projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: move test for whether level supports bitmap to correct place
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
35#include <linux/module.h>
50511da3 36#include <linux/kernel.h>
a6fb0934 37#include <linux/kthread.h>
1da177e4
LT		 38#include <linux/linkage.h>
39#include <linux/raid/md.h>
32a7627c 40#include <linux/raid/bitmap.h>
1da177e4 41#include <linux/sysctl.h>
1da177e4 42#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 43#include <linux/poll.h>
48c9c27b 44#include <linux/mutex.h>
16f17b39 45#include <linux/ctype.h>
7dfb7103 46#include <linux/freezer.h>
1da177e4
LT		 47
48#include <linux/init.h>
49
32a7627c
N		 50#include <linux/file.h>
51
1da177e4
LT		 52#ifdef CONFIG_KMOD
53#include <linux/kmod.h>
54#endif
55
56#include <asm/unaligned.h>
57
58#define MAJOR_NR MD_MAJOR
59#define MD_DRIVER
60
61/* 63 partitions with the alternate major number (mdp) */
62#define MdpMinorShift 6
63
64#define DEBUG 0
65#define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68#ifndef MODULE
69static void autostart_arrays (int part);
70#endif
71
2604b703 72static LIST_HEAD(pers_list);
1da177e4
LT		 73static DEFINE_SPINLOCK(pers_lock);
74
5e56341d
AB		 75static void md_print_devices(void);
76
77#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
1da177e4
LT		 79/*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 83 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 89 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 90 */
91
92static int sysctl_speed_limit_min = 1000;
93static int sysctl_speed_limit_max = 200000;
88202a0c
N		 94static inline int speed_min(mddev_t *mddev)
95{
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98}
99
100static inline int speed_max(mddev_t *mddev)
101{
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104}
1da177e4
LT		 105
106static struct ctl_table_header *raid_table_header;
107
108static ctl_table raid_table[] = {
109 {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
80ca3a44 114 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 115 .proc_handler = &proc_dointvec,
116 },
117 {
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
1da177e4
LT		 123 .proc_handler = &proc_dointvec,
124 },
125 { .ctl_name = 0 }
126};
127
128static ctl_table raid_dir_table[] = {
129 {
130 .ctl_name = DEV_RAID,
131 .procname = "raid",
132 .maxlen = 0,
80ca3a44 133 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 134 .child = raid_table,
135 },
136 { .ctl_name = 0 }
137};
138
139static ctl_table raid_root_table[] = {
140 {
141 .ctl_name = CTL_DEV,
142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
147 { .ctl_name = 0 }
148};
149
150static struct block_device_operations md_fops;
151
f91de92e
N		 152static int start_readonly;
153
d7603b7e
N		 154/*
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
158 * count increases.
159 *
160 * Events are:
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
163 */
2989ddbd 164static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 165static atomic_t md_event_count;
29269553 166void md_new_event(mddev_t *mddev)
d7603b7e
N		 167{
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
4508a7a7 170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
d7603b7e 171}
29269553 172EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 173
c331eb04
N		 174/* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
176 */
05381954 177static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 178{
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
181}
182
1da177e4
LT		 183/*
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
186 */
187static LIST_HEAD(all_mddevs);
188static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191/*
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
197 */
198#define ITERATE_MDDEV(mddev,tmp) \
199 \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
202 mddev = NULL;}); \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
210 tmp = tmp->next;}) \
211 )
212
213
214static int md_fail_request (request_queue_t *q, struct bio *bio)
215{
216 bio_io_error(bio, bio->bi_size);
217 return 0;
218}
219
220static inline mddev_t *mddev_get(mddev_t *mddev)
221{
222 atomic_inc(&mddev->active);
223 return mddev;
224}
225
226static void mddev_put(mddev_t *mddev)
227{
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229 return;
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
926ce2d8 232 spin_unlock(&all_mddevs_lock);
1312f40e 233 blk_cleanup_queue(mddev->queue);
eae1701f 234 kobject_unregister(&mddev->kobj);
926ce2d8
N		 235 } else
236 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 237}
238
239static mddev_t * mddev_find(dev_t unit)
240{
241 mddev_t *mddev, *new = NULL;
242
243 retry:
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
247 mddev_get(mddev);
248 spin_unlock(&all_mddevs_lock);
990a8baf 249 kfree(new);
1da177e4
LT		 250 return mddev;
251 }
252
253 if (new) {
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
256 return new;
257 }
258 spin_unlock(&all_mddevs_lock);
259
9ffae0cf 260 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 261 if (!new)
262 return NULL;
263
1da177e4
LT		 264 new->unit = unit;
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
267 else
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
df5b89b3 270 mutex_init(&new->reconfig_mutex);
1da177e4
LT		 271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
06d91a5f 275 spin_lock_init(&new->write_lock);
3d310eb7 276 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 277
278 new->queue = blk_alloc_queue(GFP_KERNEL);
279 if (!new->queue) {
280 kfree(new);
281 return NULL;
282 }
89e5c8b5 283 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
1da177e4
LT		 284
285 blk_queue_make_request(new->queue, md_fail_request);
286
287 goto retry;
288}
289
290static inline int mddev_lock(mddev_t * mddev)
291{
df5b89b3 292 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 293}
294
1da177e4
LT		 295static inline int mddev_trylock(mddev_t * mddev)
296{
df5b89b3 297 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 298}
299
300static inline void mddev_unlock(mddev_t * mddev)
301{
df5b89b3 302 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 303
005eca5e 304 md_wakeup_thread(mddev->thread);
1da177e4
LT		 305}
306
2989ddbd 307static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4
LT		 308{
309 mdk_rdev_t * rdev;
310 struct list_head *tmp;
311
312 ITERATE_RDEV(mddev,rdev,tmp) {
313 if (rdev->desc_nr == nr)
314 return rdev;
315 }
316 return NULL;
317}
318
319static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
320{
321 struct list_head *tmp;
322 mdk_rdev_t *rdev;
323
324 ITERATE_RDEV(mddev,rdev,tmp) {
325 if (rdev->bdev->bd_dev == dev)
326 return rdev;
327 }
328 return NULL;
329}
330
d9d166c2 331static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 332{
333 struct mdk_personality *pers;
d9d166c2
N		 334 list_for_each_entry(pers, &pers_list, list) {
335 if (level != LEVEL_NONE && pers->level == level)
2604b703 336 return pers;
d9d166c2
N		 337 if (strcmp(pers->name, clevel)==0)
338 return pers;
339 }
2604b703
N		 340 return NULL;
341}
342
77933d72 343static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 344{
345 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
346 return MD_NEW_SIZE_BLOCKS(size);
347}
348
349static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
350{
351 sector_t size;
352
353 size = rdev->sb_offset;
354
355 if (chunk_size)
356 size &= ~((sector_t)chunk_size/1024 - 1);
357 return size;
358}
359
360static int alloc_disk_sb(mdk_rdev_t * rdev)
361{
362 if (rdev->sb_page)
363 MD_BUG();
364
365 rdev->sb_page = alloc_page(GFP_KERNEL);
366 if (!rdev->sb_page) {
367 printk(KERN_ALERT "md: out of memory.\n");
368 return -EINVAL;
369 }
370
371 return 0;
372}
373
374static void free_disk_sb(mdk_rdev_t * rdev)
375{
376 if (rdev->sb_page) {
2d1f3b5d 377 put_page(rdev->sb_page);
1da177e4
LT		 378 rdev->sb_loaded = 0;
379 rdev->sb_page = NULL;
380 rdev->sb_offset = 0;
381 rdev->size = 0;
382 }
383}
384
385
7bfa19f2
N		 386static int super_written(struct bio *bio, unsigned int bytes_done, int error)
387{
388 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 389 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 390 if (bio->bi_size)
391 return 1;
392
3a0f5bbb
N		 393 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
394 printk("md: super_written gets error=%d, uptodate=%d\n",
395 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
396 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 397 md_error(mddev, rdev);
3a0f5bbb 398 }
7bfa19f2 399
a9701a30
N		 400 if (atomic_dec_and_test(&mddev->pending_writes))
401 wake_up(&mddev->sb_wait);
f8b58edf 402 bio_put(bio);
7bfa19f2
N		 403 return 0;
404}
405
a9701a30
N		 406static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
407{
408 struct bio *bio2 = bio->bi_private;
409 mdk_rdev_t *rdev = bio2->bi_private;
410 mddev_t *mddev = rdev->mddev;
411 if (bio->bi_size)
412 return 1;
413
414 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
415 error == -EOPNOTSUPP) {
416 unsigned long flags;
417 /* barriers don't appear to be supported :-( */
418 set_bit(BarriersNotsupp, &rdev->flags);
419 mddev->barriers_work = 0;
420 spin_lock_irqsave(&mddev->write_lock, flags);
421 bio2->bi_next = mddev->biolist;
422 mddev->biolist = bio2;
423 spin_unlock_irqrestore(&mddev->write_lock, flags);
424 wake_up(&mddev->sb_wait);
425 bio_put(bio);
426 return 0;
427 }
428 bio_put(bio2);
429 bio->bi_private = rdev;
430 return super_written(bio, bytes_done, error);
431}
432
7bfa19f2
N		 433void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
434 sector_t sector, int size, struct page *page)
435{
436 /* write first size bytes of page to sector of rdev
437 * Increment mddev->pending_writes before returning
438 * and decrement it on completion, waking up sb_wait
439 * if zero is reached.
440 * If an error occurred, call md_error
a9701a30
N		 441 *
442 * As we might need to resubmit the request if BIO_RW_BARRIER
443 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 444 */
445 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 446 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 447
448 bio->bi_bdev = rdev->bdev;
449 bio->bi_sector = sector;
450 bio_add_page(bio, page, size, 0);
451 bio->bi_private = rdev;
452 bio->bi_end_io = super_written;
a9701a30
N		 453 bio->bi_rw = rw;
454
7bfa19f2 455 atomic_inc(&mddev->pending_writes);
a9701a30
N		 456 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
457 struct bio *rbio;
458 rw |= (1<<BIO_RW_BARRIER);
459 rbio = bio_clone(bio, GFP_NOIO);
460 rbio->bi_private = bio;
461 rbio->bi_end_io = super_written_barrier;
462 submit_bio(rw, rbio);
463 } else
464 submit_bio(rw, bio);
465}
466
467void md_super_wait(mddev_t *mddev)
468{
469 /* wait for all superblock writes that were scheduled to complete.
470 * if any had to be retried (due to BARRIER problems), retry them
471 */
472 DEFINE_WAIT(wq);
473 for(;;) {
474 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
475 if (atomic_read(&mddev->pending_writes)==0)
476 break;
477 while (mddev->biolist) {
478 struct bio *bio;
479 spin_lock_irq(&mddev->write_lock);
480 bio = mddev->biolist;
481 mddev->biolist = bio->bi_next ;
482 bio->bi_next = NULL;
483 spin_unlock_irq(&mddev->write_lock);
484 submit_bio(bio->bi_rw, bio);
485 }
486 schedule();
487 }
488 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 489}
490
1da177e4
LT		 491static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
492{
493 if (bio->bi_size)
494 return 1;
495
496 complete((struct completion*)bio->bi_private);
497 return 0;
498}
499
a654b9d8 500int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 501 struct page *page, int rw)
502{
baaa2c51 503 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 504 struct completion event;
505 int ret;
506
507 rw |= (1 << BIO_RW_SYNC);
508
509 bio->bi_bdev = bdev;
510 bio->bi_sector = sector;
511 bio_add_page(bio, page, size, 0);
512 init_completion(&event);
513 bio->bi_private = &event;
514 bio->bi_end_io = bi_complete;
515 submit_bio(rw, bio);
516 wait_for_completion(&event);
517
518 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
519 bio_put(bio);
520 return ret;
521}
a8745db2 522EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 523
0002b271 524static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 525{
526 char b[BDEVNAME_SIZE];
527 if (!rdev->sb_page) {
528 MD_BUG();
529 return -EINVAL;
530 }
531 if (rdev->sb_loaded)
532 return 0;
533
534
0002b271 535 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 536 goto fail;
537 rdev->sb_loaded = 1;
538 return 0;
539
540fail:
541 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
542 bdevname(rdev->bdev,b));
543 return -EINVAL;
544}
545
546static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
547{
548 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
549 (sb1->set_uuid1 == sb2->set_uuid1) &&
550 (sb1->set_uuid2 == sb2->set_uuid2) &&
551 (sb1->set_uuid3 == sb2->set_uuid3))
552
553 return 1;
554
555 return 0;
556}
557
558
559static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
560{
561 int ret;
562 mdp_super_t *tmp1, *tmp2;
563
564 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
565 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
566
567 if (!tmp1 || !tmp2) {
568 ret = 0;
569 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
570 goto abort;
571 }
572
573 *tmp1 = *sb1;
574 *tmp2 = *sb2;
575
576 /*
577 * nr_disks is not constant
578 */
579 tmp1->nr_disks = 0;
580 tmp2->nr_disks = 0;
581
582 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
583 ret = 0;
584 else
585 ret = 1;
586
587abort:
990a8baf
JJ		 588 kfree(tmp1);
589 kfree(tmp2);
1da177e4
LT		 590 return ret;
591}
592
593static unsigned int calc_sb_csum(mdp_super_t * sb)
594{
595 unsigned int disk_csum, csum;
596
597 disk_csum = sb->sb_csum;
598 sb->sb_csum = 0;
599 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
600 sb->sb_csum = disk_csum;
601 return csum;
602}
603
604
605/*
606 * Handle superblock details.
607 * We want to be able to handle multiple superblock formats
608 * so we have a common interface to them all, and an array of
609 * different handlers.
610 * We rely on user-space to write the initial superblock, and support
611 * reading and updating of superblocks.
612 * Interface methods are:
613 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
614 * loads and validates a superblock on dev.
615 * if refdev != NULL, compare superblocks on both devices
616 * Return:
617 * 0 - dev has a superblock that is compatible with refdev
618 * 1 - dev has a superblock that is compatible and newer than refdev
619 * so dev should be used as the refdev in future
620 * -EINVAL superblock incompatible or invalid
621 * -othererror e.g. -EIO
622 *
623 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
624 * Verify that dev is acceptable into mddev.
625 * The first time, mddev->raid_disks will be 0, and data from
626 * dev should be merged in. Subsequent calls check that dev
627 * is new enough. Return 0 or -EINVAL
628 *
629 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
630 * Update the superblock for rdev with data in mddev
631 * This does not write to disc.
632 *
633 */
634
635struct super_type {
636 char *name;
637 struct module *owner;
638 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
639 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
640 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
641};
642
643/*
644 * load_super for 0.90.0
645 */
646static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
647{
648 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
649 mdp_super_t *sb;
650 int ret;
651 sector_t sb_offset;
652
653 /*
654 * Calculate the position of the superblock,
655 * it's at the end of the disk.
656 *
657 * It also happens to be a multiple of 4Kb.
658 */
659 sb_offset = calc_dev_sboffset(rdev->bdev);
660 rdev->sb_offset = sb_offset;
661
0002b271 662 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 663 if (ret) return ret;
664
665 ret = -EINVAL;
666
667 bdevname(rdev->bdev, b);
668 sb = (mdp_super_t*)page_address(rdev->sb_page);
669
670 if (sb->md_magic != MD_SB_MAGIC) {
671 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
672 b);
673 goto abort;
674 }
675
676 if (sb->major_version != 0 ||
f6705578
N		 677 sb->minor_version < 90 ||
678 sb->minor_version > 91) {
1da177e4
LT		 679 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
680 sb->major_version, sb->minor_version,
681 b);
682 goto abort;
683 }
684
685 if (sb->raid_disks <= 0)
686 goto abort;
687
688 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
689 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
690 b);
691 goto abort;
692 }
693
694 rdev->preferred_minor = sb->md_minor;
695 rdev->data_offset = 0;
0002b271 696 rdev->sb_size = MD_SB_BYTES;
1da177e4 697
e11e93fa
N		 698 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
699 if (sb->level != 1 && sb->level != 4
700 && sb->level != 5 && sb->level != 6
701 && sb->level != 10) {
702 /* FIXME use a better test */
703 printk(KERN_WARNING
704 "md: bitmaps not supported for this level.\n");
705 goto abort;
706 }
707 }
708
1da177e4
LT		 709 if (sb->level == LEVEL_MULTIPATH)
710 rdev->desc_nr = -1;
711 else
712 rdev->desc_nr = sb->this_disk.number;
713
714 if (refdev == 0)
715 ret = 1;
716 else {
717 __u64 ev1, ev2;
718 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
719 if (!uuid_equal(refsb, sb)) {
720 printk(KERN_WARNING "md: %s has different UUID to %s\n",
721 b, bdevname(refdev->bdev,b2));
722 goto abort;
723 }
724 if (!sb_equal(refsb, sb)) {
725 printk(KERN_WARNING "md: %s has same UUID"
726 " but different superblock to %s\n",
727 b, bdevname(refdev->bdev, b2));
728 goto abort;
729 }
730 ev1 = md_event(sb);
731 ev2 = md_event(refsb);
732 if (ev1 > ev2)
733 ret = 1;
734 else
735 ret = 0;
736 }
737 rdev->size = calc_dev_size(rdev, sb->chunk_size);
738
2bf071bf
N		 739 if (rdev->size < sb->size && sb->level > 1)
740 /* "this cannot possibly happen" ... */
741 ret = -EINVAL;
742
1da177e4
LT		 743 abort:
744 return ret;
745}
746
747/*
748 * validate_super for 0.90.0
749 */
750static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
751{
752 mdp_disk_t *desc;
753 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 754 __u64 ev1 = md_event(sb);
1da177e4 755
41158c7e 756 rdev->raid_disk = -1;
b2d444d7 757 rdev->flags = 0;
1da177e4
LT		 758 if (mddev->raid_disks == 0) {
759 mddev->major_version = 0;
760 mddev->minor_version = sb->minor_version;
761 mddev->patch_version = sb->patch_version;
762 mddev->persistent = ! sb->not_persistent;
763 mddev->chunk_size = sb->chunk_size;
764 mddev->ctime = sb->ctime;
765 mddev->utime = sb->utime;
766 mddev->level = sb->level;
d9d166c2 767 mddev->clevel[0] = 0;
1da177e4
LT		 768 mddev->layout = sb->layout;
769 mddev->raid_disks = sb->raid_disks;
770 mddev->size = sb->size;
07d84d10 771 mddev->events = ev1;
9223214e 772 mddev->bitmap_offset = 0;
36fa3063 773 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4 774
f6705578
N		 775 if (mddev->minor_version >= 91) {
776 mddev->reshape_position = sb->reshape_position;
777 mddev->delta_disks = sb->delta_disks;
778 mddev->new_level = sb->new_level;
779 mddev->new_layout = sb->new_layout;
780 mddev->new_chunk = sb->new_chunk;
781 } else {
782 mddev->reshape_position = MaxSector;
783 mddev->delta_disks = 0;
784 mddev->new_level = mddev->level;
785 mddev->new_layout = mddev->layout;
786 mddev->new_chunk = mddev->chunk_size;
787 }
788
1da177e4
LT		 789 if (sb->state & (1<<MD_SB_CLEAN))
790 mddev->recovery_cp = MaxSector;
791 else {
792 if (sb->events_hi == sb->cp_events_hi &&
793 sb->events_lo == sb->cp_events_lo) {
794 mddev->recovery_cp = sb->recovery_cp;
795 } else
796 mddev->recovery_cp = 0;
797 }
798
799 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
800 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
801 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
802 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
803
804 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 805
806 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
e11e93fa 807 mddev->bitmap_file == NULL)
36fa3063 808 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8 809
41158c7e
N		 810 } else if (mddev->pers == NULL) {
811 /* Insist on good event counter while assembling */
1da177e4
LT		 812 ++ev1;
813 if (ev1 < mddev->events)
814 return -EINVAL;
41158c7e
N		 815 } else if (mddev->bitmap) {
816 /* if adding to array with a bitmap, then we can accept an
817 * older device ... but not too old.
818 */
41158c7e
N		 819 if (ev1 < mddev->bitmap->events_cleared)
820 return 0;
07d84d10
N		 821 } else {
822 if (ev1 < mddev->events)
823 /* just a hot-add of a new device, leave raid_disk at -1 */
824 return 0;
825 }
41158c7e 826
1da177e4 827 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 828 desc = sb->disks + rdev->desc_nr;
829
830 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 831 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 832 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
833 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 834 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 835 rdev->raid_disk = desc->raid_disk;
836 }
8ddf9efe
N		 837 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
838 set_bit(WriteMostly, &rdev->flags);
41158c7e 839 } else /* MULTIPATH are always insync */
b2d444d7 840 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 841 return 0;
842}
843
844/*
845 * sync_super for 0.90.0
846 */
847static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
848{
849 mdp_super_t *sb;
850 struct list_head *tmp;
851 mdk_rdev_t *rdev2;
852 int next_spare = mddev->raid_disks;
19133a42 853
1da177e4
LT		 854
855 /* make rdev->sb match mddev data..
856 *
857 * 1/ zero out disks
858 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
859 * 3/ any empty disks < next_spare become removed
860 *
861 * disks[0] gets initialised to REMOVED because
862 * we cannot be sure from other fields if it has
863 * been initialised or not.
864 */
865 int i;
866 int active=0, working=0,failed=0,spare=0,nr_disks=0;
867
61181565
N		 868 rdev->sb_size = MD_SB_BYTES;
869
1da177e4
LT		 870 sb = (mdp_super_t*)page_address(rdev->sb_page);
871
872 memset(sb, 0, sizeof(*sb));
873
874 sb->md_magic = MD_SB_MAGIC;
875 sb->major_version = mddev->major_version;
1da177e4
LT		 876 sb->patch_version = mddev->patch_version;
877 sb->gvalid_words = 0; /* ignored */
878 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
879 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
880 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
881 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
882
883 sb->ctime = mddev->ctime;
884 sb->level = mddev->level;
885 sb->size = mddev->size;
886 sb->raid_disks = mddev->raid_disks;
887 sb->md_minor = mddev->md_minor;
888 sb->not_persistent = !mddev->persistent;
889 sb->utime = mddev->utime;
890 sb->state = 0;
891 sb->events_hi = (mddev->events>>32);
892 sb->events_lo = (u32)mddev->events;
893
f6705578
N		 894 if (mddev->reshape_position == MaxSector)
895 sb->minor_version = 90;
896 else {
897 sb->minor_version = 91;
898 sb->reshape_position = mddev->reshape_position;
899 sb->new_level = mddev->new_level;
900 sb->delta_disks = mddev->delta_disks;
901 sb->new_layout = mddev->new_layout;
902 sb->new_chunk = mddev->new_chunk;
903 }
904 mddev->minor_version = sb->minor_version;
1da177e4
LT		 905 if (mddev->in_sync)
906 {
907 sb->recovery_cp = mddev->recovery_cp;
908 sb->cp_events_hi = (mddev->events>>32);
909 sb->cp_events_lo = (u32)mddev->events;
910 if (mddev->recovery_cp == MaxSector)
911 sb->state = (1<< MD_SB_CLEAN);
912 } else
913 sb->recovery_cp = 0;
914
915 sb->layout = mddev->layout;
916 sb->chunk_size = mddev->chunk_size;
917
a654b9d8
N		 918 if (mddev->bitmap && mddev->bitmap_file == NULL)
919 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
920
1da177e4
LT		 921 sb->disks[0].state = (1<<MD_DISK_REMOVED);
922 ITERATE_RDEV(mddev,rdev2,tmp) {
923 mdp_disk_t *d;
86e6ffdd 924 int desc_nr;
b2d444d7
N		 925 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
926 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 927 desc_nr = rdev2->raid_disk;
1da177e4 928 else
86e6ffdd 929 desc_nr = next_spare++;
19133a42 930 rdev2->desc_nr = desc_nr;
1da177e4
LT		 931 d = &sb->disks[rdev2->desc_nr];
932 nr_disks++;
933 d->number = rdev2->desc_nr;
934 d->major = MAJOR(rdev2->bdev->bd_dev);
935 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 936 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
937 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 938 d->raid_disk = rdev2->raid_disk;
939 else
940 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 941 if (test_bit(Faulty, &rdev2->flags))
1da177e4 942 d->state = (1<<MD_DISK_FAULTY);
1be7892f 943 else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 944 d->state = (1<<MD_DISK_ACTIVE);
945 d->state |= (1<<MD_DISK_SYNC);
946 active++;
947 working++;
948 } else {
949 d->state = 0;
950 spare++;
951 working++;
952 }
8ddf9efe
N		 953 if (test_bit(WriteMostly, &rdev2->flags))
954 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 955 }
1da177e4
LT		 956 /* now set the "removed" and "faulty" bits on any missing devices */
957 for (i=0 ; i < mddev->raid_disks ; i++) {
958 mdp_disk_t *d = &sb->disks[i];
959 if (d->state == 0 && d->number == 0) {
960 d->number = i;
961 d->raid_disk = i;
962 d->state = (1<<MD_DISK_REMOVED);
963 d->state |= (1<<MD_DISK_FAULTY);
964 failed++;
965 }
966 }
967 sb->nr_disks = nr_disks;
968 sb->active_disks = active;
969 sb->working_disks = working;
970 sb->failed_disks = failed;
971 sb->spare_disks = spare;
972
973 sb->this_disk = sb->disks[rdev->desc_nr];
974 sb->sb_csum = calc_sb_csum(sb);
975}
976
977/*
978 * version 1 superblock
979 */
980
1c05b4bc 981static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 982{
1c05b4bc
N		 983 __le32 disk_csum;
984 u32 csum;
1da177e4
LT		 985 unsigned long long newcsum;
986 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 987 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 988 int i;
989
990 disk_csum = sb->sb_csum;
991 sb->sb_csum = 0;
992 newcsum = 0;
993 for (i=0; size>=4; size -= 4 )
994 newcsum += le32_to_cpu(*isuper++);
995
996 if (size == 2)
1c05b4bc 997 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 998
999 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1000 sb->sb_csum = disk_csum;
1001 return cpu_to_le32(csum);
1002}
1003
1004static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1005{
1006 struct mdp_superblock_1 *sb;
1007 int ret;
1008 sector_t sb_offset;
1009 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1010 int bmask;
1da177e4
LT		 1011
1012 /*
1013 * Calculate the position of the superblock.
1014 * It is always aligned to a 4K boundary and
1015 * depeding on minor_version, it can be:
1016 * 0: At least 8K, but less than 12K, from end of device
1017 * 1: At start of device
1018 * 2: 4K from start of device.
1019 */
1020 switch(minor_version) {
1021 case 0:
1022 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1023 sb_offset -= 8*2;
39730960 1024 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 1025 /* convert from sectors to K */
1026 sb_offset /= 2;
1027 break;
1028 case 1:
1029 sb_offset = 0;
1030 break;
1031 case 2:
1032 sb_offset = 4;
1033 break;
1034 default:
1035 return -EINVAL;
1036 }
1037 rdev->sb_offset = sb_offset;
1038
0002b271
N		 1039 /* superblock is rarely larger than 1K, but it can be larger,
1040 * and it is safe to read 4k, so we do that
1041 */
1042 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1043 if (ret) return ret;
1044
1045
1046 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1047
1048 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1049 sb->major_version != cpu_to_le32(1) ||
1050 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1051 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 1052 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1053 return -EINVAL;
1054
1055 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1056 printk("md: invalid superblock checksum on %s\n",
1057 bdevname(rdev->bdev,b));
1058 return -EINVAL;
1059 }
1060 if (le64_to_cpu(sb->data_size) < 10) {
1061 printk("md: data_size too small on %s\n",
1062 bdevname(rdev->bdev,b));
1063 return -EINVAL;
1064 }
e11e93fa
N		 1065 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1066 if (sb->level != cpu_to_le32(1) &&
1067 sb->level != cpu_to_le32(4) &&
1068 sb->level != cpu_to_le32(5) &&
1069 sb->level != cpu_to_le32(6) &&
1070 sb->level != cpu_to_le32(10)) {
1071 printk(KERN_WARNING
1072 "md: bitmaps not supported for this level.\n");
1073 return -EINVAL;
1074 }
1075 }
1076
1da177e4
LT		 1077 rdev->preferred_minor = 0xffff;
1078 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1079 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1080
0002b271 1081 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1082 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1083 if (rdev->sb_size & bmask)
1084 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1085
31b65a0d
N		 1086 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1087 rdev->desc_nr = -1;
1088 else
1089 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1090
1da177e4 1091 if (refdev == 0)
8ed75463 1092 ret = 1;
1da177e4
LT		 1093 else {
1094 __u64 ev1, ev2;
1095 struct mdp_superblock_1 *refsb =
1096 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1097
1098 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1099 sb->level != refsb->level ||
1100 sb->layout != refsb->layout ||
1101 sb->chunksize != refsb->chunksize) {
1102 printk(KERN_WARNING "md: %s has strangely different"
1103 " superblock to %s\n",
1104 bdevname(rdev->bdev,b),
1105 bdevname(refdev->bdev,b2));
1106 return -EINVAL;
1107 }
1108 ev1 = le64_to_cpu(sb->events);
1109 ev2 = le64_to_cpu(refsb->events);
1110
1111 if (ev1 > ev2)
8ed75463
N		 1112 ret = 1;
1113 else
1114 ret = 0;
1da177e4
LT		 1115 }
1116 if (minor_version)
1117 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1118 else
1119 rdev->size = rdev->sb_offset;
1120 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1121 return -EINVAL;
1122 rdev->size = le64_to_cpu(sb->data_size)/2;
1123 if (le32_to_cpu(sb->chunksize))
1124 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
2bf071bf 1125
1c05b4bc 1126 if (le64_to_cpu(sb->size) > rdev->size*2)
2bf071bf 1127 return -EINVAL;
8ed75463 1128 return ret;
1da177e4
LT		 1129}
1130
1131static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1132{
1133 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1134 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1135
41158c7e 1136 rdev->raid_disk = -1;
b2d444d7 1137 rdev->flags = 0;
1da177e4
LT		 1138 if (mddev->raid_disks == 0) {
1139 mddev->major_version = 1;
1140 mddev->patch_version = 0;
1141 mddev->persistent = 1;
1142 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1143 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1144 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1145 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1146 mddev->clevel[0] = 0;
1da177e4
LT		 1147 mddev->layout = le32_to_cpu(sb->layout);
1148 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1149 mddev->size = le64_to_cpu(sb->size)/2;
07d84d10 1150 mddev->events = ev1;
9223214e 1151 mddev->bitmap_offset = 0;
29fc7e3e 1152 mddev->default_bitmap_offset = 1024 >> 9;
1da177e4
LT		 1153
1154 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1155 memcpy(mddev->uuid, sb->set_uuid, 16);
1156
1157 mddev->max_disks = (4096-256)/2;
a654b9d8 1158
71c0805c 1159 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
e11e93fa 1160 mddev->bitmap_file == NULL )
a654b9d8 1161 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1162
f6705578
N		 1163 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1164 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1165 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1166 mddev->new_level = le32_to_cpu(sb->new_level);
1167 mddev->new_layout = le32_to_cpu(sb->new_layout);
1168 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1169 } else {
1170 mddev->reshape_position = MaxSector;
1171 mddev->delta_disks = 0;
1172 mddev->new_level = mddev->level;
1173 mddev->new_layout = mddev->layout;
1174 mddev->new_chunk = mddev->chunk_size;
1175 }
1176
41158c7e
N		 1177 } else if (mddev->pers == NULL) {
1178 /* Insist of good event counter while assembling */
1da177e4
LT		 1179 ++ev1;
1180 if (ev1 < mddev->events)
1181 return -EINVAL;
41158c7e
N		 1182 } else if (mddev->bitmap) {
1183 /* If adding to array with a bitmap, then we can accept an
1184 * older device, but not too old.
1185 */
41158c7e
N		 1186 if (ev1 < mddev->bitmap->events_cleared)
1187 return 0;
07d84d10
N		 1188 } else {
1189 if (ev1 < mddev->events)
1190 /* just a hot-add of a new device, leave raid_disk at -1 */
1191 return 0;
1192 }
1da177e4
LT		 1193 if (mddev->level != LEVEL_MULTIPATH) {
1194 int role;
1da177e4
LT		 1195 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1196 switch(role) {
1197 case 0xffff: /* spare */
1da177e4
LT		 1198 break;
1199 case 0xfffe: /* faulty */
b2d444d7 1200 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1201 break;
1202 default:
5fd6c1dc
N		 1203 if ((le32_to_cpu(sb->feature_map) &
1204 MD_FEATURE_RECOVERY_OFFSET))
1205 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1206 else
1207 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1208 rdev->raid_disk = role;
1209 break;
1210 }
8ddf9efe
N		 1211 if (sb->devflags & WriteMostly1)
1212 set_bit(WriteMostly, &rdev->flags);
41158c7e 1213 } else /* MULTIPATH are always insync */
b2d444d7 1214 set_bit(In_sync, &rdev->flags);
41158c7e 1215
1da177e4
LT		 1216 return 0;
1217}
1218
1219static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1220{
1221 struct mdp_superblock_1 *sb;
1222 struct list_head *tmp;
1223 mdk_rdev_t *rdev2;
1224 int max_dev, i;
1225 /* make rdev->sb match mddev and rdev data. */
1226
1227 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1228
1229 sb->feature_map = 0;
1230 sb->pad0 = 0;
5fd6c1dc 1231 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1232 memset(sb->pad1, 0, sizeof(sb->pad1));
1233 memset(sb->pad2, 0, sizeof(sb->pad2));
1234 memset(sb->pad3, 0, sizeof(sb->pad3));
1235
1236 sb->utime = cpu_to_le64((__u64)mddev->utime);
1237 sb->events = cpu_to_le64(mddev->events);
1238 if (mddev->in_sync)
1239 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1240 else
1241 sb->resync_offset = cpu_to_le64(0);
1242
1c05b4bc 1243 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1244
f0ca340c 1245 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
29fc7e3e 1246 sb->size = cpu_to_le64(mddev->size<<1);
f0ca340c 1247
a654b9d8
N		 1248 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1249 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1250 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1251 }
5fd6c1dc
N		 1252
1253 if (rdev->raid_disk >= 0 &&
1254 !test_bit(In_sync, &rdev->flags) &&
1255 rdev->recovery_offset > 0) {
1256 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1257 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1258 }
1259
f6705578
N		 1260 if (mddev->reshape_position != MaxSector) {
1261 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1262 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1263 sb->new_layout = cpu_to_le32(mddev->new_layout);
1264 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1265 sb->new_level = cpu_to_le32(mddev->new_level);
1266 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1267 }
a654b9d8 1268
1da177e4
LT		 1269 max_dev = 0;
1270 ITERATE_RDEV(mddev,rdev2,tmp)
1271 if (rdev2->desc_nr+1 > max_dev)
1272 max_dev = rdev2->desc_nr+1;
1273
1274 sb->max_dev = cpu_to_le32(max_dev);
1275 for (i=0; i<max_dev;i++)
1276 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1277
1278 ITERATE_RDEV(mddev,rdev2,tmp) {
1279 i = rdev2->desc_nr;
b2d444d7 1280 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1281 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1282 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1283 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
5fd6c1dc
N		 1284 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1285 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1286 else
1287 sb->dev_roles[i] = cpu_to_le16(0xffff);
1288 }
1289
1da177e4
LT		 1290 sb->sb_csum = calc_sb_1_csum(sb);
1291}
1292
1293
75c96f85 1294static struct super_type super_types[] = {
1da177e4
LT		 1295 [0] = {
1296 .name = "0.90.0",
1297 .owner = THIS_MODULE,
1298 .load_super = super_90_load,
1299 .validate_super = super_90_validate,
1300 .sync_super = super_90_sync,
1301 },
1302 [1] = {
1303 .name = "md-1",
1304 .owner = THIS_MODULE,
1305 .load_super = super_1_load,
1306 .validate_super = super_1_validate,
1307 .sync_super = super_1_sync,
1308 },
1309};
1da177e4
LT		 1310
1311static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1312{
7dd5e7c3
N		 1313 struct list_head *tmp, *tmp2;
1314 mdk_rdev_t *rdev, *rdev2;
1da177e4
LT		 1315
1316 ITERATE_RDEV(mddev1,rdev,tmp)
7dd5e7c3
N		 1317 ITERATE_RDEV(mddev2, rdev2, tmp2)
1318 if (rdev->bdev->bd_contains ==
1319 rdev2->bdev->bd_contains)
1320 return 1;
1da177e4
LT		 1321
1322 return 0;
1323}
1324
1325static LIST_HEAD(pending_raid_disks);
1326
1327static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1328{
7dd5e7c3 1329 char b[BDEVNAME_SIZE];
f637b9f9 1330 struct kobject *ko;
1edf80d3 1331 char *s;
5e55e2f5 1332 int err;
1da177e4
LT		 1333
1334 if (rdev->mddev) {
1335 MD_BUG();
1336 return -EINVAL;
1337 }
2bf071bf
N		 1338 /* make sure rdev->size exceeds mddev->size */
1339 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1340 if (mddev->pers)
1341 /* Cannot change size, so fail */
1342 return -ENOSPC;
1343 else
1344 mddev->size = rdev->size;
1345 }
1da177e4
LT		 1346
1347 /* Verify rdev->desc_nr is unique.
1348 * If it is -1, assign a free number, else
1349 * check number is not in use
1350 */
1351 if (rdev->desc_nr < 0) {
1352 int choice = 0;
1353 if (mddev->pers) choice = mddev->raid_disks;
1354 while (find_rdev_nr(mddev, choice))
1355 choice++;
1356 rdev->desc_nr = choice;
1357 } else {
1358 if (find_rdev_nr(mddev, rdev->desc_nr))
1359 return -EBUSY;
1360 }
19133a42
N		 1361 bdevname(rdev->bdev,b);
1362 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1363 return -ENOMEM;
1edf80d3
NB		 1364 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1365 *s = '!';
1da177e4 1366
1da177e4 1367 rdev->mddev = mddev;
19133a42 1368 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1369
9c791977 1370 rdev->kobj.parent = &mddev->kobj;
5e55e2f5
N		 1371 if ((err = kobject_add(&rdev->kobj)))
1372 goto fail;
86e6ffdd 1373
f637b9f9
N		 1374 if (rdev->bdev->bd_part)
1375 ko = &rdev->bdev->bd_part->kobj;
1376 else
1377 ko = &rdev->bdev->bd_disk->kobj;
5e55e2f5
N		 1378 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1379 kobject_del(&rdev->kobj);
1380 goto fail;
1381 }
1382 list_add(&rdev->same_set, &mddev->disks);
5463c790 1383 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1da177e4 1384 return 0;
5e55e2f5
N		 1385
1386 fail:
1387 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1388 b, mdname(mddev));
1389 return err;
1da177e4
LT		 1390}
1391
5792a285
N		 1392static void delayed_delete(struct work_struct *ws)
1393{
1394 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1395 kobject_del(&rdev->kobj);
1396}
1397
1da177e4
LT		 1398static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1399{
1400 char b[BDEVNAME_SIZE];
1401 if (!rdev->mddev) {
1402 MD_BUG();
1403 return;
1404 }
5463c790 1405 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1da177e4
LT		 1406 list_del_init(&rdev->same_set);
1407 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1408 rdev->mddev = NULL;
86e6ffdd 1409 sysfs_remove_link(&rdev->kobj, "block");
5792a285
N		 1410
1411 /* We need to delay this, otherwise we can deadlock when
1412 * writing to 'remove' to "dev/state"
1413 */
1414 INIT_WORK(&rdev->del_work, delayed_delete);
1415 schedule_work(&rdev->del_work);
1da177e4
LT		 1416}
1417
1418/*
1419 * prevent the device from being mounted, repartitioned or
1420 * otherwise reused by a RAID array (or any other kernel
1421 * subsystem), by bd_claiming the device.
1422 */
1423static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1424{
1425 int err = 0;
1426 struct block_device *bdev;
1427 char b[BDEVNAME_SIZE];
1428
2e7b651d 1429 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1430 if (IS_ERR(bdev)) {
1431 printk(KERN_ERR "md: could not open %s.\n",
1432 __bdevname(dev, b));
1433 return PTR_ERR(bdev);
1434 }
1435 err = bd_claim(bdev, rdev);
1436 if (err) {
1437 printk(KERN_ERR "md: could not bd_claim %s.\n",
1438 bdevname(bdev, b));
2e7b651d 1439 blkdev_put(bdev);
1da177e4
LT		 1440 return err;
1441 }
1442 rdev->bdev = bdev;
1443 return err;
1444}
1445
1446static void unlock_rdev(mdk_rdev_t *rdev)
1447{
1448 struct block_device *bdev = rdev->bdev;
1449 rdev->bdev = NULL;
1450 if (!bdev)
1451 MD_BUG();
1452 bd_release(bdev);
2e7b651d 1453 blkdev_put(bdev);
1da177e4
LT		 1454}
1455
1456void md_autodetect_dev(dev_t dev);
1457
1458static void export_rdev(mdk_rdev_t * rdev)
1459{
1460 char b[BDEVNAME_SIZE];
1461 printk(KERN_INFO "md: export_rdev(%s)\n",
1462 bdevname(rdev->bdev,b));
1463 if (rdev->mddev)
1464 MD_BUG();
1465 free_disk_sb(rdev);
1466 list_del_init(&rdev->same_set);
1467#ifndef MODULE
1468 md_autodetect_dev(rdev->bdev->bd_dev);
1469#endif
1470 unlock_rdev(rdev);
86e6ffdd 1471 kobject_put(&rdev->kobj);
1da177e4
LT		 1472}
1473
1474static void kick_rdev_from_array(mdk_rdev_t * rdev)
1475{
1476 unbind_rdev_from_array(rdev);
1477 export_rdev(rdev);
1478}
1479
1480static void export_array(mddev_t *mddev)
1481{
1482 struct list_head *tmp;
1483 mdk_rdev_t *rdev;
1484
1485 ITERATE_RDEV(mddev,rdev,tmp) {
1486 if (!rdev->mddev) {
1487 MD_BUG();
1488 continue;
1489 }
1490 kick_rdev_from_array(rdev);
1491 }
1492 if (!list_empty(&mddev->disks))
1493 MD_BUG();
1494 mddev->raid_disks = 0;
1495 mddev->major_version = 0;
1496}
1497
1498static void print_desc(mdp_disk_t *desc)
1499{
1500 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1501 desc->major,desc->minor,desc->raid_disk,desc->state);
1502}
1503
1504static void print_sb(mdp_super_t *sb)
1505{
1506 int i;
1507
1508 printk(KERN_INFO
1509 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1510 sb->major_version, sb->minor_version, sb->patch_version,
1511 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1512 sb->ctime);
1513 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1514 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1515 sb->md_minor, sb->layout, sb->chunk_size);
1516 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1517 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1518 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1519 sb->failed_disks, sb->spare_disks,
1520 sb->sb_csum, (unsigned long)sb->events_lo);
1521
1522 printk(KERN_INFO);
1523 for (i = 0; i < MD_SB_DISKS; i++) {
1524 mdp_disk_t *desc;
1525
1526 desc = sb->disks + i;
1527 if (desc->number || desc->major || desc->minor ||
1528 desc->raid_disk || (desc->state && (desc->state != 4))) {
1529 printk(" D %2d: ", i);
1530 print_desc(desc);
1531 }
1532 }
1533 printk(KERN_INFO "md: THIS: ");
1534 print_desc(&sb->this_disk);
1535
1536}
1537
1538static void print_rdev(mdk_rdev_t *rdev)
1539{
1540 char b[BDEVNAME_SIZE];
1541 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1542 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1543 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1544 rdev->desc_nr);
1da177e4
LT		 1545 if (rdev->sb_loaded) {
1546 printk(KERN_INFO "md: rdev superblock:\n");
1547 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1548 } else
1549 printk(KERN_INFO "md: no rdev superblock!\n");
1550}
1551
5e56341d 1552static void md_print_devices(void)
1da177e4
LT		 1553{
1554 struct list_head *tmp, *tmp2;
1555 mdk_rdev_t *rdev;
1556 mddev_t *mddev;
1557 char b[BDEVNAME_SIZE];
1558
1559 printk("\n");
1560 printk("md: **********************************\n");
1561 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1562 printk("md: **********************************\n");
1563 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1564
32a7627c
N		 1565 if (mddev->bitmap)
1566 bitmap_print_sb(mddev->bitmap);
1567 else
1568 printk("%s: ", mdname(mddev));
1da177e4
LT		 1569 ITERATE_RDEV(mddev,rdev,tmp2)
1570 printk("<%s>", bdevname(rdev->bdev,b));
1571 printk("\n");
1572
1573 ITERATE_RDEV(mddev,rdev,tmp2)
1574 print_rdev(rdev);
1575 }
1576 printk("md: **********************************\n");
1577 printk("\n");
1578}
1579
1580
42543769 1581static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 1582{
42543769
N		 1583 /* Update each superblock (in-memory image), but
1584 * if we are allowed to, skip spares which already
1585 * have the right event counter, or have one earlier
1586 * (which would mean they aren't being marked as dirty
1587 * with the rest of the array)
1588 */
1da177e4
LT		 1589 mdk_rdev_t *rdev;
1590 struct list_head *tmp;
1591
1592 ITERATE_RDEV(mddev,rdev,tmp) {
42543769
N		 1593 if (rdev->sb_events == mddev->events ||
1594 (nospares &&
1595 rdev->raid_disk < 0 &&
1596 (rdev->sb_events&1)==0 &&
1597 rdev->sb_events+1 == mddev->events)) {
1598 /* Don't update this superblock */
1599 rdev->sb_loaded = 2;
1600 } else {
1601 super_types[mddev->major_version].
1602 sync_super(mddev, rdev);
1603 rdev->sb_loaded = 1;
1604 }
1da177e4
LT		 1605 }
1606}
1607
850b2b42 1608static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 1609{
7bfa19f2 1610 int err;
1da177e4
LT		 1611 struct list_head *tmp;
1612 mdk_rdev_t *rdev;
06d91a5f 1613 int sync_req;
42543769 1614 int nospares = 0;
1da177e4 1615
1da177e4 1616repeat:
a9701a30 1617 spin_lock_irq(&mddev->write_lock);
84692195 1618
850b2b42
N		 1619 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1620 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1621 force_change = 1;
1622 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1623 /* just a clean<-> dirty transition, possibly leave spares alone,
1624 * though if events isn't the right even/odd, we will have to do
1625 * spares after all
1626 */
1627 nospares = 1;
1628 if (force_change)
1629 nospares = 0;
1630 if (mddev->degraded)
84692195
N		 1631 /* If the array is degraded, then skipping spares is both
1632 * dangerous and fairly pointless.
1633 * Dangerous because a device that was removed from the array
1634 * might have a event_count that still looks up-to-date,
1635 * so it can be re-added without a resync.
1636 * Pointless because if there are any spares to skip,
1637 * then a recovery will happen and soon that array won't
1638 * be degraded any more and the spare can go back to sleep then.
1639 */
850b2b42 1640 nospares = 0;
84692195 1641
06d91a5f 1642 sync_req = mddev->in_sync;
1da177e4 1643 mddev->utime = get_seconds();
42543769
N		 1644
1645 /* If this is just a dirty<->clean transition, and the array is clean
1646 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 1647 if (nospares
42543769 1648 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1031be7a
N		 1649 && (mddev->events & 1)
1650 && mddev->events != 1)
42543769
N		 1651 mddev->events--;
1652 else {
1653 /* otherwise we have to go forward and ... */
1654 mddev->events ++;
1655 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1656 /* .. if the array isn't clean, insist on an odd 'events' */
1657 if ((mddev->events&1)==0) {
1658 mddev->events++;
1659 nospares = 0;
1660 }
1661 } else {
1662 /* otherwise insist on an even 'events' (for clean states) */
1663 if ((mddev->events&1)) {
1664 mddev->events++;
1665 nospares = 0;
1666 }
1667 }
1668 }
1da177e4
LT		 1669
1670 if (!mddev->events) {
1671 /*
1672 * oops, this 64-bit counter should never wrap.
1673 * Either we are in around ~1 trillion A.C., assuming
1674 * 1 reboot per second, or we have a bug:
1675 */
1676 MD_BUG();
1677 mddev->events --;
1678 }
42543769 1679 sync_sbs(mddev, nospares);
1da177e4
LT		 1680
1681 /*
1682 * do not write anything to disk if using
1683 * nonpersistent superblocks
1684 */
06d91a5f 1685 if (!mddev->persistent) {
850b2b42 1686 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1687 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1688 wake_up(&mddev->sb_wait);
1da177e4 1689 return;
06d91a5f 1690 }
a9701a30 1691 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1692
1693 dprintk(KERN_INFO
1694 "md: updating %s RAID superblock on device (in sync %d)\n",
1695 mdname(mddev),mddev->in_sync);
1696
32a7627c 1697 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1698 ITERATE_RDEV(mddev,rdev,tmp) {
1699 char b[BDEVNAME_SIZE];
1700 dprintk(KERN_INFO "md: ");
42543769
N		 1701 if (rdev->sb_loaded != 1)
1702 continue; /* no noise on spare devices */
b2d444d7 1703 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1704 dprintk("(skipping faulty ");
1705
1706 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1707 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1708 md_super_write(mddev,rdev,
0002b271 1709 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1710 rdev->sb_page);
1711 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1712 bdevname(rdev->bdev,b),
1713 (unsigned long long)rdev->sb_offset);
42543769 1714 rdev->sb_events = mddev->events;
7bfa19f2 1715
1da177e4
LT		 1716 } else
1717 dprintk(")\n");
7bfa19f2 1718 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1719 /* only need to write one superblock... */
1720 break;
1721 }
a9701a30 1722 md_super_wait(mddev);
850b2b42 1723 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 1724
a9701a30 1725 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 1726 if (mddev->in_sync != sync_req ||
1727 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 1728 /* have to write it out again */
a9701a30 1729 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1730 goto repeat;
1731 }
850b2b42 1732 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 1733 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1734 wake_up(&mddev->sb_wait);
06d91a5f 1735
1da177e4
LT		 1736}
1737
bce74dac
N		 1738/* words written to sysfs files may, or my not, be \n terminated.
1739 * We want to accept with case. For this we use cmd_match.
1740 */
1741static int cmd_match(const char *cmd, const char *str)
1742{
1743 /* See if cmd, written into a sysfs file, matches
1744 * str. They must either be the same, or cmd can
1745 * have a trailing newline
1746 */
1747 while (*cmd && *str && *cmd == *str) {
1748 cmd++;
1749 str++;
1750 }
1751 if (*cmd == '\n')
1752 cmd++;
1753 if (*str || *cmd)
1754 return 0;
1755 return 1;
1756}
1757
86e6ffdd
N		 1758struct rdev_sysfs_entry {
1759 struct attribute attr;
1760 ssize_t (*show)(mdk_rdev_t *, char *);
1761 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1762};
1763
1764static ssize_t
96de1e66 1765state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1766{
1767 char *sep = "";
1768 int len=0;
1769
b2d444d7 1770 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1771 len+= sprintf(page+len, "%sfaulty",sep);
1772 sep = ",";
1773 }
b2d444d7 1774 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1775 len += sprintf(page+len, "%sin_sync",sep);
1776 sep = ",";
1777 }
f655675b
N		 1778 if (test_bit(WriteMostly, &rdev->flags)) {
1779 len += sprintf(page+len, "%swrite_mostly",sep);
1780 sep = ",";
1781 }
b2d444d7
N		 1782 if (!test_bit(Faulty, &rdev->flags) &&
1783 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1784 len += sprintf(page+len, "%sspare", sep);
1785 sep = ",";
1786 }
1787 return len+sprintf(page+len, "\n");
1788}
1789
45dc2de1
N		 1790static ssize_t
1791state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1792{
1793 /* can write
1794 * faulty - simulates and error
1795 * remove - disconnects the device
f655675b
N		 1796 * writemostly - sets write_mostly
1797 * -writemostly - clears write_mostly
45dc2de1
N		 1798 */
1799 int err = -EINVAL;
1800 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1801 md_error(rdev->mddev, rdev);
1802 err = 0;
1803 } else if (cmd_match(buf, "remove")) {
1804 if (rdev->raid_disk >= 0)
1805 err = -EBUSY;
1806 else {
1807 mddev_t *mddev = rdev->mddev;
1808 kick_rdev_from_array(rdev);
3f9d7b0d
N		 1809 if (mddev->pers)
1810 md_update_sb(mddev, 1);
45dc2de1
N		 1811 md_new_event(mddev);
1812 err = 0;
1813 }
f655675b
N		 1814 } else if (cmd_match(buf, "writemostly")) {
1815 set_bit(WriteMostly, &rdev->flags);
1816 err = 0;
1817 } else if (cmd_match(buf, "-writemostly")) {
1818 clear_bit(WriteMostly, &rdev->flags);
1819 err = 0;
45dc2de1
N		 1820 }
1821 return err ? err : len;
1822}
80ca3a44
N		 1823static struct rdev_sysfs_entry rdev_state =
1824__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd
N		 1825
1826static ssize_t
96de1e66 1827super_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 1828{
1829 if (rdev->sb_loaded && rdev->sb_size) {
1830 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1831 return rdev->sb_size;
1832 } else
1833 return 0;
1834}
96de1e66
N		 1835static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1836
4dbcdc75
N		 1837static ssize_t
1838errors_show(mdk_rdev_t *rdev, char *page)
1839{
1840 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1841}
1842
1843static ssize_t
1844errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1845{
1846 char *e;
1847 unsigned long n = simple_strtoul(buf, &e, 10);
1848 if (*buf && (*e == 0 || *e == '\n')) {
1849 atomic_set(&rdev->corrected_errors, n);
1850 return len;
1851 }
1852 return -EINVAL;
1853}
1854static struct rdev_sysfs_entry rdev_errors =
80ca3a44 1855__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 1856
014236d2
N		 1857static ssize_t
1858slot_show(mdk_rdev_t *rdev, char *page)
1859{
1860 if (rdev->raid_disk < 0)
1861 return sprintf(page, "none\n");
1862 else
1863 return sprintf(page, "%d\n", rdev->raid_disk);
1864}
1865
1866static ssize_t
1867slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1868{
1869 char *e;
1870 int slot = simple_strtoul(buf, &e, 10);
1871 if (strncmp(buf, "none", 4)==0)
1872 slot = -1;
1873 else if (e==buf || (*e && *e!= '\n'))
1874 return -EINVAL;
1875 if (rdev->mddev->pers)
1876 /* Cannot set slot in active array (yet) */
1877 return -EBUSY;
1878 if (slot >= rdev->mddev->raid_disks)
1879 return -ENOSPC;
1880 rdev->raid_disk = slot;
1881 /* assume it is working */
1882 rdev->flags = 0;
1883 set_bit(In_sync, &rdev->flags);
1884 return len;
1885}
1886
1887
1888static struct rdev_sysfs_entry rdev_slot =
80ca3a44 1889__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 1890
93c8cad0
N		 1891static ssize_t
1892offset_show(mdk_rdev_t *rdev, char *page)
1893{
6961ece4 1894 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 1895}
1896
1897static ssize_t
1898offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1899{
1900 char *e;
1901 unsigned long long offset = simple_strtoull(buf, &e, 10);
1902 if (e==buf || (*e && *e != '\n'))
1903 return -EINVAL;
1904 if (rdev->mddev->pers)
1905 return -EBUSY;
1906 rdev->data_offset = offset;
1907 return len;
1908}
1909
1910static struct rdev_sysfs_entry rdev_offset =
80ca3a44 1911__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 1912
83303b61
N		 1913static ssize_t
1914rdev_size_show(mdk_rdev_t *rdev, char *page)
1915{
1916 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1917}
1918
1919static ssize_t
1920rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1921{
1922 char *e;
1923 unsigned long long size = simple_strtoull(buf, &e, 10);
1924 if (e==buf || (*e && *e != '\n'))
1925 return -EINVAL;
1926 if (rdev->mddev->pers)
1927 return -EBUSY;
1928 rdev->size = size;
1929 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1930 rdev->mddev->size = size;
1931 return len;
1932}
1933
1934static struct rdev_sysfs_entry rdev_size =
80ca3a44 1935__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 1936
86e6ffdd
N		 1937static struct attribute *rdev_default_attrs[] = {
1938 &rdev_state.attr,
1939 &rdev_super.attr,
4dbcdc75 1940 &rdev_errors.attr,
014236d2 1941 &rdev_slot.attr,
93c8cad0 1942 &rdev_offset.attr,
83303b61 1943 &rdev_size.attr,
86e6ffdd
N		 1944 NULL,
1945};
1946static ssize_t
1947rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1948{
1949 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1950 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1951
1952 if (!entry->show)
1953 return -EIO;
1954 return entry->show(rdev, page);
1955}
1956
1957static ssize_t
1958rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1959 const char *page, size_t length)
1960{
1961 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1962 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1963
1964 if (!entry->store)
1965 return -EIO;
67463acb
N		 1966 if (!capable(CAP_SYS_ADMIN))
1967 return -EACCES;
86e6ffdd
N		 1968 return entry->store(rdev, page, length);
1969}
1970
1971static void rdev_free(struct kobject *ko)
1972{
1973 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1974 kfree(rdev);
1975}
1976static struct sysfs_ops rdev_sysfs_ops = {
1977 .show = rdev_attr_show,
1978 .store = rdev_attr_store,
1979};
1980static struct kobj_type rdev_ktype = {
1981 .release = rdev_free,
1982 .sysfs_ops = &rdev_sysfs_ops,
1983 .default_attrs = rdev_default_attrs,
1984};
1985
1da177e4
LT		 1986/*
1987 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1988 *
1989 * mark the device faulty if:
1990 *
1991 * - the device is nonexistent (zero size)
1992 * - the device has no valid superblock
1993 *
1994 * a faulty rdev _never_ has rdev->sb set.
1995 */
1996static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1997{
1998 char b[BDEVNAME_SIZE];
1999 int err;
2000 mdk_rdev_t *rdev;
2001 sector_t size;
2002
9ffae0cf 2003 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2004 if (!rdev) {
2005 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2006 return ERR_PTR(-ENOMEM);
2007 }
1da177e4
LT		 2008
2009 if ((err = alloc_disk_sb(rdev)))
2010 goto abort_free;
2011
2012 err = lock_rdev(rdev, newdev);
2013 if (err)
2014 goto abort_free;
2015
86e6ffdd
N		 2016 rdev->kobj.parent = NULL;
2017 rdev->kobj.ktype = &rdev_ktype;
2018 kobject_init(&rdev->kobj);
2019
1da177e4 2020 rdev->desc_nr = -1;
2b6e8459 2021 rdev->saved_raid_disk = -1;
3f9d7b0d 2022 rdev->raid_disk = -1;
b2d444d7 2023 rdev->flags = 0;
1da177e4 2024 rdev->data_offset = 0;
42543769 2025 rdev->sb_events = 0;
1da177e4 2026 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 2027 atomic_set(&rdev->read_errors, 0);
4dbcdc75 2028 atomic_set(&rdev->corrected_errors, 0);
1da177e4
LT		 2029
2030 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2031 if (!size) {
2032 printk(KERN_WARNING
2033 "md: %s has zero or unknown size, marking faulty!\n",
2034 bdevname(rdev->bdev,b));
2035 err = -EINVAL;
2036 goto abort_free;
2037 }
2038
2039 if (super_format >= 0) {
2040 err = super_types[super_format].
2041 load_super(rdev, NULL, super_minor);
2042 if (err == -EINVAL) {
2043 printk(KERN_WARNING
2044 "md: %s has invalid sb, not importing!\n",
2045 bdevname(rdev->bdev,b));
2046 goto abort_free;
2047 }
2048 if (err < 0) {
2049 printk(KERN_WARNING
2050 "md: could not read %s's sb, not importing!\n",
2051 bdevname(rdev->bdev,b));
2052 goto abort_free;
2053 }
2054 }
2055 INIT_LIST_HEAD(&rdev->same_set);
2056
2057 return rdev;
2058
2059abort_free:
2060 if (rdev->sb_page) {
2061 if (rdev->bdev)
2062 unlock_rdev(rdev);
2063 free_disk_sb(rdev);
2064 }
2065 kfree(rdev);
2066 return ERR_PTR(err);
2067}
2068
2069/*
2070 * Check a full RAID array for plausibility
2071 */
2072
2073
a757e64c 2074static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2075{
2076 int i;
2077 struct list_head *tmp;
2078 mdk_rdev_t *rdev, *freshest;
2079 char b[BDEVNAME_SIZE];
2080
2081 freshest = NULL;
2082 ITERATE_RDEV(mddev,rdev,tmp)
2083 switch (super_types[mddev->major_version].
2084 load_super(rdev, freshest, mddev->minor_version)) {
2085 case 1:
2086 freshest = rdev;
2087 break;
2088 case 0:
2089 break;
2090 default:
2091 printk( KERN_ERR \
2092 "md: fatal superblock inconsistency in %s"
2093 " -- removing from array\n",
2094 bdevname(rdev->bdev,b));
2095 kick_rdev_from_array(rdev);
2096 }
2097
2098
2099 super_types[mddev->major_version].
2100 validate_super(mddev, freshest);
2101
2102 i = 0;
2103 ITERATE_RDEV(mddev,rdev,tmp) {
2104 if (rdev != freshest)
2105 if (super_types[mddev->major_version].
2106 validate_super(mddev, rdev)) {
2107 printk(KERN_WARNING "md: kicking non-fresh %s"
2108 " from array!\n",
2109 bdevname(rdev->bdev,b));
2110 kick_rdev_from_array(rdev);
2111 continue;
2112 }
2113 if (mddev->level == LEVEL_MULTIPATH) {
2114 rdev->desc_nr = i++;
2115 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2116 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 2117 }
2118 }
2119
2120
2121
2122 if (mddev->recovery_cp != MaxSector &&
2123 mddev->level >= 1)
2124 printk(KERN_ERR "md: %s: raid array is not clean"
2125 " -- starting background reconstruction\n",
2126 mdname(mddev));
2127
1da177e4
LT		 2128}
2129
16f17b39
N		 2130static ssize_t
2131safe_delay_show(mddev_t *mddev, char *page)
2132{
2133 int msec = (mddev->safemode_delay*1000)/HZ;
2134 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2135}
2136static ssize_t
2137safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2138{
2139 int scale=1;
2140 int dot=0;
2141 int i;
2142 unsigned long msec;
2143 char buf[30];
2144 char *e;
2145 /* remove a period, and count digits after it */
2146 if (len >= sizeof(buf))
2147 return -EINVAL;
2148 strlcpy(buf, cbuf, len);
2149 buf[len] = 0;
2150 for (i=0; i<len; i++) {
2151 if (dot) {
2152 if (isdigit(buf[i])) {
2153 buf[i-1] = buf[i];
2154 scale *= 10;
2155 }
2156 buf[i] = 0;
2157 } else if (buf[i] == '.') {
2158 dot=1;
2159 buf[i] = 0;
2160 }
2161 }
2162 msec = simple_strtoul(buf, &e, 10);
2163 if (e == buf || (*e && *e != '\n'))
2164 return -EINVAL;
2165 msec = (msec * 1000) / scale;
2166 if (msec == 0)
2167 mddev->safemode_delay = 0;
2168 else {
2169 mddev->safemode_delay = (msec*HZ)/1000;
2170 if (mddev->safemode_delay == 0)
2171 mddev->safemode_delay = 1;
2172 }
2173 return len;
2174}
2175static struct md_sysfs_entry md_safe_delay =
80ca3a44 2176__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2177
eae1701f 2178static ssize_t
96de1e66 2179level_show(mddev_t *mddev, char *page)
eae1701f 2180{
2604b703 2181 struct mdk_personality *p = mddev->pers;
d9d166c2 2182 if (p)
eae1701f 2183 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2184 else if (mddev->clevel[0])
2185 return sprintf(page, "%s\n", mddev->clevel);
2186 else if (mddev->level != LEVEL_NONE)
2187 return sprintf(page, "%d\n", mddev->level);
2188 else
2189 return 0;
eae1701f
N		 2190}
2191
d9d166c2
N		 2192static ssize_t
2193level_store(mddev_t *mddev, const char *buf, size_t len)
2194{
2195 int rv = len;
2196 if (mddev->pers)
2197 return -EBUSY;
2198 if (len == 0)
2199 return 0;
2200 if (len >= sizeof(mddev->clevel))
2201 return -ENOSPC;
2202 strncpy(mddev->clevel, buf, len);
2203 if (mddev->clevel[len-1] == '\n')
2204 len--;
2205 mddev->clevel[len] = 0;
2206 mddev->level = LEVEL_NONE;
2207 return rv;
2208}
2209
2210static struct md_sysfs_entry md_level =
80ca3a44 2211__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 2212
d4dbd025
N		 2213
2214static ssize_t
2215layout_show(mddev_t *mddev, char *page)
2216{
2217 /* just a number, not meaningful for all levels */
2218 return sprintf(page, "%d\n", mddev->layout);
2219}
2220
2221static ssize_t
2222layout_store(mddev_t *mddev, const char *buf, size_t len)
2223{
2224 char *e;
2225 unsigned long n = simple_strtoul(buf, &e, 10);
2226 if (mddev->pers)
2227 return -EBUSY;
2228
2229 if (!*buf || (*e && *e != '\n'))
2230 return -EINVAL;
2231
2232 mddev->layout = n;
2233 return len;
2234}
2235static struct md_sysfs_entry md_layout =
80ca3a44 2236__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 2237
2238
eae1701f 2239static ssize_t
96de1e66 2240raid_disks_show(mddev_t *mddev, char *page)
eae1701f 2241{
bb636547
N		 2242 if (mddev->raid_disks == 0)
2243 return 0;
eae1701f
N		 2244 return sprintf(page, "%d\n", mddev->raid_disks);
2245}
2246
da943b99
N		 2247static int update_raid_disks(mddev_t *mddev, int raid_disks);
2248
2249static ssize_t
2250raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2251{
da943b99
N		 2252 char *e;
2253 int rv = 0;
2254 unsigned long n = simple_strtoul(buf, &e, 10);
2255
2256 if (!*buf || (*e && *e != '\n'))
2257 return -EINVAL;
2258
2259 if (mddev->pers)
2260 rv = update_raid_disks(mddev, n);
2261 else
2262 mddev->raid_disks = n;
2263 return rv ? rv : len;
2264}
2265static struct md_sysfs_entry md_raid_disks =
80ca3a44 2266__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 2267
3b34380a
N		 2268static ssize_t
2269chunk_size_show(mddev_t *mddev, char *page)
2270{
2271 return sprintf(page, "%d\n", mddev->chunk_size);
2272}
2273
2274static ssize_t
2275chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2276{
2277 /* can only set chunk_size if array is not yet active */
2278 char *e;
2279 unsigned long n = simple_strtoul(buf, &e, 10);
2280
2281 if (mddev->pers)
2282 return -EBUSY;
2283 if (!*buf || (*e && *e != '\n'))
2284 return -EINVAL;
2285
2286 mddev->chunk_size = n;
2287 return len;
2288}
2289static struct md_sysfs_entry md_chunk_size =
80ca3a44 2290__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 2291
a94213b1
N		 2292static ssize_t
2293resync_start_show(mddev_t *mddev, char *page)
2294{
2295 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2296}
2297
2298static ssize_t
2299resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2300{
2301 /* can only set chunk_size if array is not yet active */
2302 char *e;
2303 unsigned long long n = simple_strtoull(buf, &e, 10);
2304
2305 if (mddev->pers)
2306 return -EBUSY;
2307 if (!*buf || (*e && *e != '\n'))
2308 return -EINVAL;
2309
2310 mddev->recovery_cp = n;
2311 return len;
2312}
2313static struct md_sysfs_entry md_resync_start =
80ca3a44 2314__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 2315
9e653b63
N		 2316/*
2317 * The array state can be:
2318 *
2319 * clear
2320 * No devices, no size, no level
2321 * Equivalent to STOP_ARRAY ioctl
2322 * inactive
2323 * May have some settings, but array is not active
2324 * all IO results in error
2325 * When written, doesn't tear down array, but just stops it
2326 * suspended (not supported yet)
2327 * All IO requests will block. The array can be reconfigured.
2328 * Writing this, if accepted, will block until array is quiessent
2329 * readonly
2330 * no resync can happen. no superblocks get written.
2331 * write requests fail
2332 * read-auto
2333 * like readonly, but behaves like 'clean' on a write request.
2334 *
2335 * clean - no pending writes, but otherwise active.
2336 * When written to inactive array, starts without resync
2337 * If a write request arrives then
2338 * if metadata is known, mark 'dirty' and switch to 'active'.
2339 * if not known, block and switch to write-pending
2340 * If written to an active array that has pending writes, then fails.
2341 * active
2342 * fully active: IO and resync can be happening.
2343 * When written to inactive array, starts with resync
2344 *
2345 * write-pending
2346 * clean, but writes are blocked waiting for 'active' to be written.
2347 *
2348 * active-idle
2349 * like active, but no writes have been seen for a while (100msec).
2350 *
2351 */
2352enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2353 write_pending, active_idle, bad_word};
05381954 2354static char *array_states[] = {
9e653b63
N		 2355 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2356 "write-pending", "active-idle", NULL };
2357
2358static int match_word(const char *word, char **list)
2359{
2360 int n;
2361 for (n=0; list[n]; n++)
2362 if (cmd_match(word, list[n]))
2363 break;
2364 return n;
2365}
2366
2367static ssize_t
2368array_state_show(mddev_t *mddev, char *page)
2369{
2370 enum array_state st = inactive;
2371
2372 if (mddev->pers)
2373 switch(mddev->ro) {
2374 case 1:
2375 st = readonly;
2376 break;
2377 case 2:
2378 st = read_auto;
2379 break;
2380 case 0:
2381 if (mddev->in_sync)
2382 st = clean;
2383 else if (mddev->safemode)
2384 st = active_idle;
2385 else
2386 st = active;
2387 }
2388 else {
2389 if (list_empty(&mddev->disks) &&
2390 mddev->raid_disks == 0 &&
2391 mddev->size == 0)
2392 st = clear;
2393 else
2394 st = inactive;
2395 }
2396 return sprintf(page, "%s\n", array_states[st]);
2397}
2398
2399static int do_md_stop(mddev_t * mddev, int ro);
2400static int do_md_run(mddev_t * mddev);
2401static int restart_array(mddev_t *mddev);
2402
2403static ssize_t
2404array_state_store(mddev_t *mddev, const char *buf, size_t len)
2405{
2406 int err = -EINVAL;
2407 enum array_state st = match_word(buf, array_states);
2408 switch(st) {
2409 case bad_word:
2410 break;
2411 case clear:
2412 /* stopping an active array */
2413 if (mddev->pers) {
2414 if (atomic_read(&mddev->active) > 1)
2415 return -EBUSY;
2416 err = do_md_stop(mddev, 0);
2417 }
2418 break;
2419 case inactive:
2420 /* stopping an active array */
2421 if (mddev->pers) {
2422 if (atomic_read(&mddev->active) > 1)
2423 return -EBUSY;
2424 err = do_md_stop(mddev, 2);
2425 }
2426 break;
2427 case suspended:
2428 break; /* not supported yet */
2429 case readonly:
2430 if (mddev->pers)
2431 err = do_md_stop(mddev, 1);
2432 else {
2433 mddev->ro = 1;
2434 err = do_md_run(mddev);
2435 }
2436 break;
2437 case read_auto:
2438 /* stopping an active array */
2439 if (mddev->pers) {
2440 err = do_md_stop(mddev, 1);
2441 if (err == 0)
2442 mddev->ro = 2; /* FIXME mark devices writable */
2443 } else {
2444 mddev->ro = 2;
2445 err = do_md_run(mddev);
2446 }
2447 break;
2448 case clean:
2449 if (mddev->pers) {
2450 restart_array(mddev);
2451 spin_lock_irq(&mddev->write_lock);
2452 if (atomic_read(&mddev->writes_pending) == 0) {
2453 mddev->in_sync = 1;
850b2b42 2454 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2455 }
2456 spin_unlock_irq(&mddev->write_lock);
2457 } else {
2458 mddev->ro = 0;
2459 mddev->recovery_cp = MaxSector;
2460 err = do_md_run(mddev);
2461 }
2462 break;
2463 case active:
2464 if (mddev->pers) {
2465 restart_array(mddev);
850b2b42 2466 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
9e653b63
N		 2467 wake_up(&mddev->sb_wait);
2468 err = 0;
2469 } else {
2470 mddev->ro = 0;
2471 err = do_md_run(mddev);
2472 }
2473 break;
2474 case write_pending:
2475 case active_idle:
2476 /* these cannot be set */
2477 break;
2478 }
2479 if (err)
2480 return err;
2481 else
2482 return len;
2483}
80ca3a44
N		 2484static struct md_sysfs_entry md_array_state =
2485__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 2486
6d7ff738
N		 2487static ssize_t
2488null_show(mddev_t *mddev, char *page)
2489{
2490 return -EINVAL;
2491}
2492
2493static ssize_t
2494new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2495{
2496 /* buf must be %d:%d\n? giving major and minor numbers */
2497 /* The new device is added to the array.
2498 * If the array has a persistent superblock, we read the
2499 * superblock to initialise info and check validity.
2500 * Otherwise, only checking done is that in bind_rdev_to_array,
2501 * which mainly checks size.
2502 */
2503 char *e;
2504 int major = simple_strtoul(buf, &e, 10);
2505 int minor;
2506 dev_t dev;
2507 mdk_rdev_t *rdev;
2508 int err;
2509
2510 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2511 return -EINVAL;
2512 minor = simple_strtoul(e+1, &e, 10);
2513 if (*e && *e != '\n')
2514 return -EINVAL;
2515 dev = MKDEV(major, minor);
2516 if (major != MAJOR(dev) ||
2517 minor != MINOR(dev))
2518 return -EOVERFLOW;
2519
2520
2521 if (mddev->persistent) {
2522 rdev = md_import_device(dev, mddev->major_version,
2523 mddev->minor_version);
2524 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2525 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2526 mdk_rdev_t, same_set);
2527 err = super_types[mddev->major_version]
2528 .load_super(rdev, rdev0, mddev->minor_version);
2529 if (err < 0)
2530 goto out;
2531 }
2532 } else
2533 rdev = md_import_device(dev, -1, -1);
2534
2535 if (IS_ERR(rdev))
2536 return PTR_ERR(rdev);
2537 err = bind_rdev_to_array(rdev, mddev);
2538 out:
2539 if (err)
2540 export_rdev(rdev);
2541 return err ? err : len;
2542}
2543
2544static struct md_sysfs_entry md_new_device =
80ca3a44 2545__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 2546
9b1d1dac
PC		 2547static ssize_t
2548bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2549{
2550 char *end;
2551 unsigned long chunk, end_chunk;
2552
2553 if (!mddev->bitmap)
2554 goto out;
2555 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2556 while (*buf) {
2557 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2558 if (buf == end) break;
2559 if (*end == '-') { /* range */
2560 buf = end + 1;
2561 end_chunk = simple_strtoul(buf, &end, 0);
2562 if (buf == end) break;
2563 }
2564 if (*end && !isspace(*end)) break;
2565 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2566 buf = end;
2567 while (isspace(*buf)) buf++;
2568 }
2569 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2570out:
2571 return len;
2572}
2573
2574static struct md_sysfs_entry md_bitmap =
2575__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2576
a35b0d69
N		 2577static ssize_t
2578size_show(mddev_t *mddev, char *page)
2579{
2580 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2581}
2582
2583static int update_size(mddev_t *mddev, unsigned long size);
2584
2585static ssize_t
2586size_store(mddev_t *mddev, const char *buf, size_t len)
2587{
2588 /* If array is inactive, we can reduce the component size, but
2589 * not increase it (except from 0).
2590 * If array is active, we can try an on-line resize
2591 */
2592 char *e;
2593 int err = 0;
2594 unsigned long long size = simple_strtoull(buf, &e, 10);
2595 if (!*buf || *buf == '\n' ||
2596 (*e && *e != '\n'))
2597 return -EINVAL;
2598
2599 if (mddev->pers) {
2600 err = update_size(mddev, size);
850b2b42 2601 md_update_sb(mddev, 1);
a35b0d69
N		 2602 } else {
2603 if (mddev->size == 0 ||
2604 mddev->size > size)
2605 mddev->size = size;
2606 else
2607 err = -ENOSPC;
2608 }
2609 return err ? err : len;
2610}
2611
2612static struct md_sysfs_entry md_size =
80ca3a44 2613__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 2614
8bb93aac
N		 2615
2616/* Metdata version.
2617 * This is either 'none' for arrays with externally managed metadata,
2618 * or N.M for internally known formats
2619 */
2620static ssize_t
2621metadata_show(mddev_t *mddev, char *page)
2622{
2623 if (mddev->persistent)
2624 return sprintf(page, "%d.%d\n",
2625 mddev->major_version, mddev->minor_version);
2626 else
2627 return sprintf(page, "none\n");
2628}
2629
2630static ssize_t
2631metadata_store(mddev_t *mddev, const char *buf, size_t len)
2632{
2633 int major, minor;
2634 char *e;
2635 if (!list_empty(&mddev->disks))
2636 return -EBUSY;
2637
2638 if (cmd_match(buf, "none")) {
2639 mddev->persistent = 0;
2640 mddev->major_version = 0;
2641 mddev->minor_version = 90;
2642 return len;
2643 }
2644 major = simple_strtoul(buf, &e, 10);
2645 if (e==buf || *e != '.')
2646 return -EINVAL;
2647 buf = e+1;
2648 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 2649 if (e==buf || (*e && *e != '\n') )
8bb93aac 2650 return -EINVAL;
50511da3 2651 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 2652 return -ENOENT;
2653 mddev->major_version = major;
2654 mddev->minor_version = minor;
2655 mddev->persistent = 1;
2656 return len;
2657}
2658
2659static struct md_sysfs_entry md_metadata =
80ca3a44 2660__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 2661
24dd469d 2662static ssize_t
7eec314d 2663action_show(mddev_t *mddev, char *page)
24dd469d 2664{
7eec314d 2665 char *type = "idle";
31399d9e
N		 2666 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2667 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
ccfcc3c1
N		 2668 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2669 type = "reshape";
2670 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 2671 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2672 type = "resync";
2673 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2674 type = "check";
2675 else
2676 type = "repair";
2677 } else
2678 type = "recover";
2679 }
2680 return sprintf(page, "%s\n", type);
2681}
2682
2683static ssize_t
7eec314d 2684action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 2685{
7eec314d
N		 2686 if (!mddev->pers || !mddev->pers->sync_request)
2687 return -EINVAL;
2688
bce74dac 2689 if (cmd_match(page, "idle")) {
7eec314d
N		 2690 if (mddev->sync_thread) {
2691 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2692 md_unregister_thread(mddev->sync_thread);
2693 mddev->sync_thread = NULL;
2694 mddev->recovery = 0;
2695 }
03c902e1
N		 2696 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2697 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 2698 return -EBUSY;
03c902e1 2699 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
7eec314d 2700 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
16484bf5
N		 2701 else if (cmd_match(page, "reshape")) {
2702 int err;
2703 if (mddev->pers->start_reshape == NULL)
2704 return -EINVAL;
2705 err = mddev->pers->start_reshape(mddev);
2706 if (err)
2707 return err;
2708 } else {
bce74dac 2709 if (cmd_match(page, "check"))
7eec314d 2710 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 2711 else if (!cmd_match(page, "repair"))
7eec314d
N		 2712 return -EINVAL;
2713 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2714 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 2715 }
03c902e1 2716 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 2717 md_wakeup_thread(mddev->thread);
2718 return len;
2719}
2720
9d88883e 2721static ssize_t
96de1e66 2722mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 2723{
2724 return sprintf(page, "%llu\n",
2725 (unsigned long long) mddev->resync_mismatches);
2726}
2727
80ca3a44
N		 2728static struct md_sysfs_entry md_scan_mode =
2729__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 2730
96de1e66 2731
80ca3a44 2732static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 2733
88202a0c
N		 2734static ssize_t
2735sync_min_show(mddev_t *mddev, char *page)
2736{
2737 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2738 mddev->sync_speed_min ? "local": "system");
2739}
2740
2741static ssize_t
2742sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2743{
2744 int min;
2745 char *e;
2746 if (strncmp(buf, "system", 6)==0) {
2747 mddev->sync_speed_min = 0;
2748 return len;
2749 }
2750 min = simple_strtoul(buf, &e, 10);
2751 if (buf == e || (*e && *e != '\n') || min <= 0)
2752 return -EINVAL;
2753 mddev->sync_speed_min = min;
2754 return len;
2755}
2756
2757static struct md_sysfs_entry md_sync_min =
2758__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2759
2760static ssize_t
2761sync_max_show(mddev_t *mddev, char *page)
2762{
2763 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2764 mddev->sync_speed_max ? "local": "system");
2765}
2766
2767static ssize_t
2768sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2769{
2770 int max;
2771 char *e;
2772 if (strncmp(buf, "system", 6)==0) {
2773 mddev->sync_speed_max = 0;
2774 return len;
2775 }
2776 max = simple_strtoul(buf, &e, 10);
2777 if (buf == e || (*e && *e != '\n') || max <= 0)
2778 return -EINVAL;
2779 mddev->sync_speed_max = max;
2780 return len;
2781}
2782
2783static struct md_sysfs_entry md_sync_max =
2784__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2785
2786
2787static ssize_t
2788sync_speed_show(mddev_t *mddev, char *page)
2789{
2790 unsigned long resync, dt, db;
ff4e8d9a 2791 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
88202a0c
N		 2792 dt = ((jiffies - mddev->resync_mark) / HZ);
2793 if (!dt) dt++;
2794 db = resync - (mddev->resync_mark_cnt);
2795 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2796}
2797
80ca3a44 2798static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 2799
2800static ssize_t
2801sync_completed_show(mddev_t *mddev, char *page)
2802{
2803 unsigned long max_blocks, resync;
2804
2805 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2806 max_blocks = mddev->resync_max_sectors;
2807 else
2808 max_blocks = mddev->size << 1;
2809
2810 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2811 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2812}
2813
80ca3a44 2814static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 2815
e464eafd
N		 2816static ssize_t
2817suspend_lo_show(mddev_t *mddev, char *page)
2818{
2819 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2820}
2821
2822static ssize_t
2823suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2824{
2825 char *e;
2826 unsigned long long new = simple_strtoull(buf, &e, 10);
2827
2828 if (mddev->pers->quiesce == NULL)
2829 return -EINVAL;
2830 if (buf == e || (*e && *e != '\n'))
2831 return -EINVAL;
2832 if (new >= mddev->suspend_hi ||
2833 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2834 mddev->suspend_lo = new;
2835 mddev->pers->quiesce(mddev, 2);
2836 return len;
2837 } else
2838 return -EINVAL;
2839}
2840static struct md_sysfs_entry md_suspend_lo =
2841__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2842
2843
2844static ssize_t
2845suspend_hi_show(mddev_t *mddev, char *page)
2846{
2847 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2848}
2849
2850static ssize_t
2851suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2852{
2853 char *e;
2854 unsigned long long new = simple_strtoull(buf, &e, 10);
2855
2856 if (mddev->pers->quiesce == NULL)
2857 return -EINVAL;
2858 if (buf == e || (*e && *e != '\n'))
2859 return -EINVAL;
2860 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2861 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2862 mddev->suspend_hi = new;
2863 mddev->pers->quiesce(mddev, 1);
2864 mddev->pers->quiesce(mddev, 0);
2865 return len;
2866 } else
2867 return -EINVAL;
2868}
2869static struct md_sysfs_entry md_suspend_hi =
2870__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2871
2872
eae1701f
N		 2873static struct attribute *md_default_attrs[] = {
2874 &md_level.attr,
d4dbd025 2875 &md_layout.attr,
eae1701f 2876 &md_raid_disks.attr,
3b34380a 2877 &md_chunk_size.attr,
a35b0d69 2878 &md_size.attr,
a94213b1 2879 &md_resync_start.attr,
8bb93aac 2880 &md_metadata.attr,
6d7ff738 2881 &md_new_device.attr,
16f17b39 2882 &md_safe_delay.attr,
9e653b63 2883 &md_array_state.attr,
411036fa
N		 2884 NULL,
2885};
2886
2887static struct attribute *md_redundancy_attrs[] = {
24dd469d 2888 &md_scan_mode.attr,
9d88883e 2889 &md_mismatches.attr,
88202a0c
N		 2890 &md_sync_min.attr,
2891 &md_sync_max.attr,
2892 &md_sync_speed.attr,
2893 &md_sync_completed.attr,
e464eafd
N		 2894 &md_suspend_lo.attr,
2895 &md_suspend_hi.attr,
9b1d1dac 2896 &md_bitmap.attr,
eae1701f
N		 2897 NULL,
2898};
411036fa
N		 2899static struct attribute_group md_redundancy_group = {
2900 .name = NULL,
2901 .attrs = md_redundancy_attrs,
2902};
2903
eae1701f
N		 2904
2905static ssize_t
2906md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2907{
2908 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2909 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2910 ssize_t rv;
eae1701f
N		 2911
2912 if (!entry->show)
2913 return -EIO;
5dc5cf7d
IM		 2914 rv = mddev_lock(mddev);
2915 if (!rv) {
2916 rv = entry->show(mddev, page);
2917 mddev_unlock(mddev);
2918 }
96de1e66 2919 return rv;
eae1701f
N		 2920}
2921
2922static ssize_t
2923md_attr_store(struct kobject *kobj, struct attribute *attr,
2924 const char *page, size_t length)
2925{
2926 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2927 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 2928 ssize_t rv;
eae1701f
N		 2929
2930 if (!entry->store)
2931 return -EIO;
67463acb
N		 2932 if (!capable(CAP_SYS_ADMIN))
2933 return -EACCES;
5dc5cf7d
IM		 2934 rv = mddev_lock(mddev);
2935 if (!rv) {
2936 rv = entry->store(mddev, page, length);
2937 mddev_unlock(mddev);
2938 }
96de1e66 2939 return rv;
eae1701f
N		 2940}
2941
2942static void md_free(struct kobject *ko)
2943{
2944 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2945 kfree(mddev);
2946}
2947
2948static struct sysfs_ops md_sysfs_ops = {
2949 .show = md_attr_show,
2950 .store = md_attr_store,
2951};
2952static struct kobj_type md_ktype = {
2953 .release = md_free,
2954 .sysfs_ops = &md_sysfs_ops,
2955 .default_attrs = md_default_attrs,
2956};
2957
1da177e4
LT		 2958int mdp_major = 0;
2959
2960static struct kobject *md_probe(dev_t dev, int *part, void *data)
2961{
48c9c27b 2962 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 2963 mddev_t *mddev = mddev_find(dev);
2964 struct gendisk *disk;
2965 int partitioned = (MAJOR(dev) != MD_MAJOR);
2966 int shift = partitioned ? MdpMinorShift : 0;
2967 int unit = MINOR(dev) >> shift;
2968
2969 if (!mddev)
2970 return NULL;
2971
48c9c27b 2972 mutex_lock(&disks_mutex);
1da177e4 2973 if (mddev->gendisk) {
48c9c27b 2974 mutex_unlock(&disks_mutex);
1da177e4
LT		 2975 mddev_put(mddev);
2976 return NULL;
2977 }
2978 disk = alloc_disk(1 << shift);
2979 if (!disk) {
48c9c27b 2980 mutex_unlock(&disks_mutex);
1da177e4
LT		 2981 mddev_put(mddev);
2982 return NULL;
2983 }
2984 disk->major = MAJOR(dev);
2985 disk->first_minor = unit << shift;
ce7b0f46 2986 if (partitioned)
1da177e4 2987 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 2988 else
1da177e4 2989 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 2990 disk->fops = &md_fops;
2991 disk->private_data = mddev;
2992 disk->queue = mddev->queue;
2993 add_disk(disk);
2994 mddev->gendisk = disk;
48c9c27b 2995 mutex_unlock(&disks_mutex);
9c791977 2996 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 2997 mddev->kobj.k_name = NULL;
2998 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2999 mddev->kobj.ktype = &md_ktype;
5e55e2f5
N		 3000 if (kobject_register(&mddev->kobj))
3001 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3002 disk->disk_name);
1da177e4
LT		 3003 return NULL;
3004}
3005
1da177e4
LT		 3006static void md_safemode_timeout(unsigned long data)
3007{
3008 mddev_t *mddev = (mddev_t *) data;
3009
3010 mddev->safemode = 1;
3011 md_wakeup_thread(mddev->thread);
3012}
3013
6ff8d8ec 3014static int start_dirty_degraded;
1da177e4
LT		 3015
3016static int do_md_run(mddev_t * mddev)
3017{
2604b703 3018 int err;
1da177e4
LT		 3019 int chunk_size;
3020 struct list_head *tmp;
3021 mdk_rdev_t *rdev;
3022 struct gendisk *disk;
2604b703 3023 struct mdk_personality *pers;
1da177e4
LT		 3024 char b[BDEVNAME_SIZE];
3025
a757e64c
N		 3026 if (list_empty(&mddev->disks))
3027 /* cannot run an array with no devices.. */
1da177e4 3028 return -EINVAL;
1da177e4
LT		 3029
3030 if (mddev->pers)
3031 return -EBUSY;
3032
3033 /*
3034 * Analyze all RAID superblock(s)
3035 */
a757e64c
N		 3036 if (!mddev->raid_disks)
3037 analyze_sbs(mddev);
1da177e4
LT		 3038
3039 chunk_size = mddev->chunk_size;
2604b703
N		 3040
3041 if (chunk_size) {
1da177e4
LT		 3042 if (chunk_size > MAX_CHUNK_SIZE) {
3043 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3044 chunk_size, MAX_CHUNK_SIZE);
3045 return -EINVAL;
3046 }
3047 /*
3048 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3049 */
3050 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 3051 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 3052 return -EINVAL;
3053 }
3054 if (chunk_size < PAGE_SIZE) {
3055 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3056 chunk_size, PAGE_SIZE);
3057 return -EINVAL;
3058 }
3059
3060 /* devices must have minimum size of one chunk */
3061 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3062 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3063 continue;
3064 if (rdev->size < chunk_size / 1024) {
3065 printk(KERN_WARNING
3066 "md: Dev %s smaller than chunk_size:"
3067 " %lluk < %dk\n",
3068 bdevname(rdev->bdev,b),
3069 (unsigned long long)rdev->size,
3070 chunk_size / 1024);
3071 return -EINVAL;
3072 }
3073 }
3074 }
3075
1da177e4 3076#ifdef CONFIG_KMOD
d9d166c2
N		 3077 if (mddev->level != LEVEL_NONE)
3078 request_module("md-level-%d", mddev->level);
3079 else if (mddev->clevel[0])
3080 request_module("md-%s", mddev->clevel);
1da177e4
LT		 3081#endif
3082
3083 /*
3084 * Drop all container device buffers, from now on
3085 * the only valid external interface is through the md
3086 * device.
3087 * Also find largest hardsector size
3088 */
3089 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 3090 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3091 continue;
3092 sync_blockdev(rdev->bdev);
f98393a6 3093 invalidate_bdev(rdev->bdev);
1da177e4
LT		 3094 }
3095
3096 md_probe(mddev->unit, NULL, NULL);
3097 disk = mddev->gendisk;
3098 if (!disk)
3099 return -ENOMEM;
3100
3101 spin_lock(&pers_lock);
d9d166c2 3102 pers = find_pers(mddev->level, mddev->clevel);
2604b703 3103 if (!pers || !try_module_get(pers->owner)) {
1da177e4 3104 spin_unlock(&pers_lock);
d9d166c2
N		 3105 if (mddev->level != LEVEL_NONE)
3106 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3107 mddev->level);
3108 else
3109 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3110 mddev->clevel);
1da177e4
LT		 3111 return -EINVAL;
3112 }
2604b703 3113 mddev->pers = pers;
1da177e4 3114 spin_unlock(&pers_lock);
d9d166c2
N		 3115 mddev->level = pers->level;
3116 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 3117
f6705578 3118 if (mddev->reshape_position != MaxSector &&
63c70c4f 3119 pers->start_reshape == NULL) {
f6705578
N		 3120 /* This personality cannot handle reshaping... */
3121 mddev->pers = NULL;
3122 module_put(pers->owner);
3123 return -EINVAL;
3124 }
3125
7dd5e7c3
N		 3126 if (pers->sync_request) {
3127 /* Warn if this is a potentially silly
3128 * configuration.
3129 */
3130 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3131 mdk_rdev_t *rdev2;
3132 struct list_head *tmp2;
3133 int warned = 0;
3134 ITERATE_RDEV(mddev, rdev, tmp) {
3135 ITERATE_RDEV(mddev, rdev2, tmp2) {
3136 if (rdev < rdev2 &&
3137 rdev->bdev->bd_contains ==
3138 rdev2->bdev->bd_contains) {
3139 printk(KERN_WARNING
3140 "%s: WARNING: %s appears to be"
3141 " on the same physical disk as"
3142 " %s.\n",
3143 mdname(mddev),
3144 bdevname(rdev->bdev,b),
3145 bdevname(rdev2->bdev,b2));
3146 warned = 1;
3147 }
3148 }
3149 }
3150 if (warned)
3151 printk(KERN_WARNING
3152 "True protection against single-disk"
3153 " failure might be compromised.\n");
3154 }
3155
657390d2 3156 mddev->recovery = 0;
1da177e4 3157 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 3158 mddev->barriers_work = 1;
6ff8d8ec 3159 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 3160
f91de92e
N		 3161 if (start_readonly)
3162 mddev->ro = 2; /* read-only, but switch on first write */
3163
b15c2e57
N		 3164 err = mddev->pers->run(mddev);
3165 if (!err && mddev->pers->sync_request) {
3166 err = bitmap_create(mddev);
3167 if (err) {
3168 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3169 mdname(mddev), err);
3170 mddev->pers->stop(mddev);
3171 }
3172 }
1da177e4
LT		 3173 if (err) {
3174 printk(KERN_ERR "md: pers->run() failed ...\n");
3175 module_put(mddev->pers->owner);
3176 mddev->pers = NULL;
32a7627c
N		 3177 bitmap_destroy(mddev);
3178 return err;
1da177e4 3179 }
5e55e2f5
N		 3180 if (mddev->pers->sync_request) {
3181 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3182 printk(KERN_WARNING
3183 "md: cannot register extra attributes for %s\n",
3184 mdname(mddev));
3185 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 3186 mddev->ro = 0;
3187
1da177e4
LT		 3188 atomic_set(&mddev->writes_pending,0);
3189 mddev->safemode = 0;
3190 mddev->safemode_timer.function = md_safemode_timeout;
3191 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 3192 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 3193 mddev->in_sync = 1;
86e6ffdd
N		 3194
3195 ITERATE_RDEV(mddev,rdev,tmp)
3196 if (rdev->raid_disk >= 0) {
3197 char nm[20];
3198 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 3199 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3200 printk("md: cannot register %s for %s\n",
3201 nm, mdname(mddev));
86e6ffdd 3202 }
1da177e4
LT		 3203
3204 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3205
850b2b42
N		 3206 if (mddev->flags)
3207 md_update_sb(mddev, 0);
1da177e4
LT		 3208
3209 set_capacity(disk, mddev->array_size<<1);
3210
3211 /* If we call blk_queue_make_request here, it will
3212 * re-initialise max_sectors etc which may have been
3213 * refined inside -> run. So just set the bits we need to set.
3214 * Most initialisation happended when we called
3215 * blk_queue_make_request(..., md_fail_request)
3216 * earlier.
3217 */
3218 mddev->queue->queuedata = mddev;
3219 mddev->queue->make_request_fn = mddev->pers->make_request;
3220
5fd6c1dc
N		 3221 /* If there is a partially-recovered drive we need to
3222 * start recovery here. If we leave it to md_check_recovery,
3223 * it will remove the drives and not do the right thing
3224 */
0b8c9de0 3225 if (mddev->degraded && !mddev->sync_thread) {
5fd6c1dc
N		 3226 struct list_head *rtmp;
3227 int spares = 0;
3228 ITERATE_RDEV(mddev,rdev,rtmp)
3229 if (rdev->raid_disk >= 0 &&
3230 !test_bit(In_sync, &rdev->flags) &&
3231 !test_bit(Faulty, &rdev->flags))
3232 /* complete an interrupted recovery */
3233 spares++;
3234 if (spares && mddev->pers->sync_request) {
3235 mddev->recovery = 0;
3236 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3237 mddev->sync_thread = md_register_thread(md_do_sync,
3238 mddev,
3239 "%s_resync");
3240 if (!mddev->sync_thread) {
3241 printk(KERN_ERR "%s: could not start resync"
3242 " thread...\n",
3243 mdname(mddev));
3244 /* leave the spares where they are, it shouldn't hurt */
3245 mddev->recovery = 0;
0b8c9de0 3246 }
5fd6c1dc
N		 3247 }
3248 }
0b8c9de0
N		 3249 md_wakeup_thread(mddev->thread);
3250 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 3251
1da177e4 3252 mddev->changed = 1;
d7603b7e 3253 md_new_event(mddev);
2f471303 3254 kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
1da177e4
LT		 3255 return 0;
3256}
3257
3258static int restart_array(mddev_t *mddev)
3259{
3260 struct gendisk *disk = mddev->gendisk;
3261 int err;
3262
3263 /*
3264 * Complain if it has no devices
3265 */
3266 err = -ENXIO;
3267 if (list_empty(&mddev->disks))
3268 goto out;
3269
3270 if (mddev->pers) {
3271 err = -EBUSY;
3272 if (!mddev->ro)
3273 goto out;
3274
3275 mddev->safemode = 0;
3276 mddev->ro = 0;
3277 set_disk_ro(disk, 0);
3278
3279 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3280 mdname(mddev));
3281 /*
3282 * Kick recovery or resync if necessary
3283 */
3284 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3285 md_wakeup_thread(mddev->thread);
5fd6c1dc 3286 md_wakeup_thread(mddev->sync_thread);
1da177e4 3287 err = 0;
9e653b63 3288 } else
1da177e4 3289 err = -EINVAL;
1da177e4
LT		 3290
3291out:
3292 return err;
3293}
3294
acc55e22
N		 3295/* similar to deny_write_access, but accounts for our holding a reference
3296 * to the file ourselves */
3297static int deny_bitmap_write_access(struct file * file)
3298{
3299 struct inode *inode = file->f_mapping->host;
3300
3301 spin_lock(&inode->i_lock);
3302 if (atomic_read(&inode->i_writecount) > 1) {
3303 spin_unlock(&inode->i_lock);
3304 return -ETXTBSY;
3305 }
3306 atomic_set(&inode->i_writecount, -1);
3307 spin_unlock(&inode->i_lock);
3308
3309 return 0;
3310}
3311
3312static void restore_bitmap_write_access(struct file *file)
3313{
3314 struct inode *inode = file->f_mapping->host;
3315
3316 spin_lock(&inode->i_lock);
3317 atomic_set(&inode->i_writecount, 1);
3318 spin_unlock(&inode->i_lock);
3319}
3320
9e653b63
N		 3321/* mode:
3322 * 0 - completely stop and dis-assemble array
3323 * 1 - switch to readonly
3324 * 2 - stop but do not disassemble array
3325 */
3326static int do_md_stop(mddev_t * mddev, int mode)
1da177e4
LT		 3327{
3328 int err = 0;
3329 struct gendisk *disk = mddev->gendisk;
3330
3331 if (mddev->pers) {
3332 if (atomic_read(&mddev->active)>2) {
3333 printk("md: %s still in use.\n",mdname(mddev));
3334 return -EBUSY;
3335 }
3336
3337 if (mddev->sync_thread) {
5fd6c1dc 3338 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4
LT		 3339 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3340 md_unregister_thread(mddev->sync_thread);
3341 mddev->sync_thread = NULL;
3342 }
3343
3344 del_timer_sync(&mddev->safemode_timer);
3345
3346 invalidate_partition(disk, 0);
3347
9e653b63
N		 3348 switch(mode) {
3349 case 1: /* readonly */
1da177e4 3350 err = -ENXIO;
f91de92e 3351 if (mddev->ro==1)
1da177e4
LT		 3352 goto out;
3353 mddev->ro = 1;
9e653b63
N		 3354 break;
3355 case 0: /* disassemble */
3356 case 2: /* stop */
6b8b3e8a 3357 bitmap_flush(mddev);
a9701a30 3358 md_super_wait(mddev);
1da177e4
LT		 3359 if (mddev->ro)
3360 set_disk_ro(disk, 0);
3361 blk_queue_make_request(mddev->queue, md_fail_request);
3362 mddev->pers->stop(mddev);
d1b5380c
N		 3363 mddev->queue->merge_bvec_fn = NULL;
3364 mddev->queue->unplug_fn = NULL;
3365 mddev->queue->issue_flush_fn = NULL;
041ae52e 3366 mddev->queue->backing_dev_info.congested_fn = NULL;
411036fa
N		 3367 if (mddev->pers->sync_request)
3368 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3369
1da177e4
LT		 3370 module_put(mddev->pers->owner);
3371 mddev->pers = NULL;
0d4ca600
N		 3372
3373 set_capacity(disk, 0);
3374 mddev->changed = 1;
3375
1da177e4
LT		 3376 if (mddev->ro)
3377 mddev->ro = 0;
3378 }
850b2b42 3379 if (!mddev->in_sync || mddev->flags) {
1da177e4
LT		 3380 /* mark array as shutdown cleanly */
3381 mddev->in_sync = 1;
850b2b42 3382 md_update_sb(mddev, 1);
1da177e4 3383 }
9e653b63 3384 if (mode == 1)
1da177e4 3385 set_disk_ro(disk, 1);
5fd6c1dc 3386 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
1da177e4 3387 }
32a7627c 3388
1da177e4
LT		 3389 /*
3390 * Free resources if final stop
3391 */
9e653b63 3392 if (mode == 0) {
86e6ffdd
N		 3393 mdk_rdev_t *rdev;
3394 struct list_head *tmp;
0d4ca600 3395
1da177e4
LT		 3396 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3397
978f946b
N		 3398 bitmap_destroy(mddev);
3399 if (mddev->bitmap_file) {
acc55e22 3400 restore_bitmap_write_access(mddev->bitmap_file);
978f946b
N		 3401 fput(mddev->bitmap_file);
3402 mddev->bitmap_file = NULL;
3403 }
3404 mddev->bitmap_offset = 0;
3405
86e6ffdd
N		 3406 ITERATE_RDEV(mddev,rdev,tmp)
3407 if (rdev->raid_disk >= 0) {
3408 char nm[20];
3409 sprintf(nm, "rd%d", rdev->raid_disk);
3410 sysfs_remove_link(&mddev->kobj, nm);
3411 }
3412
5792a285
N		 3413 /* make sure all delayed_delete calls have finished */
3414 flush_scheduled_work();
3415
1da177e4
LT		 3416 export_array(mddev);
3417
3418 mddev->array_size = 0;
9e653b63
N		 3419 mddev->size = 0;
3420 mddev->raid_disks = 0;
a94213b1 3421 mddev->recovery_cp = 0;
9e653b63 3422
a8a55c38 3423 } else if (mddev->pers)
1da177e4
LT		 3424 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3425 mdname(mddev));
3426 err = 0;
d7603b7e 3427 md_new_event(mddev);
1da177e4
LT		 3428out:
3429 return err;
3430}
3431
fdee8ae4 3432#ifndef MODULE
1da177e4
LT		 3433static void autorun_array(mddev_t *mddev)
3434{
3435 mdk_rdev_t *rdev;
3436 struct list_head *tmp;
3437 int err;
3438
a757e64c 3439 if (list_empty(&mddev->disks))
1da177e4 3440 return;
1da177e4
LT		 3441
3442 printk(KERN_INFO "md: running: ");
3443
3444 ITERATE_RDEV(mddev,rdev,tmp) {
3445 char b[BDEVNAME_SIZE];
3446 printk("<%s>", bdevname(rdev->bdev,b));
3447 }
3448 printk("\n");
3449
3450 err = do_md_run (mddev);
3451 if (err) {
3452 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3453 do_md_stop (mddev, 0);
3454 }
3455}
3456
3457/*
3458 * lets try to run arrays based on all disks that have arrived
3459 * until now. (those are in pending_raid_disks)
3460 *
3461 * the method: pick the first pending disk, collect all disks with
3462 * the same UUID, remove all from the pending list and put them into
3463 * the 'same_array' list. Then order this list based on superblock
3464 * update time (freshest comes first), kick out 'old' disks and
3465 * compare superblocks. If everything's fine then run it.
3466 *
3467 * If "unit" is allocated, then bump its reference count
3468 */
3469static void autorun_devices(int part)
3470{
1da177e4
LT		 3471 struct list_head *tmp;
3472 mdk_rdev_t *rdev0, *rdev;
3473 mddev_t *mddev;
3474 char b[BDEVNAME_SIZE];
3475
3476 printk(KERN_INFO "md: autorun ...\n");
3477 while (!list_empty(&pending_raid_disks)) {
e8703fe1 3478 int unit;
1da177e4 3479 dev_t dev;
ad01c9e3 3480 LIST_HEAD(candidates);
1da177e4
LT		 3481 rdev0 = list_entry(pending_raid_disks.next,
3482 mdk_rdev_t, same_set);
3483
3484 printk(KERN_INFO "md: considering %s ...\n",
3485 bdevname(rdev0->bdev,b));
3486 INIT_LIST_HEAD(&candidates);
3487 ITERATE_RDEV_PENDING(rdev,tmp)
3488 if (super_90_load(rdev, rdev0, 0) >= 0) {
3489 printk(KERN_INFO "md: adding %s ...\n",
3490 bdevname(rdev->bdev,b));
3491 list_move(&rdev->same_set, &candidates);
3492 }
3493 /*
3494 * now we have a set of devices, with all of them having
3495 * mostly sane superblocks. It's time to allocate the
3496 * mddev.
3497 */
e8703fe1
N		 3498 if (part) {
3499 dev = MKDEV(mdp_major,
3500 rdev0->preferred_minor << MdpMinorShift);
3501 unit = MINOR(dev) >> MdpMinorShift;
3502 } else {
3503 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3504 unit = MINOR(dev);
3505 }
3506 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 3507 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3508 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3509 break;
3510 }
1da177e4
LT		 3511
3512 md_probe(dev, NULL, NULL);
3513 mddev = mddev_find(dev);
3514 if (!mddev) {
3515 printk(KERN_ERR
3516 "md: cannot allocate memory for md drive.\n");
3517 break;
3518 }
3519 if (mddev_lock(mddev))
3520 printk(KERN_WARNING "md: %s locked, cannot run\n",
3521 mdname(mddev));
3522 else if (mddev->raid_disks || mddev->major_version
3523 || !list_empty(&mddev->disks)) {
3524 printk(KERN_WARNING
3525 "md: %s already running, cannot run %s\n",
3526 mdname(mddev), bdevname(rdev0->bdev,b));
3527 mddev_unlock(mddev);
3528 } else {
3529 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3530 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3531 list_del_init(&rdev->same_set);
3532 if (bind_rdev_to_array(rdev, mddev))
3533 export_rdev(rdev);
3534 }
3535 autorun_array(mddev);
3536 mddev_unlock(mddev);
3537 }
3538 /* on success, candidates will be empty, on error
3539 * it won't...
3540 */
3541 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3542 export_rdev(rdev);
3543 mddev_put(mddev);
3544 }
3545 printk(KERN_INFO "md: ... autorun DONE.\n");
3546}
fdee8ae4 3547#endif /* !MODULE */
1da177e4 3548
1da177e4
LT		 3549static int get_version(void __user * arg)
3550{
3551 mdu_version_t ver;
3552
3553 ver.major = MD_MAJOR_VERSION;
3554 ver.minor = MD_MINOR_VERSION;
3555 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3556
3557 if (copy_to_user(arg, &ver, sizeof(ver)))
3558 return -EFAULT;
3559
3560 return 0;
3561}
3562
3563static int get_array_info(mddev_t * mddev, void __user * arg)
3564{
3565 mdu_array_info_t info;
3566 int nr,working,active,failed,spare;
3567 mdk_rdev_t *rdev;
3568 struct list_head *tmp;
3569
3570 nr=working=active=failed=spare=0;
3571 ITERATE_RDEV(mddev,rdev,tmp) {
3572 nr++;
b2d444d7 3573 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 3574 failed++;
3575 else {
3576 working++;
b2d444d7 3577 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 3578 active++;
3579 else
3580 spare++;
3581 }
3582 }
3583
3584 info.major_version = mddev->major_version;
3585 info.minor_version = mddev->minor_version;
3586 info.patch_version = MD_PATCHLEVEL_VERSION;
3587 info.ctime = mddev->ctime;
3588 info.level = mddev->level;
3589 info.size = mddev->size;
284ae7ca
N		 3590 if (info.size != mddev->size) /* overflow */
3591 info.size = -1;
1da177e4
LT		 3592 info.nr_disks = nr;
3593 info.raid_disks = mddev->raid_disks;
3594 info.md_minor = mddev->md_minor;
3595 info.not_persistent= !mddev->persistent;
3596
3597 info.utime = mddev->utime;
3598 info.state = 0;
3599 if (mddev->in_sync)
3600 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 3601 if (mddev->bitmap && mddev->bitmap_offset)
3602 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 3603 info.active_disks = active;
3604 info.working_disks = working;
3605 info.failed_disks = failed;
3606 info.spare_disks = spare;
3607
3608 info.layout = mddev->layout;
3609 info.chunk_size = mddev->chunk_size;
3610
3611 if (copy_to_user(arg, &info, sizeof(info)))
3612 return -EFAULT;
3613
3614 return 0;
3615}
3616
87162a28 3617static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 3618{
3619 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3620 char *ptr, *buf = NULL;
3621 int err = -ENOMEM;
3622
2a2275d6
N		 3623 md_allow_write(mddev);
3624
32a7627c
N		 3625 file = kmalloc(sizeof(*file), GFP_KERNEL);
3626 if (!file)
3627 goto out;
3628
3629 /* bitmap disabled, zero the first byte and copy out */
3630 if (!mddev->bitmap || !mddev->bitmap->file) {
3631 file->pathname[0] = '\0';
3632 goto copy_out;
3633 }
3634
3635 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3636 if (!buf)
3637 goto out;
3638
3639 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3640 if (!ptr)
3641 goto out;
3642
3643 strcpy(file->pathname, ptr);
3644
3645copy_out:
3646 err = 0;
3647 if (copy_to_user(arg, file, sizeof(*file)))
3648 err = -EFAULT;
3649out:
3650 kfree(buf);
3651 kfree(file);
3652 return err;
3653}
3654
1da177e4
LT		 3655static int get_disk_info(mddev_t * mddev, void __user * arg)
3656{
3657 mdu_disk_info_t info;
3658 unsigned int nr;
3659 mdk_rdev_t *rdev;
3660
3661 if (copy_from_user(&info, arg, sizeof(info)))
3662 return -EFAULT;
3663
3664 nr = info.number;
3665
3666 rdev = find_rdev_nr(mddev, nr);
3667 if (rdev) {
3668 info.major = MAJOR(rdev->bdev->bd_dev);
3669 info.minor = MINOR(rdev->bdev->bd_dev);
3670 info.raid_disk = rdev->raid_disk;
3671 info.state = 0;
b2d444d7 3672 if (test_bit(Faulty, &rdev->flags))
1da177e4 3673 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 3674 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 3675 info.state |= (1<<MD_DISK_ACTIVE);
3676 info.state |= (1<<MD_DISK_SYNC);
3677 }
8ddf9efe
N		 3678 if (test_bit(WriteMostly, &rdev->flags))
3679 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 3680 } else {
3681 info.major = info.minor = 0;
3682 info.raid_disk = -1;
3683 info.state = (1<<MD_DISK_REMOVED);
3684 }
3685
3686 if (copy_to_user(arg, &info, sizeof(info)))
3687 return -EFAULT;
3688
3689 return 0;
3690}
3691
3692static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3693{
3694 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3695 mdk_rdev_t *rdev;
3696 dev_t dev = MKDEV(info->major,info->minor);
3697
3698 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3699 return -EOVERFLOW;
3700
3701 if (!mddev->raid_disks) {
3702 int err;
3703 /* expecting a device which has a superblock */
3704 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3705 if (IS_ERR(rdev)) {
3706 printk(KERN_WARNING
3707 "md: md_import_device returned %ld\n",
3708 PTR_ERR(rdev));
3709 return PTR_ERR(rdev);
3710 }
3711 if (!list_empty(&mddev->disks)) {
3712 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3713 mdk_rdev_t, same_set);
3714 int err = super_types[mddev->major_version]
3715 .load_super(rdev, rdev0, mddev->minor_version);
3716 if (err < 0) {
3717 printk(KERN_WARNING
3718 "md: %s has different UUID to %s\n",
3719 bdevname(rdev->bdev,b),
3720 bdevname(rdev0->bdev,b2));
3721 export_rdev(rdev);
3722 return -EINVAL;
3723 }
3724 }
3725 err = bind_rdev_to_array(rdev, mddev);
3726 if (err)
3727 export_rdev(rdev);
3728 return err;
3729 }
3730
3731 /*
3732 * add_new_disk can be used once the array is assembled
3733 * to add "hot spares". They must already have a superblock
3734 * written
3735 */
3736 if (mddev->pers) {
3737 int err;
3738 if (!mddev->pers->hot_add_disk) {
3739 printk(KERN_WARNING
3740 "%s: personality does not support diskops!\n",
3741 mdname(mddev));
3742 return -EINVAL;
3743 }
7b1e35f6
N		 3744 if (mddev->persistent)
3745 rdev = md_import_device(dev, mddev->major_version,
3746 mddev->minor_version);
3747 else
3748 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 3749 if (IS_ERR(rdev)) {
3750 printk(KERN_WARNING
3751 "md: md_import_device returned %ld\n",
3752 PTR_ERR(rdev));
3753 return PTR_ERR(rdev);
3754 }
41158c7e
N		 3755 /* set save_raid_disk if appropriate */
3756 if (!mddev->persistent) {
3757 if (info->state & (1<<MD_DISK_SYNC) &&
3758 info->raid_disk < mddev->raid_disks)
3759 rdev->raid_disk = info->raid_disk;
3760 else
3761 rdev->raid_disk = -1;
3762 } else
3763 super_types[mddev->major_version].
3764 validate_super(mddev, rdev);
3765 rdev->saved_raid_disk = rdev->raid_disk;
3766
b2d444d7 3767 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 3768 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3769 set_bit(WriteMostly, &rdev->flags);
3770
1da177e4
LT		 3771 rdev->raid_disk = -1;
3772 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 3773 if (!err && !mddev->pers->hot_remove_disk) {
3774 /* If there is hot_add_disk but no hot_remove_disk
3775 * then added disks for geometry changes,
3776 * and should be added immediately.
3777 */
3778 super_types[mddev->major_version].
3779 validate_super(mddev, rdev);
3780 err = mddev->pers->hot_add_disk(mddev, rdev);
3781 if (err)
3782 unbind_rdev_from_array(rdev);
3783 }
1da177e4
LT		 3784 if (err)
3785 export_rdev(rdev);
c361777f 3786
17571284 3787 md_update_sb(mddev, 1);
c361777f 3788 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 3789 md_wakeup_thread(mddev->thread);
1da177e4
LT		 3790 return err;
3791 }
3792
3793 /* otherwise, add_new_disk is only allowed
3794 * for major_version==0 superblocks
3795 */
3796 if (mddev->major_version != 0) {
3797 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3798 mdname(mddev));
3799 return -EINVAL;
3800 }
3801
3802 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3803 int err;
3804 rdev = md_import_device (dev, -1, 0);
3805 if (IS_ERR(rdev)) {
3806 printk(KERN_WARNING
3807 "md: error, md_import_device() returned %ld\n",
3808 PTR_ERR(rdev));
3809 return PTR_ERR(rdev);
3810 }
3811 rdev->desc_nr = info->number;
3812 if (info->raid_disk < mddev->raid_disks)
3813 rdev->raid_disk = info->raid_disk;
3814 else
3815 rdev->raid_disk = -1;
3816
b2d444d7
N		 3817 rdev->flags = 0;
3818
1da177e4 3819 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 3820 if (info->state & (1<<MD_DISK_SYNC))
3821 set_bit(In_sync, &rdev->flags);
1da177e4 3822
8ddf9efe
N		 3823 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3824 set_bit(WriteMostly, &rdev->flags);
3825
1da177e4
LT		 3826 if (!mddev->persistent) {
3827 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3828 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3829 } else
3830 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3831 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3832
2bf071bf
N		 3833 err = bind_rdev_to_array(rdev, mddev);
3834 if (err) {
3835 export_rdev(rdev);
3836 return err;
3837 }
1da177e4
LT		 3838 }
3839
3840 return 0;
3841}
3842
3843static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3844{
3845 char b[BDEVNAME_SIZE];
3846 mdk_rdev_t *rdev;
3847
3848 if (!mddev->pers)
3849 return -ENODEV;
3850
3851 rdev = find_rdev(mddev, dev);
3852 if (!rdev)
3853 return -ENXIO;
3854
3855 if (rdev->raid_disk >= 0)
3856 goto busy;
3857
3858 kick_rdev_from_array(rdev);
850b2b42 3859 md_update_sb(mddev, 1);
d7603b7e 3860 md_new_event(mddev);
1da177e4
LT		 3861
3862 return 0;
3863busy:
3864 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3865 bdevname(rdev->bdev,b), mdname(mddev));
3866 return -EBUSY;
3867}
3868
3869static int hot_add_disk(mddev_t * mddev, dev_t dev)
3870{
3871 char b[BDEVNAME_SIZE];
3872 int err;
3873 unsigned int size;
3874 mdk_rdev_t *rdev;
3875
3876 if (!mddev->pers)
3877 return -ENODEV;
3878
3879 if (mddev->major_version != 0) {
3880 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3881 " version-0 superblocks.\n",
3882 mdname(mddev));
3883 return -EINVAL;
3884 }
3885 if (!mddev->pers->hot_add_disk) {
3886 printk(KERN_WARNING
3887 "%s: personality does not support diskops!\n",
3888 mdname(mddev));
3889 return -EINVAL;
3890 }
3891
3892 rdev = md_import_device (dev, -1, 0);
3893 if (IS_ERR(rdev)) {
3894 printk(KERN_WARNING
3895 "md: error, md_import_device() returned %ld\n",
3896 PTR_ERR(rdev));
3897 return -EINVAL;
3898 }
3899
3900 if (mddev->persistent)
3901 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3902 else
3903 rdev->sb_offset =
3904 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3905
3906 size = calc_dev_size(rdev, mddev->chunk_size);
3907 rdev->size = size;
3908
b2d444d7 3909 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3910 printk(KERN_WARNING
3911 "md: can not hot-add faulty %s disk to %s!\n",
3912 bdevname(rdev->bdev,b), mdname(mddev));
3913 err = -EINVAL;
3914 goto abort_export;
3915 }
b2d444d7 3916 clear_bit(In_sync, &rdev->flags);
1da177e4 3917 rdev->desc_nr = -1;
5842730d 3918 rdev->saved_raid_disk = -1;
2bf071bf
N		 3919 err = bind_rdev_to_array(rdev, mddev);
3920 if (err)
3921 goto abort_export;
1da177e4
LT		 3922
3923 /*
3924 * The rest should better be atomic, we can have disk failures
3925 * noticed in interrupt contexts ...
3926 */
3927
3928 if (rdev->desc_nr == mddev->max_disks) {
3929 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3930 mdname(mddev));
3931 err = -EBUSY;
3932 goto abort_unbind_export;
3933 }
3934
3935 rdev->raid_disk = -1;
3936
850b2b42 3937 md_update_sb(mddev, 1);
1da177e4
LT		 3938
3939 /*
3940 * Kick recovery, maybe this spare has to be added to the
3941 * array immediately.
3942 */
3943 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3944 md_wakeup_thread(mddev->thread);
d7603b7e 3945 md_new_event(mddev);
1da177e4
LT		 3946 return 0;
3947
3948abort_unbind_export:
3949 unbind_rdev_from_array(rdev);
3950
3951abort_export:
3952 export_rdev(rdev);
3953 return err;
3954}
3955
32a7627c
N		 3956static int set_bitmap_file(mddev_t *mddev, int fd)
3957{
3958 int err;
3959
36fa3063
N		 3960 if (mddev->pers) {
3961 if (!mddev->pers->quiesce)
3962 return -EBUSY;
3963 if (mddev->recovery || mddev->sync_thread)
3964 return -EBUSY;
3965 /* we should be able to change the bitmap.. */
3966 }
32a7627c 3967
32a7627c 3968
36fa3063
N		 3969 if (fd >= 0) {
3970 if (mddev->bitmap)
3971 return -EEXIST; /* cannot add when bitmap is present */
3972 mddev->bitmap_file = fget(fd);
32a7627c 3973
36fa3063
N		 3974 if (mddev->bitmap_file == NULL) {
3975 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3976 mdname(mddev));
3977 return -EBADF;
3978 }
3979
3980 err = deny_bitmap_write_access(mddev->bitmap_file);
3981 if (err) {
3982 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3983 mdname(mddev));
3984 fput(mddev->bitmap_file);
3985 mddev->bitmap_file = NULL;
3986 return err;
3987 }
a654b9d8 3988 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 3989 } else if (mddev->bitmap == NULL)
3990 return -ENOENT; /* cannot remove what isn't there */
3991 err = 0;
3992 if (mddev->pers) {
3993 mddev->pers->quiesce(mddev, 1);
3994 if (fd >= 0)
3995 err = bitmap_create(mddev);
d7375ab3 3996 if (fd < 0 || err) {
36fa3063 3997 bitmap_destroy(mddev);
d7375ab3
N		 3998 fd = -1; /* make sure to put the file */
3999 }
36fa3063 4000 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 4001 }
4002 if (fd < 0) {
acc55e22
N		 4003 if (mddev->bitmap_file) {
4004 restore_bitmap_write_access(mddev->bitmap_file);
36fa3063 4005 fput(mddev->bitmap_file);
acc55e22 4006 }
36fa3063
N		 4007 mddev->bitmap_file = NULL;
4008 }
4009
32a7627c
N		 4010 return err;
4011}
4012
1da177e4
LT		 4013/*
4014 * set_array_info is used two different ways
4015 * The original usage is when creating a new array.
4016 * In this usage, raid_disks is > 0 and it together with
4017 * level, size, not_persistent,layout,chunksize determine the
4018 * shape of the array.
4019 * This will always create an array with a type-0.90.0 superblock.
4020 * The newer usage is when assembling an array.
4021 * In this case raid_disks will be 0, and the major_version field is
4022 * use to determine which style super-blocks are to be found on the devices.
4023 * The minor and patch _version numbers are also kept incase the
4024 * super_block handler wishes to interpret them.
4025 */
4026static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4027{
4028
4029 if (info->raid_disks == 0) {
4030 /* just setting version number for superblock loading */
4031 if (info->major_version < 0 ||
50511da3 4032 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 4033 super_types[info->major_version].name == NULL) {
4034 /* maybe try to auto-load a module? */
4035 printk(KERN_INFO
4036 "md: superblock version %d not known\n",
4037 info->major_version);
4038 return -EINVAL;
4039 }
4040 mddev->major_version = info->major_version;
4041 mddev->minor_version = info->minor_version;
4042 mddev->patch_version = info->patch_version;
3f9d7b0d 4043 mddev->persistent = !info->not_persistent;
1da177e4
LT		 4044 return 0;
4045 }
4046 mddev->major_version = MD_MAJOR_VERSION;
4047 mddev->minor_version = MD_MINOR_VERSION;
4048 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4049 mddev->ctime = get_seconds();
4050
4051 mddev->level = info->level;
17115e03 4052 mddev->clevel[0] = 0;
1da177e4
LT		 4053 mddev->size = info->size;
4054 mddev->raid_disks = info->raid_disks;
4055 /* don't set md_minor, it is determined by which /dev/md* was
4056 * openned
4057 */
4058 if (info->state & (1<<MD_SB_CLEAN))
4059 mddev->recovery_cp = MaxSector;
4060 else
4061 mddev->recovery_cp = 0;
4062 mddev->persistent = ! info->not_persistent;
4063
4064 mddev->layout = info->layout;
4065 mddev->chunk_size = info->chunk_size;
4066
4067 mddev->max_disks = MD_SB_DISKS;
4068
850b2b42
N		 4069 mddev->flags = 0;
4070 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 4071
b2a2703c
N		 4072 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4073 mddev->bitmap_offset = 0;
4074
f6705578
N		 4075 mddev->reshape_position = MaxSector;
4076
1da177e4
LT		 4077 /*
4078 * Generate a 128 bit UUID
4079 */
4080 get_random_bytes(mddev->uuid, 16);
4081
f6705578
N		 4082 mddev->new_level = mddev->level;
4083 mddev->new_chunk = mddev->chunk_size;
4084 mddev->new_layout = mddev->layout;
4085 mddev->delta_disks = 0;
4086
1da177e4
LT		 4087 return 0;
4088}
4089
a35b0d69
N		 4090static int update_size(mddev_t *mddev, unsigned long size)
4091{
4092 mdk_rdev_t * rdev;
4093 int rv;
4094 struct list_head *tmp;
8ddeeae5 4095 int fit = (size == 0);
a35b0d69
N		 4096
4097 if (mddev->pers->resize == NULL)
4098 return -EINVAL;
4099 /* The "size" is the amount of each device that is used.
4100 * This can only make sense for arrays with redundancy.
4101 * linear and raid0 always use whatever space is available
4102 * We can only consider changing the size if no resync
4103 * or reconstruction is happening, and if the new size
4104 * is acceptable. It must fit before the sb_offset or,
4105 * if that is <data_offset, it must fit before the
4106 * size of each device.
4107 * If size is zero, we find the largest size that fits.
4108 */
4109 if (mddev->sync_thread)
4110 return -EBUSY;
4111 ITERATE_RDEV(mddev,rdev,tmp) {
4112 sector_t avail;
01ab5662
N		 4113 avail = rdev->size * 2;
4114
a35b0d69
N		 4115 if (fit && (size == 0 || size > avail/2))
4116 size = avail/2;
4117 if (avail < ((sector_t)size << 1))
4118 return -ENOSPC;
4119 }
4120 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4121 if (!rv) {
4122 struct block_device *bdev;
4123
4124 bdev = bdget_disk(mddev->gendisk, 0);
4125 if (bdev) {
1b1dcc1b 4126 mutex_lock(&bdev->bd_inode->i_mutex);
6d89332b 4127 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
1b1dcc1b 4128 mutex_unlock(&bdev->bd_inode->i_mutex);
a35b0d69
N		 4129 bdput(bdev);
4130 }
4131 }
4132 return rv;
4133}
4134
da943b99
N		 4135static int update_raid_disks(mddev_t *mddev, int raid_disks)
4136{
4137 int rv;
4138 /* change the number of raid disks */
63c70c4f 4139 if (mddev->pers->check_reshape == NULL)
da943b99
N		 4140 return -EINVAL;
4141 if (raid_disks <= 0 ||
4142 raid_disks >= mddev->max_disks)
4143 return -EINVAL;
63c70c4f 4144 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 4145 return -EBUSY;
63c70c4f
N		 4146 mddev->delta_disks = raid_disks - mddev->raid_disks;
4147
4148 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 4149 return rv;
4150}
4151
4152
1da177e4
LT		 4153/*
4154 * update_array_info is used to change the configuration of an
4155 * on-line array.
4156 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4157 * fields in the info are checked against the array.
4158 * Any differences that cannot be handled will cause an error.
4159 * Normally, only one change can be managed at a time.
4160 */
4161static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4162{
4163 int rv = 0;
4164 int cnt = 0;
36fa3063
N		 4165 int state = 0;
4166
4167 /* calculate expected state,ignoring low bits */
4168 if (mddev->bitmap && mddev->bitmap_offset)
4169 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 4170
4171 if (mddev->major_version != info->major_version ||
4172 mddev->minor_version != info->minor_version ||
4173/* mddev->patch_version != info->patch_version || */
4174 mddev->ctime != info->ctime ||
4175 mddev->level != info->level ||
4176/* mddev->layout != info->layout || */
4177 !mddev->persistent != info->not_persistent||
36fa3063
N		 4178 mddev->chunk_size != info->chunk_size ||
4179 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4180 ((state^info->state) & 0xfffffe00)
4181 )
1da177e4
LT		 4182 return -EINVAL;
4183 /* Check there is only one change */
284ae7ca 4184 if (info->size >= 0 && mddev->size != info->size) cnt++;
1da177e4
LT		 4185 if (mddev->raid_disks != info->raid_disks) cnt++;
4186 if (mddev->layout != info->layout) cnt++;
36fa3063 4187 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 4188 if (cnt == 0) return 0;
4189 if (cnt > 1) return -EINVAL;
4190
4191 if (mddev->layout != info->layout) {
4192 /* Change layout
4193 * we don't need to do anything at the md level, the
4194 * personality will take care of it all.
4195 */
4196 if (mddev->pers->reconfig == NULL)
4197 return -EINVAL;
4198 else
4199 return mddev->pers->reconfig(mddev, info->layout, -1);
4200 }
284ae7ca 4201 if (info->size >= 0 && mddev->size != info->size)
a35b0d69
N		 4202 rv = update_size(mddev, info->size);
4203
da943b99
N		 4204 if (mddev->raid_disks != info->raid_disks)
4205 rv = update_raid_disks(mddev, info->raid_disks);
4206
36fa3063
N		 4207 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4208 if (mddev->pers->quiesce == NULL)
4209 return -EINVAL;
4210 if (mddev->recovery || mddev->sync_thread)
4211 return -EBUSY;
4212 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4213 /* add the bitmap */
4214 if (mddev->bitmap)
4215 return -EEXIST;
4216 if (mddev->default_bitmap_offset == 0)
4217 return -EINVAL;
4218 mddev->bitmap_offset = mddev->default_bitmap_offset;
4219 mddev->pers->quiesce(mddev, 1);
4220 rv = bitmap_create(mddev);
4221 if (rv)
4222 bitmap_destroy(mddev);
4223 mddev->pers->quiesce(mddev, 0);
4224 } else {
4225 /* remove the bitmap */
4226 if (!mddev->bitmap)
4227 return -ENOENT;
4228 if (mddev->bitmap->file)
4229 return -EINVAL;
4230 mddev->pers->quiesce(mddev, 1);
4231 bitmap_destroy(mddev);
4232 mddev->pers->quiesce(mddev, 0);
4233 mddev->bitmap_offset = 0;
4234 }
4235 }
850b2b42 4236 md_update_sb(mddev, 1);
1da177e4
LT		 4237 return rv;
4238}
4239
4240static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4241{
4242 mdk_rdev_t *rdev;
4243
4244 if (mddev->pers == NULL)
4245 return -ENODEV;
4246
4247 rdev = find_rdev(mddev, dev);
4248 if (!rdev)
4249 return -ENODEV;
4250
4251 md_error(mddev, rdev);
4252 return 0;
4253}
4254
a885c8c4
CH		 4255static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4256{
4257 mddev_t *mddev = bdev->bd_disk->private_data;
4258
4259 geo->heads = 2;
4260 geo->sectors = 4;
4261 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4262 return 0;
4263}
4264
1da177e4
LT		 4265static int md_ioctl(struct inode *inode, struct file *file,
4266 unsigned int cmd, unsigned long arg)
4267{
4268 int err = 0;
4269 void __user *argp = (void __user *)arg;
1da177e4
LT		 4270 mddev_t *mddev = NULL;
4271
4272 if (!capable(CAP_SYS_ADMIN))
4273 return -EACCES;
4274
4275 /*
4276 * Commands dealing with the RAID driver but not any
4277 * particular array:
4278 */
4279 switch (cmd)
4280 {
4281 case RAID_VERSION:
4282 err = get_version(argp);
4283 goto done;
4284
4285 case PRINT_RAID_DEBUG:
4286 err = 0;
4287 md_print_devices();
4288 goto done;
4289
4290#ifndef MODULE
4291 case RAID_AUTORUN:
4292 err = 0;
4293 autostart_arrays(arg);
4294 goto done;
4295#endif
4296 default:;
4297 }
4298
4299 /*
4300 * Commands creating/starting a new array:
4301 */
4302
4303 mddev = inode->i_bdev->bd_disk->private_data;
4304
4305 if (!mddev) {
4306 BUG();
4307 goto abort;
4308 }
4309
1da177e4
LT		 4310 err = mddev_lock(mddev);
4311 if (err) {
4312 printk(KERN_INFO
4313 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4314 err, cmd);
4315 goto abort;
4316 }
4317
4318 switch (cmd)
4319 {
4320 case SET_ARRAY_INFO:
4321 {
4322 mdu_array_info_t info;
4323 if (!arg)
4324 memset(&info, 0, sizeof(info));
4325 else if (copy_from_user(&info, argp, sizeof(info))) {
4326 err = -EFAULT;
4327 goto abort_unlock;
4328 }
4329 if (mddev->pers) {
4330 err = update_array_info(mddev, &info);
4331 if (err) {
4332 printk(KERN_WARNING "md: couldn't update"
4333 " array info. %d\n", err);
4334 goto abort_unlock;
4335 }
4336 goto done_unlock;
4337 }
4338 if (!list_empty(&mddev->disks)) {
4339 printk(KERN_WARNING
4340 "md: array %s already has disks!\n",
4341 mdname(mddev));
4342 err = -EBUSY;
4343 goto abort_unlock;
4344 }
4345 if (mddev->raid_disks) {
4346 printk(KERN_WARNING
4347 "md: array %s already initialised!\n",
4348 mdname(mddev));
4349 err = -EBUSY;
4350 goto abort_unlock;
4351 }
4352 err = set_array_info(mddev, &info);
4353 if (err) {
4354 printk(KERN_WARNING "md: couldn't set"
4355 " array info. %d\n", err);
4356 goto abort_unlock;
4357 }
4358 }
4359 goto done_unlock;
4360
4361 default:;
4362 }
4363
4364 /*
4365 * Commands querying/configuring an existing array:
4366 */
32a7627c 4367 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 4368 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
32a7627c 4369 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3f9d7b0d
N		 4370 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4371 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 4372 err = -ENODEV;
4373 goto abort_unlock;
4374 }
4375
4376 /*
4377 * Commands even a read-only array can execute:
4378 */
4379 switch (cmd)
4380 {
4381 case GET_ARRAY_INFO:
4382 err = get_array_info(mddev, argp);
4383 goto done_unlock;
4384
32a7627c 4385 case GET_BITMAP_FILE:
87162a28 4386 err = get_bitmap_file(mddev, argp);
32a7627c
N		 4387 goto done_unlock;
4388
1da177e4
LT		 4389 case GET_DISK_INFO:
4390 err = get_disk_info(mddev, argp);
4391 goto done_unlock;
4392
4393 case RESTART_ARRAY_RW:
4394 err = restart_array(mddev);
4395 goto done_unlock;
4396
4397 case STOP_ARRAY:
4398 err = do_md_stop (mddev, 0);
4399 goto done_unlock;
4400
4401 case STOP_ARRAY_RO:
4402 err = do_md_stop (mddev, 1);
4403 goto done_unlock;
4404
4405 /*
4406 * We have a problem here : there is no easy way to give a CHS
4407 * virtual geometry. We currently pretend that we have a 2 heads
4408 * 4 sectors (with a BIG number of cylinders...). This drives
4409 * dosfs just mad... ;-)
4410 */
1da177e4
LT		 4411 }
4412
4413 /*
4414 * The remaining ioctls are changing the state of the
f91de92e
N		 4415 * superblock, so we do not allow them on read-only arrays.
4416 * However non-MD ioctls (e.g. get-size) will still come through
4417 * here and hit the 'default' below, so only disallow
4418 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 4419 */
f91de92e
N		 4420 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4421 mddev->ro && mddev->pers) {
4422 if (mddev->ro == 2) {
4423 mddev->ro = 0;
4424 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4425 md_wakeup_thread(mddev->thread);
4426
4427 } else {
4428 err = -EROFS;
4429 goto abort_unlock;
4430 }
1da177e4
LT		 4431 }
4432
4433 switch (cmd)
4434 {
4435 case ADD_NEW_DISK:
4436 {
4437 mdu_disk_info_t info;
4438 if (copy_from_user(&info, argp, sizeof(info)))
4439 err = -EFAULT;
4440 else
4441 err = add_new_disk(mddev, &info);
4442 goto done_unlock;
4443 }
4444
4445 case HOT_REMOVE_DISK:
4446 err = hot_remove_disk(mddev, new_decode_dev(arg));
4447 goto done_unlock;
4448
4449 case HOT_ADD_DISK:
4450 err = hot_add_disk(mddev, new_decode_dev(arg));
4451 goto done_unlock;
4452
4453 case SET_DISK_FAULTY:
4454 err = set_disk_faulty(mddev, new_decode_dev(arg));
4455 goto done_unlock;
4456
4457 case RUN_ARRAY:
4458 err = do_md_run (mddev);
4459 goto done_unlock;
4460
32a7627c
N		 4461 case SET_BITMAP_FILE:
4462 err = set_bitmap_file(mddev, (int)arg);
4463 goto done_unlock;
4464
1da177e4 4465 default:
1da177e4
LT		 4466 err = -EINVAL;
4467 goto abort_unlock;
4468 }
4469
4470done_unlock:
4471abort_unlock:
4472 mddev_unlock(mddev);
4473
4474 return err;
4475done:
4476 if (err)
4477 MD_BUG();
4478abort:
4479 return err;
4480}
4481
4482static int md_open(struct inode *inode, struct file *file)
4483{
4484 /*
4485 * Succeed if we can lock the mddev, which confirms that
4486 * it isn't being stopped right now.
4487 */
4488 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4489 int err;
4490
d63a5a74 4491 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
1da177e4
LT		 4492 goto out;
4493
4494 err = 0;
4495 mddev_get(mddev);
4496 mddev_unlock(mddev);
4497
4498 check_disk_change(inode->i_bdev);
4499 out:
4500 return err;
4501}
4502
4503static int md_release(struct inode *inode, struct file * file)
4504{
4505 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4506
52e5f9d1 4507 BUG_ON(!mddev);
1da177e4
LT		 4508 mddev_put(mddev);
4509
4510 return 0;
4511}
4512
4513static int md_media_changed(struct gendisk *disk)
4514{
4515 mddev_t *mddev = disk->private_data;
4516
4517 return mddev->changed;
4518}
4519
4520static int md_revalidate(struct gendisk *disk)
4521{
4522 mddev_t *mddev = disk->private_data;
4523
4524 mddev->changed = 0;
4525 return 0;
4526}
4527static struct block_device_operations md_fops =
4528{
4529 .owner = THIS_MODULE,
4530 .open = md_open,
4531 .release = md_release,
4532 .ioctl = md_ioctl,
a885c8c4 4533 .getgeo = md_getgeo,
1da177e4
LT		 4534 .media_changed = md_media_changed,
4535 .revalidate_disk= md_revalidate,
4536};
4537
75c96f85 4538static int md_thread(void * arg)
1da177e4
LT		 4539{
4540 mdk_thread_t *thread = arg;
4541
1da177e4
LT		 4542 /*
4543 * md_thread is a 'system-thread', it's priority should be very
4544 * high. We avoid resource deadlocks individually in each
4545 * raid personality. (RAID5 does preallocation) We also use RR and
4546 * the very same RT priority as kswapd, thus we will never get
4547 * into a priority inversion deadlock.
4548 *
4549 * we definitely have to have equal or higher priority than
4550 * bdflush, otherwise bdflush will deadlock if there are too
4551 * many dirty RAID5 blocks.
4552 */
1da177e4 4553
4b438a23 4554 current->flags |= PF_NOFREEZE;
6985c43f 4555 allow_signal(SIGKILL);
a6fb0934 4556 while (!kthread_should_stop()) {
1da177e4 4557
93588e22
N		 4558 /* We need to wait INTERRUPTIBLE so that
4559 * we don't add to the load-average.
4560 * That means we need to be sure no signals are
4561 * pending
4562 */
4563 if (signal_pending(current))
4564 flush_signals(current);
4565
4566 wait_event_interruptible_timeout
4567 (thread->wqueue,
4568 test_bit(THREAD_WAKEUP, &thread->flags)
4569 || kthread_should_stop(),
4570 thread->timeout);
1da177e4
LT		 4571
4572 clear_bit(THREAD_WAKEUP, &thread->flags);
4573
787453c2 4574 thread->run(thread->mddev);
1da177e4 4575 }
a6fb0934 4576
1da177e4
LT		 4577 return 0;
4578}
4579
4580void md_wakeup_thread(mdk_thread_t *thread)
4581{
4582 if (thread) {
4583 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4584 set_bit(THREAD_WAKEUP, &thread->flags);
4585 wake_up(&thread->wqueue);
4586 }
4587}
4588
4589mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4590 const char *name)
4591{
4592 mdk_thread_t *thread;
1da177e4 4593
9ffae0cf 4594 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 4595 if (!thread)
4596 return NULL;
4597
1da177e4
LT		 4598 init_waitqueue_head(&thread->wqueue);
4599
1da177e4
LT		 4600 thread->run = run;
4601 thread->mddev = mddev;
32a7627c 4602 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 4603 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 4604 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 4605 kfree(thread);
4606 return NULL;
4607 }
1da177e4
LT		 4608 return thread;
4609}
4610
1da177e4
LT		 4611void md_unregister_thread(mdk_thread_t *thread)
4612{
d28446fe 4613 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 4614
4615 kthread_stop(thread->tsk);
1da177e4
LT		 4616 kfree(thread);
4617}
4618
4619void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4620{
4621 if (!mddev) {
4622 MD_BUG();
4623 return;
4624 }
4625
b2d444d7 4626 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 4627 return;
32a7627c 4628/*
1da177e4
LT		 4629 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4630 mdname(mddev),
4631 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4632 __builtin_return_address(0),__builtin_return_address(1),
4633 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 4634*/
d0a0a5ee
AM		 4635 if (!mddev->pers)
4636 return;
1da177e4
LT		 4637 if (!mddev->pers->error_handler)
4638 return;
4639 mddev->pers->error_handler(mddev,rdev);
4640 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4641 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4642 md_wakeup_thread(mddev->thread);
c331eb04 4643 md_new_event_inintr(mddev);
1da177e4
LT		 4644}
4645
4646/* seq_file implementation /proc/mdstat */
4647
4648static void status_unused(struct seq_file *seq)
4649{
4650 int i = 0;
4651 mdk_rdev_t *rdev;
4652 struct list_head *tmp;
4653
4654 seq_printf(seq, "unused devices: ");
4655
4656 ITERATE_RDEV_PENDING(rdev,tmp) {
4657 char b[BDEVNAME_SIZE];
4658 i++;
4659 seq_printf(seq, "%s ",
4660 bdevname(rdev->bdev,b));
4661 }
4662 if (!i)
4663 seq_printf(seq, "<none>");
4664
4665 seq_printf(seq, "\n");
4666}
4667
4668
4669static void status_resync(struct seq_file *seq, mddev_t * mddev)
4670{
4588b42e
N		 4671 sector_t max_blocks, resync, res;
4672 unsigned long dt, db, rt;
4673 int scale;
4674 unsigned int per_milli;
1da177e4
LT		 4675
4676 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4677
4678 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4679 max_blocks = mddev->resync_max_sectors >> 1;
4680 else
4681 max_blocks = mddev->size;
4682
4683 /*
4684 * Should not happen.
4685 */
4686 if (!max_blocks) {
4687 MD_BUG();
4688 return;
4689 }
4588b42e
N		 4690 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4691 * in a sector_t, and (max_blocks>>scale) will fit in a
4692 * u32, as those are the requirements for sector_div.
4693 * Thus 'scale' must be at least 10
4694 */
4695 scale = 10;
4696 if (sizeof(sector_t) > sizeof(unsigned long)) {
4697 while ( max_blocks/2 > (1ULL<<(scale+32)))
4698 scale++;
4699 }
4700 res = (resync>>scale)*1000;
4701 sector_div(res, (u32)((max_blocks>>scale)+1));
4702
4703 per_milli = res;
1da177e4 4704 {
4588b42e 4705 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 4706 seq_printf(seq, "[");
4707 for (i = 0; i < x; i++)
4708 seq_printf(seq, "=");
4709 seq_printf(seq, ">");
4710 for (i = 0; i < y; i++)
4711 seq_printf(seq, ".");
4712 seq_printf(seq, "] ");
4713 }
4588b42e 4714 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 4715 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4716 "reshape" :
61df9d91
N		 4717 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4718 "check" :
4719 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4720 "resync" : "recovery"))),
4721 per_milli/10, per_milli % 10,
4588b42e
N		 4722 (unsigned long long) resync,
4723 (unsigned long long) max_blocks);
1da177e4
LT		 4724
4725 /*
4726 * We do not want to overflow, so the order of operands and
4727 * the * 100 / 100 trick are important. We do a +1 to be
4728 * safe against division by zero. We only estimate anyway.
4729 *
4730 * dt: time from mark until now
4731 * db: blocks written from mark until now
4732 * rt: remaining time
4733 */
4734 dt = ((jiffies - mddev->resync_mark) / HZ);
4735 if (!dt) dt++;
ff4e8d9a
N		 4736 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4737 - mddev->resync_mark_cnt;
4738 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
1da177e4
LT		 4739
4740 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4741
ff4e8d9a 4742 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 4743}
4744
4745static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4746{
4747 struct list_head *tmp;
4748 loff_t l = *pos;
4749 mddev_t *mddev;
4750
4751 if (l >= 0x10000)
4752 return NULL;
4753 if (!l--)
4754 /* header */
4755 return (void*)1;
4756
4757 spin_lock(&all_mddevs_lock);
4758 list_for_each(tmp,&all_mddevs)
4759 if (!l--) {
4760 mddev = list_entry(tmp, mddev_t, all_mddevs);
4761 mddev_get(mddev);
4762 spin_unlock(&all_mddevs_lock);
4763 return mddev;
4764 }
4765 spin_unlock(&all_mddevs_lock);
4766 if (!l--)
4767 return (void*)2;/* tail */
4768 return NULL;
4769}
4770
4771static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4772{
4773 struct list_head *tmp;
4774 mddev_t *next_mddev, *mddev = v;
4775
4776 ++*pos;
4777 if (v == (void*)2)
4778 return NULL;
4779
4780 spin_lock(&all_mddevs_lock);
4781 if (v == (void*)1)
4782 tmp = all_mddevs.next;
4783 else
4784 tmp = mddev->all_mddevs.next;
4785 if (tmp != &all_mddevs)
4786 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4787 else {
4788 next_mddev = (void*)2;
4789 *pos = 0x10000;
4790 }
4791 spin_unlock(&all_mddevs_lock);
4792
4793 if (v != (void*)1)
4794 mddev_put(mddev);
4795 return next_mddev;
4796
4797}
4798
4799static void md_seq_stop(struct seq_file *seq, void *v)
4800{
4801 mddev_t *mddev = v;
4802
4803 if (mddev && v != (void*)1 && v != (void*)2)
4804 mddev_put(mddev);
4805}
4806
d7603b7e
N		 4807struct mdstat_info {
4808 int event;
4809};
4810
1da177e4
LT		 4811static int md_seq_show(struct seq_file *seq, void *v)
4812{
4813 mddev_t *mddev = v;
4814 sector_t size;
4815 struct list_head *tmp2;
4816 mdk_rdev_t *rdev;
d7603b7e 4817 struct mdstat_info *mi = seq->private;
32a7627c 4818 struct bitmap *bitmap;
1da177e4
LT		 4819
4820 if (v == (void*)1) {
2604b703 4821 struct mdk_personality *pers;
1da177e4
LT		 4822 seq_printf(seq, "Personalities : ");
4823 spin_lock(&pers_lock);
2604b703
N		 4824 list_for_each_entry(pers, &pers_list, list)
4825 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 4826
4827 spin_unlock(&pers_lock);
4828 seq_printf(seq, "\n");
d7603b7e 4829 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 4830 return 0;
4831 }
4832 if (v == (void*)2) {
4833 status_unused(seq);
4834 return 0;
4835 }
4836
5dc5cf7d 4837 if (mddev_lock(mddev) < 0)
1da177e4 4838 return -EINTR;
5dc5cf7d 4839
1da177e4
LT		 4840 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4841 seq_printf(seq, "%s : %sactive", mdname(mddev),
4842 mddev->pers ? "" : "in");
4843 if (mddev->pers) {
f91de92e 4844 if (mddev->ro==1)
1da177e4 4845 seq_printf(seq, " (read-only)");
f91de92e
N		 4846 if (mddev->ro==2)
4847 seq_printf(seq, "(auto-read-only)");
1da177e4
LT		 4848 seq_printf(seq, " %s", mddev->pers->name);
4849 }
4850
4851 size = 0;
4852 ITERATE_RDEV(mddev,rdev,tmp2) {
4853 char b[BDEVNAME_SIZE];
4854 seq_printf(seq, " %s[%d]",
4855 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 4856 if (test_bit(WriteMostly, &rdev->flags))
4857 seq_printf(seq, "(W)");
b2d444d7 4858 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4859 seq_printf(seq, "(F)");
4860 continue;
b325a32e
N		 4861 } else if (rdev->raid_disk < 0)
4862 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 4863 size += rdev->size;
4864 }
4865
4866 if (!list_empty(&mddev->disks)) {
4867 if (mddev->pers)
4868 seq_printf(seq, "\n %llu blocks",
4869 (unsigned long long)mddev->array_size);
4870 else
4871 seq_printf(seq, "\n %llu blocks",
4872 (unsigned long long)size);
4873 }
1cd6bf19
N		 4874 if (mddev->persistent) {
4875 if (mddev->major_version != 0 ||
4876 mddev->minor_version != 90) {
4877 seq_printf(seq," super %d.%d",
4878 mddev->major_version,
4879 mddev->minor_version);
4880 }
4881 } else
4882 seq_printf(seq, " super non-persistent");
1da177e4
LT		 4883
4884 if (mddev->pers) {
4885 mddev->pers->status (seq, mddev);
4886 seq_printf(seq, "\n ");
8e1b39d6
N		 4887 if (mddev->pers->sync_request) {
4888 if (mddev->curr_resync > 2) {
4889 status_resync (seq, mddev);
4890 seq_printf(seq, "\n ");
4891 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4892 seq_printf(seq, "\tresync=DELAYED\n ");
4893 else if (mddev->recovery_cp < MaxSector)
4894 seq_printf(seq, "\tresync=PENDING\n ");
4895 }
32a7627c
N		 4896 } else
4897 seq_printf(seq, "\n ");
4898
4899 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 4900 unsigned long chunk_kb;
4901 unsigned long flags;
32a7627c
N		 4902 spin_lock_irqsave(&bitmap->lock, flags);
4903 chunk_kb = bitmap->chunksize >> 10;
4904 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4905 "%lu%s chunk",
4906 bitmap->pages - bitmap->missing_pages,
4907 bitmap->pages,
4908 (bitmap->pages - bitmap->missing_pages)
4909 << (PAGE_SHIFT - 10),
4910 chunk_kb ? chunk_kb : bitmap->chunksize,
4911 chunk_kb ? "KB" : "B");
78d742d8
N		 4912 if (bitmap->file) {
4913 seq_printf(seq, ", file: ");
c649bb9c
JS		 4914 seq_path(seq, bitmap->file->f_path.mnt,
4915 bitmap->file->f_path.dentry," \t\n");
32a7627c 4916 }
78d742d8 4917
32a7627c
N		 4918 seq_printf(seq, "\n");
4919 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 4920 }
4921
4922 seq_printf(seq, "\n");
4923 }
4924 mddev_unlock(mddev);
4925
4926 return 0;
4927}
4928
4929static struct seq_operations md_seq_ops = {
4930 .start = md_seq_start,
4931 .next = md_seq_next,
4932 .stop = md_seq_stop,
4933 .show = md_seq_show,
4934};
4935
4936static int md_seq_open(struct inode *inode, struct file *file)
4937{
4938 int error;
d7603b7e
N		 4939 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4940 if (mi == NULL)
4941 return -ENOMEM;
1da177e4
LT		 4942
4943 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 4944 if (error)
4945 kfree(mi);
4946 else {
4947 struct seq_file *p = file->private_data;
4948 p->private = mi;
4949 mi->event = atomic_read(&md_event_count);
4950 }
1da177e4
LT		 4951 return error;
4952}
4953
d7603b7e
N		 4954static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4955{
4956 struct seq_file *m = filp->private_data;
4957 struct mdstat_info *mi = m->private;
4958 int mask;
4959
4960 poll_wait(filp, &md_event_waiters, wait);
4961
4962 /* always allow read */
4963 mask = POLLIN | POLLRDNORM;
4964
4965 if (mi->event != atomic_read(&md_event_count))
4966 mask |= POLLERR | POLLPRI;
4967 return mask;
4968}
4969
fa027c2a 4970static const struct file_operations md_seq_fops = {
e24650c2 4971 .owner = THIS_MODULE,
1da177e4
LT		 4972 .open = md_seq_open,
4973 .read = seq_read,
4974 .llseek = seq_lseek,
c3f94b40 4975 .release = seq_release_private,
d7603b7e 4976 .poll = mdstat_poll,
1da177e4
LT		 4977};
4978
2604b703 4979int register_md_personality(struct mdk_personality *p)
1da177e4 4980{
1da177e4 4981 spin_lock(&pers_lock);
2604b703
N		 4982 list_add_tail(&p->list, &pers_list);
4983 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 4984 spin_unlock(&pers_lock);
4985 return 0;
4986}
4987
2604b703 4988int unregister_md_personality(struct mdk_personality *p)
1da177e4 4989{
2604b703 4990 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 4991 spin_lock(&pers_lock);
2604b703 4992 list_del_init(&p->list);
1da177e4
LT		 4993 spin_unlock(&pers_lock);
4994 return 0;
4995}
4996
4997static int is_mddev_idle(mddev_t *mddev)
4998{
4999 mdk_rdev_t * rdev;
5000 struct list_head *tmp;
5001 int idle;
5002 unsigned long curr_events;
5003
5004 idle = 1;
5005 ITERATE_RDEV(mddev,rdev,tmp) {
5006 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 5007 curr_events = disk_stat_read(disk, sectors[0]) +
5008 disk_stat_read(disk, sectors[1]) -
1da177e4 5009 atomic_read(&disk->sync_io);
c0e48521
N		 5010 /* The difference between curr_events and last_events
5011 * will be affected by any new non-sync IO (making
5012 * curr_events bigger) and any difference in the amount of
5013 * in-flight syncio (making current_events bigger or smaller)
5014 * The amount in-flight is currently limited to
5015 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
5016 * which is at most 4096 sectors.
5017 * These numbers are fairly fragile and should be made
5018 * more robust, probably by enforcing the
5019 * 'window size' that md_do_sync sort-of uses.
5020 *
1da177e4
LT		 5021 * Note: the following is an unsigned comparison.
5022 */
c0e48521 5023 if ((curr_events - rdev->last_events + 4096) > 8192) {
1da177e4
LT		 5024 rdev->last_events = curr_events;
5025 idle = 0;
5026 }
5027 }
5028 return idle;
5029}
5030
5031void md_done_sync(mddev_t *mddev, int blocks, int ok)
5032{
5033 /* another "blocks" (512byte) blocks have been synced */
5034 atomic_sub(blocks, &mddev->recovery_active);
5035 wake_up(&mddev->recovery_wait);
5036 if (!ok) {
5037 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5038 md_wakeup_thread(mddev->thread);
5039 // stop recovery, signal do_sync ....
5040 }
5041}
5042
5043
06d91a5f
N		 5044/* md_write_start(mddev, bi)
5045 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 5046 * in superblock) before writing, schedule a superblock update
5047 * and wait for it to complete.
06d91a5f 5048 */
3d310eb7 5049void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 5050{
06d91a5f 5051 if (bio_data_dir(bi) != WRITE)
3d310eb7 5052 return;
06d91a5f 5053
f91de92e
N		 5054 BUG_ON(mddev->ro == 1);
5055 if (mddev->ro == 2) {
5056 /* need to switch to read/write */
5057 mddev->ro = 0;
5058 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5059 md_wakeup_thread(mddev->thread);
5060 }
06d91a5f 5061 atomic_inc(&mddev->writes_pending);
06d91a5f 5062 if (mddev->in_sync) {
a9701a30 5063 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 5064 if (mddev->in_sync) {
5065 mddev->in_sync = 0;
850b2b42 5066 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3d310eb7
N		 5067 md_wakeup_thread(mddev->thread);
5068 }
a9701a30 5069 spin_unlock_irq(&mddev->write_lock);
06d91a5f 5070 }
850b2b42 5071 wait_event(mddev->sb_wait, mddev->flags==0);
1da177e4
LT		 5072}
5073
5074void md_write_end(mddev_t *mddev)
5075{
5076 if (atomic_dec_and_test(&mddev->writes_pending)) {
5077 if (mddev->safemode == 2)
5078 md_wakeup_thread(mddev->thread);
16f17b39 5079 else if (mddev->safemode_delay)
1da177e4
LT		 5080 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5081 }
5082}
5083
2a2275d6
N		 5084/* md_allow_write(mddev)
5085 * Calling this ensures that the array is marked 'active' so that writes
5086 * may proceed without blocking. It is important to call this before
5087 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5088 * Must be called with mddev_lock held.
5089 */
5090void md_allow_write(mddev_t *mddev)
5091{
5092 if (!mddev->pers)
5093 return;
5094 if (mddev->ro)
5095 return;
5096
5097 spin_lock_irq(&mddev->write_lock);
5098 if (mddev->in_sync) {
5099 mddev->in_sync = 0;
5100 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5101 if (mddev->safemode_delay &&
5102 mddev->safemode == 0)
5103 mddev->safemode = 1;
5104 spin_unlock_irq(&mddev->write_lock);
5105 md_update_sb(mddev, 0);
5106 } else
5107 spin_unlock_irq(&mddev->write_lock);
5108}
5109EXPORT_SYMBOL_GPL(md_allow_write);
5110
75c96f85 5111static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 5112
5113#define SYNC_MARKS 10
5114#define SYNC_MARK_STEP (3*HZ)
29269553 5115void md_do_sync(mddev_t *mddev)
1da177e4
LT		 5116{
5117 mddev_t *mddev2;
5118 unsigned int currspeed = 0,
5119 window;
57afd89f 5120 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 5121 unsigned long mark[SYNC_MARKS];
5122 sector_t mark_cnt[SYNC_MARKS];
5123 int last_mark,m;
5124 struct list_head *tmp;
5125 sector_t last_check;
57afd89f 5126 int skipped = 0;
5fd6c1dc
N		 5127 struct list_head *rtmp;
5128 mdk_rdev_t *rdev;
61df9d91 5129 char *desc;
1da177e4
LT		 5130
5131 /* just incase thread restarts... */
5132 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5133 return;
5fd6c1dc
N		 5134 if (mddev->ro) /* never try to sync a read-only array */
5135 return;
1da177e4 5136
61df9d91
N		 5137 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5138 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5139 desc = "data-check";
5140 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5141 desc = "requested-resync";
5142 else
5143 desc = "resync";
5144 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5145 desc = "reshape";
5146 else
5147 desc = "recovery";
5148
1da177e4
LT		 5149 /* we overload curr_resync somewhat here.
5150 * 0 == not engaged in resync at all
5151 * 2 == checking that there is no conflict with another sync
5152 * 1 == like 2, but have yielded to allow conflicting resync to
5153 * commense
5154 * other == active in resync - this many blocks
5155 *
5156 * Before starting a resync we must have set curr_resync to
5157 * 2, and then checked that every "conflicting" array has curr_resync
5158 * less than ours. When we find one that is the same or higher
5159 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5160 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5161 * This will mean we have to start checking from the beginning again.
5162 *
5163 */
5164
5165 do {
5166 mddev->curr_resync = 2;
5167
5168 try_again:
787453c2 5169 if (kthread_should_stop()) {
6985c43f 5170 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 5171 goto skip;
5172 }
5173 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 5174 if (mddev2 == mddev)
5175 continue;
5176 if (mddev2->curr_resync &&
5177 match_mddev_units(mddev,mddev2)) {
5178 DEFINE_WAIT(wq);
5179 if (mddev < mddev2 && mddev->curr_resync == 2) {
5180 /* arbitrarily yield */
5181 mddev->curr_resync = 1;
5182 wake_up(&resync_wait);
5183 }
5184 if (mddev > mddev2 && mddev->curr_resync == 1)
5185 /* no need to wait here, we can wait the next
5186 * time 'round when curr_resync == 2
5187 */
5188 continue;
787453c2
N		 5189 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5190 if (!kthread_should_stop() &&
8712e553 5191 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 5192 printk(KERN_INFO "md: delaying %s of %s"
5193 " until %s has finished (they"
1da177e4 5194 " share one or more physical units)\n",
61df9d91 5195 desc, mdname(mddev), mdname(mddev2));
1da177e4
LT		 5196 mddev_put(mddev2);
5197 schedule();
5198 finish_wait(&resync_wait, &wq);
5199 goto try_again;
5200 }
5201 finish_wait(&resync_wait, &wq);
5202 }
5203 }
5204 } while (mddev->curr_resync < 2);
5205
5fd6c1dc 5206 j = 0;
9d88883e 5207 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 5208 /* resync follows the size requested by the personality,
57afd89f 5209 * which defaults to physical size, but can be virtual size
1da177e4
LT		 5210 */
5211 max_sectors = mddev->resync_max_sectors;
9d88883e 5212 mddev->resync_mismatches = 0;
5fd6c1dc
N		 5213 /* we don't use the checkpoint if there's a bitmap */
5214 if (!mddev->bitmap &&
5215 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5216 j = mddev->recovery_cp;
ccfcc3c1
N		 5217 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5218 max_sectors = mddev->size << 1;
5fd6c1dc 5219 else {
1da177e4
LT		 5220 /* recovery follows the physical size of devices */
5221 max_sectors = mddev->size << 1;
5fd6c1dc
N		 5222 j = MaxSector;
5223 ITERATE_RDEV(mddev,rdev,rtmp)
5224 if (rdev->raid_disk >= 0 &&
5225 !test_bit(Faulty, &rdev->flags) &&
5226 !test_bit(In_sync, &rdev->flags) &&
5227 rdev->recovery_offset < j)
5228 j = rdev->recovery_offset;
5229 }
1da177e4 5230
61df9d91
N		 5231 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5232 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5233 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 5234 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 5235 "(but not more than %d KB/sec) for %s.\n",
5236 speed_max(mddev), desc);
1da177e4
LT		 5237
5238 is_mddev_idle(mddev); /* this also initializes IO event counters */
5fd6c1dc 5239
57afd89f 5240 io_sectors = 0;
1da177e4
LT		 5241 for (m = 0; m < SYNC_MARKS; m++) {
5242 mark[m] = jiffies;
57afd89f 5243 mark_cnt[m] = io_sectors;
1da177e4
LT		 5244 }
5245 last_mark = 0;
5246 mddev->resync_mark = mark[last_mark];
5247 mddev->resync_mark_cnt = mark_cnt[last_mark];
5248
5249 /*
5250 * Tune reconstruction:
5251 */
5252 window = 32*(PAGE_SIZE/512);
5253 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5254 window/2,(unsigned long long) max_sectors/2);
5255
5256 atomic_set(&mddev->recovery_active, 0);
5257 init_waitqueue_head(&mddev->recovery_wait);
5258 last_check = 0;
5259
5260 if (j>2) {
5261 printk(KERN_INFO
61df9d91
N		 5262 "md: resuming %s of %s from checkpoint.\n",
5263 desc, mdname(mddev));
1da177e4
LT		 5264 mddev->curr_resync = j;
5265 }
5266
5267 while (j < max_sectors) {
57afd89f 5268 sector_t sectors;
1da177e4 5269
57afd89f
N		 5270 skipped = 0;
5271 sectors = mddev->pers->sync_request(mddev, j, &skipped,
88202a0c 5272 currspeed < speed_min(mddev));
57afd89f 5273 if (sectors == 0) {
1da177e4
LT		 5274 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5275 goto out;
5276 }
57afd89f
N		 5277
5278 if (!skipped) { /* actual IO requested */
5279 io_sectors += sectors;
5280 atomic_add(sectors, &mddev->recovery_active);
5281 }
5282
1da177e4
LT		 5283 j += sectors;
5284 if (j>1) mddev->curr_resync = j;
ff4e8d9a 5285 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 5286 if (last_check == 0)
5287 /* this is the earliers that rebuilt will be
5288 * visible in /proc/mdstat
5289 */
5290 md_new_event(mddev);
57afd89f
N		 5291
5292 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 5293 continue;
5294
57afd89f 5295 last_check = io_sectors;
1da177e4
LT		 5296
5297 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5298 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5299 break;
5300
5301 repeat:
5302 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5303 /* step marks */
5304 int next = (last_mark+1) % SYNC_MARKS;
5305
5306 mddev->resync_mark = mark[next];
5307 mddev->resync_mark_cnt = mark_cnt[next];
5308 mark[next] = jiffies;
57afd89f 5309 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 5310 last_mark = next;
5311 }
5312
5313
787453c2 5314 if (kthread_should_stop()) {
1da177e4
LT		 5315 /*
5316 * got a signal, exit.
5317 */
5318 printk(KERN_INFO
5319 "md: md_do_sync() got signal ... exiting\n");
1da177e4
LT		 5320 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5321 goto out;
5322 }
5323
5324 /*
5325 * this loop exits only if either when we are slower than
5326 * the 'hard' speed limit, or the system was IO-idle for
5327 * a jiffy.
5328 * the system might be non-idle CPU-wise, but we only care
5329 * about not overloading the IO subsystem. (things like an
5330 * e2fsck being done on the RAID array should execute fast)
5331 */
5332 mddev->queue->unplug_fn(mddev->queue);
5333 cond_resched();
5334
57afd89f
N		 5335 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5336 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 5337
88202a0c
N		 5338 if (currspeed > speed_min(mddev)) {
5339 if ((currspeed > speed_max(mddev)) ||
1da177e4 5340 !is_mddev_idle(mddev)) {
c0e48521 5341 msleep(500);
1da177e4
LT		 5342 goto repeat;
5343 }
5344 }
5345 }
61df9d91 5346 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 5347 /*
5348 * this also signals 'finished resyncing' to md_stop
5349 */
5350 out:
5351 mddev->queue->unplug_fn(mddev->queue);
5352
5353 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5354
5355 /* tell personality that we are finished */
57afd89f 5356 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 5357
5358 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
ccfcc3c1 5359 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 5360 mddev->curr_resync > 2) {
5361 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5362 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5363 if (mddev->curr_resync >= mddev->recovery_cp) {
5364 printk(KERN_INFO
61df9d91
N		 5365 "md: checkpointing %s of %s.\n",
5366 desc, mdname(mddev));
5fd6c1dc
N		 5367 mddev->recovery_cp = mddev->curr_resync;
5368 }
5369 } else
5370 mddev->recovery_cp = MaxSector;
5371 } else {
5372 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5373 mddev->curr_resync = MaxSector;
5374 ITERATE_RDEV(mddev,rdev,rtmp)
5375 if (rdev->raid_disk >= 0 &&
5376 !test_bit(Faulty, &rdev->flags) &&
5377 !test_bit(In_sync, &rdev->flags) &&
5378 rdev->recovery_offset < mddev->curr_resync)
5379 rdev->recovery_offset = mddev->curr_resync;
5fd6c1dc 5380 }
1da177e4 5381 }
17571284 5382 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5383
1da177e4
LT		 5384 skip:
5385 mddev->curr_resync = 0;
5386 wake_up(&resync_wait);
5387 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5388 md_wakeup_thread(mddev->thread);
5389}
29269553 5390EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 5391
5392
b4c4c7b8
N		 5393static int remove_and_add_spares(mddev_t *mddev)
5394{
5395 mdk_rdev_t *rdev;
5396 struct list_head *rtmp;
5397 int spares = 0;
5398
5399 ITERATE_RDEV(mddev,rdev,rtmp)
5400 if (rdev->raid_disk >= 0 &&
5401 (test_bit(Faulty, &rdev->flags) ||
5402 ! test_bit(In_sync, &rdev->flags)) &&
5403 atomic_read(&rdev->nr_pending)==0) {
5404 if (mddev->pers->hot_remove_disk(
5405 mddev, rdev->raid_disk)==0) {
5406 char nm[20];
5407 sprintf(nm,"rd%d", rdev->raid_disk);
5408 sysfs_remove_link(&mddev->kobj, nm);
5409 rdev->raid_disk = -1;
5410 }
5411 }
5412
5413 if (mddev->degraded) {
5414 ITERATE_RDEV(mddev,rdev,rtmp)
5415 if (rdev->raid_disk < 0
5416 && !test_bit(Faulty, &rdev->flags)) {
5417 rdev->recovery_offset = 0;
5418 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5419 char nm[20];
5420 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 5421 if (sysfs_create_link(&mddev->kobj,
5422 &rdev->kobj, nm))
5423 printk(KERN_WARNING
5424 "md: cannot register "
5425 "%s for %s\n",
5426 nm, mdname(mddev));
b4c4c7b8
N		 5427 spares++;
5428 md_new_event(mddev);
5429 } else
5430 break;
5431 }
5432 }
5433 return spares;
5434}
1da177e4
LT		 5435/*
5436 * This routine is regularly called by all per-raid-array threads to
5437 * deal with generic issues like resync and super-block update.
5438 * Raid personalities that don't have a thread (linear/raid0) do not
5439 * need this as they never do any recovery or update the superblock.
5440 *
5441 * It does not do any resync itself, but rather "forks" off other threads
5442 * to do that as needed.
5443 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5444 * "->recovery" and create a thread at ->sync_thread.
5445 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5446 * and wakeups up this thread which will reap the thread and finish up.
5447 * This thread also removes any faulty devices (with nr_pending == 0).
5448 *
5449 * The overall approach is:
5450 * 1/ if the superblock needs updating, update it.
5451 * 2/ If a recovery thread is running, don't do anything else.
5452 * 3/ If recovery has finished, clean up, possibly marking spares active.
5453 * 4/ If there are any faulty devices, remove them.
5454 * 5/ If array is degraded, try to add spares devices
5455 * 6/ If array has spares or is not in-sync, start a resync thread.
5456 */
5457void md_check_recovery(mddev_t *mddev)
5458{
5459 mdk_rdev_t *rdev;
5460 struct list_head *rtmp;
5461
5462
5f40402d
N		 5463 if (mddev->bitmap)
5464 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 5465
5466 if (mddev->ro)
5467 return;
fca4d848
N		 5468
5469 if (signal_pending(current)) {
5470 if (mddev->pers->sync_request) {
5471 printk(KERN_INFO "md: %s in immediate safe mode\n",
5472 mdname(mddev));
5473 mddev->safemode = 2;
5474 }
5475 flush_signals(current);
5476 }
5477
1da177e4 5478 if ( ! (
850b2b42 5479 mddev->flags ||
1da177e4 5480 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 5481 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5482 (mddev->safemode == 1) ||
5483 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5484 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 5485 ))
5486 return;
fca4d848 5487
df5b89b3 5488 if (mddev_trylock(mddev)) {
b4c4c7b8 5489 int spares = 0;
fca4d848 5490
a9701a30 5491 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 5492 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5493 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5494 mddev->in_sync = 1;
850b2b42 5495 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
fca4d848
N		 5496 }
5497 if (mddev->safemode == 1)
5498 mddev->safemode = 0;
a9701a30 5499 spin_unlock_irq(&mddev->write_lock);
fca4d848 5500
850b2b42
N		 5501 if (mddev->flags)
5502 md_update_sb(mddev, 0);
06d91a5f 5503
06d91a5f 5504
1da177e4
LT		 5505 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5506 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5507 /* resync/recovery still happening */
5508 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5509 goto unlock;
5510 }
5511 if (mddev->sync_thread) {
5512 /* resync has finished, collect result */
5513 md_unregister_thread(mddev->sync_thread);
5514 mddev->sync_thread = NULL;
5515 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5516 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5517 /* success...*/
5518 /* activate any spares */
5519 mddev->pers->spare_active(mddev);
5520 }
850b2b42 5521 md_update_sb(mddev, 1);
41158c7e
N		 5522
5523 /* if array is no-longer degraded, then any saved_raid_disk
5524 * information must be scrapped
5525 */
5526 if (!mddev->degraded)
5527 ITERATE_RDEV(mddev,rdev,rtmp)
5528 rdev->saved_raid_disk = -1;
5529
1da177e4
LT		 5530 mddev->recovery = 0;
5531 /* flag recovery needed just to double check */
5532 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
d7603b7e 5533 md_new_event(mddev);
1da177e4
LT		 5534 goto unlock;
5535 }
24dd469d
N		 5536 /* Clear some bits that don't mean anything, but
5537 * might be left set
5538 */
5539 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5540 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5541 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5542 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 5543
5fd6c1dc
N		 5544 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5545 goto unlock;
1da177e4
LT		 5546 /* no recovery is running.
5547 * remove any failed drives, then
5548 * add spares if possible.
5549 * Spare are also removed and re-added, to allow
5550 * the personality to fail the re-add.
5551 */
1da177e4 5552
b4c4c7b8
N		 5553 if (mddev->reshape_position != MaxSector) {
5554 if (mddev->pers->check_reshape(mddev) != 0)
5555 /* Cannot proceed */
5556 goto unlock;
5557 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5558 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 5559 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5560 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5561 } else if (mddev->recovery_cp < MaxSector) {
5562 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5563 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5564 /* nothing to be done ... */
1da177e4 5565 goto unlock;
24dd469d 5566
1da177e4
LT		 5567 if (mddev->pers->sync_request) {
5568 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 5569 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5570 /* We are adding a device or devices to an array
5571 * which has the bitmap stored on all devices.
5572 * So make sure all bitmap pages get written
5573 */
5574 bitmap_write_all(mddev->bitmap);
5575 }
1da177e4
LT		 5576 mddev->sync_thread = md_register_thread(md_do_sync,
5577 mddev,
5578 "%s_resync");
5579 if (!mddev->sync_thread) {
5580 printk(KERN_ERR "%s: could not start resync"
5581 " thread...\n",
5582 mdname(mddev));
5583 /* leave the spares where they are, it shouldn't hurt */
5584 mddev->recovery = 0;
d7603b7e 5585 } else
1da177e4 5586 md_wakeup_thread(mddev->sync_thread);
d7603b7e 5587 md_new_event(mddev);
1da177e4
LT		 5588 }
5589 unlock:
5590 mddev_unlock(mddev);
5591 }
5592}
5593
75c96f85
AB		 5594static int md_notify_reboot(struct notifier_block *this,
5595 unsigned long code, void *x)
1da177e4
LT		 5596{
5597 struct list_head *tmp;
5598 mddev_t *mddev;
5599
5600 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5601
5602 printk(KERN_INFO "md: stopping all md devices.\n");
5603
5604 ITERATE_MDDEV(mddev,tmp)
c71d4887 5605 if (mddev_trylock(mddev)) {
1da177e4 5606 do_md_stop (mddev, 1);
c71d4887
NB		 5607 mddev_unlock(mddev);
5608 }
1da177e4
LT		 5609 /*
5610 * certain more exotic SCSI devices are known to be
5611 * volatile wrt too early system reboots. While the
5612 * right place to handle this issue is the given
5613 * driver, we do want to have a safe RAID driver ...
5614 */
5615 mdelay(1000*1);
5616 }
5617 return NOTIFY_DONE;
5618}
5619
75c96f85 5620static struct notifier_block md_notifier = {
1da177e4
LT		 5621 .notifier_call = md_notify_reboot,
5622 .next = NULL,
5623 .priority = INT_MAX, /* before any real devices */
5624};
5625
5626static void md_geninit(void)
5627{
5628 struct proc_dir_entry *p;
5629
5630 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5631
5632 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5633 if (p)
5634 p->proc_fops = &md_seq_fops;
5635}
5636
75c96f85 5637static int __init md_init(void)
1da177e4 5638{
1da177e4
LT		 5639 if (register_blkdev(MAJOR_NR, "md"))
5640 return -1;
5641 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5642 unregister_blkdev(MAJOR_NR, "md");
5643 return -1;
5644 }
e8703fe1
N		 5645 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5646 md_probe, NULL, NULL);
5647 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 5648 md_probe, NULL, NULL);
5649
1da177e4 5650 register_reboot_notifier(&md_notifier);
0b4d4147 5651 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 5652
5653 md_geninit();
5654 return (0);
5655}
5656
5657
5658#ifndef MODULE
5659
5660/*
5661 * Searches all registered partitions for autorun RAID arrays
5662 * at boot time.
5663 */
5664static dev_t detected_devices[128];
5665static int dev_cnt;
5666
5667void md_autodetect_dev(dev_t dev)
5668{
5669 if (dev_cnt >= 0 && dev_cnt < 127)
5670 detected_devices[dev_cnt++] = dev;
5671}
5672
5673
5674static void autostart_arrays(int part)
5675{
5676 mdk_rdev_t *rdev;
5677 int i;
5678
5679 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5680
5681 for (i = 0; i < dev_cnt; i++) {
5682 dev_t dev = detected_devices[i];
5683
5684 rdev = md_import_device(dev,0, 0);
5685 if (IS_ERR(rdev))
5686 continue;
5687
b2d444d7 5688 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5689 MD_BUG();
5690 continue;
5691 }
5692 list_add(&rdev->same_set, &pending_raid_disks);
5693 }
5694 dev_cnt = 0;
5695
5696 autorun_devices(part);
5697}
5698
fdee8ae4 5699#endif /* !MODULE */
1da177e4
LT		 5700
5701static __exit void md_exit(void)
5702{
5703 mddev_t *mddev;
5704 struct list_head *tmp;
8ab5e4c1 5705
e8703fe1
N		 5706 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5707 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4
LT		 5708
5709 unregister_blkdev(MAJOR_NR,"md");
5710 unregister_blkdev(mdp_major, "mdp");
5711 unregister_reboot_notifier(&md_notifier);
5712 unregister_sysctl_table(raid_table_header);
5713 remove_proc_entry("mdstat", NULL);
5714 ITERATE_MDDEV(mddev,tmp) {
5715 struct gendisk *disk = mddev->gendisk;
5716 if (!disk)
5717 continue;
5718 export_array(mddev);
5719 del_gendisk(disk);
5720 put_disk(disk);
5721 mddev->gendisk = NULL;
5722 mddev_put(mddev);
5723 }
5724}
5725
5726module_init(md_init)
5727module_exit(md_exit)
5728
f91de92e
N		 5729static int get_ro(char *buffer, struct kernel_param *kp)
5730{
5731 return sprintf(buffer, "%d", start_readonly);
5732}
5733static int set_ro(const char *val, struct kernel_param *kp)
5734{
5735 char *e;
5736 int num = simple_strtoul(val, &e, 10);
5737 if (*val && (*e == '\0' || *e == '\n')) {
5738 start_readonly = num;
4dbcdc75 5739 return 0;
f91de92e
N		 5740 }
5741 return -EINVAL;
5742}
5743
80ca3a44
N		 5744module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5745module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 5746
f91de92e 5747
1da177e4
LT		 5748EXPORT_SYMBOL(register_md_personality);
5749EXPORT_SYMBOL(unregister_md_personality);
5750EXPORT_SYMBOL(md_error);
5751EXPORT_SYMBOL(md_done_sync);
5752EXPORT_SYMBOL(md_write_start);
5753EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 5754EXPORT_SYMBOL(md_register_thread);
5755EXPORT_SYMBOL(md_unregister_thread);
5756EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 5757EXPORT_SYMBOL(md_check_recovery);
5758MODULE_LICENSE("GPL");
aa1595e9 5759MODULE_ALIAS("md");
72008652 5760MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)