projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: reinitialise more mddev fields in do_md_stop.
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
50511da3 36#include <linux/kernel.h>
a6fb0934 37#include <linux/kthread.h>
1da177e4
LT		 38#include <linux/linkage.h>
39#include <linux/raid/md.h>
32a7627c 40#include <linux/raid/bitmap.h>
1da177e4 41#include <linux/sysctl.h>
1da177e4 42#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 43#include <linux/poll.h>
48c9c27b 44#include <linux/mutex.h>
16f17b39 45#include <linux/ctype.h>
7dfb7103 46#include <linux/freezer.h>
1da177e4
LT		 47
48#include <linux/init.h>
49
32a7627c
N		 50#include <linux/file.h>
51
1da177e4
LT		 52#ifdef CONFIG_KMOD
53#include <linux/kmod.h>
54#endif
55
56#include <asm/unaligned.h>
57
58#define MAJOR_NR MD_MAJOR
59#define MD_DRIVER
60
61/* 63 partitions with the alternate major number (mdp) */
62#define MdpMinorShift 6
63
64#define DEBUG 0
65#define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68#ifndef MODULE
69static void autostart_arrays (int part);
70#endif
71
2604b703 72static LIST_HEAD(pers_list);
1da177e4
LT		 73static DEFINE_SPINLOCK(pers_lock);
74
5e56341d
AB		 75static void md_print_devices(void);
76
77#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
1da177e4
LT		 79/*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 83 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 89 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 90 */
91
92static int sysctl_speed_limit_min = 1000;
93static int sysctl_speed_limit_max = 200000;
88202a0c
N		 94static inline int speed_min(mddev_t *mddev)
95{
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98}
99
100static inline int speed_max(mddev_t *mddev)
101{
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104}
1da177e4
LT		 105
106static struct ctl_table_header *raid_table_header;
107
108static ctl_table raid_table[] = {
109 {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
80ca3a44 114 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 115 .proc_handler = &proc_dointvec,
116 },
117 {
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 123 .proc_handler = &proc_dointvec,
124 },
125 { .ctl_name = 0 }
126};
127
128static ctl_table raid_dir_table[] = {
129 {
130 .ctl_name = DEV_RAID,
131 .procname = "raid",
132 .maxlen = 0,
80ca3a44 133 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 134 .child = raid_table,
135 },
136 { .ctl_name = 0 }
137};
138
139static ctl_table raid_root_table[] = {
140 {
141 .ctl_name = CTL_DEV,
142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
147 { .ctl_name = 0 }
148};
149
150static struct block_device_operations md_fops;
151
f91de92e
N		 152static int start_readonly;
153
d7603b7e
N		 154/*
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
158 * count increases.
159 *
160 * Events are:
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
163 */
2989ddbd 164static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 165static atomic_t md_event_count;
29269553 166void md_new_event(mddev_t *mddev)
d7603b7e
N		 167{
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
4508a7a7 170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
d7603b7e 171}
29269553 172EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 173
c331eb04
N		 174/* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
176 */
05381954 177static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 178{
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
181}
182
1da177e4
LT		 183/*
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
186 */
187static LIST_HEAD(all_mddevs);
188static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191/*
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
197 */
29ac4aa3 198#define for_each_mddev(mddev,tmp) \
1da177e4
LT		 199 \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
202 mddev = NULL;}); \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
210 tmp = tmp->next;}) \
211 )
212
213
165125e1 214static int md_fail_request (struct request_queue *q, struct bio *bio)
1da177e4 215{
6712ecf8 216 bio_io_error(bio);
1da177e4
LT		 217 return 0;
218}
219
220static inline mddev_t *mddev_get(mddev_t *mddev)
221{
222 atomic_inc(&mddev->active);
223 return mddev;
224}
225
226static void mddev_put(mddev_t *mddev)
227{
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229 return;
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
926ce2d8 232 spin_unlock(&all_mddevs_lock);
1312f40e 233 blk_cleanup_queue(mddev->queue);
c10997f6 234 kobject_put(&mddev->kobj);
926ce2d8
N		 235 } else
236 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 237}
238
239static mddev_t * mddev_find(dev_t unit)
240{
241 mddev_t *mddev, *new = NULL;
242
243 retry:
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
247 mddev_get(mddev);
248 spin_unlock(&all_mddevs_lock);
990a8baf 249 kfree(new);
1da177e4
LT		 250 return mddev;
251 }
252
253 if (new) {
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
256 return new;
257 }
258 spin_unlock(&all_mddevs_lock);
259
9ffae0cf 260 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 261 if (!new)
262 return NULL;
263
1da177e4
LT		 264 new->unit = unit;
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
267 else
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
df5b89b3 270 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
06d91a5f 275 spin_lock_init(&new->write_lock);
3d310eb7 276 init_waitqueue_head(&new->sb_wait);
08a02ecd 277 new->reshape_position = MaxSector;
c6207277 278 new->resync_max = MaxSector;
d897dbf9 279 new->level = LEVEL_NONE;
1da177e4
LT		 280
281 new->queue = blk_alloc_queue(GFP_KERNEL);
282 if (!new->queue) {
283 kfree(new);
284 return NULL;
285 }
75ad23bc
NP		 286 /* Can be unlocked because the queue is new: no concurrency */
287 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
1da177e4
LT		 288
289 blk_queue_make_request(new->queue, md_fail_request);
290
291 goto retry;
292}
293
294static inline int mddev_lock(mddev_t * mddev)
295{
df5b89b3 296 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 297}
298
1da177e4
LT		 299static inline int mddev_trylock(mddev_t * mddev)
300{
df5b89b3 301 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 302}
303
304static inline void mddev_unlock(mddev_t * mddev)
305{
df5b89b3 306 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 307
005eca5e 308 md_wakeup_thread(mddev->thread);
1da177e4
LT		 309}
310
2989ddbd 311static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 312{
313 mdk_rdev_t * rdev;
314 struct list_head *tmp;
315
d089c6af 316 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 317 if (rdev->desc_nr == nr)
318 return rdev;
319 }
320 return NULL;
321}
322
323static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
324{
325 struct list_head *tmp;
326 mdk_rdev_t *rdev;
327
d089c6af 328 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 329 if (rdev->bdev->bd_dev == dev)
330 return rdev;
331 }
332 return NULL;
333}
334
d9d166c2 335static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 336{
337 struct mdk_personality *pers;
d9d166c2
N		 338 list_for_each_entry(pers, &pers_list, list) {
339 if (level != LEVEL_NONE && pers->level == level)
2604b703 340 return pers;
d9d166c2
N		 341 if (strcmp(pers->name, clevel)==0)
342 return pers;
343 }
2604b703
N		 344 return NULL;
345}
346
77933d72 347static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 348{
349 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
350 return MD_NEW_SIZE_BLOCKS(size);
351}
352
353static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
354{
355 sector_t size;
356
357 size = rdev->sb_offset;
358
359 if (chunk_size)
360 size &= ~((sector_t)chunk_size/1024 - 1);
361 return size;
362}
363
364static int alloc_disk_sb(mdk_rdev_t * rdev)
365{
366 if (rdev->sb_page)
367 MD_BUG();
368
369 rdev->sb_page = alloc_page(GFP_KERNEL);
370 if (!rdev->sb_page) {
371 printk(KERN_ALERT "md: out of memory.\n");
372 return -EINVAL;
373 }
374
375 return 0;
376}
377
378static void free_disk_sb(mdk_rdev_t * rdev)
379{
380 if (rdev->sb_page) {
2d1f3b5d 381 put_page(rdev->sb_page);
1da177e4
LT		 382 rdev->sb_loaded = 0;
383 rdev->sb_page = NULL;
384 rdev->sb_offset = 0;
385 rdev->size = 0;
386 }
387}
388
389
6712ecf8 390static void super_written(struct bio *bio, int error)
7bfa19f2
N		 391{
392 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 393 mddev_t *mddev = rdev->mddev;
7bfa19f2 394
3a0f5bbb
N		 395 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
396 printk("md: super_written gets error=%d, uptodate=%d\n",
397 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
398 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 399 md_error(mddev, rdev);
3a0f5bbb 400 }
7bfa19f2 401
a9701a30
N		 402 if (atomic_dec_and_test(&mddev->pending_writes))
403 wake_up(&mddev->sb_wait);
f8b58edf 404 bio_put(bio);
7bfa19f2
N		 405}
406
6712ecf8 407static void super_written_barrier(struct bio *bio, int error)
a9701a30
N		 408{
409 struct bio *bio2 = bio->bi_private;
410 mdk_rdev_t *rdev = bio2->bi_private;
411 mddev_t *mddev = rdev->mddev;
a9701a30
N		 412
413 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
414 error == -EOPNOTSUPP) {
415 unsigned long flags;
416 /* barriers don't appear to be supported :-( */
417 set_bit(BarriersNotsupp, &rdev->flags);
418 mddev->barriers_work = 0;
419 spin_lock_irqsave(&mddev->write_lock, flags);
420 bio2->bi_next = mddev->biolist;
421 mddev->biolist = bio2;
422 spin_unlock_irqrestore(&mddev->write_lock, flags);
423 wake_up(&mddev->sb_wait);
424 bio_put(bio);
6712ecf8
N		 425 } else {
426 bio_put(bio2);
427 bio->bi_private = rdev;
428 super_written(bio, error);
a9701a30 429 }
a9701a30
N		 430}
431
7bfa19f2
N		 432void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
433 sector_t sector, int size, struct page *page)
434{
435 /* write first size bytes of page to sector of rdev
436 * Increment mddev->pending_writes before returning
437 * and decrement it on completion, waking up sb_wait
438 * if zero is reached.
439 * If an error occurred, call md_error
a9701a30
N		 440 *
441 * As we might need to resubmit the request if BIO_RW_BARRIER
442 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 443 */
444 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 445 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 446
447 bio->bi_bdev = rdev->bdev;
448 bio->bi_sector = sector;
449 bio_add_page(bio, page, size, 0);
450 bio->bi_private = rdev;
451 bio->bi_end_io = super_written;
a9701a30
N		 452 bio->bi_rw = rw;
453
7bfa19f2 454 atomic_inc(&mddev->pending_writes);
a9701a30
N		 455 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
456 struct bio *rbio;
457 rw |= (1<<BIO_RW_BARRIER);
458 rbio = bio_clone(bio, GFP_NOIO);
459 rbio->bi_private = bio;
460 rbio->bi_end_io = super_written_barrier;
461 submit_bio(rw, rbio);
462 } else
463 submit_bio(rw, bio);
464}
465
466void md_super_wait(mddev_t *mddev)
467{
468 /* wait for all superblock writes that were scheduled to complete.
469 * if any had to be retried (due to BARRIER problems), retry them
470 */
471 DEFINE_WAIT(wq);
472 for(;;) {
473 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
474 if (atomic_read(&mddev->pending_writes)==0)
475 break;
476 while (mddev->biolist) {
477 struct bio *bio;
478 spin_lock_irq(&mddev->write_lock);
479 bio = mddev->biolist;
480 mddev->biolist = bio->bi_next ;
481 bio->bi_next = NULL;
482 spin_unlock_irq(&mddev->write_lock);
483 submit_bio(bio->bi_rw, bio);
484 }
485 schedule();
486 }
487 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 488}
489
6712ecf8 490static void bi_complete(struct bio *bio, int error)
1da177e4 491{
1da177e4 492 complete((struct completion*)bio->bi_private);
1da177e4
LT		 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
4d167f09
N		 588
589static u32 md_csum_fold(u32 csum)
590{
591 csum = (csum & 0xffff) + (csum >> 16);
592 return (csum & 0xffff) + (csum >> 16);
593}
594
1da177e4
LT		 595static unsigned int calc_sb_csum(mdp_super_t * sb)
596{
4d167f09
N		 597 u64 newcsum = 0;
598 u32 *sb32 = (u32*)sb;
599 int i;
1da177e4
LT		 600 unsigned int disk_csum, csum;
601
602 disk_csum = sb->sb_csum;
603 sb->sb_csum = 0;
4d167f09
N		 604
605 for (i = 0; i < MD_SB_BYTES/4 ; i++)
606 newcsum += sb32[i];
607 csum = (newcsum & 0xffffffff) + (newcsum>>32);
608
609
610#ifdef CONFIG_ALPHA
611 /* This used to use csum_partial, which was wrong for several
612 * reasons including that different results are returned on
613 * different architectures. It isn't critical that we get exactly
614 * the same return value as before (we always csum_fold before
615 * testing, and that removes any differences). However as we
616 * know that csum_partial always returned a 16bit value on
617 * alphas, do a fold to maximise conformity to previous behaviour.
618 */
619 sb->sb_csum = md_csum_fold(disk_csum);
620#else
1da177e4 621 sb->sb_csum = disk_csum;
4d167f09 622#endif
1da177e4
LT		 623 return csum;
624}
625
626
627/*
628 * Handle superblock details.
629 * We want to be able to handle multiple superblock formats
630 * so we have a common interface to them all, and an array of
631 * different handlers.
632 * We rely on user-space to write the initial superblock, and support
633 * reading and updating of superblocks.
634 * Interface methods are:
635 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
636 * loads and validates a superblock on dev.
637 * if refdev != NULL, compare superblocks on both devices
638 * Return:
639 * 0 - dev has a superblock that is compatible with refdev
640 * 1 - dev has a superblock that is compatible and newer than refdev
641 * so dev should be used as the refdev in future
642 * -EINVAL superblock incompatible or invalid
643 * -othererror e.g. -EIO
644 *
645 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
646 * Verify that dev is acceptable into mddev.
647 * The first time, mddev->raid_disks will be 0, and data from
648 * dev should be merged in. Subsequent calls check that dev
649 * is new enough. Return 0 or -EINVAL
650 *
651 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
652 * Update the superblock for rdev with data in mddev
653 * This does not write to disc.
654 *
655 */
656
657struct super_type {
658 char *name;
659 struct module *owner;
660 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
661 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
662 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
663};
664
665/*
666 * load_super for 0.90.0
667 */
668static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
669{
670 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
671 mdp_super_t *sb;
672 int ret;
673 sector_t sb_offset;
674
675 /*
676 * Calculate the position of the superblock,
677 * it's at the end of the disk.
678 *
679 * It also happens to be a multiple of 4Kb.
680 */
681 sb_offset = calc_dev_sboffset(rdev->bdev);
682 rdev->sb_offset = sb_offset;
683
0002b271 684 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 685 if (ret) return ret;
686
687 ret = -EINVAL;
688
689 bdevname(rdev->bdev, b);
690 sb = (mdp_super_t*)page_address(rdev->sb_page);
691
692 if (sb->md_magic != MD_SB_MAGIC) {
693 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
694 b);
695 goto abort;
696 }
697
698 if (sb->major_version != 0 ||
f6705578
N		 699 sb->minor_version < 90 ||
700 sb->minor_version > 91) {
1da177e4
LT		 701 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
702 sb->major_version, sb->minor_version,
703 b);
704 goto abort;
705 }
706
707 if (sb->raid_disks <= 0)
708 goto abort;
709
4d167f09 710 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 711 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
712 b);
713 goto abort;
714 }
715
716 rdev->preferred_minor = sb->md_minor;
717 rdev->data_offset = 0;
0002b271 718 rdev->sb_size = MD_SB_BYTES;
1da177e4 719
e11e93fa
N		 720 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
721 if (sb->level != 1 && sb->level != 4
722 && sb->level != 5 && sb->level != 6
723 && sb->level != 10) {
724 /* FIXME use a better test */
725 printk(KERN_WARNING
726 "md: bitmaps not supported for this level.\n");
727 goto abort;
728 }
729 }
730
1da177e4
LT		 731 if (sb->level == LEVEL_MULTIPATH)
732 rdev->desc_nr = -1;
733 else
734 rdev->desc_nr = sb->this_disk.number;
735
9a7b2b0f 736 if (!refdev) {
1da177e4 737 ret = 1;
9a7b2b0f 738 } else {
1da177e4
LT		 739 __u64 ev1, ev2;
740 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
741 if (!uuid_equal(refsb, sb)) {
742 printk(KERN_WARNING "md: %s has different UUID to %s\n",
743 b, bdevname(refdev->bdev,b2));
744 goto abort;
745 }
746 if (!sb_equal(refsb, sb)) {
747 printk(KERN_WARNING "md: %s has same UUID"
748 " but different superblock to %s\n",
749 b, bdevname(refdev->bdev, b2));
750 goto abort;
751 }
752 ev1 = md_event(sb);
753 ev2 = md_event(refsb);
754 if (ev1 > ev2)
755 ret = 1;
756 else
757 ret = 0;
758 }
759 rdev->size = calc_dev_size(rdev, sb->chunk_size);
760
2bf071bf
N		 761 if (rdev->size < sb->size && sb->level > 1)
762 /* "this cannot possibly happen" ... */
763 ret = -EINVAL;
764
1da177e4
LT		 765 abort:
766 return ret;
767}
768
769/*
770 * validate_super for 0.90.0
771 */
772static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
773{
774 mdp_disk_t *desc;
775 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 776 __u64 ev1 = md_event(sb);
1da177e4 777
41158c7e 778 rdev->raid_disk = -1;
c5d79adb
N		 779 clear_bit(Faulty, &rdev->flags);
780 clear_bit(In_sync, &rdev->flags);
781 clear_bit(WriteMostly, &rdev->flags);
782 clear_bit(BarriersNotsupp, &rdev->flags);
783
1da177e4
LT		 784 if (mddev->raid_disks == 0) {
785 mddev->major_version = 0;
786 mddev->minor_version = sb->minor_version;
787 mddev->patch_version = sb->patch_version;
e691063a 788 mddev->external = 0;
1da177e4
LT		 789 mddev->chunk_size = sb->chunk_size;
790 mddev->ctime = sb->ctime;
791 mddev->utime = sb->utime;
792 mddev->level = sb->level;
d9d166c2 793 mddev->clevel[0] = 0;
1da177e4
LT		 794 mddev->layout = sb->layout;
795 mddev->raid_disks = sb->raid_disks;
796 mddev->size = sb->size;
07d84d10 797 mddev->events = ev1;
9223214e 798 mddev->bitmap_offset = 0;
36fa3063 799 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 800
f6705578
N		 801 if (mddev->minor_version >= 91) {
802 mddev->reshape_position = sb->reshape_position;
803 mddev->delta_disks = sb->delta_disks;
804 mddev->new_level = sb->new_level;
805 mddev->new_layout = sb->new_layout;
806 mddev->new_chunk = sb->new_chunk;
807 } else {
808 mddev->reshape_position = MaxSector;
809 mddev->delta_disks = 0;
810 mddev->new_level = mddev->level;
811 mddev->new_layout = mddev->layout;
812 mddev->new_chunk = mddev->chunk_size;
813 }
814
1da177e4
LT		 815 if (sb->state & (1<<MD_SB_CLEAN))
816 mddev->recovery_cp = MaxSector;
817 else {
818 if (sb->events_hi == sb->cp_events_hi &&
819 sb->events_lo == sb->cp_events_lo) {
820 mddev->recovery_cp = sb->recovery_cp;
821 } else
822 mddev->recovery_cp = 0;
823 }
824
825 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
826 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
827 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
828 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
829
830 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 831
832 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
e11e93fa 833 mddev->bitmap_file == NULL)
36fa3063 834 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8 835
41158c7e
N		 836 } else if (mddev->pers == NULL) {
837 /* Insist on good event counter while assembling */
1da177e4
LT		 838 ++ev1;
839 if (ev1 < mddev->events)
840 return -EINVAL;
41158c7e
N		 841 } else if (mddev->bitmap) {
842 /* if adding to array with a bitmap, then we can accept an
843 * older device ... but not too old.
844 */
41158c7e
N		 845 if (ev1 < mddev->bitmap->events_cleared)
846 return 0;
07d84d10
N		 847 } else {
848 if (ev1 < mddev->events)
849 /* just a hot-add of a new device, leave raid_disk at -1 */
850 return 0;
851 }
41158c7e 852
1da177e4 853 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 854 desc = sb->disks + rdev->desc_nr;
855
856 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 857 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 858 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
859 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 860 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 861 rdev->raid_disk = desc->raid_disk;
862 }
8ddf9efe
N		 863 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
864 set_bit(WriteMostly, &rdev->flags);
41158c7e 865 } else /* MULTIPATH are always insync */
b2d444d7 866 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 867 return 0;
868}
869
870/*
871 * sync_super for 0.90.0
872 */
873static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
874{
875 mdp_super_t *sb;
876 struct list_head *tmp;
877 mdk_rdev_t *rdev2;
878 int next_spare = mddev->raid_disks;
19133a42 879
1da177e4
LT		 880
881 /* make rdev->sb match mddev data..
882 *
883 * 1/ zero out disks
884 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
885 * 3/ any empty disks < next_spare become removed
886 *
887 * disks[0] gets initialised to REMOVED because
888 * we cannot be sure from other fields if it has
889 * been initialised or not.
890 */
891 int i;
892 int active=0, working=0,failed=0,spare=0,nr_disks=0;
893
61181565
N		 894 rdev->sb_size = MD_SB_BYTES;
895
1da177e4
LT		 896 sb = (mdp_super_t*)page_address(rdev->sb_page);
897
898 memset(sb, 0, sizeof(*sb));
899
900 sb->md_magic = MD_SB_MAGIC;
901 sb->major_version = mddev->major_version;
1da177e4
LT		 902 sb->patch_version = mddev->patch_version;
903 sb->gvalid_words = 0; /* ignored */
904 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
905 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
906 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
907 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
908
909 sb->ctime = mddev->ctime;
910 sb->level = mddev->level;
911 sb->size = mddev->size;
912 sb->raid_disks = mddev->raid_disks;
913 sb->md_minor = mddev->md_minor;
e691063a 914 sb->not_persistent = 0;
1da177e4
LT		 915 sb->utime = mddev->utime;
916 sb->state = 0;
917 sb->events_hi = (mddev->events>>32);
918 sb->events_lo = (u32)mddev->events;
919
f6705578
N		 920 if (mddev->reshape_position == MaxSector)
921 sb->minor_version = 90;
922 else {
923 sb->minor_version = 91;
924 sb->reshape_position = mddev->reshape_position;
925 sb->new_level = mddev->new_level;
926 sb->delta_disks = mddev->delta_disks;
927 sb->new_layout = mddev->new_layout;
928 sb->new_chunk = mddev->new_chunk;
929 }
930 mddev->minor_version = sb->minor_version;
1da177e4
LT		 931 if (mddev->in_sync)
932 {
933 sb->recovery_cp = mddev->recovery_cp;
934 sb->cp_events_hi = (mddev->events>>32);
935 sb->cp_events_lo = (u32)mddev->events;
936 if (mddev->recovery_cp == MaxSector)
937 sb->state = (1<< MD_SB_CLEAN);
938 } else
939 sb->recovery_cp = 0;
940
941 sb->layout = mddev->layout;
942 sb->chunk_size = mddev->chunk_size;
943
a654b9d8
N		 944 if (mddev->bitmap && mddev->bitmap_file == NULL)
945 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
946
1da177e4 947 sb->disks[0].state = (1<<MD_DISK_REMOVED);
d089c6af 948 rdev_for_each(rdev2, tmp, mddev) {
1da177e4 949 mdp_disk_t *d;
86e6ffdd 950 int desc_nr;
b2d444d7
N		 951 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
952 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 953 desc_nr = rdev2->raid_disk;
1da177e4 954 else
86e6ffdd 955 desc_nr = next_spare++;
19133a42 956 rdev2->desc_nr = desc_nr;
1da177e4
LT		 957 d = &sb->disks[rdev2->desc_nr];
958 nr_disks++;
959 d->number = rdev2->desc_nr;
960 d->major = MAJOR(rdev2->bdev->bd_dev);
961 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 962 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
963 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 964 d->raid_disk = rdev2->raid_disk;
965 else
966 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 967 if (test_bit(Faulty, &rdev2->flags))
1da177e4 968 d->state = (1<<MD_DISK_FAULTY);
1be7892f 969 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 970 d->state = (1<<MD_DISK_ACTIVE);
971 d->state |= (1<<MD_DISK_SYNC);
972 active++;
973 working++;
974 } else {
975 d->state = 0;
976 spare++;
977 working++;
978 }
8ddf9efe
N		 979 if (test_bit(WriteMostly, &rdev2->flags))
980 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 981 }
1da177e4
LT		 982 /* now set the "removed" and "faulty" bits on any missing devices */
983 for (i=0 ; i < mddev->raid_disks ; i++) {
984 mdp_disk_t *d = &sb->disks[i];
985 if (d->state == 0 && d->number == 0) {
986 d->number = i;
987 d->raid_disk = i;
988 d->state = (1<<MD_DISK_REMOVED);
989 d->state |= (1<<MD_DISK_FAULTY);
990 failed++;
991 }
992 }
993 sb->nr_disks = nr_disks;
994 sb->active_disks = active;
995 sb->working_disks = working;
996 sb->failed_disks = failed;
997 sb->spare_disks = spare;
998
999 sb->this_disk = sb->disks[rdev->desc_nr];
1000 sb->sb_csum = calc_sb_csum(sb);
1001}
1002
1003/*
1004 * version 1 superblock
1005 */
1006
1c05b4bc 1007static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1008{
1c05b4bc
N		 1009 __le32 disk_csum;
1010 u32 csum;
1da177e4
LT		 1011 unsigned long long newcsum;
1012 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1013 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1014 int i;
1015
1016 disk_csum = sb->sb_csum;
1017 sb->sb_csum = 0;
1018 newcsum = 0;
1019 for (i=0; size>=4; size -= 4 )
1020 newcsum += le32_to_cpu(*isuper++);
1021
1022 if (size == 2)
1c05b4bc 1023 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1024
1025 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1026 sb->sb_csum = disk_csum;
1027 return cpu_to_le32(csum);
1028}
1029
1030static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1031{
1032 struct mdp_superblock_1 *sb;
1033 int ret;
1034 sector_t sb_offset;
1035 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1036 int bmask;
1da177e4
LT		 1037
1038 /*
1039 * Calculate the position of the superblock.
1040 * It is always aligned to a 4K boundary and
1041 * depeding on minor_version, it can be:
1042 * 0: At least 8K, but less than 12K, from end of device
1043 * 1: At start of device
1044 * 2: 4K from start of device.
1045 */
1046 switch(minor_version) {
1047 case 0:
1048 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1049 sb_offset -= 8*2;
39730960 1050 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 1051 /* convert from sectors to K */
1052 sb_offset /= 2;
1053 break;
1054 case 1:
1055 sb_offset = 0;
1056 break;
1057 case 2:
1058 sb_offset = 4;
1059 break;
1060 default:
1061 return -EINVAL;
1062 }
1063 rdev->sb_offset = sb_offset;
1064
0002b271
N		 1065 /* superblock is rarely larger than 1K, but it can be larger,
1066 * and it is safe to read 4k, so we do that
1067 */
1068 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1069 if (ret) return ret;
1070
1071
1072 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1073
1074 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1075 sb->major_version != cpu_to_le32(1) ||
1076 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1077 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1078 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1079 return -EINVAL;
1080
1081 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1082 printk("md: invalid superblock checksum on %s\n",
1083 bdevname(rdev->bdev,b));
1084 return -EINVAL;
1085 }
1086 if (le64_to_cpu(sb->data_size) < 10) {
1087 printk("md: data_size too small on %s\n",
1088 bdevname(rdev->bdev,b));
1089 return -EINVAL;
1090 }
e11e93fa
N		 1091 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1092 if (sb->level != cpu_to_le32(1) &&
1093 sb->level != cpu_to_le32(4) &&
1094 sb->level != cpu_to_le32(5) &&
1095 sb->level != cpu_to_le32(6) &&
1096 sb->level != cpu_to_le32(10)) {
1097 printk(KERN_WARNING
1098 "md: bitmaps not supported for this level.\n");
1099 return -EINVAL;
1100 }
1101 }
1102
1da177e4
LT		 1103 rdev->preferred_minor = 0xffff;
1104 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1105 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1106
0002b271 1107 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1108 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1109 if (rdev->sb_size & bmask)
a1801f85
N		 1110 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1111
1112 if (minor_version
1113 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1114 return -EINVAL;
0002b271 1115
31b65a0d
N		 1116 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1117 rdev->desc_nr = -1;
1118 else
1119 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1120
9a7b2b0f 1121 if (!refdev) {
8ed75463 1122 ret = 1;
9a7b2b0f 1123 } else {
1da177e4
LT		 1124 __u64 ev1, ev2;
1125 struct mdp_superblock_1 *refsb =
1126 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1127
1128 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1129 sb->level != refsb->level ||
1130 sb->layout != refsb->layout ||
1131 sb->chunksize != refsb->chunksize) {
1132 printk(KERN_WARNING "md: %s has strangely different"
1133 " superblock to %s\n",
1134 bdevname(rdev->bdev,b),
1135 bdevname(refdev->bdev,b2));
1136 return -EINVAL;
1137 }
1138 ev1 = le64_to_cpu(sb->events);
1139 ev2 = le64_to_cpu(refsb->events);
1140
1141 if (ev1 > ev2)
8ed75463
N		 1142 ret = 1;
1143 else
1144 ret = 0;
1da177e4 1145 }
a1801f85 1146 if (minor_version)
1da177e4
LT		 1147 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1148 else
1149 rdev->size = rdev->sb_offset;
1150 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1151 return -EINVAL;
1152 rdev->size = le64_to_cpu(sb->data_size)/2;
1153 if (le32_to_cpu(sb->chunksize))
1154 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf 1155
1c05b4bc 1156 if (le64_to_cpu(sb->size) > rdev->size*2)
2bf071bf 1157 return -EINVAL;
8ed75463 1158 return ret;
1da177e4
LT		 1159}
1160
1161static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1162{
1163 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1164 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1165
41158c7e 1166 rdev->raid_disk = -1;
c5d79adb
N		 1167 clear_bit(Faulty, &rdev->flags);
1168 clear_bit(In_sync, &rdev->flags);
1169 clear_bit(WriteMostly, &rdev->flags);
1170 clear_bit(BarriersNotsupp, &rdev->flags);
1171
1da177e4
LT		 1172 if (mddev->raid_disks == 0) {
1173 mddev->major_version = 1;
1174 mddev->patch_version = 0;
e691063a 1175 mddev->external = 0;
1da177e4
LT		 1176 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1177 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1178 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1179 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1180 mddev->clevel[0] = 0;
1da177e4
LT		 1181 mddev->layout = le32_to_cpu(sb->layout);
1182 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1183 mddev->size = le64_to_cpu(sb->size)/2;
07d84d10 1184 mddev->events = ev1;
9223214e 1185 mddev->bitmap_offset = 0;
29fc7e3e 1186 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1187
1188 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1189 memcpy(mddev->uuid, sb->set_uuid, 16);
1190
1191 mddev->max_disks = (4096-256)/2;
a654b9d8 1192
71c0805c 1193 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
e11e93fa 1194 mddev->bitmap_file == NULL )
a654b9d8 1195 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1196
f6705578
N		 1197 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1198 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1199 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1200 mddev->new_level = le32_to_cpu(sb->new_level);
1201 mddev->new_layout = le32_to_cpu(sb->new_layout);
1202 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1203 } else {
1204 mddev->reshape_position = MaxSector;
1205 mddev->delta_disks = 0;
1206 mddev->new_level = mddev->level;
1207 mddev->new_layout = mddev->layout;
1208 mddev->new_chunk = mddev->chunk_size;
1209 }
1210
41158c7e
N		 1211 } else if (mddev->pers == NULL) {
1212 /* Insist of good event counter while assembling */
1da177e4
LT		 1213 ++ev1;
1214 if (ev1 < mddev->events)
1215 return -EINVAL;
41158c7e
N		 1216 } else if (mddev->bitmap) {
1217 /* If adding to array with a bitmap, then we can accept an
1218 * older device, but not too old.
1219 */
41158c7e
N		 1220 if (ev1 < mddev->bitmap->events_cleared)
1221 return 0;
07d84d10
N		 1222 } else {
1223 if (ev1 < mddev->events)
1224 /* just a hot-add of a new device, leave raid_disk at -1 */
1225 return 0;
1226 }
1da177e4
LT		 1227 if (mddev->level != LEVEL_MULTIPATH) {
1228 int role;
1da177e4
LT		 1229 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1230 switch(role) {
1231 case 0xffff: /* spare */
1da177e4
LT		 1232 break;
1233 case 0xfffe: /* faulty */
b2d444d7 1234 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1235 break;
1236 default:
5fd6c1dc
N		 1237 if ((le32_to_cpu(sb->feature_map) &
1238 MD_FEATURE_RECOVERY_OFFSET))
1239 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1240 else
1241 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1242 rdev->raid_disk = role;
1243 break;
1244 }
8ddf9efe
N		 1245 if (sb->devflags & WriteMostly1)
1246 set_bit(WriteMostly, &rdev->flags);
41158c7e 1247 } else /* MULTIPATH are always insync */
b2d444d7 1248 set_bit(In_sync, &rdev->flags);
41158c7e 1249
1da177e4
LT		 1250 return 0;
1251}
1252
1253static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1254{
1255 struct mdp_superblock_1 *sb;
1256 struct list_head *tmp;
1257 mdk_rdev_t *rdev2;
1258 int max_dev, i;
1259 /* make rdev->sb match mddev and rdev data. */
1260
1261 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1262
1263 sb->feature_map = 0;
1264 sb->pad0 = 0;
5fd6c1dc 1265 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1266 memset(sb->pad1, 0, sizeof(sb->pad1));
1267 memset(sb->pad2, 0, sizeof(sb->pad2));
1268 memset(sb->pad3, 0, sizeof(sb->pad3));
1269
1270 sb->utime = cpu_to_le64((__u64)mddev->utime);
1271 sb->events = cpu_to_le64(mddev->events);
1272 if (mddev->in_sync)
1273 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1274 else
1275 sb->resync_offset = cpu_to_le64(0);
1276
1c05b4bc 1277 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1278
f0ca340c 1279 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1280 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1281
a654b9d8
N		 1282 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1283 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1284 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1285 }
5fd6c1dc
N		 1286
1287 if (rdev->raid_disk >= 0 &&
1288 !test_bit(In_sync, &rdev->flags) &&
1289 rdev->recovery_offset > 0) {
1290 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1291 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1292 }
1293
f6705578
N		 1294 if (mddev->reshape_position != MaxSector) {
1295 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1296 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1297 sb->new_layout = cpu_to_le32(mddev->new_layout);
1298 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1299 sb->new_level = cpu_to_le32(mddev->new_level);
1300 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1301 }
a654b9d8 1302
1da177e4 1303 max_dev = 0;
d089c6af 1304 rdev_for_each(rdev2, tmp, mddev)
1da177e4
LT		 1305 if (rdev2->desc_nr+1 > max_dev)
1306 max_dev = rdev2->desc_nr+1;
a778b73f
N		 1307
1308 if (max_dev > le32_to_cpu(sb->max_dev))
1309 sb->max_dev = cpu_to_le32(max_dev);
1da177e4
LT		 1310 for (i=0; i<max_dev;i++)
1311 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1312
d089c6af 1313 rdev_for_each(rdev2, tmp, mddev) {
1da177e4 1314 i = rdev2->desc_nr;
b2d444d7 1315 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1316 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1317 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1318 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1319 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1320 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1321 else
1322 sb->dev_roles[i] = cpu_to_le16(0xffff);
1323 }
1324
1da177e4
LT		 1325 sb->sb_csum = calc_sb_1_csum(sb);
1326}
1327
1328
75c96f85 1329static struct super_type super_types[] = {
1da177e4
LT		 1330 [0] = {
1331 .name = "0.90.0",
1332 .owner = THIS_MODULE,
1333 .load_super = super_90_load,
1334 .validate_super = super_90_validate,
1335 .sync_super = super_90_sync,
1336 },
1337 [1] = {
1338 .name = "md-1",
1339 .owner = THIS_MODULE,
1340 .load_super = super_1_load,
1341 .validate_super = super_1_validate,
1342 .sync_super = super_1_sync,
1343 },
1344};
1da177e4
LT		 1345
1346static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1347{
7dd5e7c3
N		 1348 struct list_head *tmp, *tmp2;
1349 mdk_rdev_t *rdev, *rdev2;
1da177e4 1350
d089c6af
N		 1351 rdev_for_each(rdev, tmp, mddev1)
1352 rdev_for_each(rdev2, tmp2, mddev2)
7dd5e7c3
N		 1353 if (rdev->bdev->bd_contains ==
1354 rdev2->bdev->bd_contains)
1355 return 1;
1da177e4
LT		 1356
1357 return 0;
1358}
1359
1360static LIST_HEAD(pending_raid_disks);
1361
1362static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1363{
7dd5e7c3 1364 char b[BDEVNAME_SIZE];
f637b9f9 1365 struct kobject *ko;
1edf80d3 1366 char *s;
5e55e2f5 1367 int err;
1da177e4
LT		 1368
1369 if (rdev->mddev) {
1370 MD_BUG();
1371 return -EINVAL;
1372 }
2bf071bf
N		 1373 /* make sure rdev->size exceeds mddev->size */
1374 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
a778b73f
N		 1375 if (mddev->pers) {
1376 /* Cannot change size, so fail
1377 * If mddev->level <= 0, then we don't care
1378 * about aligning sizes (e.g. linear)
1379 */
1380 if (mddev->level > 0)
1381 return -ENOSPC;
1382 } else
2bf071bf
N		 1383 mddev->size = rdev->size;
1384 }
1da177e4
LT		 1385
1386 /* Verify rdev->desc_nr is unique.
1387 * If it is -1, assign a free number, else
1388 * check number is not in use
1389 */
1390 if (rdev->desc_nr < 0) {
1391 int choice = 0;
1392 if (mddev->pers) choice = mddev->raid_disks;
1393 while (find_rdev_nr(mddev, choice))
1394 choice++;
1395 rdev->desc_nr = choice;
1396 } else {
1397 if (find_rdev_nr(mddev, rdev->desc_nr))
1398 return -EBUSY;
1399 }
19133a42 1400 bdevname(rdev->bdev,b);
649316b2 1401 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1402 *s = '!';
649316b2 1403
1da177e4 1404 rdev->mddev = mddev;
19133a42 1405 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1406
b2d6db58 1407 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1408 goto fail;
86e6ffdd 1409
f637b9f9 1410 if (rdev->bdev->bd_part)
edfaa7c3 1411 ko = &rdev->bdev->bd_part->dev.kobj;
f637b9f9 1412 else
edfaa7c3 1413 ko = &rdev->bdev->bd_disk->dev.kobj;
5e55e2f5
N		 1414 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1415 kobject_del(&rdev->kobj);
1416 goto fail;
1417 }
1418 list_add(&rdev->same_set, &mddev->disks);
c5d79adb 1419 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1da177e4 1420 return 0;
5e55e2f5
N		 1421
1422 fail:
1423 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1424 b, mdname(mddev));
1425 return err;
1da177e4
LT		 1426}
1427
177a99b2 1428static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1429{
1430 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1431 kobject_del(&rdev->kobj);
177a99b2 1432 kobject_put(&rdev->kobj);
5792a285
N		 1433}
1434
1da177e4
LT		 1435static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1436{
1437 char b[BDEVNAME_SIZE];
1438 if (!rdev->mddev) {
1439 MD_BUG();
1440 return;
1441 }
5463c790 1442 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1443 list_del_init(&rdev->same_set);
1444 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1445 rdev->mddev = NULL;
86e6ffdd 1446 sysfs_remove_link(&rdev->kobj, "block");
5792a285
N		 1447
1448 /* We need to delay this, otherwise we can deadlock when
1449 * writing to 'remove' to "dev/state"
1450 */
177a99b2
N		 1451 INIT_WORK(&rdev->del_work, md_delayed_delete);
1452 kobject_get(&rdev->kobj);
5792a285 1453 schedule_work(&rdev->del_work);
1da177e4
LT		 1454}
1455
1456/*
1457 * prevent the device from being mounted, repartitioned or
1458 * otherwise reused by a RAID array (or any other kernel
1459 * subsystem), by bd_claiming the device.
1460 */
c5d79adb 1461static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1462{
1463 int err = 0;
1464 struct block_device *bdev;
1465 char b[BDEVNAME_SIZE];
1466
2e7b651d 1467 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1468 if (IS_ERR(bdev)) {
1469 printk(KERN_ERR "md: could not open %s.\n",
1470 __bdevname(dev, b));
1471 return PTR_ERR(bdev);
1472 }
c5d79adb 1473 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1474 if (err) {
1475 printk(KERN_ERR "md: could not bd_claim %s.\n",
1476 bdevname(bdev, b));
2e7b651d 1477 blkdev_put(bdev);
1da177e4
LT		 1478 return err;
1479 }
c5d79adb
N		 1480 if (!shared)
1481 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1482 rdev->bdev = bdev;
1483 return err;
1484}
1485
1486static void unlock_rdev(mdk_rdev_t *rdev)
1487{
1488 struct block_device *bdev = rdev->bdev;
1489 rdev->bdev = NULL;
1490 if (!bdev)
1491 MD_BUG();
1492 bd_release(bdev);
2e7b651d 1493 blkdev_put(bdev);
1da177e4
LT		 1494}
1495
1496void md_autodetect_dev(dev_t dev);
1497
1498static void export_rdev(mdk_rdev_t * rdev)
1499{
1500 char b[BDEVNAME_SIZE];
1501 printk(KERN_INFO "md: export_rdev(%s)\n",
1502 bdevname(rdev->bdev,b));
1503 if (rdev->mddev)
1504 MD_BUG();
1505 free_disk_sb(rdev);
1506 list_del_init(&rdev->same_set);
1507#ifndef MODULE
d0fae18f
N		 1508 if (test_bit(AutoDetected, &rdev->flags))
1509 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1510#endif
1511 unlock_rdev(rdev);
86e6ffdd 1512 kobject_put(&rdev->kobj);
1da177e4
LT		 1513}
1514
1515static void kick_rdev_from_array(mdk_rdev_t * rdev)
1516{
1517 unbind_rdev_from_array(rdev);
1518 export_rdev(rdev);
1519}
1520
1521static void export_array(mddev_t *mddev)
1522{
1523 struct list_head *tmp;
1524 mdk_rdev_t *rdev;
1525
d089c6af 1526 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1527 if (!rdev->mddev) {
1528 MD_BUG();
1529 continue;
1530 }
1531 kick_rdev_from_array(rdev);
1532 }
1533 if (!list_empty(&mddev->disks))
1534 MD_BUG();
1535 mddev->raid_disks = 0;
1536 mddev->major_version = 0;
1537}
1538
1539static void print_desc(mdp_disk_t *desc)
1540{
1541 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1542 desc->major,desc->minor,desc->raid_disk,desc->state);
1543}
1544
1545static void print_sb(mdp_super_t *sb)
1546{
1547 int i;
1548
1549 printk(KERN_INFO
1550 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1551 sb->major_version, sb->minor_version, sb->patch_version,
1552 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1553 sb->ctime);
1554 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1555 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1556 sb->md_minor, sb->layout, sb->chunk_size);
1557 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1558 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1559 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1560 sb->failed_disks, sb->spare_disks,
1561 sb->sb_csum, (unsigned long)sb->events_lo);
1562
1563 printk(KERN_INFO);
1564 for (i = 0; i < MD_SB_DISKS; i++) {
1565 mdp_disk_t *desc;
1566
1567 desc = sb->disks + i;
1568 if (desc->number || desc->major || desc->minor ||
1569 desc->raid_disk || (desc->state && (desc->state != 4))) {
1570 printk(" D %2d: ", i);
1571 print_desc(desc);
1572 }
1573 }
1574 printk(KERN_INFO "md: THIS: ");
1575 print_desc(&sb->this_disk);
1576
1577}
1578
1579static void print_rdev(mdk_rdev_t *rdev)
1580{
1581 char b[BDEVNAME_SIZE];
1582 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1583 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1584 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1585 rdev->desc_nr);
1da177e4
LT		 1586 if (rdev->sb_loaded) {
1587 printk(KERN_INFO "md: rdev superblock:\n");
1588 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1589 } else
1590 printk(KERN_INFO "md: no rdev superblock!\n");
1591}
1592
5e56341d 1593static void md_print_devices(void)
1da177e4
LT		 1594{
1595 struct list_head *tmp, *tmp2;
1596 mdk_rdev_t *rdev;
1597 mddev_t *mddev;
1598 char b[BDEVNAME_SIZE];
1599
1600 printk("\n");
1601 printk("md: **********************************\n");
1602 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1603 printk("md: **********************************\n");
29ac4aa3 1604 for_each_mddev(mddev, tmp) {
1da177e4 1605
32a7627c
N		 1606 if (mddev->bitmap)
1607 bitmap_print_sb(mddev->bitmap);
1608 else
1609 printk("%s: ", mdname(mddev));
d089c6af 1610 rdev_for_each(rdev, tmp2, mddev)
1da177e4
LT		 1611 printk("<%s>", bdevname(rdev->bdev,b));
1612 printk("\n");
1613
d089c6af 1614 rdev_for_each(rdev, tmp2, mddev)
1da177e4
LT		 1615 print_rdev(rdev);
1616 }
1617 printk("md: **********************************\n");
1618 printk("\n");
1619}
1620
1621
42543769 1622static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1623{
42543769
N		 1624 /* Update each superblock (in-memory image), but
1625 * if we are allowed to, skip spares which already
1626 * have the right event counter, or have one earlier
1627 * (which would mean they aren't being marked as dirty
1628 * with the rest of the array)
1629 */
1da177e4
LT		 1630 mdk_rdev_t *rdev;
1631 struct list_head *tmp;
1632
d089c6af 1633 rdev_for_each(rdev, tmp, mddev) {
42543769
N		 1634 if (rdev->sb_events == mddev->events ||
1635 (nospares &&
1636 rdev->raid_disk < 0 &&
1637 (rdev->sb_events&1)==0 &&
1638 rdev->sb_events+1 == mddev->events)) {
1639 /* Don't update this superblock */
1640 rdev->sb_loaded = 2;
1641 } else {
1642 super_types[mddev->major_version].
1643 sync_super(mddev, rdev);
1644 rdev->sb_loaded = 1;
1645 }
1da177e4
LT		 1646 }
1647}
1648
850b2b42 1649static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 1650{
1da177e4
LT		 1651 struct list_head *tmp;
1652 mdk_rdev_t *rdev;
06d91a5f 1653 int sync_req;
42543769 1654 int nospares = 0;
1da177e4 1655
8377bc80
N		 1656 if (mddev->external)
1657 return;
1da177e4 1658repeat:
a9701a30 1659 spin_lock_irq(&mddev->write_lock);
84692195 1660
850b2b42
N		 1661 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1662 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1663 force_change = 1;
1664 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1665 /* just a clean<-> dirty transition, possibly leave spares alone,
1666 * though if events isn't the right even/odd, we will have to do
1667 * spares after all
1668 */
1669 nospares = 1;
1670 if (force_change)
1671 nospares = 0;
1672 if (mddev->degraded)
84692195
N		 1673 /* If the array is degraded, then skipping spares is both
1674 * dangerous and fairly pointless.
1675 * Dangerous because a device that was removed from the array
1676 * might have a event_count that still looks up-to-date,
1677 * so it can be re-added without a resync.
1678 * Pointless because if there are any spares to skip,
1679 * then a recovery will happen and soon that array won't
1680 * be degraded any more and the spare can go back to sleep then.
1681 */
850b2b42 1682 nospares = 0;
84692195 1683
06d91a5f 1684 sync_req = mddev->in_sync;
1da177e4 1685 mddev->utime = get_seconds();
42543769
N		 1686
1687 /* If this is just a dirty<->clean transition, and the array is clean
1688 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1689 if (nospares
42543769 1690 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1031be7a
N		 1691 && (mddev->events & 1)
1692 && mddev->events != 1)
42543769
N		 1693 mddev->events--;
1694 else {
1695 /* otherwise we have to go forward and ... */
1696 mddev->events ++;
1697 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1698 /* .. if the array isn't clean, insist on an odd 'events' */
1699 if ((mddev->events&1)==0) {
1700 mddev->events++;
1701 nospares = 0;
1702 }
1703 } else {
1704 /* otherwise insist on an even 'events' (for clean states) */
1705 if ((mddev->events&1)) {
1706 mddev->events++;
1707 nospares = 0;
1708 }
1709 }
1710 }
1da177e4
LT		 1711
1712 if (!mddev->events) {
1713 /*
1714 * oops, this 64-bit counter should never wrap.
1715 * Either we are in around ~1 trillion A.C., assuming
1716 * 1 reboot per second, or we have a bug:
1717 */
1718 MD_BUG();
1719 mddev->events --;
1720 }
1da177e4
LT		 1721
1722 /*
1723 * do not write anything to disk if using
1724 * nonpersistent superblocks
1725 */
06d91a5f 1726 if (!mddev->persistent) {
e691063a
N		 1727 if (!mddev->external)
1728 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1729
a9701a30 1730 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1731 wake_up(&mddev->sb_wait);
1da177e4 1732 return;
06d91a5f 1733 }
e691063a 1734 sync_sbs(mddev, nospares);
a9701a30 1735 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1736
1737 dprintk(KERN_INFO
1738 "md: updating %s RAID superblock on device (in sync %d)\n",
1739 mdname(mddev),mddev->in_sync);
1740
4ad13663 1741 bitmap_update_sb(mddev->bitmap);
d089c6af 1742 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1743 char b[BDEVNAME_SIZE];
1744 dprintk(KERN_INFO "md: ");
42543769
N		 1745 if (rdev->sb_loaded != 1)
1746 continue; /* no noise on spare devices */
b2d444d7 1747 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1748 dprintk("(skipping faulty ");
1749
1750 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1751 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1752 md_super_write(mddev,rdev,
0002b271 1753 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1754 rdev->sb_page);
1755 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1756 bdevname(rdev->bdev,b),
1757 (unsigned long long)rdev->sb_offset);
42543769 1758 rdev->sb_events = mddev->events;
7bfa19f2 1759
1da177e4
LT		 1760 } else
1761 dprintk(")\n");
7bfa19f2 1762 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1763 /* only need to write one superblock... */
1764 break;
1765 }
a9701a30 1766 md_super_wait(mddev);
850b2b42 1767 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 1768
a9701a30 1769 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 1770 if (mddev->in_sync != sync_req ||
1771 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 1772 /* have to write it out again */
a9701a30 1773 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1774 goto repeat;
1775 }
850b2b42 1776 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1777 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1778 wake_up(&mddev->sb_wait);
06d91a5f 1779
1da177e4
LT		 1780}
1781
bce74dac
N		 1782/* words written to sysfs files may, or my not, be \n terminated.
1783 * We want to accept with case. For this we use cmd_match.
1784 */
1785static int cmd_match(const char *cmd, const char *str)
1786{
1787 /* See if cmd, written into a sysfs file, matches
1788 * str. They must either be the same, or cmd can
1789 * have a trailing newline
1790 */
1791 while (*cmd && *str && *cmd == *str) {
1792 cmd++;
1793 str++;
1794 }
1795 if (*cmd == '\n')
1796 cmd++;
1797 if (*str || *cmd)
1798 return 0;
1799 return 1;
1800}
1801
86e6ffdd
N		 1802struct rdev_sysfs_entry {
1803 struct attribute attr;
1804 ssize_t (*show)(mdk_rdev_t *, char *);
1805 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1806};
1807
1808static ssize_t
96de1e66 1809state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1810{
1811 char *sep = "";
20a49ff6 1812 size_t len = 0;
86e6ffdd 1813
b2d444d7 1814 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1815 len+= sprintf(page+len, "%sfaulty",sep);
1816 sep = ",";
1817 }
b2d444d7 1818 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1819 len += sprintf(page+len, "%sin_sync",sep);
1820 sep = ",";
1821 }
f655675b
N		 1822 if (test_bit(WriteMostly, &rdev->flags)) {
1823 len += sprintf(page+len, "%swrite_mostly",sep);
1824 sep = ",";
1825 }
b2d444d7
N		 1826 if (!test_bit(Faulty, &rdev->flags) &&
1827 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1828 len += sprintf(page+len, "%sspare", sep);
1829 sep = ",";
1830 }
1831 return len+sprintf(page+len, "\n");
1832}
1833
45dc2de1
N		 1834static ssize_t
1835state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1836{
1837 /* can write
1838 * faulty - simulates and error
1839 * remove - disconnects the device
f655675b
N		 1840 * writemostly - sets write_mostly
1841 * -writemostly - clears write_mostly
45dc2de1
N		 1842 */
1843 int err = -EINVAL;
1844 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1845 md_error(rdev->mddev, rdev);
1846 err = 0;
1847 } else if (cmd_match(buf, "remove")) {
1848 if (rdev->raid_disk >= 0)
1849 err = -EBUSY;
1850 else {
1851 mddev_t *mddev = rdev->mddev;
1852 kick_rdev_from_array(rdev);
3f9d7b0d
N		 1853 if (mddev->pers)
1854 md_update_sb(mddev, 1);
45dc2de1
N		 1855 md_new_event(mddev);
1856 err = 0;
1857 }
f655675b
N		 1858 } else if (cmd_match(buf, "writemostly")) {
1859 set_bit(WriteMostly, &rdev->flags);
1860 err = 0;
1861 } else if (cmd_match(buf, "-writemostly")) {
1862 clear_bit(WriteMostly, &rdev->flags);
1863 err = 0;
45dc2de1
N		 1864 }
1865 return err ? err : len;
1866}
80ca3a44
N		 1867static struct rdev_sysfs_entry rdev_state =
1868__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 1869
4dbcdc75
N		 1870static ssize_t
1871errors_show(mdk_rdev_t *rdev, char *page)
1872{
1873 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1874}
1875
1876static ssize_t
1877errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1878{
1879 char *e;
1880 unsigned long n = simple_strtoul(buf, &e, 10);
1881 if (*buf && (*e == 0 || *e == '\n')) {
1882 atomic_set(&rdev->corrected_errors, n);
1883 return len;
1884 }
1885 return -EINVAL;
1886}
1887static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1888__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1889
014236d2
N		 1890static ssize_t
1891slot_show(mdk_rdev_t *rdev, char *page)
1892{
1893 if (rdev->raid_disk < 0)
1894 return sprintf(page, "none\n");
1895 else
1896 return sprintf(page, "%d\n", rdev->raid_disk);
1897}
1898
1899static ssize_t
1900slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1901{
1902 char *e;
c303da6d
N		 1903 int err;
1904 char nm[20];
014236d2
N		 1905 int slot = simple_strtoul(buf, &e, 10);
1906 if (strncmp(buf, "none", 4)==0)
1907 slot = -1;
1908 else if (e==buf || (*e && *e!= '\n'))
1909 return -EINVAL;
c303da6d
N		 1910 if (rdev->mddev->pers) {
1911 /* Setting 'slot' on an active array requires also
1912 * updating the 'rd%d' link, and communicating
1913 * with the personality with ->hot_*_disk.
1914 * For now we only support removing
1915 * failed/spare devices. This normally happens automatically,
1916 * but not when the metadata is externally managed.
1917 */
1918 if (slot != -1)
1919 return -EBUSY;
1920 if (rdev->raid_disk == -1)
1921 return -EEXIST;
1922 /* personality does all needed checks */
1923 if (rdev->mddev->pers->hot_add_disk == NULL)
1924 return -EINVAL;
1925 err = rdev->mddev->pers->
1926 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1927 if (err)
1928 return err;
1929 sprintf(nm, "rd%d", rdev->raid_disk);
1930 sysfs_remove_link(&rdev->mddev->kobj, nm);
1931 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1932 md_wakeup_thread(rdev->mddev->thread);
1933 } else {
1934 if (slot >= rdev->mddev->raid_disks)
1935 return -ENOSPC;
1936 rdev->raid_disk = slot;
1937 /* assume it is working */
c5d79adb
N		 1938 clear_bit(Faulty, &rdev->flags);
1939 clear_bit(WriteMostly, &rdev->flags);
c303da6d
N		 1940 set_bit(In_sync, &rdev->flags);
1941 }
014236d2
N		 1942 return len;
1943}
1944
1945
1946static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1947__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1948
93c8cad0
N		 1949static ssize_t
1950offset_show(mdk_rdev_t *rdev, char *page)
1951{
6961ece4 1952 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1953}
1954
1955static ssize_t
1956offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1957{
1958 char *e;
1959 unsigned long long offset = simple_strtoull(buf, &e, 10);
1960 if (e==buf || (*e && *e != '\n'))
1961 return -EINVAL;
1962 if (rdev->mddev->pers)
1963 return -EBUSY;
c5d79adb
N		 1964 if (rdev->size && rdev->mddev->external)
1965 /* Must set offset before size, so overlap checks
1966 * can be sane */
1967 return -EBUSY;
93c8cad0
N		 1968 rdev->data_offset = offset;
1969 return len;
1970}
1971
1972static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1973__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1974
83303b61
N		 1975static ssize_t
1976rdev_size_show(mdk_rdev_t *rdev, char *page)
1977{
1978 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1979}
1980
c5d79adb
N		 1981static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1982{
1983 /* check if two start/length pairs overlap */
1984 if (s1+l1 <= s2)
1985 return 0;
1986 if (s2+l2 <= s1)
1987 return 0;
1988 return 1;
1989}
1990
83303b61
N		 1991static ssize_t
1992rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1993{
1994 char *e;
1995 unsigned long long size = simple_strtoull(buf, &e, 10);
c5d79adb 1996 unsigned long long oldsize = rdev->size;
27c529bb
N		 1997 mddev_t *my_mddev = rdev->mddev;
1998
83303b61
N		 1999 if (e==buf || (*e && *e != '\n'))
2000 return -EINVAL;
27c529bb 2001 if (my_mddev->pers)
83303b61
N		 2002 return -EBUSY;
2003 rdev->size = size;
c5d79adb
N		 2004 if (size > oldsize && rdev->mddev->external) {
2005 /* need to check that all other rdevs with the same ->bdev
2006 * do not overlap. We need to unlock the mddev to avoid
2007 * a deadlock. We have already changed rdev->size, and if
2008 * we have to change it back, we will have the lock again.
2009 */
2010 mddev_t *mddev;
2011 int overlap = 0;
2012 struct list_head *tmp, *tmp2;
2013
27c529bb 2014 mddev_unlock(my_mddev);
29ac4aa3 2015 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2016 mdk_rdev_t *rdev2;
2017
2018 mddev_lock(mddev);
d089c6af 2019 rdev_for_each(rdev2, tmp2, mddev)
c5d79adb
N		 2020 if (test_bit(AllReserved, &rdev2->flags) ||
2021 (rdev->bdev == rdev2->bdev &&
2022 rdev != rdev2 &&
2023 overlaps(rdev->data_offset, rdev->size,
2024 rdev2->data_offset, rdev2->size))) {
2025 overlap = 1;
2026 break;
2027 }
2028 mddev_unlock(mddev);
2029 if (overlap) {
2030 mddev_put(mddev);
2031 break;
2032 }
2033 }
27c529bb 2034 mddev_lock(my_mddev);
c5d79adb
N		 2035 if (overlap) {
2036 /* Someone else could have slipped in a size
2037 * change here, but doing so is just silly.
2038 * We put oldsize back because we *know* it is
2039 * safe, and trust userspace not to race with
2040 * itself
2041 */
2042 rdev->size = oldsize;
2043 return -EBUSY;
2044 }
2045 }
27c529bb
N		 2046 if (size < my_mddev->size || my_mddev->size == 0)
2047 my_mddev->size = size;
83303b61
N		 2048 return len;
2049}
2050
2051static struct rdev_sysfs_entry rdev_size =
80ca3a44 2052__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2053
86e6ffdd
N		 2054static struct attribute *rdev_default_attrs[] = {
2055 &rdev_state.attr,
4dbcdc75 2056 &rdev_errors.attr,
014236d2 2057 &rdev_slot.attr,
93c8cad0 2058 &rdev_offset.attr,
83303b61 2059 &rdev_size.attr,
86e6ffdd
N		 2060 NULL,
2061};
2062static ssize_t
2063rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2064{
2065 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2066 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2067 mddev_t *mddev = rdev->mddev;
2068 ssize_t rv;
86e6ffdd
N		 2069
2070 if (!entry->show)
2071 return -EIO;
27c529bb
N		 2072
2073 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2074 if (!rv) {
2075 if (rdev->mddev == NULL)
2076 rv = -EBUSY;
2077 else
2078 rv = entry->show(rdev, page);
2079 mddev_unlock(mddev);
2080 }
2081 return rv;
86e6ffdd
N		 2082}
2083
2084static ssize_t
2085rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2086 const char *page, size_t length)
2087{
2088 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2089 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2090 ssize_t rv;
2091 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2092
2093 if (!entry->store)
2094 return -EIO;
67463acb
N		 2095 if (!capable(CAP_SYS_ADMIN))
2096 return -EACCES;
27c529bb 2097 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2098 if (!rv) {
27c529bb
N		 2099 if (rdev->mddev == NULL)
2100 rv = -EBUSY;
2101 else
2102 rv = entry->store(rdev, page, length);
6a51830e 2103 mddev_unlock(mddev);
ca388059
N		 2104 }
2105 return rv;
86e6ffdd
N		 2106}
2107
2108static void rdev_free(struct kobject *ko)
2109{
2110 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2111 kfree(rdev);
2112}
2113static struct sysfs_ops rdev_sysfs_ops = {
2114 .show = rdev_attr_show,
2115 .store = rdev_attr_store,
2116};
2117static struct kobj_type rdev_ktype = {
2118 .release = rdev_free,
2119 .sysfs_ops = &rdev_sysfs_ops,
2120 .default_attrs = rdev_default_attrs,
2121};
2122
1da177e4
LT		 2123/*
2124 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2125 *
2126 * mark the device faulty if:
2127 *
2128 * - the device is nonexistent (zero size)
2129 * - the device has no valid superblock
2130 *
2131 * a faulty rdev _never_ has rdev->sb set.
2132 */
2133static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2134{
2135 char b[BDEVNAME_SIZE];
2136 int err;
2137 mdk_rdev_t *rdev;
2138 sector_t size;
2139
9ffae0cf 2140 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2141 if (!rdev) {
2142 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2143 return ERR_PTR(-ENOMEM);
2144 }
1da177e4
LT		 2145
2146 if ((err = alloc_disk_sb(rdev)))
2147 goto abort_free;
2148
c5d79adb 2149 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2150 if (err)
2151 goto abort_free;
2152
f9cb074b 2153 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2154
1da177e4 2155 rdev->desc_nr = -1;
2b6e8459 2156 rdev->saved_raid_disk = -1;
3f9d7b0d 2157 rdev->raid_disk = -1;
b2d444d7 2158 rdev->flags = 0;
1da177e4 2159 rdev->data_offset = 0;
42543769 2160 rdev->sb_events = 0;
1da177e4 2161 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2162 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2163 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2164
2165 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2166 if (!size) {
2167 printk(KERN_WARNING
2168 "md: %s has zero or unknown size, marking faulty!\n",
2169 bdevname(rdev->bdev,b));
2170 err = -EINVAL;
2171 goto abort_free;
2172 }
2173
2174 if (super_format >= 0) {
2175 err = super_types[super_format].
2176 load_super(rdev, NULL, super_minor);
2177 if (err == -EINVAL) {
df968c4e
N		 2178 printk(KERN_WARNING
2179 "md: %s does not have a valid v%d.%d "
2180 "superblock, not importing!\n",
2181 bdevname(rdev->bdev,b),
2182 super_format, super_minor);
1da177e4
LT		 2183 goto abort_free;
2184 }
2185 if (err < 0) {
2186 printk(KERN_WARNING
2187 "md: could not read %s's sb, not importing!\n",
2188 bdevname(rdev->bdev,b));
2189 goto abort_free;
2190 }
2191 }
2192 INIT_LIST_HEAD(&rdev->same_set);
2193
2194 return rdev;
2195
2196abort_free:
2197 if (rdev->sb_page) {
2198 if (rdev->bdev)
2199 unlock_rdev(rdev);
2200 free_disk_sb(rdev);
2201 }
2202 kfree(rdev);
2203 return ERR_PTR(err);
2204}
2205
2206/*
2207 * Check a full RAID array for plausibility
2208 */
2209
2210
a757e64c 2211static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2212{
2213 int i;
2214 struct list_head *tmp;
2215 mdk_rdev_t *rdev, *freshest;
2216 char b[BDEVNAME_SIZE];
2217
2218 freshest = NULL;
d089c6af 2219 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2220 switch (super_types[mddev->major_version].
2221 load_super(rdev, freshest, mddev->minor_version)) {
2222 case 1:
2223 freshest = rdev;
2224 break;
2225 case 0:
2226 break;
2227 default:
2228 printk( KERN_ERR \
2229 "md: fatal superblock inconsistency in %s"
2230 " -- removing from array\n",
2231 bdevname(rdev->bdev,b));
2232 kick_rdev_from_array(rdev);
2233 }
2234
2235
2236 super_types[mddev->major_version].
2237 validate_super(mddev, freshest);
2238
2239 i = 0;
d089c6af 2240 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 2241 if (rdev != freshest)
2242 if (super_types[mddev->major_version].
2243 validate_super(mddev, rdev)) {
2244 printk(KERN_WARNING "md: kicking non-fresh %s"
2245 " from array!\n",
2246 bdevname(rdev->bdev,b));
2247 kick_rdev_from_array(rdev);
2248 continue;
2249 }
2250 if (mddev->level == LEVEL_MULTIPATH) {
2251 rdev->desc_nr = i++;
2252 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2253 set_bit(In_sync, &rdev->flags);
a778b73f
N		 2254 } else if (rdev->raid_disk >= mddev->raid_disks) {
2255 rdev->raid_disk = -1;
2256 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2257 }
2258 }
2259
2260
2261
2262 if (mddev->recovery_cp != MaxSector &&
2263 mddev->level >= 1)
2264 printk(KERN_ERR "md: %s: raid array is not clean"
2265 " -- starting background reconstruction\n",
2266 mdname(mddev));
2267
1da177e4
LT		 2268}
2269
16f17b39
N		 2270static ssize_t
2271safe_delay_show(mddev_t *mddev, char *page)
2272{
2273 int msec = (mddev->safemode_delay*1000)/HZ;
2274 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2275}
2276static ssize_t
2277safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2278{
2279 int scale=1;
2280 int dot=0;
2281 int i;
2282 unsigned long msec;
2283 char buf[30];
2284 char *e;
2285 /* remove a period, and count digits after it */
2286 if (len >= sizeof(buf))
2287 return -EINVAL;
2288 strlcpy(buf, cbuf, len);
2289 buf[len] = 0;
2290 for (i=0; i<len; i++) {
2291 if (dot) {
2292 if (isdigit(buf[i])) {
2293 buf[i-1] = buf[i];
2294 scale *= 10;
2295 }
2296 buf[i] = 0;
2297 } else if (buf[i] == '.') {
2298 dot=1;
2299 buf[i] = 0;
2300 }
2301 }
2302 msec = simple_strtoul(buf, &e, 10);
2303 if (e == buf || (*e && *e != '\n'))
2304 return -EINVAL;
2305 msec = (msec * 1000) / scale;
2306 if (msec == 0)
2307 mddev->safemode_delay = 0;
2308 else {
2309 mddev->safemode_delay = (msec*HZ)/1000;
2310 if (mddev->safemode_delay == 0)
2311 mddev->safemode_delay = 1;
2312 }
2313 return len;
2314}
2315static struct md_sysfs_entry md_safe_delay =
80ca3a44 2316__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2317
eae1701f 2318static ssize_t
96de1e66 2319level_show(mddev_t *mddev, char *page)
eae1701f 2320{
2604b703 2321 struct mdk_personality *p = mddev->pers;
d9d166c2 2322 if (p)
eae1701f 2323 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2324 else if (mddev->clevel[0])
2325 return sprintf(page, "%s\n", mddev->clevel);
2326 else if (mddev->level != LEVEL_NONE)
2327 return sprintf(page, "%d\n", mddev->level);
2328 else
2329 return 0;
eae1701f
N		 2330}
2331
d9d166c2
N		 2332static ssize_t
2333level_store(mddev_t *mddev, const char *buf, size_t len)
2334{
20a49ff6 2335 ssize_t rv = len;
d9d166c2
N		 2336 if (mddev->pers)
2337 return -EBUSY;
2338 if (len == 0)
2339 return 0;
2340 if (len >= sizeof(mddev->clevel))
2341 return -ENOSPC;
2342 strncpy(mddev->clevel, buf, len);
2343 if (mddev->clevel[len-1] == '\n')
2344 len--;
2345 mddev->clevel[len] = 0;
2346 mddev->level = LEVEL_NONE;
2347 return rv;
2348}
2349
2350static struct md_sysfs_entry md_level =
80ca3a44 2351__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2352
d4dbd025
N		 2353
2354static ssize_t
2355layout_show(mddev_t *mddev, char *page)
2356{
2357 /* just a number, not meaningful for all levels */
08a02ecd
N		 2358 if (mddev->reshape_position != MaxSector &&
2359 mddev->layout != mddev->new_layout)
2360 return sprintf(page, "%d (%d)\n",
2361 mddev->new_layout, mddev->layout);
d4dbd025
N		 2362 return sprintf(page, "%d\n", mddev->layout);
2363}
2364
2365static ssize_t
2366layout_store(mddev_t *mddev, const char *buf, size_t len)
2367{
2368 char *e;
2369 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 2370
2371 if (!*buf || (*e && *e != '\n'))
2372 return -EINVAL;
2373
08a02ecd
N		 2374 if (mddev->pers)
2375 return -EBUSY;
2376 if (mddev->reshape_position != MaxSector)
2377 mddev->new_layout = n;
2378 else
2379 mddev->layout = n;
d4dbd025
N		 2380 return len;
2381}
2382static struct md_sysfs_entry md_layout =
80ca3a44 2383__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2384
2385
eae1701f 2386static ssize_t
96de1e66 2387raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2388{
bb636547
N		 2389 if (mddev->raid_disks == 0)
2390 return 0;
08a02ecd
N		 2391 if (mddev->reshape_position != MaxSector &&
2392 mddev->delta_disks != 0)
2393 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2394 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 2395 return sprintf(page, "%d\n", mddev->raid_disks);
2396}
2397
da943b99
N		 2398static int update_raid_disks(mddev_t *mddev, int raid_disks);
2399
2400static ssize_t
2401raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2402{
da943b99
N		 2403 char *e;
2404 int rv = 0;
2405 unsigned long n = simple_strtoul(buf, &e, 10);
2406
2407 if (!*buf || (*e && *e != '\n'))
2408 return -EINVAL;
2409
2410 if (mddev->pers)
2411 rv = update_raid_disks(mddev, n);
08a02ecd
N		 2412 else if (mddev->reshape_position != MaxSector) {
2413 int olddisks = mddev->raid_disks - mddev->delta_disks;
2414 mddev->delta_disks = n - olddisks;
2415 mddev->raid_disks = n;
2416 } else
da943b99
N		 2417 mddev->raid_disks = n;
2418 return rv ? rv : len;
2419}
2420static struct md_sysfs_entry md_raid_disks =
80ca3a44 2421__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2422
3b34380a
N		 2423static ssize_t
2424chunk_size_show(mddev_t *mddev, char *page)
2425{
08a02ecd
N		 2426 if (mddev->reshape_position != MaxSector &&
2427 mddev->chunk_size != mddev->new_chunk)
2428 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2429 mddev->chunk_size);
3b34380a
N		 2430 return sprintf(page, "%d\n", mddev->chunk_size);
2431}
2432
2433static ssize_t
2434chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2435{
2436 /* can only set chunk_size if array is not yet active */
2437 char *e;
2438 unsigned long n = simple_strtoul(buf, &e, 10);
2439
3b34380a
N		 2440 if (!*buf || (*e && *e != '\n'))
2441 return -EINVAL;
2442
08a02ecd
N		 2443 if (mddev->pers)
2444 return -EBUSY;
2445 else if (mddev->reshape_position != MaxSector)
2446 mddev->new_chunk = n;
2447 else
2448 mddev->chunk_size = n;
3b34380a
N		 2449 return len;
2450}
2451static struct md_sysfs_entry md_chunk_size =
80ca3a44 2452__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2453
a94213b1
N		 2454static ssize_t
2455resync_start_show(mddev_t *mddev, char *page)
2456{
2457 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2458}
2459
2460static ssize_t
2461resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2462{
2463 /* can only set chunk_size if array is not yet active */
2464 char *e;
2465 unsigned long long n = simple_strtoull(buf, &e, 10);
2466
2467 if (mddev->pers)
2468 return -EBUSY;
2469 if (!*buf || (*e && *e != '\n'))
2470 return -EINVAL;
2471
2472 mddev->recovery_cp = n;
2473 return len;
2474}
2475static struct md_sysfs_entry md_resync_start =
80ca3a44 2476__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2477
9e653b63
N		 2478/*
2479 * The array state can be:
2480 *
2481 * clear
2482 * No devices, no size, no level
2483 * Equivalent to STOP_ARRAY ioctl
2484 * inactive
2485 * May have some settings, but array is not active
2486 * all IO results in error
2487 * When written, doesn't tear down array, but just stops it
2488 * suspended (not supported yet)
2489 * All IO requests will block. The array can be reconfigured.
2490 * Writing this, if accepted, will block until array is quiessent
2491 * readonly
2492 * no resync can happen. no superblocks get written.
2493 * write requests fail
2494 * read-auto
2495 * like readonly, but behaves like 'clean' on a write request.
2496 *
2497 * clean - no pending writes, but otherwise active.
2498 * When written to inactive array, starts without resync
2499 * If a write request arrives then
2500 * if metadata is known, mark 'dirty' and switch to 'active'.
2501 * if not known, block and switch to write-pending
2502 * If written to an active array that has pending writes, then fails.
2503 * active
2504 * fully active: IO and resync can be happening.
2505 * When written to inactive array, starts with resync
2506 *
2507 * write-pending
2508 * clean, but writes are blocked waiting for 'active' to be written.
2509 *
2510 * active-idle
2511 * like active, but no writes have been seen for a while (100msec).
2512 *
2513 */
2514enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2515 write_pending, active_idle, bad_word};
05381954 2516static char *array_states[] = {
9e653b63
N		 2517 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2518 "write-pending", "active-idle", NULL };
2519
2520static int match_word(const char *word, char **list)
2521{
2522 int n;
2523 for (n=0; list[n]; n++)
2524 if (cmd_match(word, list[n]))
2525 break;
2526 return n;
2527}
2528
2529static ssize_t
2530array_state_show(mddev_t *mddev, char *page)
2531{
2532 enum array_state st = inactive;
2533
2534 if (mddev->pers)
2535 switch(mddev->ro) {
2536 case 1:
2537 st = readonly;
2538 break;
2539 case 2:
2540 st = read_auto;
2541 break;
2542 case 0:
2543 if (mddev->in_sync)
2544 st = clean;
e691063a
N		 2545 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2546 st = write_pending;
9e653b63
N		 2547 else if (mddev->safemode)
2548 st = active_idle;
2549 else
2550 st = active;
2551 }
2552 else {
2553 if (list_empty(&mddev->disks) &&
2554 mddev->raid_disks == 0 &&
2555 mddev->size == 0)
2556 st = clear;
2557 else
2558 st = inactive;
2559 }
2560 return sprintf(page, "%s\n", array_states[st]);
2561}
2562
2563static int do_md_stop(mddev_t * mddev, int ro);
2564static int do_md_run(mddev_t * mddev);
2565static int restart_array(mddev_t *mddev);
2566
2567static ssize_t
2568array_state_store(mddev_t *mddev, const char *buf, size_t len)
2569{
2570 int err = -EINVAL;
2571 enum array_state st = match_word(buf, array_states);
2572 switch(st) {
2573 case bad_word:
2574 break;
2575 case clear:
2576 /* stopping an active array */
e691063a
N		 2577 if (atomic_read(&mddev->active) > 1)
2578 return -EBUSY;
2579 err = do_md_stop(mddev, 0);
9e653b63
N		 2580 break;
2581 case inactive:
2582 /* stopping an active array */
2583 if (mddev->pers) {
2584 if (atomic_read(&mddev->active) > 1)
2585 return -EBUSY;
2586 err = do_md_stop(mddev, 2);
e691063a
N		 2587 } else
2588 err = 0; /* already inactive */
9e653b63
N		 2589 break;
2590 case suspended:
2591 break; /* not supported yet */
2592 case readonly:
2593 if (mddev->pers)
2594 err = do_md_stop(mddev, 1);
2595 else {
2596 mddev->ro = 1;
2597 err = do_md_run(mddev);
2598 }
2599 break;
2600 case read_auto:
2601 /* stopping an active array */
2602 if (mddev->pers) {
2603 err = do_md_stop(mddev, 1);
2604 if (err == 0)
2605 mddev->ro = 2; /* FIXME mark devices writable */
2606 } else {
2607 mddev->ro = 2;
2608 err = do_md_run(mddev);
2609 }
2610 break;
2611 case clean:
2612 if (mddev->pers) {
2613 restart_array(mddev);
2614 spin_lock_irq(&mddev->write_lock);
2615 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 2616 if (mddev->in_sync == 0) {
2617 mddev->in_sync = 1;
2618 if (mddev->persistent)
2619 set_bit(MD_CHANGE_CLEAN,
2620 &mddev->flags);
2621 }
2622 err = 0;
2623 } else
2624 err = -EBUSY;
9e653b63
N		 2625 spin_unlock_irq(&mddev->write_lock);
2626 } else {
2627 mddev->ro = 0;
2628 mddev->recovery_cp = MaxSector;
2629 err = do_md_run(mddev);
2630 }
2631 break;
2632 case active:
2633 if (mddev->pers) {
2634 restart_array(mddev);
e691063a
N		 2635 if (mddev->external)
2636 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2637 wake_up(&mddev->sb_wait);
2638 err = 0;
2639 } else {
2640 mddev->ro = 0;
2641 err = do_md_run(mddev);
2642 }
2643 break;
2644 case write_pending:
2645 case active_idle:
2646 /* these cannot be set */
2647 break;
2648 }
2649 if (err)
2650 return err;
2651 else
2652 return len;
2653}
80ca3a44
N		 2654static struct md_sysfs_entry md_array_state =
2655__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2656
6d7ff738
N		 2657static ssize_t
2658null_show(mddev_t *mddev, char *page)
2659{
2660 return -EINVAL;
2661}
2662
2663static ssize_t
2664new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2665{
2666 /* buf must be %d:%d\n? giving major and minor numbers */
2667 /* The new device is added to the array.
2668 * If the array has a persistent superblock, we read the
2669 * superblock to initialise info and check validity.
2670 * Otherwise, only checking done is that in bind_rdev_to_array,
2671 * which mainly checks size.
2672 */
2673 char *e;
2674 int major = simple_strtoul(buf, &e, 10);
2675 int minor;
2676 dev_t dev;
2677 mdk_rdev_t *rdev;
2678 int err;
2679
2680 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2681 return -EINVAL;
2682 minor = simple_strtoul(e+1, &e, 10);
2683 if (*e && *e != '\n')
2684 return -EINVAL;
2685 dev = MKDEV(major, minor);
2686 if (major != MAJOR(dev) ||
2687 minor != MINOR(dev))
2688 return -EOVERFLOW;
2689
2690
2691 if (mddev->persistent) {
2692 rdev = md_import_device(dev, mddev->major_version,
2693 mddev->minor_version);
2694 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2695 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2696 mdk_rdev_t, same_set);
2697 err = super_types[mddev->major_version]
2698 .load_super(rdev, rdev0, mddev->minor_version);
2699 if (err < 0)
2700 goto out;
2701 }
c5d79adb
N		 2702 } else if (mddev->external)
2703 rdev = md_import_device(dev, -2, -1);
2704 else
6d7ff738
N		 2705 rdev = md_import_device(dev, -1, -1);
2706
2707 if (IS_ERR(rdev))
2708 return PTR_ERR(rdev);
2709 err = bind_rdev_to_array(rdev, mddev);
2710 out:
2711 if (err)
2712 export_rdev(rdev);
2713 return err ? err : len;
2714}
2715
2716static struct md_sysfs_entry md_new_device =
80ca3a44 2717__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2718
9b1d1dac
PC		 2719static ssize_t
2720bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2721{
2722 char *end;
2723 unsigned long chunk, end_chunk;
2724
2725 if (!mddev->bitmap)
2726 goto out;
2727 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2728 while (*buf) {
2729 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2730 if (buf == end) break;
2731 if (*end == '-') { /* range */
2732 buf = end + 1;
2733 end_chunk = simple_strtoul(buf, &end, 0);
2734 if (buf == end) break;
2735 }
2736 if (*end && !isspace(*end)) break;
2737 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2738 buf = end;
2739 while (isspace(*buf)) buf++;
2740 }
2741 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2742out:
2743 return len;
2744}
2745
2746static struct md_sysfs_entry md_bitmap =
2747__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2748
a35b0d69
N		 2749static ssize_t
2750size_show(mddev_t *mddev, char *page)
2751{
2752 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2753}
2754
2755static int update_size(mddev_t *mddev, unsigned long size);
2756
2757static ssize_t
2758size_store(mddev_t *mddev, const char *buf, size_t len)
2759{
2760 /* If array is inactive, we can reduce the component size, but
2761 * not increase it (except from 0).
2762 * If array is active, we can try an on-line resize
2763 */
2764 char *e;
2765 int err = 0;
2766 unsigned long long size = simple_strtoull(buf, &e, 10);
2767 if (!*buf || *buf == '\n' ||
2768 (*e && *e != '\n'))
2769 return -EINVAL;
2770
2771 if (mddev->pers) {
2772 err = update_size(mddev, size);
850b2b42 2773 md_update_sb(mddev, 1);
a35b0d69
N		 2774 } else {
2775 if (mddev->size == 0 ||
2776 mddev->size > size)
2777 mddev->size = size;
2778 else
2779 err = -ENOSPC;
2780 }
2781 return err ? err : len;
2782}
2783
2784static struct md_sysfs_entry md_size =
80ca3a44 2785__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2786
8bb93aac
N		 2787
2788/* Metdata version.
e691063a
N		 2789 * This is one of
2790 * 'none' for arrays with no metadata (good luck...)
2791 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 2792 * or N.M for internally known formats
2793 */
2794static ssize_t
2795metadata_show(mddev_t *mddev, char *page)
2796{
2797 if (mddev->persistent)
2798 return sprintf(page, "%d.%d\n",
2799 mddev->major_version, mddev->minor_version);
e691063a
N		 2800 else if (mddev->external)
2801 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 2802 else
2803 return sprintf(page, "none\n");
2804}
2805
2806static ssize_t
2807metadata_store(mddev_t *mddev, const char *buf, size_t len)
2808{
2809 int major, minor;
2810 char *e;
2811 if (!list_empty(&mddev->disks))
2812 return -EBUSY;
2813
2814 if (cmd_match(buf, "none")) {
2815 mddev->persistent = 0;
e691063a
N		 2816 mddev->external = 0;
2817 mddev->major_version = 0;
2818 mddev->minor_version = 90;
2819 return len;
2820 }
2821 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 2822 size_t namelen = len-9;
e691063a
N		 2823 if (namelen >= sizeof(mddev->metadata_type))
2824 namelen = sizeof(mddev->metadata_type)-1;
2825 strncpy(mddev->metadata_type, buf+9, namelen);
2826 mddev->metadata_type[namelen] = 0;
2827 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2828 mddev->metadata_type[--namelen] = 0;
2829 mddev->persistent = 0;
2830 mddev->external = 1;
8bb93aac
N		 2831 mddev->major_version = 0;
2832 mddev->minor_version = 90;
2833 return len;
2834 }
2835 major = simple_strtoul(buf, &e, 10);
2836 if (e==buf || *e != '.')
2837 return -EINVAL;
2838 buf = e+1;
2839 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 2840 if (e==buf || (*e && *e != '\n') )
8bb93aac 2841 return -EINVAL;
50511da3 2842 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 2843 return -ENOENT;
2844 mddev->major_version = major;
2845 mddev->minor_version = minor;
2846 mddev->persistent = 1;
e691063a 2847 mddev->external = 0;
8bb93aac
N		 2848 return len;
2849}
2850
2851static struct md_sysfs_entry md_metadata =
80ca3a44 2852__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2853
24dd469d 2854static ssize_t
7eec314d 2855action_show(mddev_t *mddev, char *page)
24dd469d 2856{
7eec314d 2857 char *type = "idle";
31399d9e 2858 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 2859 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 2860 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2861 type = "reshape";
2862 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2863 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2864 type = "resync";
2865 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2866 type = "check";
2867 else
2868 type = "repair";
2869 } else
2870 type = "recover";
2871 }
2872 return sprintf(page, "%s\n", type);
2873}
2874
2875static ssize_t
7eec314d 2876action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2877{
7eec314d
N		 2878 if (!mddev->pers || !mddev->pers->sync_request)
2879 return -EINVAL;
2880
bce74dac 2881 if (cmd_match(page, "idle")) {
7eec314d
N		 2882 if (mddev->sync_thread) {
2883 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2884 md_unregister_thread(mddev->sync_thread);
2885 mddev->sync_thread = NULL;
2886 mddev->recovery = 0;
2887 }
03c902e1
N		 2888 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2889 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2890 return -EBUSY;
03c902e1 2891 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2892 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2893 else if (cmd_match(page, "reshape")) {
2894 int err;
2895 if (mddev->pers->start_reshape == NULL)
2896 return -EINVAL;
2897 err = mddev->pers->start_reshape(mddev);
2898 if (err)
2899 return err;
2900 } else {
bce74dac 2901 if (cmd_match(page, "check"))
7eec314d 2902 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2903 else if (!cmd_match(page, "repair"))
7eec314d
N		 2904 return -EINVAL;
2905 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2906 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2907 }
03c902e1 2908 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2909 md_wakeup_thread(mddev->thread);
2910 return len;
2911}
2912
9d88883e 2913static ssize_t
96de1e66 2914mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2915{
2916 return sprintf(page, "%llu\n",
2917 (unsigned long long) mddev->resync_mismatches);
2918}
2919
80ca3a44
N		 2920static struct md_sysfs_entry md_scan_mode =
2921__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2922
96de1e66 2923
80ca3a44 2924static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2925
88202a0c
N		 2926static ssize_t
2927sync_min_show(mddev_t *mddev, char *page)
2928{
2929 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2930 mddev->sync_speed_min ? "local": "system");
2931}
2932
2933static ssize_t
2934sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2935{
2936 int min;
2937 char *e;
2938 if (strncmp(buf, "system", 6)==0) {
2939 mddev->sync_speed_min = 0;
2940 return len;
2941 }
2942 min = simple_strtoul(buf, &e, 10);
2943 if (buf == e || (*e && *e != '\n') || min <= 0)
2944 return -EINVAL;
2945 mddev->sync_speed_min = min;
2946 return len;
2947}
2948
2949static struct md_sysfs_entry md_sync_min =
2950__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2951
2952static ssize_t
2953sync_max_show(mddev_t *mddev, char *page)
2954{
2955 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2956 mddev->sync_speed_max ? "local": "system");
2957}
2958
2959static ssize_t
2960sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2961{
2962 int max;
2963 char *e;
2964 if (strncmp(buf, "system", 6)==0) {
2965 mddev->sync_speed_max = 0;
2966 return len;
2967 }
2968 max = simple_strtoul(buf, &e, 10);
2969 if (buf == e || (*e && *e != '\n') || max <= 0)
2970 return -EINVAL;
2971 mddev->sync_speed_max = max;
2972 return len;
2973}
2974
2975static struct md_sysfs_entry md_sync_max =
2976__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2977
d7f3d291
IP		 2978static ssize_t
2979degraded_show(mddev_t *mddev, char *page)
2980{
2981 return sprintf(page, "%d\n", mddev->degraded);
2982}
2983static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c
N		 2984
2985static ssize_t
2986sync_speed_show(mddev_t *mddev, char *page)
2987{
2988 unsigned long resync, dt, db;
ff4e8d9a 2989 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2990 dt = ((jiffies - mddev->resync_mark) / HZ);
2991 if (!dt) dt++;
2992 db = resync - (mddev->resync_mark_cnt);
2993 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2994}
2995
80ca3a44 2996static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2997
2998static ssize_t
2999sync_completed_show(mddev_t *mddev, char *page)
3000{
3001 unsigned long max_blocks, resync;
3002
3003 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3004 max_blocks = mddev->resync_max_sectors;
3005 else
3006 max_blocks = mddev->size << 1;
3007
3008 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3009 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3010}
3011
80ca3a44 3012static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3013
c6207277
N		 3014static ssize_t
3015max_sync_show(mddev_t *mddev, char *page)
3016{
3017 if (mddev->resync_max == MaxSector)
3018 return sprintf(page, "max\n");
3019 else
3020 return sprintf(page, "%llu\n",
3021 (unsigned long long)mddev->resync_max);
3022}
3023static ssize_t
3024max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3025{
3026 if (strncmp(buf, "max", 3) == 0)
3027 mddev->resync_max = MaxSector;
3028 else {
3029 char *ep;
3030 unsigned long long max = simple_strtoull(buf, &ep, 10);
3031 if (ep == buf || (*ep != 0 && *ep != '\n'))
3032 return -EINVAL;
3033 if (max < mddev->resync_max &&
3034 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3035 return -EBUSY;
3036
3037 /* Must be a multiple of chunk_size */
3038 if (mddev->chunk_size) {
3039 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3040 return -EINVAL;
3041 }
3042 mddev->resync_max = max;
3043 }
3044 wake_up(&mddev->recovery_wait);
3045 return len;
3046}
3047
3048static struct md_sysfs_entry md_max_sync =
3049__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3050
e464eafd
N		 3051static ssize_t
3052suspend_lo_show(mddev_t *mddev, char *page)
3053{
3054 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3055}
3056
3057static ssize_t
3058suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3059{
3060 char *e;
3061 unsigned long long new = simple_strtoull(buf, &e, 10);
3062
3063 if (mddev->pers->quiesce == NULL)
3064 return -EINVAL;
3065 if (buf == e || (*e && *e != '\n'))
3066 return -EINVAL;
3067 if (new >= mddev->suspend_hi ||
3068 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3069 mddev->suspend_lo = new;
3070 mddev->pers->quiesce(mddev, 2);
3071 return len;
3072 } else
3073 return -EINVAL;
3074}
3075static struct md_sysfs_entry md_suspend_lo =
3076__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3077
3078
3079static ssize_t
3080suspend_hi_show(mddev_t *mddev, char *page)
3081{
3082 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3083}
3084
3085static ssize_t
3086suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3087{
3088 char *e;
3089 unsigned long long new = simple_strtoull(buf, &e, 10);
3090
3091 if (mddev->pers->quiesce == NULL)
3092 return -EINVAL;
3093 if (buf == e || (*e && *e != '\n'))
3094 return -EINVAL;
3095 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3096 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3097 mddev->suspend_hi = new;
3098 mddev->pers->quiesce(mddev, 1);
3099 mddev->pers->quiesce(mddev, 0);
3100 return len;
3101 } else
3102 return -EINVAL;
3103}
3104static struct md_sysfs_entry md_suspend_hi =
3105__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3106
08a02ecd
N		 3107static ssize_t
3108reshape_position_show(mddev_t *mddev, char *page)
3109{
3110 if (mddev->reshape_position != MaxSector)
3111 return sprintf(page, "%llu\n",
3112 (unsigned long long)mddev->reshape_position);
3113 strcpy(page, "none\n");
3114 return 5;
3115}
3116
3117static ssize_t
3118reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3119{
3120 char *e;
3121 unsigned long long new = simple_strtoull(buf, &e, 10);
3122 if (mddev->pers)
3123 return -EBUSY;
3124 if (buf == e || (*e && *e != '\n'))
3125 return -EINVAL;
3126 mddev->reshape_position = new;
3127 mddev->delta_disks = 0;
3128 mddev->new_level = mddev->level;
3129 mddev->new_layout = mddev->layout;
3130 mddev->new_chunk = mddev->chunk_size;
3131 return len;
3132}
3133
3134static struct md_sysfs_entry md_reshape_position =
3135__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3136 reshape_position_store);
3137
e464eafd 3138
eae1701f
N		 3139static struct attribute *md_default_attrs[] = {
3140 &md_level.attr,
d4dbd025 3141 &md_layout.attr,
eae1701f 3142 &md_raid_disks.attr,
3b34380a 3143 &md_chunk_size.attr,
a35b0d69 3144 &md_size.attr,
a94213b1 3145 &md_resync_start.attr,
8bb93aac 3146 &md_metadata.attr,
6d7ff738 3147 &md_new_device.attr,
16f17b39 3148 &md_safe_delay.attr,
9e653b63 3149 &md_array_state.attr,
08a02ecd 3150 &md_reshape_position.attr,
411036fa
N		 3151 NULL,
3152};
3153
3154static struct attribute *md_redundancy_attrs[] = {
24dd469d 3155 &md_scan_mode.attr,
9d88883e 3156 &md_mismatches.attr,
88202a0c
N		 3157 &md_sync_min.attr,
3158 &md_sync_max.attr,
3159 &md_sync_speed.attr,
3160 &md_sync_completed.attr,
c6207277 3161 &md_max_sync.attr,
e464eafd
N		 3162 &md_suspend_lo.attr,
3163 &md_suspend_hi.attr,
9b1d1dac 3164 &md_bitmap.attr,
d7f3d291 3165 &md_degraded.attr,
eae1701f
N		 3166 NULL,
3167};
411036fa
N		 3168static struct attribute_group md_redundancy_group = {
3169 .name = NULL,
3170 .attrs = md_redundancy_attrs,
3171};
3172
eae1701f
N		 3173
3174static ssize_t
3175md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3176{
3177 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3178 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3179 ssize_t rv;
eae1701f
N		 3180
3181 if (!entry->show)
3182 return -EIO;
5dc5cf7d
IM		 3183 rv = mddev_lock(mddev);
3184 if (!rv) {
3185 rv = entry->show(mddev, page);
3186 mddev_unlock(mddev);
3187 }
96de1e66 3188 return rv;
eae1701f
N		 3189}
3190
3191static ssize_t
3192md_attr_store(struct kobject *kobj, struct attribute *attr,
3193 const char *page, size_t length)
3194{
3195 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3196 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 3197 ssize_t rv;
eae1701f
N		 3198
3199 if (!entry->store)
3200 return -EIO;
67463acb
N		 3201 if (!capable(CAP_SYS_ADMIN))
3202 return -EACCES;
5dc5cf7d
IM		 3203 rv = mddev_lock(mddev);
3204 if (!rv) {
3205 rv = entry->store(mddev, page, length);
3206 mddev_unlock(mddev);
3207 }
96de1e66 3208 return rv;
eae1701f
N		 3209}
3210
3211static void md_free(struct kobject *ko)
3212{
3213 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3214 kfree(mddev);
3215}
3216
3217static struct sysfs_ops md_sysfs_ops = {
3218 .show = md_attr_show,
3219 .store = md_attr_store,
3220};
3221static struct kobj_type md_ktype = {
3222 .release = md_free,
3223 .sysfs_ops = &md_sysfs_ops,
3224 .default_attrs = md_default_attrs,
3225};
3226
1da177e4
LT		 3227int mdp_major = 0;
3228
3229static struct kobject *md_probe(dev_t dev, int *part, void *data)
3230{
48c9c27b 3231 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 3232 mddev_t *mddev = mddev_find(dev);
3233 struct gendisk *disk;
3234 int partitioned = (MAJOR(dev) != MD_MAJOR);
3235 int shift = partitioned ? MdpMinorShift : 0;
3236 int unit = MINOR(dev) >> shift;
3830c62f 3237 int error;
1da177e4
LT		 3238
3239 if (!mddev)
3240 return NULL;
3241
48c9c27b 3242 mutex_lock(&disks_mutex);
1da177e4 3243 if (mddev->gendisk) {
48c9c27b 3244 mutex_unlock(&disks_mutex);
1da177e4
LT		 3245 mddev_put(mddev);
3246 return NULL;
3247 }
3248 disk = alloc_disk(1 << shift);
3249 if (!disk) {
48c9c27b 3250 mutex_unlock(&disks_mutex);
1da177e4
LT		 3251 mddev_put(mddev);
3252 return NULL;
3253 }
3254 disk->major = MAJOR(dev);
3255 disk->first_minor = unit << shift;
ce7b0f46 3256 if (partitioned)
1da177e4 3257 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 3258 else
1da177e4 3259 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 3260 disk->fops = &md_fops;
3261 disk->private_data = mddev;
3262 disk->queue = mddev->queue;
3263 add_disk(disk);
3264 mddev->gendisk = disk;
48c9c27b 3265 mutex_unlock(&disks_mutex);
edfaa7c3 3266 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3830c62f
GKH		 3267 "%s", "md");
3268 if (error)
5e55e2f5
N		 3269 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3270 disk->disk_name);
3830c62f
GKH		 3271 else
3272 kobject_uevent(&mddev->kobj, KOBJ_ADD);
1da177e4
LT		 3273 return NULL;
3274}
3275
1da177e4
LT		 3276static void md_safemode_timeout(unsigned long data)
3277{
3278 mddev_t *mddev = (mddev_t *) data;
3279
3280 mddev->safemode = 1;
3281 md_wakeup_thread(mddev->thread);
3282}
3283
6ff8d8ec 3284static int start_dirty_degraded;
1da177e4
LT		 3285
3286static int do_md_run(mddev_t * mddev)
3287{
2604b703 3288 int err;
1da177e4
LT		 3289 int chunk_size;
3290 struct list_head *tmp;
3291 mdk_rdev_t *rdev;
3292 struct gendisk *disk;
2604b703 3293 struct mdk_personality *pers;
1da177e4
LT		 3294 char b[BDEVNAME_SIZE];
3295
a757e64c
N		 3296 if (list_empty(&mddev->disks))
3297 /* cannot run an array with no devices.. */
1da177e4 3298 return -EINVAL;
1da177e4
LT		 3299
3300 if (mddev->pers)
3301 return -EBUSY;
3302
3303 /*
3304 * Analyze all RAID superblock(s)
3305 */
1ec4a939
N		 3306 if (!mddev->raid_disks) {
3307 if (!mddev->persistent)
3308 return -EINVAL;
a757e64c 3309 analyze_sbs(mddev);
1ec4a939 3310 }
1da177e4
LT		 3311
3312 chunk_size = mddev->chunk_size;
2604b703
N		 3313
3314 if (chunk_size) {
1da177e4
LT		 3315 if (chunk_size > MAX_CHUNK_SIZE) {
3316 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3317 chunk_size, MAX_CHUNK_SIZE);
3318 return -EINVAL;
3319 }
3320 /*
3321 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3322 */
3323 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 3324 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 3325 return -EINVAL;
3326 }
3327 if (chunk_size < PAGE_SIZE) {
3328 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3329 chunk_size, PAGE_SIZE);
3330 return -EINVAL;
3331 }
3332
3333 /* devices must have minimum size of one chunk */
d089c6af 3334 rdev_for_each(rdev, tmp, mddev) {
b2d444d7 3335 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3336 continue;
3337 if (rdev->size < chunk_size / 1024) {
3338 printk(KERN_WARNING
3339 "md: Dev %s smaller than chunk_size:"
3340 " %lluk < %dk\n",
3341 bdevname(rdev->bdev,b),
3342 (unsigned long long)rdev->size,
3343 chunk_size / 1024);
3344 return -EINVAL;
3345 }
3346 }
3347 }
3348
1da177e4 3349#ifdef CONFIG_KMOD
d9d166c2
N		 3350 if (mddev->level != LEVEL_NONE)
3351 request_module("md-level-%d", mddev->level);
3352 else if (mddev->clevel[0])
3353 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3354#endif
3355
3356 /*
3357 * Drop all container device buffers, from now on
3358 * the only valid external interface is through the md
3359 * device.
1da177e4 3360 */
d089c6af 3361 rdev_for_each(rdev, tmp, mddev) {
b2d444d7 3362 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3363 continue;
3364 sync_blockdev(rdev->bdev);
f98393a6 3365 invalidate_bdev(rdev->bdev);
f0d76d70
N		 3366
3367 /* perform some consistency tests on the device.
3368 * We don't want the data to overlap the metadata,
3369 * Internal Bitmap issues has handled elsewhere.
3370 */
3371 if (rdev->data_offset < rdev->sb_offset) {
3372 if (mddev->size &&
3373 rdev->data_offset + mddev->size*2
3374 > rdev->sb_offset*2) {
3375 printk("md: %s: data overlaps metadata\n",
3376 mdname(mddev));
3377 return -EINVAL;
3378 }
3379 } else {
3380 if (rdev->sb_offset*2 + rdev->sb_size/512
3381 > rdev->data_offset) {
3382 printk("md: %s: metadata overlaps data\n",
3383 mdname(mddev));
3384 return -EINVAL;
3385 }
3386 }
1da177e4
LT		 3387 }
3388
3389 md_probe(mddev->unit, NULL, NULL);
3390 disk = mddev->gendisk;
3391 if (!disk)
3392 return -ENOMEM;
3393
3394 spin_lock(&pers_lock);
d9d166c2 3395 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3396 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3397 spin_unlock(&pers_lock);
d9d166c2
N		 3398 if (mddev->level != LEVEL_NONE)
3399 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3400 mddev->level);
3401 else
3402 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3403 mddev->clevel);
1da177e4
LT		 3404 return -EINVAL;
3405 }
2604b703 3406 mddev->pers = pers;
1da177e4 3407 spin_unlock(&pers_lock);
d9d166c2
N		 3408 mddev->level = pers->level;
3409 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3410
f6705578 3411 if (mddev->reshape_position != MaxSector &&
63c70c4f 3412 pers->start_reshape == NULL) {
f6705578
N		 3413 /* This personality cannot handle reshaping... */
3414 mddev->pers = NULL;
3415 module_put(pers->owner);
3416 return -EINVAL;
3417 }
3418
7dd5e7c3
N		 3419 if (pers->sync_request) {
3420 /* Warn if this is a potentially silly
3421 * configuration.
3422 */
3423 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3424 mdk_rdev_t *rdev2;
3425 struct list_head *tmp2;
3426 int warned = 0;
d089c6af
N		 3427 rdev_for_each(rdev, tmp, mddev) {
3428 rdev_for_each(rdev2, tmp2, mddev) {
7dd5e7c3
N		 3429 if (rdev < rdev2 &&
3430 rdev->bdev->bd_contains ==
3431 rdev2->bdev->bd_contains) {
3432 printk(KERN_WARNING
3433 "%s: WARNING: %s appears to be"
3434 " on the same physical disk as"
3435 " %s.\n",
3436 mdname(mddev),
3437 bdevname(rdev->bdev,b),
3438 bdevname(rdev2->bdev,b2));
3439 warned = 1;
3440 }
3441 }
3442 }
3443 if (warned)
3444 printk(KERN_WARNING
3445 "True protection against single-disk"
3446 " failure might be compromised.\n");
3447 }
3448
657390d2 3449 mddev->recovery = 0;
1da177e4 3450 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3451 mddev->barriers_work = 1;
6ff8d8ec 3452 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3453
f91de92e
N		 3454 if (start_readonly)
3455 mddev->ro = 2; /* read-only, but switch on first write */
3456
b15c2e57
N		 3457 err = mddev->pers->run(mddev);
3458 if (!err && mddev->pers->sync_request) {
3459 err = bitmap_create(mddev);
3460 if (err) {
3461 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3462 mdname(mddev), err);
3463 mddev->pers->stop(mddev);
3464 }
3465 }
1da177e4
LT		 3466 if (err) {
3467 printk(KERN_ERR "md: pers->run() failed ...\n");
3468 module_put(mddev->pers->owner);
3469 mddev->pers = NULL;
32a7627c
N		 3470 bitmap_destroy(mddev);
3471 return err;
1da177e4 3472 }
5e55e2f5
N		 3473 if (mddev->pers->sync_request) {
3474 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3475 printk(KERN_WARNING
3476 "md: cannot register extra attributes for %s\n",
3477 mdname(mddev));
3478 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 3479 mddev->ro = 0;
3480
1da177e4
LT		 3481 atomic_set(&mddev->writes_pending,0);
3482 mddev->safemode = 0;
3483 mddev->safemode_timer.function = md_safemode_timeout;
3484 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3485 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3486 mddev->in_sync = 1;
86e6ffdd 3487
d089c6af 3488 rdev_for_each(rdev, tmp, mddev)
86e6ffdd
N		 3489 if (rdev->raid_disk >= 0) {
3490 char nm[20];
3491 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 3492 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3493 printk("md: cannot register %s for %s\n",
3494 nm, mdname(mddev));
86e6ffdd 3495 }
1da177e4
LT		 3496
3497 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3498
850b2b42
N		 3499 if (mddev->flags)
3500 md_update_sb(mddev, 0);
1da177e4
LT		 3501
3502 set_capacity(disk, mddev->array_size<<1);
3503
3504 /* If we call blk_queue_make_request here, it will
3505 * re-initialise max_sectors etc which may have been
3506 * refined inside -> run. So just set the bits we need to set.
3507 * Most initialisation happended when we called
3508 * blk_queue_make_request(..., md_fail_request)
3509 * earlier.
3510 */
3511 mddev->queue->queuedata = mddev;
3512 mddev->queue->make_request_fn = mddev->pers->make_request;
3513
5fd6c1dc
N		 3514 /* If there is a partially-recovered drive we need to
3515 * start recovery here. If we leave it to md_check_recovery,
3516 * it will remove the drives and not do the right thing
3517 */
0b8c9de0 3518 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3519 struct list_head *rtmp;
3520 int spares = 0;
d089c6af 3521 rdev_for_each(rdev, rtmp, mddev)
5fd6c1dc
N		 3522 if (rdev->raid_disk >= 0 &&
3523 !test_bit(In_sync, &rdev->flags) &&
3524 !test_bit(Faulty, &rdev->flags))
3525 /* complete an interrupted recovery */
3526 spares++;
3527 if (spares && mddev->pers->sync_request) {
3528 mddev->recovery = 0;
3529 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3530 mddev->sync_thread = md_register_thread(md_do_sync,
3531 mddev,
3532 "%s_resync");
3533 if (!mddev->sync_thread) {
3534 printk(KERN_ERR "%s: could not start resync"
3535 " thread...\n",
3536 mdname(mddev));
3537 /* leave the spares where they are, it shouldn't hurt */
3538 mddev->recovery = 0;
0b8c9de0 3539 }
5fd6c1dc
N		 3540 }
3541 }
0b8c9de0
N		 3542 md_wakeup_thread(mddev->thread);
3543 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3544
44ce6294 3545 mddev->changed = 1;
d7603b7e 3546 md_new_event(mddev);
edfaa7c3 3547 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
1da177e4
LT		 3548 return 0;
3549}
3550
3551static int restart_array(mddev_t *mddev)
3552{
3553 struct gendisk *disk = mddev->gendisk;
3554 int err;
3555
3556 /*
3557 * Complain if it has no devices
3558 */
3559 err = -ENXIO;
3560 if (list_empty(&mddev->disks))
3561 goto out;
3562
3563 if (mddev->pers) {
3564 err = -EBUSY;
3565 if (!mddev->ro)
3566 goto out;
3567
3568 mddev->safemode = 0;
3569 mddev->ro = 0;
3570 set_disk_ro(disk, 0);
3571
3572 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3573 mdname(mddev));
3574 /*
3575 * Kick recovery or resync if necessary
3576 */
3577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3578 md_wakeup_thread(mddev->thread);
5fd6c1dc 3579 md_wakeup_thread(mddev->sync_thread);
1da177e4 3580 err = 0;
9e653b63 3581 } else
1da177e4 3582 err = -EINVAL;
1da177e4
LT		 3583
3584out:
3585 return err;
3586}
3587
acc55e22
N		 3588/* similar to deny_write_access, but accounts for our holding a reference
3589 * to the file ourselves */
3590static int deny_bitmap_write_access(struct file * file)
3591{
3592 struct inode *inode = file->f_mapping->host;
3593
3594 spin_lock(&inode->i_lock);
3595 if (atomic_read(&inode->i_writecount) > 1) {
3596 spin_unlock(&inode->i_lock);
3597 return -ETXTBSY;
3598 }
3599 atomic_set(&inode->i_writecount, -1);
3600 spin_unlock(&inode->i_lock);
3601
3602 return 0;
3603}
3604
3605static void restore_bitmap_write_access(struct file *file)
3606{
3607 struct inode *inode = file->f_mapping->host;
3608
3609 spin_lock(&inode->i_lock);
3610 atomic_set(&inode->i_writecount, 1);
3611 spin_unlock(&inode->i_lock);
3612}
3613
9e653b63
N		 3614/* mode:
3615 * 0 - completely stop and dis-assemble array
3616 * 1 - switch to readonly
3617 * 2 - stop but do not disassemble array
3618 */
3619static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3620{
3621 int err = 0;
3622 struct gendisk *disk = mddev->gendisk;
3623
3624 if (mddev->pers) {
3625 if (atomic_read(&mddev->active)>2) {
3626 printk("md: %s still in use.\n",mdname(mddev));
3627 return -EBUSY;
3628 }
3629
3630 if (mddev->sync_thread) {
5fd6c1dc 3631 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3632 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3633 md_unregister_thread(mddev->sync_thread);
3634 mddev->sync_thread = NULL;
3635 }
3636
3637 del_timer_sync(&mddev->safemode_timer);
3638
3639 invalidate_partition(disk, 0);
3640
9e653b63
N		 3641 switch(mode) {
3642 case 1: /* readonly */
1da177e4 3643 err = -ENXIO;
f91de92e 3644 if (mddev->ro==1)
1da177e4
LT		 3645 goto out;
3646 mddev->ro = 1;
9e653b63
N		 3647 break;
3648 case 0: /* disassemble */
3649 case 2: /* stop */
6b8b3e8a 3650 bitmap_flush(mddev);
a9701a30 3651 md_super_wait(mddev);
1da177e4
LT		 3652 if (mddev->ro)
3653 set_disk_ro(disk, 0);
3654 blk_queue_make_request(mddev->queue, md_fail_request);
3655 mddev->pers->stop(mddev);
d1b5380c
N		 3656 mddev->queue->merge_bvec_fn = NULL;
3657 mddev->queue->unplug_fn = NULL;
041ae52e 3658 mddev->queue->backing_dev_info.congested_fn = NULL;
411036fa
N		 3659 if (mddev->pers->sync_request)
3660 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3661
1da177e4
LT		 3662 module_put(mddev->pers->owner);
3663 mddev->pers = NULL;
0d4ca600
N		 3664
3665 set_capacity(disk, 0);
44ce6294 3666 mddev->changed = 1;
0d4ca600 3667
1da177e4
LT		 3668 if (mddev->ro)
3669 mddev->ro = 0;
3670 }
850b2b42 3671 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 3672 /* mark array as shutdown cleanly */
3673 mddev->in_sync = 1;
850b2b42 3674 md_update_sb(mddev, 1);
1da177e4 3675 }
9e653b63 3676 if (mode == 1)
1da177e4 3677 set_disk_ro(disk, 1);
5fd6c1dc 3678 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3679 }
32a7627c 3680
1da177e4
LT		 3681 /*
3682 * Free resources if final stop
3683 */
9e653b63 3684 if (mode == 0) {
86e6ffdd
N		 3685 mdk_rdev_t *rdev;
3686 struct list_head *tmp;
0d4ca600 3687
1da177e4
LT		 3688 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3689
978f946b
N		 3690 bitmap_destroy(mddev);
3691 if (mddev->bitmap_file) {
acc55e22 3692 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3693 fput(mddev->bitmap_file);
3694 mddev->bitmap_file = NULL;
3695 }
3696 mddev->bitmap_offset = 0;
3697
d089c6af 3698 rdev_for_each(rdev, tmp, mddev)
86e6ffdd
N		 3699 if (rdev->raid_disk >= 0) {
3700 char nm[20];
3701 sprintf(nm, "rd%d", rdev->raid_disk);
3702 sysfs_remove_link(&mddev->kobj, nm);
3703 }
3704
177a99b2 3705 /* make sure all md_delayed_delete calls have finished */
5792a285
N		 3706 flush_scheduled_work();
3707
1da177e4
LT		 3708 export_array(mddev);
3709
3710 mddev->array_size = 0;
9e653b63
N		 3711 mddev->size = 0;
3712 mddev->raid_disks = 0;
a94213b1 3713 mddev->recovery_cp = 0;
c6207277 3714 mddev->resync_max = MaxSector;
08a02ecd 3715 mddev->reshape_position = MaxSector;
e691063a 3716 mddev->external = 0;
1ec4a939 3717 mddev->persistent = 0;
d897dbf9
N		 3718 mddev->level = LEVEL_NONE;
3719 mddev->clevel[0] = 0;
3720 mddev->flags = 0;
3721 mddev->ro = 0;
3722 mddev->metadata_type[0] = 0;
3723 mddev->chunk_size = 0;
3724 mddev->ctime = mddev->utime = 0;
3725 mddev->layout = 0;
3726 mddev->max_disks = 0;
3727 mddev->events = 0;
3728 mddev->delta_disks = 0;
3729 mddev->new_level = LEVEL_NONE;
3730 mddev->new_layout = 0;
3731 mddev->new_chunk = 0;
3732 mddev->curr_resync = 0;
3733 mddev->resync_mismatches = 0;
3734 mddev->suspend_lo = mddev->suspend_hi = 0;
3735 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3736 mddev->recovery = 0;
3737 mddev->in_sync = 0;
3738 mddev->changed = 0;
3739 mddev->degraded = 0;
3740 mddev->barriers_work = 0;
3741 mddev->safemode = 0;
9e653b63 3742
a8a55c38 3743 } else if (mddev->pers)
1da177e4
LT		 3744 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3745 mdname(mddev));
3746 err = 0;
d7603b7e 3747 md_new_event(mddev);
1da177e4
LT		 3748out:
3749 return err;
3750}
3751
fdee8ae4 3752#ifndef MODULE
1da177e4
LT		 3753static void autorun_array(mddev_t *mddev)
3754{
3755 mdk_rdev_t *rdev;
3756 struct list_head *tmp;
3757 int err;
3758
a757e64c 3759 if (list_empty(&mddev->disks))
1da177e4 3760 return;
1da177e4
LT		 3761
3762 printk(KERN_INFO "md: running: ");
3763
d089c6af 3764 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 3765 char b[BDEVNAME_SIZE];
3766 printk("<%s>", bdevname(rdev->bdev,b));
3767 }
3768 printk("\n");
3769
3770 err = do_md_run (mddev);
3771 if (err) {
3772 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3773 do_md_stop (mddev, 0);
3774 }
3775}
3776
3777/*
3778 * lets try to run arrays based on all disks that have arrived
3779 * until now. (those are in pending_raid_disks)
3780 *
3781 * the method: pick the first pending disk, collect all disks with
3782 * the same UUID, remove all from the pending list and put them into
3783 * the 'same_array' list. Then order this list based on superblock
3784 * update time (freshest comes first), kick out 'old' disks and
3785 * compare superblocks. If everything's fine then run it.
3786 *
3787 * If "unit" is allocated, then bump its reference count
3788 */
3789static void autorun_devices(int part)
3790{
1da177e4
LT		 3791 struct list_head *tmp;
3792 mdk_rdev_t *rdev0, *rdev;
3793 mddev_t *mddev;
3794 char b[BDEVNAME_SIZE];
3795
3796 printk(KERN_INFO "md: autorun ...\n");
3797 while (!list_empty(&pending_raid_disks)) {
e8703fe1 3798 int unit;
1da177e4 3799 dev_t dev;
ad01c9e3 3800 LIST_HEAD(candidates);
1da177e4
LT		 3801 rdev0 = list_entry(pending_raid_disks.next,
3802 mdk_rdev_t, same_set);
3803
3804 printk(KERN_INFO "md: considering %s ...\n",
3805 bdevname(rdev0->bdev,b));
3806 INIT_LIST_HEAD(&candidates);
73c34431 3807 rdev_for_each_list(rdev, tmp, pending_raid_disks)
1da177e4
LT		 3808 if (super_90_load(rdev, rdev0, 0) >= 0) {
3809 printk(KERN_INFO "md: adding %s ...\n",
3810 bdevname(rdev->bdev,b));
3811 list_move(&rdev->same_set, &candidates);
3812 }
3813 /*
3814 * now we have a set of devices, with all of them having
3815 * mostly sane superblocks. It's time to allocate the
3816 * mddev.
3817 */
e8703fe1
N		 3818 if (part) {
3819 dev = MKDEV(mdp_major,
3820 rdev0->preferred_minor << MdpMinorShift);
3821 unit = MINOR(dev) >> MdpMinorShift;
3822 } else {
3823 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3824 unit = MINOR(dev);
3825 }
3826 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 3827 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3828 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3829 break;
3830 }
1da177e4
LT		 3831
3832 md_probe(dev, NULL, NULL);
3833 mddev = mddev_find(dev);
3834 if (!mddev) {
3835 printk(KERN_ERR
3836 "md: cannot allocate memory for md drive.\n");
3837 break;
3838 }
3839 if (mddev_lock(mddev))
3840 printk(KERN_WARNING "md: %s locked, cannot run\n",
3841 mdname(mddev));
3842 else if (mddev->raid_disks || mddev->major_version
3843 || !list_empty(&mddev->disks)) {
3844 printk(KERN_WARNING
3845 "md: %s already running, cannot run %s\n",
3846 mdname(mddev), bdevname(rdev0->bdev,b));
3847 mddev_unlock(mddev);
3848 } else {
3849 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 3850 mddev->persistent = 1;
73c34431 3851 rdev_for_each_list(rdev, tmp, candidates) {
1da177e4
LT		 3852 list_del_init(&rdev->same_set);
3853 if (bind_rdev_to_array(rdev, mddev))
3854 export_rdev(rdev);
3855 }
3856 autorun_array(mddev);
3857 mddev_unlock(mddev);
3858 }
3859 /* on success, candidates will be empty, on error
3860 * it won't...
3861 */
73c34431 3862 rdev_for_each_list(rdev, tmp, candidates)
1da177e4
LT		 3863 export_rdev(rdev);
3864 mddev_put(mddev);
3865 }
3866 printk(KERN_INFO "md: ... autorun DONE.\n");
3867}
fdee8ae4 3868#endif /* !MODULE */
1da177e4 3869
1da177e4
LT		 3870static int get_version(void __user * arg)
3871{
3872 mdu_version_t ver;
3873
3874 ver.major = MD_MAJOR_VERSION;
3875 ver.minor = MD_MINOR_VERSION;
3876 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3877
3878 if (copy_to_user(arg, &ver, sizeof(ver)))
3879 return -EFAULT;
3880
3881 return 0;
3882}
3883
3884static int get_array_info(mddev_t * mddev, void __user * arg)
3885{
3886 mdu_array_info_t info;
3887 int nr,working,active,failed,spare;
3888 mdk_rdev_t *rdev;
3889 struct list_head *tmp;
3890
3891 nr=working=active=failed=spare=0;
d089c6af 3892 rdev_for_each(rdev, tmp, mddev) {
1da177e4 3893 nr++;
b2d444d7 3894 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3895 failed++;
3896 else {
3897 working++;
b2d444d7 3898 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3899 active++;
3900 else
3901 spare++;
3902 }
3903 }
3904
3905 info.major_version = mddev->major_version;
3906 info.minor_version = mddev->minor_version;
3907 info.patch_version = MD_PATCHLEVEL_VERSION;
3908 info.ctime = mddev->ctime;
3909 info.level = mddev->level;
3910 info.size = mddev->size;
284ae7ca
N		 3911 if (info.size != mddev->size) /* overflow */
3912 info.size = -1;
1da177e4
LT		 3913 info.nr_disks = nr;
3914 info.raid_disks = mddev->raid_disks;
3915 info.md_minor = mddev->md_minor;
3916 info.not_persistent= !mddev->persistent;
3917
3918 info.utime = mddev->utime;
3919 info.state = 0;
3920 if (mddev->in_sync)
3921 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3922 if (mddev->bitmap && mddev->bitmap_offset)
3923 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3924 info.active_disks = active;
3925 info.working_disks = working;
3926 info.failed_disks = failed;
3927 info.spare_disks = spare;
3928
3929 info.layout = mddev->layout;
3930 info.chunk_size = mddev->chunk_size;
3931
3932 if (copy_to_user(arg, &info, sizeof(info)))
3933 return -EFAULT;
3934
3935 return 0;
3936}
3937
87162a28 3938static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3939{
3940 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3941 char *ptr, *buf = NULL;
3942 int err = -ENOMEM;
3943
2a2275d6
N		 3944 md_allow_write(mddev);
3945
32a7627c
N		 3946 file = kmalloc(sizeof(*file), GFP_KERNEL);
3947 if (!file)
3948 goto out;
3949
3950 /* bitmap disabled, zero the first byte and copy out */
3951 if (!mddev->bitmap || !mddev->bitmap->file) {
3952 file->pathname[0] = '\0';
3953 goto copy_out;
3954 }
3955
3956 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3957 if (!buf)
3958 goto out;
3959
3960 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3961 if (!ptr)
3962 goto out;
3963
3964 strcpy(file->pathname, ptr);
3965
3966copy_out:
3967 err = 0;
3968 if (copy_to_user(arg, file, sizeof(*file)))
3969 err = -EFAULT;
3970out:
3971 kfree(buf);
3972 kfree(file);
3973 return err;
3974}
3975
1da177e4
LT		 3976static int get_disk_info(mddev_t * mddev, void __user * arg)
3977{
3978 mdu_disk_info_t info;
3979 unsigned int nr;
3980 mdk_rdev_t *rdev;
3981
3982 if (copy_from_user(&info, arg, sizeof(info)))
3983 return -EFAULT;
3984
3985 nr = info.number;
3986
3987 rdev = find_rdev_nr(mddev, nr);
3988 if (rdev) {
3989 info.major = MAJOR(rdev->bdev->bd_dev);
3990 info.minor = MINOR(rdev->bdev->bd_dev);
3991 info.raid_disk = rdev->raid_disk;
3992 info.state = 0;
b2d444d7 3993 if (test_bit(Faulty, &rdev->flags))
1da177e4 3994 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3995 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3996 info.state |= (1<<MD_DISK_ACTIVE);
3997 info.state |= (1<<MD_DISK_SYNC);
3998 }
8ddf9efe
N		 3999 if (test_bit(WriteMostly, &rdev->flags))
4000 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 4001 } else {
4002 info.major = info.minor = 0;
4003 info.raid_disk = -1;
4004 info.state = (1<<MD_DISK_REMOVED);
4005 }
4006
4007 if (copy_to_user(arg, &info, sizeof(info)))
4008 return -EFAULT;
4009
4010 return 0;
4011}
4012
4013static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4014{
4015 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4016 mdk_rdev_t *rdev;
4017 dev_t dev = MKDEV(info->major,info->minor);
4018
4019 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4020 return -EOVERFLOW;
4021
4022 if (!mddev->raid_disks) {
4023 int err;
4024 /* expecting a device which has a superblock */
4025 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4026 if (IS_ERR(rdev)) {
4027 printk(KERN_WARNING
4028 "md: md_import_device returned %ld\n",
4029 PTR_ERR(rdev));
4030 return PTR_ERR(rdev);
4031 }
4032 if (!list_empty(&mddev->disks)) {
4033 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4034 mdk_rdev_t, same_set);
4035 int err = super_types[mddev->major_version]
4036 .load_super(rdev, rdev0, mddev->minor_version);
4037 if (err < 0) {
4038 printk(KERN_WARNING
4039 "md: %s has different UUID to %s\n",
4040 bdevname(rdev->bdev,b),
4041 bdevname(rdev0->bdev,b2));
4042 export_rdev(rdev);
4043 return -EINVAL;
4044 }
4045 }
4046 err = bind_rdev_to_array(rdev, mddev);
4047 if (err)
4048 export_rdev(rdev);
4049 return err;
4050 }
4051
4052 /*
4053 * add_new_disk can be used once the array is assembled
4054 * to add "hot spares". They must already have a superblock
4055 * written
4056 */
4057 if (mddev->pers) {
4058 int err;
4059 if (!mddev->pers->hot_add_disk) {
4060 printk(KERN_WARNING
4061 "%s: personality does not support diskops!\n",
4062 mdname(mddev));
4063 return -EINVAL;
4064 }
7b1e35f6
N		 4065 if (mddev->persistent)
4066 rdev = md_import_device(dev, mddev->major_version,
4067 mddev->minor_version);
4068 else
4069 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 4070 if (IS_ERR(rdev)) {
4071 printk(KERN_WARNING
4072 "md: md_import_device returned %ld\n",
4073 PTR_ERR(rdev));
4074 return PTR_ERR(rdev);
4075 }
41158c7e
N		 4076 /* set save_raid_disk if appropriate */
4077 if (!mddev->persistent) {
4078 if (info->state & (1<<MD_DISK_SYNC) &&
4079 info->raid_disk < mddev->raid_disks)
4080 rdev->raid_disk = info->raid_disk;
4081 else
4082 rdev->raid_disk = -1;
4083 } else
4084 super_types[mddev->major_version].
4085 validate_super(mddev, rdev);
4086 rdev->saved_raid_disk = rdev->raid_disk;
4087
b2d444d7 4088 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 4089 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4090 set_bit(WriteMostly, &rdev->flags);
4091
1da177e4
LT		 4092 rdev->raid_disk = -1;
4093 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 4094 if (!err && !mddev->pers->hot_remove_disk) {
4095 /* If there is hot_add_disk but no hot_remove_disk
4096 * then added disks for geometry changes,
4097 * and should be added immediately.
4098 */
4099 super_types[mddev->major_version].
4100 validate_super(mddev, rdev);
4101 err = mddev->pers->hot_add_disk(mddev, rdev);
4102 if (err)
4103 unbind_rdev_from_array(rdev);
4104 }
1da177e4
LT		 4105 if (err)
4106 export_rdev(rdev);
c361777f 4107
17571284 4108 md_update_sb(mddev, 1);
c361777f 4109 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 4110 md_wakeup_thread(mddev->thread);
1da177e4
LT		 4111 return err;
4112 }
4113
4114 /* otherwise, add_new_disk is only allowed
4115 * for major_version==0 superblocks
4116 */
4117 if (mddev->major_version != 0) {
4118 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4119 mdname(mddev));
4120 return -EINVAL;
4121 }
4122
4123 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4124 int err;
4125 rdev = md_import_device (dev, -1, 0);
4126 if (IS_ERR(rdev)) {
4127 printk(KERN_WARNING
4128 "md: error, md_import_device() returned %ld\n",
4129 PTR_ERR(rdev));
4130 return PTR_ERR(rdev);
4131 }
4132 rdev->desc_nr = info->number;
4133 if (info->raid_disk < mddev->raid_disks)
4134 rdev->raid_disk = info->raid_disk;
4135 else
4136 rdev->raid_disk = -1;
4137
1da177e4 4138 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 4139 if (info->state & (1<<MD_DISK_SYNC))
4140 set_bit(In_sync, &rdev->flags);
1da177e4 4141
8ddf9efe
N		 4142 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4143 set_bit(WriteMostly, &rdev->flags);
4144
1da177e4
LT		 4145 if (!mddev->persistent) {
4146 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4147 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4148 } else
4149 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4150 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4151
2bf071bf
N		 4152 err = bind_rdev_to_array(rdev, mddev);
4153 if (err) {
4154 export_rdev(rdev);
4155 return err;
4156 }
1da177e4
LT		 4157 }
4158
4159 return 0;
4160}
4161
4162static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4163{
4164 char b[BDEVNAME_SIZE];
4165 mdk_rdev_t *rdev;
4166
4167 if (!mddev->pers)
4168 return -ENODEV;
4169
4170 rdev = find_rdev(mddev, dev);
4171 if (!rdev)
4172 return -ENXIO;
4173
4174 if (rdev->raid_disk >= 0)
4175 goto busy;
4176
4177 kick_rdev_from_array(rdev);
850b2b42 4178 md_update_sb(mddev, 1);
d7603b7e 4179 md_new_event(mddev);
1da177e4
LT		 4180
4181 return 0;
4182busy:
fdefa4d8 4183 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 4184 bdevname(rdev->bdev,b), mdname(mddev));
4185 return -EBUSY;
4186}
4187
4188static int hot_add_disk(mddev_t * mddev, dev_t dev)
4189{
4190 char b[BDEVNAME_SIZE];
4191 int err;
4192 unsigned int size;
4193 mdk_rdev_t *rdev;
4194
4195 if (!mddev->pers)
4196 return -ENODEV;
4197
4198 if (mddev->major_version != 0) {
4199 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4200 " version-0 superblocks.\n",
4201 mdname(mddev));
4202 return -EINVAL;
4203 }
4204 if (!mddev->pers->hot_add_disk) {
4205 printk(KERN_WARNING
4206 "%s: personality does not support diskops!\n",
4207 mdname(mddev));
4208 return -EINVAL;
4209 }
4210
4211 rdev = md_import_device (dev, -1, 0);
4212 if (IS_ERR(rdev)) {
4213 printk(KERN_WARNING
4214 "md: error, md_import_device() returned %ld\n",
4215 PTR_ERR(rdev));
4216 return -EINVAL;
4217 }
4218
4219 if (mddev->persistent)
4220 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4221 else
4222 rdev->sb_offset =
4223 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4224
4225 size = calc_dev_size(rdev, mddev->chunk_size);
4226 rdev->size = size;
4227
b2d444d7 4228 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4229 printk(KERN_WARNING
4230 "md: can not hot-add faulty %s disk to %s!\n",
4231 bdevname(rdev->bdev,b), mdname(mddev));
4232 err = -EINVAL;
4233 goto abort_export;
4234 }
b2d444d7 4235 clear_bit(In_sync, &rdev->flags);
1da177e4 4236 rdev->desc_nr = -1;
5842730d 4237 rdev->saved_raid_disk = -1;
2bf071bf
N		 4238 err = bind_rdev_to_array(rdev, mddev);
4239 if (err)
4240 goto abort_export;
1da177e4
LT		 4241
4242 /*
4243 * The rest should better be atomic, we can have disk failures
4244 * noticed in interrupt contexts ...
4245 */
4246
4247 if (rdev->desc_nr == mddev->max_disks) {
4248 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4249 mdname(mddev));
4250 err = -EBUSY;
4251 goto abort_unbind_export;
4252 }
4253
4254 rdev->raid_disk = -1;
4255
850b2b42 4256 md_update_sb(mddev, 1);
1da177e4
LT		 4257
4258 /*
4259 * Kick recovery, maybe this spare has to be added to the
4260 * array immediately.
4261 */
4262 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4263 md_wakeup_thread(mddev->thread);
d7603b7e 4264 md_new_event(mddev);
1da177e4
LT		 4265 return 0;
4266
4267abort_unbind_export:
4268 unbind_rdev_from_array(rdev);
4269
4270abort_export:
4271 export_rdev(rdev);
4272 return err;
4273}
4274
32a7627c
N		 4275static int set_bitmap_file(mddev_t *mddev, int fd)
4276{
4277 int err;
4278
36fa3063
N		 4279 if (mddev->pers) {
4280 if (!mddev->pers->quiesce)
4281 return -EBUSY;
4282 if (mddev->recovery || mddev->sync_thread)
4283 return -EBUSY;
4284 /* we should be able to change the bitmap.. */
4285 }
32a7627c 4286
32a7627c 4287
36fa3063
N		 4288 if (fd >= 0) {
4289 if (mddev->bitmap)
4290 return -EEXIST; /* cannot add when bitmap is present */
4291 mddev->bitmap_file = fget(fd);
32a7627c 4292
36fa3063
N		 4293 if (mddev->bitmap_file == NULL) {
4294 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4295 mdname(mddev));
4296 return -EBADF;
4297 }
4298
4299 err = deny_bitmap_write_access(mddev->bitmap_file);
4300 if (err) {
4301 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4302 mdname(mddev));
4303 fput(mddev->bitmap_file);
4304 mddev->bitmap_file = NULL;
4305 return err;
4306 }
a654b9d8 4307 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 4308 } else if (mddev->bitmap == NULL)
4309 return -ENOENT; /* cannot remove what isn't there */
4310 err = 0;
4311 if (mddev->pers) {
4312 mddev->pers->quiesce(mddev, 1);
4313 if (fd >= 0)
4314 err = bitmap_create(mddev);
d7375ab3 4315 if (fd < 0 || err) {
36fa3063 4316 bitmap_destroy(mddev);
d7375ab3
N		 4317 fd = -1; /* make sure to put the file */
4318 }
36fa3063 4319 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 4320 }
4321 if (fd < 0) {
acc55e22
N		 4322 if (mddev->bitmap_file) {
4323 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 4324 fput(mddev->bitmap_file);
acc55e22 4325 }
36fa3063
N		 4326 mddev->bitmap_file = NULL;
4327 }
4328
32a7627c
N		 4329 return err;
4330}
4331
1da177e4
LT		 4332/*
4333 * set_array_info is used two different ways
4334 * The original usage is when creating a new array.
4335 * In this usage, raid_disks is > 0 and it together with
4336 * level, size, not_persistent,layout,chunksize determine the
4337 * shape of the array.
4338 * This will always create an array with a type-0.90.0 superblock.
4339 * The newer usage is when assembling an array.
4340 * In this case raid_disks will be 0, and the major_version field is
4341 * use to determine which style super-blocks are to be found on the devices.
4342 * The minor and patch _version numbers are also kept incase the
4343 * super_block handler wishes to interpret them.
4344 */
4345static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4346{
4347
4348 if (info->raid_disks == 0) {
4349 /* just setting version number for superblock loading */
4350 if (info->major_version < 0 ||
50511da3 4351 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 4352 super_types[info->major_version].name == NULL) {
4353 /* maybe try to auto-load a module? */
4354 printk(KERN_INFO
4355 "md: superblock version %d not known\n",
4356 info->major_version);
4357 return -EINVAL;
4358 }
4359 mddev->major_version = info->major_version;
4360 mddev->minor_version = info->minor_version;
4361 mddev->patch_version = info->patch_version;
3f9d7b0d 4362 mddev->persistent = !info->not_persistent;
1da177e4
LT		 4363 return 0;
4364 }
4365 mddev->major_version = MD_MAJOR_VERSION;
4366 mddev->minor_version = MD_MINOR_VERSION;
4367 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4368 mddev->ctime = get_seconds();
4369
4370 mddev->level = info->level;
17115e03 4371 mddev->clevel[0] = 0;
1da177e4
LT		 4372 mddev->size = info->size;
4373 mddev->raid_disks = info->raid_disks;
4374 /* don't set md_minor, it is determined by which /dev/md* was
4375 * openned
4376 */
4377 if (info->state & (1<<MD_SB_CLEAN))
4378 mddev->recovery_cp = MaxSector;
4379 else
4380 mddev->recovery_cp = 0;
4381 mddev->persistent = ! info->not_persistent;
e691063a 4382 mddev->external = 0;
1da177e4
LT		 4383
4384 mddev->layout = info->layout;
4385 mddev->chunk_size = info->chunk_size;
4386
4387 mddev->max_disks = MD_SB_DISKS;
4388
e691063a
N		 4389 if (mddev->persistent)
4390 mddev->flags = 0;
850b2b42 4391 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 4392
b2a2703c
N		 4393 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4394 mddev->bitmap_offset = 0;
4395
f6705578
N		 4396 mddev->reshape_position = MaxSector;
4397
1da177e4
LT		 4398 /*
4399 * Generate a 128 bit UUID
4400 */
4401 get_random_bytes(mddev->uuid, 16);
4402
f6705578
N		 4403 mddev->new_level = mddev->level;
4404 mddev->new_chunk = mddev->chunk_size;
4405 mddev->new_layout = mddev->layout;
4406 mddev->delta_disks = 0;
4407
1da177e4
LT		 4408 return 0;
4409}
4410
a35b0d69
N		 4411static int update_size(mddev_t *mddev, unsigned long size)
4412{
4413 mdk_rdev_t * rdev;
4414 int rv;
4415 struct list_head *tmp;
8ddeeae5 4416 int fit = (size == 0);
a35b0d69
N		 4417
4418 if (mddev->pers->resize == NULL)
4419 return -EINVAL;
4420 /* The "size" is the amount of each device that is used.
4421 * This can only make sense for arrays with redundancy.
4422 * linear and raid0 always use whatever space is available
4423 * We can only consider changing the size if no resync
4424 * or reconstruction is happening, and if the new size
4425 * is acceptable. It must fit before the sb_offset or,
4426 * if that is <data_offset, it must fit before the
4427 * size of each device.
4428 * If size is zero, we find the largest size that fits.
4429 */
4430 if (mddev->sync_thread)
4431 return -EBUSY;
d089c6af 4432 rdev_for_each(rdev, tmp, mddev) {
a35b0d69 4433 sector_t avail;
01ab5662
N		 4434 avail = rdev->size * 2;
4435
a35b0d69
N		 4436 if (fit && (size == 0 || size > avail/2))
4437 size = avail/2;
4438 if (avail < ((sector_t)size << 1))
4439 return -ENOSPC;
4440 }
4441 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4442 if (!rv) {
4443 struct block_device *bdev;
4444
4445 bdev = bdget_disk(mddev->gendisk, 0);
4446 if (bdev) {
1b1dcc1b 4447 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4448 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4449 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4450 bdput(bdev);
4451 }
4452 }
4453 return rv;
4454}
4455
da943b99
N		 4456static int update_raid_disks(mddev_t *mddev, int raid_disks)
4457{
4458 int rv;
4459 /* change the number of raid disks */
63c70c4f 4460 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4461 return -EINVAL;
4462 if (raid_disks <= 0 ||
4463 raid_disks >= mddev->max_disks)
4464 return -EINVAL;
63c70c4f 4465 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4466 return -EBUSY;
63c70c4f
N		 4467 mddev->delta_disks = raid_disks - mddev->raid_disks;
4468
4469 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4470 return rv;
4471}
4472
4473
1da177e4
LT		 4474/*
4475 * update_array_info is used to change the configuration of an
4476 * on-line array.
4477 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4478 * fields in the info are checked against the array.
4479 * Any differences that cannot be handled will cause an error.
4480 * Normally, only one change can be managed at a time.
4481 */
4482static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4483{
4484 int rv = 0;
4485 int cnt = 0;
36fa3063
N		 4486 int state = 0;
4487
4488 /* calculate expected state,ignoring low bits */
4489 if (mddev->bitmap && mddev->bitmap_offset)
4490 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4491
4492 if (mddev->major_version != info->major_version ||
4493 mddev->minor_version != info->minor_version ||
4494/* mddev->patch_version != info->patch_version || */
4495 mddev->ctime != info->ctime ||
4496 mddev->level != info->level ||
4497/* mddev->layout != info->layout || */
4498 !mddev->persistent != info->not_persistent||
36fa3063
N		 4499 mddev->chunk_size != info->chunk_size ||
4500 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4501 ((state^info->state) & 0xfffffe00)
4502 )
1da177e4
LT		 4503 return -EINVAL;
4504 /* Check there is only one change */
284ae7ca 4505 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4506 if (mddev->raid_disks != info->raid_disks) cnt++;
4507 if (mddev->layout != info->layout) cnt++;
36fa3063 4508 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4509 if (cnt == 0) return 0;
4510 if (cnt > 1) return -EINVAL;
4511
4512 if (mddev->layout != info->layout) {
4513 /* Change layout
4514 * we don't need to do anything at the md level, the
4515 * personality will take care of it all.
4516 */
4517 if (mddev->pers->reconfig == NULL)
4518 return -EINVAL;
4519 else
4520 return mddev->pers->reconfig(mddev, info->layout, -1);
4521 }
284ae7ca 4522 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4523 rv = update_size(mddev, info->size);
4524
da943b99
N		 4525 if (mddev->raid_disks != info->raid_disks)
4526 rv = update_raid_disks(mddev, info->raid_disks);
4527
36fa3063
N		 4528 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4529 if (mddev->pers->quiesce == NULL)
4530 return -EINVAL;
4531 if (mddev->recovery || mddev->sync_thread)
4532 return -EBUSY;
4533 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4534 /* add the bitmap */
4535 if (mddev->bitmap)
4536 return -EEXIST;
4537 if (mddev->default_bitmap_offset == 0)
4538 return -EINVAL;
4539 mddev->bitmap_offset = mddev->default_bitmap_offset;
4540 mddev->pers->quiesce(mddev, 1);
4541 rv = bitmap_create(mddev);
4542 if (rv)
4543 bitmap_destroy(mddev);
4544 mddev->pers->quiesce(mddev, 0);
4545 } else {
4546 /* remove the bitmap */
4547 if (!mddev->bitmap)
4548 return -ENOENT;
4549 if (mddev->bitmap->file)
4550 return -EINVAL;
4551 mddev->pers->quiesce(mddev, 1);
4552 bitmap_destroy(mddev);
4553 mddev->pers->quiesce(mddev, 0);
4554 mddev->bitmap_offset = 0;
4555 }
4556 }
850b2b42 4557 md_update_sb(mddev, 1);
1da177e4
LT		 4558 return rv;
4559}
4560
4561static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4562{
4563 mdk_rdev_t *rdev;
4564
4565 if (mddev->pers == NULL)
4566 return -ENODEV;
4567
4568 rdev = find_rdev(mddev, dev);
4569 if (!rdev)
4570 return -ENODEV;
4571
4572 md_error(mddev, rdev);
4573 return 0;
4574}
4575
a885c8c4
CH		 4576static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4577{
4578 mddev_t *mddev = bdev->bd_disk->private_data;
4579
4580 geo->heads = 2;
4581 geo->sectors = 4;
4582 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4583 return 0;
4584}
4585
1da177e4
LT		 4586static int md_ioctl(struct inode *inode, struct file *file,
4587 unsigned int cmd, unsigned long arg)
4588{
4589 int err = 0;
4590 void __user *argp = (void __user *)arg;
1da177e4
LT		 4591 mddev_t *mddev = NULL;
4592
4593 if (!capable(CAP_SYS_ADMIN))
4594 return -EACCES;
4595
4596 /*
4597 * Commands dealing with the RAID driver but not any
4598 * particular array:
4599 */
4600 switch (cmd)
4601 {
4602 case RAID_VERSION:
4603 err = get_version(argp);
4604 goto done;
4605
4606 case PRINT_RAID_DEBUG:
4607 err = 0;
4608 md_print_devices();
4609 goto done;
4610
4611#ifndef MODULE
4612 case RAID_AUTORUN:
4613 err = 0;
4614 autostart_arrays(arg);
4615 goto done;
4616#endif
4617 default:;
4618 }
4619
4620 /*
4621 * Commands creating/starting a new array:
4622 */
4623
4624 mddev = inode->i_bdev->bd_disk->private_data;
4625
4626 if (!mddev) {
4627 BUG();
4628 goto abort;
4629 }
4630
1da177e4
LT		 4631 err = mddev_lock(mddev);
4632 if (err) {
4633 printk(KERN_INFO
4634 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4635 err, cmd);
4636 goto abort;
4637 }
4638
4639 switch (cmd)
4640 {
4641 case SET_ARRAY_INFO:
4642 {
4643 mdu_array_info_t info;
4644 if (!arg)
4645 memset(&info, 0, sizeof(info));
4646 else if (copy_from_user(&info, argp, sizeof(info))) {
4647 err = -EFAULT;
4648 goto abort_unlock;
4649 }
4650 if (mddev->pers) {
4651 err = update_array_info(mddev, &info);
4652 if (err) {
4653 printk(KERN_WARNING "md: couldn't update"
4654 " array info. %d\n", err);
4655 goto abort_unlock;
4656 }
4657 goto done_unlock;
4658 }
4659 if (!list_empty(&mddev->disks)) {
4660 printk(KERN_WARNING
4661 "md: array %s already has disks!\n",
4662 mdname(mddev));
4663 err = -EBUSY;
4664 goto abort_unlock;
4665 }
4666 if (mddev->raid_disks) {
4667 printk(KERN_WARNING
4668 "md: array %s already initialised!\n",
4669 mdname(mddev));
4670 err = -EBUSY;
4671 goto abort_unlock;
4672 }
4673 err = set_array_info(mddev, &info);
4674 if (err) {
4675 printk(KERN_WARNING "md: couldn't set"
4676 " array info. %d\n", err);
4677 goto abort_unlock;
4678 }
4679 }
4680 goto done_unlock;
4681
4682 default:;
4683 }
4684
4685 /*
4686 * Commands querying/configuring an existing array:
4687 */
32a7627c 4688 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 4689 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 4690 if ((!mddev->raid_disks && !mddev->external)
4691 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4692 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4693 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 4694 err = -ENODEV;
4695 goto abort_unlock;
4696 }
4697
4698 /*
4699 * Commands even a read-only array can execute:
4700 */
4701 switch (cmd)
4702 {
4703 case GET_ARRAY_INFO:
4704 err = get_array_info(mddev, argp);
4705 goto done_unlock;
4706
32a7627c 4707 case GET_BITMAP_FILE:
87162a28 4708 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4709 goto done_unlock;
4710
1da177e4
LT		 4711 case GET_DISK_INFO:
4712 err = get_disk_info(mddev, argp);
4713 goto done_unlock;
4714
4715 case RESTART_ARRAY_RW:
4716 err = restart_array(mddev);
4717 goto done_unlock;
4718
4719 case STOP_ARRAY:
4720 err = do_md_stop (mddev, 0);
4721 goto done_unlock;
4722
4723 case STOP_ARRAY_RO:
4724 err = do_md_stop (mddev, 1);
4725 goto done_unlock;
4726
4727 /*
4728 * We have a problem here : there is no easy way to give a CHS
4729 * virtual geometry. We currently pretend that we have a 2 heads
4730 * 4 sectors (with a BIG number of cylinders...). This drives
4731 * dosfs just mad... ;-)
4732 */
1da177e4
LT		 4733 }
4734
4735 /*
4736 * The remaining ioctls are changing the state of the
f91de92e
N		 4737 * superblock, so we do not allow them on read-only arrays.
4738 * However non-MD ioctls (e.g. get-size) will still come through
4739 * here and hit the 'default' below, so only disallow
4740 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4741 */
f91de92e
N		 4742 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4743 mddev->ro && mddev->pers) {
4744 if (mddev->ro == 2) {
4745 mddev->ro = 0;
4746 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4747 md_wakeup_thread(mddev->thread);
4748
4749 } else {
4750 err = -EROFS;
4751 goto abort_unlock;
4752 }
1da177e4
LT		 4753 }
4754
4755 switch (cmd)
4756 {
4757 case ADD_NEW_DISK:
4758 {
4759 mdu_disk_info_t info;
4760 if (copy_from_user(&info, argp, sizeof(info)))
4761 err = -EFAULT;
4762 else
4763 err = add_new_disk(mddev, &info);
4764 goto done_unlock;
4765 }
4766
4767 case HOT_REMOVE_DISK:
4768 err = hot_remove_disk(mddev, new_decode_dev(arg));
4769 goto done_unlock;
4770
4771 case HOT_ADD_DISK:
4772 err = hot_add_disk(mddev, new_decode_dev(arg));
4773 goto done_unlock;
4774
4775 case SET_DISK_FAULTY:
4776 err = set_disk_faulty(mddev, new_decode_dev(arg));
4777 goto done_unlock;
4778
4779 case RUN_ARRAY:
4780 err = do_md_run (mddev);
4781 goto done_unlock;
4782
32a7627c
N		 4783 case SET_BITMAP_FILE:
4784 err = set_bitmap_file(mddev, (int)arg);
4785 goto done_unlock;
4786
1da177e4 4787 default:
1da177e4
LT		 4788 err = -EINVAL;
4789 goto abort_unlock;
4790 }
4791
4792done_unlock:
4793abort_unlock:
4794 mddev_unlock(mddev);
4795
4796 return err;
4797done:
4798 if (err)
4799 MD_BUG();
4800abort:
4801 return err;
4802}
4803
4804static int md_open(struct inode *inode, struct file *file)
4805{
4806 /*
4807 * Succeed if we can lock the mddev, which confirms that
4808 * it isn't being stopped right now.
4809 */
4810 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4811 int err;
4812
d63a5a74 4813 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
1da177e4
LT		 4814 goto out;
4815
4816 err = 0;
4817 mddev_get(mddev);
4818 mddev_unlock(mddev);
4819
4820 check_disk_change(inode->i_bdev);
4821 out:
4822 return err;
4823}
4824
4825static int md_release(struct inode *inode, struct file * file)
4826{
4827 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4828
52e5f9d1 4829 BUG_ON(!mddev);
1da177e4
LT		 4830 mddev_put(mddev);
4831
4832 return 0;
4833}
4834
44ce6294
LT		 4835static int md_media_changed(struct gendisk *disk)
4836{
4837 mddev_t *mddev = disk->private_data;
4838
4839 return mddev->changed;
4840}
4841
4842static int md_revalidate(struct gendisk *disk)
4843{
4844 mddev_t *mddev = disk->private_data;
4845
4846 mddev->changed = 0;
4847 return 0;
4848}
1da177e4
LT		 4849static struct block_device_operations md_fops =
4850{
4851 .owner = THIS_MODULE,
4852 .open = md_open,
4853 .release = md_release,
4854 .ioctl = md_ioctl,
a885c8c4 4855 .getgeo = md_getgeo,
44ce6294
LT		 4856 .media_changed = md_media_changed,
4857 .revalidate_disk= md_revalidate,
1da177e4
LT		 4858};
4859
75c96f85 4860static int md_thread(void * arg)
1da177e4
LT		 4861{
4862 mdk_thread_t *thread = arg;
4863
1da177e4
LT		 4864 /*
4865 * md_thread is a 'system-thread', it's priority should be very
4866 * high. We avoid resource deadlocks individually in each
4867 * raid personality. (RAID5 does preallocation) We also use RR and
4868 * the very same RT priority as kswapd, thus we will never get
4869 * into a priority inversion deadlock.
4870 *
4871 * we definitely have to have equal or higher priority than
4872 * bdflush, otherwise bdflush will deadlock if there are too
4873 * many dirty RAID5 blocks.
4874 */
1da177e4 4875
6985c43f 4876 allow_signal(SIGKILL);
a6fb0934 4877 while (!kthread_should_stop()) {
1da177e4 4878
93588e22
N		 4879 /* We need to wait INTERRUPTIBLE so that
4880 * we don't add to the load-average.
4881 * That means we need to be sure no signals are
4882 * pending
4883 */
4884 if (signal_pending(current))
4885 flush_signals(current);
4886
4887 wait_event_interruptible_timeout
4888 (thread->wqueue,
4889 test_bit(THREAD_WAKEUP, &thread->flags)
4890 || kthread_should_stop(),
4891 thread->timeout);
1da177e4
LT		 4892
4893 clear_bit(THREAD_WAKEUP, &thread->flags);
4894
787453c2 4895 thread->run(thread->mddev);
1da177e4 4896 }
a6fb0934 4897
1da177e4
LT		 4898 return 0;
4899}
4900
4901void md_wakeup_thread(mdk_thread_t *thread)
4902{
4903 if (thread) {
4904 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4905 set_bit(THREAD_WAKEUP, &thread->flags);
4906 wake_up(&thread->wqueue);
4907 }
4908}
4909
4910mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4911 const char *name)
4912{
4913 mdk_thread_t *thread;
1da177e4 4914
9ffae0cf 4915 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4916 if (!thread)
4917 return NULL;
4918
1da177e4
LT		 4919 init_waitqueue_head(&thread->wqueue);
4920
1da177e4
LT		 4921 thread->run = run;
4922 thread->mddev = mddev;
32a7627c 4923 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4924 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4925 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4926 kfree(thread);
4927 return NULL;
4928 }
1da177e4
LT		 4929 return thread;
4930}
4931
1da177e4
LT		 4932void md_unregister_thread(mdk_thread_t *thread)
4933{
ba25f9dc 4934 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 4935
4936 kthread_stop(thread->tsk);
1da177e4
LT		 4937 kfree(thread);
4938}
4939
4940void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4941{
4942 if (!mddev) {
4943 MD_BUG();
4944 return;
4945 }
4946
b2d444d7 4947 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4948 return;
32a7627c 4949/*
1da177e4
LT		 4950 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4951 mdname(mddev),
4952 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4953 __builtin_return_address(0),__builtin_return_address(1),
4954 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4955*/
d0a0a5ee
AM		 4956 if (!mddev->pers)
4957 return;
1da177e4
LT		 4958 if (!mddev->pers->error_handler)
4959 return;
4960 mddev->pers->error_handler(mddev,rdev);
4961 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4962 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4963 md_wakeup_thread(mddev->thread);
c331eb04 4964 md_new_event_inintr(mddev);
1da177e4
LT		 4965}
4966
4967/* seq_file implementation /proc/mdstat */
4968
4969static void status_unused(struct seq_file *seq)
4970{
4971 int i = 0;
4972 mdk_rdev_t *rdev;
4973 struct list_head *tmp;
4974
4975 seq_printf(seq, "unused devices: ");
4976
73c34431 4977 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
1da177e4
LT		 4978 char b[BDEVNAME_SIZE];
4979 i++;
4980 seq_printf(seq, "%s ",
4981 bdevname(rdev->bdev,b));
4982 }
4983 if (!i)
4984 seq_printf(seq, "<none>");
4985
4986 seq_printf(seq, "\n");
4987}
4988
4989
4990static void status_resync(struct seq_file *seq, mddev_t * mddev)
4991{
4588b42e
N		 4992 sector_t max_blocks, resync, res;
4993 unsigned long dt, db, rt;
4994 int scale;
4995 unsigned int per_milli;
1da177e4
LT		 4996
4997 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4998
4999 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5000 max_blocks = mddev->resync_max_sectors >> 1;
5001 else
5002 max_blocks = mddev->size;
5003
5004 /*
5005 * Should not happen.
5006 */
5007 if (!max_blocks) {
5008 MD_BUG();
5009 return;
5010 }
4588b42e
N		 5011 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5012 * in a sector_t, and (max_blocks>>scale) will fit in a
5013 * u32, as those are the requirements for sector_div.
5014 * Thus 'scale' must be at least 10
5015 */
5016 scale = 10;
5017 if (sizeof(sector_t) > sizeof(unsigned long)) {
5018 while ( max_blocks/2 > (1ULL<<(scale+32)))
5019 scale++;
5020 }
5021 res = (resync>>scale)*1000;
5022 sector_div(res, (u32)((max_blocks>>scale)+1));
5023
5024 per_milli = res;
1da177e4 5025 {
4588b42e 5026 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 5027 seq_printf(seq, "[");
5028 for (i = 0; i < x; i++)
5029 seq_printf(seq, "=");
5030 seq_printf(seq, ">");
5031 for (i = 0; i < y; i++)
5032 seq_printf(seq, ".");
5033 seq_printf(seq, "] ");
5034 }
4588b42e 5035 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 5036 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5037 "reshape" :
61df9d91
N		 5038 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5039 "check" :
5040 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5041 "resync" : "recovery"))),
5042 per_milli/10, per_milli % 10,
4588b42e
N		 5043 (unsigned long long) resync,
5044 (unsigned long long) max_blocks);
1da177e4
LT		 5045
5046 /*
5047 * We do not want to overflow, so the order of operands and
5048 * the * 100 / 100 trick are important. We do a +1 to be
5049 * safe against division by zero. We only estimate anyway.
5050 *
5051 * dt: time from mark until now
5052 * db: blocks written from mark until now
5053 * rt: remaining time
5054 */
5055 dt = ((jiffies - mddev->resync_mark) / HZ);
5056 if (!dt) dt++;
ff4e8d9a
N		 5057 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5058 - mddev->resync_mark_cnt;
5059 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 5060
5061 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5062
ff4e8d9a 5063 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 5064}
5065
5066static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5067{
5068 struct list_head *tmp;
5069 loff_t l = *pos;
5070 mddev_t *mddev;
5071
5072 if (l >= 0x10000)
5073 return NULL;
5074 if (!l--)
5075 /* header */
5076 return (void*)1;
5077
5078 spin_lock(&all_mddevs_lock);
5079 list_for_each(tmp,&all_mddevs)
5080 if (!l--) {
5081 mddev = list_entry(tmp, mddev_t, all_mddevs);
5082 mddev_get(mddev);
5083 spin_unlock(&all_mddevs_lock);
5084 return mddev;
5085 }
5086 spin_unlock(&all_mddevs_lock);
5087 if (!l--)
5088 return (void*)2;/* tail */
5089 return NULL;
5090}
5091
5092static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5093{
5094 struct list_head *tmp;
5095 mddev_t *next_mddev, *mddev = v;
5096
5097 ++*pos;
5098 if (v == (void*)2)
5099 return NULL;
5100
5101 spin_lock(&all_mddevs_lock);
5102 if (v == (void*)1)
5103 tmp = all_mddevs.next;
5104 else
5105 tmp = mddev->all_mddevs.next;
5106 if (tmp != &all_mddevs)
5107 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5108 else {
5109 next_mddev = (void*)2;
5110 *pos = 0x10000;
5111 }
5112 spin_unlock(&all_mddevs_lock);
5113
5114 if (v != (void*)1)
5115 mddev_put(mddev);
5116 return next_mddev;
5117
5118}
5119
5120static void md_seq_stop(struct seq_file *seq, void *v)
5121{
5122 mddev_t *mddev = v;
5123
5124 if (mddev && v != (void*)1 && v != (void*)2)
5125 mddev_put(mddev);
5126}
5127
d7603b7e
N		 5128struct mdstat_info {
5129 int event;
5130};
5131
1da177e4
LT		 5132static int md_seq_show(struct seq_file *seq, void *v)
5133{
5134 mddev_t *mddev = v;
5135 sector_t size;
5136 struct list_head *tmp2;
5137 mdk_rdev_t *rdev;
d7603b7e 5138 struct mdstat_info *mi = seq->private;
32a7627c 5139 struct bitmap *bitmap;
1da177e4
LT		 5140
5141 if (v == (void*)1) {
2604b703 5142 struct mdk_personality *pers;
1da177e4
LT		 5143 seq_printf(seq, "Personalities : ");
5144 spin_lock(&pers_lock);
2604b703
N		 5145 list_for_each_entry(pers, &pers_list, list)
5146 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 5147
5148 spin_unlock(&pers_lock);
5149 seq_printf(seq, "\n");
d7603b7e 5150 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 5151 return 0;
5152 }
5153 if (v == (void*)2) {
5154 status_unused(seq);
5155 return 0;
5156 }
5157
5dc5cf7d 5158 if (mddev_lock(mddev) < 0)
1da177e4 5159 return -EINTR;
5dc5cf7d 5160
1da177e4
LT		 5161 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5162 seq_printf(seq, "%s : %sactive", mdname(mddev),
5163 mddev->pers ? "" : "in");
5164 if (mddev->pers) {
f91de92e 5165 if (mddev->ro==1)
1da177e4 5166 seq_printf(seq, " (read-only)");
f91de92e 5167 if (mddev->ro==2)
52720ae7 5168 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 5169 seq_printf(seq, " %s", mddev->pers->name);
5170 }
5171
5172 size = 0;
d089c6af 5173 rdev_for_each(rdev, tmp2, mddev) {
1da177e4
LT		 5174 char b[BDEVNAME_SIZE];
5175 seq_printf(seq, " %s[%d]",
5176 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 5177 if (test_bit(WriteMostly, &rdev->flags))
5178 seq_printf(seq, "(W)");
b2d444d7 5179 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5180 seq_printf(seq, "(F)");
5181 continue;
b325a32e
N		 5182 } else if (rdev->raid_disk < 0)
5183 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 5184 size += rdev->size;
5185 }
5186
5187 if (!list_empty(&mddev->disks)) {
5188 if (mddev->pers)
5189 seq_printf(seq, "\n %llu blocks",
5190 (unsigned long long)mddev->array_size);
5191 else
5192 seq_printf(seq, "\n %llu blocks",
5193 (unsigned long long)size);
5194 }
1cd6bf19
N		 5195 if (mddev->persistent) {
5196 if (mddev->major_version != 0 ||
5197 mddev->minor_version != 90) {
5198 seq_printf(seq," super %d.%d",
5199 mddev->major_version,
5200 mddev->minor_version);
5201 }
e691063a
N		 5202 } else if (mddev->external)
5203 seq_printf(seq, " super external:%s",
5204 mddev->metadata_type);
5205 else
1cd6bf19 5206 seq_printf(seq, " super non-persistent");
1da177e4
LT		 5207
5208 if (mddev->pers) {
5209 mddev->pers->status (seq, mddev);
5210 seq_printf(seq, "\n ");
8e1b39d6
N		 5211 if (mddev->pers->sync_request) {
5212 if (mddev->curr_resync > 2) {
5213 status_resync (seq, mddev);
5214 seq_printf(seq, "\n ");
5215 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5216 seq_printf(seq, "\tresync=DELAYED\n ");
5217 else if (mddev->recovery_cp < MaxSector)
5218 seq_printf(seq, "\tresync=PENDING\n ");
5219 }
32a7627c
N		 5220 } else
5221 seq_printf(seq, "\n ");
5222
5223 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 5224 unsigned long chunk_kb;
5225 unsigned long flags;
32a7627c
N		 5226 spin_lock_irqsave(&bitmap->lock, flags);
5227 chunk_kb = bitmap->chunksize >> 10;
5228 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5229 "%lu%s chunk",
5230 bitmap->pages - bitmap->missing_pages,
5231 bitmap->pages,
5232 (bitmap->pages - bitmap->missing_pages)
5233 << (PAGE_SHIFT - 10),
5234 chunk_kb ? chunk_kb : bitmap->chunksize,
5235 chunk_kb ? "KB" : "B");
78d742d8
N		 5236 if (bitmap->file) {
5237 seq_printf(seq, ", file: ");
c32c2f63 5238 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 5239 }
78d742d8 5240
32a7627c
N		 5241 seq_printf(seq, "\n");
5242 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 5243 }
5244
5245 seq_printf(seq, "\n");
5246 }
5247 mddev_unlock(mddev);
5248
5249 return 0;
5250}
5251
5252static struct seq_operations md_seq_ops = {
5253 .start = md_seq_start,
5254 .next = md_seq_next,
5255 .stop = md_seq_stop,
5256 .show = md_seq_show,
5257};
5258
5259static int md_seq_open(struct inode *inode, struct file *file)
5260{
5261 int error;
d7603b7e
N		 5262 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5263 if (mi == NULL)
5264 return -ENOMEM;
1da177e4
LT		 5265
5266 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 5267 if (error)
5268 kfree(mi);
5269 else {
5270 struct seq_file *p = file->private_data;
5271 p->private = mi;
5272 mi->event = atomic_read(&md_event_count);
5273 }
1da177e4
LT		 5274 return error;
5275}
5276
d7603b7e
N		 5277static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5278{
5279 struct seq_file *m = filp->private_data;
5280 struct mdstat_info *mi = m->private;
5281 int mask;
5282
5283 poll_wait(filp, &md_event_waiters, wait);
5284
5285 /* always allow read */
5286 mask = POLLIN | POLLRDNORM;
5287
5288 if (mi->event != atomic_read(&md_event_count))
5289 mask |= POLLERR | POLLPRI;
5290 return mask;
5291}
5292
fa027c2a 5293static const struct file_operations md_seq_fops = {
e24650c2 5294 .owner = THIS_MODULE,
1da177e4
LT		 5295 .open = md_seq_open,
5296 .read = seq_read,
5297 .llseek = seq_lseek,
c3f94b40 5298 .release = seq_release_private,
d7603b7e 5299 .poll = mdstat_poll,
1da177e4
LT		 5300};
5301
2604b703 5302int register_md_personality(struct mdk_personality *p)
1da177e4 5303{
1da177e4 5304 spin_lock(&pers_lock);
2604b703
N		 5305 list_add_tail(&p->list, &pers_list);
5306 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 5307 spin_unlock(&pers_lock);
5308 return 0;
5309}
5310
2604b703 5311int unregister_md_personality(struct mdk_personality *p)
1da177e4 5312{
2604b703 5313 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 5314 spin_lock(&pers_lock);
2604b703 5315 list_del_init(&p->list);
1da177e4
LT		 5316 spin_unlock(&pers_lock);
5317 return 0;
5318}
5319
5320static int is_mddev_idle(mddev_t *mddev)
5321{
5322 mdk_rdev_t * rdev;
5323 struct list_head *tmp;
5324 int idle;
713f6ab1 5325 long curr_events;
1da177e4
LT		 5326
5327 idle = 1;
d089c6af 5328 rdev_for_each(rdev, tmp, mddev) {
1da177e4 5329 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 5330 curr_events = disk_stat_read(disk, sectors[0]) +
5331 disk_stat_read(disk, sectors[1]) -
1da177e4 5332 atomic_read(&disk->sync_io);
713f6ab1
N		 5333 /* sync IO will cause sync_io to increase before the disk_stats
5334 * as sync_io is counted when a request starts, and
5335 * disk_stats is counted when it completes.
5336 * So resync activity will cause curr_events to be smaller than
5337 * when there was no such activity.
5338 * non-sync IO will cause disk_stat to increase without
5339 * increasing sync_io so curr_events will (eventually)
5340 * be larger than it was before. Once it becomes
5341 * substantially larger, the test below will cause
5342 * the array to appear non-idle, and resync will slow
5343 * down.
5344 * If there is a lot of outstanding resync activity when
5345 * we set last_event to curr_events, then all that activity
5346 * completing might cause the array to appear non-idle
5347 * and resync will be slowed down even though there might
5348 * not have been non-resync activity. This will only
5349 * happen once though. 'last_events' will soon reflect
5350 * the state where there is little or no outstanding
5351 * resync requests, and further resync activity will
5352 * always make curr_events less than last_events.
c0e48521 5353 *
1da177e4 5354 */
713f6ab1 5355 if (curr_events - rdev->last_events > 4096) {
1da177e4
LT		 5356 rdev->last_events = curr_events;
5357 idle = 0;
5358 }
5359 }
5360 return idle;
5361}
5362
5363void md_done_sync(mddev_t *mddev, int blocks, int ok)
5364{
5365 /* another "blocks" (512byte) blocks have been synced */
5366 atomic_sub(blocks, &mddev->recovery_active);
5367 wake_up(&mddev->recovery_wait);
5368 if (!ok) {
5369 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5370 md_wakeup_thread(mddev->thread);
5371 // stop recovery, signal do_sync ....
5372 }
5373}
5374
5375
06d91a5f
N		 5376/* md_write_start(mddev, bi)
5377 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 5378 * in superblock) before writing, schedule a superblock update
5379 * and wait for it to complete.
06d91a5f 5380 */
3d310eb7 5381void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 5382{
06d91a5f 5383 if (bio_data_dir(bi) != WRITE)
3d310eb7 5384 return;
06d91a5f 5385
f91de92e
N		 5386 BUG_ON(mddev->ro == 1);
5387 if (mddev->ro == 2) {
5388 /* need to switch to read/write */
5389 mddev->ro = 0;
5390 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5391 md_wakeup_thread(mddev->thread);
25156198 5392 md_wakeup_thread(mddev->sync_thread);
f91de92e 5393 }
06d91a5f 5394 atomic_inc(&mddev->writes_pending);
06d91a5f 5395 if (mddev->in_sync) {
a9701a30 5396 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5397 if (mddev->in_sync) {
5398 mddev->in_sync = 0;
850b2b42 5399 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7
N		 5400 md_wakeup_thread(mddev->thread);
5401 }
a9701a30 5402 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5403 }
850b2b42 5404 wait_event(mddev->sb_wait, mddev->flags==0);
1da177e4
LT		 5405}
5406
5407void md_write_end(mddev_t *mddev)
5408{
5409 if (atomic_dec_and_test(&mddev->writes_pending)) {
5410 if (mddev->safemode == 2)
5411 md_wakeup_thread(mddev->thread);
16f17b39 5412 else if (mddev->safemode_delay)
1da177e4
LT		 5413 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5414 }
5415}
5416
2a2275d6
N		 5417/* md_allow_write(mddev)
5418 * Calling this ensures that the array is marked 'active' so that writes
5419 * may proceed without blocking. It is important to call this before
5420 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5421 * Must be called with mddev_lock held.
5422 */
5423void md_allow_write(mddev_t *mddev)
5424{
5425 if (!mddev->pers)
5426 return;
5427 if (mddev->ro)
5428 return;
5429
5430 spin_lock_irq(&mddev->write_lock);
5431 if (mddev->in_sync) {
5432 mddev->in_sync = 0;
5433 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5434 if (mddev->safemode_delay &&
5435 mddev->safemode == 0)
5436 mddev->safemode = 1;
5437 spin_unlock_irq(&mddev->write_lock);
5438 md_update_sb(mddev, 0);
5439 } else
5440 spin_unlock_irq(&mddev->write_lock);
5441}
5442EXPORT_SYMBOL_GPL(md_allow_write);
5443
75c96f85 5444static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5445
5446#define SYNC_MARKS 10
5447#define SYNC_MARK_STEP (3*HZ)
29269553 5448void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5449{
5450 mddev_t *mddev2;
5451 unsigned int currspeed = 0,
5452 window;
57afd89f 5453 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5454 unsigned long mark[SYNC_MARKS];
5455 sector_t mark_cnt[SYNC_MARKS];
5456 int last_mark,m;
5457 struct list_head *tmp;
5458 sector_t last_check;
57afd89f 5459 int skipped = 0;
5fd6c1dc
N		 5460 struct list_head *rtmp;
5461 mdk_rdev_t *rdev;
61df9d91 5462 char *desc;
1da177e4
LT		 5463
5464 /* just incase thread restarts... */
5465 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5466 return;
5fd6c1dc
N		 5467 if (mddev->ro) /* never try to sync a read-only array */
5468 return;
1da177e4 5469
61df9d91
N		 5470 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5471 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5472 desc = "data-check";
5473 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5474 desc = "requested-resync";
5475 else
5476 desc = "resync";
5477 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5478 desc = "reshape";
5479 else
5480 desc = "recovery";
5481
1da177e4
LT		 5482 /* we overload curr_resync somewhat here.
5483 * 0 == not engaged in resync at all
5484 * 2 == checking that there is no conflict with another sync
5485 * 1 == like 2, but have yielded to allow conflicting resync to
5486 * commense
5487 * other == active in resync - this many blocks
5488 *
5489 * Before starting a resync we must have set curr_resync to
5490 * 2, and then checked that every "conflicting" array has curr_resync
5491 * less than ours. When we find one that is the same or higher
5492 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5493 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5494 * This will mean we have to start checking from the beginning again.
5495 *
5496 */
5497
5498 do {
5499 mddev->curr_resync = 2;
5500
5501 try_again:
787453c2 5502 if (kthread_should_stop()) {
6985c43f 5503 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5504 goto skip;
5505 }
29ac4aa3 5506 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 5507 if (mddev2 == mddev)
5508 continue;
5509 if (mddev2->curr_resync &&
5510 match_mddev_units(mddev,mddev2)) {
5511 DEFINE_WAIT(wq);
5512 if (mddev < mddev2 && mddev->curr_resync == 2) {
5513 /* arbitrarily yield */
5514 mddev->curr_resync = 1;
5515 wake_up(&resync_wait);
5516 }
5517 if (mddev > mddev2 && mddev->curr_resync == 1)
5518 /* no need to wait here, we can wait the next
5519 * time 'round when curr_resync == 2
5520 */
5521 continue;
787453c2
N		 5522 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5523 if (!kthread_should_stop() &&
8712e553 5524 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 5525 printk(KERN_INFO "md: delaying %s of %s"
5526 " until %s has finished (they"
1da177e4 5527 " share one or more physical units)\n",
61df9d91 5528 desc, mdname(mddev), mdname(mddev2));
1da177e4
LT		 5529 mddev_put(mddev2);
5530 schedule();
5531 finish_wait(&resync_wait, &wq);
5532 goto try_again;
5533 }
5534 finish_wait(&resync_wait, &wq);
5535 }
5536 }
5537 } while (mddev->curr_resync < 2);
5538
5fd6c1dc 5539 j = 0;
9d88883e 5540 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5541 /* resync follows the size requested by the personality,
57afd89f 5542 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5543 */
5544 max_sectors = mddev->resync_max_sectors;
9d88883e 5545 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5546 /* we don't use the checkpoint if there's a bitmap */
5547 if (!mddev->bitmap &&
5548 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5549 j = mddev->recovery_cp;
ccfcc3c1
N		 5550 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5551 max_sectors = mddev->size << 1;
5fd6c1dc 5552 else {
1da177e4
LT		 5553 /* recovery follows the physical size of devices */
5554 max_sectors = mddev->size << 1;
5fd6c1dc 5555 j = MaxSector;
d089c6af 5556 rdev_for_each(rdev, rtmp, mddev)
5fd6c1dc
N		 5557 if (rdev->raid_disk >= 0 &&
5558 !test_bit(Faulty, &rdev->flags) &&
5559 !test_bit(In_sync, &rdev->flags) &&
5560 rdev->recovery_offset < j)
5561 j = rdev->recovery_offset;
5562 }
1da177e4 5563
61df9d91
N		 5564 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5565 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5566 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 5567 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 5568 "(but not more than %d KB/sec) for %s.\n",
5569 speed_max(mddev), desc);
1da177e4
LT		 5570
5571 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5572
57afd89f 5573 io_sectors = 0;
1da177e4
LT		 5574 for (m = 0; m < SYNC_MARKS; m++) {
5575 mark[m] = jiffies;
57afd89f 5576 mark_cnt[m] = io_sectors;
1da177e4
LT		 5577 }
5578 last_mark = 0;
5579 mddev->resync_mark = mark[last_mark];
5580 mddev->resync_mark_cnt = mark_cnt[last_mark];
5581
5582 /*
5583 * Tune reconstruction:
5584 */
5585 window = 32*(PAGE_SIZE/512);
5586 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5587 window/2,(unsigned long long) max_sectors/2);
5588
5589 atomic_set(&mddev->recovery_active, 0);
5590 init_waitqueue_head(&mddev->recovery_wait);
5591 last_check = 0;
5592
5593 if (j>2) {
5594 printk(KERN_INFO
61df9d91
N		 5595 "md: resuming %s of %s from checkpoint.\n",
5596 desc, mdname(mddev));
1da177e4
LT		 5597 mddev->curr_resync = j;
5598 }
5599
5600 while (j < max_sectors) {
57afd89f 5601 sector_t sectors;
1da177e4 5602
57afd89f 5603 skipped = 0;
c6207277
N		 5604 if (j >= mddev->resync_max) {
5605 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5606 wait_event(mddev->recovery_wait,
5607 mddev->resync_max > j
5608 || kthread_should_stop());
5609 }
5610 if (kthread_should_stop())
5611 goto interrupted;
57afd89f 5612 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 5613 currspeed < speed_min(mddev));
57afd89f 5614 if (sectors == 0) {
1da177e4
LT		 5615 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5616 goto out;
5617 }
57afd89f
N		 5618
5619 if (!skipped) { /* actual IO requested */
5620 io_sectors += sectors;
5621 atomic_add(sectors, &mddev->recovery_active);
5622 }
5623
1da177e4
LT		 5624 j += sectors;
5625 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5626 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5627 if (last_check == 0)
5628 /* this is the earliers that rebuilt will be
5629 * visible in /proc/mdstat
5630 */
5631 md_new_event(mddev);
57afd89f
N		 5632
5633 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5634 continue;
5635
57afd89f 5636 last_check = io_sectors;
1da177e4
LT		 5637
5638 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5639 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5640 break;
5641
5642 repeat:
5643 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5644 /* step marks */
5645 int next = (last_mark+1) % SYNC_MARKS;
5646
5647 mddev->resync_mark = mark[next];
5648 mddev->resync_mark_cnt = mark_cnt[next];
5649 mark[next] = jiffies;
57afd89f 5650 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5651 last_mark = next;
5652 }
5653
5654
c6207277
N		 5655 if (kthread_should_stop())
5656 goto interrupted;
5657
1da177e4
LT		 5658
5659 /*
5660 * this loop exits only if either when we are slower than
5661 * the 'hard' speed limit, or the system was IO-idle for
5662 * a jiffy.
5663 * the system might be non-idle CPU-wise, but we only care
5664 * about not overloading the IO subsystem. (things like an
5665 * e2fsck being done on the RAID array should execute fast)
5666 */
2ad8b1ef 5667 blk_unplug(mddev->queue);
1da177e4
LT		 5668 cond_resched();
5669
57afd89f
N		 5670 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5671 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5672
88202a0c
N		 5673 if (currspeed > speed_min(mddev)) {
5674 if ((currspeed > speed_max(mddev)) ||
1da177e4 5675 !is_mddev_idle(mddev)) {
c0e48521 5676 msleep(500);
1da177e4
LT		 5677 goto repeat;
5678 }
5679 }
5680 }
61df9d91 5681 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 5682 /*
5683 * this also signals 'finished resyncing' to md_stop
5684 */
5685 out:
2ad8b1ef 5686 blk_unplug(mddev->queue);
1da177e4
LT		 5687
5688 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5689
5690 /* tell personality that we are finished */
57afd89f 5691 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5692
5693 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1 5694 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5695 mddev->curr_resync > 2) {
5696 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5697 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5698 if (mddev->curr_resync >= mddev->recovery_cp) {
5699 printk(KERN_INFO
61df9d91
N		 5700 "md: checkpointing %s of %s.\n",
5701 desc, mdname(mddev));
5fd6c1dc
N		 5702 mddev->recovery_cp = mddev->curr_resync;
5703 }
5704 } else
5705 mddev->recovery_cp = MaxSector;
5706 } else {
5707 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5708 mddev->curr_resync = MaxSector;
d089c6af 5709 rdev_for_each(rdev, rtmp, mddev)
5fd6c1dc
N		 5710 if (rdev->raid_disk >= 0 &&
5711 !test_bit(Faulty, &rdev->flags) &&
5712 !test_bit(In_sync, &rdev->flags) &&
5713 rdev->recovery_offset < mddev->curr_resync)
5714 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 5715 }
1da177e4 5716 }
17571284 5717 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5718
1da177e4
LT		 5719 skip:
5720 mddev->curr_resync = 0;
c6207277
N		 5721 mddev->resync_max = MaxSector;
5722 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 5723 wake_up(&resync_wait);
5724 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5725 md_wakeup_thread(mddev->thread);
c6207277
N		 5726 return;
5727
5728 interrupted:
5729 /*
5730 * got a signal, exit.
5731 */
5732 printk(KERN_INFO
5733 "md: md_do_sync() got signal ... exiting\n");
5734 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5735 goto out;
5736
1da177e4 5737}
29269553 5738EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5739
5740
b4c4c7b8
N		 5741static int remove_and_add_spares(mddev_t *mddev)
5742{
5743 mdk_rdev_t *rdev;
5744 struct list_head *rtmp;
5745 int spares = 0;
5746
d089c6af 5747 rdev_for_each(rdev, rtmp, mddev)
b4c4c7b8 5748 if (rdev->raid_disk >= 0 &&
c303da6d 5749 !mddev->external &&
b4c4c7b8
N		 5750 (test_bit(Faulty, &rdev->flags) ||
5751 ! test_bit(In_sync, &rdev->flags)) &&
5752 atomic_read(&rdev->nr_pending)==0) {
5753 if (mddev->pers->hot_remove_disk(
5754 mddev, rdev->raid_disk)==0) {
5755 char nm[20];
5756 sprintf(nm,"rd%d", rdev->raid_disk);
5757 sysfs_remove_link(&mddev->kobj, nm);
5758 rdev->raid_disk = -1;
5759 }
5760 }
5761
5762 if (mddev->degraded) {
d089c6af 5763 rdev_for_each(rdev, rtmp, mddev)
b4c4c7b8
N		 5764 if (rdev->raid_disk < 0
5765 && !test_bit(Faulty, &rdev->flags)) {
5766 rdev->recovery_offset = 0;
5767 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5768 char nm[20];
5769 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 5770 if (sysfs_create_link(&mddev->kobj,
5771 &rdev->kobj, nm))
5772 printk(KERN_WARNING
5773 "md: cannot register "
5774 "%s for %s\n",
5775 nm, mdname(mddev));
b4c4c7b8
N		 5776 spares++;
5777 md_new_event(mddev);
5778 } else
5779 break;
5780 }
5781 }
5782 return spares;
5783}
1da177e4
LT		 5784/*
5785 * This routine is regularly called by all per-raid-array threads to
5786 * deal with generic issues like resync and super-block update.
5787 * Raid personalities that don't have a thread (linear/raid0) do not
5788 * need this as they never do any recovery or update the superblock.
5789 *
5790 * It does not do any resync itself, but rather "forks" off other threads
5791 * to do that as needed.
5792 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5793 * "->recovery" and create a thread at ->sync_thread.
5794 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5795 * and wakeups up this thread which will reap the thread and finish up.
5796 * This thread also removes any faulty devices (with nr_pending == 0).
5797 *
5798 * The overall approach is:
5799 * 1/ if the superblock needs updating, update it.
5800 * 2/ If a recovery thread is running, don't do anything else.
5801 * 3/ If recovery has finished, clean up, possibly marking spares active.
5802 * 4/ If there are any faulty devices, remove them.
5803 * 5/ If array is degraded, try to add spares devices
5804 * 6/ If array has spares or is not in-sync, start a resync thread.
5805 */
5806void md_check_recovery(mddev_t *mddev)
5807{
5808 mdk_rdev_t *rdev;
5809 struct list_head *rtmp;
5810
5811
5f40402d
N		 5812 if (mddev->bitmap)
5813 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5814
5815 if (mddev->ro)
5816 return;
fca4d848
N		 5817
5818 if (signal_pending(current)) {
5819 if (mddev->pers->sync_request) {
5820 printk(KERN_INFO "md: %s in immediate safe mode\n",
5821 mdname(mddev));
5822 mddev->safemode = 2;
5823 }
5824 flush_signals(current);
5825 }
5826
1da177e4 5827 if ( ! (
e691063a 5828 (mddev->flags && !mddev->external) ||
1da177e4 5829 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5830 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5831 (mddev->safemode == 1) ||
5832 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5833 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5834 ))
5835 return;
fca4d848 5836
df5b89b3 5837 if (mddev_trylock(mddev)) {
b4c4c7b8 5838 int spares = 0;
fca4d848 5839
a9701a30 5840 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5841 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5842 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5843 mddev->in_sync = 1;
e691063a
N		 5844 if (mddev->persistent)
5845 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
fca4d848
N		 5846 }
5847 if (mddev->safemode == 1)
5848 mddev->safemode = 0;
a9701a30 5849 spin_unlock_irq(&mddev->write_lock);
fca4d848 5850
850b2b42
N		 5851 if (mddev->flags)
5852 md_update_sb(mddev, 0);
06d91a5f 5853
06d91a5f 5854
1da177e4
LT		 5855 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5856 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5857 /* resync/recovery still happening */
5858 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5859 goto unlock;
5860 }
5861 if (mddev->sync_thread) {
5862 /* resync has finished, collect result */
5863 md_unregister_thread(mddev->sync_thread);
5864 mddev->sync_thread = NULL;
5865 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5866 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5867 /* success...*/
5868 /* activate any spares */
5869 mddev->pers->spare_active(mddev);
5870 }
850b2b42 5871 md_update_sb(mddev, 1);
41158c7e
N		 5872
5873 /* if array is no-longer degraded, then any saved_raid_disk
5874 * information must be scrapped
5875 */
5876 if (!mddev->degraded)
d089c6af 5877 rdev_for_each(rdev, rtmp, mddev)
41158c7e
N		 5878 rdev->saved_raid_disk = -1;
5879
1da177e4
LT		 5880 mddev->recovery = 0;
5881 /* flag recovery needed just to double check */
5882 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5883 md_new_event(mddev);
1da177e4
LT		 5884 goto unlock;
5885 }
24dd469d
N		 5886 /* Clear some bits that don't mean anything, but
5887 * might be left set
5888 */
5889 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5890 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5891 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5892 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5893
5fd6c1dc
N		 5894 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5895 goto unlock;
1da177e4
LT		 5896 /* no recovery is running.
5897 * remove any failed drives, then
5898 * add spares if possible.
5899 * Spare are also removed and re-added, to allow
5900 * the personality to fail the re-add.
5901 */
1da177e4 5902
b4c4c7b8
N		 5903 if (mddev->reshape_position != MaxSector) {
5904 if (mddev->pers->check_reshape(mddev) != 0)
5905 /* Cannot proceed */
5906 goto unlock;
5907 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5908 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 5909 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5910 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5911 } else if (mddev->recovery_cp < MaxSector) {
5912 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5913 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5914 /* nothing to be done ... */
1da177e4 5915 goto unlock;
24dd469d 5916
1da177e4
LT		 5917 if (mddev->pers->sync_request) {
5918 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5919 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5920 /* We are adding a device or devices to an array
5921 * which has the bitmap stored on all devices.
5922 * So make sure all bitmap pages get written
5923 */
5924 bitmap_write_all(mddev->bitmap);
5925 }
1da177e4
LT		 5926 mddev->sync_thread = md_register_thread(md_do_sync,
5927 mddev,
5928 "%s_resync");
5929 if (!mddev->sync_thread) {
5930 printk(KERN_ERR "%s: could not start resync"
5931 " thread...\n",
5932 mdname(mddev));
5933 /* leave the spares where they are, it shouldn't hurt */
5934 mddev->recovery = 0;
d7603b7e 5935 } else
1da177e4 5936 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5937 md_new_event(mddev);
1da177e4
LT		 5938 }
5939 unlock:
5940 mddev_unlock(mddev);
5941 }
5942}
5943
75c96f85
AB		 5944static int md_notify_reboot(struct notifier_block *this,
5945 unsigned long code, void *x)
1da177e4
LT		 5946{
5947 struct list_head *tmp;
5948 mddev_t *mddev;
5949
5950 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5951
5952 printk(KERN_INFO "md: stopping all md devices.\n");
5953
29ac4aa3 5954 for_each_mddev(mddev, tmp)
c71d4887 5955 if (mddev_trylock(mddev)) {
1da177e4 5956 do_md_stop (mddev, 1);
c71d4887
NB		 5957 mddev_unlock(mddev);
5958 }
1da177e4
LT		 5959 /*
5960 * certain more exotic SCSI devices are known to be
5961 * volatile wrt too early system reboots. While the
5962 * right place to handle this issue is the given
5963 * driver, we do want to have a safe RAID driver ...
5964 */
5965 mdelay(1000*1);
5966 }
5967 return NOTIFY_DONE;
5968}
5969
75c96f85 5970static struct notifier_block md_notifier = {
1da177e4
LT		 5971 .notifier_call = md_notify_reboot,
5972 .next = NULL,
5973 .priority = INT_MAX, /* before any real devices */
5974};
5975
5976static void md_geninit(void)
5977{
1da177e4
LT		 5978 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5979
c7705f34 5980 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 5981}
5982
75c96f85 5983static int __init md_init(void)
1da177e4 5984{
1da177e4
LT		 5985 if (register_blkdev(MAJOR_NR, "md"))
5986 return -1;
5987 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5988 unregister_blkdev(MAJOR_NR, "md");
5989 return -1;
5990 }
e8703fe1
N		 5991 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5992 md_probe, NULL, NULL);
5993 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 5994 md_probe, NULL, NULL);
5995
1da177e4 5996 register_reboot_notifier(&md_notifier);
0b4d4147 5997 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 5998
5999 md_geninit();
6000 return (0);
6001}
6002
6003
6004#ifndef MODULE
6005
6006/*
6007 * Searches all registered partitions for autorun RAID arrays
6008 * at boot time.
6009 */
4d936ec1
ME		 6010
6011static LIST_HEAD(all_detected_devices);
6012struct detected_devices_node {
6013 struct list_head list;
6014 dev_t dev;
6015};
1da177e4
LT		 6016
6017void md_autodetect_dev(dev_t dev)
6018{
4d936ec1
ME		 6019 struct detected_devices_node *node_detected_dev;
6020
6021 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6022 if (node_detected_dev) {
6023 node_detected_dev->dev = dev;
6024 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6025 } else {
6026 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6027 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6028 }
1da177e4
LT		 6029}
6030
6031
6032static void autostart_arrays(int part)
6033{
6034 mdk_rdev_t *rdev;
4d936ec1
ME		 6035 struct detected_devices_node *node_detected_dev;
6036 dev_t dev;
6037 int i_scanned, i_passed;
1da177e4 6038
4d936ec1
ME		 6039 i_scanned = 0;
6040 i_passed = 0;
1da177e4 6041
4d936ec1 6042 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 6043
4d936ec1
ME		 6044 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6045 i_scanned++;
6046 node_detected_dev = list_entry(all_detected_devices.next,
6047 struct detected_devices_node, list);
6048 list_del(&node_detected_dev->list);
6049 dev = node_detected_dev->dev;
6050 kfree(node_detected_dev);
df968c4e 6051 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 6052 if (IS_ERR(rdev))
6053 continue;
6054
b2d444d7 6055 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6056 MD_BUG();
6057 continue;
6058 }
d0fae18f 6059 set_bit(AutoDetected, &rdev->flags);
1da177e4 6060 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 6061 i_passed++;
1da177e4 6062 }
4d936ec1
ME		 6063
6064 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6065 i_scanned, i_passed);
1da177e4
LT		 6066
6067 autorun_devices(part);
6068}
6069
fdee8ae4 6070#endif /* !MODULE */
1da177e4
LT		 6071
6072static __exit void md_exit(void)
6073{
6074 mddev_t *mddev;
6075 struct list_head *tmp;
8ab5e4c1 6076
e8703fe1
N		 6077 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6078 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4
LT		 6079
6080 unregister_blkdev(MAJOR_NR,"md");
6081 unregister_blkdev(mdp_major, "mdp");
6082 unregister_reboot_notifier(&md_notifier);
6083 unregister_sysctl_table(raid_table_header);
6084 remove_proc_entry("mdstat", NULL);
29ac4aa3 6085 for_each_mddev(mddev, tmp) {
1da177e4
LT		 6086 struct gendisk *disk = mddev->gendisk;
6087 if (!disk)
6088 continue;
6089 export_array(mddev);
6090 del_gendisk(disk);
6091 put_disk(disk);
6092 mddev->gendisk = NULL;
6093 mddev_put(mddev);
6094 }
6095}
6096
685784aa 6097subsys_initcall(md_init);
1da177e4
LT		 6098module_exit(md_exit)
6099
f91de92e
N		 6100static int get_ro(char *buffer, struct kernel_param *kp)
6101{
6102 return sprintf(buffer, "%d", start_readonly);
6103}
6104static int set_ro(const char *val, struct kernel_param *kp)
6105{
6106 char *e;
6107 int num = simple_strtoul(val, &e, 10);
6108 if (*val && (*e == '\0' || *e == '\n')) {
6109 start_readonly = num;
4dbcdc75 6110 return 0;
f91de92e
N		 6111 }
6112 return -EINVAL;
6113}
6114
80ca3a44
N		 6115module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6116module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 6117
f91de92e 6118
1da177e4
LT		 6119EXPORT_SYMBOL(register_md_personality);
6120EXPORT_SYMBOL(unregister_md_personality);
6121EXPORT_SYMBOL(md_error);
6122EXPORT_SYMBOL(md_done_sync);
6123EXPORT_SYMBOL(md_write_start);
6124EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 6125EXPORT_SYMBOL(md_register_thread);
6126EXPORT_SYMBOL(md_unregister_thread);
6127EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 6128EXPORT_SYMBOL(md_check_recovery);
6129MODULE_LICENSE("GPL");
aa1595e9 6130MODULE_ALIAS("md");
72008652 6131MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)