was
3984 * openned
3985 */
3986 if (info->state & (1<<MD_SB_CLEAN))
3987 mddev->recovery_cp = MaxSector;
3988 else
3989 mddev->recovery_cp = 0;
3990 mddev->persistent = ! info->not_persistent;
3991
3992 mddev->layout = info->layout;
3993 mddev->chunk_size = info->chunk_size;
3994
3995 mddev->max_disks = MD_SB_DISKS;
3996
850b2b42
N		 3997 mddev->flags = 0;
3998 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 3999
b2a2703c
N		 4000 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4001 mddev->bitmap_offset = 0;
4002
f6705578
N		 4003 mddev->reshape_position = MaxSector;
4004
1da177e4
LT		 4005 /*
4006 * Generate a 128 bit UUID
4007 */
4008 get_random_bytes(mddev->uuid, 16);
4009
f6705578
N		 4010 mddev->new_level = mddev->level;
4011 mddev->new_chunk = mddev->chunk_size;
4012 mddev->new_layout = mddev->layout;
4013 mddev->delta_disks = 0;
4014
1da177e4
LT		 4015 return 0;
4016}
4017
a35b0d69
N		 4018static int update_size(mddev_t *mddev, unsigned long size)
4019{
4020 mdk_rdev_t * rdev;
4021 int rv;
4022 struct list_head *tmp;
8ddeeae5 4023 int fit = (size == 0);
a35b0d69
N		 4024
4025 if (mddev->pers->resize == NULL)
4026 return -EINVAL;
4027 /* The "size" is the amount of each device that is used.
4028 * This can only make sense for arrays with redundancy.
4029 * linear and raid0 always use whatever space is available
4030 * We can only consider changing the size if no resync
4031 * or reconstruction is happening, and if the new size
4032 * is acceptable. It must fit before the sb_offset or,
4033 * if that is <data_offset, it must fit before the
4034 * size of each device.
4035 * If size is zero, we find the largest size that fits.
4036 */
4037 if (mddev->sync_thread)
4038 return -EBUSY;
4039 ITERATE_RDEV(mddev,rdev,tmp) {
4040 sector_t avail;
a35b0d69
N		 4041 if (rdev->sb_offset > rdev->data_offset)
4042 avail = (rdev->sb_offset*2) - rdev->data_offset;
4043 else
4044 avail = get_capacity(rdev->bdev->bd_disk)
4045 - rdev->data_offset;
4046 if (fit && (size == 0 || size > avail/2))
4047 size = avail/2;
4048 if (avail < ((sector_t)size << 1))
4049 return -ENOSPC;
4050 }
4051 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4052 if (!rv) {
4053 struct block_device *bdev;
4054
4055 bdev = bdget_disk(mddev->gendisk, 0);
4056 if (bdev) {
1b1dcc1b 4057 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4058 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4059 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4060 bdput(bdev);
4061 }
4062 }
4063 return rv;
4064}
4065
da943b99
N		 4066static int update_raid_disks(mddev_t *mddev, int raid_disks)
4067{
4068 int rv;
4069 /* change the number of raid disks */
63c70c4f 4070 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4071 return -EINVAL;
4072 if (raid_disks <= 0 ||
4073 raid_disks >= mddev->max_disks)
4074 return -EINVAL;
63c70c4f 4075 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4076 return -EBUSY;
63c70c4f
N		 4077 mddev->delta_disks = raid_disks - mddev->raid_disks;
4078
4079 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4080 return rv;
4081}
4082
4083
1da177e4
LT		 4084/*
4085 * update_array_info is used to change the configuration of an
4086 * on-line array.
4087 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4088 * fields in the info are checked against the array.
4089 * Any differences that cannot be handled will cause an error.
4090 * Normally, only one change can be managed at a time.
4091 */
4092static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4093{
4094 int rv = 0;
4095 int cnt = 0;
36fa3063
N		 4096 int state = 0;
4097
4098 /* calculate expected state,ignoring low bits */
4099 if (mddev->bitmap && mddev->bitmap_offset)
4100 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4101
4102 if (mddev->major_version != info->major_version ||
4103 mddev->minor_version != info->minor_version ||
4104/* mddev->patch_version != info->patch_version || */
4105 mddev->ctime != info->ctime ||
4106 mddev->level != info->level ||
4107/* mddev->layout != info->layout || */
4108 !mddev->persistent != info->not_persistent||
36fa3063
N		 4109 mddev->chunk_size != info->chunk_size ||
4110 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4111 ((state^info->state) & 0xfffffe00)
4112 )
1da177e4
LT		 4113 return -EINVAL;
4114 /* Check there is only one change */
284ae7ca 4115 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4116 if (mddev->raid_disks != info->raid_disks) cnt++;
4117 if (mddev->layout != info->layout) cnt++;
36fa3063 4118 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4119 if (cnt == 0) return 0;
4120 if (cnt > 1) return -EINVAL;
4121
4122 if (mddev->layout != info->layout) {
4123 /* Change layout
4124 * we don't need to do anything at the md level, the
4125 * personality will take care of it all.
4126 */
4127 if (mddev->pers->reconfig == NULL)
4128 return -EINVAL;
4129 else
4130 return mddev->pers->reconfig(mddev, info->layout, -1);
4131 }
284ae7ca 4132 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4133 rv = update_size(mddev, info->size);
4134
da943b99
N		 4135 if (mddev->raid_disks != info->raid_disks)
4136 rv = update_raid_disks(mddev, info->raid_disks);
4137
36fa3063
N		 4138 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4139 if (mddev->pers->quiesce == NULL)
4140 return -EINVAL;
4141 if (mddev->recovery || mddev->sync_thread)
4142 return -EBUSY;
4143 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4144 /* add the bitmap */
4145 if (mddev->bitmap)
4146 return -EEXIST;
4147 if (mddev->default_bitmap_offset == 0)
4148 return -EINVAL;
4149 mddev->bitmap_offset = mddev->default_bitmap_offset;
4150 mddev->pers->quiesce(mddev, 1);
4151 rv = bitmap_create(mddev);
4152 if (rv)
4153 bitmap_destroy(mddev);
4154 mddev->pers->quiesce(mddev, 0);
4155 } else {
4156 /* remove the bitmap */
4157 if (!mddev->bitmap)
4158 return -ENOENT;
4159 if (mddev->bitmap->file)
4160 return -EINVAL;
4161 mddev->pers->quiesce(mddev, 1);
4162 bitmap_destroy(mddev);
4163 mddev->pers->quiesce(mddev, 0);
4164 mddev->bitmap_offset = 0;
4165 }
4166 }
850b2b42 4167 md_update_sb(mddev, 1);
1da177e4
LT		 4168 return rv;
4169}
4170
4171static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4172{
4173 mdk_rdev_t *rdev;
4174
4175 if (mddev->pers == NULL)
4176 return -ENODEV;
4177
4178 rdev = find_rdev(mddev, dev);
4179 if (!rdev)
4180 return -ENODEV;
4181
4182 md_error(mddev, rdev);
4183 return 0;
4184}
4185
a885c8c4
CH		 4186static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4187{
4188 mddev_t *mddev = bdev->bd_disk->private_data;
4189
4190 geo->heads = 2;
4191 geo->sectors = 4;
4192 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4193 return 0;
4194}
4195
1da177e4
LT		 4196static int md_ioctl(struct inode *inode, struct file *file,
4197 unsigned int cmd, unsigned long arg)
4198{
4199 int err = 0;
4200 void __user *argp = (void __user *)arg;
1da177e4
LT		 4201 mddev_t *mddev = NULL;
4202
4203 if (!capable(CAP_SYS_ADMIN))
4204 return -EACCES;
4205
4206 /*
4207 * Commands dealing with the RAID driver but not any
4208 * particular array:
4209 */
4210 switch (cmd)
4211 {
4212 case RAID_VERSION:
4213 err = get_version(argp);
4214 goto done;
4215
4216 case PRINT_RAID_DEBUG:
4217 err = 0;
4218 md_print_devices();
4219 goto done;
4220
4221#ifndef MODULE
4222 case RAID_AUTORUN:
4223 err = 0;
4224 autostart_arrays(arg);
4225 goto done;
4226#endif
4227 default:;
4228 }
4229
4230 /*
4231 * Commands creating/starting a new array:
4232 */
4233
4234 mddev = inode->i_bdev->bd_disk->private_data;
4235
4236 if (!mddev) {
4237 BUG();
4238 goto abort;
4239 }
4240
1da177e4
LT		 4241 err = mddev_lock(mddev);
4242 if (err) {
4243 printk(KERN_INFO
4244 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4245 err, cmd);
4246 goto abort;
4247 }
4248
4249 switch (cmd)
4250 {
4251 case SET_ARRAY_INFO:
4252 {
4253 mdu_array_info_t info;
4254 if (!arg)
4255 memset(&info, 0, sizeof(info));
4256 else if (copy_from_user(&info, argp, sizeof(info))) {
4257 err = -EFAULT;
4258 goto abort_unlock;
4259 }
4260 if (mddev->pers) {
4261 err = update_array_info(mddev, &info);
4262 if (err) {
4263 printk(KERN_WARNING "md: couldn't update"
4264 " array info. %d\n", err);
4265 goto abort_unlock;
4266 }
4267 goto done_unlock;
4268 }
4269 if (!list_empty(&mddev->disks)) {
4270 printk(KERN_WARNING
4271 "md: array %s already has disks!\n",
4272 mdname(mddev));
4273 err = -EBUSY;
4274 goto abort_unlock;
4275 }
4276 if (mddev->raid_disks) {
4277 printk(KERN_WARNING
4278 "md: array %s already initialised!\n",
4279 mdname(mddev));
4280 err = -EBUSY;
4281 goto abort_unlock;
4282 }
4283 err = set_array_info(mddev, &info);
4284 if (err) {
4285 printk(KERN_WARNING "md: couldn't set"
4286 " array info. %d\n", err);
4287 goto abort_unlock;
4288 }
4289 }
4290 goto done_unlock;
4291
4292 default:;
4293 }
4294
4295 /*
4296 * Commands querying/configuring an existing array:
4297 */
32a7627c
N		 4298 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4299 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4300 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4301 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 4302 err = -ENODEV;
4303 goto abort_unlock;
4304 }
4305
4306 /*
4307 * Commands even a read-only array can execute:
4308 */
4309 switch (cmd)
4310 {
4311 case GET_ARRAY_INFO:
4312 err = get_array_info(mddev, argp);
4313 goto done_unlock;
4314
32a7627c 4315 case GET_BITMAP_FILE:
87162a28 4316 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4317 goto done_unlock;
4318
1da177e4
LT		 4319 case GET_DISK_INFO:
4320 err = get_disk_info(mddev, argp);
4321 goto done_unlock;
4322
4323 case RESTART_ARRAY_RW:
4324 err = restart_array(mddev);
4325 goto done_unlock;
4326
4327 case STOP_ARRAY:
4328 err = do_md_stop (mddev, 0);
4329 goto done_unlock;
4330
4331 case STOP_ARRAY_RO:
4332 err = do_md_stop (mddev, 1);
4333 goto done_unlock;
4334
4335 /*
4336 * We have a problem here : there is no easy way to give a CHS
4337 * virtual geometry. We currently pretend that we have a 2 heads
4338 * 4 sectors (with a BIG number of cylinders...). This drives
4339 * dosfs just mad... ;-)
4340 */
1da177e4
LT		 4341 }
4342
4343 /*
4344 * The remaining ioctls are changing the state of the
f91de92e
N		 4345 * superblock, so we do not allow them on read-only arrays.
4346 * However non-MD ioctls (e.g. get-size) will still come through
4347 * here and hit the 'default' below, so only disallow
4348 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4349 */
f91de92e
N		 4350 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4351 mddev->ro && mddev->pers) {
4352 if (mddev->ro == 2) {
4353 mddev->ro = 0;
4354 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4355 md_wakeup_thread(mddev->thread);
4356
4357 } else {
4358 err = -EROFS;
4359 goto abort_unlock;
4360 }
1da177e4
LT		 4361 }
4362
4363 switch (cmd)
4364 {
4365 case ADD_NEW_DISK:
4366 {
4367 mdu_disk_info_t info;
4368 if (copy_from_user(&info, argp, sizeof(info)))
4369 err = -EFAULT;
4370 else
4371 err = add_new_disk(mddev, &info);
4372 goto done_unlock;
4373 }
4374
4375 case HOT_REMOVE_DISK:
4376 err = hot_remove_disk(mddev, new_decode_dev(arg));
4377 goto done_unlock;
4378
4379 case HOT_ADD_DISK:
4380 err = hot_add_disk(mddev, new_decode_dev(arg));
4381 goto done_unlock;
4382
4383 case SET_DISK_FAULTY:
4384 err = set_disk_faulty(mddev, new_decode_dev(arg));
4385 goto done_unlock;
4386
4387 case RUN_ARRAY:
4388 err = do_md_run (mddev);
4389 goto done_unlock;
4390
32a7627c
N		 4391 case SET_BITMAP_FILE:
4392 err = set_bitmap_file(mddev, (int)arg);
4393 goto done_unlock;
4394
1da177e4 4395 default:
1da177e4
LT		 4396 err = -EINVAL;
4397 goto abort_unlock;
4398 }
4399
4400done_unlock:
4401abort_unlock:
4402 mddev_unlock(mddev);
4403
4404 return err;
4405done:
4406 if (err)
4407 MD_BUG();
4408abort:
4409 return err;
4410}
4411
4412static int md_open(struct inode *inode, struct file *file)
4413{
4414 /*
4415 * Succeed if we can lock the mddev, which confirms that
4416 * it isn't being stopped right now.
4417 */
4418 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4419 int err;
4420
4421 if ((err = mddev_lock(mddev)))
4422 goto out;
4423
4424 err = 0;
4425 mddev_get(mddev);
4426 mddev_unlock(mddev);
4427
4428 check_disk_change(inode->i_bdev);
4429 out:
4430 return err;
4431}
4432
4433static int md_release(struct inode *inode, struct file * file)
4434{
4435 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4436
4437 if (!mddev)
4438 BUG();
4439 mddev_put(mddev);
4440
4441 return 0;
4442}
4443
4444static int md_media_changed(struct gendisk *disk)
4445{
4446 mddev_t *mddev = disk->private_data;
4447
4448 return mddev->changed;
4449}
4450
4451static int md_revalidate(struct gendisk *disk)
4452{
4453 mddev_t *mddev = disk->private_data;
4454
4455 mddev->changed = 0;
4456 return 0;
4457}
4458static struct block_device_operations md_fops =
4459{
4460 .owner = THIS_MODULE,
4461 .open = md_open,
4462 .release = md_release,
4463 .ioctl = md_ioctl,
a885c8c4 4464 .getgeo = md_getgeo,
1da177e4
LT		 4465 .media_changed = md_media_changed,
4466 .revalidate_disk= md_revalidate,
4467};
4468
75c96f85 4469static int md_thread(void * arg)
1da177e4
LT		 4470{
4471 mdk_thread_t *thread = arg;
4472
1da177e4
LT		 4473 /*
4474 * md_thread is a 'system-thread', it's priority should be very
4475 * high. We avoid resource deadlocks individually in each
4476 * raid personality. (RAID5 does preallocation) We also use RR and
4477 * the very same RT priority as kswapd, thus we will never get
4478 * into a priority inversion deadlock.
4479 *
4480 * we definitely have to have equal or higher priority than
4481 * bdflush, otherwise bdflush will deadlock if there are too
4482 * many dirty RAID5 blocks.
4483 */
1da177e4 4484
6985c43f 4485 allow_signal(SIGKILL);
a6fb0934 4486 while (!kthread_should_stop()) {
1da177e4 4487
93588e22
N		 4488 /* We need to wait INTERRUPTIBLE so that
4489 * we don't add to the load-average.
4490 * That means we need to be sure no signals are
4491 * pending
4492 */
4493 if (signal_pending(current))
4494 flush_signals(current);
4495
4496 wait_event_interruptible_timeout
4497 (thread->wqueue,
4498 test_bit(THREAD_WAKEUP, &thread->flags)
4499 || kthread_should_stop(),
4500 thread->timeout);
3e1d1d28 4501 try_to_freeze();
1da177e4
LT		 4502
4503 clear_bit(THREAD_WAKEUP, &thread->flags);
4504
787453c2 4505 thread->run(thread->mddev);
1da177e4 4506 }
a6fb0934 4507
1da177e4
LT		 4508 return 0;
4509}
4510
4511void md_wakeup_thread(mdk_thread_t *thread)
4512{
4513 if (thread) {
4514 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4515 set_bit(THREAD_WAKEUP, &thread->flags);
4516 wake_up(&thread->wqueue);
4517 }
4518}
4519
4520mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4521 const char *name)
4522{
4523 mdk_thread_t *thread;
1da177e4 4524
9ffae0cf 4525 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4526 if (!thread)
4527 return NULL;
4528
1da177e4
LT		 4529 init_waitqueue_head(&thread->wqueue);
4530
1da177e4
LT		 4531 thread->run = run;
4532 thread->mddev = mddev;
32a7627c 4533 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4534 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4535 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4536 kfree(thread);
4537 return NULL;
4538 }
1da177e4
LT		 4539 return thread;
4540}
4541
1da177e4
LT		 4542void md_unregister_thread(mdk_thread_t *thread)
4543{
d28446fe 4544 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 4545
4546 kthread_stop(thread->tsk);
1da177e4
LT		 4547 kfree(thread);
4548}
4549
4550void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4551{
4552 if (!mddev) {
4553 MD_BUG();
4554 return;
4555 }
4556
b2d444d7 4557 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4558 return;
32a7627c 4559/*
1da177e4
LT		 4560 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4561 mdname(mddev),
4562 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4563 __builtin_return_address(0),__builtin_return_address(1),
4564 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4565*/
d0a0a5ee
AM		 4566 if (!mddev->pers)
4567 return;
1da177e4
LT		 4568 if (!mddev->pers->error_handler)
4569 return;
4570 mddev->pers->error_handler(mddev,rdev);
4571 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4572 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4573 md_wakeup_thread(mddev->thread);
c331eb04 4574 md_new_event_inintr(mddev);
1da177e4
LT		 4575}
4576
4577/* seq_file implementation /proc/mdstat */
4578
4579static void status_unused(struct seq_file *seq)
4580{
4581 int i = 0;
4582 mdk_rdev_t *rdev;
4583 struct list_head *tmp;
4584
4585 seq_printf(seq, "unused devices: ");
4586
4587 ITERATE_RDEV_PENDING(rdev,tmp) {
4588 char b[BDEVNAME_SIZE];
4589 i++;
4590 seq_printf(seq, "%s ",
4591 bdevname(rdev->bdev,b));
4592 }
4593 if (!i)
4594 seq_printf(seq, "<none>");
4595
4596 seq_printf(seq, "\n");
4597}
4598
4599
4600static void status_resync(struct seq_file *seq, mddev_t * mddev)
4601{
4588b42e
N		 4602 sector_t max_blocks, resync, res;
4603 unsigned long dt, db, rt;
4604 int scale;
4605 unsigned int per_milli;
1da177e4
LT		 4606
4607 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4608
4609 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4610 max_blocks = mddev->resync_max_sectors >> 1;
4611 else
4612 max_blocks = mddev->size;
4613
4614 /*
4615 * Should not happen.
4616 */
4617 if (!max_blocks) {
4618 MD_BUG();
4619 return;
4620 }
4588b42e
N		 4621 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4622 * in a sector_t, and (max_blocks>>scale) will fit in a
4623 * u32, as those are the requirements for sector_div.
4624 * Thus 'scale' must be at least 10
4625 */
4626 scale = 10;
4627 if (sizeof(sector_t) > sizeof(unsigned long)) {
4628 while ( max_blocks/2 > (1ULL<<(scale+32)))
4629 scale++;
4630 }
4631 res = (resync>>scale)*1000;
4632 sector_div(res, (u32)((max_blocks>>scale)+1));
4633
4634 per_milli = res;
1da177e4 4635 {
4588b42e 4636 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4637 seq_printf(seq, "[");
4638 for (i = 0; i < x; i++)
4639 seq_printf(seq, "=");
4640 seq_printf(seq, ">");
4641 for (i = 0; i < y; i++)
4642 seq_printf(seq, ".");
4643 seq_printf(seq, "] ");
4644 }
4588b42e 4645 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 4646 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4647 "reshape" :
61df9d91
N		 4648 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4649 "check" :
4650 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4651 "resync" : "recovery"))),
4652 per_milli/10, per_milli % 10,
4588b42e
N		 4653 (unsigned long long) resync,
4654 (unsigned long long) max_blocks);
1da177e4
LT		 4655
4656 /*
4657 * We do not want to overflow, so the order of operands and
4658 * the * 100 / 100 trick are important. We do a +1 to be
4659 * safe against division by zero. We only estimate anyway.
4660 *
4661 * dt: time from mark until now
4662 * db: blocks written from mark until now
4663 * rt: remaining time
4664 */
4665 dt = ((jiffies - mddev->resync_mark) / HZ);
4666 if (!dt) dt++;
ff4e8d9a
N		 4667 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4668 - mddev->resync_mark_cnt;
4669 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 4670
4671 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4672
ff4e8d9a 4673 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 4674}
4675
4676static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4677{
4678 struct list_head *tmp;
4679 loff_t l = *pos;
4680 mddev_t *mddev;
4681
4682 if (l >= 0x10000)
4683 return NULL;
4684 if (!l--)
4685 /* header */
4686 return (void*)1;
4687
4688 spin_lock(&all_mddevs_lock);
4689 list_for_each(tmp,&all_mddevs)
4690 if (!l--) {
4691 mddev = list_entry(tmp, mddev_t, all_mddevs);
4692 mddev_get(mddev);
4693 spin_unlock(&all_mddevs_lock);
4694 return mddev;
4695 }
4696 spin_unlock(&all_mddevs_lock);
4697 if (!l--)
4698 return (void*)2;/* tail */
4699 return NULL;
4700}
4701
4702static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4703{
4704 struct list_head *tmp;
4705 mddev_t *next_mddev, *mddev = v;
4706
4707 ++*pos;
4708 if (v == (void*)2)
4709 return NULL;
4710
4711 spin_lock(&all_mddevs_lock);
4712 if (v == (void*)1)
4713 tmp = all_mddevs.next;
4714 else
4715 tmp = mddev->all_mddevs.next;
4716 if (tmp != &all_mddevs)
4717 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4718 else {
4719 next_mddev = (void*)2;
4720 *pos = 0x10000;
4721 }
4722 spin_unlock(&all_mddevs_lock);
4723
4724 if (v != (void*)1)
4725 mddev_put(mddev);
4726 return next_mddev;
4727
4728}
4729
4730static void md_seq_stop(struct seq_file *seq, void *v)
4731{
4732 mddev_t *mddev = v;
4733
4734 if (mddev && v != (void*)1 && v != (void*)2)
4735 mddev_put(mddev);
4736}
4737
d7603b7e
N		 4738struct mdstat_info {
4739 int event;
4740};
4741
1da177e4
LT		 4742static int md_seq_show(struct seq_file *seq, void *v)
4743{
4744 mddev_t *mddev = v;
4745 sector_t size;
4746 struct list_head *tmp2;
4747 mdk_rdev_t *rdev;
d7603b7e 4748 struct mdstat_info *mi = seq->private;
32a7627c 4749 struct bitmap *bitmap;
1da177e4
LT		 4750
4751 if (v == (void*)1) {
2604b703 4752 struct mdk_personality *pers;
1da177e4
LT		 4753 seq_printf(seq, "Personalities : ");
4754 spin_lock(&pers_lock);
2604b703
N		 4755 list_for_each_entry(pers, &pers_list, list)
4756 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 4757
4758 spin_unlock(&pers_lock);
4759 seq_printf(seq, "\n");
d7603b7e 4760 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 4761 return 0;
4762 }
4763 if (v == (void*)2) {
4764 status_unused(seq);
4765 return 0;
4766 }
4767
5dc5cf7d 4768 if (mddev_lock(mddev) < 0)
1da177e4 4769 return -EINTR;
5dc5cf7d 4770
1da177e4
LT		 4771 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4772 seq_printf(seq, "%s : %sactive", mdname(mddev),
4773 mddev->pers ? "" : "in");
4774 if (mddev->pers) {
f91de92e 4775 if (mddev->ro==1)
1da177e4 4776 seq_printf(seq, " (read-only)");
f91de92e
N		 4777 if (mddev->ro==2)
4778 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 4779 seq_printf(seq, " %s", mddev->pers->name);
4780 }
4781
4782 size = 0;
4783 ITERATE_RDEV(mddev,rdev,tmp2) {
4784 char b[BDEVNAME_SIZE];
4785 seq_printf(seq, " %s[%d]",
4786 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 4787 if (test_bit(WriteMostly, &rdev->flags))
4788 seq_printf(seq, "(W)");
b2d444d7 4789 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4790 seq_printf(seq, "(F)");
4791 continue;
b325a32e
N		 4792 } else if (rdev->raid_disk < 0)
4793 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 4794 size += rdev->size;
4795 }
4796
4797 if (!list_empty(&mddev->disks)) {
4798 if (mddev->pers)
4799 seq_printf(seq, "\n %llu blocks",
4800 (unsigned long long)mddev->array_size);
4801 else
4802 seq_printf(seq, "\n %llu blocks",
4803 (unsigned long long)size);
4804 }
1cd6bf19
N		 4805 if (mddev->persistent) {
4806 if (mddev->major_version != 0 ||
4807 mddev->minor_version != 90) {
4808 seq_printf(seq," super %d.%d",
4809 mddev->major_version,
4810 mddev->minor_version);
4811 }
4812 } else
4813 seq_printf(seq, " super non-persistent");
1da177e4
LT		 4814
4815 if (mddev->pers) {
4816 mddev->pers->status (seq, mddev);
4817 seq_printf(seq, "\n ");
8e1b39d6
N		 4818 if (mddev->pers->sync_request) {
4819 if (mddev->curr_resync > 2) {
4820 status_resync (seq, mddev);
4821 seq_printf(seq, "\n ");
4822 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4823 seq_printf(seq, "\tresync=DELAYED\n ");
4824 else if (mddev->recovery_cp < MaxSector)
4825 seq_printf(seq, "\tresync=PENDING\n ");
4826 }
32a7627c
N		 4827 } else
4828 seq_printf(seq, "\n ");
4829
4830 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 4831 unsigned long chunk_kb;
4832 unsigned long flags;
32a7627c
N		 4833 spin_lock_irqsave(&bitmap->lock, flags);
4834 chunk_kb = bitmap->chunksize >> 10;
4835 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4836 "%lu%s chunk",
4837 bitmap->pages - bitmap->missing_pages,
4838 bitmap->pages,
4839 (bitmap->pages - bitmap->missing_pages)
4840 << (PAGE_SHIFT - 10),
4841 chunk_kb ? chunk_kb : bitmap->chunksize,
4842 chunk_kb ? "KB" : "B");
78d742d8
N		 4843 if (bitmap->file) {
4844 seq_printf(seq, ", file: ");
4845 seq_path(seq, bitmap->file->f_vfsmnt,
4846 bitmap->file->f_dentry," \t\n");
32a7627c 4847 }
78d742d8 4848
32a7627c
N		 4849 seq_printf(seq, "\n");
4850 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 4851 }
4852
4853 seq_printf(seq, "\n");
4854 }
4855 mddev_unlock(mddev);
4856
4857 return 0;
4858}
4859
4860static struct seq_operations md_seq_ops = {
4861 .start = md_seq_start,
4862 .next = md_seq_next,
4863 .stop = md_seq_stop,
4864 .show = md_seq_show,
4865};
4866
4867static int md_seq_open(struct inode *inode, struct file *file)
4868{
4869 int error;
d7603b7e
N		 4870 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4871 if (mi == NULL)
4872 return -ENOMEM;
1da177e4
LT		 4873
4874 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 4875 if (error)
4876 kfree(mi);
4877 else {
4878 struct seq_file *p = file->private_data;
4879 p->private = mi;
4880 mi->event = atomic_read(&md_event_count);
4881 }
1da177e4
LT		 4882 return error;
4883}
4884
d7603b7e
N		 4885static int md_seq_release(struct inode *inode, struct file *file)
4886{
4887 struct seq_file *m = file->private_data;
4888 struct mdstat_info *mi = m->private;
4889 m->private = NULL;
4890 kfree(mi);
4891 return seq_release(inode, file);
4892}
4893
4894static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4895{
4896 struct seq_file *m = filp->private_data;
4897 struct mdstat_info *mi = m->private;
4898 int mask;
4899
4900 poll_wait(filp, &md_event_waiters, wait);
4901
4902 /* always allow read */
4903 mask = POLLIN | POLLRDNORM;
4904
4905 if (mi->event != atomic_read(&md_event_count))
4906 mask |= POLLERR | POLLPRI;
4907 return mask;
4908}
4909
1da177e4
LT		 4910static struct file_operations md_seq_fops = {
4911 .open = md_seq_open,
4912 .read = seq_read,
4913 .llseek = seq_lseek,
d7603b7e
N		 4914 .release = md_seq_release,
4915 .poll = mdstat_poll,
1da177e4
LT		 4916};
4917
2604b703 4918int register_md_personality(struct mdk_personality *p)
1da177e4 4919{
1da177e4 4920 spin_lock(&pers_lock);
2604b703
N		 4921 list_add_tail(&p->list, &pers_list);
4922 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 4923 spin_unlock(&pers_lock);
4924 return 0;
4925}
4926
2604b703 4927int unregister_md_personality(struct mdk_personality *p)
1da177e4 4928{
2604b703 4929 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 4930 spin_lock(&pers_lock);
2604b703 4931 list_del_init(&p->list);
1da177e4
LT		 4932 spin_unlock(&pers_lock);
4933 return 0;
4934}
4935
4936static int is_mddev_idle(mddev_t *mddev)
4937{
4938 mdk_rdev_t * rdev;
4939 struct list_head *tmp;
4940 int idle;
4941 unsigned long curr_events;
4942
4943 idle = 1;
4944 ITERATE_RDEV(mddev,rdev,tmp) {
4945 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 4946 curr_events = disk_stat_read(disk, sectors[0]) +
4947 disk_stat_read(disk, sectors[1]) -
1da177e4 4948 atomic_read(&disk->sync_io);
c0e48521
N		 4949 /* The difference between curr_events and last_events
4950 * will be affected by any new non-sync IO (making
4951 * curr_events bigger) and any difference in the amount of
4952 * in-flight syncio (making current_events bigger or smaller)
4953 * The amount in-flight is currently limited to
4954 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4955 * which is at most 4096 sectors.
4956 * These numbers are fairly fragile and should be made
4957 * more robust, probably by enforcing the
4958 * 'window size' that md_do_sync sort-of uses.
4959 *
1da177e4
LT		 4960 * Note: the following is an unsigned comparison.
4961 */
c0e48521 4962 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 4963 rdev->last_events = curr_events;
4964 idle = 0;
4965 }
4966 }
4967 return idle;
4968}
4969
4970void md_done_sync(mddev_t *mddev, int blocks, int ok)
4971{
4972 /* another "blocks" (512byte) blocks have been synced */
4973 atomic_sub(blocks, &mddev->recovery_active);
4974 wake_up(&mddev->recovery_wait);
4975 if (!ok) {
4976 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4977 md_wakeup_thread(mddev->thread);
4978 // stop recovery, signal do_sync ....
4979 }
4980}
4981
4982
06d91a5f
N		 4983/* md_write_start(mddev, bi)
4984 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 4985 * in superblock) before writing, schedule a superblock update
4986 * and wait for it to complete.
06d91a5f 4987 */
3d310eb7 4988void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 4989{
06d91a5f 4990 if (bio_data_dir(bi) != WRITE)
3d310eb7 4991 return;
06d91a5f 4992
f91de92e
N		 4993 BUG_ON(mddev->ro == 1);
4994 if (mddev->ro == 2) {
4995 /* need to switch to read/write */
4996 mddev->ro = 0;
4997 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4998 md_wakeup_thread(mddev->thread);
4999 }
06d91a5f 5000 atomic_inc(&mddev->writes_pending);
06d91a5f 5001 if (mddev->in_sync) {
a9701a30 5002 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5003 if (mddev->in_sync) {
5004 mddev->in_sync = 0;
850b2b42 5005 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7
N		 5006 md_wakeup_thread(mddev->thread);
5007 }
a9701a30 5008 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5009 }
850b2b42 5010 wait_event(mddev->sb_wait, mddev->flags==0);
1da177e4
LT		 5011}
5012
5013void md_write_end(mddev_t *mddev)
5014{
5015 if (atomic_dec_and_test(&mddev->writes_pending)) {
5016 if (mddev->safemode == 2)
5017 md_wakeup_thread(mddev->thread);
16f17b39 5018 else if (mddev->safemode_delay)
1da177e4
LT		 5019 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5020 }
5021}
5022
75c96f85 5023static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5024
5025#define SYNC_MARKS 10
5026#define SYNC_MARK_STEP (3*HZ)
29269553 5027void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5028{
5029 mddev_t *mddev2;
5030 unsigned int currspeed = 0,
5031 window;
57afd89f 5032 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5033 unsigned long mark[SYNC_MARKS];
5034 sector_t mark_cnt[SYNC_MARKS];
5035 int last_mark,m;
5036 struct list_head *tmp;
5037 sector_t last_check;
57afd89f 5038 int skipped = 0;
5fd6c1dc
N		 5039 struct list_head *rtmp;
5040 mdk_rdev_t *rdev;
61df9d91 5041 char *desc;
1da177e4
LT		 5042
5043 /* just incase thread restarts... */
5044 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5045 return;
5fd6c1dc
N		 5046 if (mddev->ro) /* never try to sync a read-only array */
5047 return;
1da177e4 5048
61df9d91
N		 5049 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5050 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5051 desc = "data-check";
5052 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5053 desc = "requested-resync";
5054 else
5055 desc = "resync";
5056 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5057 desc = "reshape";
5058 else
5059 desc = "recovery";
5060
1da177e4
LT		 5061 /* we overload curr_resync somewhat here.
5062 * 0 == not engaged in resync at all
5063 * 2 == checking that there is no conflict with another sync
5064 * 1 == like 2, but have yielded to allow conflicting resync to
5065 * commense
5066 * other == active in resync - this many blocks
5067 *
5068 * Before starting a resync we must have set curr_resync to
5069 * 2, and then checked that every "conflicting" array has curr_resync
5070 * less than ours. When we find one that is the same or higher
5071 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5072 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5073 * This will mean we have to start checking from the beginning again.
5074 *
5075 */
5076
5077 do {
5078 mddev->curr_resync = 2;
5079
5080 try_again:
787453c2 5081 if (kthread_should_stop()) {
6985c43f 5082 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5083 goto skip;
5084 }
5085 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 5086 if (mddev2 == mddev)
5087 continue;
5088 if (mddev2->curr_resync &&
5089 match_mddev_units(mddev,mddev2)) {
5090 DEFINE_WAIT(wq);
5091 if (mddev < mddev2 && mddev->curr_resync == 2) {
5092 /* arbitrarily yield */
5093 mddev->curr_resync = 1;
5094 wake_up(&resync_wait);
5095 }
5096 if (mddev > mddev2 && mddev->curr_resync == 1)
5097 /* no need to wait here, we can wait the next
5098 * time 'round when curr_resync == 2
5099 */
5100 continue;
787453c2
N		 5101 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5102 if (!kthread_should_stop() &&
8712e553 5103 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 5104 printk(KERN_INFO "md: delaying %s of %s"
5105 " until %s has finished (they"
1da177e4 5106 " share one or more physical units)\n",
61df9d91 5107 desc, mdname(mddev), mdname(mddev2));
1da177e4
LT		 5108 mddev_put(mddev2);
5109 schedule();
5110 finish_wait(&resync_wait, &wq);
5111 goto try_again;
5112 }
5113 finish_wait(&resync_wait, &wq);
5114 }
5115 }
5116 } while (mddev->curr_resync < 2);
5117
5fd6c1dc 5118 j = 0;
9d88883e 5119 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5120 /* resync follows the size requested by the personality,
57afd89f 5121 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5122 */
5123 max_sectors = mddev->resync_max_sectors;
9d88883e 5124 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5125 /* we don't use the checkpoint if there's a bitmap */
5126 if (!mddev->bitmap &&
5127 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5128 j = mddev->recovery_cp;
ccfcc3c1
N		 5129 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5130 max_sectors = mddev->size << 1;
5fd6c1dc 5131 else {
1da177e4
LT		 5132 /* recovery follows the physical size of devices */
5133 max_sectors = mddev->size << 1;
5fd6c1dc
N		 5134 j = MaxSector;
5135 ITERATE_RDEV(mddev,rdev,rtmp)
5136 if (rdev->raid_disk >= 0 &&
5137 !test_bit(Faulty, &rdev->flags) &&
5138 !test_bit(In_sync, &rdev->flags) &&
5139 rdev->recovery_offset < j)
5140 j = rdev->recovery_offset;
5141 }
1da177e4 5142
61df9d91
N		 5143 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5144 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5145 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 5146 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 5147 "(but not more than %d KB/sec) for %s.\n",
5148 speed_max(mddev), desc);
1da177e4
LT		 5149
5150 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5151
57afd89f 5152 io_sectors = 0;
1da177e4
LT		 5153 for (m = 0; m < SYNC_MARKS; m++) {
5154 mark[m] = jiffies;
57afd89f 5155 mark_cnt[m] = io_sectors;
1da177e4
LT		 5156 }
5157 last_mark = 0;
5158 mddev->resync_mark = mark[last_mark];
5159 mddev->resync_mark_cnt = mark_cnt[last_mark];
5160
5161 /*
5162 * Tune reconstruction:
5163 */
5164 window = 32*(PAGE_SIZE/512);
5165 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5166 window/2,(unsigned long long) max_sectors/2);
5167
5168 atomic_set(&mddev->recovery_active, 0);
5169 init_waitqueue_head(&mddev->recovery_wait);
5170 last_check = 0;
5171
5172 if (j>2) {
5173 printk(KERN_INFO
61df9d91
N		 5174 "md: resuming %s of %s from checkpoint.\n",
5175 desc, mdname(mddev));
1da177e4
LT		 5176 mddev->curr_resync = j;
5177 }
5178
5179 while (j < max_sectors) {
57afd89f 5180 sector_t sectors;
1da177e4 5181
57afd89f
N		 5182 skipped = 0;
5183 sectors = mddev->pers->sync_request(mddev, j, &skipped,
88202a0c 5184 currspeed < speed_min(mddev));
57afd89f 5185 if (sectors == 0) {
1da177e4
LT		 5186 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5187 goto out;
5188 }
57afd89f
N		 5189
5190 if (!skipped) { /* actual IO requested */
5191 io_sectors += sectors;
5192 atomic_add(sectors, &mddev->recovery_active);
5193 }
5194
1da177e4
LT		 5195 j += sectors;
5196 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5197 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5198 if (last_check == 0)
5199 /* this is the earliers that rebuilt will be
5200 * visible in /proc/mdstat
5201 */
5202 md_new_event(mddev);
57afd89f
N		 5203
5204 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5205 continue;
5206
57afd89f 5207 last_check = io_sectors;
1da177e4
LT		 5208
5209 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5210 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5211 break;
5212
5213 repeat:
5214 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5215 /* step marks */
5216 int next = (last_mark+1) % SYNC_MARKS;
5217
5218 mddev->resync_mark = mark[next];
5219 mddev->resync_mark_cnt = mark_cnt[next];
5220 mark[next] = jiffies;
57afd89f 5221 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5222 last_mark = next;
5223 }
5224
5225
787453c2 5226 if (kthread_should_stop()) {
1da177e4
LT		 5227 /*
5228 * got a signal, exit.
5229 */
5230 printk(KERN_INFO
5231 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 5232 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5233 goto out;
5234 }
5235
5236 /*
5237 * this loop exits only if either when we are slower than
5238 * the 'hard' speed limit, or the system was IO-idle for
5239 * a jiffy.
5240 * the system might be non-idle CPU-wise, but we only care
5241 * about not overloading the IO subsystem. (things like an
5242 * e2fsck being done on the RAID array should execute fast)
5243 */
5244 mddev->queue->unplug_fn(mddev->queue);
5245 cond_resched();
5246
57afd89f
N		 5247 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5248 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5249
88202a0c
N		 5250 if (currspeed > speed_min(mddev)) {
5251 if ((currspeed > speed_max(mddev)) ||
1da177e4 5252 !is_mddev_idle(mddev)) {
c0e48521 5253 msleep(500);
1da177e4
LT		 5254 goto repeat;
5255 }
5256 }
5257 }
61df9d91 5258 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 5259 /*
5260 * this also signals 'finished resyncing' to md_stop
5261 */
5262 out:
5263 mddev->queue->unplug_fn(mddev->queue);
5264
5265 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5266
5267 /* tell personality that we are finished */
57afd89f 5268 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5269
5270 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1
N		 5271 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
5272 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5273 mddev->curr_resync > 2) {
5274 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5275 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5276 if (mddev->curr_resync >= mddev->recovery_cp) {
5277 printk(KERN_INFO
61df9d91
N		 5278 "md: checkpointing %s of %s.\n",
5279 desc, mdname(mddev));
5fd6c1dc
N		 5280 mddev->recovery_cp = mddev->curr_resync;
5281 }
5282 } else
5283 mddev->recovery_cp = MaxSector;
5284 } else {
5285 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5286 mddev->curr_resync = MaxSector;
5287 ITERATE_RDEV(mddev,rdev,rtmp)
5288 if (rdev->raid_disk >= 0 &&
5289 !test_bit(Faulty, &rdev->flags) &&
5290 !test_bit(In_sync, &rdev->flags) &&
5291 rdev->recovery_offset < mddev->curr_resync)
5292 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 5293 }
1da177e4
LT		 5294 }
5295
1da177e4
LT		 5296 skip:
5297 mddev->curr_resync = 0;
5298 wake_up(&resync_wait);
5299 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5300 md_wakeup_thread(mddev->thread);
5301}
29269553 5302EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5303
5304
5305/*
5306 * This routine is regularly called by all per-raid-array threads to
5307 * deal with generic issues like resync and super-block update.
5308 * Raid personalities that don't have a thread (linear/raid0) do not
5309 * need this as they never do any recovery or update the superblock.
5310 *
5311 * It does not do any resync itself, but rather "forks" off other threads
5312 * to do that as needed.
5313 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5314 * "->recovery" and create a thread at ->sync_thread.
5315 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5316 * and wakeups up this thread which will reap the thread and finish up.
5317 * This thread also removes any faulty devices (with nr_pending == 0).
5318 *
5319 * The overall approach is:
5320 * 1/ if the superblock needs updating, update it.
5321 * 2/ If a recovery thread is running, don't do anything else.
5322 * 3/ If recovery has finished, clean up, possibly marking spares active.
5323 * 4/ If there are any faulty devices, remove them.
5324 * 5/ If array is degraded, try to add spares devices
5325 * 6/ If array has spares or is not in-sync, start a resync thread.
5326 */
5327void md_check_recovery(mddev_t *mddev)
5328{
5329 mdk_rdev_t *rdev;
5330 struct list_head *rtmp;
5331
5332
5f40402d
N		 5333 if (mddev->bitmap)
5334 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5335
5336 if (mddev->ro)
5337 return;
fca4d848
N		 5338
5339 if (signal_pending(current)) {
5340 if (mddev->pers->sync_request) {
5341 printk(KERN_INFO "md: %s in immediate safe mode\n",
5342 mdname(mddev));
5343 mddev->safemode = 2;
5344 }
5345 flush_signals(current);
5346 }
5347
1da177e4 5348 if ( ! (
850b2b42 5349 mddev->flags ||
1da177e4 5350 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5351 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5352 (mddev->safemode == 1) ||
5353 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5354 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5355 ))
5356 return;
fca4d848 5357
df5b89b3 5358 if (mddev_trylock(mddev)) {
1da177e4 5359 int spares =0;
fca4d848 5360
a9701a30 5361 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5362 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5363 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5364 mddev->in_sync = 1;
850b2b42 5365 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
fca4d848
N		 5366 }
5367 if (mddev->safemode == 1)
5368 mddev->safemode = 0;
a9701a30 5369 spin_unlock_irq(&mddev->write_lock);
fca4d848 5370
850b2b42
N		 5371 if (mddev->flags)
5372 md_update_sb(mddev, 0);
06d91a5f 5373
06d91a5f 5374
1da177e4
LT		 5375 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5376 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5377 /* resync/recovery still happening */
5378 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5379 goto unlock;
5380 }
5381 if (mddev->sync_thread) {
5382 /* resync has finished, collect result */
5383 md_unregister_thread(mddev->sync_thread);
5384 mddev->sync_thread = NULL;
5385 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5386 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5387 /* success...*/
5388 /* activate any spares */
5389 mddev->pers->spare_active(mddev);
5390 }
850b2b42 5391 md_update_sb(mddev, 1);
41158c7e
N		 5392
5393 /* if array is no-longer degraded, then any saved_raid_disk
5394 * information must be scrapped
5395 */
5396 if (!mddev->degraded)
5397 ITERATE_RDEV(mddev,rdev,rtmp)
5398 rdev->saved_raid_disk = -1;
5399
1da177e4
LT		 5400 mddev->recovery = 0;
5401 /* flag recovery needed just to double check */
5402 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5403 md_new_event(mddev);
1da177e4
LT		 5404 goto unlock;
5405 }
24dd469d
N		 5406 /* Clear some bits that don't mean anything, but
5407 * might be left set
5408 */
5409 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5410 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5411 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5412 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5413
5fd6c1dc
N		 5414 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5415 goto unlock;
1da177e4
LT		 5416 /* no recovery is running.
5417 * remove any failed drives, then
5418 * add spares if possible.
5419 * Spare are also removed and re-added, to allow
5420 * the personality to fail the re-add.
5421 */
5422 ITERATE_RDEV(mddev,rdev,rtmp)
5423 if (rdev->raid_disk >= 0 &&
b2d444d7 5424 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 5425 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 5426 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5427 char nm[20];
5428 sprintf(nm,"rd%d", rdev->raid_disk);
5429 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 5430 rdev->raid_disk = -1;
86e6ffdd 5431 }
1da177e4
LT		 5432 }
5433
5434 if (mddev->degraded) {
5435 ITERATE_RDEV(mddev,rdev,rtmp)
5436 if (rdev->raid_disk < 0
b2d444d7 5437 && !test_bit(Faulty, &rdev->flags)) {
5fd6c1dc 5438 rdev->recovery_offset = 0;
86e6ffdd
N		 5439 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5440 char nm[20];
5441 sprintf(nm, "rd%d", rdev->raid_disk);
5442 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 5443 spares++;
d7603b7e 5444 md_new_event(mddev);
86e6ffdd 5445 } else
1da177e4
LT		 5446 break;
5447 }
5448 }
5449
24dd469d
N		 5450 if (spares) {
5451 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5452 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5453 } else if (mddev->recovery_cp < MaxSector) {
5454 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5455 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5456 /* nothing to be done ... */
1da177e4 5457 goto unlock;
24dd469d 5458
1da177e4
LT		 5459 if (mddev->pers->sync_request) {
5460 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5461 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5462 /* We are adding a device or devices to an array
5463 * which has the bitmap stored on all devices.
5464 * So make sure all bitmap pages get written
5465 */
5466 bitmap_write_all(mddev->bitmap);
5467 }
1da177e4
LT		 5468 mddev->sync_thread = md_register_thread(md_do_sync,
5469 mddev,
5470 "%s_resync");
5471 if (!mddev->sync_thread) {
5472 printk(KERN_ERR "%s: could not start resync"
5473 " thread...\n",
5474 mdname(mddev));
5475 /* leave the spares where they are, it shouldn't hurt */
5476 mddev->recovery = 0;
d7603b7e 5477 } else
1da177e4 5478 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5479 md_new_event(mddev);
1da177e4
LT		 5480 }
5481 unlock:
5482 mddev_unlock(mddev);
5483 }
5484}
5485
75c96f85
AB		 5486static int md_notify_reboot(struct notifier_block *this,
5487 unsigned long code, void *x)
1da177e4
LT		 5488{
5489 struct list_head *tmp;
5490 mddev_t *mddev;
5491
5492 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5493
5494 printk(KERN_INFO "md: stopping all md devices.\n");
5495
5496 ITERATE_MDDEV(mddev,tmp)
c71d4887 5497 if (mddev_trylock(mddev)) {
1da177e4 5498 do_md_stop (mddev, 1);
c71d4887
NB		 5499 mddev_unlock(mddev);
5500 }
1da177e4
LT		 5501 /*
5502 * certain more exotic SCSI devices are known to be
5503 * volatile wrt too early system reboots. While the
5504 * right place to handle this issue is the given
5505 * driver, we do want to have a safe RAID driver ...
5506 */
5507 mdelay(1000*1);
5508 }
5509 return NOTIFY_DONE;
5510}
5511
75c96f85 5512static struct notifier_block md_notifier = {
1da177e4
LT		 5513 .notifier_call = md_notify_reboot,
5514 .next = NULL,
5515 .priority = INT_MAX, /* before any real devices */
5516};
5517
5518static void md_geninit(void)
5519{
5520 struct proc_dir_entry *p;
5521
5522 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5523
5524 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5525 if (p)
5526 p->proc_fops = &md_seq_fops;
5527}
5528
75c96f85 5529static int __init md_init(void)
1da177e4 5530{
1da177e4
LT		 5531 if (register_blkdev(MAJOR_NR, "md"))
5532 return -1;
5533 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5534 unregister_blkdev(MAJOR_NR, "md");
5535 return -1;
5536 }
e8703fe1
N		 5537 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5538 md_probe, NULL, NULL);
5539 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 5540 md_probe, NULL, NULL);
5541
1da177e4
LT		 5542 register_reboot_notifier(&md_notifier);
5543 raid_table_header = register_sysctl_table(raid_root_table, 1);
5544
5545 md_geninit();
5546 return (0);
5547}
5548
5549
5550#ifndef MODULE
5551
5552/*
5553 * Searches all registered partitions for autorun RAID arrays
5554 * at boot time.
5555 */
5556static dev_t detected_devices[128];
5557static int dev_cnt;
5558
5559void md_autodetect_dev(dev_t dev)
5560{
5561 if (dev_cnt >= 0 && dev_cnt < 127)
5562 detected_devices[dev_cnt++] = dev;
5563}
5564
5565
5566static void autostart_arrays(int part)
5567{
5568 mdk_rdev_t *rdev;
5569 int i;
5570
5571 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5572
5573 for (i = 0; i < dev_cnt; i++) {
5574 dev_t dev = detected_devices[i];
5575
5576 rdev = md_import_device(dev,0, 0);
5577 if (IS_ERR(rdev))
5578 continue;
5579
b2d444d7 5580 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5581 MD_BUG();
5582 continue;
5583 }
5584 list_add(&rdev->same_set, &pending_raid_disks);
5585 }
5586 dev_cnt = 0;
5587
5588 autorun_devices(part);
5589}
5590
5591#endif
5592
5593static __exit void md_exit(void)
5594{
5595 mddev_t *mddev;
5596 struct list_head *tmp;
8ab5e4c1 5597
e8703fe1
N		 5598 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5599 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4
LT		 5600
5601 unregister_blkdev(MAJOR_NR,"md");
5602 unregister_blkdev(mdp_major, "mdp");
5603 unregister_reboot_notifier(&md_notifier);
5604 unregister_sysctl_table(raid_table_header);
5605 remove_proc_entry("mdstat", NULL);
5606 ITERATE_MDDEV(mddev,tmp) {
5607 struct gendisk *disk = mddev->gendisk;
5608 if (!disk)
5609 continue;
5610 export_array(mddev);
5611 del_gendisk(disk);
5612 put_disk(disk);
5613 mddev->gendisk = NULL;
5614 mddev_put(mddev);
5615 }
5616}
5617
5618module_init(md_init)
5619module_exit(md_exit)
5620
f91de92e
N		 5621static int get_ro(char *buffer, struct kernel_param *kp)
5622{
5623 return sprintf(buffer, "%d", start_readonly);
5624}
5625static int set_ro(const char *val, struct kernel_param *kp)
5626{
5627 char *e;
5628 int num = simple_strtoul(val, &e, 10);
5629 if (*val && (*e == '\0' || *e == '\n')) {
5630 start_readonly = num;
4dbcdc75 5631 return 0;
f91de92e
N		 5632 }
5633 return -EINVAL;
5634}
5635
80ca3a44
N		 5636module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5637module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 5638
f91de92e 5639
1da177e4
LT		 5640EXPORT_SYMBOL(register_md_personality);
5641EXPORT_SYMBOL(unregister_md_personality);
5642EXPORT_SYMBOL(md_error);
5643EXPORT_SYMBOL(md_done_sync);
5644EXPORT_SYMBOL(md_write_start);
5645EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 5646EXPORT_SYMBOL(md_register_thread);
5647EXPORT_SYMBOL(md_unregister_thread);
5648EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 5649EXPORT_SYMBOL(md_check_recovery);
5650MODULE_LICENSE("GPL");
aa1595e9 5651MODULE_ALIAS("md");
72008652 5652MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Clone of davem's net-next-2.6 tree