]> bbs.cooldavid.org Git - net-next-2.6.git/blame - drivers/md/md.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] md: remove useless ioctl warning
[net-next-2.6.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
35#include <linux/module.h>
36#include <linux/config.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
LT		 41#include <linux/sysctl.h>
42#include <linux/devfs_fs_kernel.h>
43#include <linux/buffer_head.h> /* for invalidate_bdev */
44#include <linux/suspend.h>
d7603b7e 45#include <linux/poll.h>
48c9c27b 46#include <linux/mutex.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
75/*
76 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
77 * is 1000 KB/sec, so the extra system load does not show up that much.
78 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 79 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 80 * subsystem is idle. There is also an 'absolute maximum' reconstruction
81 * speed limit - in case reconstruction slows down your system despite
82 * idle IO detection.
83 *
84 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 85 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 86 */
87
88static int sysctl_speed_limit_min = 1000;
89static int sysctl_speed_limit_max = 200000;
88202a0c
N		 90static inline int speed_min(mddev_t *mddev)
91{
92 return mddev->sync_speed_min ?
93 mddev->sync_speed_min : sysctl_speed_limit_min;
94}
95
96static inline int speed_max(mddev_t *mddev)
97{
98 return mddev->sync_speed_max ?
99 mddev->sync_speed_max : sysctl_speed_limit_max;
100}
1da177e4
LT		 101
102static struct ctl_table_header *raid_table_header;
103
104static ctl_table raid_table[] = {
105 {
106 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
107 .procname = "speed_limit_min",
108 .data = &sysctl_speed_limit_min,
109 .maxlen = sizeof(int),
110 .mode = 0644,
111 .proc_handler = &proc_dointvec,
112 },
113 {
114 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
115 .procname = "speed_limit_max",
116 .data = &sysctl_speed_limit_max,
117 .maxlen = sizeof(int),
118 .mode = 0644,
119 .proc_handler = &proc_dointvec,
120 },
121 { .ctl_name = 0 }
122};
123
124static ctl_table raid_dir_table[] = {
125 {
126 .ctl_name = DEV_RAID,
127 .procname = "raid",
128 .maxlen = 0,
129 .mode = 0555,
130 .child = raid_table,
131 },
132 { .ctl_name = 0 }
133};
134
135static ctl_table raid_root_table[] = {
136 {
137 .ctl_name = CTL_DEV,
138 .procname = "dev",
139 .maxlen = 0,
140 .mode = 0555,
141 .child = raid_dir_table,
142 },
143 { .ctl_name = 0 }
144};
145
146static struct block_device_operations md_fops;
147
f91de92e
N		 148static int start_readonly;
149
d7603b7e
N		 150/*
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
154 * count increases.
155 *
156 * Events are:
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
159 */
2989ddbd 160static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 161static atomic_t md_event_count;
29269553 162void md_new_event(mddev_t *mddev)
d7603b7e
N		 163{
164 atomic_inc(&md_event_count);
165 wake_up(&md_event_waiters);
4508a7a7 166 sysfs_notify(&mddev->kobj, NULL, "sync_action");
d7603b7e 167}
29269553 168EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 169
c331eb04
N		 170/* Alternate version that can be called from interrupts
171 * when calling sysfs_notify isn't needed.
172 */
173void md_new_event_inintr(mddev_t *mddev)
174{
175 atomic_inc(&md_event_count);
176 wake_up(&md_event_waiters);
177}
178
1da177e4
LT		 179/*
180 * Enables to iterate over all existing md arrays
181 * all_mddevs_lock protects this list.
182 */
183static LIST_HEAD(all_mddevs);
184static DEFINE_SPINLOCK(all_mddevs_lock);
185
186
187/*
188 * iterates through all used mddevs in the system.
189 * We take care to grab the all_mddevs_lock whenever navigating
190 * the list, and to always hold a refcount when unlocked.
191 * Any code which breaks out of this loop while own
192 * a reference to the current mddev and must mddev_put it.
193 */
194#define ITERATE_MDDEV(mddev,tmp) \
195 \
196 for (({ spin_lock(&all_mddevs_lock); \
197 tmp = all_mddevs.next; \
198 mddev = NULL;}); \
199 ({ if (tmp != &all_mddevs) \
200 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
201 spin_unlock(&all_mddevs_lock); \
202 if (mddev) mddev_put(mddev); \
203 mddev = list_entry(tmp, mddev_t, all_mddevs); \
204 tmp != &all_mddevs;}); \
205 ({ spin_lock(&all_mddevs_lock); \
206 tmp = tmp->next;}) \
207 )
208
209
210static int md_fail_request (request_queue_t *q, struct bio *bio)
211{
212 bio_io_error(bio, bio->bi_size);
213 return 0;
214}
215
216static inline mddev_t *mddev_get(mddev_t *mddev)
217{
218 atomic_inc(&mddev->active);
219 return mddev;
220}
221
222static void mddev_put(mddev_t *mddev)
223{
224 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
225 return;
226 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
227 list_del(&mddev->all_mddevs);
926ce2d8 228 spin_unlock(&all_mddevs_lock);
1312f40e 229 blk_cleanup_queue(mddev->queue);
eae1701f 230 kobject_unregister(&mddev->kobj);
926ce2d8
N		 231 } else
232 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 233}
234
235static mddev_t * mddev_find(dev_t unit)
236{
237 mddev_t *mddev, *new = NULL;
238
239 retry:
240 spin_lock(&all_mddevs_lock);
241 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
242 if (mddev->unit == unit) {
243 mddev_get(mddev);
244 spin_unlock(&all_mddevs_lock);
990a8baf 245 kfree(new);
1da177e4
LT		 246 return mddev;
247 }
248
249 if (new) {
250 list_add(&new->all_mddevs, &all_mddevs);
251 spin_unlock(&all_mddevs_lock);
252 return new;
253 }
254 spin_unlock(&all_mddevs_lock);
255
9ffae0cf 256 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 257 if (!new)
258 return NULL;
259
1da177e4
LT		 260 new->unit = unit;
261 if (MAJOR(unit) == MD_MAJOR)
262 new->md_minor = MINOR(unit);
263 else
264 new->md_minor = MINOR(unit) >> MdpMinorShift;
265
df5b89b3 266 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 267 INIT_LIST_HEAD(&new->disks);
268 INIT_LIST_HEAD(&new->all_mddevs);
269 init_timer(&new->safemode_timer);
270 atomic_set(&new->active, 1);
06d91a5f 271 spin_lock_init(&new->write_lock);
3d310eb7 272 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 273
274 new->queue = blk_alloc_queue(GFP_KERNEL);
275 if (!new->queue) {
276 kfree(new);
277 return NULL;
278 }
89e5c8b5 279 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
1da177e4
LT		 280
281 blk_queue_make_request(new->queue, md_fail_request);
282
283 goto retry;
284}
285
286static inline int mddev_lock(mddev_t * mddev)
287{
df5b89b3 288 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 289}
290
1da177e4
LT		 291static inline int mddev_trylock(mddev_t * mddev)
292{
df5b89b3 293 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 294}
295
296static inline void mddev_unlock(mddev_t * mddev)
297{
df5b89b3 298 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 299
005eca5e 300 md_wakeup_thread(mddev->thread);
1da177e4
LT		 301}
302
2989ddbd 303static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 304{
305 mdk_rdev_t * rdev;
306 struct list_head *tmp;
307
308 ITERATE_RDEV(mddev,rdev,tmp) {
309 if (rdev->desc_nr == nr)
310 return rdev;
311 }
312 return NULL;
313}
314
315static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
316{
317 struct list_head *tmp;
318 mdk_rdev_t *rdev;
319
320 ITERATE_RDEV(mddev,rdev,tmp) {
321 if (rdev->bdev->bd_dev == dev)
322 return rdev;
323 }
324 return NULL;
325}
326
d9d166c2 327static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 328{
329 struct mdk_personality *pers;
d9d166c2
N		 330 list_for_each_entry(pers, &pers_list, list) {
331 if (level != LEVEL_NONE && pers->level == level)
2604b703 332 return pers;
d9d166c2
N		 333 if (strcmp(pers->name, clevel)==0)
334 return pers;
335 }
2604b703
N		 336 return NULL;
337}
338
77933d72 339static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 340{
341 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
342 return MD_NEW_SIZE_BLOCKS(size);
343}
344
345static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
346{
347 sector_t size;
348
349 size = rdev->sb_offset;
350
351 if (chunk_size)
352 size &= ~((sector_t)chunk_size/1024 - 1);
353 return size;
354}
355
356static int alloc_disk_sb(mdk_rdev_t * rdev)
357{
358 if (rdev->sb_page)
359 MD_BUG();
360
361 rdev->sb_page = alloc_page(GFP_KERNEL);
362 if (!rdev->sb_page) {
363 printk(KERN_ALERT "md: out of memory.\n");
364 return -EINVAL;
365 }
366
367 return 0;
368}
369
370static void free_disk_sb(mdk_rdev_t * rdev)
371{
372 if (rdev->sb_page) {
2d1f3b5d 373 put_page(rdev->sb_page);
1da177e4
LT		 374 rdev->sb_loaded = 0;
375 rdev->sb_page = NULL;
376 rdev->sb_offset = 0;
377 rdev->size = 0;
378 }
379}
380
381
7bfa19f2
N		 382static int super_written(struct bio *bio, unsigned int bytes_done, int error)
383{
384 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 385 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 386 if (bio->bi_size)
387 return 1;
388
389 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
a9701a30 390 md_error(mddev, rdev);
7bfa19f2 391
a9701a30
N		 392 if (atomic_dec_and_test(&mddev->pending_writes))
393 wake_up(&mddev->sb_wait);
f8b58edf 394 bio_put(bio);
7bfa19f2
N		 395 return 0;
396}
397
a9701a30
N		 398static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
399{
400 struct bio *bio2 = bio->bi_private;
401 mdk_rdev_t *rdev = bio2->bi_private;
402 mddev_t *mddev = rdev->mddev;
403 if (bio->bi_size)
404 return 1;
405
406 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
407 error == -EOPNOTSUPP) {
408 unsigned long flags;
409 /* barriers don't appear to be supported :-( */
410 set_bit(BarriersNotsupp, &rdev->flags);
411 mddev->barriers_work = 0;
412 spin_lock_irqsave(&mddev->write_lock, flags);
413 bio2->bi_next = mddev->biolist;
414 mddev->biolist = bio2;
415 spin_unlock_irqrestore(&mddev->write_lock, flags);
416 wake_up(&mddev->sb_wait);
417 bio_put(bio);
418 return 0;
419 }
420 bio_put(bio2);
421 bio->bi_private = rdev;
422 return super_written(bio, bytes_done, error);
423}
424
7bfa19f2
N		 425void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
426 sector_t sector, int size, struct page *page)
427{
428 /* write first size bytes of page to sector of rdev
429 * Increment mddev->pending_writes before returning
430 * and decrement it on completion, waking up sb_wait
431 * if zero is reached.
432 * If an error occurred, call md_error
a9701a30
N		 433 *
434 * As we might need to resubmit the request if BIO_RW_BARRIER
435 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 436 */
437 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 438 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 439
440 bio->bi_bdev = rdev->bdev;
441 bio->bi_sector = sector;
442 bio_add_page(bio, page, size, 0);
443 bio->bi_private = rdev;
444 bio->bi_end_io = super_written;
a9701a30
N		 445 bio->bi_rw = rw;
446
7bfa19f2 447 atomic_inc(&mddev->pending_writes);
a9701a30
N		 448 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
449 struct bio *rbio;
450 rw |= (1<<BIO_RW_BARRIER);
451 rbio = bio_clone(bio, GFP_NOIO);
452 rbio->bi_private = bio;
453 rbio->bi_end_io = super_written_barrier;
454 submit_bio(rw, rbio);
455 } else
456 submit_bio(rw, bio);
457}
458
459void md_super_wait(mddev_t *mddev)
460{
461 /* wait for all superblock writes that were scheduled to complete.
462 * if any had to be retried (due to BARRIER problems), retry them
463 */
464 DEFINE_WAIT(wq);
465 for(;;) {
466 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
467 if (atomic_read(&mddev->pending_writes)==0)
468 break;
469 while (mddev->biolist) {
470 struct bio *bio;
471 spin_lock_irq(&mddev->write_lock);
472 bio = mddev->biolist;
473 mddev->biolist = bio->bi_next ;
474 bio->bi_next = NULL;
475 spin_unlock_irq(&mddev->write_lock);
476 submit_bio(bio->bi_rw, bio);
477 }
478 schedule();
479 }
480 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 481}
482
1da177e4
LT		 483static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
484{
485 if (bio->bi_size)
486 return 1;
487
488 complete((struct completion*)bio->bi_private);
489 return 0;
490}
491
a654b9d8 492int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 493 struct page *page, int rw)
494{
baaa2c51 495 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 496 struct completion event;
497 int ret;
498
499 rw |= (1 << BIO_RW_SYNC);
500
501 bio->bi_bdev = bdev;
502 bio->bi_sector = sector;
503 bio_add_page(bio, page, size, 0);
504 init_completion(&event);
505 bio->bi_private = &event;
506 bio->bi_end_io = bi_complete;
507 submit_bio(rw, bio);
508 wait_for_completion(&event);
509
510 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
511 bio_put(bio);
512 return ret;
513}
a8745db2 514EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 515
0002b271 516static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 517{
518 char b[BDEVNAME_SIZE];
519 if (!rdev->sb_page) {
520 MD_BUG();
521 return -EINVAL;
522 }
523 if (rdev->sb_loaded)
524 return 0;
525
526
0002b271 527 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 528 goto fail;
529 rdev->sb_loaded = 1;
530 return 0;
531
532fail:
533 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
534 bdevname(rdev->bdev,b));
535 return -EINVAL;
536}
537
538static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
539{
540 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
541 (sb1->set_uuid1 == sb2->set_uuid1) &&
542 (sb1->set_uuid2 == sb2->set_uuid2) &&
543 (sb1->set_uuid3 == sb2->set_uuid3))
544
545 return 1;
546
547 return 0;
548}
549
550
551static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
552{
553 int ret;
554 mdp_super_t *tmp1, *tmp2;
555
556 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
557 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
558
559 if (!tmp1 || !tmp2) {
560 ret = 0;
561 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
562 goto abort;
563 }
564
565 *tmp1 = *sb1;
566 *tmp2 = *sb2;
567
568 /*
569 * nr_disks is not constant
570 */
571 tmp1->nr_disks = 0;
572 tmp2->nr_disks = 0;
573
574 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
575 ret = 0;
576 else
577 ret = 1;
578
579abort:
990a8baf
JJ		 580 kfree(tmp1);
581 kfree(tmp2);
1da177e4
LT		 582 return ret;
583}
584
585static unsigned int calc_sb_csum(mdp_super_t * sb)
586{
587 unsigned int disk_csum, csum;
588
589 disk_csum = sb->sb_csum;
590 sb->sb_csum = 0;
591 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
592 sb->sb_csum = disk_csum;
593 return csum;
594}
595
596
597/*
598 * Handle superblock details.
599 * We want to be able to handle multiple superblock formats
600 * so we have a common interface to them all, and an array of
601 * different handlers.
602 * We rely on user-space to write the initial superblock, and support
603 * reading and updating of superblocks.
604 * Interface methods are:
605 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
606 * loads and validates a superblock on dev.
607 * if refdev != NULL, compare superblocks on both devices
608 * Return:
609 * 0 - dev has a superblock that is compatible with refdev
610 * 1 - dev has a superblock that is compatible and newer than refdev
611 * so dev should be used as the refdev in future
612 * -EINVAL superblock incompatible or invalid
613 * -othererror e.g. -EIO
614 *
615 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
616 * Verify that dev is acceptable into mddev.
617 * The first time, mddev->raid_disks will be 0, and data from
618 * dev should be merged in. Subsequent calls check that dev
619 * is new enough. Return 0 or -EINVAL
620 *
621 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
622 * Update the superblock for rdev with data in mddev
623 * This does not write to disc.
624 *
625 */
626
627struct super_type {
628 char *name;
629 struct module *owner;
630 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
631 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
632 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
633};
634
635/*
636 * load_super for 0.90.0
637 */
638static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
639{
640 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
641 mdp_super_t *sb;
642 int ret;
643 sector_t sb_offset;
644
645 /*
646 * Calculate the position of the superblock,
647 * it's at the end of the disk.
648 *
649 * It also happens to be a multiple of 4Kb.
650 */
651 sb_offset = calc_dev_sboffset(rdev->bdev);
652 rdev->sb_offset = sb_offset;
653
0002b271 654 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 655 if (ret) return ret;
656
657 ret = -EINVAL;
658
659 bdevname(rdev->bdev, b);
660 sb = (mdp_super_t*)page_address(rdev->sb_page);
661
662 if (sb->md_magic != MD_SB_MAGIC) {
663 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
664 b);
665 goto abort;
666 }
667
668 if (sb->major_version != 0 ||
f6705578
N		 669 sb->minor_version < 90 ||
670 sb->minor_version > 91) {
1da177e4
LT		 671 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
672 sb->major_version, sb->minor_version,
673 b);
674 goto abort;
675 }
676
677 if (sb->raid_disks <= 0)
678 goto abort;
679
680 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
681 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
682 b);
683 goto abort;
684 }
685
686 rdev->preferred_minor = sb->md_minor;
687 rdev->data_offset = 0;
0002b271 688 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 689
690 if (sb->level == LEVEL_MULTIPATH)
691 rdev->desc_nr = -1;
692 else
693 rdev->desc_nr = sb->this_disk.number;
694
695 if (refdev == 0)
696 ret = 1;
697 else {
698 __u64 ev1, ev2;
699 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
700 if (!uuid_equal(refsb, sb)) {
701 printk(KERN_WARNING "md: %s has different UUID to %s\n",
702 b, bdevname(refdev->bdev,b2));
703 goto abort;
704 }
705 if (!sb_equal(refsb, sb)) {
706 printk(KERN_WARNING "md: %s has same UUID"
707 " but different superblock to %s\n",
708 b, bdevname(refdev->bdev, b2));
709 goto abort;
710 }
711 ev1 = md_event(sb);
712 ev2 = md_event(refsb);
713 if (ev1 > ev2)
714 ret = 1;
715 else
716 ret = 0;
717 }
718 rdev->size = calc_dev_size(rdev, sb->chunk_size);
719
2bf071bf
N		 720 if (rdev->size < sb->size && sb->level > 1)
721 /* "this cannot possibly happen" ... */
722 ret = -EINVAL;
723
1da177e4
LT		 724 abort:
725 return ret;
726}
727
728/*
729 * validate_super for 0.90.0
730 */
731static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
732{
733 mdp_disk_t *desc;
734 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
735
41158c7e 736 rdev->raid_disk = -1;
b2d444d7 737 rdev->flags = 0;
1da177e4
LT		 738 if (mddev->raid_disks == 0) {
739 mddev->major_version = 0;
740 mddev->minor_version = sb->minor_version;
741 mddev->patch_version = sb->patch_version;
742 mddev->persistent = ! sb->not_persistent;
743 mddev->chunk_size = sb->chunk_size;
744 mddev->ctime = sb->ctime;
745 mddev->utime = sb->utime;
746 mddev->level = sb->level;
d9d166c2 747 mddev->clevel[0] = 0;
1da177e4
LT		 748 mddev->layout = sb->layout;
749 mddev->raid_disks = sb->raid_disks;
750 mddev->size = sb->size;
751 mddev->events = md_event(sb);
9223214e 752 mddev->bitmap_offset = 0;
36fa3063 753 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 754
f6705578
N		 755 if (mddev->minor_version >= 91) {
756 mddev->reshape_position = sb->reshape_position;
757 mddev->delta_disks = sb->delta_disks;
758 mddev->new_level = sb->new_level;
759 mddev->new_layout = sb->new_layout;
760 mddev->new_chunk = sb->new_chunk;
761 } else {
762 mddev->reshape_position = MaxSector;
763 mddev->delta_disks = 0;
764 mddev->new_level = mddev->level;
765 mddev->new_layout = mddev->layout;
766 mddev->new_chunk = mddev->chunk_size;
767 }
768
1da177e4
LT		 769 if (sb->state & (1<<MD_SB_CLEAN))
770 mddev->recovery_cp = MaxSector;
771 else {
772 if (sb->events_hi == sb->cp_events_hi &&
773 sb->events_lo == sb->cp_events_lo) {
774 mddev->recovery_cp = sb->recovery_cp;
775 } else
776 mddev->recovery_cp = 0;
777 }
778
779 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
780 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
781 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
782 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
783
784 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 785
786 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
787 mddev->bitmap_file == NULL) {
c5a10f62
N		 788 if (mddev->level != 1 && mddev->level != 4
789 && mddev->level != 5 && mddev->level != 6
6cce3b23 790 && mddev->level != 10) {
a654b9d8 791 /* FIXME use a better test */
6cce3b23 792 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 793 return -EINVAL;
794 }
36fa3063 795 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8
N		 796 }
797
41158c7e
N		 798 } else if (mddev->pers == NULL) {
799 /* Insist on good event counter while assembling */
800 __u64 ev1 = md_event(sb);
1da177e4
LT		 801 ++ev1;
802 if (ev1 < mddev->events)
803 return -EINVAL;
41158c7e
N		 804 } else if (mddev->bitmap) {
805 /* if adding to array with a bitmap, then we can accept an
806 * older device ... but not too old.
807 */
808 __u64 ev1 = md_event(sb);
809 if (ev1 < mddev->bitmap->events_cleared)
810 return 0;
811 } else /* just a hot-add of a new device, leave raid_disk at -1 */
812 return 0;
813
1da177e4 814 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 815 desc = sb->disks + rdev->desc_nr;
816
817 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 818 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 819 else if (desc->state & (1<<MD_DISK_SYNC) &&
820 desc->raid_disk < mddev->raid_disks) {
b2d444d7 821 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 822 rdev->raid_disk = desc->raid_disk;
823 }
8ddf9efe
N		 824 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
825 set_bit(WriteMostly, &rdev->flags);
41158c7e 826 } else /* MULTIPATH are always insync */
b2d444d7 827 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 828 return 0;
829}
830
831/*
832 * sync_super for 0.90.0
833 */
834static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
835{
836 mdp_super_t *sb;
837 struct list_head *tmp;
838 mdk_rdev_t *rdev2;
839 int next_spare = mddev->raid_disks;
19133a42 840
1da177e4
LT		 841
842 /* make rdev->sb match mddev data..
843 *
844 * 1/ zero out disks
845 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
846 * 3/ any empty disks < next_spare become removed
847 *
848 * disks[0] gets initialised to REMOVED because
849 * we cannot be sure from other fields if it has
850 * been initialised or not.
851 */
852 int i;
853 int active=0, working=0,failed=0,spare=0,nr_disks=0;
854
61181565
N		 855 rdev->sb_size = MD_SB_BYTES;
856
1da177e4
LT		 857 sb = (mdp_super_t*)page_address(rdev->sb_page);
858
859 memset(sb, 0, sizeof(*sb));
860
861 sb->md_magic = MD_SB_MAGIC;
862 sb->major_version = mddev->major_version;
1da177e4
LT		 863 sb->patch_version = mddev->patch_version;
864 sb->gvalid_words = 0; /* ignored */
865 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
866 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
867 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
868 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
869
870 sb->ctime = mddev->ctime;
871 sb->level = mddev->level;
872 sb->size = mddev->size;
873 sb->raid_disks = mddev->raid_disks;
874 sb->md_minor = mddev->md_minor;
875 sb->not_persistent = !mddev->persistent;
876 sb->utime = mddev->utime;
877 sb->state = 0;
878 sb->events_hi = (mddev->events>>32);
879 sb->events_lo = (u32)mddev->events;
880
f6705578
N		 881 if (mddev->reshape_position == MaxSector)
882 sb->minor_version = 90;
883 else {
884 sb->minor_version = 91;
885 sb->reshape_position = mddev->reshape_position;
886 sb->new_level = mddev->new_level;
887 sb->delta_disks = mddev->delta_disks;
888 sb->new_layout = mddev->new_layout;
889 sb->new_chunk = mddev->new_chunk;
890 }
891 mddev->minor_version = sb->minor_version;
1da177e4
LT		 892 if (mddev->in_sync)
893 {
894 sb->recovery_cp = mddev->recovery_cp;
895 sb->cp_events_hi = (mddev->events>>32);
896 sb->cp_events_lo = (u32)mddev->events;
897 if (mddev->recovery_cp == MaxSector)
898 sb->state = (1<< MD_SB_CLEAN);
899 } else
900 sb->recovery_cp = 0;
901
902 sb->layout = mddev->layout;
903 sb->chunk_size = mddev->chunk_size;
904
a654b9d8
N		 905 if (mddev->bitmap && mddev->bitmap_file == NULL)
906 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
907
1da177e4
LT		 908 sb->disks[0].state = (1<<MD_DISK_REMOVED);
909 ITERATE_RDEV(mddev,rdev2,tmp) {
910 mdp_disk_t *d;
86e6ffdd 911 int desc_nr;
b2d444d7
N		 912 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
913 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 914 desc_nr = rdev2->raid_disk;
1da177e4 915 else
86e6ffdd 916 desc_nr = next_spare++;
19133a42 917 rdev2->desc_nr = desc_nr;
1da177e4
LT		 918 d = &sb->disks[rdev2->desc_nr];
919 nr_disks++;
920 d->number = rdev2->desc_nr;
921 d->major = MAJOR(rdev2->bdev->bd_dev);
922 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 923 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
924 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 925 d->raid_disk = rdev2->raid_disk;
926 else
927 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 928 if (test_bit(Faulty, &rdev2->flags))
1da177e4 929 d->state = (1<<MD_DISK_FAULTY);
1be7892f 930 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 931 d->state = (1<<MD_DISK_ACTIVE);
932 d->state |= (1<<MD_DISK_SYNC);
933 active++;
934 working++;
935 } else {
936 d->state = 0;
937 spare++;
938 working++;
939 }
8ddf9efe
N		 940 if (test_bit(WriteMostly, &rdev2->flags))
941 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 942 }
1da177e4
LT		 943 /* now set the "removed" and "faulty" bits on any missing devices */
944 for (i=0 ; i < mddev->raid_disks ; i++) {
945 mdp_disk_t *d = &sb->disks[i];
946 if (d->state == 0 && d->number == 0) {
947 d->number = i;
948 d->raid_disk = i;
949 d->state = (1<<MD_DISK_REMOVED);
950 d->state |= (1<<MD_DISK_FAULTY);
951 failed++;
952 }
953 }
954 sb->nr_disks = nr_disks;
955 sb->active_disks = active;
956 sb->working_disks = working;
957 sb->failed_disks = failed;
958 sb->spare_disks = spare;
959
960 sb->this_disk = sb->disks[rdev->desc_nr];
961 sb->sb_csum = calc_sb_csum(sb);
962}
963
964/*
965 * version 1 superblock
966 */
967
968static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
969{
970 unsigned int disk_csum, csum;
971 unsigned long long newcsum;
972 int size = 256 + le32_to_cpu(sb->max_dev)*2;
973 unsigned int *isuper = (unsigned int*)sb;
974 int i;
975
976 disk_csum = sb->sb_csum;
977 sb->sb_csum = 0;
978 newcsum = 0;
979 for (i=0; size>=4; size -= 4 )
980 newcsum += le32_to_cpu(*isuper++);
981
982 if (size == 2)
983 newcsum += le16_to_cpu(*(unsigned short*) isuper);
984
985 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
986 sb->sb_csum = disk_csum;
987 return cpu_to_le32(csum);
988}
989
990static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
991{
992 struct mdp_superblock_1 *sb;
993 int ret;
994 sector_t sb_offset;
995 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 996 int bmask;
1da177e4
LT		 997
998 /*
999 * Calculate the position of the superblock.
1000 * It is always aligned to a 4K boundary and
1001 * depeding on minor_version, it can be:
1002 * 0: At least 8K, but less than 12K, from end of device
1003 * 1: At start of device
1004 * 2: 4K from start of device.
1005 */
1006 switch(minor_version) {
1007 case 0:
1008 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1009 sb_offset -= 8*2;
39730960 1010 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 1011 /* convert from sectors to K */
1012 sb_offset /= 2;
1013 break;
1014 case 1:
1015 sb_offset = 0;
1016 break;
1017 case 2:
1018 sb_offset = 4;
1019 break;
1020 default:
1021 return -EINVAL;
1022 }
1023 rdev->sb_offset = sb_offset;
1024
0002b271
N		 1025 /* superblock is rarely larger than 1K, but it can be larger,
1026 * and it is safe to read 4k, so we do that
1027 */
1028 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1029 if (ret) return ret;
1030
1031
1032 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1033
1034 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1035 sb->major_version != cpu_to_le32(1) ||
1036 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1037 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1038 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1039 return -EINVAL;
1040
1041 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1042 printk("md: invalid superblock checksum on %s\n",
1043 bdevname(rdev->bdev,b));
1044 return -EINVAL;
1045 }
1046 if (le64_to_cpu(sb->data_size) < 10) {
1047 printk("md: data_size too small on %s\n",
1048 bdevname(rdev->bdev,b));
1049 return -EINVAL;
1050 }
1051 rdev->preferred_minor = 0xffff;
1052 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1053 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1054
0002b271 1055 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1056 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1057 if (rdev->sb_size & bmask)
1058 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1059
1da177e4 1060 if (refdev == 0)
8ed75463 1061 ret = 1;
1da177e4
LT		 1062 else {
1063 __u64 ev1, ev2;
1064 struct mdp_superblock_1 *refsb =
1065 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1066
1067 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1068 sb->level != refsb->level ||
1069 sb->layout != refsb->layout ||
1070 sb->chunksize != refsb->chunksize) {
1071 printk(KERN_WARNING "md: %s has strangely different"
1072 " superblock to %s\n",
1073 bdevname(rdev->bdev,b),
1074 bdevname(refdev->bdev,b2));
1075 return -EINVAL;
1076 }
1077 ev1 = le64_to_cpu(sb->events);
1078 ev2 = le64_to_cpu(refsb->events);
1079
1080 if (ev1 > ev2)
8ed75463
N		 1081 ret = 1;
1082 else
1083 ret = 0;
1da177e4
LT		 1084 }
1085 if (minor_version)
1086 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1087 else
1088 rdev->size = rdev->sb_offset;
1089 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1090 return -EINVAL;
1091 rdev->size = le64_to_cpu(sb->data_size)/2;
1092 if (le32_to_cpu(sb->chunksize))
1093 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf
N		 1094
1095 if (le32_to_cpu(sb->size) > rdev->size*2)
1096 return -EINVAL;
8ed75463 1097 return ret;
1da177e4
LT		 1098}
1099
1100static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1101{
1102 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1103
41158c7e 1104 rdev->raid_disk = -1;
b2d444d7 1105 rdev->flags = 0;
1da177e4
LT		 1106 if (mddev->raid_disks == 0) {
1107 mddev->major_version = 1;
1108 mddev->patch_version = 0;
1109 mddev->persistent = 1;
1110 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1111 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1112 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1113 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1114 mddev->clevel[0] = 0;
1da177e4
LT		 1115 mddev->layout = le32_to_cpu(sb->layout);
1116 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1117 mddev->size = le64_to_cpu(sb->size)/2;
1118 mddev->events = le64_to_cpu(sb->events);
9223214e 1119 mddev->bitmap_offset = 0;
29fc7e3e 1120 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1121
1122 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1123 memcpy(mddev->uuid, sb->set_uuid, 16);
1124
1125 mddev->max_disks = (4096-256)/2;
a654b9d8 1126
71c0805c 1127 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
a654b9d8 1128 mddev->bitmap_file == NULL ) {
6cce3b23
N		 1129 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1130 && mddev->level != 10) {
1131 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 1132 return -EINVAL;
1133 }
1134 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1135 }
f6705578
N		 1136 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1137 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1138 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1139 mddev->new_level = le32_to_cpu(sb->new_level);
1140 mddev->new_layout = le32_to_cpu(sb->new_layout);
1141 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1142 } else {
1143 mddev->reshape_position = MaxSector;
1144 mddev->delta_disks = 0;
1145 mddev->new_level = mddev->level;
1146 mddev->new_layout = mddev->layout;
1147 mddev->new_chunk = mddev->chunk_size;
1148 }
1149
41158c7e
N		 1150 } else if (mddev->pers == NULL) {
1151 /* Insist of good event counter while assembling */
1152 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4
LT		 1153 ++ev1;
1154 if (ev1 < mddev->events)
1155 return -EINVAL;
41158c7e
N		 1156 } else if (mddev->bitmap) {
1157 /* If adding to array with a bitmap, then we can accept an
1158 * older device, but not too old.
1159 */
1160 __u64 ev1 = le64_to_cpu(sb->events);
1161 if (ev1 < mddev->bitmap->events_cleared)
1162 return 0;
1163 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1164 return 0;
1da177e4
LT		 1165
1166 if (mddev->level != LEVEL_MULTIPATH) {
1167 int role;
1168 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1169 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1170 switch(role) {
1171 case 0xffff: /* spare */
1da177e4
LT		 1172 break;
1173 case 0xfffe: /* faulty */
b2d444d7 1174 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1175 break;
1176 default:
b2d444d7 1177 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1178 rdev->raid_disk = role;
1179 break;
1180 }
8ddf9efe
N		 1181 if (sb->devflags & WriteMostly1)
1182 set_bit(WriteMostly, &rdev->flags);
41158c7e 1183 } else /* MULTIPATH are always insync */
b2d444d7 1184 set_bit(In_sync, &rdev->flags);
41158c7e 1185
1da177e4
LT		 1186 return 0;
1187}
1188
1189static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1190{
1191 struct mdp_superblock_1 *sb;
1192 struct list_head *tmp;
1193 mdk_rdev_t *rdev2;
1194 int max_dev, i;
1195 /* make rdev->sb match mddev and rdev data. */
1196
1197 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1198
1199 sb->feature_map = 0;
1200 sb->pad0 = 0;
1201 memset(sb->pad1, 0, sizeof(sb->pad1));
1202 memset(sb->pad2, 0, sizeof(sb->pad2));
1203 memset(sb->pad3, 0, sizeof(sb->pad3));
1204
1205 sb->utime = cpu_to_le64((__u64)mddev->utime);
1206 sb->events = cpu_to_le64(mddev->events);
1207 if (mddev->in_sync)
1208 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1209 else
1210 sb->resync_offset = cpu_to_le64(0);
1211
4dbcdc75
N		 1212 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1213
f0ca340c 1214 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1215 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1216
a654b9d8
N		 1217 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1218 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1219 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1220 }
f6705578
N		 1221 if (mddev->reshape_position != MaxSector) {
1222 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1223 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1224 sb->new_layout = cpu_to_le32(mddev->new_layout);
1225 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1226 sb->new_level = cpu_to_le32(mddev->new_level);
1227 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1228 }
a654b9d8 1229
1da177e4
LT		 1230 max_dev = 0;
1231 ITERATE_RDEV(mddev,rdev2,tmp)
1232 if (rdev2->desc_nr+1 > max_dev)
1233 max_dev = rdev2->desc_nr+1;
1234
1235 sb->max_dev = cpu_to_le32(max_dev);
1236 for (i=0; i<max_dev;i++)
1237 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1238
1239 ITERATE_RDEV(mddev,rdev2,tmp) {
1240 i = rdev2->desc_nr;
b2d444d7 1241 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1242 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1243 else if (test_bit(In_sync, &rdev2->flags))
1da177e4
LT		 1244 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1245 else
1246 sb->dev_roles[i] = cpu_to_le16(0xffff);
1247 }
1248
1249 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1250 sb->sb_csum = calc_sb_1_csum(sb);
1251}
1252
1253
75c96f85 1254static struct super_type super_types[] = {
1da177e4
LT		 1255 [0] = {
1256 .name = "0.90.0",
1257 .owner = THIS_MODULE,
1258 .load_super = super_90_load,
1259 .validate_super = super_90_validate,
1260 .sync_super = super_90_sync,
1261 },
1262 [1] = {
1263 .name = "md-1",
1264 .owner = THIS_MODULE,
1265 .load_super = super_1_load,
1266 .validate_super = super_1_validate,
1267 .sync_super = super_1_sync,
1268 },
1269};
1270
1271static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1272{
1273 struct list_head *tmp;
1274 mdk_rdev_t *rdev;
1275
1276 ITERATE_RDEV(mddev,rdev,tmp)
1277 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1278 return rdev;
1279
1280 return NULL;
1281}
1282
1283static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1284{
1285 struct list_head *tmp;
1286 mdk_rdev_t *rdev;
1287
1288 ITERATE_RDEV(mddev1,rdev,tmp)
1289 if (match_dev_unit(mddev2, rdev))
1290 return 1;
1291
1292 return 0;
1293}
1294
1295static LIST_HEAD(pending_raid_disks);
1296
1297static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1298{
1299 mdk_rdev_t *same_pdev;
1300 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
f637b9f9 1301 struct kobject *ko;
1edf80d3 1302 char *s;
1da177e4
LT		 1303
1304 if (rdev->mddev) {
1305 MD_BUG();
1306 return -EINVAL;
1307 }
2bf071bf
N		 1308 /* make sure rdev->size exceeds mddev->size */
1309 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1310 if (mddev->pers)
1311 /* Cannot change size, so fail */
1312 return -ENOSPC;
1313 else
1314 mddev->size = rdev->size;
1315 }
1da177e4
LT		 1316 same_pdev = match_dev_unit(mddev, rdev);
1317 if (same_pdev)
1318 printk(KERN_WARNING
1319 "%s: WARNING: %s appears to be on the same physical"
1320 " disk as %s. True\n protection against single-disk"
1321 " failure might be compromised.\n",
1322 mdname(mddev), bdevname(rdev->bdev,b),
1323 bdevname(same_pdev->bdev,b2));
1324
1325 /* Verify rdev->desc_nr is unique.
1326 * If it is -1, assign a free number, else
1327 * check number is not in use
1328 */
1329 if (rdev->desc_nr < 0) {
1330 int choice = 0;
1331 if (mddev->pers) choice = mddev->raid_disks;
1332 while (find_rdev_nr(mddev, choice))
1333 choice++;
1334 rdev->desc_nr = choice;
1335 } else {
1336 if (find_rdev_nr(mddev, rdev->desc_nr))
1337 return -EBUSY;
1338 }
19133a42
N		 1339 bdevname(rdev->bdev,b);
1340 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1341 return -ENOMEM;
1edf80d3
NB		 1342 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1343 *s = '!';
1da177e4
LT		 1344
1345 list_add(&rdev->same_set, &mddev->disks);
1346 rdev->mddev = mddev;
19133a42 1347 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1348
9c791977 1349 rdev->kobj.parent = &mddev->kobj;
86e6ffdd
N		 1350 kobject_add(&rdev->kobj);
1351
f637b9f9
N		 1352 if (rdev->bdev->bd_part)
1353 ko = &rdev->bdev->bd_part->kobj;
1354 else
1355 ko = &rdev->bdev->bd_disk->kobj;
1356 sysfs_create_link(&rdev->kobj, ko, "block");
5463c790 1357 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1da177e4
LT		 1358 return 0;
1359}
1360
1361static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1362{
1363 char b[BDEVNAME_SIZE];
1364 if (!rdev->mddev) {
1365 MD_BUG();
1366 return;
1367 }
5463c790 1368 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1369 list_del_init(&rdev->same_set);
1370 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1371 rdev->mddev = NULL;
86e6ffdd
N		 1372 sysfs_remove_link(&rdev->kobj, "block");
1373 kobject_del(&rdev->kobj);
1da177e4
LT		 1374}
1375
1376/*
1377 * prevent the device from being mounted, repartitioned or
1378 * otherwise reused by a RAID array (or any other kernel
1379 * subsystem), by bd_claiming the device.
1380 */
1381static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1382{
1383 int err = 0;
1384 struct block_device *bdev;
1385 char b[BDEVNAME_SIZE];
1386
1387 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1388 if (IS_ERR(bdev)) {
1389 printk(KERN_ERR "md: could not open %s.\n",
1390 __bdevname(dev, b));
1391 return PTR_ERR(bdev);
1392 }
1393 err = bd_claim(bdev, rdev);
1394 if (err) {
1395 printk(KERN_ERR "md: could not bd_claim %s.\n",
1396 bdevname(bdev, b));
1397 blkdev_put(bdev);
1398 return err;
1399 }
1400 rdev->bdev = bdev;
1401 return err;
1402}
1403
1404static void unlock_rdev(mdk_rdev_t *rdev)
1405{
1406 struct block_device *bdev = rdev->bdev;
1407 rdev->bdev = NULL;
1408 if (!bdev)
1409 MD_BUG();
1410 bd_release(bdev);
1411 blkdev_put(bdev);
1412}
1413
1414void md_autodetect_dev(dev_t dev);
1415
1416static void export_rdev(mdk_rdev_t * rdev)
1417{
1418 char b[BDEVNAME_SIZE];
1419 printk(KERN_INFO "md: export_rdev(%s)\n",
1420 bdevname(rdev->bdev,b));
1421 if (rdev->mddev)
1422 MD_BUG();
1423 free_disk_sb(rdev);
1424 list_del_init(&rdev->same_set);
1425#ifndef MODULE
1426 md_autodetect_dev(rdev->bdev->bd_dev);
1427#endif
1428 unlock_rdev(rdev);
86e6ffdd 1429 kobject_put(&rdev->kobj);
1da177e4
LT		 1430}
1431
1432static void kick_rdev_from_array(mdk_rdev_t * rdev)
1433{
1434 unbind_rdev_from_array(rdev);
1435 export_rdev(rdev);
1436}
1437
1438static void export_array(mddev_t *mddev)
1439{
1440 struct list_head *tmp;
1441 mdk_rdev_t *rdev;
1442
1443 ITERATE_RDEV(mddev,rdev,tmp) {
1444 if (!rdev->mddev) {
1445 MD_BUG();
1446 continue;
1447 }
1448 kick_rdev_from_array(rdev);
1449 }
1450 if (!list_empty(&mddev->disks))
1451 MD_BUG();
1452 mddev->raid_disks = 0;
1453 mddev->major_version = 0;
1454}
1455
1456static void print_desc(mdp_disk_t *desc)
1457{
1458 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1459 desc->major,desc->minor,desc->raid_disk,desc->state);
1460}
1461
1462static void print_sb(mdp_super_t *sb)
1463{
1464 int i;
1465
1466 printk(KERN_INFO
1467 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1468 sb->major_version, sb->minor_version, sb->patch_version,
1469 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1470 sb->ctime);
1471 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1472 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1473 sb->md_minor, sb->layout, sb->chunk_size);
1474 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1475 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1476 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1477 sb->failed_disks, sb->spare_disks,
1478 sb->sb_csum, (unsigned long)sb->events_lo);
1479
1480 printk(KERN_INFO);
1481 for (i = 0; i < MD_SB_DISKS; i++) {
1482 mdp_disk_t *desc;
1483
1484 desc = sb->disks + i;
1485 if (desc->number || desc->major || desc->minor ||
1486 desc->raid_disk || (desc->state && (desc->state != 4))) {
1487 printk(" D %2d: ", i);
1488 print_desc(desc);
1489 }
1490 }
1491 printk(KERN_INFO "md: THIS: ");
1492 print_desc(&sb->this_disk);
1493
1494}
1495
1496static void print_rdev(mdk_rdev_t *rdev)
1497{
1498 char b[BDEVNAME_SIZE];
1499 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1500 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1501 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1502 rdev->desc_nr);
1da177e4
LT		 1503 if (rdev->sb_loaded) {
1504 printk(KERN_INFO "md: rdev superblock:\n");
1505 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1506 } else
1507 printk(KERN_INFO "md: no rdev superblock!\n");
1508}
1509
1510void md_print_devices(void)
1511{
1512 struct list_head *tmp, *tmp2;
1513 mdk_rdev_t *rdev;
1514 mddev_t *mddev;
1515 char b[BDEVNAME_SIZE];
1516
1517 printk("\n");
1518 printk("md: **********************************\n");
1519 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1520 printk("md: **********************************\n");
1521 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1522
32a7627c
N		 1523 if (mddev->bitmap)
1524 bitmap_print_sb(mddev->bitmap);
1525 else
1526 printk("%s: ", mdname(mddev));
1da177e4
LT		 1527 ITERATE_RDEV(mddev,rdev,tmp2)
1528 printk("<%s>", bdevname(rdev->bdev,b));
1529 printk("\n");
1530
1531 ITERATE_RDEV(mddev,rdev,tmp2)
1532 print_rdev(rdev);
1533 }
1534 printk("md: **********************************\n");
1535 printk("\n");
1536}
1537
1538
1da177e4
LT		 1539static void sync_sbs(mddev_t * mddev)
1540{
1541 mdk_rdev_t *rdev;
1542 struct list_head *tmp;
1543
1544 ITERATE_RDEV(mddev,rdev,tmp) {
1545 super_types[mddev->major_version].
1546 sync_super(mddev, rdev);
1547 rdev->sb_loaded = 1;
1548 }
1549}
1550
f6705578 1551void md_update_sb(mddev_t * mddev)
1da177e4 1552{
7bfa19f2 1553 int err;
1da177e4
LT		 1554 struct list_head *tmp;
1555 mdk_rdev_t *rdev;
06d91a5f 1556 int sync_req;
1da177e4 1557
1da177e4 1558repeat:
a9701a30 1559 spin_lock_irq(&mddev->write_lock);
06d91a5f 1560 sync_req = mddev->in_sync;
1da177e4
LT		 1561 mddev->utime = get_seconds();
1562 mddev->events ++;
1563
1564 if (!mddev->events) {
1565 /*
1566 * oops, this 64-bit counter should never wrap.
1567 * Either we are in around ~1 trillion A.C., assuming
1568 * 1 reboot per second, or we have a bug:
1569 */
1570 MD_BUG();
1571 mddev->events --;
1572 }
7bfa19f2 1573 mddev->sb_dirty = 2;
1da177e4
LT		 1574 sync_sbs(mddev);
1575
1576 /*
1577 * do not write anything to disk if using
1578 * nonpersistent superblocks
1579 */
06d91a5f
N		 1580 if (!mddev->persistent) {
1581 mddev->sb_dirty = 0;
a9701a30 1582 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1583 wake_up(&mddev->sb_wait);
1da177e4 1584 return;
06d91a5f 1585 }
a9701a30 1586 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1587
1588 dprintk(KERN_INFO
1589 "md: updating %s RAID superblock on device (in sync %d)\n",
1590 mdname(mddev),mddev->in_sync);
1591
32a7627c 1592 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1593 ITERATE_RDEV(mddev,rdev,tmp) {
1594 char b[BDEVNAME_SIZE];
1595 dprintk(KERN_INFO "md: ");
b2d444d7 1596 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1597 dprintk("(skipping faulty ");
1598
1599 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1600 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1601 md_super_write(mddev,rdev,
0002b271 1602 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1603 rdev->sb_page);
1604 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1605 bdevname(rdev->bdev,b),
1606 (unsigned long long)rdev->sb_offset);
1607
1da177e4
LT		 1608 } else
1609 dprintk(")\n");
7bfa19f2 1610 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1611 /* only need to write one superblock... */
1612 break;
1613 }
a9701a30 1614 md_super_wait(mddev);
7bfa19f2
N		 1615 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1616
a9701a30 1617 spin_lock_irq(&mddev->write_lock);
7bfa19f2 1618 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
06d91a5f 1619 /* have to write it out again */
a9701a30 1620 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1621 goto repeat;
1622 }
1623 mddev->sb_dirty = 0;
a9701a30 1624 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1625 wake_up(&mddev->sb_wait);
06d91a5f 1626
1da177e4 1627}
f6705578 1628EXPORT_SYMBOL_GPL(md_update_sb);
1da177e4 1629
bce74dac
N		 1630/* words written to sysfs files may, or my not, be \n terminated.
1631 * We want to accept with case. For this we use cmd_match.
1632 */
1633static int cmd_match(const char *cmd, const char *str)
1634{
1635 /* See if cmd, written into a sysfs file, matches
1636 * str. They must either be the same, or cmd can
1637 * have a trailing newline
1638 */
1639 while (*cmd && *str && *cmd == *str) {
1640 cmd++;
1641 str++;
1642 }
1643 if (*cmd == '\n')
1644 cmd++;
1645 if (*str || *cmd)
1646 return 0;
1647 return 1;
1648}
1649
86e6ffdd
N		 1650struct rdev_sysfs_entry {
1651 struct attribute attr;
1652 ssize_t (*show)(mdk_rdev_t *, char *);
1653 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1654};
1655
1656static ssize_t
96de1e66 1657state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1658{
1659 char *sep = "";
1660 int len=0;
1661
b2d444d7 1662 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1663 len+= sprintf(page+len, "%sfaulty",sep);
1664 sep = ",";
1665 }
b2d444d7 1666 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1667 len += sprintf(page+len, "%sin_sync",sep);
1668 sep = ",";
1669 }
b2d444d7
N		 1670 if (!test_bit(Faulty, &rdev->flags) &&
1671 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1672 len += sprintf(page+len, "%sspare", sep);
1673 sep = ",";
1674 }
1675 return len+sprintf(page+len, "\n");
1676}
1677
96de1e66
N		 1678static struct rdev_sysfs_entry
1679rdev_state = __ATTR_RO(state);
86e6ffdd
N		 1680
1681static ssize_t
96de1e66 1682super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1683{
1684 if (rdev->sb_loaded && rdev->sb_size) {
1685 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1686 return rdev->sb_size;
1687 } else
1688 return 0;
1689}
96de1e66
N		 1690static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1691
4dbcdc75
N		 1692static ssize_t
1693errors_show(mdk_rdev_t *rdev, char *page)
1694{
1695 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1696}
1697
1698static ssize_t
1699errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1700{
1701 char *e;
1702 unsigned long n = simple_strtoul(buf, &e, 10);
1703 if (*buf && (*e == 0 || *e == '\n')) {
1704 atomic_set(&rdev->corrected_errors, n);
1705 return len;
1706 }
1707 return -EINVAL;
1708}
1709static struct rdev_sysfs_entry rdev_errors =
1710__ATTR(errors, 0644, errors_show, errors_store);
1711
014236d2
N		 1712static ssize_t
1713slot_show(mdk_rdev_t *rdev, char *page)
1714{
1715 if (rdev->raid_disk < 0)
1716 return sprintf(page, "none\n");
1717 else
1718 return sprintf(page, "%d\n", rdev->raid_disk);
1719}
1720
1721static ssize_t
1722slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1723{
1724 char *e;
1725 int slot = simple_strtoul(buf, &e, 10);
1726 if (strncmp(buf, "none", 4)==0)
1727 slot = -1;
1728 else if (e==buf || (*e && *e!= '\n'))
1729 return -EINVAL;
1730 if (rdev->mddev->pers)
1731 /* Cannot set slot in active array (yet) */
1732 return -EBUSY;
1733 if (slot >= rdev->mddev->raid_disks)
1734 return -ENOSPC;
1735 rdev->raid_disk = slot;
1736 /* assume it is working */
1737 rdev->flags = 0;
1738 set_bit(In_sync, &rdev->flags);
1739 return len;
1740}
1741
1742
1743static struct rdev_sysfs_entry rdev_slot =
1744__ATTR(slot, 0644, slot_show, slot_store);
1745
93c8cad0
N		 1746static ssize_t
1747offset_show(mdk_rdev_t *rdev, char *page)
1748{
6961ece4 1749 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1750}
1751
1752static ssize_t
1753offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1754{
1755 char *e;
1756 unsigned long long offset = simple_strtoull(buf, &e, 10);
1757 if (e==buf || (*e && *e != '\n'))
1758 return -EINVAL;
1759 if (rdev->mddev->pers)
1760 return -EBUSY;
1761 rdev->data_offset = offset;
1762 return len;
1763}
1764
1765static struct rdev_sysfs_entry rdev_offset =
1766__ATTR(offset, 0644, offset_show, offset_store);
1767
83303b61
N		 1768static ssize_t
1769rdev_size_show(mdk_rdev_t *rdev, char *page)
1770{
1771 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1772}
1773
1774static ssize_t
1775rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1776{
1777 char *e;
1778 unsigned long long size = simple_strtoull(buf, &e, 10);
1779 if (e==buf || (*e && *e != '\n'))
1780 return -EINVAL;
1781 if (rdev->mddev->pers)
1782 return -EBUSY;
1783 rdev->size = size;
1784 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1785 rdev->mddev->size = size;
1786 return len;
1787}
1788
1789static struct rdev_sysfs_entry rdev_size =
1790__ATTR(size, 0644, rdev_size_show, rdev_size_store);
1791
86e6ffdd
N		 1792static struct attribute *rdev_default_attrs[] = {
1793 &rdev_state.attr,
1794 &rdev_super.attr,
4dbcdc75 1795 &rdev_errors.attr,
014236d2 1796 &rdev_slot.attr,
93c8cad0 1797 &rdev_offset.attr,
83303b61 1798 &rdev_size.attr,
86e6ffdd
N		 1799 NULL,
1800};
1801static ssize_t
1802rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1803{
1804 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1805 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1806
1807 if (!entry->show)
1808 return -EIO;
1809 return entry->show(rdev, page);
1810}
1811
1812static ssize_t
1813rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1814 const char *page, size_t length)
1815{
1816 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1817 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1818
1819 if (!entry->store)
1820 return -EIO;
1821 return entry->store(rdev, page, length);
1822}
1823
1824static void rdev_free(struct kobject *ko)
1825{
1826 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1827 kfree(rdev);
1828}
1829static struct sysfs_ops rdev_sysfs_ops = {
1830 .show = rdev_attr_show,
1831 .store = rdev_attr_store,
1832};
1833static struct kobj_type rdev_ktype = {
1834 .release = rdev_free,
1835 .sysfs_ops = &rdev_sysfs_ops,
1836 .default_attrs = rdev_default_attrs,
1837};
1838
1da177e4
LT		 1839/*
1840 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1841 *
1842 * mark the device faulty if:
1843 *
1844 * - the device is nonexistent (zero size)
1845 * - the device has no valid superblock
1846 *
1847 * a faulty rdev _never_ has rdev->sb set.
1848 */
1849static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1850{
1851 char b[BDEVNAME_SIZE];
1852 int err;
1853 mdk_rdev_t *rdev;
1854 sector_t size;
1855
9ffae0cf 1856 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 1857 if (!rdev) {
1858 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1859 return ERR_PTR(-ENOMEM);
1860 }
1da177e4
LT		 1861
1862 if ((err = alloc_disk_sb(rdev)))
1863 goto abort_free;
1864
1865 err = lock_rdev(rdev, newdev);
1866 if (err)
1867 goto abort_free;
1868
86e6ffdd
N		 1869 rdev->kobj.parent = NULL;
1870 rdev->kobj.ktype = &rdev_ktype;
1871 kobject_init(&rdev->kobj);
1872
1da177e4 1873 rdev->desc_nr = -1;
b2d444d7 1874 rdev->flags = 0;
1da177e4
LT		 1875 rdev->data_offset = 0;
1876 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 1877 atomic_set(&rdev->read_errors, 0);
4dbcdc75 1878 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 1879
1880 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1881 if (!size) {
1882 printk(KERN_WARNING
1883 "md: %s has zero or unknown size, marking faulty!\n",
1884 bdevname(rdev->bdev,b));
1885 err = -EINVAL;
1886 goto abort_free;
1887 }
1888
1889 if (super_format >= 0) {
1890 err = super_types[super_format].
1891 load_super(rdev, NULL, super_minor);
1892 if (err == -EINVAL) {
1893 printk(KERN_WARNING
1894 "md: %s has invalid sb, not importing!\n",
1895 bdevname(rdev->bdev,b));
1896 goto abort_free;
1897 }
1898 if (err < 0) {
1899 printk(KERN_WARNING
1900 "md: could not read %s's sb, not importing!\n",
1901 bdevname(rdev->bdev,b));
1902 goto abort_free;
1903 }
1904 }
1905 INIT_LIST_HEAD(&rdev->same_set);
1906
1907 return rdev;
1908
1909abort_free:
1910 if (rdev->sb_page) {
1911 if (rdev->bdev)
1912 unlock_rdev(rdev);
1913 free_disk_sb(rdev);
1914 }
1915 kfree(rdev);
1916 return ERR_PTR(err);
1917}
1918
1919/*
1920 * Check a full RAID array for plausibility
1921 */
1922
1923
a757e64c 1924static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 1925{
1926 int i;
1927 struct list_head *tmp;
1928 mdk_rdev_t *rdev, *freshest;
1929 char b[BDEVNAME_SIZE];
1930
1931 freshest = NULL;
1932 ITERATE_RDEV(mddev,rdev,tmp)
1933 switch (super_types[mddev->major_version].
1934 load_super(rdev, freshest, mddev->minor_version)) {
1935 case 1:
1936 freshest = rdev;
1937 break;
1938 case 0:
1939 break;
1940 default:
1941 printk( KERN_ERR \
1942 "md: fatal superblock inconsistency in %s"
1943 " -- removing from array\n",
1944 bdevname(rdev->bdev,b));
1945 kick_rdev_from_array(rdev);
1946 }
1947
1948
1949 super_types[mddev->major_version].
1950 validate_super(mddev, freshest);
1951
1952 i = 0;
1953 ITERATE_RDEV(mddev,rdev,tmp) {
1954 if (rdev != freshest)
1955 if (super_types[mddev->major_version].
1956 validate_super(mddev, rdev)) {
1957 printk(KERN_WARNING "md: kicking non-fresh %s"
1958 " from array!\n",
1959 bdevname(rdev->bdev,b));
1960 kick_rdev_from_array(rdev);
1961 continue;
1962 }
1963 if (mddev->level == LEVEL_MULTIPATH) {
1964 rdev->desc_nr = i++;
1965 rdev->raid_disk = rdev->desc_nr;
b2d444d7 1966 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1967 }
1968 }
1969
1970
1971
1972 if (mddev->recovery_cp != MaxSector &&
1973 mddev->level >= 1)
1974 printk(KERN_ERR "md: %s: raid array is not clean"
1975 " -- starting background reconstruction\n",
1976 mdname(mddev));
1977
1da177e4
LT		 1978}
1979
eae1701f 1980static ssize_t
96de1e66 1981level_show(mddev_t *mddev, char *page)
eae1701f 1982{
2604b703 1983 struct mdk_personality *p = mddev->pers;
d9d166c2 1984 if (p)
eae1701f 1985 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 1986 else if (mddev->clevel[0])
1987 return sprintf(page, "%s\n", mddev->clevel);
1988 else if (mddev->level != LEVEL_NONE)
1989 return sprintf(page, "%d\n", mddev->level);
1990 else
1991 return 0;
eae1701f
N		 1992}
1993
d9d166c2
N		 1994static ssize_t
1995level_store(mddev_t *mddev, const char *buf, size_t len)
1996{
1997 int rv = len;
1998 if (mddev->pers)
1999 return -EBUSY;
2000 if (len == 0)
2001 return 0;
2002 if (len >= sizeof(mddev->clevel))
2003 return -ENOSPC;
2004 strncpy(mddev->clevel, buf, len);
2005 if (mddev->clevel[len-1] == '\n')
2006 len--;
2007 mddev->clevel[len] = 0;
2008 mddev->level = LEVEL_NONE;
2009 return rv;
2010}
2011
2012static struct md_sysfs_entry md_level =
2013__ATTR(level, 0644, level_show, level_store);
eae1701f
N		 2014
2015static ssize_t
96de1e66 2016raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2017{
bb636547
N		 2018 if (mddev->raid_disks == 0)
2019 return 0;
eae1701f
N		 2020 return sprintf(page, "%d\n", mddev->raid_disks);
2021}
2022
da943b99
N		 2023static int update_raid_disks(mddev_t *mddev, int raid_disks);
2024
2025static ssize_t
2026raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2027{
2028 /* can only set raid_disks if array is not yet active */
2029 char *e;
2030 int rv = 0;
2031 unsigned long n = simple_strtoul(buf, &e, 10);
2032
2033 if (!*buf || (*e && *e != '\n'))
2034 return -EINVAL;
2035
2036 if (mddev->pers)
2037 rv = update_raid_disks(mddev, n);
2038 else
2039 mddev->raid_disks = n;
2040 return rv ? rv : len;
2041}
2042static struct md_sysfs_entry md_raid_disks =
2043__ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
eae1701f 2044
3b34380a
N		 2045static ssize_t
2046chunk_size_show(mddev_t *mddev, char *page)
2047{
2048 return sprintf(page, "%d\n", mddev->chunk_size);
2049}
2050
2051static ssize_t
2052chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2053{
2054 /* can only set chunk_size if array is not yet active */
2055 char *e;
2056 unsigned long n = simple_strtoul(buf, &e, 10);
2057
2058 if (mddev->pers)
2059 return -EBUSY;
2060 if (!*buf || (*e && *e != '\n'))
2061 return -EINVAL;
2062
2063 mddev->chunk_size = n;
2064 return len;
2065}
2066static struct md_sysfs_entry md_chunk_size =
2067__ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2068
6d7ff738
N		 2069static ssize_t
2070null_show(mddev_t *mddev, char *page)
2071{
2072 return -EINVAL;
2073}
2074
2075static ssize_t
2076new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2077{
2078 /* buf must be %d:%d\n? giving major and minor numbers */
2079 /* The new device is added to the array.
2080 * If the array has a persistent superblock, we read the
2081 * superblock to initialise info and check validity.
2082 * Otherwise, only checking done is that in bind_rdev_to_array,
2083 * which mainly checks size.
2084 */
2085 char *e;
2086 int major = simple_strtoul(buf, &e, 10);
2087 int minor;
2088 dev_t dev;
2089 mdk_rdev_t *rdev;
2090 int err;
2091
2092 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2093 return -EINVAL;
2094 minor = simple_strtoul(e+1, &e, 10);
2095 if (*e && *e != '\n')
2096 return -EINVAL;
2097 dev = MKDEV(major, minor);
2098 if (major != MAJOR(dev) ||
2099 minor != MINOR(dev))
2100 return -EOVERFLOW;
2101
2102
2103 if (mddev->persistent) {
2104 rdev = md_import_device(dev, mddev->major_version,
2105 mddev->minor_version);
2106 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2107 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2108 mdk_rdev_t, same_set);
2109 err = super_types[mddev->major_version]
2110 .load_super(rdev, rdev0, mddev->minor_version);
2111 if (err < 0)
2112 goto out;
2113 }
2114 } else
2115 rdev = md_import_device(dev, -1, -1);
2116
2117 if (IS_ERR(rdev))
2118 return PTR_ERR(rdev);
2119 err = bind_rdev_to_array(rdev, mddev);
2120 out:
2121 if (err)
2122 export_rdev(rdev);
2123 return err ? err : len;
2124}
2125
2126static struct md_sysfs_entry md_new_device =
2127__ATTR(new_dev, 0200, null_show, new_dev_store);
3b34380a 2128
a35b0d69
N		 2129static ssize_t
2130size_show(mddev_t *mddev, char *page)
2131{
2132 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2133}
2134
2135static int update_size(mddev_t *mddev, unsigned long size);
2136
2137static ssize_t
2138size_store(mddev_t *mddev, const char *buf, size_t len)
2139{
2140 /* If array is inactive, we can reduce the component size, but
2141 * not increase it (except from 0).
2142 * If array is active, we can try an on-line resize
2143 */
2144 char *e;
2145 int err = 0;
2146 unsigned long long size = simple_strtoull(buf, &e, 10);
2147 if (!*buf || *buf == '\n' ||
2148 (*e && *e != '\n'))
2149 return -EINVAL;
2150
2151 if (mddev->pers) {
2152 err = update_size(mddev, size);
2153 md_update_sb(mddev);
2154 } else {
2155 if (mddev->size == 0 ||
2156 mddev->size > size)
2157 mddev->size = size;
2158 else
2159 err = -ENOSPC;
2160 }
2161 return err ? err : len;
2162}
2163
2164static struct md_sysfs_entry md_size =
2165__ATTR(component_size, 0644, size_show, size_store);
2166
8bb93aac
N		 2167
2168/* Metdata version.
2169 * This is either 'none' for arrays with externally managed metadata,
2170 * or N.M for internally known formats
2171 */
2172static ssize_t
2173metadata_show(mddev_t *mddev, char *page)
2174{
2175 if (mddev->persistent)
2176 return sprintf(page, "%d.%d\n",
2177 mddev->major_version, mddev->minor_version);
2178 else
2179 return sprintf(page, "none\n");
2180}
2181
2182static ssize_t
2183metadata_store(mddev_t *mddev, const char *buf, size_t len)
2184{
2185 int major, minor;
2186 char *e;
2187 if (!list_empty(&mddev->disks))
2188 return -EBUSY;
2189
2190 if (cmd_match(buf, "none")) {
2191 mddev->persistent = 0;
2192 mddev->major_version = 0;
2193 mddev->minor_version = 90;
2194 return len;
2195 }
2196 major = simple_strtoul(buf, &e, 10);
2197 if (e==buf || *e != '.')
2198 return -EINVAL;
2199 buf = e+1;
2200 minor = simple_strtoul(buf, &e, 10);
2201 if (e==buf || *e != '\n')
2202 return -EINVAL;
2203 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2204 super_types[major].name == NULL)
2205 return -ENOENT;
2206 mddev->major_version = major;
2207 mddev->minor_version = minor;
2208 mddev->persistent = 1;
2209 return len;
2210}
2211
2212static struct md_sysfs_entry md_metadata =
2213__ATTR(metadata_version, 0644, metadata_show, metadata_store);
2214
24dd469d 2215static ssize_t
7eec314d 2216action_show(mddev_t *mddev, char *page)
24dd469d 2217{
7eec314d 2218 char *type = "idle";
31399d9e
N		 2219 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2220 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
ccfcc3c1
N		 2221 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2222 type = "reshape";
2223 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2224 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2225 type = "resync";
2226 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2227 type = "check";
2228 else
2229 type = "repair";
2230 } else
2231 type = "recover";
2232 }
2233 return sprintf(page, "%s\n", type);
2234}
2235
2236static ssize_t
7eec314d 2237action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2238{
7eec314d
N		 2239 if (!mddev->pers || !mddev->pers->sync_request)
2240 return -EINVAL;
2241
bce74dac 2242 if (cmd_match(page, "idle")) {
7eec314d
N		 2243 if (mddev->sync_thread) {
2244 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2245 md_unregister_thread(mddev->sync_thread);
2246 mddev->sync_thread = NULL;
2247 mddev->recovery = 0;
2248 }
03c902e1
N		 2249 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2250 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2251 return -EBUSY;
03c902e1 2252 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2253 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2254 else if (cmd_match(page, "reshape")) {
2255 int err;
2256 if (mddev->pers->start_reshape == NULL)
2257 return -EINVAL;
2258 err = mddev->pers->start_reshape(mddev);
2259 if (err)
2260 return err;
2261 } else {
bce74dac 2262 if (cmd_match(page, "check"))
7eec314d 2263 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2264 else if (!cmd_match(page, "repair"))
7eec314d
N		 2265 return -EINVAL;
2266 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2267 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2268 }
03c902e1 2269 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2270 md_wakeup_thread(mddev->thread);
2271 return len;
2272}
2273
9d88883e 2274static ssize_t
96de1e66 2275mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2276{
2277 return sprintf(page, "%llu\n",
2278 (unsigned long long) mddev->resync_mismatches);
2279}
2280
96de1e66 2281static struct md_sysfs_entry
7eec314d 2282md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2283
96de1e66
N		 2284
2285static struct md_sysfs_entry
2286md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2287
88202a0c
N		 2288static ssize_t
2289sync_min_show(mddev_t *mddev, char *page)
2290{
2291 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2292 mddev->sync_speed_min ? "local": "system");
2293}
2294
2295static ssize_t
2296sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2297{
2298 int min;
2299 char *e;
2300 if (strncmp(buf, "system", 6)==0) {
2301 mddev->sync_speed_min = 0;
2302 return len;
2303 }
2304 min = simple_strtoul(buf, &e, 10);
2305 if (buf == e || (*e && *e != '\n') || min <= 0)
2306 return -EINVAL;
2307 mddev->sync_speed_min = min;
2308 return len;
2309}
2310
2311static struct md_sysfs_entry md_sync_min =
2312__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2313
2314static ssize_t
2315sync_max_show(mddev_t *mddev, char *page)
2316{
2317 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2318 mddev->sync_speed_max ? "local": "system");
2319}
2320
2321static ssize_t
2322sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2323{
2324 int max;
2325 char *e;
2326 if (strncmp(buf, "system", 6)==0) {
2327 mddev->sync_speed_max = 0;
2328 return len;
2329 }
2330 max = simple_strtoul(buf, &e, 10);
2331 if (buf == e || (*e && *e != '\n') || max <= 0)
2332 return -EINVAL;
2333 mddev->sync_speed_max = max;
2334 return len;
2335}
2336
2337static struct md_sysfs_entry md_sync_max =
2338__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2339
2340
2341static ssize_t
2342sync_speed_show(mddev_t *mddev, char *page)
2343{
2344 unsigned long resync, dt, db;
2345 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2346 dt = ((jiffies - mddev->resync_mark) / HZ);
2347 if (!dt) dt++;
2348 db = resync - (mddev->resync_mark_cnt);
2349 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2350}
2351
2352static struct md_sysfs_entry
2353md_sync_speed = __ATTR_RO(sync_speed);
2354
2355static ssize_t
2356sync_completed_show(mddev_t *mddev, char *page)
2357{
2358 unsigned long max_blocks, resync;
2359
2360 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2361 max_blocks = mddev->resync_max_sectors;
2362 else
2363 max_blocks = mddev->size << 1;
2364
2365 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2366 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2367}
2368
2369static struct md_sysfs_entry
2370md_sync_completed = __ATTR_RO(sync_completed);
2371
e464eafd
N		 2372static ssize_t
2373suspend_lo_show(mddev_t *mddev, char *page)
2374{
2375 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2376}
2377
2378static ssize_t
2379suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2380{
2381 char *e;
2382 unsigned long long new = simple_strtoull(buf, &e, 10);
2383
2384 if (mddev->pers->quiesce == NULL)
2385 return -EINVAL;
2386 if (buf == e || (*e && *e != '\n'))
2387 return -EINVAL;
2388 if (new >= mddev->suspend_hi ||
2389 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2390 mddev->suspend_lo = new;
2391 mddev->pers->quiesce(mddev, 2);
2392 return len;
2393 } else
2394 return -EINVAL;
2395}
2396static struct md_sysfs_entry md_suspend_lo =
2397__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2398
2399
2400static ssize_t
2401suspend_hi_show(mddev_t *mddev, char *page)
2402{
2403 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2404}
2405
2406static ssize_t
2407suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2408{
2409 char *e;
2410 unsigned long long new = simple_strtoull(buf, &e, 10);
2411
2412 if (mddev->pers->quiesce == NULL)
2413 return -EINVAL;
2414 if (buf == e || (*e && *e != '\n'))
2415 return -EINVAL;
2416 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2417 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2418 mddev->suspend_hi = new;
2419 mddev->pers->quiesce(mddev, 1);
2420 mddev->pers->quiesce(mddev, 0);
2421 return len;
2422 } else
2423 return -EINVAL;
2424}
2425static struct md_sysfs_entry md_suspend_hi =
2426__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2427
2428
eae1701f
N		 2429static struct attribute *md_default_attrs[] = {
2430 &md_level.attr,
2431 &md_raid_disks.attr,
3b34380a 2432 &md_chunk_size.attr,
a35b0d69 2433 &md_size.attr,
8bb93aac 2434 &md_metadata.attr,
6d7ff738 2435 &md_new_device.attr,
411036fa
N		 2436 NULL,
2437};
2438
2439static struct attribute *md_redundancy_attrs[] = {
24dd469d 2440 &md_scan_mode.attr,
9d88883e 2441 &md_mismatches.attr,
88202a0c
N		 2442 &md_sync_min.attr,
2443 &md_sync_max.attr,
2444 &md_sync_speed.attr,
2445 &md_sync_completed.attr,
e464eafd
N		 2446 &md_suspend_lo.attr,
2447 &md_suspend_hi.attr,
eae1701f
N		 2448 NULL,
2449};
411036fa
N		 2450static struct attribute_group md_redundancy_group = {
2451 .name = NULL,
2452 .attrs = md_redundancy_attrs,
2453};
2454
eae1701f
N		 2455
2456static ssize_t
2457md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2458{
2459 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2460 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2461 ssize_t rv;
eae1701f
N		 2462
2463 if (!entry->show)
2464 return -EIO;
5dc5cf7d
IM		 2465 rv = mddev_lock(mddev);
2466 if (!rv) {
2467 rv = entry->show(mddev, page);
2468 mddev_unlock(mddev);
2469 }
96de1e66 2470 return rv;
eae1701f
N		 2471}
2472
2473static ssize_t
2474md_attr_store(struct kobject *kobj, struct attribute *attr,
2475 const char *page, size_t length)
2476{
2477 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2478 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2479 ssize_t rv;
eae1701f
N		 2480
2481 if (!entry->store)
2482 return -EIO;
5dc5cf7d
IM		 2483 rv = mddev_lock(mddev);
2484 if (!rv) {
2485 rv = entry->store(mddev, page, length);
2486 mddev_unlock(mddev);
2487 }
96de1e66 2488 return rv;
eae1701f
N		 2489}
2490
2491static void md_free(struct kobject *ko)
2492{
2493 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2494 kfree(mddev);
2495}
2496
2497static struct sysfs_ops md_sysfs_ops = {
2498 .show = md_attr_show,
2499 .store = md_attr_store,
2500};
2501static struct kobj_type md_ktype = {
2502 .release = md_free,
2503 .sysfs_ops = &md_sysfs_ops,
2504 .default_attrs = md_default_attrs,
2505};
2506
1da177e4
LT		 2507int mdp_major = 0;
2508
2509static struct kobject *md_probe(dev_t dev, int *part, void *data)
2510{
48c9c27b 2511 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 2512 mddev_t *mddev = mddev_find(dev);
2513 struct gendisk *disk;
2514 int partitioned = (MAJOR(dev) != MD_MAJOR);
2515 int shift = partitioned ? MdpMinorShift : 0;
2516 int unit = MINOR(dev) >> shift;
2517
2518 if (!mddev)
2519 return NULL;
2520
48c9c27b 2521 mutex_lock(&disks_mutex);
1da177e4 2522 if (mddev->gendisk) {
48c9c27b 2523 mutex_unlock(&disks_mutex);
1da177e4
LT		 2524 mddev_put(mddev);
2525 return NULL;
2526 }
2527 disk = alloc_disk(1 << shift);
2528 if (!disk) {
48c9c27b 2529 mutex_unlock(&disks_mutex);
1da177e4
LT		 2530 mddev_put(mddev);
2531 return NULL;
2532 }
2533 disk->major = MAJOR(dev);
2534 disk->first_minor = unit << shift;
2535 if (partitioned) {
2536 sprintf(disk->disk_name, "md_d%d", unit);
2537 sprintf(disk->devfs_name, "md/d%d", unit);
2538 } else {
2539 sprintf(disk->disk_name, "md%d", unit);
2540 sprintf(disk->devfs_name, "md/%d", unit);
2541 }
2542 disk->fops = &md_fops;
2543 disk->private_data = mddev;
2544 disk->queue = mddev->queue;
2545 add_disk(disk);
2546 mddev->gendisk = disk;
48c9c27b 2547 mutex_unlock(&disks_mutex);
9c791977 2548 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 2549 mddev->kobj.k_name = NULL;
2550 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2551 mddev->kobj.ktype = &md_ktype;
2552 kobject_register(&mddev->kobj);
1da177e4
LT		 2553 return NULL;
2554}
2555
2556void md_wakeup_thread(mdk_thread_t *thread);
2557
2558static void md_safemode_timeout(unsigned long data)
2559{
2560 mddev_t *mddev = (mddev_t *) data;
2561
2562 mddev->safemode = 1;
2563 md_wakeup_thread(mddev->thread);
2564}
2565
6ff8d8ec 2566static int start_dirty_degraded;
1da177e4
LT		 2567
2568static int do_md_run(mddev_t * mddev)
2569{
2604b703 2570 int err;
1da177e4
LT		 2571 int chunk_size;
2572 struct list_head *tmp;
2573 mdk_rdev_t *rdev;
2574 struct gendisk *disk;
2604b703 2575 struct mdk_personality *pers;
1da177e4
LT		 2576 char b[BDEVNAME_SIZE];
2577
a757e64c
N		 2578 if (list_empty(&mddev->disks))
2579 /* cannot run an array with no devices.. */
1da177e4 2580 return -EINVAL;
1da177e4
LT		 2581
2582 if (mddev->pers)
2583 return -EBUSY;
2584
2585 /*
2586 * Analyze all RAID superblock(s)
2587 */
a757e64c
N		 2588 if (!mddev->raid_disks)
2589 analyze_sbs(mddev);
1da177e4
LT		 2590
2591 chunk_size = mddev->chunk_size;
2604b703
N		 2592
2593 if (chunk_size) {
1da177e4
LT		 2594 if (chunk_size > MAX_CHUNK_SIZE) {
2595 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2596 chunk_size, MAX_CHUNK_SIZE);
2597 return -EINVAL;
2598 }
2599 /*
2600 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2601 */
2602 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 2603 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 2604 return -EINVAL;
2605 }
2606 if (chunk_size < PAGE_SIZE) {
2607 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2608 chunk_size, PAGE_SIZE);
2609 return -EINVAL;
2610 }
2611
2612 /* devices must have minimum size of one chunk */
2613 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2614 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2615 continue;
2616 if (rdev->size < chunk_size / 1024) {
2617 printk(KERN_WARNING
2618 "md: Dev %s smaller than chunk_size:"
2619 " %lluk < %dk\n",
2620 bdevname(rdev->bdev,b),
2621 (unsigned long long)rdev->size,
2622 chunk_size / 1024);
2623 return -EINVAL;
2624 }
2625 }
2626 }
2627
1da177e4 2628#ifdef CONFIG_KMOD
d9d166c2
N		 2629 if (mddev->level != LEVEL_NONE)
2630 request_module("md-level-%d", mddev->level);
2631 else if (mddev->clevel[0])
2632 request_module("md-%s", mddev->clevel);
1da177e4
LT		 2633#endif
2634
2635 /*
2636 * Drop all container device buffers, from now on
2637 * the only valid external interface is through the md
2638 * device.
2639 * Also find largest hardsector size
2640 */
2641 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2642 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2643 continue;
2644 sync_blockdev(rdev->bdev);
2645 invalidate_bdev(rdev->bdev, 0);
2646 }
2647
2648 md_probe(mddev->unit, NULL, NULL);
2649 disk = mddev->gendisk;
2650 if (!disk)
2651 return -ENOMEM;
2652
2653 spin_lock(&pers_lock);
d9d166c2 2654 pers = find_pers(mddev->level, mddev->clevel);
2604b703 2655 if (!pers || !try_module_get(pers->owner)) {
1da177e4 2656 spin_unlock(&pers_lock);
d9d166c2
N		 2657 if (mddev->level != LEVEL_NONE)
2658 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2659 mddev->level);
2660 else
2661 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2662 mddev->clevel);
1da177e4
LT		 2663 return -EINVAL;
2664 }
2604b703 2665 mddev->pers = pers;
1da177e4 2666 spin_unlock(&pers_lock);
d9d166c2
N		 2667 mddev->level = pers->level;
2668 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 2669
f6705578 2670 if (mddev->reshape_position != MaxSector &&
63c70c4f 2671 pers->start_reshape == NULL) {
f6705578
N		 2672 /* This personality cannot handle reshaping... */
2673 mddev->pers = NULL;
2674 module_put(pers->owner);
2675 return -EINVAL;
2676 }
2677
657390d2 2678 mddev->recovery = 0;
1da177e4 2679 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 2680 mddev->barriers_work = 1;
6ff8d8ec 2681 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 2682
f91de92e
N		 2683 if (start_readonly)
2684 mddev->ro = 2; /* read-only, but switch on first write */
2685
b15c2e57
N		 2686 err = mddev->pers->run(mddev);
2687 if (!err && mddev->pers->sync_request) {
2688 err = bitmap_create(mddev);
2689 if (err) {
2690 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2691 mdname(mddev), err);
2692 mddev->pers->stop(mddev);
2693 }
2694 }
1da177e4
LT		 2695 if (err) {
2696 printk(KERN_ERR "md: pers->run() failed ...\n");
2697 module_put(mddev->pers->owner);
2698 mddev->pers = NULL;
32a7627c
N		 2699 bitmap_destroy(mddev);
2700 return err;
1da177e4 2701 }
411036fa
N		 2702 if (mddev->pers->sync_request)
2703 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
fd9d49ca
N		 2704 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2705 mddev->ro = 0;
2706
1da177e4
LT		 2707 atomic_set(&mddev->writes_pending,0);
2708 mddev->safemode = 0;
2709 mddev->safemode_timer.function = md_safemode_timeout;
2710 mddev->safemode_timer.data = (unsigned long) mddev;
2711 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2712 mddev->in_sync = 1;
86e6ffdd
N		 2713
2714 ITERATE_RDEV(mddev,rdev,tmp)
2715 if (rdev->raid_disk >= 0) {
2716 char nm[20];
2717 sprintf(nm, "rd%d", rdev->raid_disk);
2718 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2719 }
1da177e4
LT		 2720
2721 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2722 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2723
2724 if (mddev->sb_dirty)
2725 md_update_sb(mddev);
2726
2727 set_capacity(disk, mddev->array_size<<1);
2728
2729 /* If we call blk_queue_make_request here, it will
2730 * re-initialise max_sectors etc which may have been
2731 * refined inside -> run. So just set the bits we need to set.
2732 * Most initialisation happended when we called
2733 * blk_queue_make_request(..., md_fail_request)
2734 * earlier.
2735 */
2736 mddev->queue->queuedata = mddev;
2737 mddev->queue->make_request_fn = mddev->pers->make_request;
2738
2739 mddev->changed = 1;
d7603b7e 2740 md_new_event(mddev);
1da177e4
LT		 2741 return 0;
2742}
2743
2744static int restart_array(mddev_t *mddev)
2745{
2746 struct gendisk *disk = mddev->gendisk;
2747 int err;
2748
2749 /*
2750 * Complain if it has no devices
2751 */
2752 err = -ENXIO;
2753 if (list_empty(&mddev->disks))
2754 goto out;
2755
2756 if (mddev->pers) {
2757 err = -EBUSY;
2758 if (!mddev->ro)
2759 goto out;
2760
2761 mddev->safemode = 0;
2762 mddev->ro = 0;
2763 set_disk_ro(disk, 0);
2764
2765 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2766 mdname(mddev));
2767 /*
2768 * Kick recovery or resync if necessary
2769 */
2770 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2771 md_wakeup_thread(mddev->thread);
2772 err = 0;
2773 } else {
2774 printk(KERN_ERR "md: %s has no personality assigned.\n",
2775 mdname(mddev));
2776 err = -EINVAL;
2777 }
2778
2779out:
2780 return err;
2781}
2782
2783static int do_md_stop(mddev_t * mddev, int ro)
2784{
2785 int err = 0;
2786 struct gendisk *disk = mddev->gendisk;
2787
2788 if (mddev->pers) {
2789 if (atomic_read(&mddev->active)>2) {
2790 printk("md: %s still in use.\n",mdname(mddev));
2791 return -EBUSY;
2792 }
2793
2794 if (mddev->sync_thread) {
2795 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2796 md_unregister_thread(mddev->sync_thread);
2797 mddev->sync_thread = NULL;
2798 }
2799
2800 del_timer_sync(&mddev->safemode_timer);
2801
2802 invalidate_partition(disk, 0);
2803
2804 if (ro) {
2805 err = -ENXIO;
f91de92e 2806 if (mddev->ro==1)
1da177e4
LT		 2807 goto out;
2808 mddev->ro = 1;
2809 } else {
6b8b3e8a 2810 bitmap_flush(mddev);
a9701a30 2811 md_super_wait(mddev);
1da177e4
LT		 2812 if (mddev->ro)
2813 set_disk_ro(disk, 0);
2814 blk_queue_make_request(mddev->queue, md_fail_request);
2815 mddev->pers->stop(mddev);
411036fa
N		 2816 if (mddev->pers->sync_request)
2817 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2818
1da177e4
LT		 2819 module_put(mddev->pers->owner);
2820 mddev->pers = NULL;
2821 if (mddev->ro)
2822 mddev->ro = 0;
2823 }
2824 if (!mddev->in_sync) {
2825 /* mark array as shutdown cleanly */
2826 mddev->in_sync = 1;
2827 md_update_sb(mddev);
2828 }
2829 if (ro)
2830 set_disk_ro(disk, 1);
2831 }
32a7627c 2832
1da177e4
LT		 2833 /*
2834 * Free resources if final stop
2835 */
2836 if (!ro) {
86e6ffdd
N		 2837 mdk_rdev_t *rdev;
2838 struct list_head *tmp;
1da177e4
LT		 2839 struct gendisk *disk;
2840 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2841
978f946b
N		 2842 bitmap_destroy(mddev);
2843 if (mddev->bitmap_file) {
2844 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2845 fput(mddev->bitmap_file);
2846 mddev->bitmap_file = NULL;
2847 }
2848 mddev->bitmap_offset = 0;
2849
86e6ffdd
N		 2850 ITERATE_RDEV(mddev,rdev,tmp)
2851 if (rdev->raid_disk >= 0) {
2852 char nm[20];
2853 sprintf(nm, "rd%d", rdev->raid_disk);
2854 sysfs_remove_link(&mddev->kobj, nm);
2855 }
2856
1da177e4
LT		 2857 export_array(mddev);
2858
2859 mddev->array_size = 0;
2860 disk = mddev->gendisk;
2861 if (disk)
2862 set_capacity(disk, 0);
2863 mddev->changed = 1;
2864 } else
2865 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2866 mdname(mddev));
2867 err = 0;
d7603b7e 2868 md_new_event(mddev);
1da177e4
LT		 2869out:
2870 return err;
2871}
2872
2873static void autorun_array(mddev_t *mddev)
2874{
2875 mdk_rdev_t *rdev;
2876 struct list_head *tmp;
2877 int err;
2878
a757e64c 2879 if (list_empty(&mddev->disks))
1da177e4 2880 return;
1da177e4
LT		 2881
2882 printk(KERN_INFO "md: running: ");
2883
2884 ITERATE_RDEV(mddev,rdev,tmp) {
2885 char b[BDEVNAME_SIZE];
2886 printk("<%s>", bdevname(rdev->bdev,b));
2887 }
2888 printk("\n");
2889
2890 err = do_md_run (mddev);
2891 if (err) {
2892 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2893 do_md_stop (mddev, 0);
2894 }
2895}
2896
2897/*
2898 * lets try to run arrays based on all disks that have arrived
2899 * until now. (those are in pending_raid_disks)
2900 *
2901 * the method: pick the first pending disk, collect all disks with
2902 * the same UUID, remove all from the pending list and put them into
2903 * the 'same_array' list. Then order this list based on superblock
2904 * update time (freshest comes first), kick out 'old' disks and
2905 * compare superblocks. If everything's fine then run it.
2906 *
2907 * If "unit" is allocated, then bump its reference count
2908 */
2909static void autorun_devices(int part)
2910{
1da177e4
LT		 2911 struct list_head *tmp;
2912 mdk_rdev_t *rdev0, *rdev;
2913 mddev_t *mddev;
2914 char b[BDEVNAME_SIZE];
2915
2916 printk(KERN_INFO "md: autorun ...\n");
2917 while (!list_empty(&pending_raid_disks)) {
2918 dev_t dev;
ad01c9e3 2919 LIST_HEAD(candidates);
1da177e4
LT		 2920 rdev0 = list_entry(pending_raid_disks.next,
2921 mdk_rdev_t, same_set);
2922
2923 printk(KERN_INFO "md: considering %s ...\n",
2924 bdevname(rdev0->bdev,b));
2925 INIT_LIST_HEAD(&candidates);
2926 ITERATE_RDEV_PENDING(rdev,tmp)
2927 if (super_90_load(rdev, rdev0, 0) >= 0) {
2928 printk(KERN_INFO "md: adding %s ...\n",
2929 bdevname(rdev->bdev,b));
2930 list_move(&rdev->same_set, &candidates);
2931 }
2932 /*
2933 * now we have a set of devices, with all of them having
2934 * mostly sane superblocks. It's time to allocate the
2935 * mddev.
2936 */
2937 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2938 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2939 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2940 break;
2941 }
2942 if (part)
2943 dev = MKDEV(mdp_major,
2944 rdev0->preferred_minor << MdpMinorShift);
2945 else
2946 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2947
2948 md_probe(dev, NULL, NULL);
2949 mddev = mddev_find(dev);
2950 if (!mddev) {
2951 printk(KERN_ERR
2952 "md: cannot allocate memory for md drive.\n");
2953 break;
2954 }
2955 if (mddev_lock(mddev))
2956 printk(KERN_WARNING "md: %s locked, cannot run\n",
2957 mdname(mddev));
2958 else if (mddev->raid_disks || mddev->major_version
2959 || !list_empty(&mddev->disks)) {
2960 printk(KERN_WARNING
2961 "md: %s already running, cannot run %s\n",
2962 mdname(mddev), bdevname(rdev0->bdev,b));
2963 mddev_unlock(mddev);
2964 } else {
2965 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2966 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2967 list_del_init(&rdev->same_set);
2968 if (bind_rdev_to_array(rdev, mddev))
2969 export_rdev(rdev);
2970 }
2971 autorun_array(mddev);
2972 mddev_unlock(mddev);
2973 }
2974 /* on success, candidates will be empty, on error
2975 * it won't...
2976 */
2977 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2978 export_rdev(rdev);
2979 mddev_put(mddev);
2980 }
2981 printk(KERN_INFO "md: ... autorun DONE.\n");
2982}
2983
2984/*
2985 * import RAID devices based on one partition
2986 * if possible, the array gets run as well.
2987 */
2988
2989static int autostart_array(dev_t startdev)
2990{
2991 char b[BDEVNAME_SIZE];
2992 int err = -EINVAL, i;
2993 mdp_super_t *sb = NULL;
2994 mdk_rdev_t *start_rdev = NULL, *rdev;
2995
2996 start_rdev = md_import_device(startdev, 0, 0);
2997 if (IS_ERR(start_rdev))
2998 return err;
2999
3000
3001 /* NOTE: this can only work for 0.90.0 superblocks */
3002 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
3003 if (sb->major_version != 0 ||
3004 sb->minor_version != 90 ) {
3005 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
3006 export_rdev(start_rdev);
3007 return err;
3008 }
3009
b2d444d7 3010 if (test_bit(Faulty, &start_rdev->flags)) {
1da177e4
LT		 3011 printk(KERN_WARNING
3012 "md: can not autostart based on faulty %s!\n",
3013 bdevname(start_rdev->bdev,b));
3014 export_rdev(start_rdev);
3015 return err;
3016 }
3017 list_add(&start_rdev->same_set, &pending_raid_disks);
3018
3019 for (i = 0; i < MD_SB_DISKS; i++) {
3020 mdp_disk_t *desc = sb->disks + i;
3021 dev_t dev = MKDEV(desc->major, desc->minor);
3022
3023 if (!dev)
3024 continue;
3025 if (dev == startdev)
3026 continue;
3027 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3028 continue;
3029 rdev = md_import_device(dev, 0, 0);
3030 if (IS_ERR(rdev))
3031 continue;
3032
3033 list_add(&rdev->same_set, &pending_raid_disks);
3034 }
3035
3036 /*
3037 * possibly return codes
3038 */
3039 autorun_devices(0);
3040 return 0;
3041
3042}
3043
3044
3045static int get_version(void __user * arg)
3046{
3047 mdu_version_t ver;
3048
3049 ver.major = MD_MAJOR_VERSION;
3050 ver.minor = MD_MINOR_VERSION;
3051 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3052
3053 if (copy_to_user(arg, &ver, sizeof(ver)))
3054 return -EFAULT;
3055
3056 return 0;
3057}
3058
3059static int get_array_info(mddev_t * mddev, void __user * arg)
3060{
3061 mdu_array_info_t info;
3062 int nr,working,active,failed,spare;
3063 mdk_rdev_t *rdev;
3064 struct list_head *tmp;
3065
3066 nr=working=active=failed=spare=0;
3067 ITERATE_RDEV(mddev,rdev,tmp) {
3068 nr++;
b2d444d7 3069 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3070 failed++;
3071 else {
3072 working++;
b2d444d7 3073 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3074 active++;
3075 else
3076 spare++;
3077 }
3078 }
3079
3080 info.major_version = mddev->major_version;
3081 info.minor_version = mddev->minor_version;
3082 info.patch_version = MD_PATCHLEVEL_VERSION;
3083 info.ctime = mddev->ctime;
3084 info.level = mddev->level;
3085 info.size = mddev->size;
284ae7ca
N		 3086 if (info.size != mddev->size) /* overflow */
3087 info.size = -1;
1da177e4
LT		 3088 info.nr_disks = nr;
3089 info.raid_disks = mddev->raid_disks;
3090 info.md_minor = mddev->md_minor;
3091 info.not_persistent= !mddev->persistent;
3092
3093 info.utime = mddev->utime;
3094 info.state = 0;
3095 if (mddev->in_sync)
3096 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3097 if (mddev->bitmap && mddev->bitmap_offset)
3098 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3099 info.active_disks = active;
3100 info.working_disks = working;
3101 info.failed_disks = failed;
3102 info.spare_disks = spare;
3103
3104 info.layout = mddev->layout;
3105 info.chunk_size = mddev->chunk_size;
3106
3107 if (copy_to_user(arg, &info, sizeof(info)))
3108 return -EFAULT;
3109
3110 return 0;
3111}
3112
87162a28 3113static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3114{
3115 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3116 char *ptr, *buf = NULL;
3117 int err = -ENOMEM;
3118
3119 file = kmalloc(sizeof(*file), GFP_KERNEL);
3120 if (!file)
3121 goto out;
3122
3123 /* bitmap disabled, zero the first byte and copy out */
3124 if (!mddev->bitmap || !mddev->bitmap->file) {
3125 file->pathname[0] = '\0';
3126 goto copy_out;
3127 }
3128
3129 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3130 if (!buf)
3131 goto out;
3132
3133 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3134 if (!ptr)
3135 goto out;
3136
3137 strcpy(file->pathname, ptr);
3138
3139copy_out:
3140 err = 0;
3141 if (copy_to_user(arg, file, sizeof(*file)))
3142 err = -EFAULT;
3143out:
3144 kfree(buf);
3145 kfree(file);
3146 return err;
3147}
3148
1da177e4
LT		 3149static int get_disk_info(mddev_t * mddev, void __user * arg)
3150{
3151 mdu_disk_info_t info;
3152 unsigned int nr;
3153 mdk_rdev_t *rdev;
3154
3155 if (copy_from_user(&info, arg, sizeof(info)))
3156 return -EFAULT;
3157
3158 nr = info.number;
3159
3160 rdev = find_rdev_nr(mddev, nr);
3161 if (rdev) {
3162 info.major = MAJOR(rdev->bdev->bd_dev);
3163 info.minor = MINOR(rdev->bdev->bd_dev);
3164 info.raid_disk = rdev->raid_disk;
3165 info.state = 0;
b2d444d7 3166 if (test_bit(Faulty, &rdev->flags))
1da177e4 3167 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3168 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3169 info.state |= (1<<MD_DISK_ACTIVE);
3170 info.state |= (1<<MD_DISK_SYNC);
3171 }
8ddf9efe
N		 3172 if (test_bit(WriteMostly, &rdev->flags))
3173 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3174 } else {
3175 info.major = info.minor = 0;
3176 info.raid_disk = -1;
3177 info.state = (1<<MD_DISK_REMOVED);
3178 }
3179
3180 if (copy_to_user(arg, &info, sizeof(info)))
3181 return -EFAULT;
3182
3183 return 0;
3184}
3185
3186static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3187{
3188 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3189 mdk_rdev_t *rdev;
3190 dev_t dev = MKDEV(info->major,info->minor);
3191
3192 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3193 return -EOVERFLOW;
3194
3195 if (!mddev->raid_disks) {
3196 int err;
3197 /* expecting a device which has a superblock */
3198 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3199 if (IS_ERR(rdev)) {
3200 printk(KERN_WARNING
3201 "md: md_import_device returned %ld\n",
3202 PTR_ERR(rdev));
3203 return PTR_ERR(rdev);
3204 }
3205 if (!list_empty(&mddev->disks)) {
3206 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3207 mdk_rdev_t, same_set);
3208 int err = super_types[mddev->major_version]
3209 .load_super(rdev, rdev0, mddev->minor_version);
3210 if (err < 0) {
3211 printk(KERN_WARNING
3212 "md: %s has different UUID to %s\n",
3213 bdevname(rdev->bdev,b),
3214 bdevname(rdev0->bdev,b2));
3215 export_rdev(rdev);
3216 return -EINVAL;
3217 }
3218 }
3219 err = bind_rdev_to_array(rdev, mddev);
3220 if (err)
3221 export_rdev(rdev);
3222 return err;
3223 }
3224
3225 /*
3226 * add_new_disk can be used once the array is assembled
3227 * to add "hot spares". They must already have a superblock
3228 * written
3229 */
3230 if (mddev->pers) {
3231 int err;
3232 if (!mddev->pers->hot_add_disk) {
3233 printk(KERN_WARNING
3234 "%s: personality does not support diskops!\n",
3235 mdname(mddev));
3236 return -EINVAL;
3237 }
7b1e35f6
N		 3238 if (mddev->persistent)
3239 rdev = md_import_device(dev, mddev->major_version,
3240 mddev->minor_version);
3241 else
3242 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 3243 if (IS_ERR(rdev)) {
3244 printk(KERN_WARNING
3245 "md: md_import_device returned %ld\n",
3246 PTR_ERR(rdev));
3247 return PTR_ERR(rdev);
3248 }
41158c7e
N		 3249 /* set save_raid_disk if appropriate */
3250 if (!mddev->persistent) {
3251 if (info->state & (1<<MD_DISK_SYNC) &&
3252 info->raid_disk < mddev->raid_disks)
3253 rdev->raid_disk = info->raid_disk;
3254 else
3255 rdev->raid_disk = -1;
3256 } else
3257 super_types[mddev->major_version].
3258 validate_super(mddev, rdev);
3259 rdev->saved_raid_disk = rdev->raid_disk;
3260
b2d444d7 3261 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 3262 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3263 set_bit(WriteMostly, &rdev->flags);
3264
1da177e4
LT		 3265 rdev->raid_disk = -1;
3266 err = bind_rdev_to_array(rdev, mddev);
3267 if (err)
3268 export_rdev(rdev);
c361777f
N		 3269
3270 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 3271 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3272 return err;
3273 }
3274
3275 /* otherwise, add_new_disk is only allowed
3276 * for major_version==0 superblocks
3277 */
3278 if (mddev->major_version != 0) {
3279 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3280 mdname(mddev));
3281 return -EINVAL;
3282 }
3283
3284 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3285 int err;
3286 rdev = md_import_device (dev, -1, 0);
3287 if (IS_ERR(rdev)) {
3288 printk(KERN_WARNING
3289 "md: error, md_import_device() returned %ld\n",
3290 PTR_ERR(rdev));
3291 return PTR_ERR(rdev);
3292 }
3293 rdev->desc_nr = info->number;
3294 if (info->raid_disk < mddev->raid_disks)
3295 rdev->raid_disk = info->raid_disk;
3296 else
3297 rdev->raid_disk = -1;
3298
b2d444d7
N		 3299 rdev->flags = 0;
3300
1da177e4 3301 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 3302 if (info->state & (1<<MD_DISK_SYNC))
3303 set_bit(In_sync, &rdev->flags);
1da177e4 3304
8ddf9efe
N		 3305 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3306 set_bit(WriteMostly, &rdev->flags);
3307
1da177e4
LT		 3308 if (!mddev->persistent) {
3309 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3310 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3311 } else
3312 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3313 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3314
2bf071bf
N		 3315 err = bind_rdev_to_array(rdev, mddev);
3316 if (err) {
3317 export_rdev(rdev);
3318 return err;
3319 }
1da177e4
LT		 3320 }
3321
3322 return 0;
3323}
3324
3325static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3326{
3327 char b[BDEVNAME_SIZE];
3328 mdk_rdev_t *rdev;
3329
3330 if (!mddev->pers)
3331 return -ENODEV;
3332
3333 rdev = find_rdev(mddev, dev);
3334 if (!rdev)
3335 return -ENXIO;
3336
3337 if (rdev->raid_disk >= 0)
3338 goto busy;
3339
3340 kick_rdev_from_array(rdev);
3341 md_update_sb(mddev);
d7603b7e 3342 md_new_event(mddev);
1da177e4
LT		 3343
3344 return 0;
3345busy:
3346 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3347 bdevname(rdev->bdev,b), mdname(mddev));
3348 return -EBUSY;
3349}
3350
3351static int hot_add_disk(mddev_t * mddev, dev_t dev)
3352{
3353 char b[BDEVNAME_SIZE];
3354 int err;
3355 unsigned int size;
3356 mdk_rdev_t *rdev;
3357
3358 if (!mddev->pers)
3359 return -ENODEV;
3360
3361 if (mddev->major_version != 0) {
3362 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3363 " version-0 superblocks.\n",
3364 mdname(mddev));
3365 return -EINVAL;
3366 }
3367 if (!mddev->pers->hot_add_disk) {
3368 printk(KERN_WARNING
3369 "%s: personality does not support diskops!\n",
3370 mdname(mddev));
3371 return -EINVAL;
3372 }
3373
3374 rdev = md_import_device (dev, -1, 0);
3375 if (IS_ERR(rdev)) {
3376 printk(KERN_WARNING
3377 "md: error, md_import_device() returned %ld\n",
3378 PTR_ERR(rdev));
3379 return -EINVAL;
3380 }
3381
3382 if (mddev->persistent)
3383 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3384 else
3385 rdev->sb_offset =
3386 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3387
3388 size = calc_dev_size(rdev, mddev->chunk_size);
3389 rdev->size = size;
3390
b2d444d7 3391 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3392 printk(KERN_WARNING
3393 "md: can not hot-add faulty %s disk to %s!\n",
3394 bdevname(rdev->bdev,b), mdname(mddev));
3395 err = -EINVAL;
3396 goto abort_export;
3397 }
b2d444d7 3398 clear_bit(In_sync, &rdev->flags);
1da177e4 3399 rdev->desc_nr = -1;
2bf071bf
N		 3400 err = bind_rdev_to_array(rdev, mddev);
3401 if (err)
3402 goto abort_export;
1da177e4
LT		 3403
3404 /*
3405 * The rest should better be atomic, we can have disk failures
3406 * noticed in interrupt contexts ...
3407 */
3408
3409 if (rdev->desc_nr == mddev->max_disks) {
3410 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3411 mdname(mddev));
3412 err = -EBUSY;
3413 goto abort_unbind_export;
3414 }
3415
3416 rdev->raid_disk = -1;
3417
3418 md_update_sb(mddev);
3419
3420 /*
3421 * Kick recovery, maybe this spare has to be added to the
3422 * array immediately.
3423 */
3424 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3425 md_wakeup_thread(mddev->thread);
d7603b7e 3426 md_new_event(mddev);
1da177e4
LT		 3427 return 0;
3428
3429abort_unbind_export:
3430 unbind_rdev_from_array(rdev);
3431
3432abort_export:
3433 export_rdev(rdev);
3434 return err;
3435}
3436
32a7627c
N		 3437/* similar to deny_write_access, but accounts for our holding a reference
3438 * to the file ourselves */
3439static int deny_bitmap_write_access(struct file * file)
3440{
3441 struct inode *inode = file->f_mapping->host;
3442
3443 spin_lock(&inode->i_lock);
3444 if (atomic_read(&inode->i_writecount) > 1) {
3445 spin_unlock(&inode->i_lock);
3446 return -ETXTBSY;
3447 }
3448 atomic_set(&inode->i_writecount, -1);
3449 spin_unlock(&inode->i_lock);
3450
3451 return 0;
3452}
3453
3454static int set_bitmap_file(mddev_t *mddev, int fd)
3455{
3456 int err;
3457
36fa3063
N		 3458 if (mddev->pers) {
3459 if (!mddev->pers->quiesce)
3460 return -EBUSY;
3461 if (mddev->recovery || mddev->sync_thread)
3462 return -EBUSY;
3463 /* we should be able to change the bitmap.. */
3464 }
32a7627c 3465
32a7627c 3466
36fa3063
N		 3467 if (fd >= 0) {
3468 if (mddev->bitmap)
3469 return -EEXIST; /* cannot add when bitmap is present */
3470 mddev->bitmap_file = fget(fd);
32a7627c 3471
36fa3063
N		 3472 if (mddev->bitmap_file == NULL) {
3473 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3474 mdname(mddev));
3475 return -EBADF;
3476 }
3477
3478 err = deny_bitmap_write_access(mddev->bitmap_file);
3479 if (err) {
3480 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3481 mdname(mddev));
3482 fput(mddev->bitmap_file);
3483 mddev->bitmap_file = NULL;
3484 return err;
3485 }
a654b9d8 3486 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 3487 } else if (mddev->bitmap == NULL)
3488 return -ENOENT; /* cannot remove what isn't there */
3489 err = 0;
3490 if (mddev->pers) {
3491 mddev->pers->quiesce(mddev, 1);
3492 if (fd >= 0)
3493 err = bitmap_create(mddev);
3494 if (fd < 0 || err)
3495 bitmap_destroy(mddev);
3496 mddev->pers->quiesce(mddev, 0);
3497 } else if (fd < 0) {
3498 if (mddev->bitmap_file)
3499 fput(mddev->bitmap_file);
3500 mddev->bitmap_file = NULL;
3501 }
3502
32a7627c
N		 3503 return err;
3504}
3505
1da177e4
LT		 3506/*
3507 * set_array_info is used two different ways
3508 * The original usage is when creating a new array.
3509 * In this usage, raid_disks is > 0 and it together with
3510 * level, size, not_persistent,layout,chunksize determine the
3511 * shape of the array.
3512 * This will always create an array with a type-0.90.0 superblock.
3513 * The newer usage is when assembling an array.
3514 * In this case raid_disks will be 0, and the major_version field is
3515 * use to determine which style super-blocks are to be found on the devices.
3516 * The minor and patch _version numbers are also kept incase the
3517 * super_block handler wishes to interpret them.
3518 */
3519static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3520{
3521
3522 if (info->raid_disks == 0) {
3523 /* just setting version number for superblock loading */
3524 if (info->major_version < 0 ||
3525 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3526 super_types[info->major_version].name == NULL) {
3527 /* maybe try to auto-load a module? */
3528 printk(KERN_INFO
3529 "md: superblock version %d not known\n",
3530 info->major_version);
3531 return -EINVAL;
3532 }
3533 mddev->major_version = info->major_version;
3534 mddev->minor_version = info->minor_version;
3535 mddev->patch_version = info->patch_version;
3536 return 0;
3537 }
3538 mddev->major_version = MD_MAJOR_VERSION;
3539 mddev->minor_version = MD_MINOR_VERSION;
3540 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3541 mddev->ctime = get_seconds();
3542
3543 mddev->level = info->level;
17115e03 3544 mddev->clevel[0] = 0;
1da177e4
LT		 3545 mddev->size = info->size;
3546 mddev->raid_disks = info->raid_disks;
3547 /* don't set md_minor, it is determined by which /dev/md* was
3548 * openned
3549 */
3550 if (info->state & (1<<MD_SB_CLEAN))
3551 mddev->recovery_cp = MaxSector;
3552 else
3553 mddev->recovery_cp = 0;
3554 mddev->persistent = ! info->not_persistent;
3555
3556 mddev->layout = info->layout;
3557 mddev->chunk_size = info->chunk_size;
3558
3559 mddev->max_disks = MD_SB_DISKS;
3560
3561 mddev->sb_dirty = 1;
3562
b2a2703c
N		 3563 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3564 mddev->bitmap_offset = 0;
3565
f6705578
N		 3566 mddev->reshape_position = MaxSector;
3567
1da177e4
LT		 3568 /*
3569 * Generate a 128 bit UUID
3570 */
3571 get_random_bytes(mddev->uuid, 16);
3572
f6705578
N		 3573 mddev->new_level = mddev->level;
3574 mddev->new_chunk = mddev->chunk_size;
3575 mddev->new_layout = mddev->layout;
3576 mddev->delta_disks = 0;
3577
1da177e4
LT		 3578 return 0;
3579}
3580
a35b0d69
N		 3581static int update_size(mddev_t *mddev, unsigned long size)
3582{
3583 mdk_rdev_t * rdev;
3584 int rv;
3585 struct list_head *tmp;
8ddeeae5 3586 int fit = (size == 0);
a35b0d69
N		 3587
3588 if (mddev->pers->resize == NULL)
3589 return -EINVAL;
3590 /* The "size" is the amount of each device that is used.
3591 * This can only make sense for arrays with redundancy.
3592 * linear and raid0 always use whatever space is available
3593 * We can only consider changing the size if no resync
3594 * or reconstruction is happening, and if the new size
3595 * is acceptable. It must fit before the sb_offset or,
3596 * if that is <data_offset, it must fit before the
3597 * size of each device.
3598 * If size is zero, we find the largest size that fits.
3599 */
3600 if (mddev->sync_thread)
3601 return -EBUSY;
3602 ITERATE_RDEV(mddev,rdev,tmp) {
3603 sector_t avail;
a35b0d69
N		 3604 if (rdev->sb_offset > rdev->data_offset)
3605 avail = (rdev->sb_offset*2) - rdev->data_offset;
3606 else
3607 avail = get_capacity(rdev->bdev->bd_disk)
3608 - rdev->data_offset;
3609 if (fit && (size == 0 || size > avail/2))
3610 size = avail/2;
3611 if (avail < ((sector_t)size << 1))
3612 return -ENOSPC;
3613 }
3614 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3615 if (!rv) {
3616 struct block_device *bdev;
3617
3618 bdev = bdget_disk(mddev->gendisk, 0);
3619 if (bdev) {
1b1dcc1b 3620 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 3621 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 3622 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 3623 bdput(bdev);
3624 }
3625 }
3626 return rv;
3627}
3628
da943b99
N		 3629static int update_raid_disks(mddev_t *mddev, int raid_disks)
3630{
3631 int rv;
3632 /* change the number of raid disks */
63c70c4f 3633 if (mddev->pers->check_reshape == NULL)
da943b99
N		 3634 return -EINVAL;
3635 if (raid_disks <= 0 ||
3636 raid_disks >= mddev->max_disks)
3637 return -EINVAL;
63c70c4f 3638 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 3639 return -EBUSY;
63c70c4f
N		 3640 mddev->delta_disks = raid_disks - mddev->raid_disks;
3641
3642 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 3643 return rv;
3644}
3645
3646
1da177e4
LT		 3647/*
3648 * update_array_info is used to change the configuration of an
3649 * on-line array.
3650 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3651 * fields in the info are checked against the array.
3652 * Any differences that cannot be handled will cause an error.
3653 * Normally, only one change can be managed at a time.
3654 */
3655static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3656{
3657 int rv = 0;
3658 int cnt = 0;
36fa3063
N		 3659 int state = 0;
3660
3661 /* calculate expected state,ignoring low bits */
3662 if (mddev->bitmap && mddev->bitmap_offset)
3663 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3664
3665 if (mddev->major_version != info->major_version ||
3666 mddev->minor_version != info->minor_version ||
3667/* mddev->patch_version != info->patch_version || */
3668 mddev->ctime != info->ctime ||
3669 mddev->level != info->level ||
3670/* mddev->layout != info->layout || */
3671 !mddev->persistent != info->not_persistent||
36fa3063
N		 3672 mddev->chunk_size != info->chunk_size ||
3673 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3674 ((state^info->state) & 0xfffffe00)
3675 )
1da177e4
LT		 3676 return -EINVAL;
3677 /* Check there is only one change */
284ae7ca 3678 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 3679 if (mddev->raid_disks != info->raid_disks) cnt++;
3680 if (mddev->layout != info->layout) cnt++;
36fa3063 3681 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 3682 if (cnt == 0) return 0;
3683 if (cnt > 1) return -EINVAL;
3684
3685 if (mddev->layout != info->layout) {
3686 /* Change layout
3687 * we don't need to do anything at the md level, the
3688 * personality will take care of it all.
3689 */
3690 if (mddev->pers->reconfig == NULL)
3691 return -EINVAL;
3692 else
3693 return mddev->pers->reconfig(mddev, info->layout, -1);
3694 }
284ae7ca 3695 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 3696 rv = update_size(mddev, info->size);
3697
da943b99
N		 3698 if (mddev->raid_disks != info->raid_disks)
3699 rv = update_raid_disks(mddev, info->raid_disks);
3700
36fa3063
N		 3701 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3702 if (mddev->pers->quiesce == NULL)
3703 return -EINVAL;
3704 if (mddev->recovery || mddev->sync_thread)
3705 return -EBUSY;
3706 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3707 /* add the bitmap */
3708 if (mddev->bitmap)
3709 return -EEXIST;
3710 if (mddev->default_bitmap_offset == 0)
3711 return -EINVAL;
3712 mddev->bitmap_offset = mddev->default_bitmap_offset;
3713 mddev->pers->quiesce(mddev, 1);
3714 rv = bitmap_create(mddev);
3715 if (rv)
3716 bitmap_destroy(mddev);
3717 mddev->pers->quiesce(mddev, 0);
3718 } else {
3719 /* remove the bitmap */
3720 if (!mddev->bitmap)
3721 return -ENOENT;
3722 if (mddev->bitmap->file)
3723 return -EINVAL;
3724 mddev->pers->quiesce(mddev, 1);
3725 bitmap_destroy(mddev);
3726 mddev->pers->quiesce(mddev, 0);
3727 mddev->bitmap_offset = 0;
3728 }
3729 }
1da177e4
LT		 3730 md_update_sb(mddev);
3731 return rv;
3732}
3733
3734static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3735{
3736 mdk_rdev_t *rdev;
3737
3738 if (mddev->pers == NULL)
3739 return -ENODEV;
3740
3741 rdev = find_rdev(mddev, dev);
3742 if (!rdev)
3743 return -ENODEV;
3744
3745 md_error(mddev, rdev);
3746 return 0;
3747}
3748
a885c8c4
CH		 3749static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3750{
3751 mddev_t *mddev = bdev->bd_disk->private_data;
3752
3753 geo->heads = 2;
3754 geo->sectors = 4;
3755 geo->cylinders = get_capacity(mddev->gendisk) / 8;
3756 return 0;
3757}
3758
1da177e4
LT		 3759static int md_ioctl(struct inode *inode, struct file *file,
3760 unsigned int cmd, unsigned long arg)
3761{
3762 int err = 0;
3763 void __user *argp = (void __user *)arg;
1da177e4
LT		 3764 mddev_t *mddev = NULL;
3765
3766 if (!capable(CAP_SYS_ADMIN))
3767 return -EACCES;
3768
3769 /*
3770 * Commands dealing with the RAID driver but not any
3771 * particular array:
3772 */
3773 switch (cmd)
3774 {
3775 case RAID_VERSION:
3776 err = get_version(argp);
3777 goto done;
3778
3779 case PRINT_RAID_DEBUG:
3780 err = 0;
3781 md_print_devices();
3782 goto done;
3783
3784#ifndef MODULE
3785 case RAID_AUTORUN:
3786 err = 0;
3787 autostart_arrays(arg);
3788 goto done;
3789#endif
3790 default:;
3791 }
3792
3793 /*
3794 * Commands creating/starting a new array:
3795 */
3796
3797 mddev = inode->i_bdev->bd_disk->private_data;
3798
3799 if (!mddev) {
3800 BUG();
3801 goto abort;
3802 }
3803
3804
3805 if (cmd == START_ARRAY) {
3806 /* START_ARRAY doesn't need to lock the array as autostart_array
3807 * does the locking, and it could even be a different array
3808 */
3809 static int cnt = 3;
3810 if (cnt > 0 ) {
3811 printk(KERN_WARNING
3812 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
e8a00334 3813 "This will not be supported beyond July 2006\n",
1da177e4
LT		 3814 current->comm, current->pid);
3815 cnt--;
3816 }
3817 err = autostart_array(new_decode_dev(arg));
3818 if (err) {
3819 printk(KERN_WARNING "md: autostart failed!\n");
3820 goto abort;
3821 }
3822 goto done;
3823 }
3824
3825 err = mddev_lock(mddev);
3826 if (err) {
3827 printk(KERN_INFO
3828 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3829 err, cmd);
3830 goto abort;
3831 }
3832
3833 switch (cmd)
3834 {
3835 case SET_ARRAY_INFO:
3836 {
3837 mdu_array_info_t info;
3838 if (!arg)
3839 memset(&info, 0, sizeof(info));
3840 else if (copy_from_user(&info, argp, sizeof(info))) {
3841 err = -EFAULT;
3842 goto abort_unlock;
3843 }
3844 if (mddev->pers) {
3845 err = update_array_info(mddev, &info);
3846 if (err) {
3847 printk(KERN_WARNING "md: couldn't update"
3848 " array info. %d\n", err);
3849 goto abort_unlock;
3850 }
3851 goto done_unlock;
3852 }
3853 if (!list_empty(&mddev->disks)) {
3854 printk(KERN_WARNING
3855 "md: array %s already has disks!\n",
3856 mdname(mddev));
3857 err = -EBUSY;
3858 goto abort_unlock;
3859 }
3860 if (mddev->raid_disks) {
3861 printk(KERN_WARNING
3862 "md: array %s already initialised!\n",
3863 mdname(mddev));
3864 err = -EBUSY;
3865 goto abort_unlock;
3866 }
3867 err = set_array_info(mddev, &info);
3868 if (err) {
3869 printk(KERN_WARNING "md: couldn't set"
3870 " array info. %d\n", err);
3871 goto abort_unlock;
3872 }
3873 }
3874 goto done_unlock;
3875
3876 default:;
3877 }
3878
3879 /*
3880 * Commands querying/configuring an existing array:
3881 */
32a7627c
N		 3882 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3883 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3884 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3885 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 3886 err = -ENODEV;
3887 goto abort_unlock;
3888 }
3889
3890 /*
3891 * Commands even a read-only array can execute:
3892 */
3893 switch (cmd)
3894 {
3895 case GET_ARRAY_INFO:
3896 err = get_array_info(mddev, argp);
3897 goto done_unlock;
3898
32a7627c 3899 case GET_BITMAP_FILE:
87162a28 3900 err = get_bitmap_file(mddev, argp);
32a7627c
N		 3901 goto done_unlock;
3902
1da177e4
LT		 3903 case GET_DISK_INFO:
3904 err = get_disk_info(mddev, argp);
3905 goto done_unlock;
3906
3907 case RESTART_ARRAY_RW:
3908 err = restart_array(mddev);
3909 goto done_unlock;
3910
3911 case STOP_ARRAY:
3912 err = do_md_stop (mddev, 0);
3913 goto done_unlock;
3914
3915 case STOP_ARRAY_RO:
3916 err = do_md_stop (mddev, 1);
3917 goto done_unlock;
3918
3919 /*
3920 * We have a problem here : there is no easy way to give a CHS
3921 * virtual geometry. We currently pretend that we have a 2 heads
3922 * 4 sectors (with a BIG number of cylinders...). This drives
3923 * dosfs just mad... ;-)
3924 */
1da177e4
LT		 3925 }
3926
3927 /*
3928 * The remaining ioctls are changing the state of the
f91de92e
N		 3929 * superblock, so we do not allow them on read-only arrays.
3930 * However non-MD ioctls (e.g. get-size) will still come through
3931 * here and hit the 'default' below, so only disallow
3932 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 3933 */
f91de92e
N		 3934 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3935 mddev->ro && mddev->pers) {
3936 if (mddev->ro == 2) {
3937 mddev->ro = 0;
3938 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3939 md_wakeup_thread(mddev->thread);
3940
3941 } else {
3942 err = -EROFS;
3943 goto abort_unlock;
3944 }
1da177e4
LT		 3945 }
3946
3947 switch (cmd)
3948 {
3949 case ADD_NEW_DISK:
3950 {
3951 mdu_disk_info_t info;
3952 if (copy_from_user(&info, argp, sizeof(info)))
3953 err = -EFAULT;
3954 else
3955 err = add_new_disk(mddev, &info);
3956 goto done_unlock;
3957 }
3958
3959 case HOT_REMOVE_DISK:
3960 err = hot_remove_disk(mddev, new_decode_dev(arg));
3961 goto done_unlock;
3962
3963 case HOT_ADD_DISK:
3964 err = hot_add_disk(mddev, new_decode_dev(arg));
3965 goto done_unlock;
3966
3967 case SET_DISK_FAULTY:
3968 err = set_disk_faulty(mddev, new_decode_dev(arg));
3969 goto done_unlock;
3970
3971 case RUN_ARRAY:
3972 err = do_md_run (mddev);
3973 goto done_unlock;
3974
32a7627c
N		 3975 case SET_BITMAP_FILE:
3976 err = set_bitmap_file(mddev, (int)arg);
3977 goto done_unlock;
3978
1da177e4 3979 default:
1da177e4
LT		 3980 err = -EINVAL;
3981 goto abort_unlock;
3982 }
3983
3984done_unlock:
3985abort_unlock:
3986 mddev_unlock(mddev);
3987
3988 return err;
3989done:
3990 if (err)
3991 MD_BUG();
3992abort:
3993 return err;
3994}
3995
3996static int md_open(struct inode *inode, struct file *file)
3997{
3998 /*
3999 * Succeed if we can lock the mddev, which confirms that
4000 * it isn't being stopped right now.
4001 */
4002 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4003 int err;
4004
4005 if ((err = mddev_lock(mddev)))
4006 goto out;
4007
4008 err = 0;
4009 mddev_get(mddev);
4010 mddev_unlock(mddev);
4011
4012 check_disk_change(inode->i_bdev);
4013 out:
4014 return err;
4015}
4016
4017static int md_release(struct inode *inode, struct file * file)
4018{
4019 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4020
4021 if (!mddev)
4022 BUG();
4023 mddev_put(mddev);
4024
4025 return 0;
4026}
4027
4028static int md_media_changed(struct gendisk *disk)
4029{
4030 mddev_t *mddev = disk->private_data;
4031
4032 return mddev->changed;
4033}
4034
4035static int md_revalidate(struct gendisk *disk)
4036{
4037 mddev_t *mddev = disk->private_data;
4038
4039 mddev->changed = 0;
4040 return 0;
4041}
4042static struct block_device_operations md_fops =
4043{
4044 .owner = THIS_MODULE,
4045 .open = md_open,
4046 .release = md_release,
4047 .ioctl = md_ioctl,
a885c8c4 4048 .getgeo = md_getgeo,
1da177e4
LT		 4049 .media_changed = md_media_changed,
4050 .revalidate_disk= md_revalidate,
4051};
4052
75c96f85 4053static int md_thread(void * arg)
1da177e4
LT		 4054{
4055 mdk_thread_t *thread = arg;
4056
1da177e4
LT		 4057 /*
4058 * md_thread is a 'system-thread', it's priority should be very
4059 * high. We avoid resource deadlocks individually in each
4060 * raid personality. (RAID5 does preallocation) We also use RR and
4061 * the very same RT priority as kswapd, thus we will never get
4062 * into a priority inversion deadlock.
4063 *
4064 * we definitely have to have equal or higher priority than
4065 * bdflush, otherwise bdflush will deadlock if there are too
4066 * many dirty RAID5 blocks.
4067 */
1da177e4 4068
6985c43f 4069 allow_signal(SIGKILL);
a6fb0934 4070 while (!kthread_should_stop()) {
1da177e4 4071
93588e22
N		 4072 /* We need to wait INTERRUPTIBLE so that
4073 * we don't add to the load-average.
4074 * That means we need to be sure no signals are
4075 * pending
4076 */
4077 if (signal_pending(current))
4078 flush_signals(current);
4079
4080 wait_event_interruptible_timeout
4081 (thread->wqueue,
4082 test_bit(THREAD_WAKEUP, &thread->flags)
4083 || kthread_should_stop(),
4084 thread->timeout);
3e1d1d28 4085 try_to_freeze();
1da177e4
LT		 4086
4087 clear_bit(THREAD_WAKEUP, &thread->flags);
4088
787453c2 4089 thread->run(thread->mddev);
1da177e4 4090 }
a6fb0934 4091
1da177e4
LT		 4092 return 0;
4093}
4094
4095void md_wakeup_thread(mdk_thread_t *thread)
4096{
4097 if (thread) {
4098 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4099 set_bit(THREAD_WAKEUP, &thread->flags);
4100 wake_up(&thread->wqueue);
4101 }
4102}
4103
4104mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4105 const char *name)
4106{
4107 mdk_thread_t *thread;
1da177e4 4108
9ffae0cf 4109 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4110 if (!thread)
4111 return NULL;
4112
1da177e4
LT		 4113 init_waitqueue_head(&thread->wqueue);
4114
1da177e4
LT		 4115 thread->run = run;
4116 thread->mddev = mddev;
32a7627c 4117 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4118 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4119 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4120 kfree(thread);
4121 return NULL;
4122 }
1da177e4
LT		 4123 return thread;
4124}
4125
1da177e4
LT		 4126void md_unregister_thread(mdk_thread_t *thread)
4127{
d28446fe 4128 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 4129
4130 kthread_stop(thread->tsk);
1da177e4
LT		 4131 kfree(thread);
4132}
4133
4134void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4135{
4136 if (!mddev) {
4137 MD_BUG();
4138 return;
4139 }
4140
b2d444d7 4141 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4142 return;
32a7627c 4143/*
1da177e4
LT		 4144 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4145 mdname(mddev),
4146 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4147 __builtin_return_address(0),__builtin_return_address(1),
4148 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4149*/
1da177e4
LT		 4150 if (!mddev->pers->error_handler)
4151 return;
4152 mddev->pers->error_handler(mddev,rdev);
4153 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4155 md_wakeup_thread(mddev->thread);
c331eb04 4156 md_new_event_inintr(mddev);
1da177e4
LT		 4157}
4158
4159/* seq_file implementation /proc/mdstat */
4160
4161static void status_unused(struct seq_file *seq)
4162{
4163 int i = 0;
4164 mdk_rdev_t *rdev;
4165 struct list_head *tmp;
4166
4167 seq_printf(seq, "unused devices: ");
4168
4169 ITERATE_RDEV_PENDING(rdev,tmp) {
4170 char b[BDEVNAME_SIZE];
4171 i++;
4172 seq_printf(seq, "%s ",
4173 bdevname(rdev->bdev,b));
4174 }
4175 if (!i)
4176 seq_printf(seq, "<none>");
4177
4178 seq_printf(seq, "\n");
4179}
4180
4181
4182static void status_resync(struct seq_file *seq, mddev_t * mddev)
4183{
4588b42e
N		 4184 sector_t max_blocks, resync, res;
4185 unsigned long dt, db, rt;
4186 int scale;
4187 unsigned int per_milli;
1da177e4
LT		 4188
4189 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4190
4191 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4192 max_blocks = mddev->resync_max_sectors >> 1;
4193 else
4194 max_blocks = mddev->size;
4195
4196 /*
4197 * Should not happen.
4198 */
4199 if (!max_blocks) {
4200 MD_BUG();
4201 return;
4202 }
4588b42e
N		 4203 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4204 * in a sector_t, and (max_blocks>>scale) will fit in a
4205 * u32, as those are the requirements for sector_div.
4206 * Thus 'scale' must be at least 10
4207 */
4208 scale = 10;
4209 if (sizeof(sector_t) > sizeof(unsigned long)) {
4210 while ( max_blocks/2 > (1ULL<<(scale+32)))
4211 scale++;
4212 }
4213 res = (resync>>scale)*1000;
4214 sector_div(res, (u32)((max_blocks>>scale)+1));
4215
4216 per_milli = res;
1da177e4 4217 {
4588b42e 4218 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4219 seq_printf(seq, "[");
4220 for (i = 0; i < x; i++)
4221 seq_printf(seq, "=");
4222 seq_printf(seq, ">");
4223 for (i = 0; i < y; i++)
4224 seq_printf(seq, ".");
4225 seq_printf(seq, "] ");
4226 }
4588b42e 4227 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 4228 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4229 "reshape" :
1da177e4 4230 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
ccfcc3c1 4231 "resync" : "recovery")),
4588b42e
N		 4232 per_milli/10, per_milli % 10,
4233 (unsigned long long) resync,
4234 (unsigned long long) max_blocks);
1da177e4
LT		 4235
4236 /*
4237 * We do not want to overflow, so the order of operands and
4238 * the * 100 / 100 trick are important. We do a +1 to be
4239 * safe against division by zero. We only estimate anyway.
4240 *
4241 * dt: time from mark until now
4242 * db: blocks written from mark until now
4243 * rt: remaining time
4244 */
4245 dt = ((jiffies - mddev->resync_mark) / HZ);
4246 if (!dt) dt++;
4247 db = resync - (mddev->resync_mark_cnt/2);
4588b42e 4248 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100;
1da177e4
LT		 4249
4250 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4251
4252 seq_printf(seq, " speed=%ldK/sec", db/dt);
4253}
4254
4255static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4256{
4257 struct list_head *tmp;
4258 loff_t l = *pos;
4259 mddev_t *mddev;
4260
4261 if (l >= 0x10000)
4262 return NULL;
4263 if (!l--)
4264 /* header */
4265 return (void*)1;
4266
4267 spin_lock(&all_mddevs_lock);
4268 list_for_each(tmp,&all_mddevs)
4269 if (!l--) {
4270 mddev = list_entry(tmp, mddev_t, all_mddevs);
4271 mddev_get(mddev);
4272 spin_unlock(&all_mddevs_lock);
4273 return mddev;
4274 }
4275 spin_unlock(&all_mddevs_lock);
4276 if (!l--)
4277 return (void*)2;/* tail */
4278 return NULL;
4279}
4280
4281static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4282{
4283 struct list_head *tmp;
4284 mddev_t *next_mddev, *mddev = v;
4285
4286 ++*pos;
4287 if (v == (void*)2)
4288 return NULL;
4289
4290 spin_lock(&all_mddevs_lock);
4291 if (v == (void*)1)
4292 tmp = all_mddevs.next;
4293 else
4294 tmp = mddev->all_mddevs.next;
4295 if (tmp != &all_mddevs)
4296 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4297 else {
4298 next_mddev = (void*)2;
4299 *pos = 0x10000;
4300 }
4301 spin_unlock(&all_mddevs_lock);
4302
4303 if (v != (void*)1)
4304 mddev_put(mddev);
4305 return next_mddev;
4306
4307}
4308
4309static void md_seq_stop(struct seq_file *seq, void *v)
4310{
4311 mddev_t *mddev = v;
4312
4313 if (mddev && v != (void*)1 && v != (void*)2)
4314 mddev_put(mddev);
4315}
4316
d7603b7e
N		 4317struct mdstat_info {
4318 int event;
4319};
4320
1da177e4
LT		 4321static int md_seq_show(struct seq_file *seq, void *v)
4322{
4323 mddev_t *mddev = v;
4324 sector_t size;
4325 struct list_head *tmp2;
4326 mdk_rdev_t *rdev;
d7603b7e 4327 struct mdstat_info *mi = seq->private;
32a7627c 4328 struct bitmap *bitmap;
1da177e4
LT		 4329
4330 if (v == (void*)1) {
2604b703 4331 struct mdk_personality *pers;
1da177e4
LT		 4332 seq_printf(seq, "Personalities : ");
4333 spin_lock(&pers_lock);
2604b703
N		 4334 list_for_each_entry(pers, &pers_list, list)
4335 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 4336
4337 spin_unlock(&pers_lock);
4338 seq_printf(seq, "\n");
d7603b7e 4339 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 4340 return 0;
4341 }
4342 if (v == (void*)2) {
4343 status_unused(seq);
4344 return 0;
4345 }
4346
5dc5cf7d 4347 if (mddev_lock(mddev) < 0)
1da177e4 4348 return -EINTR;
5dc5cf7d 4349
1da177e4
LT		 4350 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4351 seq_printf(seq, "%s : %sactive", mdname(mddev),
4352 mddev->pers ? "" : "in");
4353 if (mddev->pers) {
f91de92e 4354 if (mddev->ro==1)
1da177e4 4355 seq_printf(seq, " (read-only)");
f91de92e
N		 4356 if (mddev->ro==2)
4357 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 4358 seq_printf(seq, " %s", mddev->pers->name);
4359 }
4360
4361 size = 0;
4362 ITERATE_RDEV(mddev,rdev,tmp2) {
4363 char b[BDEVNAME_SIZE];
4364 seq_printf(seq, " %s[%d]",
4365 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 4366 if (test_bit(WriteMostly, &rdev->flags))
4367 seq_printf(seq, "(W)");
b2d444d7 4368 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4369 seq_printf(seq, "(F)");
4370 continue;
b325a32e
N		 4371 } else if (rdev->raid_disk < 0)
4372 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 4373 size += rdev->size;
4374 }
4375
4376 if (!list_empty(&mddev->disks)) {
4377 if (mddev->pers)
4378 seq_printf(seq, "\n %llu blocks",
4379 (unsigned long long)mddev->array_size);
4380 else
4381 seq_printf(seq, "\n %llu blocks",
4382 (unsigned long long)size);
4383 }
1cd6bf19
N		 4384 if (mddev->persistent) {
4385 if (mddev->major_version != 0 ||
4386 mddev->minor_version != 90) {
4387 seq_printf(seq," super %d.%d",
4388 mddev->major_version,
4389 mddev->minor_version);
4390 }
4391 } else
4392 seq_printf(seq, " super non-persistent");
1da177e4
LT		 4393
4394 if (mddev->pers) {
4395 mddev->pers->status (seq, mddev);
4396 seq_printf(seq, "\n ");
8e1b39d6
N		 4397 if (mddev->pers->sync_request) {
4398 if (mddev->curr_resync > 2) {
4399 status_resync (seq, mddev);
4400 seq_printf(seq, "\n ");
4401 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4402 seq_printf(seq, "\tresync=DELAYED\n ");
4403 else if (mddev->recovery_cp < MaxSector)
4404 seq_printf(seq, "\tresync=PENDING\n ");
4405 }
32a7627c
N		 4406 } else
4407 seq_printf(seq, "\n ");
4408
4409 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 4410 unsigned long chunk_kb;
4411 unsigned long flags;
32a7627c
N		 4412 spin_lock_irqsave(&bitmap->lock, flags);
4413 chunk_kb = bitmap->chunksize >> 10;
4414 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4415 "%lu%s chunk",
4416 bitmap->pages - bitmap->missing_pages,
4417 bitmap->pages,
4418 (bitmap->pages - bitmap->missing_pages)
4419 << (PAGE_SHIFT - 10),
4420 chunk_kb ? chunk_kb : bitmap->chunksize,
4421 chunk_kb ? "KB" : "B");
78d742d8
N		 4422 if (bitmap->file) {
4423 seq_printf(seq, ", file: ");
4424 seq_path(seq, bitmap->file->f_vfsmnt,
4425 bitmap->file->f_dentry," \t\n");
32a7627c 4426 }
78d742d8 4427
32a7627c
N		 4428 seq_printf(seq, "\n");
4429 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 4430 }
4431
4432 seq_printf(seq, "\n");
4433 }
4434 mddev_unlock(mddev);
4435
4436 return 0;
4437}
4438
4439static struct seq_operations md_seq_ops = {
4440 .start = md_seq_start,
4441 .next = md_seq_next,
4442 .stop = md_seq_stop,
4443 .show = md_seq_show,
4444};
4445
4446static int md_seq_open(struct inode *inode, struct file *file)
4447{
4448 int error;
d7603b7e
N		 4449 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4450 if (mi == NULL)
4451 return -ENOMEM;
1da177e4
LT		 4452
4453 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 4454 if (error)
4455 kfree(mi);
4456 else {
4457 struct seq_file *p = file->private_data;
4458 p->private = mi;
4459 mi->event = atomic_read(&md_event_count);
4460 }
1da177e4
LT		 4461 return error;
4462}
4463
d7603b7e
N		 4464static int md_seq_release(struct inode *inode, struct file *file)
4465{
4466 struct seq_file *m = file->private_data;
4467 struct mdstat_info *mi = m->private;
4468 m->private = NULL;
4469 kfree(mi);
4470 return seq_release(inode, file);
4471}
4472
4473static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4474{
4475 struct seq_file *m = filp->private_data;
4476 struct mdstat_info *mi = m->private;
4477 int mask;
4478
4479 poll_wait(filp, &md_event_waiters, wait);
4480
4481 /* always allow read */
4482 mask = POLLIN | POLLRDNORM;
4483
4484 if (mi->event != atomic_read(&md_event_count))
4485 mask |= POLLERR | POLLPRI;
4486 return mask;
4487}
4488
1da177e4
LT		 4489static struct file_operations md_seq_fops = {
4490 .open = md_seq_open,
4491 .read = seq_read,
4492 .llseek = seq_lseek,
d7603b7e
N		 4493 .release = md_seq_release,
4494 .poll = mdstat_poll,
1da177e4
LT		 4495};
4496
2604b703 4497int register_md_personality(struct mdk_personality *p)
1da177e4 4498{
1da177e4 4499 spin_lock(&pers_lock);
2604b703
N		 4500 list_add_tail(&p->list, &pers_list);
4501 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 4502 spin_unlock(&pers_lock);
4503 return 0;
4504}
4505
2604b703 4506int unregister_md_personality(struct mdk_personality *p)
1da177e4 4507{
2604b703 4508 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 4509 spin_lock(&pers_lock);
2604b703 4510 list_del_init(&p->list);
1da177e4
LT		 4511 spin_unlock(&pers_lock);
4512 return 0;
4513}
4514
4515static int is_mddev_idle(mddev_t *mddev)
4516{
4517 mdk_rdev_t * rdev;
4518 struct list_head *tmp;
4519 int idle;
4520 unsigned long curr_events;
4521
4522 idle = 1;
4523 ITERATE_RDEV(mddev,rdev,tmp) {
4524 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 4525 curr_events = disk_stat_read(disk, sectors[0]) +
4526 disk_stat_read(disk, sectors[1]) -
1da177e4 4527 atomic_read(&disk->sync_io);
c0e48521
N		 4528 /* The difference between curr_events and last_events
4529 * will be affected by any new non-sync IO (making
4530 * curr_events bigger) and any difference in the amount of
4531 * in-flight syncio (making current_events bigger or smaller)
4532 * The amount in-flight is currently limited to
4533 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4534 * which is at most 4096 sectors.
4535 * These numbers are fairly fragile and should be made
4536 * more robust, probably by enforcing the
4537 * 'window size' that md_do_sync sort-of uses.
4538 *
1da177e4
LT		 4539 * Note: the following is an unsigned comparison.
4540 */
c0e48521 4541 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 4542 rdev->last_events = curr_events;
4543 idle = 0;
4544 }
4545 }
4546 return idle;
4547}
4548
4549void md_done_sync(mddev_t *mddev, int blocks, int ok)
4550{
4551 /* another "blocks" (512byte) blocks have been synced */
4552 atomic_sub(blocks, &mddev->recovery_active);
4553 wake_up(&mddev->recovery_wait);
4554 if (!ok) {
4555 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4556 md_wakeup_thread(mddev->thread);
4557 // stop recovery, signal do_sync ....
4558 }
4559}
4560
4561
06d91a5f
N		 4562/* md_write_start(mddev, bi)
4563 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 4564 * in superblock) before writing, schedule a superblock update
4565 * and wait for it to complete.
06d91a5f 4566 */
3d310eb7 4567void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 4568{
06d91a5f 4569 if (bio_data_dir(bi) != WRITE)
3d310eb7 4570 return;
06d91a5f 4571
f91de92e
N		 4572 BUG_ON(mddev->ro == 1);
4573 if (mddev->ro == 2) {
4574 /* need to switch to read/write */
4575 mddev->ro = 0;
4576 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4577 md_wakeup_thread(mddev->thread);
4578 }
06d91a5f 4579 atomic_inc(&mddev->writes_pending);
06d91a5f 4580 if (mddev->in_sync) {
a9701a30 4581 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 4582 if (mddev->in_sync) {
4583 mddev->in_sync = 0;
4584 mddev->sb_dirty = 1;
4585 md_wakeup_thread(mddev->thread);
4586 }
a9701a30 4587 spin_unlock_irq(&mddev->write_lock);
06d91a5f 4588 }
3d310eb7 4589 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
1da177e4
LT		 4590}
4591
4592void md_write_end(mddev_t *mddev)
4593{
4594 if (atomic_dec_and_test(&mddev->writes_pending)) {
4595 if (mddev->safemode == 2)
4596 md_wakeup_thread(mddev->thread);
4597 else
4598 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4599 }
4600}
4601
75c96f85 4602static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 4603
4604#define SYNC_MARKS 10
4605#define SYNC_MARK_STEP (3*HZ)
29269553 4606void md_do_sync(mddev_t *mddev)
1da177e4
LT		 4607{
4608 mddev_t *mddev2;
4609 unsigned int currspeed = 0,
4610 window;
57afd89f 4611 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 4612 unsigned long mark[SYNC_MARKS];
4613 sector_t mark_cnt[SYNC_MARKS];
4614 int last_mark,m;
4615 struct list_head *tmp;
4616 sector_t last_check;
57afd89f 4617 int skipped = 0;
1da177e4
LT		 4618
4619 /* just incase thread restarts... */
4620 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4621 return;
4622
4623 /* we overload curr_resync somewhat here.
4624 * 0 == not engaged in resync at all
4625 * 2 == checking that there is no conflict with another sync
4626 * 1 == like 2, but have yielded to allow conflicting resync to
4627 * commense
4628 * other == active in resync - this many blocks
4629 *
4630 * Before starting a resync we must have set curr_resync to
4631 * 2, and then checked that every "conflicting" array has curr_resync
4632 * less than ours. When we find one that is the same or higher
4633 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4634 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4635 * This will mean we have to start checking from the beginning again.
4636 *
4637 */
4638
4639 do {
4640 mddev->curr_resync = 2;
4641
4642 try_again:
787453c2 4643 if (kthread_should_stop()) {
6985c43f 4644 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 4645 goto skip;
4646 }
4647 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 4648 if (mddev2 == mddev)
4649 continue;
4650 if (mddev2->curr_resync &&
4651 match_mddev_units(mddev,mddev2)) {
4652 DEFINE_WAIT(wq);
4653 if (mddev < mddev2 && mddev->curr_resync == 2) {
4654 /* arbitrarily yield */
4655 mddev->curr_resync = 1;
4656 wake_up(&resync_wait);
4657 }
4658 if (mddev > mddev2 && mddev->curr_resync == 1)
4659 /* no need to wait here, we can wait the next
4660 * time 'round when curr_resync == 2
4661 */
4662 continue;
787453c2
N		 4663 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4664 if (!kthread_should_stop() &&
8712e553 4665 mddev2->curr_resync >= mddev->curr_resync) {
1da177e4
LT		 4666 printk(KERN_INFO "md: delaying resync of %s"
4667 " until %s has finished resync (they"
4668 " share one or more physical units)\n",
4669 mdname(mddev), mdname(mddev2));
4670 mddev_put(mddev2);
4671 schedule();
4672 finish_wait(&resync_wait, &wq);
4673 goto try_again;
4674 }
4675 finish_wait(&resync_wait, &wq);
4676 }
4677 }
4678 } while (mddev->curr_resync < 2);
4679
9d88883e 4680 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 4681 /* resync follows the size requested by the personality,
57afd89f 4682 * which defaults to physical size, but can be virtual size
1da177e4
LT		 4683 */
4684 max_sectors = mddev->resync_max_sectors;
9d88883e 4685 mddev->resync_mismatches = 0;
ccfcc3c1
N		 4686 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4687 max_sectors = mddev->size << 1;
4688 else
1da177e4
LT		 4689 /* recovery follows the physical size of devices */
4690 max_sectors = mddev->size << 1;
4691
4692 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4693 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
88202a0c 4694 " %d KB/sec/disc.\n", speed_min(mddev));
338cec32 4695 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
1da177e4 4696 "(but not more than %d KB/sec) for reconstruction.\n",
88202a0c 4697 speed_max(mddev));
1da177e4
LT		 4698
4699 is_mddev_idle(mddev); /* this also initializes IO event counters */
32a7627c 4700 /* we don't use the checkpoint if there's a bitmap */
24dd469d
N		 4701 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4702 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1da177e4
LT		 4703 j = mddev->recovery_cp;
4704 else
4705 j = 0;
57afd89f 4706 io_sectors = 0;
1da177e4
LT		 4707 for (m = 0; m < SYNC_MARKS; m++) {
4708 mark[m] = jiffies;
57afd89f 4709 mark_cnt[m] = io_sectors;
1da177e4
LT		 4710 }
4711 last_mark = 0;
4712 mddev->resync_mark = mark[last_mark];
4713 mddev->resync_mark_cnt = mark_cnt[last_mark];
4714
4715 /*
4716 * Tune reconstruction:
4717 */
4718 window = 32*(PAGE_SIZE/512);
4719 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4720 window/2,(unsigned long long) max_sectors/2);
4721
4722 atomic_set(&mddev->recovery_active, 0);
4723 init_waitqueue_head(&mddev->recovery_wait);
4724 last_check = 0;
4725
4726 if (j>2) {
4727 printk(KERN_INFO
4728 "md: resuming recovery of %s from checkpoint.\n",
4729 mdname(mddev));
4730 mddev->curr_resync = j;
4731 }
4732
4733 while (j < max_sectors) {
57afd89f 4734 sector_t sectors;
1da177e4 4735
57afd89f
N		 4736 skipped = 0;
4737 sectors = mddev->pers->sync_request(mddev, j, &skipped,
88202a0c 4738 currspeed < speed_min(mddev));
57afd89f 4739 if (sectors == 0) {
1da177e4
LT		 4740 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4741 goto out;
4742 }
57afd89f
N		 4743
4744 if (!skipped) { /* actual IO requested */
4745 io_sectors += sectors;
4746 atomic_add(sectors, &mddev->recovery_active);
4747 }
4748
1da177e4
LT		 4749 j += sectors;
4750 if (j>1) mddev->curr_resync = j;
d7603b7e
N		 4751 if (last_check == 0)
4752 /* this is the earliers that rebuilt will be
4753 * visible in /proc/mdstat
4754 */
4755 md_new_event(mddev);
57afd89f
N		 4756
4757 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 4758 continue;
4759
57afd89f 4760 last_check = io_sectors;
1da177e4
LT		 4761
4762 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4763 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4764 break;
4765
4766 repeat:
4767 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4768 /* step marks */
4769 int next = (last_mark+1) % SYNC_MARKS;
4770
4771 mddev->resync_mark = mark[next];
4772 mddev->resync_mark_cnt = mark_cnt[next];
4773 mark[next] = jiffies;
57afd89f 4774 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 4775 last_mark = next;
4776 }
4777
4778
787453c2 4779 if (kthread_should_stop()) {
1da177e4
LT		 4780 /*
4781 * got a signal, exit.
4782 */
4783 printk(KERN_INFO
4784 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 4785 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4786 goto out;
4787 }
4788
4789 /*
4790 * this loop exits only if either when we are slower than
4791 * the 'hard' speed limit, or the system was IO-idle for
4792 * a jiffy.
4793 * the system might be non-idle CPU-wise, but we only care
4794 * about not overloading the IO subsystem. (things like an
4795 * e2fsck being done on the RAID array should execute fast)
4796 */
4797 mddev->queue->unplug_fn(mddev->queue);
4798 cond_resched();
4799
57afd89f
N		 4800 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4801 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 4802
88202a0c
N		 4803 if (currspeed > speed_min(mddev)) {
4804 if ((currspeed > speed_max(mddev)) ||
1da177e4 4805 !is_mddev_idle(mddev)) {
c0e48521 4806 msleep(500);
1da177e4
LT		 4807 goto repeat;
4808 }
4809 }
4810 }
4811 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4812 /*
4813 * this also signals 'finished resyncing' to md_stop
4814 */
4815 out:
4816 mddev->queue->unplug_fn(mddev->queue);
4817
4818 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4819
4820 /* tell personality that we are finished */
57afd89f 4821 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 4822
4823 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1
N		 4824 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
4825 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
1da177e4
LT		 4826 mddev->curr_resync > 2 &&
4827 mddev->curr_resync >= mddev->recovery_cp) {
4828 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4829 printk(KERN_INFO
4830 "md: checkpointing recovery of %s.\n",
4831 mdname(mddev));
4832 mddev->recovery_cp = mddev->curr_resync;
4833 } else
4834 mddev->recovery_cp = MaxSector;
4835 }
4836
1da177e4
LT		 4837 skip:
4838 mddev->curr_resync = 0;
4839 wake_up(&resync_wait);
4840 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4841 md_wakeup_thread(mddev->thread);
4842}
29269553 4843EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 4844
4845
4846/*
4847 * This routine is regularly called by all per-raid-array threads to
4848 * deal with generic issues like resync and super-block update.
4849 * Raid personalities that don't have a thread (linear/raid0) do not
4850 * need this as they never do any recovery or update the superblock.
4851 *
4852 * It does not do any resync itself, but rather "forks" off other threads
4853 * to do that as needed.
4854 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4855 * "->recovery" and create a thread at ->sync_thread.
4856 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4857 * and wakeups up this thread which will reap the thread and finish up.
4858 * This thread also removes any faulty devices (with nr_pending == 0).
4859 *
4860 * The overall approach is:
4861 * 1/ if the superblock needs updating, update it.
4862 * 2/ If a recovery thread is running, don't do anything else.
4863 * 3/ If recovery has finished, clean up, possibly marking spares active.
4864 * 4/ If there are any faulty devices, remove them.
4865 * 5/ If array is degraded, try to add spares devices
4866 * 6/ If array has spares or is not in-sync, start a resync thread.
4867 */
4868void md_check_recovery(mddev_t *mddev)
4869{
4870 mdk_rdev_t *rdev;
4871 struct list_head *rtmp;
4872
4873
5f40402d
N		 4874 if (mddev->bitmap)
4875 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 4876
4877 if (mddev->ro)
4878 return;
fca4d848
N		 4879
4880 if (signal_pending(current)) {
4881 if (mddev->pers->sync_request) {
4882 printk(KERN_INFO "md: %s in immediate safe mode\n",
4883 mdname(mddev));
4884 mddev->safemode = 2;
4885 }
4886 flush_signals(current);
4887 }
4888
1da177e4
LT		 4889 if ( ! (
4890 mddev->sb_dirty ||
4891 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 4892 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4893 (mddev->safemode == 1) ||
4894 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4895 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 4896 ))
4897 return;
fca4d848 4898
df5b89b3 4899 if (mddev_trylock(mddev)) {
1da177e4 4900 int spares =0;
fca4d848 4901
a9701a30 4902 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 4903 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4904 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4905 mddev->in_sync = 1;
4906 mddev->sb_dirty = 1;
4907 }
4908 if (mddev->safemode == 1)
4909 mddev->safemode = 0;
a9701a30 4910 spin_unlock_irq(&mddev->write_lock);
fca4d848 4911
1da177e4
LT		 4912 if (mddev->sb_dirty)
4913 md_update_sb(mddev);
06d91a5f 4914
06d91a5f 4915
1da177e4
LT		 4916 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4917 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4918 /* resync/recovery still happening */
4919 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4920 goto unlock;
4921 }
4922 if (mddev->sync_thread) {
4923 /* resync has finished, collect result */
4924 md_unregister_thread(mddev->sync_thread);
4925 mddev->sync_thread = NULL;
4926 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4927 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4928 /* success...*/
4929 /* activate any spares */
4930 mddev->pers->spare_active(mddev);
4931 }
4932 md_update_sb(mddev);
41158c7e
N		 4933
4934 /* if array is no-longer degraded, then any saved_raid_disk
4935 * information must be scrapped
4936 */
4937 if (!mddev->degraded)
4938 ITERATE_RDEV(mddev,rdev,rtmp)
4939 rdev->saved_raid_disk = -1;
4940
1da177e4
LT		 4941 mddev->recovery = 0;
4942 /* flag recovery needed just to double check */
4943 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 4944 md_new_event(mddev);
1da177e4
LT		 4945 goto unlock;
4946 }
24dd469d
N		 4947 /* Clear some bits that don't mean anything, but
4948 * might be left set
4949 */
4950 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4951 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4952 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4953 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4
LT		 4954
4955 /* no recovery is running.
4956 * remove any failed drives, then
4957 * add spares if possible.
4958 * Spare are also removed and re-added, to allow
4959 * the personality to fail the re-add.
4960 */
4961 ITERATE_RDEV(mddev,rdev,rtmp)
4962 if (rdev->raid_disk >= 0 &&
b2d444d7 4963 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 4964 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 4965 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4966 char nm[20];
4967 sprintf(nm,"rd%d", rdev->raid_disk);
4968 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 4969 rdev->raid_disk = -1;
86e6ffdd 4970 }
1da177e4
LT		 4971 }
4972
4973 if (mddev->degraded) {
4974 ITERATE_RDEV(mddev,rdev,rtmp)
4975 if (rdev->raid_disk < 0
b2d444d7 4976 && !test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 4977 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4978 char nm[20];
4979 sprintf(nm, "rd%d", rdev->raid_disk);
4980 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 4981 spares++;
d7603b7e 4982 md_new_event(mddev);
86e6ffdd 4983 } else
1da177e4
LT		 4984 break;
4985 }
4986 }
4987
24dd469d
N		 4988 if (spares) {
4989 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4990 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4991 } else if (mddev->recovery_cp < MaxSector) {
4992 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4993 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4994 /* nothing to be done ... */
1da177e4 4995 goto unlock;
24dd469d 4996
1da177e4
LT		 4997 if (mddev->pers->sync_request) {
4998 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 4999 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5000 /* We are adding a device or devices to an array
5001 * which has the bitmap stored on all devices.
5002 * So make sure all bitmap pages get written
5003 */
5004 bitmap_write_all(mddev->bitmap);
5005 }
1da177e4
LT		 5006 mddev->sync_thread = md_register_thread(md_do_sync,
5007 mddev,
5008 "%s_resync");
5009 if (!mddev->sync_thread) {
5010 printk(KERN_ERR "%s: could not start resync"
5011 " thread...\n",
5012 mdname(mddev));
5013 /* leave the spares where they are, it shouldn't hurt */
5014 mddev->recovery = 0;
d7603b7e 5015 } else
1da177e4 5016 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5017 md_new_event(mddev);
1da177e4
LT		 5018 }
5019 unlock:
5020 mddev_unlock(mddev);
5021 }
5022}
5023
75c96f85
AB		 5024static int md_notify_reboot(struct notifier_block *this,
5025 unsigned long code, void *x)
1da177e4
LT		 5026{
5027 struct list_head *tmp;
5028 mddev_t *mddev;
5029
5030 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5031
5032 printk(KERN_INFO "md: stopping all md devices.\n");
5033
5034 ITERATE_MDDEV(mddev,tmp)
c71d4887 5035 if (mddev_trylock(mddev)) {
1da177e4 5036 do_md_stop (mddev, 1);
c71d4887
NB		 5037 mddev_unlock(mddev);
5038 }
1da177e4
LT		 5039 /*
5040 * certain more exotic SCSI devices are known to be
5041 * volatile wrt too early system reboots. While the
5042 * right place to handle this issue is the given
5043 * driver, we do want to have a safe RAID driver ...
5044 */
5045 mdelay(1000*1);
5046 }
5047 return NOTIFY_DONE;
5048}
5049
75c96f85 5050static struct notifier_block md_notifier = {
1da177e4
LT		 5051 .notifier_call = md_notify_reboot,
5052 .next = NULL,
5053 .priority = INT_MAX, /* before any real devices */
5054};
5055
5056static void md_geninit(void)
5057{
5058 struct proc_dir_entry *p;
5059
5060 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5061
5062 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5063 if (p)
5064 p->proc_fops = &md_seq_fops;
5065}
5066
75c96f85 5067static int __init md_init(void)
1da177e4
LT		 5068{
5069 int minor;
5070
5071 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5072 " MD_SB_DISKS=%d\n",
5073 MD_MAJOR_VERSION, MD_MINOR_VERSION,
5074 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
bd926c63 5075 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
32a7627c 5076 BITMAP_MINOR);
1da177e4
LT		 5077
5078 if (register_blkdev(MAJOR_NR, "md"))
5079 return -1;
5080 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5081 unregister_blkdev(MAJOR_NR, "md");
5082 return -1;
5083 }
5084 devfs_mk_dir("md");
5085 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5086 md_probe, NULL, NULL);
5087 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5088 md_probe, NULL, NULL);
5089
5090 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5091 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
5092 S_IFBLK|S_IRUSR|S_IWUSR,
5093 "md/%d", minor);
5094
5095 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5096 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
5097 S_IFBLK|S_IRUSR|S_IWUSR,
5098 "md/mdp%d", minor);
5099
5100
5101 register_reboot_notifier(&md_notifier);
5102 raid_table_header = register_sysctl_table(raid_root_table, 1);
5103
5104 md_geninit();
5105 return (0);
5106}
5107
5108
5109#ifndef MODULE
5110
5111/*
5112 * Searches all registered partitions for autorun RAID arrays
5113 * at boot time.
5114 */
5115static dev_t detected_devices[128];
5116static int dev_cnt;
5117
5118void md_autodetect_dev(dev_t dev)
5119{
5120 if (dev_cnt >= 0 && dev_cnt < 127)
5121 detected_devices[dev_cnt++] = dev;
5122}
5123
5124
5125static void autostart_arrays(int part)
5126{
5127 mdk_rdev_t *rdev;
5128 int i;
5129
5130 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5131
5132 for (i = 0; i < dev_cnt; i++) {
5133 dev_t dev = detected_devices[i];
5134
5135 rdev = md_import_device(dev,0, 0);
5136 if (IS_ERR(rdev))
5137 continue;
5138
b2d444d7 5139 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5140 MD_BUG();
5141 continue;
5142 }
5143 list_add(&rdev->same_set, &pending_raid_disks);
5144 }
5145 dev_cnt = 0;
5146
5147 autorun_devices(part);
5148}
5149
5150#endif
5151
5152static __exit void md_exit(void)
5153{
5154 mddev_t *mddev;
5155 struct list_head *tmp;
5156 int i;
5157 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5158 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
5159 for (i=0; i < MAX_MD_DEVS; i++)
5160 devfs_remove("md/%d", i);
5161 for (i=0; i < MAX_MD_DEVS; i++)
5162 devfs_remove("md/d%d", i);
5163
5164 devfs_remove("md");
5165
5166 unregister_blkdev(MAJOR_NR,"md");
5167 unregister_blkdev(mdp_major, "mdp");
5168 unregister_reboot_notifier(&md_notifier);
5169 unregister_sysctl_table(raid_table_header);
5170 remove_proc_entry("mdstat", NULL);
5171 ITERATE_MDDEV(mddev,tmp) {
5172 struct gendisk *disk = mddev->gendisk;
5173 if (!disk)
5174 continue;
5175 export_array(mddev);
5176 del_gendisk(disk);
5177 put_disk(disk);
5178 mddev->gendisk = NULL;
5179 mddev_put(mddev);
5180 }
5181}
5182
5183module_init(md_init)
5184module_exit(md_exit)
5185
f91de92e
N		 5186static int get_ro(char *buffer, struct kernel_param *kp)
5187{
5188 return sprintf(buffer, "%d", start_readonly);
5189}
5190static int set_ro(const char *val, struct kernel_param *kp)
5191{
5192 char *e;
5193 int num = simple_strtoul(val, &e, 10);
5194 if (*val && (*e == '\0' || *e == '\n')) {
5195 start_readonly = num;
4dbcdc75 5196 return 0;
f91de92e
N		 5197 }
5198 return -EINVAL;
5199}
5200
5201module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
6ff8d8ec
N		 5202module_param(start_dirty_degraded, int, 0644);
5203
f91de92e 5204
1da177e4
LT		 5205EXPORT_SYMBOL(register_md_personality);
5206EXPORT_SYMBOL(unregister_md_personality);
5207EXPORT_SYMBOL(md_error);
5208EXPORT_SYMBOL(md_done_sync);
5209EXPORT_SYMBOL(md_write_start);
5210EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 5211EXPORT_SYMBOL(md_register_thread);
5212EXPORT_SYMBOL(md_unregister_thread);
5213EXPORT_SYMBOL(md_wakeup_thread);
5214EXPORT_SYMBOL(md_print_devices);
5215EXPORT_SYMBOL(md_check_recovery);
5216MODULE_LICENSE("GPL");
aa1595e9 5217MODULE_ALIAS("md");
72008652 5218MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Clone of davem's net-next-2.6 tree