projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] md: Fix device-size updates in md
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
35#include <linux/module.h>
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>
1da177e4
LT		 46
47#include <linux/init.h>
48
32a7627c
N		 49#include <linux/file.h>
50
1da177e4
LT		 51#ifdef CONFIG_KMOD
52#include <linux/kmod.h>
53#endif
54
55#include <asm/unaligned.h>
56
57#define MAJOR_NR MD_MAJOR
58#define MD_DRIVER
59
60/* 63 partitions with the alternate major number (mdp) */
61#define MdpMinorShift 6
62
63#define DEBUG 0
64#define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67#ifndef MODULE
68static void autostart_arrays (int part);
69#endif
70
2604b703 71static LIST_HEAD(pers_list);
1da177e4
LT		 72static DEFINE_SPINLOCK(pers_lock);
73
74/*
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 78 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
81 * idle IO detection.
82 *
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 84 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 85 */
86
87static int sysctl_speed_limit_min = 1000;
88static int sysctl_speed_limit_max = 200000;
88202a0c
N		 89static inline int speed_min(mddev_t *mddev)
90{
91 return mddev->sync_speed_min ?
92 mddev->sync_speed_min : sysctl_speed_limit_min;
93}
94
95static inline int speed_max(mddev_t *mddev)
96{
97 return mddev->sync_speed_max ?
98 mddev->sync_speed_max : sysctl_speed_limit_max;
99}
1da177e4
LT		 100
101static struct ctl_table_header *raid_table_header;
102
103static ctl_table raid_table[] = {
104 {
105 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
106 .procname = "speed_limit_min",
107 .data = &sysctl_speed_limit_min,
108 .maxlen = sizeof(int),
109 .mode = 0644,
110 .proc_handler = &proc_dointvec,
111 },
112 {
113 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
114 .procname = "speed_limit_max",
115 .data = &sysctl_speed_limit_max,
116 .maxlen = sizeof(int),
117 .mode = 0644,
118 .proc_handler = &proc_dointvec,
119 },
120 { .ctl_name = 0 }
121};
122
123static ctl_table raid_dir_table[] = {
124 {
125 .ctl_name = DEV_RAID,
126 .procname = "raid",
127 .maxlen = 0,
128 .mode = 0555,
129 .child = raid_table,
130 },
131 { .ctl_name = 0 }
132};
133
134static ctl_table raid_root_table[] = {
135 {
136 .ctl_name = CTL_DEV,
137 .procname = "dev",
138 .maxlen = 0,
139 .mode = 0555,
140 .child = raid_dir_table,
141 },
142 { .ctl_name = 0 }
143};
144
145static struct block_device_operations md_fops;
146
f91de92e
N		 147static int start_readonly;
148
d7603b7e
N		 149/*
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
153 * count increases.
154 *
155 * Events are:
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
158 */
2989ddbd 159static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 160static atomic_t md_event_count;
07dbd377 161static void md_new_event(mddev_t *mddev)
d7603b7e
N		 162{
163 atomic_inc(&md_event_count);
164 wake_up(&md_event_waiters);
165}
166
1da177e4
LT		 167/*
168 * Enables to iterate over all existing md arrays
169 * all_mddevs_lock protects this list.
170 */
171static LIST_HEAD(all_mddevs);
172static DEFINE_SPINLOCK(all_mddevs_lock);
173
174
175/*
176 * iterates through all used mddevs in the system.
177 * We take care to grab the all_mddevs_lock whenever navigating
178 * the list, and to always hold a refcount when unlocked.
179 * Any code which breaks out of this loop while own
180 * a reference to the current mddev and must mddev_put it.
181 */
182#define ITERATE_MDDEV(mddev,tmp) \
183 \
184 for (({ spin_lock(&all_mddevs_lock); \
185 tmp = all_mddevs.next; \
186 mddev = NULL;}); \
187 ({ if (tmp != &all_mddevs) \
188 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
189 spin_unlock(&all_mddevs_lock); \
190 if (mddev) mddev_put(mddev); \
191 mddev = list_entry(tmp, mddev_t, all_mddevs); \
192 tmp != &all_mddevs;}); \
193 ({ spin_lock(&all_mddevs_lock); \
194 tmp = tmp->next;}) \
195 )
196
197
198static int md_fail_request (request_queue_t *q, struct bio *bio)
199{
200 bio_io_error(bio, bio->bi_size);
201 return 0;
202}
203
204static inline mddev_t *mddev_get(mddev_t *mddev)
205{
206 atomic_inc(&mddev->active);
207 return mddev;
208}
209
210static void mddev_put(mddev_t *mddev)
211{
212 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
213 return;
214 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
215 list_del(&mddev->all_mddevs);
216 blk_put_queue(mddev->queue);
eae1701f 217 kobject_unregister(&mddev->kobj);
1da177e4
LT		 218 }
219 spin_unlock(&all_mddevs_lock);
220}
221
222static mddev_t * mddev_find(dev_t unit)
223{
224 mddev_t *mddev, *new = NULL;
225
226 retry:
227 spin_lock(&all_mddevs_lock);
228 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
229 if (mddev->unit == unit) {
230 mddev_get(mddev);
231 spin_unlock(&all_mddevs_lock);
990a8baf 232 kfree(new);
1da177e4
LT		 233 return mddev;
234 }
235
236 if (new) {
237 list_add(&new->all_mddevs, &all_mddevs);
238 spin_unlock(&all_mddevs_lock);
239 return new;
240 }
241 spin_unlock(&all_mddevs_lock);
242
9ffae0cf 243 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 244 if (!new)
245 return NULL;
246
1da177e4
LT		 247 new->unit = unit;
248 if (MAJOR(unit) == MD_MAJOR)
249 new->md_minor = MINOR(unit);
250 else
251 new->md_minor = MINOR(unit) >> MdpMinorShift;
252
253 init_MUTEX(&new->reconfig_sem);
254 INIT_LIST_HEAD(&new->disks);
255 INIT_LIST_HEAD(&new->all_mddevs);
256 init_timer(&new->safemode_timer);
257 atomic_set(&new->active, 1);
06d91a5f 258 spin_lock_init(&new->write_lock);
3d310eb7 259 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 260
261 new->queue = blk_alloc_queue(GFP_KERNEL);
262 if (!new->queue) {
263 kfree(new);
264 return NULL;
265 }
266
267 blk_queue_make_request(new->queue, md_fail_request);
268
269 goto retry;
270}
271
272static inline int mddev_lock(mddev_t * mddev)
273{
274 return down_interruptible(&mddev->reconfig_sem);
275}
276
277static inline void mddev_lock_uninterruptible(mddev_t * mddev)
278{
279 down(&mddev->reconfig_sem);
280}
281
282static inline int mddev_trylock(mddev_t * mddev)
283{
284 return down_trylock(&mddev->reconfig_sem);
285}
286
287static inline void mddev_unlock(mddev_t * mddev)
288{
289 up(&mddev->reconfig_sem);
290
005eca5e 291 md_wakeup_thread(mddev->thread);
1da177e4
LT		 292}
293
2989ddbd 294static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 295{
296 mdk_rdev_t * rdev;
297 struct list_head *tmp;
298
299 ITERATE_RDEV(mddev,rdev,tmp) {
300 if (rdev->desc_nr == nr)
301 return rdev;
302 }
303 return NULL;
304}
305
306static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
307{
308 struct list_head *tmp;
309 mdk_rdev_t *rdev;
310
311 ITERATE_RDEV(mddev,rdev,tmp) {
312 if (rdev->bdev->bd_dev == dev)
313 return rdev;
314 }
315 return NULL;
316}
317
d9d166c2 318static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 319{
320 struct mdk_personality *pers;
d9d166c2
N		 321 list_for_each_entry(pers, &pers_list, list) {
322 if (level != LEVEL_NONE && pers->level == level)
2604b703 323 return pers;
d9d166c2
N		 324 if (strcmp(pers->name, clevel)==0)
325 return pers;
326 }
2604b703
N		 327 return NULL;
328}
329
77933d72 330static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 331{
332 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
333 return MD_NEW_SIZE_BLOCKS(size);
334}
335
336static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
337{
338 sector_t size;
339
340 size = rdev->sb_offset;
341
342 if (chunk_size)
343 size &= ~((sector_t)chunk_size/1024 - 1);
344 return size;
345}
346
347static int alloc_disk_sb(mdk_rdev_t * rdev)
348{
349 if (rdev->sb_page)
350 MD_BUG();
351
352 rdev->sb_page = alloc_page(GFP_KERNEL);
353 if (!rdev->sb_page) {
354 printk(KERN_ALERT "md: out of memory.\n");
355 return -EINVAL;
356 }
357
358 return 0;
359}
360
361static void free_disk_sb(mdk_rdev_t * rdev)
362{
363 if (rdev->sb_page) {
2d1f3b5d 364 put_page(rdev->sb_page);
1da177e4
LT		 365 rdev->sb_loaded = 0;
366 rdev->sb_page = NULL;
367 rdev->sb_offset = 0;
368 rdev->size = 0;
369 }
370}
371
372
7bfa19f2
N		 373static int super_written(struct bio *bio, unsigned int bytes_done, int error)
374{
375 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 376 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 377 if (bio->bi_size)
378 return 1;
379
380 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
a9701a30 381 md_error(mddev, rdev);
7bfa19f2 382
a9701a30
N		 383 if (atomic_dec_and_test(&mddev->pending_writes))
384 wake_up(&mddev->sb_wait);
f8b58edf 385 bio_put(bio);
7bfa19f2
N		 386 return 0;
387}
388
a9701a30
N		 389static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
390{
391 struct bio *bio2 = bio->bi_private;
392 mdk_rdev_t *rdev = bio2->bi_private;
393 mddev_t *mddev = rdev->mddev;
394 if (bio->bi_size)
395 return 1;
396
397 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
398 error == -EOPNOTSUPP) {
399 unsigned long flags;
400 /* barriers don't appear to be supported :-( */
401 set_bit(BarriersNotsupp, &rdev->flags);
402 mddev->barriers_work = 0;
403 spin_lock_irqsave(&mddev->write_lock, flags);
404 bio2->bi_next = mddev->biolist;
405 mddev->biolist = bio2;
406 spin_unlock_irqrestore(&mddev->write_lock, flags);
407 wake_up(&mddev->sb_wait);
408 bio_put(bio);
409 return 0;
410 }
411 bio_put(bio2);
412 bio->bi_private = rdev;
413 return super_written(bio, bytes_done, error);
414}
415
7bfa19f2
N		 416void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
417 sector_t sector, int size, struct page *page)
418{
419 /* write first size bytes of page to sector of rdev
420 * Increment mddev->pending_writes before returning
421 * and decrement it on completion, waking up sb_wait
422 * if zero is reached.
423 * If an error occurred, call md_error
a9701a30
N		 424 *
425 * As we might need to resubmit the request if BIO_RW_BARRIER
426 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 427 */
428 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 429 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 430
431 bio->bi_bdev = rdev->bdev;
432 bio->bi_sector = sector;
433 bio_add_page(bio, page, size, 0);
434 bio->bi_private = rdev;
435 bio->bi_end_io = super_written;
a9701a30
N		 436 bio->bi_rw = rw;
437
7bfa19f2 438 atomic_inc(&mddev->pending_writes);
a9701a30
N		 439 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
440 struct bio *rbio;
441 rw |= (1<<BIO_RW_BARRIER);
442 rbio = bio_clone(bio, GFP_NOIO);
443 rbio->bi_private = bio;
444 rbio->bi_end_io = super_written_barrier;
445 submit_bio(rw, rbio);
446 } else
447 submit_bio(rw, bio);
448}
449
450void md_super_wait(mddev_t *mddev)
451{
452 /* wait for all superblock writes that were scheduled to complete.
453 * if any had to be retried (due to BARRIER problems), retry them
454 */
455 DEFINE_WAIT(wq);
456 for(;;) {
457 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
458 if (atomic_read(&mddev->pending_writes)==0)
459 break;
460 while (mddev->biolist) {
461 struct bio *bio;
462 spin_lock_irq(&mddev->write_lock);
463 bio = mddev->biolist;
464 mddev->biolist = bio->bi_next ;
465 bio->bi_next = NULL;
466 spin_unlock_irq(&mddev->write_lock);
467 submit_bio(bio->bi_rw, bio);
468 }
469 schedule();
470 }
471 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 472}
473
1da177e4
LT		 474static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
475{
476 if (bio->bi_size)
477 return 1;
478
479 complete((struct completion*)bio->bi_private);
480 return 0;
481}
482
a654b9d8 483int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 484 struct page *page, int rw)
485{
baaa2c51 486 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 487 struct completion event;
488 int ret;
489
490 rw |= (1 << BIO_RW_SYNC);
491
492 bio->bi_bdev = bdev;
493 bio->bi_sector = sector;
494 bio_add_page(bio, page, size, 0);
495 init_completion(&event);
496 bio->bi_private = &event;
497 bio->bi_end_io = bi_complete;
498 submit_bio(rw, bio);
499 wait_for_completion(&event);
500
501 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
502 bio_put(bio);
503 return ret;
504}
a8745db2 505EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 506
0002b271 507static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 508{
509 char b[BDEVNAME_SIZE];
510 if (!rdev->sb_page) {
511 MD_BUG();
512 return -EINVAL;
513 }
514 if (rdev->sb_loaded)
515 return 0;
516
517
0002b271 518 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 519 goto fail;
520 rdev->sb_loaded = 1;
521 return 0;
522
523fail:
524 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
525 bdevname(rdev->bdev,b));
526 return -EINVAL;
527}
528
529static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
530{
531 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
532 (sb1->set_uuid1 == sb2->set_uuid1) &&
533 (sb1->set_uuid2 == sb2->set_uuid2) &&
534 (sb1->set_uuid3 == sb2->set_uuid3))
535
536 return 1;
537
538 return 0;
539}
540
541
542static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
543{
544 int ret;
545 mdp_super_t *tmp1, *tmp2;
546
547 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
548 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
549
550 if (!tmp1 || !tmp2) {
551 ret = 0;
552 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
553 goto abort;
554 }
555
556 *tmp1 = *sb1;
557 *tmp2 = *sb2;
558
559 /*
560 * nr_disks is not constant
561 */
562 tmp1->nr_disks = 0;
563 tmp2->nr_disks = 0;
564
565 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
566 ret = 0;
567 else
568 ret = 1;
569
570abort:
990a8baf
JJ		 571 kfree(tmp1);
572 kfree(tmp2);
1da177e4
LT		 573 return ret;
574}
575
576static unsigned int calc_sb_csum(mdp_super_t * sb)
577{
578 unsigned int disk_csum, csum;
579
580 disk_csum = sb->sb_csum;
581 sb->sb_csum = 0;
582 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
583 sb->sb_csum = disk_csum;
584 return csum;
585}
586
587
588/*
589 * Handle superblock details.
590 * We want to be able to handle multiple superblock formats
591 * so we have a common interface to them all, and an array of
592 * different handlers.
593 * We rely on user-space to write the initial superblock, and support
594 * reading and updating of superblocks.
595 * Interface methods are:
596 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
597 * loads and validates a superblock on dev.
598 * if refdev != NULL, compare superblocks on both devices
599 * Return:
600 * 0 - dev has a superblock that is compatible with refdev
601 * 1 - dev has a superblock that is compatible and newer than refdev
602 * so dev should be used as the refdev in future
603 * -EINVAL superblock incompatible or invalid
604 * -othererror e.g. -EIO
605 *
606 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
607 * Verify that dev is acceptable into mddev.
608 * The first time, mddev->raid_disks will be 0, and data from
609 * dev should be merged in. Subsequent calls check that dev
610 * is new enough. Return 0 or -EINVAL
611 *
612 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
613 * Update the superblock for rdev with data in mddev
614 * This does not write to disc.
615 *
616 */
617
618struct super_type {
619 char *name;
620 struct module *owner;
621 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
622 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
623 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
624};
625
626/*
627 * load_super for 0.90.0
628 */
629static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
630{
631 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
632 mdp_super_t *sb;
633 int ret;
634 sector_t sb_offset;
635
636 /*
637 * Calculate the position of the superblock,
638 * it's at the end of the disk.
639 *
640 * It also happens to be a multiple of 4Kb.
641 */
642 sb_offset = calc_dev_sboffset(rdev->bdev);
643 rdev->sb_offset = sb_offset;
644
0002b271 645 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 646 if (ret) return ret;
647
648 ret = -EINVAL;
649
650 bdevname(rdev->bdev, b);
651 sb = (mdp_super_t*)page_address(rdev->sb_page);
652
653 if (sb->md_magic != MD_SB_MAGIC) {
654 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
655 b);
656 goto abort;
657 }
658
659 if (sb->major_version != 0 ||
660 sb->minor_version != 90) {
661 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
662 sb->major_version, sb->minor_version,
663 b);
664 goto abort;
665 }
666
667 if (sb->raid_disks <= 0)
668 goto abort;
669
670 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
671 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
672 b);
673 goto abort;
674 }
675
676 rdev->preferred_minor = sb->md_minor;
677 rdev->data_offset = 0;
0002b271 678 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 679
680 if (sb->level == LEVEL_MULTIPATH)
681 rdev->desc_nr = -1;
682 else
683 rdev->desc_nr = sb->this_disk.number;
684
685 if (refdev == 0)
686 ret = 1;
687 else {
688 __u64 ev1, ev2;
689 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
690 if (!uuid_equal(refsb, sb)) {
691 printk(KERN_WARNING "md: %s has different UUID to %s\n",
692 b, bdevname(refdev->bdev,b2));
693 goto abort;
694 }
695 if (!sb_equal(refsb, sb)) {
696 printk(KERN_WARNING "md: %s has same UUID"
697 " but different superblock to %s\n",
698 b, bdevname(refdev->bdev, b2));
699 goto abort;
700 }
701 ev1 = md_event(sb);
702 ev2 = md_event(refsb);
703 if (ev1 > ev2)
704 ret = 1;
705 else
706 ret = 0;
707 }
708 rdev->size = calc_dev_size(rdev, sb->chunk_size);
709
2bf071bf
N		 710 if (rdev->size < sb->size && sb->level > 1)
711 /* "this cannot possibly happen" ... */
712 ret = -EINVAL;
713
1da177e4
LT		 714 abort:
715 return ret;
716}
717
718/*
719 * validate_super for 0.90.0
720 */
721static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
722{
723 mdp_disk_t *desc;
724 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
725
41158c7e 726 rdev->raid_disk = -1;
b2d444d7 727 rdev->flags = 0;
1da177e4
LT		 728 if (mddev->raid_disks == 0) {
729 mddev->major_version = 0;
730 mddev->minor_version = sb->minor_version;
731 mddev->patch_version = sb->patch_version;
732 mddev->persistent = ! sb->not_persistent;
733 mddev->chunk_size = sb->chunk_size;
734 mddev->ctime = sb->ctime;
735 mddev->utime = sb->utime;
736 mddev->level = sb->level;
d9d166c2 737 mddev->clevel[0] = 0;
1da177e4
LT		 738 mddev->layout = sb->layout;
739 mddev->raid_disks = sb->raid_disks;
740 mddev->size = sb->size;
741 mddev->events = md_event(sb);
9223214e 742 mddev->bitmap_offset = 0;
36fa3063 743 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4
LT		 744
745 if (sb->state & (1<<MD_SB_CLEAN))
746 mddev->recovery_cp = MaxSector;
747 else {
748 if (sb->events_hi == sb->cp_events_hi &&
749 sb->events_lo == sb->cp_events_lo) {
750 mddev->recovery_cp = sb->recovery_cp;
751 } else
752 mddev->recovery_cp = 0;
753 }
754
755 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
756 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
757 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
758 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
759
760 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 761
762 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
763 mddev->bitmap_file == NULL) {
6cce3b23
N		 764 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
765 && mddev->level != 10) {
a654b9d8 766 /* FIXME use a better test */
6cce3b23 767 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 768 return -EINVAL;
769 }
36fa3063 770 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8
N		 771 }
772
41158c7e
N		 773 } else if (mddev->pers == NULL) {
774 /* Insist on good event counter while assembling */
775 __u64 ev1 = md_event(sb);
1da177e4
LT		 776 ++ev1;
777 if (ev1 < mddev->events)
778 return -EINVAL;
41158c7e
N		 779 } else if (mddev->bitmap) {
780 /* if adding to array with a bitmap, then we can accept an
781 * older device ... but not too old.
782 */
783 __u64 ev1 = md_event(sb);
784 if (ev1 < mddev->bitmap->events_cleared)
785 return 0;
786 } else /* just a hot-add of a new device, leave raid_disk at -1 */
787 return 0;
788
1da177e4 789 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 790 desc = sb->disks + rdev->desc_nr;
791
792 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 793 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 794 else if (desc->state & (1<<MD_DISK_SYNC) &&
795 desc->raid_disk < mddev->raid_disks) {
b2d444d7 796 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 797 rdev->raid_disk = desc->raid_disk;
798 }
8ddf9efe
N		 799 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
800 set_bit(WriteMostly, &rdev->flags);
41158c7e 801 } else /* MULTIPATH are always insync */
b2d444d7 802 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 803 return 0;
804}
805
806/*
807 * sync_super for 0.90.0
808 */
809static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
810{
811 mdp_super_t *sb;
812 struct list_head *tmp;
813 mdk_rdev_t *rdev2;
814 int next_spare = mddev->raid_disks;
19133a42 815
1da177e4
LT		 816
817 /* make rdev->sb match mddev data..
818 *
819 * 1/ zero out disks
820 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
821 * 3/ any empty disks < next_spare become removed
822 *
823 * disks[0] gets initialised to REMOVED because
824 * we cannot be sure from other fields if it has
825 * been initialised or not.
826 */
827 int i;
828 int active=0, working=0,failed=0,spare=0,nr_disks=0;
829
61181565
N		 830 rdev->sb_size = MD_SB_BYTES;
831
1da177e4
LT		 832 sb = (mdp_super_t*)page_address(rdev->sb_page);
833
834 memset(sb, 0, sizeof(*sb));
835
836 sb->md_magic = MD_SB_MAGIC;
837 sb->major_version = mddev->major_version;
838 sb->minor_version = mddev->minor_version;
839 sb->patch_version = mddev->patch_version;
840 sb->gvalid_words = 0; /* ignored */
841 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
842 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
843 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
844 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
845
846 sb->ctime = mddev->ctime;
847 sb->level = mddev->level;
848 sb->size = mddev->size;
849 sb->raid_disks = mddev->raid_disks;
850 sb->md_minor = mddev->md_minor;
851 sb->not_persistent = !mddev->persistent;
852 sb->utime = mddev->utime;
853 sb->state = 0;
854 sb->events_hi = (mddev->events>>32);
855 sb->events_lo = (u32)mddev->events;
856
857 if (mddev->in_sync)
858 {
859 sb->recovery_cp = mddev->recovery_cp;
860 sb->cp_events_hi = (mddev->events>>32);
861 sb->cp_events_lo = (u32)mddev->events;
862 if (mddev->recovery_cp == MaxSector)
863 sb->state = (1<< MD_SB_CLEAN);
864 } else
865 sb->recovery_cp = 0;
866
867 sb->layout = mddev->layout;
868 sb->chunk_size = mddev->chunk_size;
869
a654b9d8
N		 870 if (mddev->bitmap && mddev->bitmap_file == NULL)
871 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
872
1da177e4
LT		 873 sb->disks[0].state = (1<<MD_DISK_REMOVED);
874 ITERATE_RDEV(mddev,rdev2,tmp) {
875 mdp_disk_t *d;
86e6ffdd 876 int desc_nr;
b2d444d7
N		 877 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
878 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 879 desc_nr = rdev2->raid_disk;
1da177e4 880 else
86e6ffdd 881 desc_nr = next_spare++;
19133a42 882 rdev2->desc_nr = desc_nr;
1da177e4
LT		 883 d = &sb->disks[rdev2->desc_nr];
884 nr_disks++;
885 d->number = rdev2->desc_nr;
886 d->major = MAJOR(rdev2->bdev->bd_dev);
887 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 888 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
889 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 890 d->raid_disk = rdev2->raid_disk;
891 else
892 d->raid_disk = rdev2->desc_nr; /* compatibility */
b2d444d7 893 if (test_bit(Faulty, &rdev2->flags)) {
1da177e4
LT		 894 d->state = (1<<MD_DISK_FAULTY);
895 failed++;
b2d444d7 896 } else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 897 d->state = (1<<MD_DISK_ACTIVE);
898 d->state |= (1<<MD_DISK_SYNC);
899 active++;
900 working++;
901 } else {
902 d->state = 0;
903 spare++;
904 working++;
905 }
8ddf9efe
N		 906 if (test_bit(WriteMostly, &rdev2->flags))
907 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 908 }
1da177e4
LT		 909 /* now set the "removed" and "faulty" bits on any missing devices */
910 for (i=0 ; i < mddev->raid_disks ; i++) {
911 mdp_disk_t *d = &sb->disks[i];
912 if (d->state == 0 && d->number == 0) {
913 d->number = i;
914 d->raid_disk = i;
915 d->state = (1<<MD_DISK_REMOVED);
916 d->state |= (1<<MD_DISK_FAULTY);
917 failed++;
918 }
919 }
920 sb->nr_disks = nr_disks;
921 sb->active_disks = active;
922 sb->working_disks = working;
923 sb->failed_disks = failed;
924 sb->spare_disks = spare;
925
926 sb->this_disk = sb->disks[rdev->desc_nr];
927 sb->sb_csum = calc_sb_csum(sb);
928}
929
930/*
931 * version 1 superblock
932 */
933
934static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
935{
936 unsigned int disk_csum, csum;
937 unsigned long long newcsum;
938 int size = 256 + le32_to_cpu(sb->max_dev)*2;
939 unsigned int *isuper = (unsigned int*)sb;
940 int i;
941
942 disk_csum = sb->sb_csum;
943 sb->sb_csum = 0;
944 newcsum = 0;
945 for (i=0; size>=4; size -= 4 )
946 newcsum += le32_to_cpu(*isuper++);
947
948 if (size == 2)
949 newcsum += le16_to_cpu(*(unsigned short*) isuper);
950
951 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
952 sb->sb_csum = disk_csum;
953 return cpu_to_le32(csum);
954}
955
956static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
957{
958 struct mdp_superblock_1 *sb;
959 int ret;
960 sector_t sb_offset;
961 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 962 int bmask;
1da177e4
LT		 963
964 /*
965 * Calculate the position of the superblock.
966 * It is always aligned to a 4K boundary and
967 * depeding on minor_version, it can be:
968 * 0: At least 8K, but less than 12K, from end of device
969 * 1: At start of device
970 * 2: 4K from start of device.
971 */
972 switch(minor_version) {
973 case 0:
974 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
975 sb_offset -= 8*2;
39730960 976 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 977 /* convert from sectors to K */
978 sb_offset /= 2;
979 break;
980 case 1:
981 sb_offset = 0;
982 break;
983 case 2:
984 sb_offset = 4;
985 break;
986 default:
987 return -EINVAL;
988 }
989 rdev->sb_offset = sb_offset;
990
0002b271
N		 991 /* superblock is rarely larger than 1K, but it can be larger,
992 * and it is safe to read 4k, so we do that
993 */
994 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 995 if (ret) return ret;
996
997
998 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
999
1000 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1001 sb->major_version != cpu_to_le32(1) ||
1002 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1003 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1004 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1005 return -EINVAL;
1006
1007 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1008 printk("md: invalid superblock checksum on %s\n",
1009 bdevname(rdev->bdev,b));
1010 return -EINVAL;
1011 }
1012 if (le64_to_cpu(sb->data_size) < 10) {
1013 printk("md: data_size too small on %s\n",
1014 bdevname(rdev->bdev,b));
1015 return -EINVAL;
1016 }
1017 rdev->preferred_minor = 0xffff;
1018 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1019 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1020
0002b271 1021 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1022 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1023 if (rdev->sb_size & bmask)
1024 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1025
1da177e4
LT		 1026 if (refdev == 0)
1027 return 1;
1028 else {
1029 __u64 ev1, ev2;
1030 struct mdp_superblock_1 *refsb =
1031 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1032
1033 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1034 sb->level != refsb->level ||
1035 sb->layout != refsb->layout ||
1036 sb->chunksize != refsb->chunksize) {
1037 printk(KERN_WARNING "md: %s has strangely different"
1038 " superblock to %s\n",
1039 bdevname(rdev->bdev,b),
1040 bdevname(refdev->bdev,b2));
1041 return -EINVAL;
1042 }
1043 ev1 = le64_to_cpu(sb->events);
1044 ev2 = le64_to_cpu(refsb->events);
1045
1046 if (ev1 > ev2)
1047 return 1;
1048 }
1049 if (minor_version)
1050 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1051 else
1052 rdev->size = rdev->sb_offset;
1053 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1054 return -EINVAL;
1055 rdev->size = le64_to_cpu(sb->data_size)/2;
1056 if (le32_to_cpu(sb->chunksize))
1057 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf
N		 1058
1059 if (le32_to_cpu(sb->size) > rdev->size*2)
1060 return -EINVAL;
1da177e4
LT		 1061 return 0;
1062}
1063
1064static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1065{
1066 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1067
41158c7e 1068 rdev->raid_disk = -1;
b2d444d7 1069 rdev->flags = 0;
1da177e4
LT		 1070 if (mddev->raid_disks == 0) {
1071 mddev->major_version = 1;
1072 mddev->patch_version = 0;
1073 mddev->persistent = 1;
1074 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1075 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1076 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1077 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1078 mddev->clevel[0] = 0;
1da177e4
LT		 1079 mddev->layout = le32_to_cpu(sb->layout);
1080 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1081 mddev->size = le64_to_cpu(sb->size)/2;
1082 mddev->events = le64_to_cpu(sb->events);
9223214e 1083 mddev->bitmap_offset = 0;
53e87fbb 1084 mddev->default_bitmap_offset = 1024;
1da177e4
LT		 1085
1086 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1087 memcpy(mddev->uuid, sb->set_uuid, 16);
1088
1089 mddev->max_disks = (4096-256)/2;
a654b9d8 1090
71c0805c 1091 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
a654b9d8 1092 mddev->bitmap_file == NULL ) {
6cce3b23
N		 1093 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1094 && mddev->level != 10) {
1095 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 1096 return -EINVAL;
1097 }
1098 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1099 }
41158c7e
N		 1100 } else if (mddev->pers == NULL) {
1101 /* Insist of good event counter while assembling */
1102 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4
LT		 1103 ++ev1;
1104 if (ev1 < mddev->events)
1105 return -EINVAL;
41158c7e
N		 1106 } else if (mddev->bitmap) {
1107 /* If adding to array with a bitmap, then we can accept an
1108 * older device, but not too old.
1109 */
1110 __u64 ev1 = le64_to_cpu(sb->events);
1111 if (ev1 < mddev->bitmap->events_cleared)
1112 return 0;
1113 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1114 return 0;
1da177e4
LT		 1115
1116 if (mddev->level != LEVEL_MULTIPATH) {
1117 int role;
1118 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1119 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1120 switch(role) {
1121 case 0xffff: /* spare */
1da177e4
LT		 1122 break;
1123 case 0xfffe: /* faulty */
b2d444d7 1124 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1125 break;
1126 default:
b2d444d7 1127 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1128 rdev->raid_disk = role;
1129 break;
1130 }
8ddf9efe
N		 1131 if (sb->devflags & WriteMostly1)
1132 set_bit(WriteMostly, &rdev->flags);
41158c7e 1133 } else /* MULTIPATH are always insync */
b2d444d7 1134 set_bit(In_sync, &rdev->flags);
41158c7e 1135
1da177e4
LT		 1136 return 0;
1137}
1138
1139static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1140{
1141 struct mdp_superblock_1 *sb;
1142 struct list_head *tmp;
1143 mdk_rdev_t *rdev2;
1144 int max_dev, i;
1145 /* make rdev->sb match mddev and rdev data. */
1146
1147 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1148
1149 sb->feature_map = 0;
1150 sb->pad0 = 0;
1151 memset(sb->pad1, 0, sizeof(sb->pad1));
1152 memset(sb->pad2, 0, sizeof(sb->pad2));
1153 memset(sb->pad3, 0, sizeof(sb->pad3));
1154
1155 sb->utime = cpu_to_le64((__u64)mddev->utime);
1156 sb->events = cpu_to_le64(mddev->events);
1157 if (mddev->in_sync)
1158 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1159 else
1160 sb->resync_offset = cpu_to_le64(0);
1161
4dbcdc75
N		 1162 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1163
a654b9d8
N		 1164 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1165 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1166 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8
N		 1167 }
1168
1da177e4
LT		 1169 max_dev = 0;
1170 ITERATE_RDEV(mddev,rdev2,tmp)
1171 if (rdev2->desc_nr+1 > max_dev)
1172 max_dev = rdev2->desc_nr+1;
1173
1174 sb->max_dev = cpu_to_le32(max_dev);
1175 for (i=0; i<max_dev;i++)
1176 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1177
1178 ITERATE_RDEV(mddev,rdev2,tmp) {
1179 i = rdev2->desc_nr;
b2d444d7 1180 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1181 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1182 else if (test_bit(In_sync, &rdev2->flags))
1da177e4
LT		 1183 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1184 else
1185 sb->dev_roles[i] = cpu_to_le16(0xffff);
1186 }
1187
1188 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1189 sb->sb_csum = calc_sb_1_csum(sb);
1190}
1191
1192
75c96f85 1193static struct super_type super_types[] = {
1da177e4
LT		 1194 [0] = {
1195 .name = "0.90.0",
1196 .owner = THIS_MODULE,
1197 .load_super = super_90_load,
1198 .validate_super = super_90_validate,
1199 .sync_super = super_90_sync,
1200 },
1201 [1] = {
1202 .name = "md-1",
1203 .owner = THIS_MODULE,
1204 .load_super = super_1_load,
1205 .validate_super = super_1_validate,
1206 .sync_super = super_1_sync,
1207 },
1208};
1209
1210static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1211{
1212 struct list_head *tmp;
1213 mdk_rdev_t *rdev;
1214
1215 ITERATE_RDEV(mddev,rdev,tmp)
1216 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1217 return rdev;
1218
1219 return NULL;
1220}
1221
1222static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1223{
1224 struct list_head *tmp;
1225 mdk_rdev_t *rdev;
1226
1227 ITERATE_RDEV(mddev1,rdev,tmp)
1228 if (match_dev_unit(mddev2, rdev))
1229 return 1;
1230
1231 return 0;
1232}
1233
1234static LIST_HEAD(pending_raid_disks);
1235
1236static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1237{
1238 mdk_rdev_t *same_pdev;
1239 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
f637b9f9 1240 struct kobject *ko;
1edf80d3 1241 char *s;
1da177e4
LT		 1242
1243 if (rdev->mddev) {
1244 MD_BUG();
1245 return -EINVAL;
1246 }
2bf071bf
N		 1247 /* make sure rdev->size exceeds mddev->size */
1248 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1249 if (mddev->pers)
1250 /* Cannot change size, so fail */
1251 return -ENOSPC;
1252 else
1253 mddev->size = rdev->size;
1254 }
1da177e4
LT		 1255 same_pdev = match_dev_unit(mddev, rdev);
1256 if (same_pdev)
1257 printk(KERN_WARNING
1258 "%s: WARNING: %s appears to be on the same physical"
1259 " disk as %s. True\n protection against single-disk"
1260 " failure might be compromised.\n",
1261 mdname(mddev), bdevname(rdev->bdev,b),
1262 bdevname(same_pdev->bdev,b2));
1263
1264 /* Verify rdev->desc_nr is unique.
1265 * If it is -1, assign a free number, else
1266 * check number is not in use
1267 */
1268 if (rdev->desc_nr < 0) {
1269 int choice = 0;
1270 if (mddev->pers) choice = mddev->raid_disks;
1271 while (find_rdev_nr(mddev, choice))
1272 choice++;
1273 rdev->desc_nr = choice;
1274 } else {
1275 if (find_rdev_nr(mddev, rdev->desc_nr))
1276 return -EBUSY;
1277 }
19133a42
N		 1278 bdevname(rdev->bdev,b);
1279 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1280 return -ENOMEM;
1edf80d3
NB		 1281 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1282 *s = '!';
1da177e4
LT		 1283
1284 list_add(&rdev->same_set, &mddev->disks);
1285 rdev->mddev = mddev;
19133a42 1286 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1287
9c791977 1288 rdev->kobj.parent = &mddev->kobj;
86e6ffdd
N		 1289 kobject_add(&rdev->kobj);
1290
f637b9f9
N		 1291 if (rdev->bdev->bd_part)
1292 ko = &rdev->bdev->bd_part->kobj;
1293 else
1294 ko = &rdev->bdev->bd_disk->kobj;
1295 sysfs_create_link(&rdev->kobj, ko, "block");
1da177e4
LT		 1296 return 0;
1297}
1298
1299static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1300{
1301 char b[BDEVNAME_SIZE];
1302 if (!rdev->mddev) {
1303 MD_BUG();
1304 return;
1305 }
1306 list_del_init(&rdev->same_set);
1307 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1308 rdev->mddev = NULL;
86e6ffdd
N		 1309 sysfs_remove_link(&rdev->kobj, "block");
1310 kobject_del(&rdev->kobj);
1da177e4
LT		 1311}
1312
1313/*
1314 * prevent the device from being mounted, repartitioned or
1315 * otherwise reused by a RAID array (or any other kernel
1316 * subsystem), by bd_claiming the device.
1317 */
1318static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1319{
1320 int err = 0;
1321 struct block_device *bdev;
1322 char b[BDEVNAME_SIZE];
1323
1324 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1325 if (IS_ERR(bdev)) {
1326 printk(KERN_ERR "md: could not open %s.\n",
1327 __bdevname(dev, b));
1328 return PTR_ERR(bdev);
1329 }
1330 err = bd_claim(bdev, rdev);
1331 if (err) {
1332 printk(KERN_ERR "md: could not bd_claim %s.\n",
1333 bdevname(bdev, b));
1334 blkdev_put(bdev);
1335 return err;
1336 }
1337 rdev->bdev = bdev;
1338 return err;
1339}
1340
1341static void unlock_rdev(mdk_rdev_t *rdev)
1342{
1343 struct block_device *bdev = rdev->bdev;
1344 rdev->bdev = NULL;
1345 if (!bdev)
1346 MD_BUG();
1347 bd_release(bdev);
1348 blkdev_put(bdev);
1349}
1350
1351void md_autodetect_dev(dev_t dev);
1352
1353static void export_rdev(mdk_rdev_t * rdev)
1354{
1355 char b[BDEVNAME_SIZE];
1356 printk(KERN_INFO "md: export_rdev(%s)\n",
1357 bdevname(rdev->bdev,b));
1358 if (rdev->mddev)
1359 MD_BUG();
1360 free_disk_sb(rdev);
1361 list_del_init(&rdev->same_set);
1362#ifndef MODULE
1363 md_autodetect_dev(rdev->bdev->bd_dev);
1364#endif
1365 unlock_rdev(rdev);
86e6ffdd 1366 kobject_put(&rdev->kobj);
1da177e4
LT		 1367}
1368
1369static void kick_rdev_from_array(mdk_rdev_t * rdev)
1370{
1371 unbind_rdev_from_array(rdev);
1372 export_rdev(rdev);
1373}
1374
1375static void export_array(mddev_t *mddev)
1376{
1377 struct list_head *tmp;
1378 mdk_rdev_t *rdev;
1379
1380 ITERATE_RDEV(mddev,rdev,tmp) {
1381 if (!rdev->mddev) {
1382 MD_BUG();
1383 continue;
1384 }
1385 kick_rdev_from_array(rdev);
1386 }
1387 if (!list_empty(&mddev->disks))
1388 MD_BUG();
1389 mddev->raid_disks = 0;
1390 mddev->major_version = 0;
1391}
1392
1393static void print_desc(mdp_disk_t *desc)
1394{
1395 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1396 desc->major,desc->minor,desc->raid_disk,desc->state);
1397}
1398
1399static void print_sb(mdp_super_t *sb)
1400{
1401 int i;
1402
1403 printk(KERN_INFO
1404 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1405 sb->major_version, sb->minor_version, sb->patch_version,
1406 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1407 sb->ctime);
1408 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1409 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1410 sb->md_minor, sb->layout, sb->chunk_size);
1411 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1412 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1413 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1414 sb->failed_disks, sb->spare_disks,
1415 sb->sb_csum, (unsigned long)sb->events_lo);
1416
1417 printk(KERN_INFO);
1418 for (i = 0; i < MD_SB_DISKS; i++) {
1419 mdp_disk_t *desc;
1420
1421 desc = sb->disks + i;
1422 if (desc->number || desc->major || desc->minor ||
1423 desc->raid_disk || (desc->state && (desc->state != 4))) {
1424 printk(" D %2d: ", i);
1425 print_desc(desc);
1426 }
1427 }
1428 printk(KERN_INFO "md: THIS: ");
1429 print_desc(&sb->this_disk);
1430
1431}
1432
1433static void print_rdev(mdk_rdev_t *rdev)
1434{
1435 char b[BDEVNAME_SIZE];
1436 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1437 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1438 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1439 rdev->desc_nr);
1da177e4
LT		 1440 if (rdev->sb_loaded) {
1441 printk(KERN_INFO "md: rdev superblock:\n");
1442 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1443 } else
1444 printk(KERN_INFO "md: no rdev superblock!\n");
1445}
1446
1447void md_print_devices(void)
1448{
1449 struct list_head *tmp, *tmp2;
1450 mdk_rdev_t *rdev;
1451 mddev_t *mddev;
1452 char b[BDEVNAME_SIZE];
1453
1454 printk("\n");
1455 printk("md: **********************************\n");
1456 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1457 printk("md: **********************************\n");
1458 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1459
32a7627c
N		 1460 if (mddev->bitmap)
1461 bitmap_print_sb(mddev->bitmap);
1462 else
1463 printk("%s: ", mdname(mddev));
1da177e4
LT		 1464 ITERATE_RDEV(mddev,rdev,tmp2)
1465 printk("<%s>", bdevname(rdev->bdev,b));
1466 printk("\n");
1467
1468 ITERATE_RDEV(mddev,rdev,tmp2)
1469 print_rdev(rdev);
1470 }
1471 printk("md: **********************************\n");
1472 printk("\n");
1473}
1474
1475
1da177e4
LT		 1476static void sync_sbs(mddev_t * mddev)
1477{
1478 mdk_rdev_t *rdev;
1479 struct list_head *tmp;
1480
1481 ITERATE_RDEV(mddev,rdev,tmp) {
1482 super_types[mddev->major_version].
1483 sync_super(mddev, rdev);
1484 rdev->sb_loaded = 1;
1485 }
1486}
1487
1488static void md_update_sb(mddev_t * mddev)
1489{
7bfa19f2 1490 int err;
1da177e4
LT		 1491 struct list_head *tmp;
1492 mdk_rdev_t *rdev;
06d91a5f 1493 int sync_req;
1da177e4 1494
1da177e4 1495repeat:
a9701a30 1496 spin_lock_irq(&mddev->write_lock);
06d91a5f 1497 sync_req = mddev->in_sync;
1da177e4
LT		 1498 mddev->utime = get_seconds();
1499 mddev->events ++;
1500
1501 if (!mddev->events) {
1502 /*
1503 * oops, this 64-bit counter should never wrap.
1504 * Either we are in around ~1 trillion A.C., assuming
1505 * 1 reboot per second, or we have a bug:
1506 */
1507 MD_BUG();
1508 mddev->events --;
1509 }
7bfa19f2 1510 mddev->sb_dirty = 2;
1da177e4
LT		 1511 sync_sbs(mddev);
1512
1513 /*
1514 * do not write anything to disk if using
1515 * nonpersistent superblocks
1516 */
06d91a5f
N		 1517 if (!mddev->persistent) {
1518 mddev->sb_dirty = 0;
a9701a30 1519 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1520 wake_up(&mddev->sb_wait);
1da177e4 1521 return;
06d91a5f 1522 }
a9701a30 1523 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1524
1525 dprintk(KERN_INFO
1526 "md: updating %s RAID superblock on device (in sync %d)\n",
1527 mdname(mddev),mddev->in_sync);
1528
32a7627c 1529 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1530 ITERATE_RDEV(mddev,rdev,tmp) {
1531 char b[BDEVNAME_SIZE];
1532 dprintk(KERN_INFO "md: ");
b2d444d7 1533 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1534 dprintk("(skipping faulty ");
1535
1536 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1537 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1538 md_super_write(mddev,rdev,
0002b271 1539 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1540 rdev->sb_page);
1541 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1542 bdevname(rdev->bdev,b),
1543 (unsigned long long)rdev->sb_offset);
1544
1da177e4
LT		 1545 } else
1546 dprintk(")\n");
7bfa19f2 1547 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1548 /* only need to write one superblock... */
1549 break;
1550 }
a9701a30 1551 md_super_wait(mddev);
7bfa19f2
N		 1552 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1553
a9701a30 1554 spin_lock_irq(&mddev->write_lock);
7bfa19f2 1555 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
06d91a5f 1556 /* have to write it out again */
a9701a30 1557 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1558 goto repeat;
1559 }
1560 mddev->sb_dirty = 0;
a9701a30 1561 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1562 wake_up(&mddev->sb_wait);
06d91a5f 1563
1da177e4
LT		 1564}
1565
bce74dac
N		 1566/* words written to sysfs files may, or my not, be \n terminated.
1567 * We want to accept with case. For this we use cmd_match.
1568 */
1569static int cmd_match(const char *cmd, const char *str)
1570{
1571 /* See if cmd, written into a sysfs file, matches
1572 * str. They must either be the same, or cmd can
1573 * have a trailing newline
1574 */
1575 while (*cmd && *str && *cmd == *str) {
1576 cmd++;
1577 str++;
1578 }
1579 if (*cmd == '\n')
1580 cmd++;
1581 if (*str || *cmd)
1582 return 0;
1583 return 1;
1584}
1585
86e6ffdd
N		 1586struct rdev_sysfs_entry {
1587 struct attribute attr;
1588 ssize_t (*show)(mdk_rdev_t *, char *);
1589 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1590};
1591
1592static ssize_t
96de1e66 1593state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1594{
1595 char *sep = "";
1596 int len=0;
1597
b2d444d7 1598 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1599 len+= sprintf(page+len, "%sfaulty",sep);
1600 sep = ",";
1601 }
b2d444d7 1602 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1603 len += sprintf(page+len, "%sin_sync",sep);
1604 sep = ",";
1605 }
b2d444d7
N		 1606 if (!test_bit(Faulty, &rdev->flags) &&
1607 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1608 len += sprintf(page+len, "%sspare", sep);
1609 sep = ",";
1610 }
1611 return len+sprintf(page+len, "\n");
1612}
1613
96de1e66
N		 1614static struct rdev_sysfs_entry
1615rdev_state = __ATTR_RO(state);
86e6ffdd
N		 1616
1617static ssize_t
96de1e66 1618super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1619{
1620 if (rdev->sb_loaded && rdev->sb_size) {
1621 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1622 return rdev->sb_size;
1623 } else
1624 return 0;
1625}
96de1e66
N		 1626static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1627
4dbcdc75
N		 1628static ssize_t
1629errors_show(mdk_rdev_t *rdev, char *page)
1630{
1631 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1632}
1633
1634static ssize_t
1635errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1636{
1637 char *e;
1638 unsigned long n = simple_strtoul(buf, &e, 10);
1639 if (*buf && (*e == 0 || *e == '\n')) {
1640 atomic_set(&rdev->corrected_errors, n);
1641 return len;
1642 }
1643 return -EINVAL;
1644}
1645static struct rdev_sysfs_entry rdev_errors =
1646__ATTR(errors, 0644, errors_show, errors_store);
1647
014236d2
N		 1648static ssize_t
1649slot_show(mdk_rdev_t *rdev, char *page)
1650{
1651 if (rdev->raid_disk < 0)
1652 return sprintf(page, "none\n");
1653 else
1654 return sprintf(page, "%d\n", rdev->raid_disk);
1655}
1656
1657static ssize_t
1658slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1659{
1660 char *e;
1661 int slot = simple_strtoul(buf, &e, 10);
1662 if (strncmp(buf, "none", 4)==0)
1663 slot = -1;
1664 else if (e==buf || (*e && *e!= '\n'))
1665 return -EINVAL;
1666 if (rdev->mddev->pers)
1667 /* Cannot set slot in active array (yet) */
1668 return -EBUSY;
1669 if (slot >= rdev->mddev->raid_disks)
1670 return -ENOSPC;
1671 rdev->raid_disk = slot;
1672 /* assume it is working */
1673 rdev->flags = 0;
1674 set_bit(In_sync, &rdev->flags);
1675 return len;
1676}
1677
1678
1679static struct rdev_sysfs_entry rdev_slot =
1680__ATTR(slot, 0644, slot_show, slot_store);
1681
93c8cad0
N		 1682static ssize_t
1683offset_show(mdk_rdev_t *rdev, char *page)
1684{
6961ece4 1685 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1686}
1687
1688static ssize_t
1689offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1690{
1691 char *e;
1692 unsigned long long offset = simple_strtoull(buf, &e, 10);
1693 if (e==buf || (*e && *e != '\n'))
1694 return -EINVAL;
1695 if (rdev->mddev->pers)
1696 return -EBUSY;
1697 rdev->data_offset = offset;
1698 return len;
1699}
1700
1701static struct rdev_sysfs_entry rdev_offset =
1702__ATTR(offset, 0644, offset_show, offset_store);
1703
83303b61
N		 1704static ssize_t
1705rdev_size_show(mdk_rdev_t *rdev, char *page)
1706{
1707 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1708}
1709
1710static ssize_t
1711rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1712{
1713 char *e;
1714 unsigned long long size = simple_strtoull(buf, &e, 10);
1715 if (e==buf || (*e && *e != '\n'))
1716 return -EINVAL;
1717 if (rdev->mddev->pers)
1718 return -EBUSY;
1719 rdev->size = size;
1720 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1721 rdev->mddev->size = size;
1722 return len;
1723}
1724
1725static struct rdev_sysfs_entry rdev_size =
1726__ATTR(size, 0644, rdev_size_show, rdev_size_store);
1727
86e6ffdd
N		 1728static struct attribute *rdev_default_attrs[] = {
1729 &rdev_state.attr,
1730 &rdev_super.attr,
4dbcdc75 1731 &rdev_errors.attr,
014236d2 1732 &rdev_slot.attr,
93c8cad0 1733 &rdev_offset.attr,
83303b61 1734 &rdev_size.attr,
86e6ffdd
N		 1735 NULL,
1736};
1737static ssize_t
1738rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1739{
1740 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1741 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1742
1743 if (!entry->show)
1744 return -EIO;
1745 return entry->show(rdev, page);
1746}
1747
1748static ssize_t
1749rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1750 const char *page, size_t length)
1751{
1752 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1753 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1754
1755 if (!entry->store)
1756 return -EIO;
1757 return entry->store(rdev, page, length);
1758}
1759
1760static void rdev_free(struct kobject *ko)
1761{
1762 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1763 kfree(rdev);
1764}
1765static struct sysfs_ops rdev_sysfs_ops = {
1766 .show = rdev_attr_show,
1767 .store = rdev_attr_store,
1768};
1769static struct kobj_type rdev_ktype = {
1770 .release = rdev_free,
1771 .sysfs_ops = &rdev_sysfs_ops,
1772 .default_attrs = rdev_default_attrs,
1773};
1774
1da177e4
LT		 1775/*
1776 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1777 *
1778 * mark the device faulty if:
1779 *
1780 * - the device is nonexistent (zero size)
1781 * - the device has no valid superblock
1782 *
1783 * a faulty rdev _never_ has rdev->sb set.
1784 */
1785static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1786{
1787 char b[BDEVNAME_SIZE];
1788 int err;
1789 mdk_rdev_t *rdev;
1790 sector_t size;
1791
9ffae0cf 1792 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 1793 if (!rdev) {
1794 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1795 return ERR_PTR(-ENOMEM);
1796 }
1da177e4
LT		 1797
1798 if ((err = alloc_disk_sb(rdev)))
1799 goto abort_free;
1800
1801 err = lock_rdev(rdev, newdev);
1802 if (err)
1803 goto abort_free;
1804
86e6ffdd
N		 1805 rdev->kobj.parent = NULL;
1806 rdev->kobj.ktype = &rdev_ktype;
1807 kobject_init(&rdev->kobj);
1808
1da177e4 1809 rdev->desc_nr = -1;
b2d444d7 1810 rdev->flags = 0;
1da177e4
LT		 1811 rdev->data_offset = 0;
1812 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 1813 atomic_set(&rdev->read_errors, 0);
4dbcdc75 1814 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 1815
1816 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1817 if (!size) {
1818 printk(KERN_WARNING
1819 "md: %s has zero or unknown size, marking faulty!\n",
1820 bdevname(rdev->bdev,b));
1821 err = -EINVAL;
1822 goto abort_free;
1823 }
1824
1825 if (super_format >= 0) {
1826 err = super_types[super_format].
1827 load_super(rdev, NULL, super_minor);
1828 if (err == -EINVAL) {
1829 printk(KERN_WARNING
1830 "md: %s has invalid sb, not importing!\n",
1831 bdevname(rdev->bdev,b));
1832 goto abort_free;
1833 }
1834 if (err < 0) {
1835 printk(KERN_WARNING
1836 "md: could not read %s's sb, not importing!\n",
1837 bdevname(rdev->bdev,b));
1838 goto abort_free;
1839 }
1840 }
1841 INIT_LIST_HEAD(&rdev->same_set);
1842
1843 return rdev;
1844
1845abort_free:
1846 if (rdev->sb_page) {
1847 if (rdev->bdev)
1848 unlock_rdev(rdev);
1849 free_disk_sb(rdev);
1850 }
1851 kfree(rdev);
1852 return ERR_PTR(err);
1853}
1854
1855/*
1856 * Check a full RAID array for plausibility
1857 */
1858
1859
a757e64c 1860static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 1861{
1862 int i;
1863 struct list_head *tmp;
1864 mdk_rdev_t *rdev, *freshest;
1865 char b[BDEVNAME_SIZE];
1866
1867 freshest = NULL;
1868 ITERATE_RDEV(mddev,rdev,tmp)
1869 switch (super_types[mddev->major_version].
1870 load_super(rdev, freshest, mddev->minor_version)) {
1871 case 1:
1872 freshest = rdev;
1873 break;
1874 case 0:
1875 break;
1876 default:
1877 printk( KERN_ERR \
1878 "md: fatal superblock inconsistency in %s"
1879 " -- removing from array\n",
1880 bdevname(rdev->bdev,b));
1881 kick_rdev_from_array(rdev);
1882 }
1883
1884
1885 super_types[mddev->major_version].
1886 validate_super(mddev, freshest);
1887
1888 i = 0;
1889 ITERATE_RDEV(mddev,rdev,tmp) {
1890 if (rdev != freshest)
1891 if (super_types[mddev->major_version].
1892 validate_super(mddev, rdev)) {
1893 printk(KERN_WARNING "md: kicking non-fresh %s"
1894 " from array!\n",
1895 bdevname(rdev->bdev,b));
1896 kick_rdev_from_array(rdev);
1897 continue;
1898 }
1899 if (mddev->level == LEVEL_MULTIPATH) {
1900 rdev->desc_nr = i++;
1901 rdev->raid_disk = rdev->desc_nr;
b2d444d7 1902 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1903 }
1904 }
1905
1906
1907
1908 if (mddev->recovery_cp != MaxSector &&
1909 mddev->level >= 1)
1910 printk(KERN_ERR "md: %s: raid array is not clean"
1911 " -- starting background reconstruction\n",
1912 mdname(mddev));
1913
1da177e4
LT		 1914}
1915
eae1701f 1916static ssize_t
96de1e66 1917level_show(mddev_t *mddev, char *page)
eae1701f 1918{
2604b703 1919 struct mdk_personality *p = mddev->pers;
d9d166c2 1920 if (p)
eae1701f 1921 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 1922 else if (mddev->clevel[0])
1923 return sprintf(page, "%s\n", mddev->clevel);
1924 else if (mddev->level != LEVEL_NONE)
1925 return sprintf(page, "%d\n", mddev->level);
1926 else
1927 return 0;
eae1701f
N		 1928}
1929
d9d166c2
N		 1930static ssize_t
1931level_store(mddev_t *mddev, const char *buf, size_t len)
1932{
1933 int rv = len;
1934 if (mddev->pers)
1935 return -EBUSY;
1936 if (len == 0)
1937 return 0;
1938 if (len >= sizeof(mddev->clevel))
1939 return -ENOSPC;
1940 strncpy(mddev->clevel, buf, len);
1941 if (mddev->clevel[len-1] == '\n')
1942 len--;
1943 mddev->clevel[len] = 0;
1944 mddev->level = LEVEL_NONE;
1945 return rv;
1946}
1947
1948static struct md_sysfs_entry md_level =
1949__ATTR(level, 0644, level_show, level_store);
eae1701f
N		 1950
1951static ssize_t
96de1e66 1952raid_disks_show(mddev_t *mddev, char *page)
eae1701f 1953{
bb636547
N		 1954 if (mddev->raid_disks == 0)
1955 return 0;
eae1701f
N		 1956 return sprintf(page, "%d\n", mddev->raid_disks);
1957}
1958
da943b99
N		 1959static int update_raid_disks(mddev_t *mddev, int raid_disks);
1960
1961static ssize_t
1962raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1963{
1964 /* can only set raid_disks if array is not yet active */
1965 char *e;
1966 int rv = 0;
1967 unsigned long n = simple_strtoul(buf, &e, 10);
1968
1969 if (!*buf || (*e && *e != '\n'))
1970 return -EINVAL;
1971
1972 if (mddev->pers)
1973 rv = update_raid_disks(mddev, n);
1974 else
1975 mddev->raid_disks = n;
1976 return rv ? rv : len;
1977}
1978static struct md_sysfs_entry md_raid_disks =
1979__ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
eae1701f 1980
3b34380a
N		 1981static ssize_t
1982chunk_size_show(mddev_t *mddev, char *page)
1983{
1984 return sprintf(page, "%d\n", mddev->chunk_size);
1985}
1986
1987static ssize_t
1988chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1989{
1990 /* can only set chunk_size if array is not yet active */
1991 char *e;
1992 unsigned long n = simple_strtoul(buf, &e, 10);
1993
1994 if (mddev->pers)
1995 return -EBUSY;
1996 if (!*buf || (*e && *e != '\n'))
1997 return -EINVAL;
1998
1999 mddev->chunk_size = n;
2000 return len;
2001}
2002static struct md_sysfs_entry md_chunk_size =
2003__ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2004
6d7ff738
N		 2005static ssize_t
2006null_show(mddev_t *mddev, char *page)
2007{
2008 return -EINVAL;
2009}
2010
2011static ssize_t
2012new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2013{
2014 /* buf must be %d:%d\n? giving major and minor numbers */
2015 /* The new device is added to the array.
2016 * If the array has a persistent superblock, we read the
2017 * superblock to initialise info and check validity.
2018 * Otherwise, only checking done is that in bind_rdev_to_array,
2019 * which mainly checks size.
2020 */
2021 char *e;
2022 int major = simple_strtoul(buf, &e, 10);
2023 int minor;
2024 dev_t dev;
2025 mdk_rdev_t *rdev;
2026 int err;
2027
2028 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2029 return -EINVAL;
2030 minor = simple_strtoul(e+1, &e, 10);
2031 if (*e && *e != '\n')
2032 return -EINVAL;
2033 dev = MKDEV(major, minor);
2034 if (major != MAJOR(dev) ||
2035 minor != MINOR(dev))
2036 return -EOVERFLOW;
2037
2038
2039 if (mddev->persistent) {
2040 rdev = md_import_device(dev, mddev->major_version,
2041 mddev->minor_version);
2042 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2043 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2044 mdk_rdev_t, same_set);
2045 err = super_types[mddev->major_version]
2046 .load_super(rdev, rdev0, mddev->minor_version);
2047 if (err < 0)
2048 goto out;
2049 }
2050 } else
2051 rdev = md_import_device(dev, -1, -1);
2052
2053 if (IS_ERR(rdev))
2054 return PTR_ERR(rdev);
2055 err = bind_rdev_to_array(rdev, mddev);
2056 out:
2057 if (err)
2058 export_rdev(rdev);
2059 return err ? err : len;
2060}
2061
2062static struct md_sysfs_entry md_new_device =
2063__ATTR(new_dev, 0200, null_show, new_dev_store);
3b34380a 2064
a35b0d69
N		 2065static ssize_t
2066size_show(mddev_t *mddev, char *page)
2067{
2068 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2069}
2070
2071static int update_size(mddev_t *mddev, unsigned long size);
2072
2073static ssize_t
2074size_store(mddev_t *mddev, const char *buf, size_t len)
2075{
2076 /* If array is inactive, we can reduce the component size, but
2077 * not increase it (except from 0).
2078 * If array is active, we can try an on-line resize
2079 */
2080 char *e;
2081 int err = 0;
2082 unsigned long long size = simple_strtoull(buf, &e, 10);
2083 if (!*buf || *buf == '\n' ||
2084 (*e && *e != '\n'))
2085 return -EINVAL;
2086
2087 if (mddev->pers) {
2088 err = update_size(mddev, size);
2089 md_update_sb(mddev);
2090 } else {
2091 if (mddev->size == 0 ||
2092 mddev->size > size)
2093 mddev->size = size;
2094 else
2095 err = -ENOSPC;
2096 }
2097 return err ? err : len;
2098}
2099
2100static struct md_sysfs_entry md_size =
2101__ATTR(component_size, 0644, size_show, size_store);
2102
8bb93aac
N		 2103
2104/* Metdata version.
2105 * This is either 'none' for arrays with externally managed metadata,
2106 * or N.M for internally known formats
2107 */
2108static ssize_t
2109metadata_show(mddev_t *mddev, char *page)
2110{
2111 if (mddev->persistent)
2112 return sprintf(page, "%d.%d\n",
2113 mddev->major_version, mddev->minor_version);
2114 else
2115 return sprintf(page, "none\n");
2116}
2117
2118static ssize_t
2119metadata_store(mddev_t *mddev, const char *buf, size_t len)
2120{
2121 int major, minor;
2122 char *e;
2123 if (!list_empty(&mddev->disks))
2124 return -EBUSY;
2125
2126 if (cmd_match(buf, "none")) {
2127 mddev->persistent = 0;
2128 mddev->major_version = 0;
2129 mddev->minor_version = 90;
2130 return len;
2131 }
2132 major = simple_strtoul(buf, &e, 10);
2133 if (e==buf || *e != '.')
2134 return -EINVAL;
2135 buf = e+1;
2136 minor = simple_strtoul(buf, &e, 10);
2137 if (e==buf || *e != '\n')
2138 return -EINVAL;
2139 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2140 super_types[major].name == NULL)
2141 return -ENOENT;
2142 mddev->major_version = major;
2143 mddev->minor_version = minor;
2144 mddev->persistent = 1;
2145 return len;
2146}
2147
2148static struct md_sysfs_entry md_metadata =
2149__ATTR(metadata_version, 0644, metadata_show, metadata_store);
2150
24dd469d 2151static ssize_t
7eec314d 2152action_show(mddev_t *mddev, char *page)
24dd469d 2153{
7eec314d 2154 char *type = "idle";
31399d9e
N		 2155 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2156 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2157 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2158 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2159 type = "resync";
2160 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2161 type = "check";
2162 else
2163 type = "repair";
2164 } else
2165 type = "recover";
2166 }
2167 return sprintf(page, "%s\n", type);
2168}
2169
2170static ssize_t
7eec314d 2171action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2172{
7eec314d
N		 2173 if (!mddev->pers || !mddev->pers->sync_request)
2174 return -EINVAL;
2175
bce74dac 2176 if (cmd_match(page, "idle")) {
7eec314d
N		 2177 if (mddev->sync_thread) {
2178 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2179 md_unregister_thread(mddev->sync_thread);
2180 mddev->sync_thread = NULL;
2181 mddev->recovery = 0;
2182 }
03c902e1
N		 2183 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2184 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2185 return -EBUSY;
03c902e1 2186 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d
N		 2187 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2188 else {
bce74dac 2189 if (cmd_match(page, "check"))
7eec314d 2190 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
bce74dac 2191 else if (cmd_match(page, "repair"))
7eec314d
N		 2192 return -EINVAL;
2193 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2194 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2195 }
03c902e1 2196 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2197 md_wakeup_thread(mddev->thread);
2198 return len;
2199}
2200
9d88883e 2201static ssize_t
96de1e66 2202mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2203{
2204 return sprintf(page, "%llu\n",
2205 (unsigned long long) mddev->resync_mismatches);
2206}
2207
96de1e66 2208static struct md_sysfs_entry
7eec314d 2209md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2210
96de1e66
N		 2211
2212static struct md_sysfs_entry
2213md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2214
88202a0c
N		 2215static ssize_t
2216sync_min_show(mddev_t *mddev, char *page)
2217{
2218 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2219 mddev->sync_speed_min ? "local": "system");
2220}
2221
2222static ssize_t
2223sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2224{
2225 int min;
2226 char *e;
2227 if (strncmp(buf, "system", 6)==0) {
2228 mddev->sync_speed_min = 0;
2229 return len;
2230 }
2231 min = simple_strtoul(buf, &e, 10);
2232 if (buf == e || (*e && *e != '\n') || min <= 0)
2233 return -EINVAL;
2234 mddev->sync_speed_min = min;
2235 return len;
2236}
2237
2238static struct md_sysfs_entry md_sync_min =
2239__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2240
2241static ssize_t
2242sync_max_show(mddev_t *mddev, char *page)
2243{
2244 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2245 mddev->sync_speed_max ? "local": "system");
2246}
2247
2248static ssize_t
2249sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2250{
2251 int max;
2252 char *e;
2253 if (strncmp(buf, "system", 6)==0) {
2254 mddev->sync_speed_max = 0;
2255 return len;
2256 }
2257 max = simple_strtoul(buf, &e, 10);
2258 if (buf == e || (*e && *e != '\n') || max <= 0)
2259 return -EINVAL;
2260 mddev->sync_speed_max = max;
2261 return len;
2262}
2263
2264static struct md_sysfs_entry md_sync_max =
2265__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2266
2267
2268static ssize_t
2269sync_speed_show(mddev_t *mddev, char *page)
2270{
2271 unsigned long resync, dt, db;
2272 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2273 dt = ((jiffies - mddev->resync_mark) / HZ);
2274 if (!dt) dt++;
2275 db = resync - (mddev->resync_mark_cnt);
2276 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2277}
2278
2279static struct md_sysfs_entry
2280md_sync_speed = __ATTR_RO(sync_speed);
2281
2282static ssize_t
2283sync_completed_show(mddev_t *mddev, char *page)
2284{
2285 unsigned long max_blocks, resync;
2286
2287 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2288 max_blocks = mddev->resync_max_sectors;
2289 else
2290 max_blocks = mddev->size << 1;
2291
2292 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2293 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2294}
2295
2296static struct md_sysfs_entry
2297md_sync_completed = __ATTR_RO(sync_completed);
2298
eae1701f
N		 2299static struct attribute *md_default_attrs[] = {
2300 &md_level.attr,
2301 &md_raid_disks.attr,
3b34380a 2302 &md_chunk_size.attr,
a35b0d69 2303 &md_size.attr,
8bb93aac 2304 &md_metadata.attr,
6d7ff738 2305 &md_new_device.attr,
411036fa
N		 2306 NULL,
2307};
2308
2309static struct attribute *md_redundancy_attrs[] = {
24dd469d 2310 &md_scan_mode.attr,
9d88883e 2311 &md_mismatches.attr,
88202a0c
N		 2312 &md_sync_min.attr,
2313 &md_sync_max.attr,
2314 &md_sync_speed.attr,
2315 &md_sync_completed.attr,
eae1701f
N		 2316 NULL,
2317};
411036fa
N		 2318static struct attribute_group md_redundancy_group = {
2319 .name = NULL,
2320 .attrs = md_redundancy_attrs,
2321};
2322
eae1701f
N		 2323
2324static ssize_t
2325md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2326{
2327 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2328 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2329 ssize_t rv;
eae1701f
N		 2330
2331 if (!entry->show)
2332 return -EIO;
96de1e66
N		 2333 mddev_lock(mddev);
2334 rv = entry->show(mddev, page);
2335 mddev_unlock(mddev);
2336 return rv;
eae1701f
N		 2337}
2338
2339static ssize_t
2340md_attr_store(struct kobject *kobj, struct attribute *attr,
2341 const char *page, size_t length)
2342{
2343 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2344 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2345 ssize_t rv;
eae1701f
N		 2346
2347 if (!entry->store)
2348 return -EIO;
96de1e66
N		 2349 mddev_lock(mddev);
2350 rv = entry->store(mddev, page, length);
2351 mddev_unlock(mddev);
2352 return rv;
eae1701f
N		 2353}
2354
2355static void md_free(struct kobject *ko)
2356{
2357 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2358 kfree(mddev);
2359}
2360
2361static struct sysfs_ops md_sysfs_ops = {
2362 .show = md_attr_show,
2363 .store = md_attr_store,
2364};
2365static struct kobj_type md_ktype = {
2366 .release = md_free,
2367 .sysfs_ops = &md_sysfs_ops,
2368 .default_attrs = md_default_attrs,
2369};
2370
1da177e4
LT		 2371int mdp_major = 0;
2372
2373static struct kobject *md_probe(dev_t dev, int *part, void *data)
2374{
2375 static DECLARE_MUTEX(disks_sem);
2376 mddev_t *mddev = mddev_find(dev);
2377 struct gendisk *disk;
2378 int partitioned = (MAJOR(dev) != MD_MAJOR);
2379 int shift = partitioned ? MdpMinorShift : 0;
2380 int unit = MINOR(dev) >> shift;
2381
2382 if (!mddev)
2383 return NULL;
2384
2385 down(&disks_sem);
2386 if (mddev->gendisk) {
2387 up(&disks_sem);
2388 mddev_put(mddev);
2389 return NULL;
2390 }
2391 disk = alloc_disk(1 << shift);
2392 if (!disk) {
2393 up(&disks_sem);
2394 mddev_put(mddev);
2395 return NULL;
2396 }
2397 disk->major = MAJOR(dev);
2398 disk->first_minor = unit << shift;
2399 if (partitioned) {
2400 sprintf(disk->disk_name, "md_d%d", unit);
2401 sprintf(disk->devfs_name, "md/d%d", unit);
2402 } else {
2403 sprintf(disk->disk_name, "md%d", unit);
2404 sprintf(disk->devfs_name, "md/%d", unit);
2405 }
2406 disk->fops = &md_fops;
2407 disk->private_data = mddev;
2408 disk->queue = mddev->queue;
2409 add_disk(disk);
2410 mddev->gendisk = disk;
2411 up(&disks_sem);
9c791977 2412 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 2413 mddev->kobj.k_name = NULL;
2414 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2415 mddev->kobj.ktype = &md_ktype;
2416 kobject_register(&mddev->kobj);
1da177e4
LT		 2417 return NULL;
2418}
2419
2420void md_wakeup_thread(mdk_thread_t *thread);
2421
2422static void md_safemode_timeout(unsigned long data)
2423{
2424 mddev_t *mddev = (mddev_t *) data;
2425
2426 mddev->safemode = 1;
2427 md_wakeup_thread(mddev->thread);
2428}
2429
6ff8d8ec 2430static int start_dirty_degraded;
1da177e4
LT		 2431
2432static int do_md_run(mddev_t * mddev)
2433{
2604b703 2434 int err;
1da177e4
LT		 2435 int chunk_size;
2436 struct list_head *tmp;
2437 mdk_rdev_t *rdev;
2438 struct gendisk *disk;
2604b703 2439 struct mdk_personality *pers;
1da177e4
LT		 2440 char b[BDEVNAME_SIZE];
2441
a757e64c
N		 2442 if (list_empty(&mddev->disks))
2443 /* cannot run an array with no devices.. */
1da177e4 2444 return -EINVAL;
1da177e4
LT		 2445
2446 if (mddev->pers)
2447 return -EBUSY;
2448
2449 /*
2450 * Analyze all RAID superblock(s)
2451 */
a757e64c
N		 2452 if (!mddev->raid_disks)
2453 analyze_sbs(mddev);
1da177e4
LT		 2454
2455 chunk_size = mddev->chunk_size;
2604b703
N		 2456
2457 if (chunk_size) {
1da177e4
LT		 2458 if (chunk_size > MAX_CHUNK_SIZE) {
2459 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2460 chunk_size, MAX_CHUNK_SIZE);
2461 return -EINVAL;
2462 }
2463 /*
2464 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2465 */
2466 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 2467 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 2468 return -EINVAL;
2469 }
2470 if (chunk_size < PAGE_SIZE) {
2471 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2472 chunk_size, PAGE_SIZE);
2473 return -EINVAL;
2474 }
2475
2476 /* devices must have minimum size of one chunk */
2477 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2478 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2479 continue;
2480 if (rdev->size < chunk_size / 1024) {
2481 printk(KERN_WARNING
2482 "md: Dev %s smaller than chunk_size:"
2483 " %lluk < %dk\n",
2484 bdevname(rdev->bdev,b),
2485 (unsigned long long)rdev->size,
2486 chunk_size / 1024);
2487 return -EINVAL;
2488 }
2489 }
2490 }
2491
1da177e4 2492#ifdef CONFIG_KMOD
d9d166c2
N		 2493 if (mddev->level != LEVEL_NONE)
2494 request_module("md-level-%d", mddev->level);
2495 else if (mddev->clevel[0])
2496 request_module("md-%s", mddev->clevel);
1da177e4
LT		 2497#endif
2498
2499 /*
2500 * Drop all container device buffers, from now on
2501 * the only valid external interface is through the md
2502 * device.
2503 * Also find largest hardsector size
2504 */
2505 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2506 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2507 continue;
2508 sync_blockdev(rdev->bdev);
2509 invalidate_bdev(rdev->bdev, 0);
2510 }
2511
2512 md_probe(mddev->unit, NULL, NULL);
2513 disk = mddev->gendisk;
2514 if (!disk)
2515 return -ENOMEM;
2516
2517 spin_lock(&pers_lock);
d9d166c2 2518 pers = find_pers(mddev->level, mddev->clevel);
2604b703 2519 if (!pers || !try_module_get(pers->owner)) {
1da177e4 2520 spin_unlock(&pers_lock);
d9d166c2
N		 2521 if (mddev->level != LEVEL_NONE)
2522 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2523 mddev->level);
2524 else
2525 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2526 mddev->clevel);
1da177e4
LT		 2527 return -EINVAL;
2528 }
2604b703 2529 mddev->pers = pers;
1da177e4 2530 spin_unlock(&pers_lock);
d9d166c2
N		 2531 mddev->level = pers->level;
2532 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 2533
657390d2 2534 mddev->recovery = 0;
1da177e4 2535 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 2536 mddev->barriers_work = 1;
6ff8d8ec 2537 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 2538
f91de92e
N		 2539 if (start_readonly)
2540 mddev->ro = 2; /* read-only, but switch on first write */
2541
b15c2e57
N		 2542 err = mddev->pers->run(mddev);
2543 if (!err && mddev->pers->sync_request) {
2544 err = bitmap_create(mddev);
2545 if (err) {
2546 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2547 mdname(mddev), err);
2548 mddev->pers->stop(mddev);
2549 }
2550 }
1da177e4
LT		 2551 if (err) {
2552 printk(KERN_ERR "md: pers->run() failed ...\n");
2553 module_put(mddev->pers->owner);
2554 mddev->pers = NULL;
32a7627c
N		 2555 bitmap_destroy(mddev);
2556 return err;
1da177e4 2557 }
411036fa
N		 2558 if (mddev->pers->sync_request)
2559 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
fd9d49ca
N		 2560 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2561 mddev->ro = 0;
2562
1da177e4
LT		 2563 atomic_set(&mddev->writes_pending,0);
2564 mddev->safemode = 0;
2565 mddev->safemode_timer.function = md_safemode_timeout;
2566 mddev->safemode_timer.data = (unsigned long) mddev;
2567 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2568 mddev->in_sync = 1;
86e6ffdd
N		 2569
2570 ITERATE_RDEV(mddev,rdev,tmp)
2571 if (rdev->raid_disk >= 0) {
2572 char nm[20];
2573 sprintf(nm, "rd%d", rdev->raid_disk);
2574 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2575 }
1da177e4
LT		 2576
2577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2578 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2579
2580 if (mddev->sb_dirty)
2581 md_update_sb(mddev);
2582
2583 set_capacity(disk, mddev->array_size<<1);
2584
2585 /* If we call blk_queue_make_request here, it will
2586 * re-initialise max_sectors etc which may have been
2587 * refined inside -> run. So just set the bits we need to set.
2588 * Most initialisation happended when we called
2589 * blk_queue_make_request(..., md_fail_request)
2590 * earlier.
2591 */
2592 mddev->queue->queuedata = mddev;
2593 mddev->queue->make_request_fn = mddev->pers->make_request;
2594
2595 mddev->changed = 1;
d7603b7e 2596 md_new_event(mddev);
1da177e4
LT		 2597 return 0;
2598}
2599
2600static int restart_array(mddev_t *mddev)
2601{
2602 struct gendisk *disk = mddev->gendisk;
2603 int err;
2604
2605 /*
2606 * Complain if it has no devices
2607 */
2608 err = -ENXIO;
2609 if (list_empty(&mddev->disks))
2610 goto out;
2611
2612 if (mddev->pers) {
2613 err = -EBUSY;
2614 if (!mddev->ro)
2615 goto out;
2616
2617 mddev->safemode = 0;
2618 mddev->ro = 0;
2619 set_disk_ro(disk, 0);
2620
2621 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2622 mdname(mddev));
2623 /*
2624 * Kick recovery or resync if necessary
2625 */
2626 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2627 md_wakeup_thread(mddev->thread);
2628 err = 0;
2629 } else {
2630 printk(KERN_ERR "md: %s has no personality assigned.\n",
2631 mdname(mddev));
2632 err = -EINVAL;
2633 }
2634
2635out:
2636 return err;
2637}
2638
2639static int do_md_stop(mddev_t * mddev, int ro)
2640{
2641 int err = 0;
2642 struct gendisk *disk = mddev->gendisk;
2643
2644 if (mddev->pers) {
2645 if (atomic_read(&mddev->active)>2) {
2646 printk("md: %s still in use.\n",mdname(mddev));
2647 return -EBUSY;
2648 }
2649
2650 if (mddev->sync_thread) {
2651 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2652 md_unregister_thread(mddev->sync_thread);
2653 mddev->sync_thread = NULL;
2654 }
2655
2656 del_timer_sync(&mddev->safemode_timer);
2657
2658 invalidate_partition(disk, 0);
2659
2660 if (ro) {
2661 err = -ENXIO;
f91de92e 2662 if (mddev->ro==1)
1da177e4
LT		 2663 goto out;
2664 mddev->ro = 1;
2665 } else {
6b8b3e8a 2666 bitmap_flush(mddev);
a9701a30 2667 md_super_wait(mddev);
1da177e4
LT		 2668 if (mddev->ro)
2669 set_disk_ro(disk, 0);
2670 blk_queue_make_request(mddev->queue, md_fail_request);
2671 mddev->pers->stop(mddev);
411036fa
N		 2672 if (mddev->pers->sync_request)
2673 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2674
1da177e4
LT		 2675 module_put(mddev->pers->owner);
2676 mddev->pers = NULL;
2677 if (mddev->ro)
2678 mddev->ro = 0;
2679 }
2680 if (!mddev->in_sync) {
2681 /* mark array as shutdown cleanly */
2682 mddev->in_sync = 1;
2683 md_update_sb(mddev);
2684 }
2685 if (ro)
2686 set_disk_ro(disk, 1);
2687 }
32a7627c
N		 2688
2689 bitmap_destroy(mddev);
2690 if (mddev->bitmap_file) {
2691 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2692 fput(mddev->bitmap_file);
2693 mddev->bitmap_file = NULL;
2694 }
9223214e 2695 mddev->bitmap_offset = 0;
32a7627c 2696
1da177e4
LT		 2697 /*
2698 * Free resources if final stop
2699 */
2700 if (!ro) {
86e6ffdd
N		 2701 mdk_rdev_t *rdev;
2702 struct list_head *tmp;
1da177e4
LT		 2703 struct gendisk *disk;
2704 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2705
86e6ffdd
N		 2706 ITERATE_RDEV(mddev,rdev,tmp)
2707 if (rdev->raid_disk >= 0) {
2708 char nm[20];
2709 sprintf(nm, "rd%d", rdev->raid_disk);
2710 sysfs_remove_link(&mddev->kobj, nm);
2711 }
2712
1da177e4
LT		 2713 export_array(mddev);
2714
2715 mddev->array_size = 0;
2716 disk = mddev->gendisk;
2717 if (disk)
2718 set_capacity(disk, 0);
2719 mddev->changed = 1;
2720 } else
2721 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2722 mdname(mddev));
2723 err = 0;
d7603b7e 2724 md_new_event(mddev);
1da177e4
LT		 2725out:
2726 return err;
2727}
2728
2729static void autorun_array(mddev_t *mddev)
2730{
2731 mdk_rdev_t *rdev;
2732 struct list_head *tmp;
2733 int err;
2734
a757e64c 2735 if (list_empty(&mddev->disks))
1da177e4 2736 return;
1da177e4
LT		 2737
2738 printk(KERN_INFO "md: running: ");
2739
2740 ITERATE_RDEV(mddev,rdev,tmp) {
2741 char b[BDEVNAME_SIZE];
2742 printk("<%s>", bdevname(rdev->bdev,b));
2743 }
2744 printk("\n");
2745
2746 err = do_md_run (mddev);
2747 if (err) {
2748 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2749 do_md_stop (mddev, 0);
2750 }
2751}
2752
2753/*
2754 * lets try to run arrays based on all disks that have arrived
2755 * until now. (those are in pending_raid_disks)
2756 *
2757 * the method: pick the first pending disk, collect all disks with
2758 * the same UUID, remove all from the pending list and put them into
2759 * the 'same_array' list. Then order this list based on superblock
2760 * update time (freshest comes first), kick out 'old' disks and
2761 * compare superblocks. If everything's fine then run it.
2762 *
2763 * If "unit" is allocated, then bump its reference count
2764 */
2765static void autorun_devices(int part)
2766{
2767 struct list_head candidates;
2768 struct list_head *tmp;
2769 mdk_rdev_t *rdev0, *rdev;
2770 mddev_t *mddev;
2771 char b[BDEVNAME_SIZE];
2772
2773 printk(KERN_INFO "md: autorun ...\n");
2774 while (!list_empty(&pending_raid_disks)) {
2775 dev_t dev;
2776 rdev0 = list_entry(pending_raid_disks.next,
2777 mdk_rdev_t, same_set);
2778
2779 printk(KERN_INFO "md: considering %s ...\n",
2780 bdevname(rdev0->bdev,b));
2781 INIT_LIST_HEAD(&candidates);
2782 ITERATE_RDEV_PENDING(rdev,tmp)
2783 if (super_90_load(rdev, rdev0, 0) >= 0) {
2784 printk(KERN_INFO "md: adding %s ...\n",
2785 bdevname(rdev->bdev,b));
2786 list_move(&rdev->same_set, &candidates);
2787 }
2788 /*
2789 * now we have a set of devices, with all of them having
2790 * mostly sane superblocks. It's time to allocate the
2791 * mddev.
2792 */
2793 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2794 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2795 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2796 break;
2797 }
2798 if (part)
2799 dev = MKDEV(mdp_major,
2800 rdev0->preferred_minor << MdpMinorShift);
2801 else
2802 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2803
2804 md_probe(dev, NULL, NULL);
2805 mddev = mddev_find(dev);
2806 if (!mddev) {
2807 printk(KERN_ERR
2808 "md: cannot allocate memory for md drive.\n");
2809 break;
2810 }
2811 if (mddev_lock(mddev))
2812 printk(KERN_WARNING "md: %s locked, cannot run\n",
2813 mdname(mddev));
2814 else if (mddev->raid_disks || mddev->major_version
2815 || !list_empty(&mddev->disks)) {
2816 printk(KERN_WARNING
2817 "md: %s already running, cannot run %s\n",
2818 mdname(mddev), bdevname(rdev0->bdev,b));
2819 mddev_unlock(mddev);
2820 } else {
2821 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2822 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2823 list_del_init(&rdev->same_set);
2824 if (bind_rdev_to_array(rdev, mddev))
2825 export_rdev(rdev);
2826 }
2827 autorun_array(mddev);
2828 mddev_unlock(mddev);
2829 }
2830 /* on success, candidates will be empty, on error
2831 * it won't...
2832 */
2833 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2834 export_rdev(rdev);
2835 mddev_put(mddev);
2836 }
2837 printk(KERN_INFO "md: ... autorun DONE.\n");
2838}
2839
2840/*
2841 * import RAID devices based on one partition
2842 * if possible, the array gets run as well.
2843 */
2844
2845static int autostart_array(dev_t startdev)
2846{
2847 char b[BDEVNAME_SIZE];
2848 int err = -EINVAL, i;
2849 mdp_super_t *sb = NULL;
2850 mdk_rdev_t *start_rdev = NULL, *rdev;
2851
2852 start_rdev = md_import_device(startdev, 0, 0);
2853 if (IS_ERR(start_rdev))
2854 return err;
2855
2856
2857 /* NOTE: this can only work for 0.90.0 superblocks */
2858 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2859 if (sb->major_version != 0 ||
2860 sb->minor_version != 90 ) {
2861 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2862 export_rdev(start_rdev);
2863 return err;
2864 }
2865
b2d444d7 2866 if (test_bit(Faulty, &start_rdev->flags)) {
1da177e4
LT		 2867 printk(KERN_WARNING
2868 "md: can not autostart based on faulty %s!\n",
2869 bdevname(start_rdev->bdev,b));
2870 export_rdev(start_rdev);
2871 return err;
2872 }
2873 list_add(&start_rdev->same_set, &pending_raid_disks);
2874
2875 for (i = 0; i < MD_SB_DISKS; i++) {
2876 mdp_disk_t *desc = sb->disks + i;
2877 dev_t dev = MKDEV(desc->major, desc->minor);
2878
2879 if (!dev)
2880 continue;
2881 if (dev == startdev)
2882 continue;
2883 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2884 continue;
2885 rdev = md_import_device(dev, 0, 0);
2886 if (IS_ERR(rdev))
2887 continue;
2888
2889 list_add(&rdev->same_set, &pending_raid_disks);
2890 }
2891
2892 /*
2893 * possibly return codes
2894 */
2895 autorun_devices(0);
2896 return 0;
2897
2898}
2899
2900
2901static int get_version(void __user * arg)
2902{
2903 mdu_version_t ver;
2904
2905 ver.major = MD_MAJOR_VERSION;
2906 ver.minor = MD_MINOR_VERSION;
2907 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2908
2909 if (copy_to_user(arg, &ver, sizeof(ver)))
2910 return -EFAULT;
2911
2912 return 0;
2913}
2914
2915static int get_array_info(mddev_t * mddev, void __user * arg)
2916{
2917 mdu_array_info_t info;
2918 int nr,working,active,failed,spare;
2919 mdk_rdev_t *rdev;
2920 struct list_head *tmp;
2921
2922 nr=working=active=failed=spare=0;
2923 ITERATE_RDEV(mddev,rdev,tmp) {
2924 nr++;
b2d444d7 2925 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2926 failed++;
2927 else {
2928 working++;
b2d444d7 2929 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 2930 active++;
2931 else
2932 spare++;
2933 }
2934 }
2935
2936 info.major_version = mddev->major_version;
2937 info.minor_version = mddev->minor_version;
2938 info.patch_version = MD_PATCHLEVEL_VERSION;
2939 info.ctime = mddev->ctime;
2940 info.level = mddev->level;
2941 info.size = mddev->size;
2942 info.nr_disks = nr;
2943 info.raid_disks = mddev->raid_disks;
2944 info.md_minor = mddev->md_minor;
2945 info.not_persistent= !mddev->persistent;
2946
2947 info.utime = mddev->utime;
2948 info.state = 0;
2949 if (mddev->in_sync)
2950 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 2951 if (mddev->bitmap && mddev->bitmap_offset)
2952 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 2953 info.active_disks = active;
2954 info.working_disks = working;
2955 info.failed_disks = failed;
2956 info.spare_disks = spare;
2957
2958 info.layout = mddev->layout;
2959 info.chunk_size = mddev->chunk_size;
2960
2961 if (copy_to_user(arg, &info, sizeof(info)))
2962 return -EFAULT;
2963
2964 return 0;
2965}
2966
87162a28 2967static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 2968{
2969 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2970 char *ptr, *buf = NULL;
2971 int err = -ENOMEM;
2972
2973 file = kmalloc(sizeof(*file), GFP_KERNEL);
2974 if (!file)
2975 goto out;
2976
2977 /* bitmap disabled, zero the first byte and copy out */
2978 if (!mddev->bitmap || !mddev->bitmap->file) {
2979 file->pathname[0] = '\0';
2980 goto copy_out;
2981 }
2982
2983 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2984 if (!buf)
2985 goto out;
2986
2987 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2988 if (!ptr)
2989 goto out;
2990
2991 strcpy(file->pathname, ptr);
2992
2993copy_out:
2994 err = 0;
2995 if (copy_to_user(arg, file, sizeof(*file)))
2996 err = -EFAULT;
2997out:
2998 kfree(buf);
2999 kfree(file);
3000 return err;
3001}
3002
1da177e4
LT		 3003static int get_disk_info(mddev_t * mddev, void __user * arg)
3004{
3005 mdu_disk_info_t info;
3006 unsigned int nr;
3007 mdk_rdev_t *rdev;
3008
3009 if (copy_from_user(&info, arg, sizeof(info)))
3010 return -EFAULT;
3011
3012 nr = info.number;
3013
3014 rdev = find_rdev_nr(mddev, nr);
3015 if (rdev) {
3016 info.major = MAJOR(rdev->bdev->bd_dev);
3017 info.minor = MINOR(rdev->bdev->bd_dev);
3018 info.raid_disk = rdev->raid_disk;
3019 info.state = 0;
b2d444d7 3020 if (test_bit(Faulty, &rdev->flags))
1da177e4 3021 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3022 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3023 info.state |= (1<<MD_DISK_ACTIVE);
3024 info.state |= (1<<MD_DISK_SYNC);
3025 }
8ddf9efe
N		 3026 if (test_bit(WriteMostly, &rdev->flags))
3027 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3028 } else {
3029 info.major = info.minor = 0;
3030 info.raid_disk = -1;
3031 info.state = (1<<MD_DISK_REMOVED);
3032 }
3033
3034 if (copy_to_user(arg, &info, sizeof(info)))
3035 return -EFAULT;
3036
3037 return 0;
3038}
3039
3040static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3041{
3042 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3043 mdk_rdev_t *rdev;
3044 dev_t dev = MKDEV(info->major,info->minor);
3045
3046 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3047 return -EOVERFLOW;
3048
3049 if (!mddev->raid_disks) {
3050 int err;
3051 /* expecting a device which has a superblock */
3052 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3053 if (IS_ERR(rdev)) {
3054 printk(KERN_WARNING
3055 "md: md_import_device returned %ld\n",
3056 PTR_ERR(rdev));
3057 return PTR_ERR(rdev);
3058 }
3059 if (!list_empty(&mddev->disks)) {
3060 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3061 mdk_rdev_t, same_set);
3062 int err = super_types[mddev->major_version]
3063 .load_super(rdev, rdev0, mddev->minor_version);
3064 if (err < 0) {
3065 printk(KERN_WARNING
3066 "md: %s has different UUID to %s\n",
3067 bdevname(rdev->bdev,b),
3068 bdevname(rdev0->bdev,b2));
3069 export_rdev(rdev);
3070 return -EINVAL;
3071 }
3072 }
3073 err = bind_rdev_to_array(rdev, mddev);
3074 if (err)
3075 export_rdev(rdev);
3076 return err;
3077 }
3078
3079 /*
3080 * add_new_disk can be used once the array is assembled
3081 * to add "hot spares". They must already have a superblock
3082 * written
3083 */
3084 if (mddev->pers) {
3085 int err;
3086 if (!mddev->pers->hot_add_disk) {
3087 printk(KERN_WARNING
3088 "%s: personality does not support diskops!\n",
3089 mdname(mddev));
3090 return -EINVAL;
3091 }
7b1e35f6
N		 3092 if (mddev->persistent)
3093 rdev = md_import_device(dev, mddev->major_version,
3094 mddev->minor_version);
3095 else
3096 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 3097 if (IS_ERR(rdev)) {
3098 printk(KERN_WARNING
3099 "md: md_import_device returned %ld\n",
3100 PTR_ERR(rdev));
3101 return PTR_ERR(rdev);
3102 }
41158c7e
N		 3103 /* set save_raid_disk if appropriate */
3104 if (!mddev->persistent) {
3105 if (info->state & (1<<MD_DISK_SYNC) &&
3106 info->raid_disk < mddev->raid_disks)
3107 rdev->raid_disk = info->raid_disk;
3108 else
3109 rdev->raid_disk = -1;
3110 } else
3111 super_types[mddev->major_version].
3112 validate_super(mddev, rdev);
3113 rdev->saved_raid_disk = rdev->raid_disk;
3114
b2d444d7 3115 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 3116 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3117 set_bit(WriteMostly, &rdev->flags);
3118
1da177e4
LT		 3119 rdev->raid_disk = -1;
3120 err = bind_rdev_to_array(rdev, mddev);
3121 if (err)
3122 export_rdev(rdev);
c361777f
N		 3123
3124 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 3125 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3126 return err;
3127 }
3128
3129 /* otherwise, add_new_disk is only allowed
3130 * for major_version==0 superblocks
3131 */
3132 if (mddev->major_version != 0) {
3133 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3134 mdname(mddev));
3135 return -EINVAL;
3136 }
3137
3138 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3139 int err;
3140 rdev = md_import_device (dev, -1, 0);
3141 if (IS_ERR(rdev)) {
3142 printk(KERN_WARNING
3143 "md: error, md_import_device() returned %ld\n",
3144 PTR_ERR(rdev));
3145 return PTR_ERR(rdev);
3146 }
3147 rdev->desc_nr = info->number;
3148 if (info->raid_disk < mddev->raid_disks)
3149 rdev->raid_disk = info->raid_disk;
3150 else
3151 rdev->raid_disk = -1;
3152
b2d444d7
N		 3153 rdev->flags = 0;
3154
1da177e4 3155 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 3156 if (info->state & (1<<MD_DISK_SYNC))
3157 set_bit(In_sync, &rdev->flags);
1da177e4 3158
8ddf9efe
N		 3159 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3160 set_bit(WriteMostly, &rdev->flags);
3161
1da177e4
LT		 3162 if (!mddev->persistent) {
3163 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3164 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3165 } else
3166 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3167 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3168
2bf071bf
N		 3169 err = bind_rdev_to_array(rdev, mddev);
3170 if (err) {
3171 export_rdev(rdev);
3172 return err;
3173 }
1da177e4
LT		 3174 }
3175
3176 return 0;
3177}
3178
3179static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3180{
3181 char b[BDEVNAME_SIZE];
3182 mdk_rdev_t *rdev;
3183
3184 if (!mddev->pers)
3185 return -ENODEV;
3186
3187 rdev = find_rdev(mddev, dev);
3188 if (!rdev)
3189 return -ENXIO;
3190
3191 if (rdev->raid_disk >= 0)
3192 goto busy;
3193
3194 kick_rdev_from_array(rdev);
3195 md_update_sb(mddev);
d7603b7e 3196 md_new_event(mddev);
1da177e4
LT		 3197
3198 return 0;
3199busy:
3200 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3201 bdevname(rdev->bdev,b), mdname(mddev));
3202 return -EBUSY;
3203}
3204
3205static int hot_add_disk(mddev_t * mddev, dev_t dev)
3206{
3207 char b[BDEVNAME_SIZE];
3208 int err;
3209 unsigned int size;
3210 mdk_rdev_t *rdev;
3211
3212 if (!mddev->pers)
3213 return -ENODEV;
3214
3215 if (mddev->major_version != 0) {
3216 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3217 " version-0 superblocks.\n",
3218 mdname(mddev));
3219 return -EINVAL;
3220 }
3221 if (!mddev->pers->hot_add_disk) {
3222 printk(KERN_WARNING
3223 "%s: personality does not support diskops!\n",
3224 mdname(mddev));
3225 return -EINVAL;
3226 }
3227
3228 rdev = md_import_device (dev, -1, 0);
3229 if (IS_ERR(rdev)) {
3230 printk(KERN_WARNING
3231 "md: error, md_import_device() returned %ld\n",
3232 PTR_ERR(rdev));
3233 return -EINVAL;
3234 }
3235
3236 if (mddev->persistent)
3237 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3238 else
3239 rdev->sb_offset =
3240 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3241
3242 size = calc_dev_size(rdev, mddev->chunk_size);
3243 rdev->size = size;
3244
b2d444d7 3245 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3246 printk(KERN_WARNING
3247 "md: can not hot-add faulty %s disk to %s!\n",
3248 bdevname(rdev->bdev,b), mdname(mddev));
3249 err = -EINVAL;
3250 goto abort_export;
3251 }
b2d444d7 3252 clear_bit(In_sync, &rdev->flags);
1da177e4 3253 rdev->desc_nr = -1;
2bf071bf
N		 3254 err = bind_rdev_to_array(rdev, mddev);
3255 if (err)
3256 goto abort_export;
1da177e4
LT		 3257
3258 /*
3259 * The rest should better be atomic, we can have disk failures
3260 * noticed in interrupt contexts ...
3261 */
3262
3263 if (rdev->desc_nr == mddev->max_disks) {
3264 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3265 mdname(mddev));
3266 err = -EBUSY;
3267 goto abort_unbind_export;
3268 }
3269
3270 rdev->raid_disk = -1;
3271
3272 md_update_sb(mddev);
3273
3274 /*
3275 * Kick recovery, maybe this spare has to be added to the
3276 * array immediately.
3277 */
3278 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3279 md_wakeup_thread(mddev->thread);
d7603b7e 3280 md_new_event(mddev);
1da177e4
LT		 3281 return 0;
3282
3283abort_unbind_export:
3284 unbind_rdev_from_array(rdev);
3285
3286abort_export:
3287 export_rdev(rdev);
3288 return err;
3289}
3290
32a7627c
N		 3291/* similar to deny_write_access, but accounts for our holding a reference
3292 * to the file ourselves */
3293static int deny_bitmap_write_access(struct file * file)
3294{
3295 struct inode *inode = file->f_mapping->host;
3296
3297 spin_lock(&inode->i_lock);
3298 if (atomic_read(&inode->i_writecount) > 1) {
3299 spin_unlock(&inode->i_lock);
3300 return -ETXTBSY;
3301 }
3302 atomic_set(&inode->i_writecount, -1);
3303 spin_unlock(&inode->i_lock);
3304
3305 return 0;
3306}
3307
3308static int set_bitmap_file(mddev_t *mddev, int fd)
3309{
3310 int err;
3311
36fa3063
N		 3312 if (mddev->pers) {
3313 if (!mddev->pers->quiesce)
3314 return -EBUSY;
3315 if (mddev->recovery || mddev->sync_thread)
3316 return -EBUSY;
3317 /* we should be able to change the bitmap.. */
3318 }
32a7627c 3319
32a7627c 3320
36fa3063
N		 3321 if (fd >= 0) {
3322 if (mddev->bitmap)
3323 return -EEXIST; /* cannot add when bitmap is present */
3324 mddev->bitmap_file = fget(fd);
32a7627c 3325
36fa3063
N		 3326 if (mddev->bitmap_file == NULL) {
3327 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3328 mdname(mddev));
3329 return -EBADF;
3330 }
3331
3332 err = deny_bitmap_write_access(mddev->bitmap_file);
3333 if (err) {
3334 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3335 mdname(mddev));
3336 fput(mddev->bitmap_file);
3337 mddev->bitmap_file = NULL;
3338 return err;
3339 }
a654b9d8 3340 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 3341 } else if (mddev->bitmap == NULL)
3342 return -ENOENT; /* cannot remove what isn't there */
3343 err = 0;
3344 if (mddev->pers) {
3345 mddev->pers->quiesce(mddev, 1);
3346 if (fd >= 0)
3347 err = bitmap_create(mddev);
3348 if (fd < 0 || err)
3349 bitmap_destroy(mddev);
3350 mddev->pers->quiesce(mddev, 0);
3351 } else if (fd < 0) {
3352 if (mddev->bitmap_file)
3353 fput(mddev->bitmap_file);
3354 mddev->bitmap_file = NULL;
3355 }
3356
32a7627c
N		 3357 return err;
3358}
3359
1da177e4
LT		 3360/*
3361 * set_array_info is used two different ways
3362 * The original usage is when creating a new array.
3363 * In this usage, raid_disks is > 0 and it together with
3364 * level, size, not_persistent,layout,chunksize determine the
3365 * shape of the array.
3366 * This will always create an array with a type-0.90.0 superblock.
3367 * The newer usage is when assembling an array.
3368 * In this case raid_disks will be 0, and the major_version field is
3369 * use to determine which style super-blocks are to be found on the devices.
3370 * The minor and patch _version numbers are also kept incase the
3371 * super_block handler wishes to interpret them.
3372 */
3373static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3374{
3375
3376 if (info->raid_disks == 0) {
3377 /* just setting version number for superblock loading */
3378 if (info->major_version < 0 ||
3379 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3380 super_types[info->major_version].name == NULL) {
3381 /* maybe try to auto-load a module? */
3382 printk(KERN_INFO
3383 "md: superblock version %d not known\n",
3384 info->major_version);
3385 return -EINVAL;
3386 }
3387 mddev->major_version = info->major_version;
3388 mddev->minor_version = info->minor_version;
3389 mddev->patch_version = info->patch_version;
3390 return 0;
3391 }
3392 mddev->major_version = MD_MAJOR_VERSION;
3393 mddev->minor_version = MD_MINOR_VERSION;
3394 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3395 mddev->ctime = get_seconds();
3396
3397 mddev->level = info->level;
17115e03 3398 mddev->clevel[0] = 0;
1da177e4
LT		 3399 mddev->size = info->size;
3400 mddev->raid_disks = info->raid_disks;
3401 /* don't set md_minor, it is determined by which /dev/md* was
3402 * openned
3403 */
3404 if (info->state & (1<<MD_SB_CLEAN))
3405 mddev->recovery_cp = MaxSector;
3406 else
3407 mddev->recovery_cp = 0;
3408 mddev->persistent = ! info->not_persistent;
3409
3410 mddev->layout = info->layout;
3411 mddev->chunk_size = info->chunk_size;
3412
3413 mddev->max_disks = MD_SB_DISKS;
3414
3415 mddev->sb_dirty = 1;
3416
b2a2703c
N		 3417 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3418 mddev->bitmap_offset = 0;
3419
1da177e4
LT		 3420 /*
3421 * Generate a 128 bit UUID
3422 */
3423 get_random_bytes(mddev->uuid, 16);
3424
3425 return 0;
3426}
3427
a35b0d69
N		 3428static int update_size(mddev_t *mddev, unsigned long size)
3429{
3430 mdk_rdev_t * rdev;
3431 int rv;
3432 struct list_head *tmp;
3433
3434 if (mddev->pers->resize == NULL)
3435 return -EINVAL;
3436 /* The "size" is the amount of each device that is used.
3437 * This can only make sense for arrays with redundancy.
3438 * linear and raid0 always use whatever space is available
3439 * We can only consider changing the size if no resync
3440 * or reconstruction is happening, and if the new size
3441 * is acceptable. It must fit before the sb_offset or,
3442 * if that is <data_offset, it must fit before the
3443 * size of each device.
3444 * If size is zero, we find the largest size that fits.
3445 */
3446 if (mddev->sync_thread)
3447 return -EBUSY;
3448 ITERATE_RDEV(mddev,rdev,tmp) {
3449 sector_t avail;
3450 int fit = (size == 0);
3451 if (rdev->sb_offset > rdev->data_offset)
3452 avail = (rdev->sb_offset*2) - rdev->data_offset;
3453 else
3454 avail = get_capacity(rdev->bdev->bd_disk)
3455 - rdev->data_offset;
3456 if (fit && (size == 0 || size > avail/2))
3457 size = avail/2;
3458 if (avail < ((sector_t)size << 1))
3459 return -ENOSPC;
3460 }
3461 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3462 if (!rv) {
3463 struct block_device *bdev;
3464
3465 bdev = bdget_disk(mddev->gendisk, 0);
3466 if (bdev) {
1b1dcc1b 3467 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 3468 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 3469 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 3470 bdput(bdev);
3471 }
3472 }
3473 return rv;
3474}
3475
da943b99
N		 3476static int update_raid_disks(mddev_t *mddev, int raid_disks)
3477{
3478 int rv;
3479 /* change the number of raid disks */
3480 if (mddev->pers->reshape == NULL)
3481 return -EINVAL;
3482 if (raid_disks <= 0 ||
3483 raid_disks >= mddev->max_disks)
3484 return -EINVAL;
3485 if (mddev->sync_thread)
3486 return -EBUSY;
3487 rv = mddev->pers->reshape(mddev, raid_disks);
da943b99
N		 3488 return rv;
3489}
3490
3491
1da177e4
LT		 3492/*
3493 * update_array_info is used to change the configuration of an
3494 * on-line array.
3495 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3496 * fields in the info are checked against the array.
3497 * Any differences that cannot be handled will cause an error.
3498 * Normally, only one change can be managed at a time.
3499 */
3500static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3501{
3502 int rv = 0;
3503 int cnt = 0;
36fa3063
N		 3504 int state = 0;
3505
3506 /* calculate expected state,ignoring low bits */
3507 if (mddev->bitmap && mddev->bitmap_offset)
3508 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3509
3510 if (mddev->major_version != info->major_version ||
3511 mddev->minor_version != info->minor_version ||
3512/* mddev->patch_version != info->patch_version || */
3513 mddev->ctime != info->ctime ||
3514 mddev->level != info->level ||
3515/* mddev->layout != info->layout || */
3516 !mddev->persistent != info->not_persistent||
36fa3063
N		 3517 mddev->chunk_size != info->chunk_size ||
3518 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3519 ((state^info->state) & 0xfffffe00)
3520 )
1da177e4
LT		 3521 return -EINVAL;
3522 /* Check there is only one change */
3523 if (mddev->size != info->size) cnt++;
3524 if (mddev->raid_disks != info->raid_disks) cnt++;
3525 if (mddev->layout != info->layout) cnt++;
36fa3063 3526 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 3527 if (cnt == 0) return 0;
3528 if (cnt > 1) return -EINVAL;
3529
3530 if (mddev->layout != info->layout) {
3531 /* Change layout
3532 * we don't need to do anything at the md level, the
3533 * personality will take care of it all.
3534 */
3535 if (mddev->pers->reconfig == NULL)
3536 return -EINVAL;
3537 else
3538 return mddev->pers->reconfig(mddev, info->layout, -1);
3539 }
a35b0d69
N		 3540 if (mddev->size != info->size)
3541 rv = update_size(mddev, info->size);
3542
da943b99
N		 3543 if (mddev->raid_disks != info->raid_disks)
3544 rv = update_raid_disks(mddev, info->raid_disks);
3545
36fa3063
N		 3546 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3547 if (mddev->pers->quiesce == NULL)
3548 return -EINVAL;
3549 if (mddev->recovery || mddev->sync_thread)
3550 return -EBUSY;
3551 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3552 /* add the bitmap */
3553 if (mddev->bitmap)
3554 return -EEXIST;
3555 if (mddev->default_bitmap_offset == 0)
3556 return -EINVAL;
3557 mddev->bitmap_offset = mddev->default_bitmap_offset;
3558 mddev->pers->quiesce(mddev, 1);
3559 rv = bitmap_create(mddev);
3560 if (rv)
3561 bitmap_destroy(mddev);
3562 mddev->pers->quiesce(mddev, 0);
3563 } else {
3564 /* remove the bitmap */
3565 if (!mddev->bitmap)
3566 return -ENOENT;
3567 if (mddev->bitmap->file)
3568 return -EINVAL;
3569 mddev->pers->quiesce(mddev, 1);
3570 bitmap_destroy(mddev);
3571 mddev->pers->quiesce(mddev, 0);
3572 mddev->bitmap_offset = 0;
3573 }
3574 }
1da177e4
LT		 3575 md_update_sb(mddev);
3576 return rv;
3577}
3578
3579static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3580{
3581 mdk_rdev_t *rdev;
3582
3583 if (mddev->pers == NULL)
3584 return -ENODEV;
3585
3586 rdev = find_rdev(mddev, dev);
3587 if (!rdev)
3588 return -ENODEV;
3589
3590 md_error(mddev, rdev);
3591 return 0;
3592}
3593
a885c8c4
CH		 3594static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3595{
3596 mddev_t *mddev = bdev->bd_disk->private_data;
3597
3598 geo->heads = 2;
3599 geo->sectors = 4;
3600 geo->cylinders = get_capacity(mddev->gendisk) / 8;
3601 return 0;
3602}
3603
1da177e4
LT		 3604static int md_ioctl(struct inode *inode, struct file *file,
3605 unsigned int cmd, unsigned long arg)
3606{
3607 int err = 0;
3608 void __user *argp = (void __user *)arg;
1da177e4
LT		 3609 mddev_t *mddev = NULL;
3610
3611 if (!capable(CAP_SYS_ADMIN))
3612 return -EACCES;
3613
3614 /*
3615 * Commands dealing with the RAID driver but not any
3616 * particular array:
3617 */
3618 switch (cmd)
3619 {
3620 case RAID_VERSION:
3621 err = get_version(argp);
3622 goto done;
3623
3624 case PRINT_RAID_DEBUG:
3625 err = 0;
3626 md_print_devices();
3627 goto done;
3628
3629#ifndef MODULE
3630 case RAID_AUTORUN:
3631 err = 0;
3632 autostart_arrays(arg);
3633 goto done;
3634#endif
3635 default:;
3636 }
3637
3638 /*
3639 * Commands creating/starting a new array:
3640 */
3641
3642 mddev = inode->i_bdev->bd_disk->private_data;
3643
3644 if (!mddev) {
3645 BUG();
3646 goto abort;
3647 }
3648
3649
3650 if (cmd == START_ARRAY) {
3651 /* START_ARRAY doesn't need to lock the array as autostart_array
3652 * does the locking, and it could even be a different array
3653 */
3654 static int cnt = 3;
3655 if (cnt > 0 ) {
3656 printk(KERN_WARNING
3657 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
e8a00334 3658 "This will not be supported beyond July 2006\n",
1da177e4
LT		 3659 current->comm, current->pid);
3660 cnt--;
3661 }
3662 err = autostart_array(new_decode_dev(arg));
3663 if (err) {
3664 printk(KERN_WARNING "md: autostart failed!\n");
3665 goto abort;
3666 }
3667 goto done;
3668 }
3669
3670 err = mddev_lock(mddev);
3671 if (err) {
3672 printk(KERN_INFO
3673 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3674 err, cmd);
3675 goto abort;
3676 }
3677
3678 switch (cmd)
3679 {
3680 case SET_ARRAY_INFO:
3681 {
3682 mdu_array_info_t info;
3683 if (!arg)
3684 memset(&info, 0, sizeof(info));
3685 else if (copy_from_user(&info, argp, sizeof(info))) {
3686 err = -EFAULT;
3687 goto abort_unlock;
3688 }
3689 if (mddev->pers) {
3690 err = update_array_info(mddev, &info);
3691 if (err) {
3692 printk(KERN_WARNING "md: couldn't update"
3693 " array info. %d\n", err);
3694 goto abort_unlock;
3695 }
3696 goto done_unlock;
3697 }
3698 if (!list_empty(&mddev->disks)) {
3699 printk(KERN_WARNING
3700 "md: array %s already has disks!\n",
3701 mdname(mddev));
3702 err = -EBUSY;
3703 goto abort_unlock;
3704 }
3705 if (mddev->raid_disks) {
3706 printk(KERN_WARNING
3707 "md: array %s already initialised!\n",
3708 mdname(mddev));
3709 err = -EBUSY;
3710 goto abort_unlock;
3711 }
3712 err = set_array_info(mddev, &info);
3713 if (err) {
3714 printk(KERN_WARNING "md: couldn't set"
3715 " array info. %d\n", err);
3716 goto abort_unlock;
3717 }
3718 }
3719 goto done_unlock;
3720
3721 default:;
3722 }
3723
3724 /*
3725 * Commands querying/configuring an existing array:
3726 */
32a7627c
N		 3727 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3728 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3729 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3730 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 3731 err = -ENODEV;
3732 goto abort_unlock;
3733 }
3734
3735 /*
3736 * Commands even a read-only array can execute:
3737 */
3738 switch (cmd)
3739 {
3740 case GET_ARRAY_INFO:
3741 err = get_array_info(mddev, argp);
3742 goto done_unlock;
3743
32a7627c 3744 case GET_BITMAP_FILE:
87162a28 3745 err = get_bitmap_file(mddev, argp);
32a7627c
N		 3746 goto done_unlock;
3747
1da177e4
LT		 3748 case GET_DISK_INFO:
3749 err = get_disk_info(mddev, argp);
3750 goto done_unlock;
3751
3752 case RESTART_ARRAY_RW:
3753 err = restart_array(mddev);
3754 goto done_unlock;
3755
3756 case STOP_ARRAY:
3757 err = do_md_stop (mddev, 0);
3758 goto done_unlock;
3759
3760 case STOP_ARRAY_RO:
3761 err = do_md_stop (mddev, 1);
3762 goto done_unlock;
3763
3764 /*
3765 * We have a problem here : there is no easy way to give a CHS
3766 * virtual geometry. We currently pretend that we have a 2 heads
3767 * 4 sectors (with a BIG number of cylinders...). This drives
3768 * dosfs just mad... ;-)
3769 */
1da177e4
LT		 3770 }
3771
3772 /*
3773 * The remaining ioctls are changing the state of the
f91de92e
N		 3774 * superblock, so we do not allow them on read-only arrays.
3775 * However non-MD ioctls (e.g. get-size) will still come through
3776 * here and hit the 'default' below, so only disallow
3777 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 3778 */
f91de92e
N		 3779 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3780 mddev->ro && mddev->pers) {
3781 if (mddev->ro == 2) {
3782 mddev->ro = 0;
3783 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3784 md_wakeup_thread(mddev->thread);
3785
3786 } else {
3787 err = -EROFS;
3788 goto abort_unlock;
3789 }
1da177e4
LT		 3790 }
3791
3792 switch (cmd)
3793 {
3794 case ADD_NEW_DISK:
3795 {
3796 mdu_disk_info_t info;
3797 if (copy_from_user(&info, argp, sizeof(info)))
3798 err = -EFAULT;
3799 else
3800 err = add_new_disk(mddev, &info);
3801 goto done_unlock;
3802 }
3803
3804 case HOT_REMOVE_DISK:
3805 err = hot_remove_disk(mddev, new_decode_dev(arg));
3806 goto done_unlock;
3807
3808 case HOT_ADD_DISK:
3809 err = hot_add_disk(mddev, new_decode_dev(arg));
3810 goto done_unlock;
3811
3812 case SET_DISK_FAULTY:
3813 err = set_disk_faulty(mddev, new_decode_dev(arg));
3814 goto done_unlock;
3815
3816 case RUN_ARRAY:
3817 err = do_md_run (mddev);
3818 goto done_unlock;
3819
32a7627c
N		 3820 case SET_BITMAP_FILE:
3821 err = set_bitmap_file(mddev, (int)arg);
3822 goto done_unlock;
3823
1da177e4
LT		 3824 default:
3825 if (_IOC_TYPE(cmd) == MD_MAJOR)
3826 printk(KERN_WARNING "md: %s(pid %d) used"
3827 " obsolete MD ioctl, upgrade your"
3828 " software to use new ictls.\n",
3829 current->comm, current->pid);
3830 err = -EINVAL;
3831 goto abort_unlock;
3832 }
3833
3834done_unlock:
3835abort_unlock:
3836 mddev_unlock(mddev);
3837
3838 return err;
3839done:
3840 if (err)
3841 MD_BUG();
3842abort:
3843 return err;
3844}
3845
3846static int md_open(struct inode *inode, struct file *file)
3847{
3848 /*
3849 * Succeed if we can lock the mddev, which confirms that
3850 * it isn't being stopped right now.
3851 */
3852 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3853 int err;
3854
3855 if ((err = mddev_lock(mddev)))
3856 goto out;
3857
3858 err = 0;
3859 mddev_get(mddev);
3860 mddev_unlock(mddev);
3861
3862 check_disk_change(inode->i_bdev);
3863 out:
3864 return err;
3865}
3866
3867static int md_release(struct inode *inode, struct file * file)
3868{
3869 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3870
3871 if (!mddev)
3872 BUG();
3873 mddev_put(mddev);
3874
3875 return 0;
3876}
3877
3878static int md_media_changed(struct gendisk *disk)
3879{
3880 mddev_t *mddev = disk->private_data;
3881
3882 return mddev->changed;
3883}
3884
3885static int md_revalidate(struct gendisk *disk)
3886{
3887 mddev_t *mddev = disk->private_data;
3888
3889 mddev->changed = 0;
3890 return 0;
3891}
3892static struct block_device_operations md_fops =
3893{
3894 .owner = THIS_MODULE,
3895 .open = md_open,
3896 .release = md_release,
3897 .ioctl = md_ioctl,
a885c8c4 3898 .getgeo = md_getgeo,
1da177e4
LT		 3899 .media_changed = md_media_changed,
3900 .revalidate_disk= md_revalidate,
3901};
3902
75c96f85 3903static int md_thread(void * arg)
1da177e4
LT		 3904{
3905 mdk_thread_t *thread = arg;
3906
1da177e4
LT		 3907 /*
3908 * md_thread is a 'system-thread', it's priority should be very
3909 * high. We avoid resource deadlocks individually in each
3910 * raid personality. (RAID5 does preallocation) We also use RR and
3911 * the very same RT priority as kswapd, thus we will never get
3912 * into a priority inversion deadlock.
3913 *
3914 * we definitely have to have equal or higher priority than
3915 * bdflush, otherwise bdflush will deadlock if there are too
3916 * many dirty RAID5 blocks.
3917 */
1da177e4 3918
6985c43f 3919 allow_signal(SIGKILL);
a6fb0934 3920 while (!kthread_should_stop()) {
1da177e4 3921
93588e22
N		 3922 /* We need to wait INTERRUPTIBLE so that
3923 * we don't add to the load-average.
3924 * That means we need to be sure no signals are
3925 * pending
3926 */
3927 if (signal_pending(current))
3928 flush_signals(current);
3929
3930 wait_event_interruptible_timeout
3931 (thread->wqueue,
3932 test_bit(THREAD_WAKEUP, &thread->flags)
3933 || kthread_should_stop(),
3934 thread->timeout);
3e1d1d28 3935 try_to_freeze();
1da177e4
LT		 3936
3937 clear_bit(THREAD_WAKEUP, &thread->flags);
3938
787453c2 3939 thread->run(thread->mddev);
1da177e4 3940 }
a6fb0934 3941
1da177e4
LT		 3942 return 0;
3943}
3944
3945void md_wakeup_thread(mdk_thread_t *thread)
3946{
3947 if (thread) {
3948 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3949 set_bit(THREAD_WAKEUP, &thread->flags);
3950 wake_up(&thread->wqueue);
3951 }
3952}
3953
3954mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3955 const char *name)
3956{
3957 mdk_thread_t *thread;
1da177e4 3958
9ffae0cf 3959 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 3960 if (!thread)
3961 return NULL;
3962
1da177e4
LT		 3963 init_waitqueue_head(&thread->wqueue);
3964
1da177e4
LT		 3965 thread->run = run;
3966 thread->mddev = mddev;
32a7627c 3967 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 3968 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 3969 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 3970 kfree(thread);
3971 return NULL;
3972 }
1da177e4
LT		 3973 return thread;
3974}
3975
1da177e4
LT		 3976void md_unregister_thread(mdk_thread_t *thread)
3977{
d28446fe 3978 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 3979
3980 kthread_stop(thread->tsk);
1da177e4
LT		 3981 kfree(thread);
3982}
3983
3984void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3985{
3986 if (!mddev) {
3987 MD_BUG();
3988 return;
3989 }
3990
b2d444d7 3991 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 3992 return;
32a7627c 3993/*
1da177e4
LT		 3994 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3995 mdname(mddev),
3996 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3997 __builtin_return_address(0),__builtin_return_address(1),
3998 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 3999*/
1da177e4
LT		 4000 if (!mddev->pers->error_handler)
4001 return;
4002 mddev->pers->error_handler(mddev,rdev);
4003 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4004 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4005 md_wakeup_thread(mddev->thread);
d7603b7e 4006 md_new_event(mddev);
1da177e4
LT		 4007}
4008
4009/* seq_file implementation /proc/mdstat */
4010
4011static void status_unused(struct seq_file *seq)
4012{
4013 int i = 0;
4014 mdk_rdev_t *rdev;
4015 struct list_head *tmp;
4016
4017 seq_printf(seq, "unused devices: ");
4018
4019 ITERATE_RDEV_PENDING(rdev,tmp) {
4020 char b[BDEVNAME_SIZE];
4021 i++;
4022 seq_printf(seq, "%s ",
4023 bdevname(rdev->bdev,b));
4024 }
4025 if (!i)
4026 seq_printf(seq, "<none>");
4027
4028 seq_printf(seq, "\n");
4029}
4030
4031
4032static void status_resync(struct seq_file *seq, mddev_t * mddev)
4033{
4034 unsigned long max_blocks, resync, res, dt, db, rt;
4035
4036 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4037
4038 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4039 max_blocks = mddev->resync_max_sectors >> 1;
4040 else
4041 max_blocks = mddev->size;
4042
4043 /*
4044 * Should not happen.
4045 */
4046 if (!max_blocks) {
4047 MD_BUG();
4048 return;
4049 }
4050 res = (resync/1024)*1000/(max_blocks/1024 + 1);
4051 {
4052 int i, x = res/50, y = 20-x;
4053 seq_printf(seq, "[");
4054 for (i = 0; i < x; i++)
4055 seq_printf(seq, "=");
4056 seq_printf(seq, ">");
4057 for (i = 0; i < y; i++)
4058 seq_printf(seq, ".");
4059 seq_printf(seq, "] ");
4060 }
4061 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
4062 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4063 "resync" : "recovery"),
4064 res/10, res % 10, resync, max_blocks);
4065
4066 /*
4067 * We do not want to overflow, so the order of operands and
4068 * the * 100 / 100 trick are important. We do a +1 to be
4069 * safe against division by zero. We only estimate anyway.
4070 *
4071 * dt: time from mark until now
4072 * db: blocks written from mark until now
4073 * rt: remaining time
4074 */
4075 dt = ((jiffies - mddev->resync_mark) / HZ);
4076 if (!dt) dt++;
4077 db = resync - (mddev->resync_mark_cnt/2);
4078 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
4079
4080 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4081
4082 seq_printf(seq, " speed=%ldK/sec", db/dt);
4083}
4084
4085static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4086{
4087 struct list_head *tmp;
4088 loff_t l = *pos;
4089 mddev_t *mddev;
4090
4091 if (l >= 0x10000)
4092 return NULL;
4093 if (!l--)
4094 /* header */
4095 return (void*)1;
4096
4097 spin_lock(&all_mddevs_lock);
4098 list_for_each(tmp,&all_mddevs)
4099 if (!l--) {
4100 mddev = list_entry(tmp, mddev_t, all_mddevs);
4101 mddev_get(mddev);
4102 spin_unlock(&all_mddevs_lock);
4103 return mddev;
4104 }
4105 spin_unlock(&all_mddevs_lock);
4106 if (!l--)
4107 return (void*)2;/* tail */
4108 return NULL;
4109}
4110
4111static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4112{
4113 struct list_head *tmp;
4114 mddev_t *next_mddev, *mddev = v;
4115
4116 ++*pos;
4117 if (v == (void*)2)
4118 return NULL;
4119
4120 spin_lock(&all_mddevs_lock);
4121 if (v == (void*)1)
4122 tmp = all_mddevs.next;
4123 else
4124 tmp = mddev->all_mddevs.next;
4125 if (tmp != &all_mddevs)
4126 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4127 else {
4128 next_mddev = (void*)2;
4129 *pos = 0x10000;
4130 }
4131 spin_unlock(&all_mddevs_lock);
4132
4133 if (v != (void*)1)
4134 mddev_put(mddev);
4135 return next_mddev;
4136
4137}
4138
4139static void md_seq_stop(struct seq_file *seq, void *v)
4140{
4141 mddev_t *mddev = v;
4142
4143 if (mddev && v != (void*)1 && v != (void*)2)
4144 mddev_put(mddev);
4145}
4146
d7603b7e
N		 4147struct mdstat_info {
4148 int event;
4149};
4150
1da177e4
LT		 4151static int md_seq_show(struct seq_file *seq, void *v)
4152{
4153 mddev_t *mddev = v;
4154 sector_t size;
4155 struct list_head *tmp2;
4156 mdk_rdev_t *rdev;
d7603b7e 4157 struct mdstat_info *mi = seq->private;
32a7627c 4158 struct bitmap *bitmap;
1da177e4
LT		 4159
4160 if (v == (void*)1) {
2604b703 4161 struct mdk_personality *pers;
1da177e4
LT		 4162 seq_printf(seq, "Personalities : ");
4163 spin_lock(&pers_lock);
2604b703
N		 4164 list_for_each_entry(pers, &pers_list, list)
4165 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 4166
4167 spin_unlock(&pers_lock);
4168 seq_printf(seq, "\n");
d7603b7e 4169 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 4170 return 0;
4171 }
4172 if (v == (void*)2) {
4173 status_unused(seq);
4174 return 0;
4175 }
4176
4177 if (mddev_lock(mddev)!=0)
4178 return -EINTR;
4179 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4180 seq_printf(seq, "%s : %sactive", mdname(mddev),
4181 mddev->pers ? "" : "in");
4182 if (mddev->pers) {
f91de92e 4183 if (mddev->ro==1)
1da177e4 4184 seq_printf(seq, " (read-only)");
f91de92e
N		 4185 if (mddev->ro==2)
4186 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 4187 seq_printf(seq, " %s", mddev->pers->name);
4188 }
4189
4190 size = 0;
4191 ITERATE_RDEV(mddev,rdev,tmp2) {
4192 char b[BDEVNAME_SIZE];
4193 seq_printf(seq, " %s[%d]",
4194 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 4195 if (test_bit(WriteMostly, &rdev->flags))
4196 seq_printf(seq, "(W)");
b2d444d7 4197 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4198 seq_printf(seq, "(F)");
4199 continue;
b325a32e
N		 4200 } else if (rdev->raid_disk < 0)
4201 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 4202 size += rdev->size;
4203 }
4204
4205 if (!list_empty(&mddev->disks)) {
4206 if (mddev->pers)
4207 seq_printf(seq, "\n %llu blocks",
4208 (unsigned long long)mddev->array_size);
4209 else
4210 seq_printf(seq, "\n %llu blocks",
4211 (unsigned long long)size);
4212 }
1cd6bf19
N		 4213 if (mddev->persistent) {
4214 if (mddev->major_version != 0 ||
4215 mddev->minor_version != 90) {
4216 seq_printf(seq," super %d.%d",
4217 mddev->major_version,
4218 mddev->minor_version);
4219 }
4220 } else
4221 seq_printf(seq, " super non-persistent");
1da177e4
LT		 4222
4223 if (mddev->pers) {
4224 mddev->pers->status (seq, mddev);
4225 seq_printf(seq, "\n ");
8e1b39d6
N		 4226 if (mddev->pers->sync_request) {
4227 if (mddev->curr_resync > 2) {
4228 status_resync (seq, mddev);
4229 seq_printf(seq, "\n ");
4230 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4231 seq_printf(seq, "\tresync=DELAYED\n ");
4232 else if (mddev->recovery_cp < MaxSector)
4233 seq_printf(seq, "\tresync=PENDING\n ");
4234 }
32a7627c
N		 4235 } else
4236 seq_printf(seq, "\n ");
4237
4238 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 4239 unsigned long chunk_kb;
4240 unsigned long flags;
32a7627c
N		 4241 spin_lock_irqsave(&bitmap->lock, flags);
4242 chunk_kb = bitmap->chunksize >> 10;
4243 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4244 "%lu%s chunk",
4245 bitmap->pages - bitmap->missing_pages,
4246 bitmap->pages,
4247 (bitmap->pages - bitmap->missing_pages)
4248 << (PAGE_SHIFT - 10),
4249 chunk_kb ? chunk_kb : bitmap->chunksize,
4250 chunk_kb ? "KB" : "B");
78d742d8
N		 4251 if (bitmap->file) {
4252 seq_printf(seq, ", file: ");
4253 seq_path(seq, bitmap->file->f_vfsmnt,
4254 bitmap->file->f_dentry," \t\n");
32a7627c 4255 }
78d742d8 4256
32a7627c
N		 4257 seq_printf(seq, "\n");
4258 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 4259 }
4260
4261 seq_printf(seq, "\n");
4262 }
4263 mddev_unlock(mddev);
4264
4265 return 0;
4266}
4267
4268static struct seq_operations md_seq_ops = {
4269 .start = md_seq_start,
4270 .next = md_seq_next,
4271 .stop = md_seq_stop,
4272 .show = md_seq_show,
4273};
4274
4275static int md_seq_open(struct inode *inode, struct file *file)
4276{
4277 int error;
d7603b7e
N		 4278 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4279 if (mi == NULL)
4280 return -ENOMEM;
1da177e4
LT		 4281
4282 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 4283 if (error)
4284 kfree(mi);
4285 else {
4286 struct seq_file *p = file->private_data;
4287 p->private = mi;
4288 mi->event = atomic_read(&md_event_count);
4289 }
1da177e4
LT		 4290 return error;
4291}
4292
d7603b7e
N		 4293static int md_seq_release(struct inode *inode, struct file *file)
4294{
4295 struct seq_file *m = file->private_data;
4296 struct mdstat_info *mi = m->private;
4297 m->private = NULL;
4298 kfree(mi);
4299 return seq_release(inode, file);
4300}
4301
4302static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4303{
4304 struct seq_file *m = filp->private_data;
4305 struct mdstat_info *mi = m->private;
4306 int mask;
4307
4308 poll_wait(filp, &md_event_waiters, wait);
4309
4310 /* always allow read */
4311 mask = POLLIN | POLLRDNORM;
4312
4313 if (mi->event != atomic_read(&md_event_count))
4314 mask |= POLLERR | POLLPRI;
4315 return mask;
4316}
4317
1da177e4
LT		 4318static struct file_operations md_seq_fops = {
4319 .open = md_seq_open,
4320 .read = seq_read,
4321 .llseek = seq_lseek,
d7603b7e
N		 4322 .release = md_seq_release,
4323 .poll = mdstat_poll,
1da177e4
LT		 4324};
4325
2604b703 4326int register_md_personality(struct mdk_personality *p)
1da177e4 4327{
1da177e4 4328 spin_lock(&pers_lock);
2604b703
N		 4329 list_add_tail(&p->list, &pers_list);
4330 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 4331 spin_unlock(&pers_lock);
4332 return 0;
4333}
4334
2604b703 4335int unregister_md_personality(struct mdk_personality *p)
1da177e4 4336{
2604b703 4337 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 4338 spin_lock(&pers_lock);
2604b703 4339 list_del_init(&p->list);
1da177e4
LT		 4340 spin_unlock(&pers_lock);
4341 return 0;
4342}
4343
4344static int is_mddev_idle(mddev_t *mddev)
4345{
4346 mdk_rdev_t * rdev;
4347 struct list_head *tmp;
4348 int idle;
4349 unsigned long curr_events;
4350
4351 idle = 1;
4352 ITERATE_RDEV(mddev,rdev,tmp) {
4353 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 4354 curr_events = disk_stat_read(disk, sectors[0]) +
4355 disk_stat_read(disk, sectors[1]) -
1da177e4 4356 atomic_read(&disk->sync_io);
c0e48521
N		 4357 /* The difference between curr_events and last_events
4358 * will be affected by any new non-sync IO (making
4359 * curr_events bigger) and any difference in the amount of
4360 * in-flight syncio (making current_events bigger or smaller)
4361 * The amount in-flight is currently limited to
4362 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4363 * which is at most 4096 sectors.
4364 * These numbers are fairly fragile and should be made
4365 * more robust, probably by enforcing the
4366 * 'window size' that md_do_sync sort-of uses.
4367 *
1da177e4
LT		 4368 * Note: the following is an unsigned comparison.
4369 */
c0e48521 4370 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 4371 rdev->last_events = curr_events;
4372 idle = 0;
4373 }
4374 }
4375 return idle;
4376}
4377
4378void md_done_sync(mddev_t *mddev, int blocks, int ok)
4379{
4380 /* another "blocks" (512byte) blocks have been synced */
4381 atomic_sub(blocks, &mddev->recovery_active);
4382 wake_up(&mddev->recovery_wait);
4383 if (!ok) {
4384 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4385 md_wakeup_thread(mddev->thread);
4386 // stop recovery, signal do_sync ....
4387 }
4388}
4389
4390
06d91a5f
N		 4391/* md_write_start(mddev, bi)
4392 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 4393 * in superblock) before writing, schedule a superblock update
4394 * and wait for it to complete.
06d91a5f 4395 */
3d310eb7 4396void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 4397{
06d91a5f 4398 if (bio_data_dir(bi) != WRITE)
3d310eb7 4399 return;
06d91a5f 4400
f91de92e
N		 4401 BUG_ON(mddev->ro == 1);
4402 if (mddev->ro == 2) {
4403 /* need to switch to read/write */
4404 mddev->ro = 0;
4405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4406 md_wakeup_thread(mddev->thread);
4407 }
06d91a5f 4408 atomic_inc(&mddev->writes_pending);
06d91a5f 4409 if (mddev->in_sync) {
a9701a30 4410 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 4411 if (mddev->in_sync) {
4412 mddev->in_sync = 0;
4413 mddev->sb_dirty = 1;
4414 md_wakeup_thread(mddev->thread);
4415 }
a9701a30 4416 spin_unlock_irq(&mddev->write_lock);
06d91a5f 4417 }
3d310eb7 4418 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
1da177e4
LT		 4419}
4420
4421void md_write_end(mddev_t *mddev)
4422{
4423 if (atomic_dec_and_test(&mddev->writes_pending)) {
4424 if (mddev->safemode == 2)
4425 md_wakeup_thread(mddev->thread);
4426 else
4427 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4428 }
4429}
4430
75c96f85 4431static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 4432
4433#define SYNC_MARKS 10
4434#define SYNC_MARK_STEP (3*HZ)
4435static void md_do_sync(mddev_t *mddev)
4436{
4437 mddev_t *mddev2;
4438 unsigned int currspeed = 0,
4439 window;
57afd89f 4440 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 4441 unsigned long mark[SYNC_MARKS];
4442 sector_t mark_cnt[SYNC_MARKS];
4443 int last_mark,m;
4444 struct list_head *tmp;
4445 sector_t last_check;
57afd89f 4446 int skipped = 0;
1da177e4
LT		 4447
4448 /* just incase thread restarts... */
4449 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4450 return;
4451
4452 /* we overload curr_resync somewhat here.
4453 * 0 == not engaged in resync at all
4454 * 2 == checking that there is no conflict with another sync
4455 * 1 == like 2, but have yielded to allow conflicting resync to
4456 * commense
4457 * other == active in resync - this many blocks
4458 *
4459 * Before starting a resync we must have set curr_resync to
4460 * 2, and then checked that every "conflicting" array has curr_resync
4461 * less than ours. When we find one that is the same or higher
4462 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4463 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4464 * This will mean we have to start checking from the beginning again.
4465 *
4466 */
4467
4468 do {
4469 mddev->curr_resync = 2;
4470
4471 try_again:
787453c2 4472 if (kthread_should_stop()) {
6985c43f 4473 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 4474 goto skip;
4475 }
4476 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 4477 if (mddev2 == mddev)
4478 continue;
4479 if (mddev2->curr_resync &&
4480 match_mddev_units(mddev,mddev2)) {
4481 DEFINE_WAIT(wq);
4482 if (mddev < mddev2 && mddev->curr_resync == 2) {
4483 /* arbitrarily yield */
4484 mddev->curr_resync = 1;
4485 wake_up(&resync_wait);
4486 }
4487 if (mddev > mddev2 && mddev->curr_resync == 1)
4488 /* no need to wait here, we can wait the next
4489 * time 'round when curr_resync == 2
4490 */
4491 continue;
787453c2
N		 4492 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4493 if (!kthread_should_stop() &&
8712e553 4494 mddev2->curr_resync >= mddev->curr_resync) {
1da177e4
LT		 4495 printk(KERN_INFO "md: delaying resync of %s"
4496 " until %s has finished resync (they"
4497 " share one or more physical units)\n",
4498 mdname(mddev), mdname(mddev2));
4499 mddev_put(mddev2);
4500 schedule();
4501 finish_wait(&resync_wait, &wq);
4502 goto try_again;
4503 }
4504 finish_wait(&resync_wait, &wq);
4505 }
4506 }
4507 } while (mddev->curr_resync < 2);
4508
9d88883e 4509 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 4510 /* resync follows the size requested by the personality,
57afd89f 4511 * which defaults to physical size, but can be virtual size
1da177e4
LT		 4512 */
4513 max_sectors = mddev->resync_max_sectors;
9d88883e
N		 4514 mddev->resync_mismatches = 0;
4515 } else
1da177e4
LT		 4516 /* recovery follows the physical size of devices */
4517 max_sectors = mddev->size << 1;
4518
4519 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4520 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
88202a0c 4521 " %d KB/sec/disc.\n", speed_min(mddev));
338cec32 4522 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
1da177e4 4523 "(but not more than %d KB/sec) for reconstruction.\n",
88202a0c 4524 speed_max(mddev));
1da177e4
LT		 4525
4526 is_mddev_idle(mddev); /* this also initializes IO event counters */
32a7627c 4527 /* we don't use the checkpoint if there's a bitmap */
24dd469d
N		 4528 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4529 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1da177e4
LT		 4530 j = mddev->recovery_cp;
4531 else
4532 j = 0;
57afd89f 4533 io_sectors = 0;
1da177e4
LT		 4534 for (m = 0; m < SYNC_MARKS; m++) {
4535 mark[m] = jiffies;
57afd89f 4536 mark_cnt[m] = io_sectors;
1da177e4
LT		 4537 }
4538 last_mark = 0;
4539 mddev->resync_mark = mark[last_mark];
4540 mddev->resync_mark_cnt = mark_cnt[last_mark];
4541
4542 /*
4543 * Tune reconstruction:
4544 */
4545 window = 32*(PAGE_SIZE/512);
4546 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4547 window/2,(unsigned long long) max_sectors/2);
4548
4549 atomic_set(&mddev->recovery_active, 0);
4550 init_waitqueue_head(&mddev->recovery_wait);
4551 last_check = 0;
4552
4553 if (j>2) {
4554 printk(KERN_INFO
4555 "md: resuming recovery of %s from checkpoint.\n",
4556 mdname(mddev));
4557 mddev->curr_resync = j;
4558 }
4559
4560 while (j < max_sectors) {
57afd89f 4561 sector_t sectors;
1da177e4 4562
57afd89f
N		 4563 skipped = 0;
4564 sectors = mddev->pers->sync_request(mddev, j, &skipped,
88202a0c 4565 currspeed < speed_min(mddev));
57afd89f 4566 if (sectors == 0) {
1da177e4
LT		 4567 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4568 goto out;
4569 }
57afd89f
N		 4570
4571 if (!skipped) { /* actual IO requested */
4572 io_sectors += sectors;
4573 atomic_add(sectors, &mddev->recovery_active);
4574 }
4575
1da177e4
LT		 4576 j += sectors;
4577 if (j>1) mddev->curr_resync = j;
d7603b7e
N		 4578 if (last_check == 0)
4579 /* this is the earliers that rebuilt will be
4580 * visible in /proc/mdstat
4581 */
4582 md_new_event(mddev);
57afd89f
N		 4583
4584 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 4585 continue;
4586
57afd89f 4587 last_check = io_sectors;
1da177e4
LT		 4588
4589 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4590 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4591 break;
4592
4593 repeat:
4594 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4595 /* step marks */
4596 int next = (last_mark+1) % SYNC_MARKS;
4597
4598 mddev->resync_mark = mark[next];
4599 mddev->resync_mark_cnt = mark_cnt[next];
4600 mark[next] = jiffies;
57afd89f 4601 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 4602 last_mark = next;
4603 }
4604
4605
787453c2 4606 if (kthread_should_stop()) {
1da177e4
LT		 4607 /*
4608 * got a signal, exit.
4609 */
4610 printk(KERN_INFO
4611 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 4612 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4613 goto out;
4614 }
4615
4616 /*
4617 * this loop exits only if either when we are slower than
4618 * the 'hard' speed limit, or the system was IO-idle for
4619 * a jiffy.
4620 * the system might be non-idle CPU-wise, but we only care
4621 * about not overloading the IO subsystem. (things like an
4622 * e2fsck being done on the RAID array should execute fast)
4623 */
4624 mddev->queue->unplug_fn(mddev->queue);
4625 cond_resched();
4626
57afd89f
N		 4627 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4628 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 4629
88202a0c
N		 4630 if (currspeed > speed_min(mddev)) {
4631 if ((currspeed > speed_max(mddev)) ||
1da177e4 4632 !is_mddev_idle(mddev)) {
c0e48521 4633 msleep(500);
1da177e4
LT		 4634 goto repeat;
4635 }
4636 }
4637 }
4638 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4639 /*
4640 * this also signals 'finished resyncing' to md_stop
4641 */
4642 out:
4643 mddev->queue->unplug_fn(mddev->queue);
4644
4645 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4646
4647 /* tell personality that we are finished */
57afd89f 4648 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 4649
4650 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4651 mddev->curr_resync > 2 &&
4652 mddev->curr_resync >= mddev->recovery_cp) {
4653 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4654 printk(KERN_INFO
4655 "md: checkpointing recovery of %s.\n",
4656 mdname(mddev));
4657 mddev->recovery_cp = mddev->curr_resync;
4658 } else
4659 mddev->recovery_cp = MaxSector;
4660 }
4661
1da177e4
LT		 4662 skip:
4663 mddev->curr_resync = 0;
4664 wake_up(&resync_wait);
4665 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4666 md_wakeup_thread(mddev->thread);
4667}
4668
4669
4670/*
4671 * This routine is regularly called by all per-raid-array threads to
4672 * deal with generic issues like resync and super-block update.
4673 * Raid personalities that don't have a thread (linear/raid0) do not
4674 * need this as they never do any recovery or update the superblock.
4675 *
4676 * It does not do any resync itself, but rather "forks" off other threads
4677 * to do that as needed.
4678 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4679 * "->recovery" and create a thread at ->sync_thread.
4680 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4681 * and wakeups up this thread which will reap the thread and finish up.
4682 * This thread also removes any faulty devices (with nr_pending == 0).
4683 *
4684 * The overall approach is:
4685 * 1/ if the superblock needs updating, update it.
4686 * 2/ If a recovery thread is running, don't do anything else.
4687 * 3/ If recovery has finished, clean up, possibly marking spares active.
4688 * 4/ If there are any faulty devices, remove them.
4689 * 5/ If array is degraded, try to add spares devices
4690 * 6/ If array has spares or is not in-sync, start a resync thread.
4691 */
4692void md_check_recovery(mddev_t *mddev)
4693{
4694 mdk_rdev_t *rdev;
4695 struct list_head *rtmp;
4696
4697
5f40402d
N		 4698 if (mddev->bitmap)
4699 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 4700
4701 if (mddev->ro)
4702 return;
fca4d848
N		 4703
4704 if (signal_pending(current)) {
4705 if (mddev->pers->sync_request) {
4706 printk(KERN_INFO "md: %s in immediate safe mode\n",
4707 mdname(mddev));
4708 mddev->safemode = 2;
4709 }
4710 flush_signals(current);
4711 }
4712
1da177e4
LT		 4713 if ( ! (
4714 mddev->sb_dirty ||
4715 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 4716 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4717 (mddev->safemode == 1) ||
4718 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4719 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 4720 ))
4721 return;
fca4d848 4722
1da177e4
LT		 4723 if (mddev_trylock(mddev)==0) {
4724 int spares =0;
fca4d848 4725
a9701a30 4726 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 4727 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4728 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4729 mddev->in_sync = 1;
4730 mddev->sb_dirty = 1;
4731 }
4732 if (mddev->safemode == 1)
4733 mddev->safemode = 0;
a9701a30 4734 spin_unlock_irq(&mddev->write_lock);
fca4d848 4735
1da177e4
LT		 4736 if (mddev->sb_dirty)
4737 md_update_sb(mddev);
06d91a5f 4738
06d91a5f 4739
1da177e4
LT		 4740 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4741 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4742 /* resync/recovery still happening */
4743 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4744 goto unlock;
4745 }
4746 if (mddev->sync_thread) {
4747 /* resync has finished, collect result */
4748 md_unregister_thread(mddev->sync_thread);
4749 mddev->sync_thread = NULL;
4750 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4751 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4752 /* success...*/
4753 /* activate any spares */
4754 mddev->pers->spare_active(mddev);
4755 }
4756 md_update_sb(mddev);
41158c7e
N		 4757
4758 /* if array is no-longer degraded, then any saved_raid_disk
4759 * information must be scrapped
4760 */
4761 if (!mddev->degraded)
4762 ITERATE_RDEV(mddev,rdev,rtmp)
4763 rdev->saved_raid_disk = -1;
4764
1da177e4
LT		 4765 mddev->recovery = 0;
4766 /* flag recovery needed just to double check */
4767 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 4768 md_new_event(mddev);
1da177e4
LT		 4769 goto unlock;
4770 }
24dd469d
N		 4771 /* Clear some bits that don't mean anything, but
4772 * might be left set
4773 */
4774 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4775 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4776 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4777 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4
LT		 4778
4779 /* no recovery is running.
4780 * remove any failed drives, then
4781 * add spares if possible.
4782 * Spare are also removed and re-added, to allow
4783 * the personality to fail the re-add.
4784 */
4785 ITERATE_RDEV(mddev,rdev,rtmp)
4786 if (rdev->raid_disk >= 0 &&
b2d444d7 4787 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 4788 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 4789 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4790 char nm[20];
4791 sprintf(nm,"rd%d", rdev->raid_disk);
4792 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 4793 rdev->raid_disk = -1;
86e6ffdd 4794 }
1da177e4
LT		 4795 }
4796
4797 if (mddev->degraded) {
4798 ITERATE_RDEV(mddev,rdev,rtmp)
4799 if (rdev->raid_disk < 0
b2d444d7 4800 && !test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 4801 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4802 char nm[20];
4803 sprintf(nm, "rd%d", rdev->raid_disk);
4804 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 4805 spares++;
d7603b7e 4806 md_new_event(mddev);
86e6ffdd 4807 } else
1da177e4
LT		 4808 break;
4809 }
4810 }
4811
24dd469d
N		 4812 if (spares) {
4813 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4814 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4815 } else if (mddev->recovery_cp < MaxSector) {
4816 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4817 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4818 /* nothing to be done ... */
1da177e4 4819 goto unlock;
24dd469d 4820
1da177e4
LT		 4821 if (mddev->pers->sync_request) {
4822 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 4823 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4824 /* We are adding a device or devices to an array
4825 * which has the bitmap stored on all devices.
4826 * So make sure all bitmap pages get written
4827 */
4828 bitmap_write_all(mddev->bitmap);
4829 }
1da177e4
LT		 4830 mddev->sync_thread = md_register_thread(md_do_sync,
4831 mddev,
4832 "%s_resync");
4833 if (!mddev->sync_thread) {
4834 printk(KERN_ERR "%s: could not start resync"
4835 " thread...\n",
4836 mdname(mddev));
4837 /* leave the spares where they are, it shouldn't hurt */
4838 mddev->recovery = 0;
d7603b7e 4839 } else
1da177e4 4840 md_wakeup_thread(mddev->sync_thread);
d7603b7e 4841 md_new_event(mddev);
1da177e4
LT		 4842 }
4843 unlock:
4844 mddev_unlock(mddev);
4845 }
4846}
4847
75c96f85
AB		 4848static int md_notify_reboot(struct notifier_block *this,
4849 unsigned long code, void *x)
1da177e4
LT		 4850{
4851 struct list_head *tmp;
4852 mddev_t *mddev;
4853
4854 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4855
4856 printk(KERN_INFO "md: stopping all md devices.\n");
4857
4858 ITERATE_MDDEV(mddev,tmp)
4859 if (mddev_trylock(mddev)==0)
4860 do_md_stop (mddev, 1);
4861 /*
4862 * certain more exotic SCSI devices are known to be
4863 * volatile wrt too early system reboots. While the
4864 * right place to handle this issue is the given
4865 * driver, we do want to have a safe RAID driver ...
4866 */
4867 mdelay(1000*1);
4868 }
4869 return NOTIFY_DONE;
4870}
4871
75c96f85 4872static struct notifier_block md_notifier = {
1da177e4
LT		 4873 .notifier_call = md_notify_reboot,
4874 .next = NULL,
4875 .priority = INT_MAX, /* before any real devices */
4876};
4877
4878static void md_geninit(void)
4879{
4880 struct proc_dir_entry *p;
4881
4882 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4883
4884 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4885 if (p)
4886 p->proc_fops = &md_seq_fops;
4887}
4888
75c96f85 4889static int __init md_init(void)
1da177e4
LT		 4890{
4891 int minor;
4892
4893 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4894 " MD_SB_DISKS=%d\n",
4895 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4896 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
bd926c63 4897 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
32a7627c 4898 BITMAP_MINOR);
1da177e4
LT		 4899
4900 if (register_blkdev(MAJOR_NR, "md"))
4901 return -1;
4902 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4903 unregister_blkdev(MAJOR_NR, "md");
4904 return -1;
4905 }
4906 devfs_mk_dir("md");
4907 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4908 md_probe, NULL, NULL);
4909 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4910 md_probe, NULL, NULL);
4911
4912 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4913 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4914 S_IFBLK|S_IRUSR|S_IWUSR,
4915 "md/%d", minor);
4916
4917 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4918 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4919 S_IFBLK|S_IRUSR|S_IWUSR,
4920 "md/mdp%d", minor);
4921
4922
4923 register_reboot_notifier(&md_notifier);
4924 raid_table_header = register_sysctl_table(raid_root_table, 1);
4925
4926 md_geninit();
4927 return (0);
4928}
4929
4930
4931#ifndef MODULE
4932
4933/*
4934 * Searches all registered partitions for autorun RAID arrays
4935 * at boot time.
4936 */
4937static dev_t detected_devices[128];
4938static int dev_cnt;
4939
4940void md_autodetect_dev(dev_t dev)
4941{
4942 if (dev_cnt >= 0 && dev_cnt < 127)
4943 detected_devices[dev_cnt++] = dev;
4944}
4945
4946
4947static void autostart_arrays(int part)
4948{
4949 mdk_rdev_t *rdev;
4950 int i;
4951
4952 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4953
4954 for (i = 0; i < dev_cnt; i++) {
4955 dev_t dev = detected_devices[i];
4956
4957 rdev = md_import_device(dev,0, 0);
4958 if (IS_ERR(rdev))
4959 continue;
4960
b2d444d7 4961 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4962 MD_BUG();
4963 continue;
4964 }
4965 list_add(&rdev->same_set, &pending_raid_disks);
4966 }
4967 dev_cnt = 0;
4968
4969 autorun_devices(part);
4970}
4971
4972#endif
4973
4974static __exit void md_exit(void)
4975{
4976 mddev_t *mddev;
4977 struct list_head *tmp;
4978 int i;
4979 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4980 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4981 for (i=0; i < MAX_MD_DEVS; i++)
4982 devfs_remove("md/%d", i);
4983 for (i=0; i < MAX_MD_DEVS; i++)
4984 devfs_remove("md/d%d", i);
4985
4986 devfs_remove("md");
4987
4988 unregister_blkdev(MAJOR_NR,"md");
4989 unregister_blkdev(mdp_major, "mdp");
4990 unregister_reboot_notifier(&md_notifier);
4991 unregister_sysctl_table(raid_table_header);
4992 remove_proc_entry("mdstat", NULL);
4993 ITERATE_MDDEV(mddev,tmp) {
4994 struct gendisk *disk = mddev->gendisk;
4995 if (!disk)
4996 continue;
4997 export_array(mddev);
4998 del_gendisk(disk);
4999 put_disk(disk);
5000 mddev->gendisk = NULL;
5001 mddev_put(mddev);
5002 }
5003}
5004
5005module_init(md_init)
5006module_exit(md_exit)
5007
f91de92e
N		 5008static int get_ro(char *buffer, struct kernel_param *kp)
5009{
5010 return sprintf(buffer, "%d", start_readonly);
5011}
5012static int set_ro(const char *val, struct kernel_param *kp)
5013{
5014 char *e;
5015 int num = simple_strtoul(val, &e, 10);
5016 if (*val && (*e == '\0' || *e == '\n')) {
5017 start_readonly = num;
4dbcdc75 5018 return 0;
f91de92e
N		 5019 }
5020 return -EINVAL;
5021}
5022
5023module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
6ff8d8ec
N		 5024module_param(start_dirty_degraded, int, 0644);
5025
f91de92e 5026
1da177e4
LT		 5027EXPORT_SYMBOL(register_md_personality);
5028EXPORT_SYMBOL(unregister_md_personality);
5029EXPORT_SYMBOL(md_error);
5030EXPORT_SYMBOL(md_done_sync);
5031EXPORT_SYMBOL(md_write_start);
5032EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 5033EXPORT_SYMBOL(md_register_thread);
5034EXPORT_SYMBOL(md_unregister_thread);
5035EXPORT_SYMBOL(md_wakeup_thread);
5036EXPORT_SYMBOL(md_print_devices);
5037EXPORT_SYMBOL(md_check_recovery);
5038MODULE_LICENSE("GPL");
aa1595e9 5039MODULE_ALIAS("md");
72008652 5040MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)