projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] md: define and use safe_put_page for 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.
84 */
85
86static int sysctl_speed_limit_min = 1000;
87static int sysctl_speed_limit_max = 200000;
88
89static struct ctl_table_header *raid_table_header;
90
91static ctl_table raid_table[] = {
92 {
93 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
94 .procname = "speed_limit_min",
95 .data = &sysctl_speed_limit_min,
96 .maxlen = sizeof(int),
97 .mode = 0644,
98 .proc_handler = &proc_dointvec,
99 },
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
102 .procname = "speed_limit_max",
103 .data = &sysctl_speed_limit_max,
104 .maxlen = sizeof(int),
105 .mode = 0644,
106 .proc_handler = &proc_dointvec,
107 },
108 { .ctl_name = 0 }
109};
110
111static ctl_table raid_dir_table[] = {
112 {
113 .ctl_name = DEV_RAID,
114 .procname = "raid",
115 .maxlen = 0,
116 .mode = 0555,
117 .child = raid_table,
118 },
119 { .ctl_name = 0 }
120};
121
122static ctl_table raid_root_table[] = {
123 {
124 .ctl_name = CTL_DEV,
125 .procname = "dev",
126 .maxlen = 0,
127 .mode = 0555,
128 .child = raid_dir_table,
129 },
130 { .ctl_name = 0 }
131};
132
133static struct block_device_operations md_fops;
134
f91de92e
N		 135static int start_readonly;
136
d7603b7e
N		 137/*
138 * We have a system wide 'event count' that is incremented
139 * on any 'interesting' event, and readers of /proc/mdstat
140 * can use 'poll' or 'select' to find out when the event
141 * count increases.
142 *
143 * Events are:
144 * start array, stop array, error, add device, remove device,
145 * start build, activate spare
146 */
147DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
148static atomic_t md_event_count;
149void md_new_event(mddev_t *mddev)
150{
151 atomic_inc(&md_event_count);
152 wake_up(&md_event_waiters);
153}
154
1da177e4
LT		 155/*
156 * Enables to iterate over all existing md arrays
157 * all_mddevs_lock protects this list.
158 */
159static LIST_HEAD(all_mddevs);
160static DEFINE_SPINLOCK(all_mddevs_lock);
161
162
163/*
164 * iterates through all used mddevs in the system.
165 * We take care to grab the all_mddevs_lock whenever navigating
166 * the list, and to always hold a refcount when unlocked.
167 * Any code which breaks out of this loop while own
168 * a reference to the current mddev and must mddev_put it.
169 */
170#define ITERATE_MDDEV(mddev,tmp) \
171 \
172 for (({ spin_lock(&all_mddevs_lock); \
173 tmp = all_mddevs.next; \
174 mddev = NULL;}); \
175 ({ if (tmp != &all_mddevs) \
176 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
177 spin_unlock(&all_mddevs_lock); \
178 if (mddev) mddev_put(mddev); \
179 mddev = list_entry(tmp, mddev_t, all_mddevs); \
180 tmp != &all_mddevs;}); \
181 ({ spin_lock(&all_mddevs_lock); \
182 tmp = tmp->next;}) \
183 )
184
185
186static int md_fail_request (request_queue_t *q, struct bio *bio)
187{
188 bio_io_error(bio, bio->bi_size);
189 return 0;
190}
191
192static inline mddev_t *mddev_get(mddev_t *mddev)
193{
194 atomic_inc(&mddev->active);
195 return mddev;
196}
197
198static void mddev_put(mddev_t *mddev)
199{
200 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
201 return;
202 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
203 list_del(&mddev->all_mddevs);
204 blk_put_queue(mddev->queue);
eae1701f 205 kobject_unregister(&mddev->kobj);
1da177e4
LT		 206 }
207 spin_unlock(&all_mddevs_lock);
208}
209
210static mddev_t * mddev_find(dev_t unit)
211{
212 mddev_t *mddev, *new = NULL;
213
214 retry:
215 spin_lock(&all_mddevs_lock);
216 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
217 if (mddev->unit == unit) {
218 mddev_get(mddev);
219 spin_unlock(&all_mddevs_lock);
990a8baf 220 kfree(new);
1da177e4
LT		 221 return mddev;
222 }
223
224 if (new) {
225 list_add(&new->all_mddevs, &all_mddevs);
226 spin_unlock(&all_mddevs_lock);
227 return new;
228 }
229 spin_unlock(&all_mddevs_lock);
230
9ffae0cf 231 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 232 if (!new)
233 return NULL;
234
1da177e4
LT		 235 new->unit = unit;
236 if (MAJOR(unit) == MD_MAJOR)
237 new->md_minor = MINOR(unit);
238 else
239 new->md_minor = MINOR(unit) >> MdpMinorShift;
240
241 init_MUTEX(&new->reconfig_sem);
242 INIT_LIST_HEAD(&new->disks);
243 INIT_LIST_HEAD(&new->all_mddevs);
244 init_timer(&new->safemode_timer);
245 atomic_set(&new->active, 1);
06d91a5f 246 spin_lock_init(&new->write_lock);
3d310eb7 247 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 248
249 new->queue = blk_alloc_queue(GFP_KERNEL);
250 if (!new->queue) {
251 kfree(new);
252 return NULL;
253 }
254
255 blk_queue_make_request(new->queue, md_fail_request);
256
257 goto retry;
258}
259
260static inline int mddev_lock(mddev_t * mddev)
261{
262 return down_interruptible(&mddev->reconfig_sem);
263}
264
265static inline void mddev_lock_uninterruptible(mddev_t * mddev)
266{
267 down(&mddev->reconfig_sem);
268}
269
270static inline int mddev_trylock(mddev_t * mddev)
271{
272 return down_trylock(&mddev->reconfig_sem);
273}
274
275static inline void mddev_unlock(mddev_t * mddev)
276{
277 up(&mddev->reconfig_sem);
278
005eca5e 279 md_wakeup_thread(mddev->thread);
1da177e4
LT		 280}
281
282mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
283{
284 mdk_rdev_t * rdev;
285 struct list_head *tmp;
286
287 ITERATE_RDEV(mddev,rdev,tmp) {
288 if (rdev->desc_nr == nr)
289 return rdev;
290 }
291 return NULL;
292}
293
294static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
295{
296 struct list_head *tmp;
297 mdk_rdev_t *rdev;
298
299 ITERATE_RDEV(mddev,rdev,tmp) {
300 if (rdev->bdev->bd_dev == dev)
301 return rdev;
302 }
303 return NULL;
304}
305
2604b703
N		 306static struct mdk_personality *find_pers(int level)
307{
308 struct mdk_personality *pers;
309 list_for_each_entry(pers, &pers_list, list)
310 if (pers->level == level)
311 return pers;
312 return NULL;
313}
314
77933d72 315static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 316{
317 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
318 return MD_NEW_SIZE_BLOCKS(size);
319}
320
321static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
322{
323 sector_t size;
324
325 size = rdev->sb_offset;
326
327 if (chunk_size)
328 size &= ~((sector_t)chunk_size/1024 - 1);
329 return size;
330}
331
332static int alloc_disk_sb(mdk_rdev_t * rdev)
333{
334 if (rdev->sb_page)
335 MD_BUG();
336
337 rdev->sb_page = alloc_page(GFP_KERNEL);
338 if (!rdev->sb_page) {
339 printk(KERN_ALERT "md: out of memory.\n");
340 return -EINVAL;
341 }
342
343 return 0;
344}
345
346static void free_disk_sb(mdk_rdev_t * rdev)
347{
348 if (rdev->sb_page) {
2d1f3b5d 349 put_page(rdev->sb_page);
1da177e4
LT		 350 rdev->sb_loaded = 0;
351 rdev->sb_page = NULL;
352 rdev->sb_offset = 0;
353 rdev->size = 0;
354 }
355}
356
357
7bfa19f2
N		 358static int super_written(struct bio *bio, unsigned int bytes_done, int error)
359{
360 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 361 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 362 if (bio->bi_size)
363 return 1;
364
365 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
a9701a30 366 md_error(mddev, rdev);
7bfa19f2 367
a9701a30
N		 368 if (atomic_dec_and_test(&mddev->pending_writes))
369 wake_up(&mddev->sb_wait);
f8b58edf 370 bio_put(bio);
7bfa19f2
N		 371 return 0;
372}
373
a9701a30
N		 374static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
375{
376 struct bio *bio2 = bio->bi_private;
377 mdk_rdev_t *rdev = bio2->bi_private;
378 mddev_t *mddev = rdev->mddev;
379 if (bio->bi_size)
380 return 1;
381
382 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
383 error == -EOPNOTSUPP) {
384 unsigned long flags;
385 /* barriers don't appear to be supported :-( */
386 set_bit(BarriersNotsupp, &rdev->flags);
387 mddev->barriers_work = 0;
388 spin_lock_irqsave(&mddev->write_lock, flags);
389 bio2->bi_next = mddev->biolist;
390 mddev->biolist = bio2;
391 spin_unlock_irqrestore(&mddev->write_lock, flags);
392 wake_up(&mddev->sb_wait);
393 bio_put(bio);
394 return 0;
395 }
396 bio_put(bio2);
397 bio->bi_private = rdev;
398 return super_written(bio, bytes_done, error);
399}
400
7bfa19f2
N		 401void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
402 sector_t sector, int size, struct page *page)
403{
404 /* write first size bytes of page to sector of rdev
405 * Increment mddev->pending_writes before returning
406 * and decrement it on completion, waking up sb_wait
407 * if zero is reached.
408 * If an error occurred, call md_error
a9701a30
N		 409 *
410 * As we might need to resubmit the request if BIO_RW_BARRIER
411 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 412 */
413 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 414 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 415
416 bio->bi_bdev = rdev->bdev;
417 bio->bi_sector = sector;
418 bio_add_page(bio, page, size, 0);
419 bio->bi_private = rdev;
420 bio->bi_end_io = super_written;
a9701a30
N		 421 bio->bi_rw = rw;
422
7bfa19f2 423 atomic_inc(&mddev->pending_writes);
a9701a30
N		 424 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
425 struct bio *rbio;
426 rw |= (1<<BIO_RW_BARRIER);
427 rbio = bio_clone(bio, GFP_NOIO);
428 rbio->bi_private = bio;
429 rbio->bi_end_io = super_written_barrier;
430 submit_bio(rw, rbio);
431 } else
432 submit_bio(rw, bio);
433}
434
435void md_super_wait(mddev_t *mddev)
436{
437 /* wait for all superblock writes that were scheduled to complete.
438 * if any had to be retried (due to BARRIER problems), retry them
439 */
440 DEFINE_WAIT(wq);
441 for(;;) {
442 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
443 if (atomic_read(&mddev->pending_writes)==0)
444 break;
445 while (mddev->biolist) {
446 struct bio *bio;
447 spin_lock_irq(&mddev->write_lock);
448 bio = mddev->biolist;
449 mddev->biolist = bio->bi_next ;
450 bio->bi_next = NULL;
451 spin_unlock_irq(&mddev->write_lock);
452 submit_bio(bio->bi_rw, bio);
453 }
454 schedule();
455 }
456 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 457}
458
1da177e4
LT		 459static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
460{
461 if (bio->bi_size)
462 return 1;
463
464 complete((struct completion*)bio->bi_private);
465 return 0;
466}
467
a654b9d8 468int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 469 struct page *page, int rw)
470{
baaa2c51 471 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 472 struct completion event;
473 int ret;
474
475 rw |= (1 << BIO_RW_SYNC);
476
477 bio->bi_bdev = bdev;
478 bio->bi_sector = sector;
479 bio_add_page(bio, page, size, 0);
480 init_completion(&event);
481 bio->bi_private = &event;
482 bio->bi_end_io = bi_complete;
483 submit_bio(rw, bio);
484 wait_for_completion(&event);
485
486 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
487 bio_put(bio);
488 return ret;
489}
a8745db2 490EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 491
0002b271 492static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 493{
494 char b[BDEVNAME_SIZE];
495 if (!rdev->sb_page) {
496 MD_BUG();
497 return -EINVAL;
498 }
499 if (rdev->sb_loaded)
500 return 0;
501
502
0002b271 503 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 504 goto fail;
505 rdev->sb_loaded = 1;
506 return 0;
507
508fail:
509 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
510 bdevname(rdev->bdev,b));
511 return -EINVAL;
512}
513
514static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
515{
516 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
517 (sb1->set_uuid1 == sb2->set_uuid1) &&
518 (sb1->set_uuid2 == sb2->set_uuid2) &&
519 (sb1->set_uuid3 == sb2->set_uuid3))
520
521 return 1;
522
523 return 0;
524}
525
526
527static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
528{
529 int ret;
530 mdp_super_t *tmp1, *tmp2;
531
532 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
533 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
534
535 if (!tmp1 || !tmp2) {
536 ret = 0;
537 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
538 goto abort;
539 }
540
541 *tmp1 = *sb1;
542 *tmp2 = *sb2;
543
544 /*
545 * nr_disks is not constant
546 */
547 tmp1->nr_disks = 0;
548 tmp2->nr_disks = 0;
549
550 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
551 ret = 0;
552 else
553 ret = 1;
554
555abort:
990a8baf
JJ		 556 kfree(tmp1);
557 kfree(tmp2);
1da177e4
LT		 558 return ret;
559}
560
561static unsigned int calc_sb_csum(mdp_super_t * sb)
562{
563 unsigned int disk_csum, csum;
564
565 disk_csum = sb->sb_csum;
566 sb->sb_csum = 0;
567 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
568 sb->sb_csum = disk_csum;
569 return csum;
570}
571
572
573/*
574 * Handle superblock details.
575 * We want to be able to handle multiple superblock formats
576 * so we have a common interface to them all, and an array of
577 * different handlers.
578 * We rely on user-space to write the initial superblock, and support
579 * reading and updating of superblocks.
580 * Interface methods are:
581 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
582 * loads and validates a superblock on dev.
583 * if refdev != NULL, compare superblocks on both devices
584 * Return:
585 * 0 - dev has a superblock that is compatible with refdev
586 * 1 - dev has a superblock that is compatible and newer than refdev
587 * so dev should be used as the refdev in future
588 * -EINVAL superblock incompatible or invalid
589 * -othererror e.g. -EIO
590 *
591 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
592 * Verify that dev is acceptable into mddev.
593 * The first time, mddev->raid_disks will be 0, and data from
594 * dev should be merged in. Subsequent calls check that dev
595 * is new enough. Return 0 or -EINVAL
596 *
597 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
598 * Update the superblock for rdev with data in mddev
599 * This does not write to disc.
600 *
601 */
602
603struct super_type {
604 char *name;
605 struct module *owner;
606 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
607 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
608 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
609};
610
611/*
612 * load_super for 0.90.0
613 */
614static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
615{
616 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
617 mdp_super_t *sb;
618 int ret;
619 sector_t sb_offset;
620
621 /*
622 * Calculate the position of the superblock,
623 * it's at the end of the disk.
624 *
625 * It also happens to be a multiple of 4Kb.
626 */
627 sb_offset = calc_dev_sboffset(rdev->bdev);
628 rdev->sb_offset = sb_offset;
629
0002b271 630 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 631 if (ret) return ret;
632
633 ret = -EINVAL;
634
635 bdevname(rdev->bdev, b);
636 sb = (mdp_super_t*)page_address(rdev->sb_page);
637
638 if (sb->md_magic != MD_SB_MAGIC) {
639 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
640 b);
641 goto abort;
642 }
643
644 if (sb->major_version != 0 ||
645 sb->minor_version != 90) {
646 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
647 sb->major_version, sb->minor_version,
648 b);
649 goto abort;
650 }
651
652 if (sb->raid_disks <= 0)
653 goto abort;
654
655 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
656 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
657 b);
658 goto abort;
659 }
660
661 rdev->preferred_minor = sb->md_minor;
662 rdev->data_offset = 0;
0002b271 663 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 664
665 if (sb->level == LEVEL_MULTIPATH)
666 rdev->desc_nr = -1;
667 else
668 rdev->desc_nr = sb->this_disk.number;
669
670 if (refdev == 0)
671 ret = 1;
672 else {
673 __u64 ev1, ev2;
674 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
675 if (!uuid_equal(refsb, sb)) {
676 printk(KERN_WARNING "md: %s has different UUID to %s\n",
677 b, bdevname(refdev->bdev,b2));
678 goto abort;
679 }
680 if (!sb_equal(refsb, sb)) {
681 printk(KERN_WARNING "md: %s has same UUID"
682 " but different superblock to %s\n",
683 b, bdevname(refdev->bdev, b2));
684 goto abort;
685 }
686 ev1 = md_event(sb);
687 ev2 = md_event(refsb);
688 if (ev1 > ev2)
689 ret = 1;
690 else
691 ret = 0;
692 }
693 rdev->size = calc_dev_size(rdev, sb->chunk_size);
694
695 abort:
696 return ret;
697}
698
699/*
700 * validate_super for 0.90.0
701 */
702static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
703{
704 mdp_disk_t *desc;
705 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
706
41158c7e 707 rdev->raid_disk = -1;
b2d444d7 708 rdev->flags = 0;
1da177e4
LT		 709 if (mddev->raid_disks == 0) {
710 mddev->major_version = 0;
711 mddev->minor_version = sb->minor_version;
712 mddev->patch_version = sb->patch_version;
713 mddev->persistent = ! sb->not_persistent;
714 mddev->chunk_size = sb->chunk_size;
715 mddev->ctime = sb->ctime;
716 mddev->utime = sb->utime;
717 mddev->level = sb->level;
718 mddev->layout = sb->layout;
719 mddev->raid_disks = sb->raid_disks;
720 mddev->size = sb->size;
721 mddev->events = md_event(sb);
9223214e 722 mddev->bitmap_offset = 0;
36fa3063 723 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4
LT		 724
725 if (sb->state & (1<<MD_SB_CLEAN))
726 mddev->recovery_cp = MaxSector;
727 else {
728 if (sb->events_hi == sb->cp_events_hi &&
729 sb->events_lo == sb->cp_events_lo) {
730 mddev->recovery_cp = sb->recovery_cp;
731 } else
732 mddev->recovery_cp = 0;
733 }
734
735 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
736 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
737 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
738 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
739
740 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 741
742 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
743 mddev->bitmap_file == NULL) {
6cce3b23
N		 744 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
745 && mddev->level != 10) {
a654b9d8 746 /* FIXME use a better test */
6cce3b23 747 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 748 return -EINVAL;
749 }
36fa3063 750 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8
N		 751 }
752
41158c7e
N		 753 } else if (mddev->pers == NULL) {
754 /* Insist on good event counter while assembling */
755 __u64 ev1 = md_event(sb);
1da177e4
LT		 756 ++ev1;
757 if (ev1 < mddev->events)
758 return -EINVAL;
41158c7e
N		 759 } else if (mddev->bitmap) {
760 /* if adding to array with a bitmap, then we can accept an
761 * older device ... but not too old.
762 */
763 __u64 ev1 = md_event(sb);
764 if (ev1 < mddev->bitmap->events_cleared)
765 return 0;
766 } else /* just a hot-add of a new device, leave raid_disk at -1 */
767 return 0;
768
1da177e4 769 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 770 desc = sb->disks + rdev->desc_nr;
771
772 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 773 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 774 else if (desc->state & (1<<MD_DISK_SYNC) &&
775 desc->raid_disk < mddev->raid_disks) {
b2d444d7 776 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 777 rdev->raid_disk = desc->raid_disk;
778 }
8ddf9efe
N		 779 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
780 set_bit(WriteMostly, &rdev->flags);
41158c7e 781 } else /* MULTIPATH are always insync */
b2d444d7 782 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 783 return 0;
784}
785
786/*
787 * sync_super for 0.90.0
788 */
789static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
790{
791 mdp_super_t *sb;
792 struct list_head *tmp;
793 mdk_rdev_t *rdev2;
794 int next_spare = mddev->raid_disks;
19133a42 795
1da177e4
LT		 796
797 /* make rdev->sb match mddev data..
798 *
799 * 1/ zero out disks
800 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
801 * 3/ any empty disks < next_spare become removed
802 *
803 * disks[0] gets initialised to REMOVED because
804 * we cannot be sure from other fields if it has
805 * been initialised or not.
806 */
807 int i;
808 int active=0, working=0,failed=0,spare=0,nr_disks=0;
809
61181565
N		 810 rdev->sb_size = MD_SB_BYTES;
811
1da177e4
LT		 812 sb = (mdp_super_t*)page_address(rdev->sb_page);
813
814 memset(sb, 0, sizeof(*sb));
815
816 sb->md_magic = MD_SB_MAGIC;
817 sb->major_version = mddev->major_version;
818 sb->minor_version = mddev->minor_version;
819 sb->patch_version = mddev->patch_version;
820 sb->gvalid_words = 0; /* ignored */
821 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
822 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
823 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
824 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
825
826 sb->ctime = mddev->ctime;
827 sb->level = mddev->level;
828 sb->size = mddev->size;
829 sb->raid_disks = mddev->raid_disks;
830 sb->md_minor = mddev->md_minor;
831 sb->not_persistent = !mddev->persistent;
832 sb->utime = mddev->utime;
833 sb->state = 0;
834 sb->events_hi = (mddev->events>>32);
835 sb->events_lo = (u32)mddev->events;
836
837 if (mddev->in_sync)
838 {
839 sb->recovery_cp = mddev->recovery_cp;
840 sb->cp_events_hi = (mddev->events>>32);
841 sb->cp_events_lo = (u32)mddev->events;
842 if (mddev->recovery_cp == MaxSector)
843 sb->state = (1<< MD_SB_CLEAN);
844 } else
845 sb->recovery_cp = 0;
846
847 sb->layout = mddev->layout;
848 sb->chunk_size = mddev->chunk_size;
849
a654b9d8
N		 850 if (mddev->bitmap && mddev->bitmap_file == NULL)
851 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
852
1da177e4
LT		 853 sb->disks[0].state = (1<<MD_DISK_REMOVED);
854 ITERATE_RDEV(mddev,rdev2,tmp) {
855 mdp_disk_t *d;
86e6ffdd 856 int desc_nr;
b2d444d7
N		 857 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
858 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 859 desc_nr = rdev2->raid_disk;
1da177e4 860 else
86e6ffdd 861 desc_nr = next_spare++;
19133a42 862 rdev2->desc_nr = desc_nr;
1da177e4
LT		 863 d = &sb->disks[rdev2->desc_nr];
864 nr_disks++;
865 d->number = rdev2->desc_nr;
866 d->major = MAJOR(rdev2->bdev->bd_dev);
867 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 868 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
869 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 870 d->raid_disk = rdev2->raid_disk;
871 else
872 d->raid_disk = rdev2->desc_nr; /* compatibility */
b2d444d7 873 if (test_bit(Faulty, &rdev2->flags)) {
1da177e4
LT		 874 d->state = (1<<MD_DISK_FAULTY);
875 failed++;
b2d444d7 876 } else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 877 d->state = (1<<MD_DISK_ACTIVE);
878 d->state |= (1<<MD_DISK_SYNC);
879 active++;
880 working++;
881 } else {
882 d->state = 0;
883 spare++;
884 working++;
885 }
8ddf9efe
N		 886 if (test_bit(WriteMostly, &rdev2->flags))
887 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 888 }
1da177e4
LT		 889 /* now set the "removed" and "faulty" bits on any missing devices */
890 for (i=0 ; i < mddev->raid_disks ; i++) {
891 mdp_disk_t *d = &sb->disks[i];
892 if (d->state == 0 && d->number == 0) {
893 d->number = i;
894 d->raid_disk = i;
895 d->state = (1<<MD_DISK_REMOVED);
896 d->state |= (1<<MD_DISK_FAULTY);
897 failed++;
898 }
899 }
900 sb->nr_disks = nr_disks;
901 sb->active_disks = active;
902 sb->working_disks = working;
903 sb->failed_disks = failed;
904 sb->spare_disks = spare;
905
906 sb->this_disk = sb->disks[rdev->desc_nr];
907 sb->sb_csum = calc_sb_csum(sb);
908}
909
910/*
911 * version 1 superblock
912 */
913
914static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
915{
916 unsigned int disk_csum, csum;
917 unsigned long long newcsum;
918 int size = 256 + le32_to_cpu(sb->max_dev)*2;
919 unsigned int *isuper = (unsigned int*)sb;
920 int i;
921
922 disk_csum = sb->sb_csum;
923 sb->sb_csum = 0;
924 newcsum = 0;
925 for (i=0; size>=4; size -= 4 )
926 newcsum += le32_to_cpu(*isuper++);
927
928 if (size == 2)
929 newcsum += le16_to_cpu(*(unsigned short*) isuper);
930
931 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
932 sb->sb_csum = disk_csum;
933 return cpu_to_le32(csum);
934}
935
936static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
937{
938 struct mdp_superblock_1 *sb;
939 int ret;
940 sector_t sb_offset;
941 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 942 int bmask;
1da177e4
LT		 943
944 /*
945 * Calculate the position of the superblock.
946 * It is always aligned to a 4K boundary and
947 * depeding on minor_version, it can be:
948 * 0: At least 8K, but less than 12K, from end of device
949 * 1: At start of device
950 * 2: 4K from start of device.
951 */
952 switch(minor_version) {
953 case 0:
954 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
955 sb_offset -= 8*2;
39730960 956 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 957 /* convert from sectors to K */
958 sb_offset /= 2;
959 break;
960 case 1:
961 sb_offset = 0;
962 break;
963 case 2:
964 sb_offset = 4;
965 break;
966 default:
967 return -EINVAL;
968 }
969 rdev->sb_offset = sb_offset;
970
0002b271
N		 971 /* superblock is rarely larger than 1K, but it can be larger,
972 * and it is safe to read 4k, so we do that
973 */
974 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 975 if (ret) return ret;
976
977
978 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
979
980 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
981 sb->major_version != cpu_to_le32(1) ||
982 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
983 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 984 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 985 return -EINVAL;
986
987 if (calc_sb_1_csum(sb) != sb->sb_csum) {
988 printk("md: invalid superblock checksum on %s\n",
989 bdevname(rdev->bdev,b));
990 return -EINVAL;
991 }
992 if (le64_to_cpu(sb->data_size) < 10) {
993 printk("md: data_size too small on %s\n",
994 bdevname(rdev->bdev,b));
995 return -EINVAL;
996 }
997 rdev->preferred_minor = 0xffff;
998 rdev->data_offset = le64_to_cpu(sb->data_offset);
999
0002b271 1000 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1001 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1002 if (rdev->sb_size & bmask)
1003 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1004
1da177e4
LT		 1005 if (refdev == 0)
1006 return 1;
1007 else {
1008 __u64 ev1, ev2;
1009 struct mdp_superblock_1 *refsb =
1010 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1011
1012 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1013 sb->level != refsb->level ||
1014 sb->layout != refsb->layout ||
1015 sb->chunksize != refsb->chunksize) {
1016 printk(KERN_WARNING "md: %s has strangely different"
1017 " superblock to %s\n",
1018 bdevname(rdev->bdev,b),
1019 bdevname(refdev->bdev,b2));
1020 return -EINVAL;
1021 }
1022 ev1 = le64_to_cpu(sb->events);
1023 ev2 = le64_to_cpu(refsb->events);
1024
1025 if (ev1 > ev2)
1026 return 1;
1027 }
1028 if (minor_version)
1029 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1030 else
1031 rdev->size = rdev->sb_offset;
1032 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1033 return -EINVAL;
1034 rdev->size = le64_to_cpu(sb->data_size)/2;
1035 if (le32_to_cpu(sb->chunksize))
1036 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1037 return 0;
1038}
1039
1040static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1041{
1042 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1043
41158c7e 1044 rdev->raid_disk = -1;
b2d444d7 1045 rdev->flags = 0;
1da177e4
LT		 1046 if (mddev->raid_disks == 0) {
1047 mddev->major_version = 1;
1048 mddev->patch_version = 0;
1049 mddev->persistent = 1;
1050 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1051 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1052 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1053 mddev->level = le32_to_cpu(sb->level);
1054 mddev->layout = le32_to_cpu(sb->layout);
1055 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1056 mddev->size = le64_to_cpu(sb->size)/2;
1057 mddev->events = le64_to_cpu(sb->events);
9223214e 1058 mddev->bitmap_offset = 0;
53e87fbb 1059 mddev->default_bitmap_offset = 1024;
1da177e4
LT		 1060
1061 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1062 memcpy(mddev->uuid, sb->set_uuid, 16);
1063
1064 mddev->max_disks = (4096-256)/2;
a654b9d8 1065
71c0805c 1066 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
a654b9d8 1067 mddev->bitmap_file == NULL ) {
6cce3b23
N		 1068 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1069 && mddev->level != 10) {
1070 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
a654b9d8
N		 1071 return -EINVAL;
1072 }
1073 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1074 }
41158c7e
N		 1075 } else if (mddev->pers == NULL) {
1076 /* Insist of good event counter while assembling */
1077 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4
LT		 1078 ++ev1;
1079 if (ev1 < mddev->events)
1080 return -EINVAL;
41158c7e
N		 1081 } else if (mddev->bitmap) {
1082 /* If adding to array with a bitmap, then we can accept an
1083 * older device, but not too old.
1084 */
1085 __u64 ev1 = le64_to_cpu(sb->events);
1086 if (ev1 < mddev->bitmap->events_cleared)
1087 return 0;
1088 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1089 return 0;
1da177e4
LT		 1090
1091 if (mddev->level != LEVEL_MULTIPATH) {
1092 int role;
1093 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1094 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1095 switch(role) {
1096 case 0xffff: /* spare */
1da177e4
LT		 1097 break;
1098 case 0xfffe: /* faulty */
b2d444d7 1099 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1100 break;
1101 default:
b2d444d7 1102 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1103 rdev->raid_disk = role;
1104 break;
1105 }
8ddf9efe
N		 1106 if (sb->devflags & WriteMostly1)
1107 set_bit(WriteMostly, &rdev->flags);
41158c7e 1108 } else /* MULTIPATH are always insync */
b2d444d7 1109 set_bit(In_sync, &rdev->flags);
41158c7e 1110
1da177e4
LT		 1111 return 0;
1112}
1113
1114static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1115{
1116 struct mdp_superblock_1 *sb;
1117 struct list_head *tmp;
1118 mdk_rdev_t *rdev2;
1119 int max_dev, i;
1120 /* make rdev->sb match mddev and rdev data. */
1121
1122 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1123
1124 sb->feature_map = 0;
1125 sb->pad0 = 0;
1126 memset(sb->pad1, 0, sizeof(sb->pad1));
1127 memset(sb->pad2, 0, sizeof(sb->pad2));
1128 memset(sb->pad3, 0, sizeof(sb->pad3));
1129
1130 sb->utime = cpu_to_le64((__u64)mddev->utime);
1131 sb->events = cpu_to_le64(mddev->events);
1132 if (mddev->in_sync)
1133 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1134 else
1135 sb->resync_offset = cpu_to_le64(0);
1136
a654b9d8
N		 1137 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1138 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1139 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8
N		 1140 }
1141
1da177e4
LT		 1142 max_dev = 0;
1143 ITERATE_RDEV(mddev,rdev2,tmp)
1144 if (rdev2->desc_nr+1 > max_dev)
1145 max_dev = rdev2->desc_nr+1;
1146
1147 sb->max_dev = cpu_to_le32(max_dev);
1148 for (i=0; i<max_dev;i++)
1149 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1150
1151 ITERATE_RDEV(mddev,rdev2,tmp) {
1152 i = rdev2->desc_nr;
b2d444d7 1153 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1154 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1155 else if (test_bit(In_sync, &rdev2->flags))
1da177e4
LT		 1156 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1157 else
1158 sb->dev_roles[i] = cpu_to_le16(0xffff);
1159 }
1160
1161 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1162 sb->sb_csum = calc_sb_1_csum(sb);
1163}
1164
1165
75c96f85 1166static struct super_type super_types[] = {
1da177e4
LT		 1167 [0] = {
1168 .name = "0.90.0",
1169 .owner = THIS_MODULE,
1170 .load_super = super_90_load,
1171 .validate_super = super_90_validate,
1172 .sync_super = super_90_sync,
1173 },
1174 [1] = {
1175 .name = "md-1",
1176 .owner = THIS_MODULE,
1177 .load_super = super_1_load,
1178 .validate_super = super_1_validate,
1179 .sync_super = super_1_sync,
1180 },
1181};
1182
1183static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1184{
1185 struct list_head *tmp;
1186 mdk_rdev_t *rdev;
1187
1188 ITERATE_RDEV(mddev,rdev,tmp)
1189 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1190 return rdev;
1191
1192 return NULL;
1193}
1194
1195static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1196{
1197 struct list_head *tmp;
1198 mdk_rdev_t *rdev;
1199
1200 ITERATE_RDEV(mddev1,rdev,tmp)
1201 if (match_dev_unit(mddev2, rdev))
1202 return 1;
1203
1204 return 0;
1205}
1206
1207static LIST_HEAD(pending_raid_disks);
1208
1209static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1210{
1211 mdk_rdev_t *same_pdev;
1212 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
f637b9f9 1213 struct kobject *ko;
1da177e4
LT		 1214
1215 if (rdev->mddev) {
1216 MD_BUG();
1217 return -EINVAL;
1218 }
1219 same_pdev = match_dev_unit(mddev, rdev);
1220 if (same_pdev)
1221 printk(KERN_WARNING
1222 "%s: WARNING: %s appears to be on the same physical"
1223 " disk as %s. True\n protection against single-disk"
1224 " failure might be compromised.\n",
1225 mdname(mddev), bdevname(rdev->bdev,b),
1226 bdevname(same_pdev->bdev,b2));
1227
1228 /* Verify rdev->desc_nr is unique.
1229 * If it is -1, assign a free number, else
1230 * check number is not in use
1231 */
1232 if (rdev->desc_nr < 0) {
1233 int choice = 0;
1234 if (mddev->pers) choice = mddev->raid_disks;
1235 while (find_rdev_nr(mddev, choice))
1236 choice++;
1237 rdev->desc_nr = choice;
1238 } else {
1239 if (find_rdev_nr(mddev, rdev->desc_nr))
1240 return -EBUSY;
1241 }
19133a42
N		 1242 bdevname(rdev->bdev,b);
1243 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1244 return -ENOMEM;
1da177e4
LT		 1245
1246 list_add(&rdev->same_set, &mddev->disks);
1247 rdev->mddev = mddev;
19133a42 1248 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1249
9c791977 1250 rdev->kobj.parent = &mddev->kobj;
86e6ffdd
N		 1251 kobject_add(&rdev->kobj);
1252
f637b9f9
N		 1253 if (rdev->bdev->bd_part)
1254 ko = &rdev->bdev->bd_part->kobj;
1255 else
1256 ko = &rdev->bdev->bd_disk->kobj;
1257 sysfs_create_link(&rdev->kobj, ko, "block");
1da177e4
LT		 1258 return 0;
1259}
1260
1261static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1262{
1263 char b[BDEVNAME_SIZE];
1264 if (!rdev->mddev) {
1265 MD_BUG();
1266 return;
1267 }
1268 list_del_init(&rdev->same_set);
1269 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1270 rdev->mddev = NULL;
86e6ffdd
N		 1271 sysfs_remove_link(&rdev->kobj, "block");
1272 kobject_del(&rdev->kobj);
1da177e4
LT		 1273}
1274
1275/*
1276 * prevent the device from being mounted, repartitioned or
1277 * otherwise reused by a RAID array (or any other kernel
1278 * subsystem), by bd_claiming the device.
1279 */
1280static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1281{
1282 int err = 0;
1283 struct block_device *bdev;
1284 char b[BDEVNAME_SIZE];
1285
1286 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1287 if (IS_ERR(bdev)) {
1288 printk(KERN_ERR "md: could not open %s.\n",
1289 __bdevname(dev, b));
1290 return PTR_ERR(bdev);
1291 }
1292 err = bd_claim(bdev, rdev);
1293 if (err) {
1294 printk(KERN_ERR "md: could not bd_claim %s.\n",
1295 bdevname(bdev, b));
1296 blkdev_put(bdev);
1297 return err;
1298 }
1299 rdev->bdev = bdev;
1300 return err;
1301}
1302
1303static void unlock_rdev(mdk_rdev_t *rdev)
1304{
1305 struct block_device *bdev = rdev->bdev;
1306 rdev->bdev = NULL;
1307 if (!bdev)
1308 MD_BUG();
1309 bd_release(bdev);
1310 blkdev_put(bdev);
1311}
1312
1313void md_autodetect_dev(dev_t dev);
1314
1315static void export_rdev(mdk_rdev_t * rdev)
1316{
1317 char b[BDEVNAME_SIZE];
1318 printk(KERN_INFO "md: export_rdev(%s)\n",
1319 bdevname(rdev->bdev,b));
1320 if (rdev->mddev)
1321 MD_BUG();
1322 free_disk_sb(rdev);
1323 list_del_init(&rdev->same_set);
1324#ifndef MODULE
1325 md_autodetect_dev(rdev->bdev->bd_dev);
1326#endif
1327 unlock_rdev(rdev);
86e6ffdd 1328 kobject_put(&rdev->kobj);
1da177e4
LT		 1329}
1330
1331static void kick_rdev_from_array(mdk_rdev_t * rdev)
1332{
1333 unbind_rdev_from_array(rdev);
1334 export_rdev(rdev);
1335}
1336
1337static void export_array(mddev_t *mddev)
1338{
1339 struct list_head *tmp;
1340 mdk_rdev_t *rdev;
1341
1342 ITERATE_RDEV(mddev,rdev,tmp) {
1343 if (!rdev->mddev) {
1344 MD_BUG();
1345 continue;
1346 }
1347 kick_rdev_from_array(rdev);
1348 }
1349 if (!list_empty(&mddev->disks))
1350 MD_BUG();
1351 mddev->raid_disks = 0;
1352 mddev->major_version = 0;
1353}
1354
1355static void print_desc(mdp_disk_t *desc)
1356{
1357 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1358 desc->major,desc->minor,desc->raid_disk,desc->state);
1359}
1360
1361static void print_sb(mdp_super_t *sb)
1362{
1363 int i;
1364
1365 printk(KERN_INFO
1366 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1367 sb->major_version, sb->minor_version, sb->patch_version,
1368 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1369 sb->ctime);
1370 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1371 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1372 sb->md_minor, sb->layout, sb->chunk_size);
1373 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1374 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1375 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1376 sb->failed_disks, sb->spare_disks,
1377 sb->sb_csum, (unsigned long)sb->events_lo);
1378
1379 printk(KERN_INFO);
1380 for (i = 0; i < MD_SB_DISKS; i++) {
1381 mdp_disk_t *desc;
1382
1383 desc = sb->disks + i;
1384 if (desc->number || desc->major || desc->minor ||
1385 desc->raid_disk || (desc->state && (desc->state != 4))) {
1386 printk(" D %2d: ", i);
1387 print_desc(desc);
1388 }
1389 }
1390 printk(KERN_INFO "md: THIS: ");
1391 print_desc(&sb->this_disk);
1392
1393}
1394
1395static void print_rdev(mdk_rdev_t *rdev)
1396{
1397 char b[BDEVNAME_SIZE];
1398 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1399 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1400 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1401 rdev->desc_nr);
1da177e4
LT		 1402 if (rdev->sb_loaded) {
1403 printk(KERN_INFO "md: rdev superblock:\n");
1404 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1405 } else
1406 printk(KERN_INFO "md: no rdev superblock!\n");
1407}
1408
1409void md_print_devices(void)
1410{
1411 struct list_head *tmp, *tmp2;
1412 mdk_rdev_t *rdev;
1413 mddev_t *mddev;
1414 char b[BDEVNAME_SIZE];
1415
1416 printk("\n");
1417 printk("md: **********************************\n");
1418 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1419 printk("md: **********************************\n");
1420 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1421
32a7627c
N		 1422 if (mddev->bitmap)
1423 bitmap_print_sb(mddev->bitmap);
1424 else
1425 printk("%s: ", mdname(mddev));
1da177e4
LT		 1426 ITERATE_RDEV(mddev,rdev,tmp2)
1427 printk("<%s>", bdevname(rdev->bdev,b));
1428 printk("\n");
1429
1430 ITERATE_RDEV(mddev,rdev,tmp2)
1431 print_rdev(rdev);
1432 }
1433 printk("md: **********************************\n");
1434 printk("\n");
1435}
1436
1437
1da177e4
LT		 1438static void sync_sbs(mddev_t * mddev)
1439{
1440 mdk_rdev_t *rdev;
1441 struct list_head *tmp;
1442
1443 ITERATE_RDEV(mddev,rdev,tmp) {
1444 super_types[mddev->major_version].
1445 sync_super(mddev, rdev);
1446 rdev->sb_loaded = 1;
1447 }
1448}
1449
1450static void md_update_sb(mddev_t * mddev)
1451{
7bfa19f2 1452 int err;
1da177e4
LT		 1453 struct list_head *tmp;
1454 mdk_rdev_t *rdev;
06d91a5f 1455 int sync_req;
1da177e4 1456
1da177e4 1457repeat:
a9701a30 1458 spin_lock_irq(&mddev->write_lock);
06d91a5f 1459 sync_req = mddev->in_sync;
1da177e4
LT		 1460 mddev->utime = get_seconds();
1461 mddev->events ++;
1462
1463 if (!mddev->events) {
1464 /*
1465 * oops, this 64-bit counter should never wrap.
1466 * Either we are in around ~1 trillion A.C., assuming
1467 * 1 reboot per second, or we have a bug:
1468 */
1469 MD_BUG();
1470 mddev->events --;
1471 }
7bfa19f2 1472 mddev->sb_dirty = 2;
1da177e4
LT		 1473 sync_sbs(mddev);
1474
1475 /*
1476 * do not write anything to disk if using
1477 * nonpersistent superblocks
1478 */
06d91a5f
N		 1479 if (!mddev->persistent) {
1480 mddev->sb_dirty = 0;
a9701a30 1481 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1482 wake_up(&mddev->sb_wait);
1da177e4 1483 return;
06d91a5f 1484 }
a9701a30 1485 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1486
1487 dprintk(KERN_INFO
1488 "md: updating %s RAID superblock on device (in sync %d)\n",
1489 mdname(mddev),mddev->in_sync);
1490
32a7627c 1491 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1492 ITERATE_RDEV(mddev,rdev,tmp) {
1493 char b[BDEVNAME_SIZE];
1494 dprintk(KERN_INFO "md: ");
b2d444d7 1495 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1496 dprintk("(skipping faulty ");
1497
1498 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1499 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1500 md_super_write(mddev,rdev,
0002b271 1501 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1502 rdev->sb_page);
1503 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1504 bdevname(rdev->bdev,b),
1505 (unsigned long long)rdev->sb_offset);
1506
1da177e4
LT		 1507 } else
1508 dprintk(")\n");
7bfa19f2 1509 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1510 /* only need to write one superblock... */
1511 break;
1512 }
a9701a30 1513 md_super_wait(mddev);
7bfa19f2
N		 1514 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1515
a9701a30 1516 spin_lock_irq(&mddev->write_lock);
7bfa19f2 1517 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
06d91a5f 1518 /* have to write it out again */
a9701a30 1519 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1520 goto repeat;
1521 }
1522 mddev->sb_dirty = 0;
a9701a30 1523 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1524 wake_up(&mddev->sb_wait);
06d91a5f 1525
1da177e4
LT		 1526}
1527
86e6ffdd
N		 1528struct rdev_sysfs_entry {
1529 struct attribute attr;
1530 ssize_t (*show)(mdk_rdev_t *, char *);
1531 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1532};
1533
1534static ssize_t
96de1e66 1535state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1536{
1537 char *sep = "";
1538 int len=0;
1539
b2d444d7 1540 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1541 len+= sprintf(page+len, "%sfaulty",sep);
1542 sep = ",";
1543 }
b2d444d7 1544 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1545 len += sprintf(page+len, "%sin_sync",sep);
1546 sep = ",";
1547 }
b2d444d7
N		 1548 if (!test_bit(Faulty, &rdev->flags) &&
1549 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1550 len += sprintf(page+len, "%sspare", sep);
1551 sep = ",";
1552 }
1553 return len+sprintf(page+len, "\n");
1554}
1555
96de1e66
N		 1556static struct rdev_sysfs_entry
1557rdev_state = __ATTR_RO(state);
86e6ffdd
N		 1558
1559static ssize_t
96de1e66 1560super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1561{
1562 if (rdev->sb_loaded && rdev->sb_size) {
1563 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1564 return rdev->sb_size;
1565 } else
1566 return 0;
1567}
96de1e66
N		 1568static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1569
86e6ffdd
N		 1570static struct attribute *rdev_default_attrs[] = {
1571 &rdev_state.attr,
1572 &rdev_super.attr,
1573 NULL,
1574};
1575static ssize_t
1576rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1577{
1578 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1579 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1580
1581 if (!entry->show)
1582 return -EIO;
1583 return entry->show(rdev, page);
1584}
1585
1586static ssize_t
1587rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1588 const char *page, size_t length)
1589{
1590 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1591 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1592
1593 if (!entry->store)
1594 return -EIO;
1595 return entry->store(rdev, page, length);
1596}
1597
1598static void rdev_free(struct kobject *ko)
1599{
1600 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1601 kfree(rdev);
1602}
1603static struct sysfs_ops rdev_sysfs_ops = {
1604 .show = rdev_attr_show,
1605 .store = rdev_attr_store,
1606};
1607static struct kobj_type rdev_ktype = {
1608 .release = rdev_free,
1609 .sysfs_ops = &rdev_sysfs_ops,
1610 .default_attrs = rdev_default_attrs,
1611};
1612
1da177e4
LT		 1613/*
1614 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1615 *
1616 * mark the device faulty if:
1617 *
1618 * - the device is nonexistent (zero size)
1619 * - the device has no valid superblock
1620 *
1621 * a faulty rdev _never_ has rdev->sb set.
1622 */
1623static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1624{
1625 char b[BDEVNAME_SIZE];
1626 int err;
1627 mdk_rdev_t *rdev;
1628 sector_t size;
1629
9ffae0cf 1630 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 1631 if (!rdev) {
1632 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1633 return ERR_PTR(-ENOMEM);
1634 }
1da177e4
LT		 1635
1636 if ((err = alloc_disk_sb(rdev)))
1637 goto abort_free;
1638
1639 err = lock_rdev(rdev, newdev);
1640 if (err)
1641 goto abort_free;
1642
86e6ffdd
N		 1643 rdev->kobj.parent = NULL;
1644 rdev->kobj.ktype = &rdev_ktype;
1645 kobject_init(&rdev->kobj);
1646
1da177e4 1647 rdev->desc_nr = -1;
b2d444d7 1648 rdev->flags = 0;
1da177e4
LT		 1649 rdev->data_offset = 0;
1650 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 1651 atomic_set(&rdev->read_errors, 0);
1da177e4
LT		 1652
1653 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1654 if (!size) {
1655 printk(KERN_WARNING
1656 "md: %s has zero or unknown size, marking faulty!\n",
1657 bdevname(rdev->bdev,b));
1658 err = -EINVAL;
1659 goto abort_free;
1660 }
1661
1662 if (super_format >= 0) {
1663 err = super_types[super_format].
1664 load_super(rdev, NULL, super_minor);
1665 if (err == -EINVAL) {
1666 printk(KERN_WARNING
1667 "md: %s has invalid sb, not importing!\n",
1668 bdevname(rdev->bdev,b));
1669 goto abort_free;
1670 }
1671 if (err < 0) {
1672 printk(KERN_WARNING
1673 "md: could not read %s's sb, not importing!\n",
1674 bdevname(rdev->bdev,b));
1675 goto abort_free;
1676 }
1677 }
1678 INIT_LIST_HEAD(&rdev->same_set);
1679
1680 return rdev;
1681
1682abort_free:
1683 if (rdev->sb_page) {
1684 if (rdev->bdev)
1685 unlock_rdev(rdev);
1686 free_disk_sb(rdev);
1687 }
1688 kfree(rdev);
1689 return ERR_PTR(err);
1690}
1691
1692/*
1693 * Check a full RAID array for plausibility
1694 */
1695
1696
a757e64c 1697static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 1698{
1699 int i;
1700 struct list_head *tmp;
1701 mdk_rdev_t *rdev, *freshest;
1702 char b[BDEVNAME_SIZE];
1703
1704 freshest = NULL;
1705 ITERATE_RDEV(mddev,rdev,tmp)
1706 switch (super_types[mddev->major_version].
1707 load_super(rdev, freshest, mddev->minor_version)) {
1708 case 1:
1709 freshest = rdev;
1710 break;
1711 case 0:
1712 break;
1713 default:
1714 printk( KERN_ERR \
1715 "md: fatal superblock inconsistency in %s"
1716 " -- removing from array\n",
1717 bdevname(rdev->bdev,b));
1718 kick_rdev_from_array(rdev);
1719 }
1720
1721
1722 super_types[mddev->major_version].
1723 validate_super(mddev, freshest);
1724
1725 i = 0;
1726 ITERATE_RDEV(mddev,rdev,tmp) {
1727 if (rdev != freshest)
1728 if (super_types[mddev->major_version].
1729 validate_super(mddev, rdev)) {
1730 printk(KERN_WARNING "md: kicking non-fresh %s"
1731 " from array!\n",
1732 bdevname(rdev->bdev,b));
1733 kick_rdev_from_array(rdev);
1734 continue;
1735 }
1736 if (mddev->level == LEVEL_MULTIPATH) {
1737 rdev->desc_nr = i++;
1738 rdev->raid_disk = rdev->desc_nr;
b2d444d7 1739 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1740 }
1741 }
1742
1743
1744
1745 if (mddev->recovery_cp != MaxSector &&
1746 mddev->level >= 1)
1747 printk(KERN_ERR "md: %s: raid array is not clean"
1748 " -- starting background reconstruction\n",
1749 mdname(mddev));
1750
1da177e4
LT		 1751}
1752
eae1701f 1753static ssize_t
96de1e66 1754level_show(mddev_t *mddev, char *page)
eae1701f 1755{
2604b703 1756 struct mdk_personality *p = mddev->pers;
bb636547 1757 if (p == NULL && mddev->raid_disks == 0)
eae1701f
N		 1758 return 0;
1759 if (mddev->level >= 0)
bcb97940 1760 return sprintf(page, "raid%d\n", mddev->level);
eae1701f
N		 1761 else
1762 return sprintf(page, "%s\n", p->name);
1763}
1764
96de1e66 1765static struct md_sysfs_entry md_level = __ATTR_RO(level);
eae1701f
N		 1766
1767static ssize_t
96de1e66 1768raid_disks_show(mddev_t *mddev, char *page)
eae1701f 1769{
bb636547
N		 1770 if (mddev->raid_disks == 0)
1771 return 0;
eae1701f
N		 1772 return sprintf(page, "%d\n", mddev->raid_disks);
1773}
1774
96de1e66 1775static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
eae1701f 1776
24dd469d 1777static ssize_t
7eec314d 1778action_show(mddev_t *mddev, char *page)
24dd469d 1779{
7eec314d 1780 char *type = "idle";
31399d9e
N		 1781 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1782 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1783 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 1784 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1785 type = "resync";
1786 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1787 type = "check";
1788 else
1789 type = "repair";
1790 } else
1791 type = "recover";
1792 }
1793 return sprintf(page, "%s\n", type);
1794}
1795
1796static ssize_t
7eec314d 1797action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 1798{
7eec314d
N		 1799 if (!mddev->pers || !mddev->pers->sync_request)
1800 return -EINVAL;
1801
1802 if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1803 if (mddev->sync_thread) {
1804 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1805 md_unregister_thread(mddev->sync_thread);
1806 mddev->sync_thread = NULL;
1807 mddev->recovery = 0;
1808 }
1809 return len;
1810 }
31399d9e
N		 1811
1812 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1813 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 1814 return -EBUSY;
7eec314d
N		 1815 if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1816 strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1817 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1818 else {
1819 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1820 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1821 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1822 return -EINVAL;
1823 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1824 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1825 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1826 }
24dd469d
N		 1827 md_wakeup_thread(mddev->thread);
1828 return len;
1829}
1830
9d88883e 1831static ssize_t
96de1e66 1832mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 1833{
1834 return sprintf(page, "%llu\n",
1835 (unsigned long long) mddev->resync_mismatches);
1836}
1837
96de1e66 1838static struct md_sysfs_entry
7eec314d 1839md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 1840
96de1e66
N		 1841
1842static struct md_sysfs_entry
1843md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 1844
eae1701f
N		 1845static struct attribute *md_default_attrs[] = {
1846 &md_level.attr,
1847 &md_raid_disks.attr,
411036fa
N		 1848 NULL,
1849};
1850
1851static struct attribute *md_redundancy_attrs[] = {
24dd469d 1852 &md_scan_mode.attr,
9d88883e 1853 &md_mismatches.attr,
eae1701f
N		 1854 NULL,
1855};
411036fa
N		 1856static struct attribute_group md_redundancy_group = {
1857 .name = NULL,
1858 .attrs = md_redundancy_attrs,
1859};
1860
eae1701f
N		 1861
1862static ssize_t
1863md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1864{
1865 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1866 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 1867 ssize_t rv;
eae1701f
N		 1868
1869 if (!entry->show)
1870 return -EIO;
96de1e66
N		 1871 mddev_lock(mddev);
1872 rv = entry->show(mddev, page);
1873 mddev_unlock(mddev);
1874 return rv;
eae1701f
N		 1875}
1876
1877static ssize_t
1878md_attr_store(struct kobject *kobj, struct attribute *attr,
1879 const char *page, size_t length)
1880{
1881 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1882 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 1883 ssize_t rv;
eae1701f
N		 1884
1885 if (!entry->store)
1886 return -EIO;
96de1e66
N		 1887 mddev_lock(mddev);
1888 rv = entry->store(mddev, page, length);
1889 mddev_unlock(mddev);
1890 return rv;
eae1701f
N		 1891}
1892
1893static void md_free(struct kobject *ko)
1894{
1895 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1896 kfree(mddev);
1897}
1898
1899static struct sysfs_ops md_sysfs_ops = {
1900 .show = md_attr_show,
1901 .store = md_attr_store,
1902};
1903static struct kobj_type md_ktype = {
1904 .release = md_free,
1905 .sysfs_ops = &md_sysfs_ops,
1906 .default_attrs = md_default_attrs,
1907};
1908
1da177e4
LT		 1909int mdp_major = 0;
1910
1911static struct kobject *md_probe(dev_t dev, int *part, void *data)
1912{
1913 static DECLARE_MUTEX(disks_sem);
1914 mddev_t *mddev = mddev_find(dev);
1915 struct gendisk *disk;
1916 int partitioned = (MAJOR(dev) != MD_MAJOR);
1917 int shift = partitioned ? MdpMinorShift : 0;
1918 int unit = MINOR(dev) >> shift;
1919
1920 if (!mddev)
1921 return NULL;
1922
1923 down(&disks_sem);
1924 if (mddev->gendisk) {
1925 up(&disks_sem);
1926 mddev_put(mddev);
1927 return NULL;
1928 }
1929 disk = alloc_disk(1 << shift);
1930 if (!disk) {
1931 up(&disks_sem);
1932 mddev_put(mddev);
1933 return NULL;
1934 }
1935 disk->major = MAJOR(dev);
1936 disk->first_minor = unit << shift;
1937 if (partitioned) {
1938 sprintf(disk->disk_name, "md_d%d", unit);
1939 sprintf(disk->devfs_name, "md/d%d", unit);
1940 } else {
1941 sprintf(disk->disk_name, "md%d", unit);
1942 sprintf(disk->devfs_name, "md/%d", unit);
1943 }
1944 disk->fops = &md_fops;
1945 disk->private_data = mddev;
1946 disk->queue = mddev->queue;
1947 add_disk(disk);
1948 mddev->gendisk = disk;
1949 up(&disks_sem);
9c791977 1950 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 1951 mddev->kobj.k_name = NULL;
1952 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1953 mddev->kobj.ktype = &md_ktype;
1954 kobject_register(&mddev->kobj);
1da177e4
LT		 1955 return NULL;
1956}
1957
1958void md_wakeup_thread(mdk_thread_t *thread);
1959
1960static void md_safemode_timeout(unsigned long data)
1961{
1962 mddev_t *mddev = (mddev_t *) data;
1963
1964 mddev->safemode = 1;
1965 md_wakeup_thread(mddev->thread);
1966}
1967
6ff8d8ec 1968static int start_dirty_degraded;
1da177e4
LT		 1969
1970static int do_md_run(mddev_t * mddev)
1971{
2604b703 1972 int err;
1da177e4
LT		 1973 int chunk_size;
1974 struct list_head *tmp;
1975 mdk_rdev_t *rdev;
1976 struct gendisk *disk;
2604b703 1977 struct mdk_personality *pers;
1da177e4
LT		 1978 char b[BDEVNAME_SIZE];
1979
a757e64c
N		 1980 if (list_empty(&mddev->disks))
1981 /* cannot run an array with no devices.. */
1da177e4 1982 return -EINVAL;
1da177e4
LT		 1983
1984 if (mddev->pers)
1985 return -EBUSY;
1986
1987 /*
1988 * Analyze all RAID superblock(s)
1989 */
a757e64c
N		 1990 if (!mddev->raid_disks)
1991 analyze_sbs(mddev);
1da177e4
LT		 1992
1993 chunk_size = mddev->chunk_size;
2604b703
N		 1994
1995 if (chunk_size) {
1da177e4
LT		 1996 if (chunk_size > MAX_CHUNK_SIZE) {
1997 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1998 chunk_size, MAX_CHUNK_SIZE);
1999 return -EINVAL;
2000 }
2001 /*
2002 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2003 */
2004 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 2005 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 2006 return -EINVAL;
2007 }
2008 if (chunk_size < PAGE_SIZE) {
2009 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2010 chunk_size, PAGE_SIZE);
2011 return -EINVAL;
2012 }
2013
2014 /* devices must have minimum size of one chunk */
2015 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2016 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2017 continue;
2018 if (rdev->size < chunk_size / 1024) {
2019 printk(KERN_WARNING
2020 "md: Dev %s smaller than chunk_size:"
2021 " %lluk < %dk\n",
2022 bdevname(rdev->bdev,b),
2023 (unsigned long long)rdev->size,
2024 chunk_size / 1024);
2025 return -EINVAL;
2026 }
2027 }
2028 }
2029
1da177e4 2030#ifdef CONFIG_KMOD
2604b703 2031 request_module("md-level-%d", mddev->level);
1da177e4
LT		 2032#endif
2033
2034 /*
2035 * Drop all container device buffers, from now on
2036 * the only valid external interface is through the md
2037 * device.
2038 * Also find largest hardsector size
2039 */
2040 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2041 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2042 continue;
2043 sync_blockdev(rdev->bdev);
2044 invalidate_bdev(rdev->bdev, 0);
2045 }
2046
2047 md_probe(mddev->unit, NULL, NULL);
2048 disk = mddev->gendisk;
2049 if (!disk)
2050 return -ENOMEM;
2051
2052 spin_lock(&pers_lock);
2604b703
N		 2053 pers = find_pers(mddev->level);
2054 if (!pers || !try_module_get(pers->owner)) {
1da177e4 2055 spin_unlock(&pers_lock);
2604b703
N		 2056 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2057 mddev->level);
1da177e4
LT		 2058 return -EINVAL;
2059 }
2604b703 2060 mddev->pers = pers;
1da177e4
LT		 2061 spin_unlock(&pers_lock);
2062
657390d2 2063 mddev->recovery = 0;
1da177e4 2064 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 2065 mddev->barriers_work = 1;
6ff8d8ec 2066 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 2067
f91de92e
N		 2068 if (start_readonly)
2069 mddev->ro = 2; /* read-only, but switch on first write */
2070
b15c2e57
N		 2071 err = mddev->pers->run(mddev);
2072 if (!err && mddev->pers->sync_request) {
2073 err = bitmap_create(mddev);
2074 if (err) {
2075 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2076 mdname(mddev), err);
2077 mddev->pers->stop(mddev);
2078 }
2079 }
1da177e4
LT		 2080 if (err) {
2081 printk(KERN_ERR "md: pers->run() failed ...\n");
2082 module_put(mddev->pers->owner);
2083 mddev->pers = NULL;
32a7627c
N		 2084 bitmap_destroy(mddev);
2085 return err;
1da177e4 2086 }
411036fa
N		 2087 if (mddev->pers->sync_request)
2088 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
fd9d49ca
N		 2089 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2090 mddev->ro = 0;
2091
1da177e4
LT		 2092 atomic_set(&mddev->writes_pending,0);
2093 mddev->safemode = 0;
2094 mddev->safemode_timer.function = md_safemode_timeout;
2095 mddev->safemode_timer.data = (unsigned long) mddev;
2096 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2097 mddev->in_sync = 1;
86e6ffdd
N		 2098
2099 ITERATE_RDEV(mddev,rdev,tmp)
2100 if (rdev->raid_disk >= 0) {
2101 char nm[20];
2102 sprintf(nm, "rd%d", rdev->raid_disk);
2103 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2104 }
1da177e4
LT		 2105
2106 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2107 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2108
2109 if (mddev->sb_dirty)
2110 md_update_sb(mddev);
2111
2112 set_capacity(disk, mddev->array_size<<1);
2113
2114 /* If we call blk_queue_make_request here, it will
2115 * re-initialise max_sectors etc which may have been
2116 * refined inside -> run. So just set the bits we need to set.
2117 * Most initialisation happended when we called
2118 * blk_queue_make_request(..., md_fail_request)
2119 * earlier.
2120 */
2121 mddev->queue->queuedata = mddev;
2122 mddev->queue->make_request_fn = mddev->pers->make_request;
2123
2124 mddev->changed = 1;
d7603b7e 2125 md_new_event(mddev);
1da177e4
LT		 2126 return 0;
2127}
2128
2129static int restart_array(mddev_t *mddev)
2130{
2131 struct gendisk *disk = mddev->gendisk;
2132 int err;
2133
2134 /*
2135 * Complain if it has no devices
2136 */
2137 err = -ENXIO;
2138 if (list_empty(&mddev->disks))
2139 goto out;
2140
2141 if (mddev->pers) {
2142 err = -EBUSY;
2143 if (!mddev->ro)
2144 goto out;
2145
2146 mddev->safemode = 0;
2147 mddev->ro = 0;
2148 set_disk_ro(disk, 0);
2149
2150 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2151 mdname(mddev));
2152 /*
2153 * Kick recovery or resync if necessary
2154 */
2155 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2156 md_wakeup_thread(mddev->thread);
2157 err = 0;
2158 } else {
2159 printk(KERN_ERR "md: %s has no personality assigned.\n",
2160 mdname(mddev));
2161 err = -EINVAL;
2162 }
2163
2164out:
2165 return err;
2166}
2167
2168static int do_md_stop(mddev_t * mddev, int ro)
2169{
2170 int err = 0;
2171 struct gendisk *disk = mddev->gendisk;
2172
2173 if (mddev->pers) {
2174 if (atomic_read(&mddev->active)>2) {
2175 printk("md: %s still in use.\n",mdname(mddev));
2176 return -EBUSY;
2177 }
2178
2179 if (mddev->sync_thread) {
2180 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2181 md_unregister_thread(mddev->sync_thread);
2182 mddev->sync_thread = NULL;
2183 }
2184
2185 del_timer_sync(&mddev->safemode_timer);
2186
2187 invalidate_partition(disk, 0);
2188
2189 if (ro) {
2190 err = -ENXIO;
f91de92e 2191 if (mddev->ro==1)
1da177e4
LT		 2192 goto out;
2193 mddev->ro = 1;
2194 } else {
6b8b3e8a 2195 bitmap_flush(mddev);
a9701a30 2196 md_super_wait(mddev);
1da177e4
LT		 2197 if (mddev->ro)
2198 set_disk_ro(disk, 0);
2199 blk_queue_make_request(mddev->queue, md_fail_request);
2200 mddev->pers->stop(mddev);
411036fa
N		 2201 if (mddev->pers->sync_request)
2202 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2203
1da177e4
LT		 2204 module_put(mddev->pers->owner);
2205 mddev->pers = NULL;
2206 if (mddev->ro)
2207 mddev->ro = 0;
2208 }
2209 if (!mddev->in_sync) {
2210 /* mark array as shutdown cleanly */
2211 mddev->in_sync = 1;
2212 md_update_sb(mddev);
2213 }
2214 if (ro)
2215 set_disk_ro(disk, 1);
2216 }
32a7627c
N		 2217
2218 bitmap_destroy(mddev);
2219 if (mddev->bitmap_file) {
2220 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2221 fput(mddev->bitmap_file);
2222 mddev->bitmap_file = NULL;
2223 }
9223214e 2224 mddev->bitmap_offset = 0;
32a7627c 2225
1da177e4
LT		 2226 /*
2227 * Free resources if final stop
2228 */
2229 if (!ro) {
86e6ffdd
N		 2230 mdk_rdev_t *rdev;
2231 struct list_head *tmp;
1da177e4
LT		 2232 struct gendisk *disk;
2233 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2234
86e6ffdd
N		 2235 ITERATE_RDEV(mddev,rdev,tmp)
2236 if (rdev->raid_disk >= 0) {
2237 char nm[20];
2238 sprintf(nm, "rd%d", rdev->raid_disk);
2239 sysfs_remove_link(&mddev->kobj, nm);
2240 }
2241
1da177e4
LT		 2242 export_array(mddev);
2243
2244 mddev->array_size = 0;
2245 disk = mddev->gendisk;
2246 if (disk)
2247 set_capacity(disk, 0);
2248 mddev->changed = 1;
2249 } else
2250 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2251 mdname(mddev));
2252 err = 0;
d7603b7e 2253 md_new_event(mddev);
1da177e4
LT		 2254out:
2255 return err;
2256}
2257
2258static void autorun_array(mddev_t *mddev)
2259{
2260 mdk_rdev_t *rdev;
2261 struct list_head *tmp;
2262 int err;
2263
a757e64c 2264 if (list_empty(&mddev->disks))
1da177e4 2265 return;
1da177e4
LT		 2266
2267 printk(KERN_INFO "md: running: ");
2268
2269 ITERATE_RDEV(mddev,rdev,tmp) {
2270 char b[BDEVNAME_SIZE];
2271 printk("<%s>", bdevname(rdev->bdev,b));
2272 }
2273 printk("\n");
2274
2275 err = do_md_run (mddev);
2276 if (err) {
2277 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2278 do_md_stop (mddev, 0);
2279 }
2280}
2281
2282/*
2283 * lets try to run arrays based on all disks that have arrived
2284 * until now. (those are in pending_raid_disks)
2285 *
2286 * the method: pick the first pending disk, collect all disks with
2287 * the same UUID, remove all from the pending list and put them into
2288 * the 'same_array' list. Then order this list based on superblock
2289 * update time (freshest comes first), kick out 'old' disks and
2290 * compare superblocks. If everything's fine then run it.
2291 *
2292 * If "unit" is allocated, then bump its reference count
2293 */
2294static void autorun_devices(int part)
2295{
2296 struct list_head candidates;
2297 struct list_head *tmp;
2298 mdk_rdev_t *rdev0, *rdev;
2299 mddev_t *mddev;
2300 char b[BDEVNAME_SIZE];
2301
2302 printk(KERN_INFO "md: autorun ...\n");
2303 while (!list_empty(&pending_raid_disks)) {
2304 dev_t dev;
2305 rdev0 = list_entry(pending_raid_disks.next,
2306 mdk_rdev_t, same_set);
2307
2308 printk(KERN_INFO "md: considering %s ...\n",
2309 bdevname(rdev0->bdev,b));
2310 INIT_LIST_HEAD(&candidates);
2311 ITERATE_RDEV_PENDING(rdev,tmp)
2312 if (super_90_load(rdev, rdev0, 0) >= 0) {
2313 printk(KERN_INFO "md: adding %s ...\n",
2314 bdevname(rdev->bdev,b));
2315 list_move(&rdev->same_set, &candidates);
2316 }
2317 /*
2318 * now we have a set of devices, with all of them having
2319 * mostly sane superblocks. It's time to allocate the
2320 * mddev.
2321 */
2322 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2323 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2324 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2325 break;
2326 }
2327 if (part)
2328 dev = MKDEV(mdp_major,
2329 rdev0->preferred_minor << MdpMinorShift);
2330 else
2331 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2332
2333 md_probe(dev, NULL, NULL);
2334 mddev = mddev_find(dev);
2335 if (!mddev) {
2336 printk(KERN_ERR
2337 "md: cannot allocate memory for md drive.\n");
2338 break;
2339 }
2340 if (mddev_lock(mddev))
2341 printk(KERN_WARNING "md: %s locked, cannot run\n",
2342 mdname(mddev));
2343 else if (mddev->raid_disks || mddev->major_version
2344 || !list_empty(&mddev->disks)) {
2345 printk(KERN_WARNING
2346 "md: %s already running, cannot run %s\n",
2347 mdname(mddev), bdevname(rdev0->bdev,b));
2348 mddev_unlock(mddev);
2349 } else {
2350 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2351 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2352 list_del_init(&rdev->same_set);
2353 if (bind_rdev_to_array(rdev, mddev))
2354 export_rdev(rdev);
2355 }
2356 autorun_array(mddev);
2357 mddev_unlock(mddev);
2358 }
2359 /* on success, candidates will be empty, on error
2360 * it won't...
2361 */
2362 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2363 export_rdev(rdev);
2364 mddev_put(mddev);
2365 }
2366 printk(KERN_INFO "md: ... autorun DONE.\n");
2367}
2368
2369/*
2370 * import RAID devices based on one partition
2371 * if possible, the array gets run as well.
2372 */
2373
2374static int autostart_array(dev_t startdev)
2375{
2376 char b[BDEVNAME_SIZE];
2377 int err = -EINVAL, i;
2378 mdp_super_t *sb = NULL;
2379 mdk_rdev_t *start_rdev = NULL, *rdev;
2380
2381 start_rdev = md_import_device(startdev, 0, 0);
2382 if (IS_ERR(start_rdev))
2383 return err;
2384
2385
2386 /* NOTE: this can only work for 0.90.0 superblocks */
2387 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2388 if (sb->major_version != 0 ||
2389 sb->minor_version != 90 ) {
2390 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2391 export_rdev(start_rdev);
2392 return err;
2393 }
2394
b2d444d7 2395 if (test_bit(Faulty, &start_rdev->flags)) {
1da177e4
LT		 2396 printk(KERN_WARNING
2397 "md: can not autostart based on faulty %s!\n",
2398 bdevname(start_rdev->bdev,b));
2399 export_rdev(start_rdev);
2400 return err;
2401 }
2402 list_add(&start_rdev->same_set, &pending_raid_disks);
2403
2404 for (i = 0; i < MD_SB_DISKS; i++) {
2405 mdp_disk_t *desc = sb->disks + i;
2406 dev_t dev = MKDEV(desc->major, desc->minor);
2407
2408 if (!dev)
2409 continue;
2410 if (dev == startdev)
2411 continue;
2412 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2413 continue;
2414 rdev = md_import_device(dev, 0, 0);
2415 if (IS_ERR(rdev))
2416 continue;
2417
2418 list_add(&rdev->same_set, &pending_raid_disks);
2419 }
2420
2421 /*
2422 * possibly return codes
2423 */
2424 autorun_devices(0);
2425 return 0;
2426
2427}
2428
2429
2430static int get_version(void __user * arg)
2431{
2432 mdu_version_t ver;
2433
2434 ver.major = MD_MAJOR_VERSION;
2435 ver.minor = MD_MINOR_VERSION;
2436 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2437
2438 if (copy_to_user(arg, &ver, sizeof(ver)))
2439 return -EFAULT;
2440
2441 return 0;
2442}
2443
2444static int get_array_info(mddev_t * mddev, void __user * arg)
2445{
2446 mdu_array_info_t info;
2447 int nr,working,active,failed,spare;
2448 mdk_rdev_t *rdev;
2449 struct list_head *tmp;
2450
2451 nr=working=active=failed=spare=0;
2452 ITERATE_RDEV(mddev,rdev,tmp) {
2453 nr++;
b2d444d7 2454 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2455 failed++;
2456 else {
2457 working++;
b2d444d7 2458 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 2459 active++;
2460 else
2461 spare++;
2462 }
2463 }
2464
2465 info.major_version = mddev->major_version;
2466 info.minor_version = mddev->minor_version;
2467 info.patch_version = MD_PATCHLEVEL_VERSION;
2468 info.ctime = mddev->ctime;
2469 info.level = mddev->level;
2470 info.size = mddev->size;
2471 info.nr_disks = nr;
2472 info.raid_disks = mddev->raid_disks;
2473 info.md_minor = mddev->md_minor;
2474 info.not_persistent= !mddev->persistent;
2475
2476 info.utime = mddev->utime;
2477 info.state = 0;
2478 if (mddev->in_sync)
2479 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 2480 if (mddev->bitmap && mddev->bitmap_offset)
2481 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 2482 info.active_disks = active;
2483 info.working_disks = working;
2484 info.failed_disks = failed;
2485 info.spare_disks = spare;
2486
2487 info.layout = mddev->layout;
2488 info.chunk_size = mddev->chunk_size;
2489
2490 if (copy_to_user(arg, &info, sizeof(info)))
2491 return -EFAULT;
2492
2493 return 0;
2494}
2495
87162a28 2496static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 2497{
2498 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2499 char *ptr, *buf = NULL;
2500 int err = -ENOMEM;
2501
2502 file = kmalloc(sizeof(*file), GFP_KERNEL);
2503 if (!file)
2504 goto out;
2505
2506 /* bitmap disabled, zero the first byte and copy out */
2507 if (!mddev->bitmap || !mddev->bitmap->file) {
2508 file->pathname[0] = '\0';
2509 goto copy_out;
2510 }
2511
2512 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2513 if (!buf)
2514 goto out;
2515
2516 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2517 if (!ptr)
2518 goto out;
2519
2520 strcpy(file->pathname, ptr);
2521
2522copy_out:
2523 err = 0;
2524 if (copy_to_user(arg, file, sizeof(*file)))
2525 err = -EFAULT;
2526out:
2527 kfree(buf);
2528 kfree(file);
2529 return err;
2530}
2531
1da177e4
LT		 2532static int get_disk_info(mddev_t * mddev, void __user * arg)
2533{
2534 mdu_disk_info_t info;
2535 unsigned int nr;
2536 mdk_rdev_t *rdev;
2537
2538 if (copy_from_user(&info, arg, sizeof(info)))
2539 return -EFAULT;
2540
2541 nr = info.number;
2542
2543 rdev = find_rdev_nr(mddev, nr);
2544 if (rdev) {
2545 info.major = MAJOR(rdev->bdev->bd_dev);
2546 info.minor = MINOR(rdev->bdev->bd_dev);
2547 info.raid_disk = rdev->raid_disk;
2548 info.state = 0;
b2d444d7 2549 if (test_bit(Faulty, &rdev->flags))
1da177e4 2550 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 2551 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 2552 info.state |= (1<<MD_DISK_ACTIVE);
2553 info.state |= (1<<MD_DISK_SYNC);
2554 }
8ddf9efe
N		 2555 if (test_bit(WriteMostly, &rdev->flags))
2556 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 2557 } else {
2558 info.major = info.minor = 0;
2559 info.raid_disk = -1;
2560 info.state = (1<<MD_DISK_REMOVED);
2561 }
2562
2563 if (copy_to_user(arg, &info, sizeof(info)))
2564 return -EFAULT;
2565
2566 return 0;
2567}
2568
2569static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2570{
2571 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2572 mdk_rdev_t *rdev;
2573 dev_t dev = MKDEV(info->major,info->minor);
2574
2575 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2576 return -EOVERFLOW;
2577
2578 if (!mddev->raid_disks) {
2579 int err;
2580 /* expecting a device which has a superblock */
2581 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2582 if (IS_ERR(rdev)) {
2583 printk(KERN_WARNING
2584 "md: md_import_device returned %ld\n",
2585 PTR_ERR(rdev));
2586 return PTR_ERR(rdev);
2587 }
2588 if (!list_empty(&mddev->disks)) {
2589 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2590 mdk_rdev_t, same_set);
2591 int err = super_types[mddev->major_version]
2592 .load_super(rdev, rdev0, mddev->minor_version);
2593 if (err < 0) {
2594 printk(KERN_WARNING
2595 "md: %s has different UUID to %s\n",
2596 bdevname(rdev->bdev,b),
2597 bdevname(rdev0->bdev,b2));
2598 export_rdev(rdev);
2599 return -EINVAL;
2600 }
2601 }
2602 err = bind_rdev_to_array(rdev, mddev);
2603 if (err)
2604 export_rdev(rdev);
2605 return err;
2606 }
2607
2608 /*
2609 * add_new_disk can be used once the array is assembled
2610 * to add "hot spares". They must already have a superblock
2611 * written
2612 */
2613 if (mddev->pers) {
2614 int err;
2615 if (!mddev->pers->hot_add_disk) {
2616 printk(KERN_WARNING
2617 "%s: personality does not support diskops!\n",
2618 mdname(mddev));
2619 return -EINVAL;
2620 }
7b1e35f6
N		 2621 if (mddev->persistent)
2622 rdev = md_import_device(dev, mddev->major_version,
2623 mddev->minor_version);
2624 else
2625 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 2626 if (IS_ERR(rdev)) {
2627 printk(KERN_WARNING
2628 "md: md_import_device returned %ld\n",
2629 PTR_ERR(rdev));
2630 return PTR_ERR(rdev);
2631 }
41158c7e
N		 2632 /* set save_raid_disk if appropriate */
2633 if (!mddev->persistent) {
2634 if (info->state & (1<<MD_DISK_SYNC) &&
2635 info->raid_disk < mddev->raid_disks)
2636 rdev->raid_disk = info->raid_disk;
2637 else
2638 rdev->raid_disk = -1;
2639 } else
2640 super_types[mddev->major_version].
2641 validate_super(mddev, rdev);
2642 rdev->saved_raid_disk = rdev->raid_disk;
2643
b2d444d7 2644 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 2645 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2646 set_bit(WriteMostly, &rdev->flags);
2647
1da177e4
LT		 2648 rdev->raid_disk = -1;
2649 err = bind_rdev_to_array(rdev, mddev);
2650 if (err)
2651 export_rdev(rdev);
c361777f
N		 2652
2653 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2654 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2655 return err;
2656 }
2657
2658 /* otherwise, add_new_disk is only allowed
2659 * for major_version==0 superblocks
2660 */
2661 if (mddev->major_version != 0) {
2662 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2663 mdname(mddev));
2664 return -EINVAL;
2665 }
2666
2667 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2668 int err;
2669 rdev = md_import_device (dev, -1, 0);
2670 if (IS_ERR(rdev)) {
2671 printk(KERN_WARNING
2672 "md: error, md_import_device() returned %ld\n",
2673 PTR_ERR(rdev));
2674 return PTR_ERR(rdev);
2675 }
2676 rdev->desc_nr = info->number;
2677 if (info->raid_disk < mddev->raid_disks)
2678 rdev->raid_disk = info->raid_disk;
2679 else
2680 rdev->raid_disk = -1;
2681
b2d444d7
N		 2682 rdev->flags = 0;
2683
1da177e4 2684 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 2685 if (info->state & (1<<MD_DISK_SYNC))
2686 set_bit(In_sync, &rdev->flags);
1da177e4 2687
8ddf9efe
N		 2688 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2689 set_bit(WriteMostly, &rdev->flags);
2690
1da177e4
LT		 2691 err = bind_rdev_to_array(rdev, mddev);
2692 if (err) {
2693 export_rdev(rdev);
2694 return err;
2695 }
2696
2697 if (!mddev->persistent) {
2698 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2699 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2700 } else
2701 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2702 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2703
2704 if (!mddev->size || (mddev->size > rdev->size))
2705 mddev->size = rdev->size;
2706 }
2707
2708 return 0;
2709}
2710
2711static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2712{
2713 char b[BDEVNAME_SIZE];
2714 mdk_rdev_t *rdev;
2715
2716 if (!mddev->pers)
2717 return -ENODEV;
2718
2719 rdev = find_rdev(mddev, dev);
2720 if (!rdev)
2721 return -ENXIO;
2722
2723 if (rdev->raid_disk >= 0)
2724 goto busy;
2725
2726 kick_rdev_from_array(rdev);
2727 md_update_sb(mddev);
d7603b7e 2728 md_new_event(mddev);
1da177e4
LT		 2729
2730 return 0;
2731busy:
2732 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2733 bdevname(rdev->bdev,b), mdname(mddev));
2734 return -EBUSY;
2735}
2736
2737static int hot_add_disk(mddev_t * mddev, dev_t dev)
2738{
2739 char b[BDEVNAME_SIZE];
2740 int err;
2741 unsigned int size;
2742 mdk_rdev_t *rdev;
2743
2744 if (!mddev->pers)
2745 return -ENODEV;
2746
2747 if (mddev->major_version != 0) {
2748 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2749 " version-0 superblocks.\n",
2750 mdname(mddev));
2751 return -EINVAL;
2752 }
2753 if (!mddev->pers->hot_add_disk) {
2754 printk(KERN_WARNING
2755 "%s: personality does not support diskops!\n",
2756 mdname(mddev));
2757 return -EINVAL;
2758 }
2759
2760 rdev = md_import_device (dev, -1, 0);
2761 if (IS_ERR(rdev)) {
2762 printk(KERN_WARNING
2763 "md: error, md_import_device() returned %ld\n",
2764 PTR_ERR(rdev));
2765 return -EINVAL;
2766 }
2767
2768 if (mddev->persistent)
2769 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2770 else
2771 rdev->sb_offset =
2772 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2773
2774 size = calc_dev_size(rdev, mddev->chunk_size);
2775 rdev->size = size;
2776
2777 if (size < mddev->size) {
2778 printk(KERN_WARNING
2779 "%s: disk size %llu blocks < array size %llu\n",
2780 mdname(mddev), (unsigned long long)size,
2781 (unsigned long long)mddev->size);
2782 err = -ENOSPC;
2783 goto abort_export;
2784 }
2785
b2d444d7 2786 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 2787 printk(KERN_WARNING
2788 "md: can not hot-add faulty %s disk to %s!\n",
2789 bdevname(rdev->bdev,b), mdname(mddev));
2790 err = -EINVAL;
2791 goto abort_export;
2792 }
b2d444d7 2793 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2794 rdev->desc_nr = -1;
2795 bind_rdev_to_array(rdev, mddev);
2796
2797 /*
2798 * The rest should better be atomic, we can have disk failures
2799 * noticed in interrupt contexts ...
2800 */
2801
2802 if (rdev->desc_nr == mddev->max_disks) {
2803 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2804 mdname(mddev));
2805 err = -EBUSY;
2806 goto abort_unbind_export;
2807 }
2808
2809 rdev->raid_disk = -1;
2810
2811 md_update_sb(mddev);
2812
2813 /*
2814 * Kick recovery, maybe this spare has to be added to the
2815 * array immediately.
2816 */
2817 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2818 md_wakeup_thread(mddev->thread);
d7603b7e 2819 md_new_event(mddev);
1da177e4
LT		 2820 return 0;
2821
2822abort_unbind_export:
2823 unbind_rdev_from_array(rdev);
2824
2825abort_export:
2826 export_rdev(rdev);
2827 return err;
2828}
2829
32a7627c
N		 2830/* similar to deny_write_access, but accounts for our holding a reference
2831 * to the file ourselves */
2832static int deny_bitmap_write_access(struct file * file)
2833{
2834 struct inode *inode = file->f_mapping->host;
2835
2836 spin_lock(&inode->i_lock);
2837 if (atomic_read(&inode->i_writecount) > 1) {
2838 spin_unlock(&inode->i_lock);
2839 return -ETXTBSY;
2840 }
2841 atomic_set(&inode->i_writecount, -1);
2842 spin_unlock(&inode->i_lock);
2843
2844 return 0;
2845}
2846
2847static int set_bitmap_file(mddev_t *mddev, int fd)
2848{
2849 int err;
2850
36fa3063
N		 2851 if (mddev->pers) {
2852 if (!mddev->pers->quiesce)
2853 return -EBUSY;
2854 if (mddev->recovery || mddev->sync_thread)
2855 return -EBUSY;
2856 /* we should be able to change the bitmap.. */
2857 }
32a7627c 2858
32a7627c 2859
36fa3063
N		 2860 if (fd >= 0) {
2861 if (mddev->bitmap)
2862 return -EEXIST; /* cannot add when bitmap is present */
2863 mddev->bitmap_file = fget(fd);
32a7627c 2864
36fa3063
N		 2865 if (mddev->bitmap_file == NULL) {
2866 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2867 mdname(mddev));
2868 return -EBADF;
2869 }
2870
2871 err = deny_bitmap_write_access(mddev->bitmap_file);
2872 if (err) {
2873 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2874 mdname(mddev));
2875 fput(mddev->bitmap_file);
2876 mddev->bitmap_file = NULL;
2877 return err;
2878 }
a654b9d8 2879 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 2880 } else if (mddev->bitmap == NULL)
2881 return -ENOENT; /* cannot remove what isn't there */
2882 err = 0;
2883 if (mddev->pers) {
2884 mddev->pers->quiesce(mddev, 1);
2885 if (fd >= 0)
2886 err = bitmap_create(mddev);
2887 if (fd < 0 || err)
2888 bitmap_destroy(mddev);
2889 mddev->pers->quiesce(mddev, 0);
2890 } else if (fd < 0) {
2891 if (mddev->bitmap_file)
2892 fput(mddev->bitmap_file);
2893 mddev->bitmap_file = NULL;
2894 }
2895
32a7627c
N		 2896 return err;
2897}
2898
1da177e4
LT		 2899/*
2900 * set_array_info is used two different ways
2901 * The original usage is when creating a new array.
2902 * In this usage, raid_disks is > 0 and it together with
2903 * level, size, not_persistent,layout,chunksize determine the
2904 * shape of the array.
2905 * This will always create an array with a type-0.90.0 superblock.
2906 * The newer usage is when assembling an array.
2907 * In this case raid_disks will be 0, and the major_version field is
2908 * use to determine which style super-blocks are to be found on the devices.
2909 * The minor and patch _version numbers are also kept incase the
2910 * super_block handler wishes to interpret them.
2911 */
2912static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2913{
2914
2915 if (info->raid_disks == 0) {
2916 /* just setting version number for superblock loading */
2917 if (info->major_version < 0 ||
2918 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2919 super_types[info->major_version].name == NULL) {
2920 /* maybe try to auto-load a module? */
2921 printk(KERN_INFO
2922 "md: superblock version %d not known\n",
2923 info->major_version);
2924 return -EINVAL;
2925 }
2926 mddev->major_version = info->major_version;
2927 mddev->minor_version = info->minor_version;
2928 mddev->patch_version = info->patch_version;
2929 return 0;
2930 }
2931 mddev->major_version = MD_MAJOR_VERSION;
2932 mddev->minor_version = MD_MINOR_VERSION;
2933 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2934 mddev->ctime = get_seconds();
2935
2936 mddev->level = info->level;
2937 mddev->size = info->size;
2938 mddev->raid_disks = info->raid_disks;
2939 /* don't set md_minor, it is determined by which /dev/md* was
2940 * openned
2941 */
2942 if (info->state & (1<<MD_SB_CLEAN))
2943 mddev->recovery_cp = MaxSector;
2944 else
2945 mddev->recovery_cp = 0;
2946 mddev->persistent = ! info->not_persistent;
2947
2948 mddev->layout = info->layout;
2949 mddev->chunk_size = info->chunk_size;
2950
2951 mddev->max_disks = MD_SB_DISKS;
2952
2953 mddev->sb_dirty = 1;
2954
b2a2703c
N		 2955 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2956 mddev->bitmap_offset = 0;
2957
1da177e4
LT		 2958 /*
2959 * Generate a 128 bit UUID
2960 */
2961 get_random_bytes(mddev->uuid, 16);
2962
2963 return 0;
2964}
2965
2966/*
2967 * update_array_info is used to change the configuration of an
2968 * on-line array.
2969 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2970 * fields in the info are checked against the array.
2971 * Any differences that cannot be handled will cause an error.
2972 * Normally, only one change can be managed at a time.
2973 */
2974static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2975{
2976 int rv = 0;
2977 int cnt = 0;
36fa3063
N		 2978 int state = 0;
2979
2980 /* calculate expected state,ignoring low bits */
2981 if (mddev->bitmap && mddev->bitmap_offset)
2982 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 2983
2984 if (mddev->major_version != info->major_version ||
2985 mddev->minor_version != info->minor_version ||
2986/* mddev->patch_version != info->patch_version || */
2987 mddev->ctime != info->ctime ||
2988 mddev->level != info->level ||
2989/* mddev->layout != info->layout || */
2990 !mddev->persistent != info->not_persistent||
36fa3063
N		 2991 mddev->chunk_size != info->chunk_size ||
2992 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2993 ((state^info->state) & 0xfffffe00)
2994 )
1da177e4
LT		 2995 return -EINVAL;
2996 /* Check there is only one change */
2997 if (mddev->size != info->size) cnt++;
2998 if (mddev->raid_disks != info->raid_disks) cnt++;
2999 if (mddev->layout != info->layout) cnt++;
36fa3063 3000 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 3001 if (cnt == 0) return 0;
3002 if (cnt > 1) return -EINVAL;
3003
3004 if (mddev->layout != info->layout) {
3005 /* Change layout
3006 * we don't need to do anything at the md level, the
3007 * personality will take care of it all.
3008 */
3009 if (mddev->pers->reconfig == NULL)
3010 return -EINVAL;
3011 else
3012 return mddev->pers->reconfig(mddev, info->layout, -1);
3013 }
3014 if (mddev->size != info->size) {
3015 mdk_rdev_t * rdev;
3016 struct list_head *tmp;
3017 if (mddev->pers->resize == NULL)
3018 return -EINVAL;
3019 /* The "size" is the amount of each device that is used.
3020 * This can only make sense for arrays with redundancy.
3021 * linear and raid0 always use whatever space is available
3022 * We can only consider changing the size if no resync
3023 * or reconstruction is happening, and if the new size
3024 * is acceptable. It must fit before the sb_offset or,
3025 * if that is <data_offset, it must fit before the
3026 * size of each device.
3027 * If size is zero, we find the largest size that fits.
3028 */
3029 if (mddev->sync_thread)
3030 return -EBUSY;
3031 ITERATE_RDEV(mddev,rdev,tmp) {
3032 sector_t avail;
3033 int fit = (info->size == 0);
3034 if (rdev->sb_offset > rdev->data_offset)
3035 avail = (rdev->sb_offset*2) - rdev->data_offset;
3036 else
3037 avail = get_capacity(rdev->bdev->bd_disk)
3038 - rdev->data_offset;
3039 if (fit && (info->size == 0 || info->size > avail/2))
3040 info->size = avail/2;
3041 if (avail < ((sector_t)info->size << 1))
3042 return -ENOSPC;
3043 }
3044 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3045 if (!rv) {
3046 struct block_device *bdev;
3047
3048 bdev = bdget_disk(mddev->gendisk, 0);
3049 if (bdev) {
3050 down(&bdev->bd_inode->i_sem);
3051 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3052 up(&bdev->bd_inode->i_sem);
3053 bdput(bdev);
3054 }
3055 }
3056 }
3057 if (mddev->raid_disks != info->raid_disks) {
3058 /* change the number of raid disks */
3059 if (mddev->pers->reshape == NULL)
3060 return -EINVAL;
3061 if (info->raid_disks <= 0 ||
3062 info->raid_disks >= mddev->max_disks)
3063 return -EINVAL;
3064 if (mddev->sync_thread)
3065 return -EBUSY;
3066 rv = mddev->pers->reshape(mddev, info->raid_disks);
3067 if (!rv) {
3068 struct block_device *bdev;
3069
3070 bdev = bdget_disk(mddev->gendisk, 0);
3071 if (bdev) {
3072 down(&bdev->bd_inode->i_sem);
3073 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3074 up(&bdev->bd_inode->i_sem);
3075 bdput(bdev);
3076 }
3077 }
3078 }
36fa3063
N		 3079 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3080 if (mddev->pers->quiesce == NULL)
3081 return -EINVAL;
3082 if (mddev->recovery || mddev->sync_thread)
3083 return -EBUSY;
3084 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3085 /* add the bitmap */
3086 if (mddev->bitmap)
3087 return -EEXIST;
3088 if (mddev->default_bitmap_offset == 0)
3089 return -EINVAL;
3090 mddev->bitmap_offset = mddev->default_bitmap_offset;
3091 mddev->pers->quiesce(mddev, 1);
3092 rv = bitmap_create(mddev);
3093 if (rv)
3094 bitmap_destroy(mddev);
3095 mddev->pers->quiesce(mddev, 0);
3096 } else {
3097 /* remove the bitmap */
3098 if (!mddev->bitmap)
3099 return -ENOENT;
3100 if (mddev->bitmap->file)
3101 return -EINVAL;
3102 mddev->pers->quiesce(mddev, 1);
3103 bitmap_destroy(mddev);
3104 mddev->pers->quiesce(mddev, 0);
3105 mddev->bitmap_offset = 0;
3106 }
3107 }
1da177e4
LT		 3108 md_update_sb(mddev);
3109 return rv;
3110}
3111
3112static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3113{
3114 mdk_rdev_t *rdev;
3115
3116 if (mddev->pers == NULL)
3117 return -ENODEV;
3118
3119 rdev = find_rdev(mddev, dev);
3120 if (!rdev)
3121 return -ENODEV;
3122
3123 md_error(mddev, rdev);
3124 return 0;
3125}
3126
3127static int md_ioctl(struct inode *inode, struct file *file,
3128 unsigned int cmd, unsigned long arg)
3129{
3130 int err = 0;
3131 void __user *argp = (void __user *)arg;
3132 struct hd_geometry __user *loc = argp;
3133 mddev_t *mddev = NULL;
3134
3135 if (!capable(CAP_SYS_ADMIN))
3136 return -EACCES;
3137
3138 /*
3139 * Commands dealing with the RAID driver but not any
3140 * particular array:
3141 */
3142 switch (cmd)
3143 {
3144 case RAID_VERSION:
3145 err = get_version(argp);
3146 goto done;
3147
3148 case PRINT_RAID_DEBUG:
3149 err = 0;
3150 md_print_devices();
3151 goto done;
3152
3153#ifndef MODULE
3154 case RAID_AUTORUN:
3155 err = 0;
3156 autostart_arrays(arg);
3157 goto done;
3158#endif
3159 default:;
3160 }
3161
3162 /*
3163 * Commands creating/starting a new array:
3164 */
3165
3166 mddev = inode->i_bdev->bd_disk->private_data;
3167
3168 if (!mddev) {
3169 BUG();
3170 goto abort;
3171 }
3172
3173
3174 if (cmd == START_ARRAY) {
3175 /* START_ARRAY doesn't need to lock the array as autostart_array
3176 * does the locking, and it could even be a different array
3177 */
3178 static int cnt = 3;
3179 if (cnt > 0 ) {
3180 printk(KERN_WARNING
3181 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
e8a00334 3182 "This will not be supported beyond July 2006\n",
1da177e4
LT		 3183 current->comm, current->pid);
3184 cnt--;
3185 }
3186 err = autostart_array(new_decode_dev(arg));
3187 if (err) {
3188 printk(KERN_WARNING "md: autostart failed!\n");
3189 goto abort;
3190 }
3191 goto done;
3192 }
3193
3194 err = mddev_lock(mddev);
3195 if (err) {
3196 printk(KERN_INFO
3197 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3198 err, cmd);
3199 goto abort;
3200 }
3201
3202 switch (cmd)
3203 {
3204 case SET_ARRAY_INFO:
3205 {
3206 mdu_array_info_t info;
3207 if (!arg)
3208 memset(&info, 0, sizeof(info));
3209 else if (copy_from_user(&info, argp, sizeof(info))) {
3210 err = -EFAULT;
3211 goto abort_unlock;
3212 }
3213 if (mddev->pers) {
3214 err = update_array_info(mddev, &info);
3215 if (err) {
3216 printk(KERN_WARNING "md: couldn't update"
3217 " array info. %d\n", err);
3218 goto abort_unlock;
3219 }
3220 goto done_unlock;
3221 }
3222 if (!list_empty(&mddev->disks)) {
3223 printk(KERN_WARNING
3224 "md: array %s already has disks!\n",
3225 mdname(mddev));
3226 err = -EBUSY;
3227 goto abort_unlock;
3228 }
3229 if (mddev->raid_disks) {
3230 printk(KERN_WARNING
3231 "md: array %s already initialised!\n",
3232 mdname(mddev));
3233 err = -EBUSY;
3234 goto abort_unlock;
3235 }
3236 err = set_array_info(mddev, &info);
3237 if (err) {
3238 printk(KERN_WARNING "md: couldn't set"
3239 " array info. %d\n", err);
3240 goto abort_unlock;
3241 }
3242 }
3243 goto done_unlock;
3244
3245 default:;
3246 }
3247
3248 /*
3249 * Commands querying/configuring an existing array:
3250 */
32a7627c
N		 3251 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3252 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3253 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3254 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 3255 err = -ENODEV;
3256 goto abort_unlock;
3257 }
3258
3259 /*
3260 * Commands even a read-only array can execute:
3261 */
3262 switch (cmd)
3263 {
3264 case GET_ARRAY_INFO:
3265 err = get_array_info(mddev, argp);
3266 goto done_unlock;
3267
32a7627c 3268 case GET_BITMAP_FILE:
87162a28 3269 err = get_bitmap_file(mddev, argp);
32a7627c
N		 3270 goto done_unlock;
3271
1da177e4
LT		 3272 case GET_DISK_INFO:
3273 err = get_disk_info(mddev, argp);
3274 goto done_unlock;
3275
3276 case RESTART_ARRAY_RW:
3277 err = restart_array(mddev);
3278 goto done_unlock;
3279
3280 case STOP_ARRAY:
3281 err = do_md_stop (mddev, 0);
3282 goto done_unlock;
3283
3284 case STOP_ARRAY_RO:
3285 err = do_md_stop (mddev, 1);
3286 goto done_unlock;
3287
3288 /*
3289 * We have a problem here : there is no easy way to give a CHS
3290 * virtual geometry. We currently pretend that we have a 2 heads
3291 * 4 sectors (with a BIG number of cylinders...). This drives
3292 * dosfs just mad... ;-)
3293 */
3294 case HDIO_GETGEO:
3295 if (!loc) {
3296 err = -EINVAL;
3297 goto abort_unlock;
3298 }
3299 err = put_user (2, (char __user *) &loc->heads);
3300 if (err)
3301 goto abort_unlock;
3302 err = put_user (4, (char __user *) &loc->sectors);
3303 if (err)
3304 goto abort_unlock;
3305 err = put_user(get_capacity(mddev->gendisk)/8,
3306 (short __user *) &loc->cylinders);
3307 if (err)
3308 goto abort_unlock;
3309 err = put_user (get_start_sect(inode->i_bdev),
3310 (long __user *) &loc->start);
3311 goto done_unlock;
3312 }
3313
3314 /*
3315 * The remaining ioctls are changing the state of the
f91de92e
N		 3316 * superblock, so we do not allow them on read-only arrays.
3317 * However non-MD ioctls (e.g. get-size) will still come through
3318 * here and hit the 'default' below, so only disallow
3319 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 3320 */
f91de92e
N		 3321 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3322 mddev->ro && mddev->pers) {
3323 if (mddev->ro == 2) {
3324 mddev->ro = 0;
3325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3326 md_wakeup_thread(mddev->thread);
3327
3328 } else {
3329 err = -EROFS;
3330 goto abort_unlock;
3331 }
1da177e4
LT		 3332 }
3333
3334 switch (cmd)
3335 {
3336 case ADD_NEW_DISK:
3337 {
3338 mdu_disk_info_t info;
3339 if (copy_from_user(&info, argp, sizeof(info)))
3340 err = -EFAULT;
3341 else
3342 err = add_new_disk(mddev, &info);
3343 goto done_unlock;
3344 }
3345
3346 case HOT_REMOVE_DISK:
3347 err = hot_remove_disk(mddev, new_decode_dev(arg));
3348 goto done_unlock;
3349
3350 case HOT_ADD_DISK:
3351 err = hot_add_disk(mddev, new_decode_dev(arg));
3352 goto done_unlock;
3353
3354 case SET_DISK_FAULTY:
3355 err = set_disk_faulty(mddev, new_decode_dev(arg));
3356 goto done_unlock;
3357
3358 case RUN_ARRAY:
3359 err = do_md_run (mddev);
3360 goto done_unlock;
3361
32a7627c
N		 3362 case SET_BITMAP_FILE:
3363 err = set_bitmap_file(mddev, (int)arg);
3364 goto done_unlock;
3365
1da177e4
LT		 3366 default:
3367 if (_IOC_TYPE(cmd) == MD_MAJOR)
3368 printk(KERN_WARNING "md: %s(pid %d) used"
3369 " obsolete MD ioctl, upgrade your"
3370 " software to use new ictls.\n",
3371 current->comm, current->pid);
3372 err = -EINVAL;
3373 goto abort_unlock;
3374 }
3375
3376done_unlock:
3377abort_unlock:
3378 mddev_unlock(mddev);
3379
3380 return err;
3381done:
3382 if (err)
3383 MD_BUG();
3384abort:
3385 return err;
3386}
3387
3388static int md_open(struct inode *inode, struct file *file)
3389{
3390 /*
3391 * Succeed if we can lock the mddev, which confirms that
3392 * it isn't being stopped right now.
3393 */
3394 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3395 int err;
3396
3397 if ((err = mddev_lock(mddev)))
3398 goto out;
3399
3400 err = 0;
3401 mddev_get(mddev);
3402 mddev_unlock(mddev);
3403
3404 check_disk_change(inode->i_bdev);
3405 out:
3406 return err;
3407}
3408
3409static int md_release(struct inode *inode, struct file * file)
3410{
3411 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3412
3413 if (!mddev)
3414 BUG();
3415 mddev_put(mddev);
3416
3417 return 0;
3418}
3419
3420static int md_media_changed(struct gendisk *disk)
3421{
3422 mddev_t *mddev = disk->private_data;
3423
3424 return mddev->changed;
3425}
3426
3427static int md_revalidate(struct gendisk *disk)
3428{
3429 mddev_t *mddev = disk->private_data;
3430
3431 mddev->changed = 0;
3432 return 0;
3433}
3434static struct block_device_operations md_fops =
3435{
3436 .owner = THIS_MODULE,
3437 .open = md_open,
3438 .release = md_release,
3439 .ioctl = md_ioctl,
3440 .media_changed = md_media_changed,
3441 .revalidate_disk= md_revalidate,
3442};
3443
75c96f85 3444static int md_thread(void * arg)
1da177e4
LT		 3445{
3446 mdk_thread_t *thread = arg;
3447
1da177e4
LT		 3448 /*
3449 * md_thread is a 'system-thread', it's priority should be very
3450 * high. We avoid resource deadlocks individually in each
3451 * raid personality. (RAID5 does preallocation) We also use RR and
3452 * the very same RT priority as kswapd, thus we will never get
3453 * into a priority inversion deadlock.
3454 *
3455 * we definitely have to have equal or higher priority than
3456 * bdflush, otherwise bdflush will deadlock if there are too
3457 * many dirty RAID5 blocks.
3458 */
1da177e4 3459
6985c43f 3460 allow_signal(SIGKILL);
a6fb0934 3461 while (!kthread_should_stop()) {
1da177e4 3462
93588e22
N		 3463 /* We need to wait INTERRUPTIBLE so that
3464 * we don't add to the load-average.
3465 * That means we need to be sure no signals are
3466 * pending
3467 */
3468 if (signal_pending(current))
3469 flush_signals(current);
3470
3471 wait_event_interruptible_timeout
3472 (thread->wqueue,
3473 test_bit(THREAD_WAKEUP, &thread->flags)
3474 || kthread_should_stop(),
3475 thread->timeout);
3e1d1d28 3476 try_to_freeze();
1da177e4
LT		 3477
3478 clear_bit(THREAD_WAKEUP, &thread->flags);
3479
787453c2 3480 thread->run(thread->mddev);
1da177e4 3481 }
a6fb0934 3482
1da177e4
LT		 3483 return 0;
3484}
3485
3486void md_wakeup_thread(mdk_thread_t *thread)
3487{
3488 if (thread) {
3489 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3490 set_bit(THREAD_WAKEUP, &thread->flags);
3491 wake_up(&thread->wqueue);
3492 }
3493}
3494
3495mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3496 const char *name)
3497{
3498 mdk_thread_t *thread;
1da177e4 3499
9ffae0cf 3500 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 3501 if (!thread)
3502 return NULL;
3503
1da177e4
LT		 3504 init_waitqueue_head(&thread->wqueue);
3505
1da177e4
LT		 3506 thread->run = run;
3507 thread->mddev = mddev;
32a7627c 3508 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 3509 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 3510 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 3511 kfree(thread);
3512 return NULL;
3513 }
1da177e4
LT		 3514 return thread;
3515}
3516
1da177e4
LT		 3517void md_unregister_thread(mdk_thread_t *thread)
3518{
d28446fe 3519 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 3520
3521 kthread_stop(thread->tsk);
1da177e4
LT		 3522 kfree(thread);
3523}
3524
3525void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3526{
3527 if (!mddev) {
3528 MD_BUG();
3529 return;
3530 }
3531
b2d444d7 3532 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 3533 return;
32a7627c 3534/*
1da177e4
LT		 3535 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3536 mdname(mddev),
3537 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3538 __builtin_return_address(0),__builtin_return_address(1),
3539 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 3540*/
1da177e4
LT		 3541 if (!mddev->pers->error_handler)
3542 return;
3543 mddev->pers->error_handler(mddev,rdev);
3544 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3545 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3546 md_wakeup_thread(mddev->thread);
d7603b7e 3547 md_new_event(mddev);
1da177e4
LT		 3548}
3549
3550/* seq_file implementation /proc/mdstat */
3551
3552static void status_unused(struct seq_file *seq)
3553{
3554 int i = 0;
3555 mdk_rdev_t *rdev;
3556 struct list_head *tmp;
3557
3558 seq_printf(seq, "unused devices: ");
3559
3560 ITERATE_RDEV_PENDING(rdev,tmp) {
3561 char b[BDEVNAME_SIZE];
3562 i++;
3563 seq_printf(seq, "%s ",
3564 bdevname(rdev->bdev,b));
3565 }
3566 if (!i)
3567 seq_printf(seq, "<none>");
3568
3569 seq_printf(seq, "\n");
3570}
3571
3572
3573static void status_resync(struct seq_file *seq, mddev_t * mddev)
3574{
3575 unsigned long max_blocks, resync, res, dt, db, rt;
3576
3577 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3578
3579 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3580 max_blocks = mddev->resync_max_sectors >> 1;
3581 else
3582 max_blocks = mddev->size;
3583
3584 /*
3585 * Should not happen.
3586 */
3587 if (!max_blocks) {
3588 MD_BUG();
3589 return;
3590 }
3591 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3592 {
3593 int i, x = res/50, y = 20-x;
3594 seq_printf(seq, "[");
3595 for (i = 0; i < x; i++)
3596 seq_printf(seq, "=");
3597 seq_printf(seq, ">");
3598 for (i = 0; i < y; i++)
3599 seq_printf(seq, ".");
3600 seq_printf(seq, "] ");
3601 }
3602 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3603 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3604 "resync" : "recovery"),
3605 res/10, res % 10, resync, max_blocks);
3606
3607 /*
3608 * We do not want to overflow, so the order of operands and
3609 * the * 100 / 100 trick are important. We do a +1 to be
3610 * safe against division by zero. We only estimate anyway.
3611 *
3612 * dt: time from mark until now
3613 * db: blocks written from mark until now
3614 * rt: remaining time
3615 */
3616 dt = ((jiffies - mddev->resync_mark) / HZ);
3617 if (!dt) dt++;
3618 db = resync - (mddev->resync_mark_cnt/2);
3619 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3620
3621 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3622
3623 seq_printf(seq, " speed=%ldK/sec", db/dt);
3624}
3625
3626static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3627{
3628 struct list_head *tmp;
3629 loff_t l = *pos;
3630 mddev_t *mddev;
3631
3632 if (l >= 0x10000)
3633 return NULL;
3634 if (!l--)
3635 /* header */
3636 return (void*)1;
3637
3638 spin_lock(&all_mddevs_lock);
3639 list_for_each(tmp,&all_mddevs)
3640 if (!l--) {
3641 mddev = list_entry(tmp, mddev_t, all_mddevs);
3642 mddev_get(mddev);
3643 spin_unlock(&all_mddevs_lock);
3644 return mddev;
3645 }
3646 spin_unlock(&all_mddevs_lock);
3647 if (!l--)
3648 return (void*)2;/* tail */
3649 return NULL;
3650}
3651
3652static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3653{
3654 struct list_head *tmp;
3655 mddev_t *next_mddev, *mddev = v;
3656
3657 ++*pos;
3658 if (v == (void*)2)
3659 return NULL;
3660
3661 spin_lock(&all_mddevs_lock);
3662 if (v == (void*)1)
3663 tmp = all_mddevs.next;
3664 else
3665 tmp = mddev->all_mddevs.next;
3666 if (tmp != &all_mddevs)
3667 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3668 else {
3669 next_mddev = (void*)2;
3670 *pos = 0x10000;
3671 }
3672 spin_unlock(&all_mddevs_lock);
3673
3674 if (v != (void*)1)
3675 mddev_put(mddev);
3676 return next_mddev;
3677
3678}
3679
3680static void md_seq_stop(struct seq_file *seq, void *v)
3681{
3682 mddev_t *mddev = v;
3683
3684 if (mddev && v != (void*)1 && v != (void*)2)
3685 mddev_put(mddev);
3686}
3687
d7603b7e
N		 3688struct mdstat_info {
3689 int event;
3690};
3691
1da177e4
LT		 3692static int md_seq_show(struct seq_file *seq, void *v)
3693{
3694 mddev_t *mddev = v;
3695 sector_t size;
3696 struct list_head *tmp2;
3697 mdk_rdev_t *rdev;
d7603b7e 3698 struct mdstat_info *mi = seq->private;
32a7627c 3699 struct bitmap *bitmap;
1da177e4
LT		 3700
3701 if (v == (void*)1) {
2604b703 3702 struct mdk_personality *pers;
1da177e4
LT		 3703 seq_printf(seq, "Personalities : ");
3704 spin_lock(&pers_lock);
2604b703
N		 3705 list_for_each_entry(pers, &pers_list, list)
3706 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 3707
3708 spin_unlock(&pers_lock);
3709 seq_printf(seq, "\n");
d7603b7e 3710 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 3711 return 0;
3712 }
3713 if (v == (void*)2) {
3714 status_unused(seq);
3715 return 0;
3716 }
3717
3718 if (mddev_lock(mddev)!=0)
3719 return -EINTR;
3720 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3721 seq_printf(seq, "%s : %sactive", mdname(mddev),
3722 mddev->pers ? "" : "in");
3723 if (mddev->pers) {
f91de92e 3724 if (mddev->ro==1)
1da177e4 3725 seq_printf(seq, " (read-only)");
f91de92e
N		 3726 if (mddev->ro==2)
3727 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 3728 seq_printf(seq, " %s", mddev->pers->name);
3729 }
3730
3731 size = 0;
3732 ITERATE_RDEV(mddev,rdev,tmp2) {
3733 char b[BDEVNAME_SIZE];
3734 seq_printf(seq, " %s[%d]",
3735 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 3736 if (test_bit(WriteMostly, &rdev->flags))
3737 seq_printf(seq, "(W)");
b2d444d7 3738 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3739 seq_printf(seq, "(F)");
3740 continue;
b325a32e
N		 3741 } else if (rdev->raid_disk < 0)
3742 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 3743 size += rdev->size;
3744 }
3745
3746 if (!list_empty(&mddev->disks)) {
3747 if (mddev->pers)
3748 seq_printf(seq, "\n %llu blocks",
3749 (unsigned long long)mddev->array_size);
3750 else
3751 seq_printf(seq, "\n %llu blocks",
3752 (unsigned long long)size);
3753 }
1cd6bf19
N		 3754 if (mddev->persistent) {
3755 if (mddev->major_version != 0 ||
3756 mddev->minor_version != 90) {
3757 seq_printf(seq," super %d.%d",
3758 mddev->major_version,
3759 mddev->minor_version);
3760 }
3761 } else
3762 seq_printf(seq, " super non-persistent");
1da177e4
LT		 3763
3764 if (mddev->pers) {
3765 mddev->pers->status (seq, mddev);
3766 seq_printf(seq, "\n ");
8e1b39d6
N		 3767 if (mddev->pers->sync_request) {
3768 if (mddev->curr_resync > 2) {
3769 status_resync (seq, mddev);
3770 seq_printf(seq, "\n ");
3771 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3772 seq_printf(seq, "\tresync=DELAYED\n ");
3773 else if (mddev->recovery_cp < MaxSector)
3774 seq_printf(seq, "\tresync=PENDING\n ");
3775 }
32a7627c
N		 3776 } else
3777 seq_printf(seq, "\n ");
3778
3779 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 3780 unsigned long chunk_kb;
3781 unsigned long flags;
32a7627c
N		 3782 spin_lock_irqsave(&bitmap->lock, flags);
3783 chunk_kb = bitmap->chunksize >> 10;
3784 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3785 "%lu%s chunk",
3786 bitmap->pages - bitmap->missing_pages,
3787 bitmap->pages,
3788 (bitmap->pages - bitmap->missing_pages)
3789 << (PAGE_SHIFT - 10),
3790 chunk_kb ? chunk_kb : bitmap->chunksize,
3791 chunk_kb ? "KB" : "B");
78d742d8
N		 3792 if (bitmap->file) {
3793 seq_printf(seq, ", file: ");
3794 seq_path(seq, bitmap->file->f_vfsmnt,
3795 bitmap->file->f_dentry," \t\n");
32a7627c 3796 }
78d742d8 3797
32a7627c
N		 3798 seq_printf(seq, "\n");
3799 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 3800 }
3801
3802 seq_printf(seq, "\n");
3803 }
3804 mddev_unlock(mddev);
3805
3806 return 0;
3807}
3808
3809static struct seq_operations md_seq_ops = {
3810 .start = md_seq_start,
3811 .next = md_seq_next,
3812 .stop = md_seq_stop,
3813 .show = md_seq_show,
3814};
3815
3816static int md_seq_open(struct inode *inode, struct file *file)
3817{
3818 int error;
d7603b7e
N		 3819 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3820 if (mi == NULL)
3821 return -ENOMEM;
1da177e4
LT		 3822
3823 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 3824 if (error)
3825 kfree(mi);
3826 else {
3827 struct seq_file *p = file->private_data;
3828 p->private = mi;
3829 mi->event = atomic_read(&md_event_count);
3830 }
1da177e4
LT		 3831 return error;
3832}
3833
d7603b7e
N		 3834static int md_seq_release(struct inode *inode, struct file *file)
3835{
3836 struct seq_file *m = file->private_data;
3837 struct mdstat_info *mi = m->private;
3838 m->private = NULL;
3839 kfree(mi);
3840 return seq_release(inode, file);
3841}
3842
3843static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
3844{
3845 struct seq_file *m = filp->private_data;
3846 struct mdstat_info *mi = m->private;
3847 int mask;
3848
3849 poll_wait(filp, &md_event_waiters, wait);
3850
3851 /* always allow read */
3852 mask = POLLIN | POLLRDNORM;
3853
3854 if (mi->event != atomic_read(&md_event_count))
3855 mask |= POLLERR | POLLPRI;
3856 return mask;
3857}
3858
1da177e4
LT		 3859static struct file_operations md_seq_fops = {
3860 .open = md_seq_open,
3861 .read = seq_read,
3862 .llseek = seq_lseek,
d7603b7e
N		 3863 .release = md_seq_release,
3864 .poll = mdstat_poll,
1da177e4
LT		 3865};
3866
2604b703 3867int register_md_personality(struct mdk_personality *p)
1da177e4 3868{
1da177e4 3869 spin_lock(&pers_lock);
2604b703
N		 3870 list_add_tail(&p->list, &pers_list);
3871 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 3872 spin_unlock(&pers_lock);
3873 return 0;
3874}
3875
2604b703 3876int unregister_md_personality(struct mdk_personality *p)
1da177e4 3877{
2604b703 3878 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 3879 spin_lock(&pers_lock);
2604b703 3880 list_del_init(&p->list);
1da177e4
LT		 3881 spin_unlock(&pers_lock);
3882 return 0;
3883}
3884
3885static int is_mddev_idle(mddev_t *mddev)
3886{
3887 mdk_rdev_t * rdev;
3888 struct list_head *tmp;
3889 int idle;
3890 unsigned long curr_events;
3891
3892 idle = 1;
3893 ITERATE_RDEV(mddev,rdev,tmp) {
3894 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 3895 curr_events = disk_stat_read(disk, sectors[0]) +
3896 disk_stat_read(disk, sectors[1]) -
1da177e4 3897 atomic_read(&disk->sync_io);
c0e48521
N		 3898 /* The difference between curr_events and last_events
3899 * will be affected by any new non-sync IO (making
3900 * curr_events bigger) and any difference in the amount of
3901 * in-flight syncio (making current_events bigger or smaller)
3902 * The amount in-flight is currently limited to
3903 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3904 * which is at most 4096 sectors.
3905 * These numbers are fairly fragile and should be made
3906 * more robust, probably by enforcing the
3907 * 'window size' that md_do_sync sort-of uses.
3908 *
1da177e4
LT		 3909 * Note: the following is an unsigned comparison.
3910 */
c0e48521 3911 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 3912 rdev->last_events = curr_events;
3913 idle = 0;
3914 }
3915 }
3916 return idle;
3917}
3918
3919void md_done_sync(mddev_t *mddev, int blocks, int ok)
3920{
3921 /* another "blocks" (512byte) blocks have been synced */
3922 atomic_sub(blocks, &mddev->recovery_active);
3923 wake_up(&mddev->recovery_wait);
3924 if (!ok) {
3925 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3926 md_wakeup_thread(mddev->thread);
3927 // stop recovery, signal do_sync ....
3928 }
3929}
3930
3931
06d91a5f
N		 3932/* md_write_start(mddev, bi)
3933 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 3934 * in superblock) before writing, schedule a superblock update
3935 * and wait for it to complete.
06d91a5f 3936 */
3d310eb7 3937void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 3938{
06d91a5f 3939 if (bio_data_dir(bi) != WRITE)
3d310eb7 3940 return;
06d91a5f 3941
f91de92e
N		 3942 BUG_ON(mddev->ro == 1);
3943 if (mddev->ro == 2) {
3944 /* need to switch to read/write */
3945 mddev->ro = 0;
3946 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3947 md_wakeup_thread(mddev->thread);
3948 }
06d91a5f 3949 atomic_inc(&mddev->writes_pending);
06d91a5f 3950 if (mddev->in_sync) {
a9701a30 3951 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 3952 if (mddev->in_sync) {
3953 mddev->in_sync = 0;
3954 mddev->sb_dirty = 1;
3955 md_wakeup_thread(mddev->thread);
3956 }
a9701a30 3957 spin_unlock_irq(&mddev->write_lock);
06d91a5f 3958 }
3d310eb7 3959 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
1da177e4
LT		 3960}
3961
3962void md_write_end(mddev_t *mddev)
3963{
3964 if (atomic_dec_and_test(&mddev->writes_pending)) {
3965 if (mddev->safemode == 2)
3966 md_wakeup_thread(mddev->thread);
3967 else
3968 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3969 }
3970}
3971
75c96f85 3972static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 3973
3974#define SYNC_MARKS 10
3975#define SYNC_MARK_STEP (3*HZ)
3976static void md_do_sync(mddev_t *mddev)
3977{
3978 mddev_t *mddev2;
3979 unsigned int currspeed = 0,
3980 window;
57afd89f 3981 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 3982 unsigned long mark[SYNC_MARKS];
3983 sector_t mark_cnt[SYNC_MARKS];
3984 int last_mark,m;
3985 struct list_head *tmp;
3986 sector_t last_check;
57afd89f 3987 int skipped = 0;
1da177e4
LT		 3988
3989 /* just incase thread restarts... */
3990 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3991 return;
3992
3993 /* we overload curr_resync somewhat here.
3994 * 0 == not engaged in resync at all
3995 * 2 == checking that there is no conflict with another sync
3996 * 1 == like 2, but have yielded to allow conflicting resync to
3997 * commense
3998 * other == active in resync - this many blocks
3999 *
4000 * Before starting a resync we must have set curr_resync to
4001 * 2, and then checked that every "conflicting" array has curr_resync
4002 * less than ours. When we find one that is the same or higher
4003 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4004 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4005 * This will mean we have to start checking from the beginning again.
4006 *
4007 */
4008
4009 do {
4010 mddev->curr_resync = 2;
4011
4012 try_again:
787453c2 4013 if (kthread_should_stop()) {
6985c43f 4014 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 4015 goto skip;
4016 }
4017 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 4018 if (mddev2 == mddev)
4019 continue;
4020 if (mddev2->curr_resync &&
4021 match_mddev_units(mddev,mddev2)) {
4022 DEFINE_WAIT(wq);
4023 if (mddev < mddev2 && mddev->curr_resync == 2) {
4024 /* arbitrarily yield */
4025 mddev->curr_resync = 1;
4026 wake_up(&resync_wait);
4027 }
4028 if (mddev > mddev2 && mddev->curr_resync == 1)
4029 /* no need to wait here, we can wait the next
4030 * time 'round when curr_resync == 2
4031 */
4032 continue;
787453c2
N		 4033 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4034 if (!kthread_should_stop() &&
8712e553 4035 mddev2->curr_resync >= mddev->curr_resync) {
1da177e4
LT		 4036 printk(KERN_INFO "md: delaying resync of %s"
4037 " until %s has finished resync (they"
4038 " share one or more physical units)\n",
4039 mdname(mddev), mdname(mddev2));
4040 mddev_put(mddev2);
4041 schedule();
4042 finish_wait(&resync_wait, &wq);
4043 goto try_again;
4044 }
4045 finish_wait(&resync_wait, &wq);
4046 }
4047 }
4048 } while (mddev->curr_resync < 2);
4049
9d88883e 4050 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 4051 /* resync follows the size requested by the personality,
57afd89f 4052 * which defaults to physical size, but can be virtual size
1da177e4
LT		 4053 */
4054 max_sectors = mddev->resync_max_sectors;
9d88883e
N		 4055 mddev->resync_mismatches = 0;
4056 } else
1da177e4
LT		 4057 /* recovery follows the physical size of devices */
4058 max_sectors = mddev->size << 1;
4059
4060 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4061 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4062 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
338cec32 4063 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
1da177e4
LT		 4064 "(but not more than %d KB/sec) for reconstruction.\n",
4065 sysctl_speed_limit_max);
4066
4067 is_mddev_idle(mddev); /* this also initializes IO event counters */
32a7627c 4068 /* we don't use the checkpoint if there's a bitmap */
24dd469d
N		 4069 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4070 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1da177e4
LT		 4071 j = mddev->recovery_cp;
4072 else
4073 j = 0;
57afd89f 4074 io_sectors = 0;
1da177e4
LT		 4075 for (m = 0; m < SYNC_MARKS; m++) {
4076 mark[m] = jiffies;
57afd89f 4077 mark_cnt[m] = io_sectors;
1da177e4
LT		 4078 }
4079 last_mark = 0;
4080 mddev->resync_mark = mark[last_mark];
4081 mddev->resync_mark_cnt = mark_cnt[last_mark];
4082
4083 /*
4084 * Tune reconstruction:
4085 */
4086 window = 32*(PAGE_SIZE/512);
4087 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4088 window/2,(unsigned long long) max_sectors/2);
4089
4090 atomic_set(&mddev->recovery_active, 0);
4091 init_waitqueue_head(&mddev->recovery_wait);
4092 last_check = 0;
4093
4094 if (j>2) {
4095 printk(KERN_INFO
4096 "md: resuming recovery of %s from checkpoint.\n",
4097 mdname(mddev));
4098 mddev->curr_resync = j;
4099 }
4100
4101 while (j < max_sectors) {
57afd89f 4102 sector_t sectors;
1da177e4 4103
57afd89f
N		 4104 skipped = 0;
4105 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4106 currspeed < sysctl_speed_limit_min);
4107 if (sectors == 0) {
1da177e4
LT		 4108 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4109 goto out;
4110 }
57afd89f
N		 4111
4112 if (!skipped) { /* actual IO requested */
4113 io_sectors += sectors;
4114 atomic_add(sectors, &mddev->recovery_active);
4115 }
4116
1da177e4
LT		 4117 j += sectors;
4118 if (j>1) mddev->curr_resync = j;
d7603b7e
N		 4119 if (last_check == 0)
4120 /* this is the earliers that rebuilt will be
4121 * visible in /proc/mdstat
4122 */
4123 md_new_event(mddev);
57afd89f
N		 4124
4125 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 4126 continue;
4127
57afd89f 4128 last_check = io_sectors;
1da177e4
LT		 4129
4130 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4131 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4132 break;
4133
4134 repeat:
4135 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4136 /* step marks */
4137 int next = (last_mark+1) % SYNC_MARKS;
4138
4139 mddev->resync_mark = mark[next];
4140 mddev->resync_mark_cnt = mark_cnt[next];
4141 mark[next] = jiffies;
57afd89f 4142 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 4143 last_mark = next;
4144 }
4145
4146
787453c2 4147 if (kthread_should_stop()) {
1da177e4
LT		 4148 /*
4149 * got a signal, exit.
4150 */
4151 printk(KERN_INFO
4152 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 4153 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4154 goto out;
4155 }
4156
4157 /*
4158 * this loop exits only if either when we are slower than
4159 * the 'hard' speed limit, or the system was IO-idle for
4160 * a jiffy.
4161 * the system might be non-idle CPU-wise, but we only care
4162 * about not overloading the IO subsystem. (things like an
4163 * e2fsck being done on the RAID array should execute fast)
4164 */
4165 mddev->queue->unplug_fn(mddev->queue);
4166 cond_resched();
4167
57afd89f
N		 4168 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4169 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4
LT		 4170
4171 if (currspeed > sysctl_speed_limit_min) {
4172 if ((currspeed > sysctl_speed_limit_max) ||
4173 !is_mddev_idle(mddev)) {
c0e48521 4174 msleep(500);
1da177e4
LT		 4175 goto repeat;
4176 }
4177 }
4178 }
4179 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4180 /*
4181 * this also signals 'finished resyncing' to md_stop
4182 */
4183 out:
4184 mddev->queue->unplug_fn(mddev->queue);
4185
4186 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4187
4188 /* tell personality that we are finished */
57afd89f 4189 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 4190
4191 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4192 mddev->curr_resync > 2 &&
4193 mddev->curr_resync >= mddev->recovery_cp) {
4194 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4195 printk(KERN_INFO
4196 "md: checkpointing recovery of %s.\n",
4197 mdname(mddev));
4198 mddev->recovery_cp = mddev->curr_resync;
4199 } else
4200 mddev->recovery_cp = MaxSector;
4201 }
4202
1da177e4
LT		 4203 skip:
4204 mddev->curr_resync = 0;
4205 wake_up(&resync_wait);
4206 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4207 md_wakeup_thread(mddev->thread);
4208}
4209
4210
4211/*
4212 * This routine is regularly called by all per-raid-array threads to
4213 * deal with generic issues like resync and super-block update.
4214 * Raid personalities that don't have a thread (linear/raid0) do not
4215 * need this as they never do any recovery or update the superblock.
4216 *
4217 * It does not do any resync itself, but rather "forks" off other threads
4218 * to do that as needed.
4219 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4220 * "->recovery" and create a thread at ->sync_thread.
4221 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4222 * and wakeups up this thread which will reap the thread and finish up.
4223 * This thread also removes any faulty devices (with nr_pending == 0).
4224 *
4225 * The overall approach is:
4226 * 1/ if the superblock needs updating, update it.
4227 * 2/ If a recovery thread is running, don't do anything else.
4228 * 3/ If recovery has finished, clean up, possibly marking spares active.
4229 * 4/ If there are any faulty devices, remove them.
4230 * 5/ If array is degraded, try to add spares devices
4231 * 6/ If array has spares or is not in-sync, start a resync thread.
4232 */
4233void md_check_recovery(mddev_t *mddev)
4234{
4235 mdk_rdev_t *rdev;
4236 struct list_head *rtmp;
4237
4238
5f40402d
N		 4239 if (mddev->bitmap)
4240 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 4241
4242 if (mddev->ro)
4243 return;
fca4d848
N		 4244
4245 if (signal_pending(current)) {
4246 if (mddev->pers->sync_request) {
4247 printk(KERN_INFO "md: %s in immediate safe mode\n",
4248 mdname(mddev));
4249 mddev->safemode = 2;
4250 }
4251 flush_signals(current);
4252 }
4253
1da177e4
LT		 4254 if ( ! (
4255 mddev->sb_dirty ||
4256 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 4257 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4258 (mddev->safemode == 1) ||
4259 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4260 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 4261 ))
4262 return;
fca4d848 4263
1da177e4
LT		 4264 if (mddev_trylock(mddev)==0) {
4265 int spares =0;
fca4d848 4266
a9701a30 4267 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 4268 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4269 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4270 mddev->in_sync = 1;
4271 mddev->sb_dirty = 1;
4272 }
4273 if (mddev->safemode == 1)
4274 mddev->safemode = 0;
a9701a30 4275 spin_unlock_irq(&mddev->write_lock);
fca4d848 4276
1da177e4
LT		 4277 if (mddev->sb_dirty)
4278 md_update_sb(mddev);
06d91a5f 4279
06d91a5f 4280
1da177e4
LT		 4281 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4282 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4283 /* resync/recovery still happening */
4284 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4285 goto unlock;
4286 }
4287 if (mddev->sync_thread) {
4288 /* resync has finished, collect result */
4289 md_unregister_thread(mddev->sync_thread);
4290 mddev->sync_thread = NULL;
4291 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4292 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4293 /* success...*/
4294 /* activate any spares */
4295 mddev->pers->spare_active(mddev);
4296 }
4297 md_update_sb(mddev);
41158c7e
N		 4298
4299 /* if array is no-longer degraded, then any saved_raid_disk
4300 * information must be scrapped
4301 */
4302 if (!mddev->degraded)
4303 ITERATE_RDEV(mddev,rdev,rtmp)
4304 rdev->saved_raid_disk = -1;
4305
1da177e4
LT		 4306 mddev->recovery = 0;
4307 /* flag recovery needed just to double check */
4308 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 4309 md_new_event(mddev);
1da177e4
LT		 4310 goto unlock;
4311 }
24dd469d
N		 4312 /* Clear some bits that don't mean anything, but
4313 * might be left set
4314 */
4315 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4316 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4317 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4318 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4
LT		 4319
4320 /* no recovery is running.
4321 * remove any failed drives, then
4322 * add spares if possible.
4323 * Spare are also removed and re-added, to allow
4324 * the personality to fail the re-add.
4325 */
4326 ITERATE_RDEV(mddev,rdev,rtmp)
4327 if (rdev->raid_disk >= 0 &&
b2d444d7 4328 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 4329 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 4330 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4331 char nm[20];
4332 sprintf(nm,"rd%d", rdev->raid_disk);
4333 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 4334 rdev->raid_disk = -1;
86e6ffdd 4335 }
1da177e4
LT		 4336 }
4337
4338 if (mddev->degraded) {
4339 ITERATE_RDEV(mddev,rdev,rtmp)
4340 if (rdev->raid_disk < 0
b2d444d7 4341 && !test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 4342 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4343 char nm[20];
4344 sprintf(nm, "rd%d", rdev->raid_disk);
4345 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 4346 spares++;
d7603b7e 4347 md_new_event(mddev);
86e6ffdd 4348 } else
1da177e4
LT		 4349 break;
4350 }
4351 }
4352
24dd469d
N		 4353 if (spares) {
4354 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4355 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4356 } else if (mddev->recovery_cp < MaxSector) {
4357 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4358 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4359 /* nothing to be done ... */
1da177e4 4360 goto unlock;
24dd469d 4361
1da177e4
LT		 4362 if (mddev->pers->sync_request) {
4363 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 4364 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4365 /* We are adding a device or devices to an array
4366 * which has the bitmap stored on all devices.
4367 * So make sure all bitmap pages get written
4368 */
4369 bitmap_write_all(mddev->bitmap);
4370 }
1da177e4
LT		 4371 mddev->sync_thread = md_register_thread(md_do_sync,
4372 mddev,
4373 "%s_resync");
4374 if (!mddev->sync_thread) {
4375 printk(KERN_ERR "%s: could not start resync"
4376 " thread...\n",
4377 mdname(mddev));
4378 /* leave the spares where they are, it shouldn't hurt */
4379 mddev->recovery = 0;
d7603b7e 4380 } else
1da177e4 4381 md_wakeup_thread(mddev->sync_thread);
d7603b7e 4382 md_new_event(mddev);
1da177e4
LT		 4383 }
4384 unlock:
4385 mddev_unlock(mddev);
4386 }
4387}
4388
75c96f85
AB		 4389static int md_notify_reboot(struct notifier_block *this,
4390 unsigned long code, void *x)
1da177e4
LT		 4391{
4392 struct list_head *tmp;
4393 mddev_t *mddev;
4394
4395 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4396
4397 printk(KERN_INFO "md: stopping all md devices.\n");
4398
4399 ITERATE_MDDEV(mddev,tmp)
4400 if (mddev_trylock(mddev)==0)
4401 do_md_stop (mddev, 1);
4402 /*
4403 * certain more exotic SCSI devices are known to be
4404 * volatile wrt too early system reboots. While the
4405 * right place to handle this issue is the given
4406 * driver, we do want to have a safe RAID driver ...
4407 */
4408 mdelay(1000*1);
4409 }
4410 return NOTIFY_DONE;
4411}
4412
75c96f85 4413static struct notifier_block md_notifier = {
1da177e4
LT		 4414 .notifier_call = md_notify_reboot,
4415 .next = NULL,
4416 .priority = INT_MAX, /* before any real devices */
4417};
4418
4419static void md_geninit(void)
4420{
4421 struct proc_dir_entry *p;
4422
4423 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4424
4425 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4426 if (p)
4427 p->proc_fops = &md_seq_fops;
4428}
4429
75c96f85 4430static int __init md_init(void)
1da177e4
LT		 4431{
4432 int minor;
4433
4434 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4435 " MD_SB_DISKS=%d\n",
4436 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4437 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
bd926c63 4438 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
32a7627c 4439 BITMAP_MINOR);
1da177e4
LT		 4440
4441 if (register_blkdev(MAJOR_NR, "md"))
4442 return -1;
4443 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4444 unregister_blkdev(MAJOR_NR, "md");
4445 return -1;
4446 }
4447 devfs_mk_dir("md");
4448 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4449 md_probe, NULL, NULL);
4450 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4451 md_probe, NULL, NULL);
4452
4453 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4454 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4455 S_IFBLK|S_IRUSR|S_IWUSR,
4456 "md/%d", minor);
4457
4458 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4459 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4460 S_IFBLK|S_IRUSR|S_IWUSR,
4461 "md/mdp%d", minor);
4462
4463
4464 register_reboot_notifier(&md_notifier);
4465 raid_table_header = register_sysctl_table(raid_root_table, 1);
4466
4467 md_geninit();
4468 return (0);
4469}
4470
4471
4472#ifndef MODULE
4473
4474/*
4475 * Searches all registered partitions for autorun RAID arrays
4476 * at boot time.
4477 */
4478static dev_t detected_devices[128];
4479static int dev_cnt;
4480
4481void md_autodetect_dev(dev_t dev)
4482{
4483 if (dev_cnt >= 0 && dev_cnt < 127)
4484 detected_devices[dev_cnt++] = dev;
4485}
4486
4487
4488static void autostart_arrays(int part)
4489{
4490 mdk_rdev_t *rdev;
4491 int i;
4492
4493 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4494
4495 for (i = 0; i < dev_cnt; i++) {
4496 dev_t dev = detected_devices[i];
4497
4498 rdev = md_import_device(dev,0, 0);
4499 if (IS_ERR(rdev))
4500 continue;
4501
b2d444d7 4502 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4503 MD_BUG();
4504 continue;
4505 }
4506 list_add(&rdev->same_set, &pending_raid_disks);
4507 }
4508 dev_cnt = 0;
4509
4510 autorun_devices(part);
4511}
4512
4513#endif
4514
4515static __exit void md_exit(void)
4516{
4517 mddev_t *mddev;
4518 struct list_head *tmp;
4519 int i;
4520 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4521 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4522 for (i=0; i < MAX_MD_DEVS; i++)
4523 devfs_remove("md/%d", i);
4524 for (i=0; i < MAX_MD_DEVS; i++)
4525 devfs_remove("md/d%d", i);
4526
4527 devfs_remove("md");
4528
4529 unregister_blkdev(MAJOR_NR,"md");
4530 unregister_blkdev(mdp_major, "mdp");
4531 unregister_reboot_notifier(&md_notifier);
4532 unregister_sysctl_table(raid_table_header);
4533 remove_proc_entry("mdstat", NULL);
4534 ITERATE_MDDEV(mddev,tmp) {
4535 struct gendisk *disk = mddev->gendisk;
4536 if (!disk)
4537 continue;
4538 export_array(mddev);
4539 del_gendisk(disk);
4540 put_disk(disk);
4541 mddev->gendisk = NULL;
4542 mddev_put(mddev);
4543 }
4544}
4545
4546module_init(md_init)
4547module_exit(md_exit)
4548
f91de92e
N		 4549static int get_ro(char *buffer, struct kernel_param *kp)
4550{
4551 return sprintf(buffer, "%d", start_readonly);
4552}
4553static int set_ro(const char *val, struct kernel_param *kp)
4554{
4555 char *e;
4556 int num = simple_strtoul(val, &e, 10);
4557 if (*val && (*e == '\0' || *e == '\n')) {
4558 start_readonly = num;
4559 return 0;;
4560 }
4561 return -EINVAL;
4562}
4563
4564module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
6ff8d8ec
N		 4565module_param(start_dirty_degraded, int, 0644);
4566
f91de92e 4567
1da177e4
LT		 4568EXPORT_SYMBOL(register_md_personality);
4569EXPORT_SYMBOL(unregister_md_personality);
4570EXPORT_SYMBOL(md_error);
4571EXPORT_SYMBOL(md_done_sync);
4572EXPORT_SYMBOL(md_write_start);
4573EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 4574EXPORT_SYMBOL(md_register_thread);
4575EXPORT_SYMBOL(md_unregister_thread);
4576EXPORT_SYMBOL(md_wakeup_thread);
4577EXPORT_SYMBOL(md_print_devices);
4578EXPORT_SYMBOL(md_check_recovery);
4579MODULE_LICENSE("GPL");
aa1595e9 4580MODULE_ALIAS("md");
72008652 4581MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)