projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] md: support BIO_RW_BARRIER for md/raid1
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
35#include <linux/module.h>
36#include <linux/config.h>
a6fb0934 37#include <linux/kthread.h>
1da177e4
LT		 38#include <linux/linkage.h>
39#include <linux/raid/md.h>
32a7627c 40#include <linux/raid/bitmap.h>
1da177e4
LT		 41#include <linux/sysctl.h>
42#include <linux/devfs_fs_kernel.h>
43#include <linux/buffer_head.h> /* for invalidate_bdev */
44#include <linux/suspend.h>
45
46#include <linux/init.h>
47
32a7627c
N		 48#include <linux/file.h>
49
1da177e4
LT		 50#ifdef CONFIG_KMOD
51#include <linux/kmod.h>
52#endif
53
54#include <asm/unaligned.h>
55
56#define MAJOR_NR MD_MAJOR
57#define MD_DRIVER
58
59/* 63 partitions with the alternate major number (mdp) */
60#define MdpMinorShift 6
61
62#define DEBUG 0
63#define dprintk(x...) ((void)(DEBUG && printk(x)))
64
65
66#ifndef MODULE
67static void autostart_arrays (int part);
68#endif
69
70static mdk_personality_t *pers[MAX_PERSONALITY];
71static DEFINE_SPINLOCK(pers_lock);
72
73/*
74 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
75 * is 1000 KB/sec, so the extra system load does not show up that much.
76 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 77 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 78 * subsystem is idle. There is also an 'absolute maximum' reconstruction
79 * speed limit - in case reconstruction slows down your system despite
80 * idle IO detection.
81 *
82 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
83 */
84
85static int sysctl_speed_limit_min = 1000;
86static int sysctl_speed_limit_max = 200000;
87
88static struct ctl_table_header *raid_table_header;
89
90static ctl_table raid_table[] = {
91 {
92 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
93 .procname = "speed_limit_min",
94 .data = &sysctl_speed_limit_min,
95 .maxlen = sizeof(int),
96 .mode = 0644,
97 .proc_handler = &proc_dointvec,
98 },
99 {
100 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
101 .procname = "speed_limit_max",
102 .data = &sysctl_speed_limit_max,
103 .maxlen = sizeof(int),
104 .mode = 0644,
105 .proc_handler = &proc_dointvec,
106 },
107 { .ctl_name = 0 }
108};
109
110static ctl_table raid_dir_table[] = {
111 {
112 .ctl_name = DEV_RAID,
113 .procname = "raid",
114 .maxlen = 0,
115 .mode = 0555,
116 .child = raid_table,
117 },
118 { .ctl_name = 0 }
119};
120
121static ctl_table raid_root_table[] = {
122 {
123 .ctl_name = CTL_DEV,
124 .procname = "dev",
125 .maxlen = 0,
126 .mode = 0555,
127 .child = raid_dir_table,
128 },
129 { .ctl_name = 0 }
130};
131
132static struct block_device_operations md_fops;
133
134/*
135 * Enables to iterate over all existing md arrays
136 * all_mddevs_lock protects this list.
137 */
138static LIST_HEAD(all_mddevs);
139static DEFINE_SPINLOCK(all_mddevs_lock);
140
141
142/*
143 * iterates through all used mddevs in the system.
144 * We take care to grab the all_mddevs_lock whenever navigating
145 * the list, and to always hold a refcount when unlocked.
146 * Any code which breaks out of this loop while own
147 * a reference to the current mddev and must mddev_put it.
148 */
149#define ITERATE_MDDEV(mddev,tmp) \
150 \
151 for (({ spin_lock(&all_mddevs_lock); \
152 tmp = all_mddevs.next; \
153 mddev = NULL;}); \
154 ({ if (tmp != &all_mddevs) \
155 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
156 spin_unlock(&all_mddevs_lock); \
157 if (mddev) mddev_put(mddev); \
158 mddev = list_entry(tmp, mddev_t, all_mddevs); \
159 tmp != &all_mddevs;}); \
160 ({ spin_lock(&all_mddevs_lock); \
161 tmp = tmp->next;}) \
162 )
163
164
165static int md_fail_request (request_queue_t *q, struct bio *bio)
166{
167 bio_io_error(bio, bio->bi_size);
168 return 0;
169}
170
171static inline mddev_t *mddev_get(mddev_t *mddev)
172{
173 atomic_inc(&mddev->active);
174 return mddev;
175}
176
177static void mddev_put(mddev_t *mddev)
178{
179 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
180 return;
181 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
182 list_del(&mddev->all_mddevs);
183 blk_put_queue(mddev->queue);
eae1701f 184 kobject_unregister(&mddev->kobj);
1da177e4
LT		 185 }
186 spin_unlock(&all_mddevs_lock);
187}
188
189static mddev_t * mddev_find(dev_t unit)
190{
191 mddev_t *mddev, *new = NULL;
192
193 retry:
194 spin_lock(&all_mddevs_lock);
195 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
196 if (mddev->unit == unit) {
197 mddev_get(mddev);
198 spin_unlock(&all_mddevs_lock);
990a8baf 199 kfree(new);
1da177e4
LT		 200 return mddev;
201 }
202
203 if (new) {
204 list_add(&new->all_mddevs, &all_mddevs);
205 spin_unlock(&all_mddevs_lock);
206 return new;
207 }
208 spin_unlock(&all_mddevs_lock);
209
210 new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
211 if (!new)
212 return NULL;
213
214 memset(new, 0, sizeof(*new));
215
216 new->unit = unit;
217 if (MAJOR(unit) == MD_MAJOR)
218 new->md_minor = MINOR(unit);
219 else
220 new->md_minor = MINOR(unit) >> MdpMinorShift;
221
222 init_MUTEX(&new->reconfig_sem);
223 INIT_LIST_HEAD(&new->disks);
224 INIT_LIST_HEAD(&new->all_mddevs);
225 init_timer(&new->safemode_timer);
226 atomic_set(&new->active, 1);
06d91a5f 227 spin_lock_init(&new->write_lock);
3d310eb7 228 init_waitqueue_head(&new->sb_wait);
1da177e4
LT		 229
230 new->queue = blk_alloc_queue(GFP_KERNEL);
231 if (!new->queue) {
232 kfree(new);
233 return NULL;
234 }
235
236 blk_queue_make_request(new->queue, md_fail_request);
237
238 goto retry;
239}
240
241static inline int mddev_lock(mddev_t * mddev)
242{
243 return down_interruptible(&mddev->reconfig_sem);
244}
245
246static inline void mddev_lock_uninterruptible(mddev_t * mddev)
247{
248 down(&mddev->reconfig_sem);
249}
250
251static inline int mddev_trylock(mddev_t * mddev)
252{
253 return down_trylock(&mddev->reconfig_sem);
254}
255
256static inline void mddev_unlock(mddev_t * mddev)
257{
258 up(&mddev->reconfig_sem);
259
005eca5e 260 md_wakeup_thread(mddev->thread);
1da177e4
LT		 261}
262
263mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
264{
265 mdk_rdev_t * rdev;
266 struct list_head *tmp;
267
268 ITERATE_RDEV(mddev,rdev,tmp) {
269 if (rdev->desc_nr == nr)
270 return rdev;
271 }
272 return NULL;
273}
274
275static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
276{
277 struct list_head *tmp;
278 mdk_rdev_t *rdev;
279
280 ITERATE_RDEV(mddev,rdev,tmp) {
281 if (rdev->bdev->bd_dev == dev)
282 return rdev;
283 }
284 return NULL;
285}
286
77933d72 287static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4
LT		 288{
289 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
290 return MD_NEW_SIZE_BLOCKS(size);
291}
292
293static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
294{
295 sector_t size;
296
297 size = rdev->sb_offset;
298
299 if (chunk_size)
300 size &= ~((sector_t)chunk_size/1024 - 1);
301 return size;
302}
303
304static int alloc_disk_sb(mdk_rdev_t * rdev)
305{
306 if (rdev->sb_page)
307 MD_BUG();
308
309 rdev->sb_page = alloc_page(GFP_KERNEL);
310 if (!rdev->sb_page) {
311 printk(KERN_ALERT "md: out of memory.\n");
312 return -EINVAL;
313 }
314
315 return 0;
316}
317
318static void free_disk_sb(mdk_rdev_t * rdev)
319{
320 if (rdev->sb_page) {
321 page_cache_release(rdev->sb_page);
322 rdev->sb_loaded = 0;
323 rdev->sb_page = NULL;
324 rdev->sb_offset = 0;
325 rdev->size = 0;
326 }
327}
328
329
7bfa19f2
N		 330static int super_written(struct bio *bio, unsigned int bytes_done, int error)
331{
332 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 333 mddev_t *mddev = rdev->mddev;
7bfa19f2
N		 334 if (bio->bi_size)
335 return 1;
336
337 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
a9701a30 338 md_error(mddev, rdev);
7bfa19f2 339
a9701a30
N		 340 if (atomic_dec_and_test(&mddev->pending_writes))
341 wake_up(&mddev->sb_wait);
f8b58edf 342 bio_put(bio);
7bfa19f2
N		 343 return 0;
344}
345
a9701a30
N		 346static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
347{
348 struct bio *bio2 = bio->bi_private;
349 mdk_rdev_t *rdev = bio2->bi_private;
350 mddev_t *mddev = rdev->mddev;
351 if (bio->bi_size)
352 return 1;
353
354 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
355 error == -EOPNOTSUPP) {
356 unsigned long flags;
357 /* barriers don't appear to be supported :-( */
358 set_bit(BarriersNotsupp, &rdev->flags);
359 mddev->barriers_work = 0;
360 spin_lock_irqsave(&mddev->write_lock, flags);
361 bio2->bi_next = mddev->biolist;
362 mddev->biolist = bio2;
363 spin_unlock_irqrestore(&mddev->write_lock, flags);
364 wake_up(&mddev->sb_wait);
365 bio_put(bio);
366 return 0;
367 }
368 bio_put(bio2);
369 bio->bi_private = rdev;
370 return super_written(bio, bytes_done, error);
371}
372
7bfa19f2
N		 373void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
374 sector_t sector, int size, struct page *page)
375{
376 /* write first size bytes of page to sector of rdev
377 * Increment mddev->pending_writes before returning
378 * and decrement it on completion, waking up sb_wait
379 * if zero is reached.
380 * If an error occurred, call md_error
a9701a30
N		 381 *
382 * As we might need to resubmit the request if BIO_RW_BARRIER
383 * causes ENOTSUPP, we allocate a spare bio...
7bfa19f2
N		 384 */
385 struct bio *bio = bio_alloc(GFP_NOIO, 1);
a9701a30 386 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
7bfa19f2
N		 387
388 bio->bi_bdev = rdev->bdev;
389 bio->bi_sector = sector;
390 bio_add_page(bio, page, size, 0);
391 bio->bi_private = rdev;
392 bio->bi_end_io = super_written;
a9701a30
N		 393 bio->bi_rw = rw;
394
7bfa19f2 395 atomic_inc(&mddev->pending_writes);
a9701a30
N		 396 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
397 struct bio *rbio;
398 rw |= (1<<BIO_RW_BARRIER);
399 rbio = bio_clone(bio, GFP_NOIO);
400 rbio->bi_private = bio;
401 rbio->bi_end_io = super_written_barrier;
402 submit_bio(rw, rbio);
403 } else
404 submit_bio(rw, bio);
405}
406
407void md_super_wait(mddev_t *mddev)
408{
409 /* wait for all superblock writes that were scheduled to complete.
410 * if any had to be retried (due to BARRIER problems), retry them
411 */
412 DEFINE_WAIT(wq);
413 for(;;) {
414 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
415 if (atomic_read(&mddev->pending_writes)==0)
416 break;
417 while (mddev->biolist) {
418 struct bio *bio;
419 spin_lock_irq(&mddev->write_lock);
420 bio = mddev->biolist;
421 mddev->biolist = bio->bi_next ;
422 bio->bi_next = NULL;
423 spin_unlock_irq(&mddev->write_lock);
424 submit_bio(bio->bi_rw, bio);
425 }
426 schedule();
427 }
428 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 429}
430
1da177e4
LT		 431static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
432{
433 if (bio->bi_size)
434 return 1;
435
436 complete((struct completion*)bio->bi_private);
437 return 0;
438}
439
a654b9d8 440int sync_page_io(struct block_device *bdev, sector_t sector, int size,
1da177e4
LT		 441 struct page *page, int rw)
442{
baaa2c51 443 struct bio *bio = bio_alloc(GFP_NOIO, 1);
1da177e4
LT		 444 struct completion event;
445 int ret;
446
447 rw |= (1 << BIO_RW_SYNC);
448
449 bio->bi_bdev = bdev;
450 bio->bi_sector = sector;
451 bio_add_page(bio, page, size, 0);
452 init_completion(&event);
453 bio->bi_private = &event;
454 bio->bi_end_io = bi_complete;
455 submit_bio(rw, bio);
456 wait_for_completion(&event);
457
458 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
459 bio_put(bio);
460 return ret;
461}
462
0002b271 463static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 464{
465 char b[BDEVNAME_SIZE];
466 if (!rdev->sb_page) {
467 MD_BUG();
468 return -EINVAL;
469 }
470 if (rdev->sb_loaded)
471 return 0;
472
473
0002b271 474 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
1da177e4
LT		 475 goto fail;
476 rdev->sb_loaded = 1;
477 return 0;
478
479fail:
480 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
481 bdevname(rdev->bdev,b));
482 return -EINVAL;
483}
484
485static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
486{
487 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
488 (sb1->set_uuid1 == sb2->set_uuid1) &&
489 (sb1->set_uuid2 == sb2->set_uuid2) &&
490 (sb1->set_uuid3 == sb2->set_uuid3))
491
492 return 1;
493
494 return 0;
495}
496
497
498static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
499{
500 int ret;
501 mdp_super_t *tmp1, *tmp2;
502
503 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
504 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
505
506 if (!tmp1 || !tmp2) {
507 ret = 0;
508 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
509 goto abort;
510 }
511
512 *tmp1 = *sb1;
513 *tmp2 = *sb2;
514
515 /*
516 * nr_disks is not constant
517 */
518 tmp1->nr_disks = 0;
519 tmp2->nr_disks = 0;
520
521 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
522 ret = 0;
523 else
524 ret = 1;
525
526abort:
990a8baf
JJ		 527 kfree(tmp1);
528 kfree(tmp2);
1da177e4
LT		 529 return ret;
530}
531
532static unsigned int calc_sb_csum(mdp_super_t * sb)
533{
534 unsigned int disk_csum, csum;
535
536 disk_csum = sb->sb_csum;
537 sb->sb_csum = 0;
538 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
539 sb->sb_csum = disk_csum;
540 return csum;
541}
542
543
544/*
545 * Handle superblock details.
546 * We want to be able to handle multiple superblock formats
547 * so we have a common interface to them all, and an array of
548 * different handlers.
549 * We rely on user-space to write the initial superblock, and support
550 * reading and updating of superblocks.
551 * Interface methods are:
552 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
553 * loads and validates a superblock on dev.
554 * if refdev != NULL, compare superblocks on both devices
555 * Return:
556 * 0 - dev has a superblock that is compatible with refdev
557 * 1 - dev has a superblock that is compatible and newer than refdev
558 * so dev should be used as the refdev in future
559 * -EINVAL superblock incompatible or invalid
560 * -othererror e.g. -EIO
561 *
562 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
563 * Verify that dev is acceptable into mddev.
564 * The first time, mddev->raid_disks will be 0, and data from
565 * dev should be merged in. Subsequent calls check that dev
566 * is new enough. Return 0 or -EINVAL
567 *
568 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
569 * Update the superblock for rdev with data in mddev
570 * This does not write to disc.
571 *
572 */
573
574struct super_type {
575 char *name;
576 struct module *owner;
577 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
578 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
579 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
580};
581
582/*
583 * load_super for 0.90.0
584 */
585static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
586{
587 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
588 mdp_super_t *sb;
589 int ret;
590 sector_t sb_offset;
591
592 /*
593 * Calculate the position of the superblock,
594 * it's at the end of the disk.
595 *
596 * It also happens to be a multiple of 4Kb.
597 */
598 sb_offset = calc_dev_sboffset(rdev->bdev);
599 rdev->sb_offset = sb_offset;
600
0002b271 601 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 602 if (ret) return ret;
603
604 ret = -EINVAL;
605
606 bdevname(rdev->bdev, b);
607 sb = (mdp_super_t*)page_address(rdev->sb_page);
608
609 if (sb->md_magic != MD_SB_MAGIC) {
610 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
611 b);
612 goto abort;
613 }
614
615 if (sb->major_version != 0 ||
616 sb->minor_version != 90) {
617 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
618 sb->major_version, sb->minor_version,
619 b);
620 goto abort;
621 }
622
623 if (sb->raid_disks <= 0)
624 goto abort;
625
626 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
627 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
628 b);
629 goto abort;
630 }
631
632 rdev->preferred_minor = sb->md_minor;
633 rdev->data_offset = 0;
0002b271 634 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 635
636 if (sb->level == LEVEL_MULTIPATH)
637 rdev->desc_nr = -1;
638 else
639 rdev->desc_nr = sb->this_disk.number;
640
641 if (refdev == 0)
642 ret = 1;
643 else {
644 __u64 ev1, ev2;
645 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
646 if (!uuid_equal(refsb, sb)) {
647 printk(KERN_WARNING "md: %s has different UUID to %s\n",
648 b, bdevname(refdev->bdev,b2));
649 goto abort;
650 }
651 if (!sb_equal(refsb, sb)) {
652 printk(KERN_WARNING "md: %s has same UUID"
653 " but different superblock to %s\n",
654 b, bdevname(refdev->bdev, b2));
655 goto abort;
656 }
657 ev1 = md_event(sb);
658 ev2 = md_event(refsb);
659 if (ev1 > ev2)
660 ret = 1;
661 else
662 ret = 0;
663 }
664 rdev->size = calc_dev_size(rdev, sb->chunk_size);
665
666 abort:
667 return ret;
668}
669
670/*
671 * validate_super for 0.90.0
672 */
673static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
674{
675 mdp_disk_t *desc;
676 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
677
41158c7e 678 rdev->raid_disk = -1;
b2d444d7 679 rdev->flags = 0;
1da177e4
LT		 680 if (mddev->raid_disks == 0) {
681 mddev->major_version = 0;
682 mddev->minor_version = sb->minor_version;
683 mddev->patch_version = sb->patch_version;
684 mddev->persistent = ! sb->not_persistent;
685 mddev->chunk_size = sb->chunk_size;
686 mddev->ctime = sb->ctime;
687 mddev->utime = sb->utime;
688 mddev->level = sb->level;
689 mddev->layout = sb->layout;
690 mddev->raid_disks = sb->raid_disks;
691 mddev->size = sb->size;
692 mddev->events = md_event(sb);
9223214e 693 mddev->bitmap_offset = 0;
36fa3063 694 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
1da177e4
LT		 695
696 if (sb->state & (1<<MD_SB_CLEAN))
697 mddev->recovery_cp = MaxSector;
698 else {
699 if (sb->events_hi == sb->cp_events_hi &&
700 sb->events_lo == sb->cp_events_lo) {
701 mddev->recovery_cp = sb->recovery_cp;
702 } else
703 mddev->recovery_cp = 0;
704 }
705
706 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
707 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
708 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
709 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
710
711 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 712
713 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
714 mddev->bitmap_file == NULL) {
934ce7c8 715 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
a654b9d8
N		 716 /* FIXME use a better test */
717 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
718 return -EINVAL;
719 }
36fa3063 720 mddev->bitmap_offset = mddev->default_bitmap_offset;
a654b9d8
N		 721 }
722
41158c7e
N		 723 } else if (mddev->pers == NULL) {
724 /* Insist on good event counter while assembling */
725 __u64 ev1 = md_event(sb);
1da177e4
LT		 726 ++ev1;
727 if (ev1 < mddev->events)
728 return -EINVAL;
41158c7e
N		 729 } else if (mddev->bitmap) {
730 /* if adding to array with a bitmap, then we can accept an
731 * older device ... but not too old.
732 */
733 __u64 ev1 = md_event(sb);
734 if (ev1 < mddev->bitmap->events_cleared)
735 return 0;
736 } else /* just a hot-add of a new device, leave raid_disk at -1 */
737 return 0;
738
1da177e4 739 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 740 desc = sb->disks + rdev->desc_nr;
741
742 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 743 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 744 else if (desc->state & (1<<MD_DISK_SYNC) &&
745 desc->raid_disk < mddev->raid_disks) {
b2d444d7 746 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 747 rdev->raid_disk = desc->raid_disk;
748 }
8ddf9efe
N		 749 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
750 set_bit(WriteMostly, &rdev->flags);
41158c7e 751 } else /* MULTIPATH are always insync */
b2d444d7 752 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 753 return 0;
754}
755
756/*
757 * sync_super for 0.90.0
758 */
759static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
760{
761 mdp_super_t *sb;
762 struct list_head *tmp;
763 mdk_rdev_t *rdev2;
764 int next_spare = mddev->raid_disks;
31399d9e 765 char nm[20];
1da177e4
LT		 766
767 /* make rdev->sb match mddev data..
768 *
769 * 1/ zero out disks
770 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
771 * 3/ any empty disks < next_spare become removed
772 *
773 * disks[0] gets initialised to REMOVED because
774 * we cannot be sure from other fields if it has
775 * been initialised or not.
776 */
777 int i;
778 int active=0, working=0,failed=0,spare=0,nr_disks=0;
86e6ffdd 779 unsigned int fixdesc=0;
1da177e4 780
61181565
N		 781 rdev->sb_size = MD_SB_BYTES;
782
1da177e4
LT		 783 sb = (mdp_super_t*)page_address(rdev->sb_page);
784
785 memset(sb, 0, sizeof(*sb));
786
787 sb->md_magic = MD_SB_MAGIC;
788 sb->major_version = mddev->major_version;
789 sb->minor_version = mddev->minor_version;
790 sb->patch_version = mddev->patch_version;
791 sb->gvalid_words = 0; /* ignored */
792 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
793 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
794 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
795 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
796
797 sb->ctime = mddev->ctime;
798 sb->level = mddev->level;
799 sb->size = mddev->size;
800 sb->raid_disks = mddev->raid_disks;
801 sb->md_minor = mddev->md_minor;
802 sb->not_persistent = !mddev->persistent;
803 sb->utime = mddev->utime;
804 sb->state = 0;
805 sb->events_hi = (mddev->events>>32);
806 sb->events_lo = (u32)mddev->events;
807
808 if (mddev->in_sync)
809 {
810 sb->recovery_cp = mddev->recovery_cp;
811 sb->cp_events_hi = (mddev->events>>32);
812 sb->cp_events_lo = (u32)mddev->events;
813 if (mddev->recovery_cp == MaxSector)
814 sb->state = (1<< MD_SB_CLEAN);
815 } else
816 sb->recovery_cp = 0;
817
818 sb->layout = mddev->layout;
819 sb->chunk_size = mddev->chunk_size;
820
a654b9d8
N		 821 if (mddev->bitmap && mddev->bitmap_file == NULL)
822 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
823
1da177e4
LT		 824 sb->disks[0].state = (1<<MD_DISK_REMOVED);
825 ITERATE_RDEV(mddev,rdev2,tmp) {
826 mdp_disk_t *d;
86e6ffdd 827 int desc_nr;
b2d444d7
N		 828 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
829 && !test_bit(Faulty, &rdev2->flags))
86e6ffdd 830 desc_nr = rdev2->raid_disk;
1da177e4 831 else
86e6ffdd
N		 832 desc_nr = next_spare++;
833 if (desc_nr != rdev2->desc_nr) {
834 fixdesc |= (1 << desc_nr);
835 rdev2->desc_nr = desc_nr;
836 if (rdev2->raid_disk >= 0) {
86e6ffdd
N		 837 sprintf(nm, "rd%d", rdev2->raid_disk);
838 sysfs_remove_link(&mddev->kobj, nm);
839 }
840 sysfs_remove_link(&rdev2->kobj, "block");
841 kobject_del(&rdev2->kobj);
842 }
1da177e4
LT		 843 d = &sb->disks[rdev2->desc_nr];
844 nr_disks++;
845 d->number = rdev2->desc_nr;
846 d->major = MAJOR(rdev2->bdev->bd_dev);
847 d->minor = MINOR(rdev2->bdev->bd_dev);
b2d444d7
N		 848 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
849 && !test_bit(Faulty, &rdev2->flags))
1da177e4
LT		 850 d->raid_disk = rdev2->raid_disk;
851 else
852 d->raid_disk = rdev2->desc_nr; /* compatibility */
b2d444d7 853 if (test_bit(Faulty, &rdev2->flags)) {
1da177e4
LT		 854 d->state = (1<<MD_DISK_FAULTY);
855 failed++;
b2d444d7 856 } else if (test_bit(In_sync, &rdev2->flags)) {
1da177e4
LT		 857 d->state = (1<<MD_DISK_ACTIVE);
858 d->state |= (1<<MD_DISK_SYNC);
859 active++;
860 working++;
861 } else {
862 d->state = 0;
863 spare++;
864 working++;
865 }
8ddf9efe
N		 866 if (test_bit(WriteMostly, &rdev2->flags))
867 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 868 }
86e6ffdd
N		 869 if (fixdesc)
870 ITERATE_RDEV(mddev,rdev2,tmp)
871 if (fixdesc & (1<<rdev2->desc_nr)) {
872 snprintf(rdev2->kobj.name, KOBJ_NAME_LEN, "dev%d",
873 rdev2->desc_nr);
9c791977
N		 874 /* kobject_add gets a ref on the parent, so
875 * we have to drop the one we already have
876 */
86e6ffdd 877 kobject_add(&rdev2->kobj);
9c791977 878 kobject_put(rdev->kobj.parent);
86e6ffdd
N		 879 sysfs_create_link(&rdev2->kobj,
880 &rdev2->bdev->bd_disk->kobj,
881 "block");
882 if (rdev2->raid_disk >= 0) {
86e6ffdd
N		 883 sprintf(nm, "rd%d", rdev2->raid_disk);
884 sysfs_create_link(&mddev->kobj,
885 &rdev2->kobj, nm);
886 }
887 }
1da177e4
LT		 888 /* now set the "removed" and "faulty" bits on any missing devices */
889 for (i=0 ; i < mddev->raid_disks ; i++) {
890 mdp_disk_t *d = &sb->disks[i];
891 if (d->state == 0 && d->number == 0) {
892 d->number = i;
893 d->raid_disk = i;
894 d->state = (1<<MD_DISK_REMOVED);
895 d->state |= (1<<MD_DISK_FAULTY);
896 failed++;
897 }
898 }
899 sb->nr_disks = nr_disks;
900 sb->active_disks = active;
901 sb->working_disks = working;
902 sb->failed_disks = failed;
903 sb->spare_disks = spare;
904
905 sb->this_disk = sb->disks[rdev->desc_nr];
906 sb->sb_csum = calc_sb_csum(sb);
907}
908
909/*
910 * version 1 superblock
911 */
912
913static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
914{
915 unsigned int disk_csum, csum;
916 unsigned long long newcsum;
917 int size = 256 + le32_to_cpu(sb->max_dev)*2;
918 unsigned int *isuper = (unsigned int*)sb;
919 int i;
920
921 disk_csum = sb->sb_csum;
922 sb->sb_csum = 0;
923 newcsum = 0;
924 for (i=0; size>=4; size -= 4 )
925 newcsum += le32_to_cpu(*isuper++);
926
927 if (size == 2)
928 newcsum += le16_to_cpu(*(unsigned short*) isuper);
929
930 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
931 sb->sb_csum = disk_csum;
932 return cpu_to_le32(csum);
933}
934
935static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
936{
937 struct mdp_superblock_1 *sb;
938 int ret;
939 sector_t sb_offset;
940 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 941 int bmask;
1da177e4
LT		 942
943 /*
944 * Calculate the position of the superblock.
945 * It is always aligned to a 4K boundary and
946 * depeding on minor_version, it can be:
947 * 0: At least 8K, but less than 12K, from end of device
948 * 1: At start of device
949 * 2: 4K from start of device.
950 */
951 switch(minor_version) {
952 case 0:
953 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
954 sb_offset -= 8*2;
39730960 955 sb_offset &= ~(sector_t)(4*2-1);
1da177e4
LT		 956 /* convert from sectors to K */
957 sb_offset /= 2;
958 break;
959 case 1:
960 sb_offset = 0;
961 break;
962 case 2:
963 sb_offset = 4;
964 break;
965 default:
966 return -EINVAL;
967 }
968 rdev->sb_offset = sb_offset;
969
0002b271
N		 970 /* superblock is rarely larger than 1K, but it can be larger,
971 * and it is safe to read 4k, so we do that
972 */
973 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 974 if (ret) return ret;
975
976
977 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
978
979 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
980 sb->major_version != cpu_to_le32(1) ||
981 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
982 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
71c0805c 983 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 984 return -EINVAL;
985
986 if (calc_sb_1_csum(sb) != sb->sb_csum) {
987 printk("md: invalid superblock checksum on %s\n",
988 bdevname(rdev->bdev,b));
989 return -EINVAL;
990 }
991 if (le64_to_cpu(sb->data_size) < 10) {
992 printk("md: data_size too small on %s\n",
993 bdevname(rdev->bdev,b));
994 return -EINVAL;
995 }
996 rdev->preferred_minor = 0xffff;
997 rdev->data_offset = le64_to_cpu(sb->data_offset);
998
0002b271 999 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
720a3dc3 1000 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
0002b271
N		 1001 if (rdev->sb_size & bmask)
1002 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1003
1da177e4
LT		 1004 if (refdev == 0)
1005 return 1;
1006 else {
1007 __u64 ev1, ev2;
1008 struct mdp_superblock_1 *refsb =
1009 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1010
1011 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1012 sb->level != refsb->level ||
1013 sb->layout != refsb->layout ||
1014 sb->chunksize != refsb->chunksize) {
1015 printk(KERN_WARNING "md: %s has strangely different"
1016 " superblock to %s\n",
1017 bdevname(rdev->bdev,b),
1018 bdevname(refdev->bdev,b2));
1019 return -EINVAL;
1020 }
1021 ev1 = le64_to_cpu(sb->events);
1022 ev2 = le64_to_cpu(refsb->events);
1023
1024 if (ev1 > ev2)
1025 return 1;
1026 }
1027 if (minor_version)
1028 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1029 else
1030 rdev->size = rdev->sb_offset;
1031 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1032 return -EINVAL;
1033 rdev->size = le64_to_cpu(sb->data_size)/2;
1034 if (le32_to_cpu(sb->chunksize))
1035 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1036 return 0;
1037}
1038
1039static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1040{
1041 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1042
41158c7e 1043 rdev->raid_disk = -1;
b2d444d7 1044 rdev->flags = 0;
1da177e4
LT		 1045 if (mddev->raid_disks == 0) {
1046 mddev->major_version = 1;
1047 mddev->patch_version = 0;
1048 mddev->persistent = 1;
1049 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1050 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1051 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1052 mddev->level = le32_to_cpu(sb->level);
1053 mddev->layout = le32_to_cpu(sb->layout);
1054 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1055 mddev->size = le64_to_cpu(sb->size)/2;
1056 mddev->events = le64_to_cpu(sb->events);
9223214e 1057 mddev->bitmap_offset = 0;
36fa3063 1058 mddev->default_bitmap_offset = 0;
53e87fbb 1059 mddev->default_bitmap_offset = 1024;
1da177e4
LT		 1060
1061 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1062 memcpy(mddev->uuid, sb->set_uuid, 16);
1063
1064 mddev->max_disks = (4096-256)/2;
a654b9d8 1065
71c0805c 1066 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
a654b9d8
N		 1067 mddev->bitmap_file == NULL ) {
1068 if (mddev->level != 1) {
1069 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1070 return -EINVAL;
1071 }
1072 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1073 }
41158c7e
N		 1074 } else if (mddev->pers == NULL) {
1075 /* Insist of good event counter while assembling */
1076 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4
LT		 1077 ++ev1;
1078 if (ev1 < mddev->events)
1079 return -EINVAL;
41158c7e
N		 1080 } else if (mddev->bitmap) {
1081 /* If adding to array with a bitmap, then we can accept an
1082 * older device, but not too old.
1083 */
1084 __u64 ev1 = le64_to_cpu(sb->events);
1085 if (ev1 < mddev->bitmap->events_cleared)
1086 return 0;
1087 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1088 return 0;
1da177e4
LT		 1089
1090 if (mddev->level != LEVEL_MULTIPATH) {
1091 int role;
1092 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1093 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1094 switch(role) {
1095 case 0xffff: /* spare */
1da177e4
LT		 1096 break;
1097 case 0xfffe: /* faulty */
b2d444d7 1098 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1099 break;
1100 default:
b2d444d7 1101 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1102 rdev->raid_disk = role;
1103 break;
1104 }
8ddf9efe
N		 1105 if (sb->devflags & WriteMostly1)
1106 set_bit(WriteMostly, &rdev->flags);
41158c7e 1107 } else /* MULTIPATH are always insync */
b2d444d7 1108 set_bit(In_sync, &rdev->flags);
41158c7e 1109
1da177e4
LT		 1110 return 0;
1111}
1112
1113static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1114{
1115 struct mdp_superblock_1 *sb;
1116 struct list_head *tmp;
1117 mdk_rdev_t *rdev2;
1118 int max_dev, i;
1119 /* make rdev->sb match mddev and rdev data. */
1120
1121 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1122
1123 sb->feature_map = 0;
1124 sb->pad0 = 0;
1125 memset(sb->pad1, 0, sizeof(sb->pad1));
1126 memset(sb->pad2, 0, sizeof(sb->pad2));
1127 memset(sb->pad3, 0, sizeof(sb->pad3));
1128
1129 sb->utime = cpu_to_le64((__u64)mddev->utime);
1130 sb->events = cpu_to_le64(mddev->events);
1131 if (mddev->in_sync)
1132 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1133 else
1134 sb->resync_offset = cpu_to_le64(0);
1135
a654b9d8
N		 1136 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1137 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
71c0805c 1138 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8
N		 1139 }
1140
1da177e4
LT		 1141 max_dev = 0;
1142 ITERATE_RDEV(mddev,rdev2,tmp)
1143 if (rdev2->desc_nr+1 > max_dev)
1144 max_dev = rdev2->desc_nr+1;
1145
1146 sb->max_dev = cpu_to_le32(max_dev);
1147 for (i=0; i<max_dev;i++)
1148 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1149
1150 ITERATE_RDEV(mddev,rdev2,tmp) {
1151 i = rdev2->desc_nr;
b2d444d7 1152 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1153 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1154 else if (test_bit(In_sync, &rdev2->flags))
1da177e4
LT		 1155 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1156 else
1157 sb->dev_roles[i] = cpu_to_le16(0xffff);
1158 }
1159
1160 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1161 sb->sb_csum = calc_sb_1_csum(sb);
1162}
1163
1164
75c96f85 1165static struct super_type super_types[] = {
1da177e4
LT		 1166 [0] = {
1167 .name = "0.90.0",
1168 .owner = THIS_MODULE,
1169 .load_super = super_90_load,
1170 .validate_super = super_90_validate,
1171 .sync_super = super_90_sync,
1172 },
1173 [1] = {
1174 .name = "md-1",
1175 .owner = THIS_MODULE,
1176 .load_super = super_1_load,
1177 .validate_super = super_1_validate,
1178 .sync_super = super_1_sync,
1179 },
1180};
1181
1182static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1183{
1184 struct list_head *tmp;
1185 mdk_rdev_t *rdev;
1186
1187 ITERATE_RDEV(mddev,rdev,tmp)
1188 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1189 return rdev;
1190
1191 return NULL;
1192}
1193
1194static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1195{
1196 struct list_head *tmp;
1197 mdk_rdev_t *rdev;
1198
1199 ITERATE_RDEV(mddev1,rdev,tmp)
1200 if (match_dev_unit(mddev2, rdev))
1201 return 1;
1202
1203 return 0;
1204}
1205
1206static LIST_HEAD(pending_raid_disks);
1207
1208static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1209{
1210 mdk_rdev_t *same_pdev;
1211 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1212
1213 if (rdev->mddev) {
1214 MD_BUG();
1215 return -EINVAL;
1216 }
1217 same_pdev = match_dev_unit(mddev, rdev);
1218 if (same_pdev)
1219 printk(KERN_WARNING
1220 "%s: WARNING: %s appears to be on the same physical"
1221 " disk as %s. True\n protection against single-disk"
1222 " failure might be compromised.\n",
1223 mdname(mddev), bdevname(rdev->bdev,b),
1224 bdevname(same_pdev->bdev,b2));
1225
1226 /* Verify rdev->desc_nr is unique.
1227 * If it is -1, assign a free number, else
1228 * check number is not in use
1229 */
1230 if (rdev->desc_nr < 0) {
1231 int choice = 0;
1232 if (mddev->pers) choice = mddev->raid_disks;
1233 while (find_rdev_nr(mddev, choice))
1234 choice++;
1235 rdev->desc_nr = choice;
1236 } else {
1237 if (find_rdev_nr(mddev, rdev->desc_nr))
1238 return -EBUSY;
1239 }
1240
1241 list_add(&rdev->same_set, &mddev->disks);
1242 rdev->mddev = mddev;
1243 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
86e6ffdd
N		 1244
1245 rdev->kobj.k_name = NULL;
1246 snprintf(rdev->kobj.name, KOBJ_NAME_LEN, "dev%d", rdev->desc_nr);
9c791977 1247 rdev->kobj.parent = &mddev->kobj;
86e6ffdd
N		 1248 kobject_add(&rdev->kobj);
1249
1250 sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1da177e4
LT		 1251 return 0;
1252}
1253
1254static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1255{
1256 char b[BDEVNAME_SIZE];
1257 if (!rdev->mddev) {
1258 MD_BUG();
1259 return;
1260 }
1261 list_del_init(&rdev->same_set);
1262 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1263 rdev->mddev = NULL;
86e6ffdd
N		 1264 sysfs_remove_link(&rdev->kobj, "block");
1265 kobject_del(&rdev->kobj);
1da177e4
LT		 1266}
1267
1268/*
1269 * prevent the device from being mounted, repartitioned or
1270 * otherwise reused by a RAID array (or any other kernel
1271 * subsystem), by bd_claiming the device.
1272 */
1273static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1274{
1275 int err = 0;
1276 struct block_device *bdev;
1277 char b[BDEVNAME_SIZE];
1278
1279 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1280 if (IS_ERR(bdev)) {
1281 printk(KERN_ERR "md: could not open %s.\n",
1282 __bdevname(dev, b));
1283 return PTR_ERR(bdev);
1284 }
1285 err = bd_claim(bdev, rdev);
1286 if (err) {
1287 printk(KERN_ERR "md: could not bd_claim %s.\n",
1288 bdevname(bdev, b));
1289 blkdev_put(bdev);
1290 return err;
1291 }
1292 rdev->bdev = bdev;
1293 return err;
1294}
1295
1296static void unlock_rdev(mdk_rdev_t *rdev)
1297{
1298 struct block_device *bdev = rdev->bdev;
1299 rdev->bdev = NULL;
1300 if (!bdev)
1301 MD_BUG();
1302 bd_release(bdev);
1303 blkdev_put(bdev);
1304}
1305
1306void md_autodetect_dev(dev_t dev);
1307
1308static void export_rdev(mdk_rdev_t * rdev)
1309{
1310 char b[BDEVNAME_SIZE];
1311 printk(KERN_INFO "md: export_rdev(%s)\n",
1312 bdevname(rdev->bdev,b));
1313 if (rdev->mddev)
1314 MD_BUG();
1315 free_disk_sb(rdev);
1316 list_del_init(&rdev->same_set);
1317#ifndef MODULE
1318 md_autodetect_dev(rdev->bdev->bd_dev);
1319#endif
1320 unlock_rdev(rdev);
86e6ffdd 1321 kobject_put(&rdev->kobj);
1da177e4
LT		 1322}
1323
1324static void kick_rdev_from_array(mdk_rdev_t * rdev)
1325{
1326 unbind_rdev_from_array(rdev);
1327 export_rdev(rdev);
1328}
1329
1330static void export_array(mddev_t *mddev)
1331{
1332 struct list_head *tmp;
1333 mdk_rdev_t *rdev;
1334
1335 ITERATE_RDEV(mddev,rdev,tmp) {
1336 if (!rdev->mddev) {
1337 MD_BUG();
1338 continue;
1339 }
1340 kick_rdev_from_array(rdev);
1341 }
1342 if (!list_empty(&mddev->disks))
1343 MD_BUG();
1344 mddev->raid_disks = 0;
1345 mddev->major_version = 0;
1346}
1347
1348static void print_desc(mdp_disk_t *desc)
1349{
1350 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1351 desc->major,desc->minor,desc->raid_disk,desc->state);
1352}
1353
1354static void print_sb(mdp_super_t *sb)
1355{
1356 int i;
1357
1358 printk(KERN_INFO
1359 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1360 sb->major_version, sb->minor_version, sb->patch_version,
1361 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1362 sb->ctime);
1363 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1364 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1365 sb->md_minor, sb->layout, sb->chunk_size);
1366 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1367 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1368 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1369 sb->failed_disks, sb->spare_disks,
1370 sb->sb_csum, (unsigned long)sb->events_lo);
1371
1372 printk(KERN_INFO);
1373 for (i = 0; i < MD_SB_DISKS; i++) {
1374 mdp_disk_t *desc;
1375
1376 desc = sb->disks + i;
1377 if (desc->number || desc->major || desc->minor ||
1378 desc->raid_disk || (desc->state && (desc->state != 4))) {
1379 printk(" D %2d: ", i);
1380 print_desc(desc);
1381 }
1382 }
1383 printk(KERN_INFO "md: THIS: ");
1384 print_desc(&sb->this_disk);
1385
1386}
1387
1388static void print_rdev(mdk_rdev_t *rdev)
1389{
1390 char b[BDEVNAME_SIZE];
1391 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1392 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
b2d444d7
N		 1393 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1394 rdev->desc_nr);
1da177e4
LT		 1395 if (rdev->sb_loaded) {
1396 printk(KERN_INFO "md: rdev superblock:\n");
1397 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1398 } else
1399 printk(KERN_INFO "md: no rdev superblock!\n");
1400}
1401
1402void md_print_devices(void)
1403{
1404 struct list_head *tmp, *tmp2;
1405 mdk_rdev_t *rdev;
1406 mddev_t *mddev;
1407 char b[BDEVNAME_SIZE];
1408
1409 printk("\n");
1410 printk("md: **********************************\n");
1411 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1412 printk("md: **********************************\n");
1413 ITERATE_MDDEV(mddev,tmp) {
1da177e4 1414
32a7627c
N		 1415 if (mddev->bitmap)
1416 bitmap_print_sb(mddev->bitmap);
1417 else
1418 printk("%s: ", mdname(mddev));
1da177e4
LT		 1419 ITERATE_RDEV(mddev,rdev,tmp2)
1420 printk("<%s>", bdevname(rdev->bdev,b));
1421 printk("\n");
1422
1423 ITERATE_RDEV(mddev,rdev,tmp2)
1424 print_rdev(rdev);
1425 }
1426 printk("md: **********************************\n");
1427 printk("\n");
1428}
1429
1430
1da177e4
LT		 1431static void sync_sbs(mddev_t * mddev)
1432{
1433 mdk_rdev_t *rdev;
1434 struct list_head *tmp;
1435
1436 ITERATE_RDEV(mddev,rdev,tmp) {
1437 super_types[mddev->major_version].
1438 sync_super(mddev, rdev);
1439 rdev->sb_loaded = 1;
1440 }
1441}
1442
1443static void md_update_sb(mddev_t * mddev)
1444{
7bfa19f2 1445 int err;
1da177e4
LT		 1446 struct list_head *tmp;
1447 mdk_rdev_t *rdev;
06d91a5f 1448 int sync_req;
1da177e4 1449
1da177e4 1450repeat:
a9701a30 1451 spin_lock_irq(&mddev->write_lock);
06d91a5f 1452 sync_req = mddev->in_sync;
1da177e4
LT		 1453 mddev->utime = get_seconds();
1454 mddev->events ++;
1455
1456 if (!mddev->events) {
1457 /*
1458 * oops, this 64-bit counter should never wrap.
1459 * Either we are in around ~1 trillion A.C., assuming
1460 * 1 reboot per second, or we have a bug:
1461 */
1462 MD_BUG();
1463 mddev->events --;
1464 }
7bfa19f2 1465 mddev->sb_dirty = 2;
1da177e4
LT		 1466 sync_sbs(mddev);
1467
1468 /*
1469 * do not write anything to disk if using
1470 * nonpersistent superblocks
1471 */
06d91a5f
N		 1472 if (!mddev->persistent) {
1473 mddev->sb_dirty = 0;
a9701a30 1474 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1475 wake_up(&mddev->sb_wait);
1da177e4 1476 return;
06d91a5f 1477 }
a9701a30 1478 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 1479
1480 dprintk(KERN_INFO
1481 "md: updating %s RAID superblock on device (in sync %d)\n",
1482 mdname(mddev),mddev->in_sync);
1483
32a7627c 1484 err = bitmap_update_sb(mddev->bitmap);
1da177e4
LT		 1485 ITERATE_RDEV(mddev,rdev,tmp) {
1486 char b[BDEVNAME_SIZE];
1487 dprintk(KERN_INFO "md: ");
b2d444d7 1488 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 1489 dprintk("(skipping faulty ");
1490
1491 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 1492 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 1493 md_super_write(mddev,rdev,
0002b271 1494 rdev->sb_offset<<1, rdev->sb_size,
7bfa19f2
N		 1495 rdev->sb_page);
1496 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1497 bdevname(rdev->bdev,b),
1498 (unsigned long long)rdev->sb_offset);
1499
1da177e4
LT		 1500 } else
1501 dprintk(")\n");
7bfa19f2 1502 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 1503 /* only need to write one superblock... */
1504 break;
1505 }
a9701a30 1506 md_super_wait(mddev);
7bfa19f2
N		 1507 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1508
a9701a30 1509 spin_lock_irq(&mddev->write_lock);
7bfa19f2 1510 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
06d91a5f 1511 /* have to write it out again */
a9701a30 1512 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 1513 goto repeat;
1514 }
1515 mddev->sb_dirty = 0;
a9701a30 1516 spin_unlock_irq(&mddev->write_lock);
3d310eb7 1517 wake_up(&mddev->sb_wait);
06d91a5f 1518
1da177e4
LT		 1519}
1520
86e6ffdd
N		 1521struct rdev_sysfs_entry {
1522 struct attribute attr;
1523 ssize_t (*show)(mdk_rdev_t *, char *);
1524 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1525};
1526
1527static ssize_t
1528rdev_show_state(mdk_rdev_t *rdev, char *page)
1529{
1530 char *sep = "";
1531 int len=0;
1532
b2d444d7 1533 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 1534 len+= sprintf(page+len, "%sfaulty",sep);
1535 sep = ",";
1536 }
b2d444d7 1537 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1538 len += sprintf(page+len, "%sin_sync",sep);
1539 sep = ",";
1540 }
b2d444d7
N		 1541 if (!test_bit(Faulty, &rdev->flags) &&
1542 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 1543 len += sprintf(page+len, "%sspare", sep);
1544 sep = ",";
1545 }
1546 return len+sprintf(page+len, "\n");
1547}
1548
1549static struct rdev_sysfs_entry rdev_state = {
1550 .attr = {.name = "state", .mode = S_IRUGO },
1551 .show = rdev_show_state,
1552};
1553
1554static ssize_t
1555rdev_show_super(mdk_rdev_t *rdev, char *page)
1556{
1557 if (rdev->sb_loaded && rdev->sb_size) {
1558 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1559 return rdev->sb_size;
1560 } else
1561 return 0;
1562}
1563static struct rdev_sysfs_entry rdev_super = {
1564 .attr = {.name = "super", .mode = S_IRUGO },
1565 .show = rdev_show_super,
1566};
1567static struct attribute *rdev_default_attrs[] = {
1568 &rdev_state.attr,
1569 &rdev_super.attr,
1570 NULL,
1571};
1572static ssize_t
1573rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1574{
1575 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1576 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1577
1578 if (!entry->show)
1579 return -EIO;
1580 return entry->show(rdev, page);
1581}
1582
1583static ssize_t
1584rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1585 const char *page, size_t length)
1586{
1587 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1588 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1589
1590 if (!entry->store)
1591 return -EIO;
1592 return entry->store(rdev, page, length);
1593}
1594
1595static void rdev_free(struct kobject *ko)
1596{
1597 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1598 kfree(rdev);
1599}
1600static struct sysfs_ops rdev_sysfs_ops = {
1601 .show = rdev_attr_show,
1602 .store = rdev_attr_store,
1603};
1604static struct kobj_type rdev_ktype = {
1605 .release = rdev_free,
1606 .sysfs_ops = &rdev_sysfs_ops,
1607 .default_attrs = rdev_default_attrs,
1608};
1609
1da177e4
LT		 1610/*
1611 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1612 *
1613 * mark the device faulty if:
1614 *
1615 * - the device is nonexistent (zero size)
1616 * - the device has no valid superblock
1617 *
1618 * a faulty rdev _never_ has rdev->sb set.
1619 */
1620static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1621{
1622 char b[BDEVNAME_SIZE];
1623 int err;
1624 mdk_rdev_t *rdev;
1625 sector_t size;
1626
1627 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1628 if (!rdev) {
1629 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1630 return ERR_PTR(-ENOMEM);
1631 }
1632 memset(rdev, 0, sizeof(*rdev));
1633
1634 if ((err = alloc_disk_sb(rdev)))
1635 goto abort_free;
1636
1637 err = lock_rdev(rdev, newdev);
1638 if (err)
1639 goto abort_free;
1640
86e6ffdd
N		 1641 rdev->kobj.parent = NULL;
1642 rdev->kobj.ktype = &rdev_ktype;
1643 kobject_init(&rdev->kobj);
1644
1da177e4 1645 rdev->desc_nr = -1;
b2d444d7 1646 rdev->flags = 0;
1da177e4
LT		 1647 rdev->data_offset = 0;
1648 atomic_set(&rdev->nr_pending, 0);
ba22dcbf 1649 atomic_set(&rdev->read_errors, 0);
1da177e4
LT		 1650
1651 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1652 if (!size) {
1653 printk(KERN_WARNING
1654 "md: %s has zero or unknown size, marking faulty!\n",
1655 bdevname(rdev->bdev,b));
1656 err = -EINVAL;
1657 goto abort_free;
1658 }
1659
1660 if (super_format >= 0) {
1661 err = super_types[super_format].
1662 load_super(rdev, NULL, super_minor);
1663 if (err == -EINVAL) {
1664 printk(KERN_WARNING
1665 "md: %s has invalid sb, not importing!\n",
1666 bdevname(rdev->bdev,b));
1667 goto abort_free;
1668 }
1669 if (err < 0) {
1670 printk(KERN_WARNING
1671 "md: could not read %s's sb, not importing!\n",
1672 bdevname(rdev->bdev,b));
1673 goto abort_free;
1674 }
1675 }
1676 INIT_LIST_HEAD(&rdev->same_set);
1677
1678 return rdev;
1679
1680abort_free:
1681 if (rdev->sb_page) {
1682 if (rdev->bdev)
1683 unlock_rdev(rdev);
1684 free_disk_sb(rdev);
1685 }
1686 kfree(rdev);
1687 return ERR_PTR(err);
1688}
1689
1690/*
1691 * Check a full RAID array for plausibility
1692 */
1693
1694
a757e64c 1695static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 1696{
1697 int i;
1698 struct list_head *tmp;
1699 mdk_rdev_t *rdev, *freshest;
1700 char b[BDEVNAME_SIZE];
1701
1702 freshest = NULL;
1703 ITERATE_RDEV(mddev,rdev,tmp)
1704 switch (super_types[mddev->major_version].
1705 load_super(rdev, freshest, mddev->minor_version)) {
1706 case 1:
1707 freshest = rdev;
1708 break;
1709 case 0:
1710 break;
1711 default:
1712 printk( KERN_ERR \
1713 "md: fatal superblock inconsistency in %s"
1714 " -- removing from array\n",
1715 bdevname(rdev->bdev,b));
1716 kick_rdev_from_array(rdev);
1717 }
1718
1719
1720 super_types[mddev->major_version].
1721 validate_super(mddev, freshest);
1722
1723 i = 0;
1724 ITERATE_RDEV(mddev,rdev,tmp) {
1725 if (rdev != freshest)
1726 if (super_types[mddev->major_version].
1727 validate_super(mddev, rdev)) {
1728 printk(KERN_WARNING "md: kicking non-fresh %s"
1729 " from array!\n",
1730 bdevname(rdev->bdev,b));
1731 kick_rdev_from_array(rdev);
1732 continue;
1733 }
1734 if (mddev->level == LEVEL_MULTIPATH) {
1735 rdev->desc_nr = i++;
1736 rdev->raid_disk = rdev->desc_nr;
b2d444d7 1737 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1738 }
1739 }
1740
1741
1742
1743 if (mddev->recovery_cp != MaxSector &&
1744 mddev->level >= 1)
1745 printk(KERN_ERR "md: %s: raid array is not clean"
1746 " -- starting background reconstruction\n",
1747 mdname(mddev));
1748
1da177e4
LT		 1749}
1750
eae1701f
N		 1751static ssize_t
1752md_show_level(mddev_t *mddev, char *page)
1753{
1754 mdk_personality_t *p = mddev->pers;
1755 if (p == NULL)
1756 return 0;
1757 if (mddev->level >= 0)
1758 return sprintf(page, "RAID-%d\n", mddev->level);
1759 else
1760 return sprintf(page, "%s\n", p->name);
1761}
1762
1763static struct md_sysfs_entry md_level = {
1764 .attr = {.name = "level", .mode = S_IRUGO },
1765 .show = md_show_level,
1766};
1767
1768static ssize_t
1769md_show_rdisks(mddev_t *mddev, char *page)
1770{
1771 return sprintf(page, "%d\n", mddev->raid_disks);
1772}
1773
1774static struct md_sysfs_entry md_raid_disks = {
1775 .attr = {.name = "raid_disks", .mode = S_IRUGO },
1776 .show = md_show_rdisks,
1777};
1778
24dd469d
N		 1779static ssize_t
1780md_show_scan(mddev_t *mddev, char *page)
1781{
1782 char *type = "none";
31399d9e
N		 1783 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1784 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1785 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 1786 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1787 type = "resync";
1788 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1789 type = "check";
1790 else
1791 type = "repair";
1792 } else
1793 type = "recover";
1794 }
1795 return sprintf(page, "%s\n", type);
1796}
1797
1798static ssize_t
1799md_store_scan(mddev_t *mddev, const char *page, size_t len)
1800{
1801 int canscan=0;
31399d9e
N		 1802
1803 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1804 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d
N		 1805 return -EBUSY;
1806 down(&mddev->reconfig_sem);
1807 if (mddev->pers && mddev->pers->sync_request)
1808 canscan=1;
1809 up(&mddev->reconfig_sem);
1810 if (!canscan)
1811 return -EINVAL;
1812
1813 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1814 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1815 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1816 return -EINVAL;
1817 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1818 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1819 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1820 md_wakeup_thread(mddev->thread);
1821 return len;
1822}
1823
9d88883e
N		 1824static ssize_t
1825md_show_mismatch(mddev_t *mddev, char *page)
1826{
1827 return sprintf(page, "%llu\n",
1828 (unsigned long long) mddev->resync_mismatches);
1829}
1830
24dd469d
N		 1831static struct md_sysfs_entry md_scan_mode = {
1832 .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1833 .show = md_show_scan,
1834 .store = md_store_scan,
1835};
1836
9d88883e
N		 1837static struct md_sysfs_entry md_mismatches = {
1838 .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1839 .show = md_show_mismatch,
1840};
1841
eae1701f
N		 1842static struct attribute *md_default_attrs[] = {
1843 &md_level.attr,
1844 &md_raid_disks.attr,
24dd469d 1845 &md_scan_mode.attr,
9d88883e 1846 &md_mismatches.attr,
eae1701f
N		 1847 NULL,
1848};
1849
1850static ssize_t
1851md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1852{
1853 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1854 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1855
1856 if (!entry->show)
1857 return -EIO;
1858 return entry->show(mddev, page);
1859}
1860
1861static ssize_t
1862md_attr_store(struct kobject *kobj, struct attribute *attr,
1863 const char *page, size_t length)
1864{
1865 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1866 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1867
1868 if (!entry->store)
1869 return -EIO;
1870 return entry->store(mddev, page, length);
1871}
1872
1873static void md_free(struct kobject *ko)
1874{
1875 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1876 kfree(mddev);
1877}
1878
1879static struct sysfs_ops md_sysfs_ops = {
1880 .show = md_attr_show,
1881 .store = md_attr_store,
1882};
1883static struct kobj_type md_ktype = {
1884 .release = md_free,
1885 .sysfs_ops = &md_sysfs_ops,
1886 .default_attrs = md_default_attrs,
1887};
1888
1da177e4
LT		 1889int mdp_major = 0;
1890
1891static struct kobject *md_probe(dev_t dev, int *part, void *data)
1892{
1893 static DECLARE_MUTEX(disks_sem);
1894 mddev_t *mddev = mddev_find(dev);
1895 struct gendisk *disk;
1896 int partitioned = (MAJOR(dev) != MD_MAJOR);
1897 int shift = partitioned ? MdpMinorShift : 0;
1898 int unit = MINOR(dev) >> shift;
1899
1900 if (!mddev)
1901 return NULL;
1902
1903 down(&disks_sem);
1904 if (mddev->gendisk) {
1905 up(&disks_sem);
1906 mddev_put(mddev);
1907 return NULL;
1908 }
1909 disk = alloc_disk(1 << shift);
1910 if (!disk) {
1911 up(&disks_sem);
1912 mddev_put(mddev);
1913 return NULL;
1914 }
1915 disk->major = MAJOR(dev);
1916 disk->first_minor = unit << shift;
1917 if (partitioned) {
1918 sprintf(disk->disk_name, "md_d%d", unit);
1919 sprintf(disk->devfs_name, "md/d%d", unit);
1920 } else {
1921 sprintf(disk->disk_name, "md%d", unit);
1922 sprintf(disk->devfs_name, "md/%d", unit);
1923 }
1924 disk->fops = &md_fops;
1925 disk->private_data = mddev;
1926 disk->queue = mddev->queue;
1927 add_disk(disk);
1928 mddev->gendisk = disk;
1929 up(&disks_sem);
9c791977 1930 mddev->kobj.parent = &disk->kobj;
eae1701f
N		 1931 mddev->kobj.k_name = NULL;
1932 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1933 mddev->kobj.ktype = &md_ktype;
1934 kobject_register(&mddev->kobj);
1da177e4
LT		 1935 return NULL;
1936}
1937
1938void md_wakeup_thread(mdk_thread_t *thread);
1939
1940static void md_safemode_timeout(unsigned long data)
1941{
1942 mddev_t *mddev = (mddev_t *) data;
1943
1944 mddev->safemode = 1;
1945 md_wakeup_thread(mddev->thread);
1946}
1947
1948
1949static int do_md_run(mddev_t * mddev)
1950{
1951 int pnum, err;
1952 int chunk_size;
1953 struct list_head *tmp;
1954 mdk_rdev_t *rdev;
1955 struct gendisk *disk;
1956 char b[BDEVNAME_SIZE];
1957
a757e64c
N		 1958 if (list_empty(&mddev->disks))
1959 /* cannot run an array with no devices.. */
1da177e4 1960 return -EINVAL;
1da177e4
LT		 1961
1962 if (mddev->pers)
1963 return -EBUSY;
1964
1965 /*
1966 * Analyze all RAID superblock(s)
1967 */
a757e64c
N		 1968 if (!mddev->raid_disks)
1969 analyze_sbs(mddev);
1da177e4
LT		 1970
1971 chunk_size = mddev->chunk_size;
1972 pnum = level_to_pers(mddev->level);
1973
1974 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1975 if (!chunk_size) {
1976 /*
1977 * 'default chunksize' in the old md code used to
1978 * be PAGE_SIZE, baaad.
1979 * we abort here to be on the safe side. We don't
1980 * want to continue the bad practice.
1981 */
1982 printk(KERN_ERR
1983 "no chunksize specified, see 'man raidtab'\n");
1984 return -EINVAL;
1985 }
1986 if (chunk_size > MAX_CHUNK_SIZE) {
1987 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1988 chunk_size, MAX_CHUNK_SIZE);
1989 return -EINVAL;
1990 }
1991 /*
1992 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1993 */
1994 if ( (1 << ffz(~chunk_size)) != chunk_size) {
a757e64c 1995 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1da177e4
LT		 1996 return -EINVAL;
1997 }
1998 if (chunk_size < PAGE_SIZE) {
1999 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2000 chunk_size, PAGE_SIZE);
2001 return -EINVAL;
2002 }
2003
2004 /* devices must have minimum size of one chunk */
2005 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2006 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2007 continue;
2008 if (rdev->size < chunk_size / 1024) {
2009 printk(KERN_WARNING
2010 "md: Dev %s smaller than chunk_size:"
2011 " %lluk < %dk\n",
2012 bdevname(rdev->bdev,b),
2013 (unsigned long long)rdev->size,
2014 chunk_size / 1024);
2015 return -EINVAL;
2016 }
2017 }
2018 }
2019
1da177e4
LT		 2020#ifdef CONFIG_KMOD
2021 if (!pers[pnum])
2022 {
2023 request_module("md-personality-%d", pnum);
2024 }
2025#endif
2026
2027 /*
2028 * Drop all container device buffers, from now on
2029 * the only valid external interface is through the md
2030 * device.
2031 * Also find largest hardsector size
2032 */
2033 ITERATE_RDEV(mddev,rdev,tmp) {
b2d444d7 2034 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2035 continue;
2036 sync_blockdev(rdev->bdev);
2037 invalidate_bdev(rdev->bdev, 0);
2038 }
2039
2040 md_probe(mddev->unit, NULL, NULL);
2041 disk = mddev->gendisk;
2042 if (!disk)
2043 return -ENOMEM;
2044
2045 spin_lock(&pers_lock);
2046 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2047 spin_unlock(&pers_lock);
2048 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2049 pnum);
2050 return -EINVAL;
2051 }
2052
2053 mddev->pers = pers[pnum];
2054 spin_unlock(&pers_lock);
2055
657390d2 2056 mddev->recovery = 0;
1da177e4 2057 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
a9701a30 2058 mddev->barriers_work = 1;
1da177e4 2059
32a7627c
N		 2060 /* before we start the array running, initialise the bitmap */
2061 err = bitmap_create(mddev);
2062 if (err)
2063 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2064 mdname(mddev), err);
2065 else
2066 err = mddev->pers->run(mddev);
1da177e4
LT		 2067 if (err) {
2068 printk(KERN_ERR "md: pers->run() failed ...\n");
2069 module_put(mddev->pers->owner);
2070 mddev->pers = NULL;
32a7627c
N		 2071 bitmap_destroy(mddev);
2072 return err;
1da177e4
LT		 2073 }
2074 atomic_set(&mddev->writes_pending,0);
2075 mddev->safemode = 0;
2076 mddev->safemode_timer.function = md_safemode_timeout;
2077 mddev->safemode_timer.data = (unsigned long) mddev;
2078 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2079 mddev->in_sync = 1;
86e6ffdd
N		 2080
2081 ITERATE_RDEV(mddev,rdev,tmp)
2082 if (rdev->raid_disk >= 0) {
2083 char nm[20];
2084 sprintf(nm, "rd%d", rdev->raid_disk);
2085 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2086 }
1da177e4
LT		 2087
2088 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2089 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2090
2091 if (mddev->sb_dirty)
2092 md_update_sb(mddev);
2093
2094 set_capacity(disk, mddev->array_size<<1);
2095
2096 /* If we call blk_queue_make_request here, it will
2097 * re-initialise max_sectors etc which may have been
2098 * refined inside -> run. So just set the bits we need to set.
2099 * Most initialisation happended when we called
2100 * blk_queue_make_request(..., md_fail_request)
2101 * earlier.
2102 */
2103 mddev->queue->queuedata = mddev;
2104 mddev->queue->make_request_fn = mddev->pers->make_request;
2105
2106 mddev->changed = 1;
2107 return 0;
2108}
2109
2110static int restart_array(mddev_t *mddev)
2111{
2112 struct gendisk *disk = mddev->gendisk;
2113 int err;
2114
2115 /*
2116 * Complain if it has no devices
2117 */
2118 err = -ENXIO;
2119 if (list_empty(&mddev->disks))
2120 goto out;
2121
2122 if (mddev->pers) {
2123 err = -EBUSY;
2124 if (!mddev->ro)
2125 goto out;
2126
2127 mddev->safemode = 0;
2128 mddev->ro = 0;
2129 set_disk_ro(disk, 0);
2130
2131 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2132 mdname(mddev));
2133 /*
2134 * Kick recovery or resync if necessary
2135 */
2136 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2137 md_wakeup_thread(mddev->thread);
2138 err = 0;
2139 } else {
2140 printk(KERN_ERR "md: %s has no personality assigned.\n",
2141 mdname(mddev));
2142 err = -EINVAL;
2143 }
2144
2145out:
2146 return err;
2147}
2148
2149static int do_md_stop(mddev_t * mddev, int ro)
2150{
2151 int err = 0;
2152 struct gendisk *disk = mddev->gendisk;
2153
2154 if (mddev->pers) {
2155 if (atomic_read(&mddev->active)>2) {
2156 printk("md: %s still in use.\n",mdname(mddev));
2157 return -EBUSY;
2158 }
2159
2160 if (mddev->sync_thread) {
2161 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2162 md_unregister_thread(mddev->sync_thread);
2163 mddev->sync_thread = NULL;
2164 }
2165
2166 del_timer_sync(&mddev->safemode_timer);
2167
2168 invalidate_partition(disk, 0);
2169
2170 if (ro) {
2171 err = -ENXIO;
2172 if (mddev->ro)
2173 goto out;
2174 mddev->ro = 1;
2175 } else {
6b8b3e8a 2176 bitmap_flush(mddev);
a9701a30 2177 md_super_wait(mddev);
1da177e4
LT		 2178 if (mddev->ro)
2179 set_disk_ro(disk, 0);
2180 blk_queue_make_request(mddev->queue, md_fail_request);
2181 mddev->pers->stop(mddev);
2182 module_put(mddev->pers->owner);
2183 mddev->pers = NULL;
2184 if (mddev->ro)
2185 mddev->ro = 0;
2186 }
2187 if (!mddev->in_sync) {
2188 /* mark array as shutdown cleanly */
2189 mddev->in_sync = 1;
2190 md_update_sb(mddev);
2191 }
2192 if (ro)
2193 set_disk_ro(disk, 1);
2194 }
32a7627c
N		 2195
2196 bitmap_destroy(mddev);
2197 if (mddev->bitmap_file) {
2198 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2199 fput(mddev->bitmap_file);
2200 mddev->bitmap_file = NULL;
2201 }
9223214e 2202 mddev->bitmap_offset = 0;
32a7627c 2203
1da177e4
LT		 2204 /*
2205 * Free resources if final stop
2206 */
2207 if (!ro) {
86e6ffdd
N		 2208 mdk_rdev_t *rdev;
2209 struct list_head *tmp;
1da177e4
LT		 2210 struct gendisk *disk;
2211 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2212
86e6ffdd
N		 2213 ITERATE_RDEV(mddev,rdev,tmp)
2214 if (rdev->raid_disk >= 0) {
2215 char nm[20];
2216 sprintf(nm, "rd%d", rdev->raid_disk);
2217 sysfs_remove_link(&mddev->kobj, nm);
2218 }
2219
1da177e4
LT		 2220 export_array(mddev);
2221
2222 mddev->array_size = 0;
2223 disk = mddev->gendisk;
2224 if (disk)
2225 set_capacity(disk, 0);
2226 mddev->changed = 1;
2227 } else
2228 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2229 mdname(mddev));
2230 err = 0;
2231out:
2232 return err;
2233}
2234
2235static void autorun_array(mddev_t *mddev)
2236{
2237 mdk_rdev_t *rdev;
2238 struct list_head *tmp;
2239 int err;
2240
a757e64c 2241 if (list_empty(&mddev->disks))
1da177e4 2242 return;
1da177e4
LT		 2243
2244 printk(KERN_INFO "md: running: ");
2245
2246 ITERATE_RDEV(mddev,rdev,tmp) {
2247 char b[BDEVNAME_SIZE];
2248 printk("<%s>", bdevname(rdev->bdev,b));
2249 }
2250 printk("\n");
2251
2252 err = do_md_run (mddev);
2253 if (err) {
2254 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2255 do_md_stop (mddev, 0);
2256 }
2257}
2258
2259/*
2260 * lets try to run arrays based on all disks that have arrived
2261 * until now. (those are in pending_raid_disks)
2262 *
2263 * the method: pick the first pending disk, collect all disks with
2264 * the same UUID, remove all from the pending list and put them into
2265 * the 'same_array' list. Then order this list based on superblock
2266 * update time (freshest comes first), kick out 'old' disks and
2267 * compare superblocks. If everything's fine then run it.
2268 *
2269 * If "unit" is allocated, then bump its reference count
2270 */
2271static void autorun_devices(int part)
2272{
2273 struct list_head candidates;
2274 struct list_head *tmp;
2275 mdk_rdev_t *rdev0, *rdev;
2276 mddev_t *mddev;
2277 char b[BDEVNAME_SIZE];
2278
2279 printk(KERN_INFO "md: autorun ...\n");
2280 while (!list_empty(&pending_raid_disks)) {
2281 dev_t dev;
2282 rdev0 = list_entry(pending_raid_disks.next,
2283 mdk_rdev_t, same_set);
2284
2285 printk(KERN_INFO "md: considering %s ...\n",
2286 bdevname(rdev0->bdev,b));
2287 INIT_LIST_HEAD(&candidates);
2288 ITERATE_RDEV_PENDING(rdev,tmp)
2289 if (super_90_load(rdev, rdev0, 0) >= 0) {
2290 printk(KERN_INFO "md: adding %s ...\n",
2291 bdevname(rdev->bdev,b));
2292 list_move(&rdev->same_set, &candidates);
2293 }
2294 /*
2295 * now we have a set of devices, with all of them having
2296 * mostly sane superblocks. It's time to allocate the
2297 * mddev.
2298 */
2299 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2300 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2301 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2302 break;
2303 }
2304 if (part)
2305 dev = MKDEV(mdp_major,
2306 rdev0->preferred_minor << MdpMinorShift);
2307 else
2308 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2309
2310 md_probe(dev, NULL, NULL);
2311 mddev = mddev_find(dev);
2312 if (!mddev) {
2313 printk(KERN_ERR
2314 "md: cannot allocate memory for md drive.\n");
2315 break;
2316 }
2317 if (mddev_lock(mddev))
2318 printk(KERN_WARNING "md: %s locked, cannot run\n",
2319 mdname(mddev));
2320 else if (mddev->raid_disks || mddev->major_version
2321 || !list_empty(&mddev->disks)) {
2322 printk(KERN_WARNING
2323 "md: %s already running, cannot run %s\n",
2324 mdname(mddev), bdevname(rdev0->bdev,b));
2325 mddev_unlock(mddev);
2326 } else {
2327 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2328 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2329 list_del_init(&rdev->same_set);
2330 if (bind_rdev_to_array(rdev, mddev))
2331 export_rdev(rdev);
2332 }
2333 autorun_array(mddev);
2334 mddev_unlock(mddev);
2335 }
2336 /* on success, candidates will be empty, on error
2337 * it won't...
2338 */
2339 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2340 export_rdev(rdev);
2341 mddev_put(mddev);
2342 }
2343 printk(KERN_INFO "md: ... autorun DONE.\n");
2344}
2345
2346/*
2347 * import RAID devices based on one partition
2348 * if possible, the array gets run as well.
2349 */
2350
2351static int autostart_array(dev_t startdev)
2352{
2353 char b[BDEVNAME_SIZE];
2354 int err = -EINVAL, i;
2355 mdp_super_t *sb = NULL;
2356 mdk_rdev_t *start_rdev = NULL, *rdev;
2357
2358 start_rdev = md_import_device(startdev, 0, 0);
2359 if (IS_ERR(start_rdev))
2360 return err;
2361
2362
2363 /* NOTE: this can only work for 0.90.0 superblocks */
2364 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2365 if (sb->major_version != 0 ||
2366 sb->minor_version != 90 ) {
2367 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2368 export_rdev(start_rdev);
2369 return err;
2370 }
2371
b2d444d7 2372 if (test_bit(Faulty, &start_rdev->flags)) {
1da177e4
LT		 2373 printk(KERN_WARNING
2374 "md: can not autostart based on faulty %s!\n",
2375 bdevname(start_rdev->bdev,b));
2376 export_rdev(start_rdev);
2377 return err;
2378 }
2379 list_add(&start_rdev->same_set, &pending_raid_disks);
2380
2381 for (i = 0; i < MD_SB_DISKS; i++) {
2382 mdp_disk_t *desc = sb->disks + i;
2383 dev_t dev = MKDEV(desc->major, desc->minor);
2384
2385 if (!dev)
2386 continue;
2387 if (dev == startdev)
2388 continue;
2389 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2390 continue;
2391 rdev = md_import_device(dev, 0, 0);
2392 if (IS_ERR(rdev))
2393 continue;
2394
2395 list_add(&rdev->same_set, &pending_raid_disks);
2396 }
2397
2398 /*
2399 * possibly return codes
2400 */
2401 autorun_devices(0);
2402 return 0;
2403
2404}
2405
2406
2407static int get_version(void __user * arg)
2408{
2409 mdu_version_t ver;
2410
2411 ver.major = MD_MAJOR_VERSION;
2412 ver.minor = MD_MINOR_VERSION;
2413 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2414
2415 if (copy_to_user(arg, &ver, sizeof(ver)))
2416 return -EFAULT;
2417
2418 return 0;
2419}
2420
2421static int get_array_info(mddev_t * mddev, void __user * arg)
2422{
2423 mdu_array_info_t info;
2424 int nr,working,active,failed,spare;
2425 mdk_rdev_t *rdev;
2426 struct list_head *tmp;
2427
2428 nr=working=active=failed=spare=0;
2429 ITERATE_RDEV(mddev,rdev,tmp) {
2430 nr++;
b2d444d7 2431 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2432 failed++;
2433 else {
2434 working++;
b2d444d7 2435 if (test_bit(In_sync, &rdev->flags))
1da177e4
LT		 2436 active++;
2437 else
2438 spare++;
2439 }
2440 }
2441
2442 info.major_version = mddev->major_version;
2443 info.minor_version = mddev->minor_version;
2444 info.patch_version = MD_PATCHLEVEL_VERSION;
2445 info.ctime = mddev->ctime;
2446 info.level = mddev->level;
2447 info.size = mddev->size;
2448 info.nr_disks = nr;
2449 info.raid_disks = mddev->raid_disks;
2450 info.md_minor = mddev->md_minor;
2451 info.not_persistent= !mddev->persistent;
2452
2453 info.utime = mddev->utime;
2454 info.state = 0;
2455 if (mddev->in_sync)
2456 info.state = (1<<MD_SB_CLEAN);
36fa3063
N		 2457 if (mddev->bitmap && mddev->bitmap_offset)
2458 info.state = (1<<MD_SB_BITMAP_PRESENT);
1da177e4
LT		 2459 info.active_disks = active;
2460 info.working_disks = working;
2461 info.failed_disks = failed;
2462 info.spare_disks = spare;
2463
2464 info.layout = mddev->layout;
2465 info.chunk_size = mddev->chunk_size;
2466
2467 if (copy_to_user(arg, &info, sizeof(info)))
2468 return -EFAULT;
2469
2470 return 0;
2471}
2472
87162a28 2473static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 2474{
2475 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2476 char *ptr, *buf = NULL;
2477 int err = -ENOMEM;
2478
2479 file = kmalloc(sizeof(*file), GFP_KERNEL);
2480 if (!file)
2481 goto out;
2482
2483 /* bitmap disabled, zero the first byte and copy out */
2484 if (!mddev->bitmap || !mddev->bitmap->file) {
2485 file->pathname[0] = '\0';
2486 goto copy_out;
2487 }
2488
2489 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2490 if (!buf)
2491 goto out;
2492
2493 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2494 if (!ptr)
2495 goto out;
2496
2497 strcpy(file->pathname, ptr);
2498
2499copy_out:
2500 err = 0;
2501 if (copy_to_user(arg, file, sizeof(*file)))
2502 err = -EFAULT;
2503out:
2504 kfree(buf);
2505 kfree(file);
2506 return err;
2507}
2508
1da177e4
LT		 2509static int get_disk_info(mddev_t * mddev, void __user * arg)
2510{
2511 mdu_disk_info_t info;
2512 unsigned int nr;
2513 mdk_rdev_t *rdev;
2514
2515 if (copy_from_user(&info, arg, sizeof(info)))
2516 return -EFAULT;
2517
2518 nr = info.number;
2519
2520 rdev = find_rdev_nr(mddev, nr);
2521 if (rdev) {
2522 info.major = MAJOR(rdev->bdev->bd_dev);
2523 info.minor = MINOR(rdev->bdev->bd_dev);
2524 info.raid_disk = rdev->raid_disk;
2525 info.state = 0;
b2d444d7 2526 if (test_bit(Faulty, &rdev->flags))
1da177e4 2527 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 2528 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 2529 info.state |= (1<<MD_DISK_ACTIVE);
2530 info.state |= (1<<MD_DISK_SYNC);
2531 }
8ddf9efe
N		 2532 if (test_bit(WriteMostly, &rdev->flags))
2533 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 2534 } else {
2535 info.major = info.minor = 0;
2536 info.raid_disk = -1;
2537 info.state = (1<<MD_DISK_REMOVED);
2538 }
2539
2540 if (copy_to_user(arg, &info, sizeof(info)))
2541 return -EFAULT;
2542
2543 return 0;
2544}
2545
2546static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2547{
2548 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2549 mdk_rdev_t *rdev;
2550 dev_t dev = MKDEV(info->major,info->minor);
2551
2552 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2553 return -EOVERFLOW;
2554
2555 if (!mddev->raid_disks) {
2556 int err;
2557 /* expecting a device which has a superblock */
2558 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2559 if (IS_ERR(rdev)) {
2560 printk(KERN_WARNING
2561 "md: md_import_device returned %ld\n",
2562 PTR_ERR(rdev));
2563 return PTR_ERR(rdev);
2564 }
2565 if (!list_empty(&mddev->disks)) {
2566 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2567 mdk_rdev_t, same_set);
2568 int err = super_types[mddev->major_version]
2569 .load_super(rdev, rdev0, mddev->minor_version);
2570 if (err < 0) {
2571 printk(KERN_WARNING
2572 "md: %s has different UUID to %s\n",
2573 bdevname(rdev->bdev,b),
2574 bdevname(rdev0->bdev,b2));
2575 export_rdev(rdev);
2576 return -EINVAL;
2577 }
2578 }
2579 err = bind_rdev_to_array(rdev, mddev);
2580 if (err)
2581 export_rdev(rdev);
2582 return err;
2583 }
2584
2585 /*
2586 * add_new_disk can be used once the array is assembled
2587 * to add "hot spares". They must already have a superblock
2588 * written
2589 */
2590 if (mddev->pers) {
2591 int err;
2592 if (!mddev->pers->hot_add_disk) {
2593 printk(KERN_WARNING
2594 "%s: personality does not support diskops!\n",
2595 mdname(mddev));
2596 return -EINVAL;
2597 }
7b1e35f6
N		 2598 if (mddev->persistent)
2599 rdev = md_import_device(dev, mddev->major_version,
2600 mddev->minor_version);
2601 else
2602 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 2603 if (IS_ERR(rdev)) {
2604 printk(KERN_WARNING
2605 "md: md_import_device returned %ld\n",
2606 PTR_ERR(rdev));
2607 return PTR_ERR(rdev);
2608 }
41158c7e
N		 2609 /* set save_raid_disk if appropriate */
2610 if (!mddev->persistent) {
2611 if (info->state & (1<<MD_DISK_SYNC) &&
2612 info->raid_disk < mddev->raid_disks)
2613 rdev->raid_disk = info->raid_disk;
2614 else
2615 rdev->raid_disk = -1;
2616 } else
2617 super_types[mddev->major_version].
2618 validate_super(mddev, rdev);
2619 rdev->saved_raid_disk = rdev->raid_disk;
2620
b2d444d7 2621 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 2622 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2623 set_bit(WriteMostly, &rdev->flags);
2624
1da177e4
LT		 2625 rdev->raid_disk = -1;
2626 err = bind_rdev_to_array(rdev, mddev);
2627 if (err)
2628 export_rdev(rdev);
c361777f
N		 2629
2630 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 2631 md_wakeup_thread(mddev->thread);
1da177e4
LT		 2632 return err;
2633 }
2634
2635 /* otherwise, add_new_disk is only allowed
2636 * for major_version==0 superblocks
2637 */
2638 if (mddev->major_version != 0) {
2639 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2640 mdname(mddev));
2641 return -EINVAL;
2642 }
2643
2644 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2645 int err;
2646 rdev = md_import_device (dev, -1, 0);
2647 if (IS_ERR(rdev)) {
2648 printk(KERN_WARNING
2649 "md: error, md_import_device() returned %ld\n",
2650 PTR_ERR(rdev));
2651 return PTR_ERR(rdev);
2652 }
2653 rdev->desc_nr = info->number;
2654 if (info->raid_disk < mddev->raid_disks)
2655 rdev->raid_disk = info->raid_disk;
2656 else
2657 rdev->raid_disk = -1;
2658
b2d444d7
N		 2659 rdev->flags = 0;
2660
1da177e4 2661 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 2662 if (info->state & (1<<MD_DISK_SYNC))
2663 set_bit(In_sync, &rdev->flags);
1da177e4 2664
8ddf9efe
N		 2665 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2666 set_bit(WriteMostly, &rdev->flags);
2667
1da177e4
LT		 2668 err = bind_rdev_to_array(rdev, mddev);
2669 if (err) {
2670 export_rdev(rdev);
2671 return err;
2672 }
2673
2674 if (!mddev->persistent) {
2675 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2676 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2677 } else
2678 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2679 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2680
2681 if (!mddev->size || (mddev->size > rdev->size))
2682 mddev->size = rdev->size;
2683 }
2684
2685 return 0;
2686}
2687
2688static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2689{
2690 char b[BDEVNAME_SIZE];
2691 mdk_rdev_t *rdev;
2692
2693 if (!mddev->pers)
2694 return -ENODEV;
2695
2696 rdev = find_rdev(mddev, dev);
2697 if (!rdev)
2698 return -ENXIO;
2699
2700 if (rdev->raid_disk >= 0)
2701 goto busy;
2702
2703 kick_rdev_from_array(rdev);
2704 md_update_sb(mddev);
2705
2706 return 0;
2707busy:
2708 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2709 bdevname(rdev->bdev,b), mdname(mddev));
2710 return -EBUSY;
2711}
2712
2713static int hot_add_disk(mddev_t * mddev, dev_t dev)
2714{
2715 char b[BDEVNAME_SIZE];
2716 int err;
2717 unsigned int size;
2718 mdk_rdev_t *rdev;
2719
2720 if (!mddev->pers)
2721 return -ENODEV;
2722
2723 if (mddev->major_version != 0) {
2724 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2725 " version-0 superblocks.\n",
2726 mdname(mddev));
2727 return -EINVAL;
2728 }
2729 if (!mddev->pers->hot_add_disk) {
2730 printk(KERN_WARNING
2731 "%s: personality does not support diskops!\n",
2732 mdname(mddev));
2733 return -EINVAL;
2734 }
2735
2736 rdev = md_import_device (dev, -1, 0);
2737 if (IS_ERR(rdev)) {
2738 printk(KERN_WARNING
2739 "md: error, md_import_device() returned %ld\n",
2740 PTR_ERR(rdev));
2741 return -EINVAL;
2742 }
2743
2744 if (mddev->persistent)
2745 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2746 else
2747 rdev->sb_offset =
2748 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2749
2750 size = calc_dev_size(rdev, mddev->chunk_size);
2751 rdev->size = size;
2752
2753 if (size < mddev->size) {
2754 printk(KERN_WARNING
2755 "%s: disk size %llu blocks < array size %llu\n",
2756 mdname(mddev), (unsigned long long)size,
2757 (unsigned long long)mddev->size);
2758 err = -ENOSPC;
2759 goto abort_export;
2760 }
2761
b2d444d7 2762 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 2763 printk(KERN_WARNING
2764 "md: can not hot-add faulty %s disk to %s!\n",
2765 bdevname(rdev->bdev,b), mdname(mddev));
2766 err = -EINVAL;
2767 goto abort_export;
2768 }
b2d444d7 2769 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2770 rdev->desc_nr = -1;
2771 bind_rdev_to_array(rdev, mddev);
2772
2773 /*
2774 * The rest should better be atomic, we can have disk failures
2775 * noticed in interrupt contexts ...
2776 */
2777
2778 if (rdev->desc_nr == mddev->max_disks) {
2779 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2780 mdname(mddev));
2781 err = -EBUSY;
2782 goto abort_unbind_export;
2783 }
2784
2785 rdev->raid_disk = -1;
2786
2787 md_update_sb(mddev);
2788
2789 /*
2790 * Kick recovery, maybe this spare has to be added to the
2791 * array immediately.
2792 */
2793 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2794 md_wakeup_thread(mddev->thread);
2795
2796 return 0;
2797
2798abort_unbind_export:
2799 unbind_rdev_from_array(rdev);
2800
2801abort_export:
2802 export_rdev(rdev);
2803 return err;
2804}
2805
32a7627c
N		 2806/* similar to deny_write_access, but accounts for our holding a reference
2807 * to the file ourselves */
2808static int deny_bitmap_write_access(struct file * file)
2809{
2810 struct inode *inode = file->f_mapping->host;
2811
2812 spin_lock(&inode->i_lock);
2813 if (atomic_read(&inode->i_writecount) > 1) {
2814 spin_unlock(&inode->i_lock);
2815 return -ETXTBSY;
2816 }
2817 atomic_set(&inode->i_writecount, -1);
2818 spin_unlock(&inode->i_lock);
2819
2820 return 0;
2821}
2822
2823static int set_bitmap_file(mddev_t *mddev, int fd)
2824{
2825 int err;
2826
36fa3063
N		 2827 if (mddev->pers) {
2828 if (!mddev->pers->quiesce)
2829 return -EBUSY;
2830 if (mddev->recovery || mddev->sync_thread)
2831 return -EBUSY;
2832 /* we should be able to change the bitmap.. */
2833 }
32a7627c 2834
32a7627c 2835
36fa3063
N		 2836 if (fd >= 0) {
2837 if (mddev->bitmap)
2838 return -EEXIST; /* cannot add when bitmap is present */
2839 mddev->bitmap_file = fget(fd);
32a7627c 2840
36fa3063
N		 2841 if (mddev->bitmap_file == NULL) {
2842 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2843 mdname(mddev));
2844 return -EBADF;
2845 }
2846
2847 err = deny_bitmap_write_access(mddev->bitmap_file);
2848 if (err) {
2849 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2850 mdname(mddev));
2851 fput(mddev->bitmap_file);
2852 mddev->bitmap_file = NULL;
2853 return err;
2854 }
a654b9d8 2855 mddev->bitmap_offset = 0; /* file overrides offset */
36fa3063
N		 2856 } else if (mddev->bitmap == NULL)
2857 return -ENOENT; /* cannot remove what isn't there */
2858 err = 0;
2859 if (mddev->pers) {
2860 mddev->pers->quiesce(mddev, 1);
2861 if (fd >= 0)
2862 err = bitmap_create(mddev);
2863 if (fd < 0 || err)
2864 bitmap_destroy(mddev);
2865 mddev->pers->quiesce(mddev, 0);
2866 } else if (fd < 0) {
2867 if (mddev->bitmap_file)
2868 fput(mddev->bitmap_file);
2869 mddev->bitmap_file = NULL;
2870 }
2871
32a7627c
N		 2872 return err;
2873}
2874
1da177e4
LT		 2875/*
2876 * set_array_info is used two different ways
2877 * The original usage is when creating a new array.
2878 * In this usage, raid_disks is > 0 and it together with
2879 * level, size, not_persistent,layout,chunksize determine the
2880 * shape of the array.
2881 * This will always create an array with a type-0.90.0 superblock.
2882 * The newer usage is when assembling an array.
2883 * In this case raid_disks will be 0, and the major_version field is
2884 * use to determine which style super-blocks are to be found on the devices.
2885 * The minor and patch _version numbers are also kept incase the
2886 * super_block handler wishes to interpret them.
2887 */
2888static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2889{
2890
2891 if (info->raid_disks == 0) {
2892 /* just setting version number for superblock loading */
2893 if (info->major_version < 0 ||
2894 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2895 super_types[info->major_version].name == NULL) {
2896 /* maybe try to auto-load a module? */
2897 printk(KERN_INFO
2898 "md: superblock version %d not known\n",
2899 info->major_version);
2900 return -EINVAL;
2901 }
2902 mddev->major_version = info->major_version;
2903 mddev->minor_version = info->minor_version;
2904 mddev->patch_version = info->patch_version;
2905 return 0;
2906 }
2907 mddev->major_version = MD_MAJOR_VERSION;
2908 mddev->minor_version = MD_MINOR_VERSION;
2909 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2910 mddev->ctime = get_seconds();
2911
2912 mddev->level = info->level;
2913 mddev->size = info->size;
2914 mddev->raid_disks = info->raid_disks;
2915 /* don't set md_minor, it is determined by which /dev/md* was
2916 * openned
2917 */
2918 if (info->state & (1<<MD_SB_CLEAN))
2919 mddev->recovery_cp = MaxSector;
2920 else
2921 mddev->recovery_cp = 0;
2922 mddev->persistent = ! info->not_persistent;
2923
2924 mddev->layout = info->layout;
2925 mddev->chunk_size = info->chunk_size;
2926
2927 mddev->max_disks = MD_SB_DISKS;
2928
2929 mddev->sb_dirty = 1;
2930
2931 /*
2932 * Generate a 128 bit UUID
2933 */
2934 get_random_bytes(mddev->uuid, 16);
2935
2936 return 0;
2937}
2938
2939/*
2940 * update_array_info is used to change the configuration of an
2941 * on-line array.
2942 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2943 * fields in the info are checked against the array.
2944 * Any differences that cannot be handled will cause an error.
2945 * Normally, only one change can be managed at a time.
2946 */
2947static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2948{
2949 int rv = 0;
2950 int cnt = 0;
36fa3063
N		 2951 int state = 0;
2952
2953 /* calculate expected state,ignoring low bits */
2954 if (mddev->bitmap && mddev->bitmap_offset)
2955 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 2956
2957 if (mddev->major_version != info->major_version ||
2958 mddev->minor_version != info->minor_version ||
2959/* mddev->patch_version != info->patch_version || */
2960 mddev->ctime != info->ctime ||
2961 mddev->level != info->level ||
2962/* mddev->layout != info->layout || */
2963 !mddev->persistent != info->not_persistent||
36fa3063
N		 2964 mddev->chunk_size != info->chunk_size ||
2965 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2966 ((state^info->state) & 0xfffffe00)
2967 )
1da177e4
LT		 2968 return -EINVAL;
2969 /* Check there is only one change */
2970 if (mddev->size != info->size) cnt++;
2971 if (mddev->raid_disks != info->raid_disks) cnt++;
2972 if (mddev->layout != info->layout) cnt++;
36fa3063 2973 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
1da177e4
LT		 2974 if (cnt == 0) return 0;
2975 if (cnt > 1) return -EINVAL;
2976
2977 if (mddev->layout != info->layout) {
2978 /* Change layout
2979 * we don't need to do anything at the md level, the
2980 * personality will take care of it all.
2981 */
2982 if (mddev->pers->reconfig == NULL)
2983 return -EINVAL;
2984 else
2985 return mddev->pers->reconfig(mddev, info->layout, -1);
2986 }
2987 if (mddev->size != info->size) {
2988 mdk_rdev_t * rdev;
2989 struct list_head *tmp;
2990 if (mddev->pers->resize == NULL)
2991 return -EINVAL;
2992 /* The "size" is the amount of each device that is used.
2993 * This can only make sense for arrays with redundancy.
2994 * linear and raid0 always use whatever space is available
2995 * We can only consider changing the size if no resync
2996 * or reconstruction is happening, and if the new size
2997 * is acceptable. It must fit before the sb_offset or,
2998 * if that is <data_offset, it must fit before the
2999 * size of each device.
3000 * If size is zero, we find the largest size that fits.
3001 */
3002 if (mddev->sync_thread)
3003 return -EBUSY;
3004 ITERATE_RDEV(mddev,rdev,tmp) {
3005 sector_t avail;
3006 int fit = (info->size == 0);
3007 if (rdev->sb_offset > rdev->data_offset)
3008 avail = (rdev->sb_offset*2) - rdev->data_offset;
3009 else
3010 avail = get_capacity(rdev->bdev->bd_disk)
3011 - rdev->data_offset;
3012 if (fit && (info->size == 0 || info->size > avail/2))
3013 info->size = avail/2;
3014 if (avail < ((sector_t)info->size << 1))
3015 return -ENOSPC;
3016 }
3017 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3018 if (!rv) {
3019 struct block_device *bdev;
3020
3021 bdev = bdget_disk(mddev->gendisk, 0);
3022 if (bdev) {
3023 down(&bdev->bd_inode->i_sem);
3024 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3025 up(&bdev->bd_inode->i_sem);
3026 bdput(bdev);
3027 }
3028 }
3029 }
3030 if (mddev->raid_disks != info->raid_disks) {
3031 /* change the number of raid disks */
3032 if (mddev->pers->reshape == NULL)
3033 return -EINVAL;
3034 if (info->raid_disks <= 0 ||
3035 info->raid_disks >= mddev->max_disks)
3036 return -EINVAL;
3037 if (mddev->sync_thread)
3038 return -EBUSY;
3039 rv = mddev->pers->reshape(mddev, info->raid_disks);
3040 if (!rv) {
3041 struct block_device *bdev;
3042
3043 bdev = bdget_disk(mddev->gendisk, 0);
3044 if (bdev) {
3045 down(&bdev->bd_inode->i_sem);
3046 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3047 up(&bdev->bd_inode->i_sem);
3048 bdput(bdev);
3049 }
3050 }
3051 }
36fa3063
N		 3052 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3053 if (mddev->pers->quiesce == NULL)
3054 return -EINVAL;
3055 if (mddev->recovery || mddev->sync_thread)
3056 return -EBUSY;
3057 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3058 /* add the bitmap */
3059 if (mddev->bitmap)
3060 return -EEXIST;
3061 if (mddev->default_bitmap_offset == 0)
3062 return -EINVAL;
3063 mddev->bitmap_offset = mddev->default_bitmap_offset;
3064 mddev->pers->quiesce(mddev, 1);
3065 rv = bitmap_create(mddev);
3066 if (rv)
3067 bitmap_destroy(mddev);
3068 mddev->pers->quiesce(mddev, 0);
3069 } else {
3070 /* remove the bitmap */
3071 if (!mddev->bitmap)
3072 return -ENOENT;
3073 if (mddev->bitmap->file)
3074 return -EINVAL;
3075 mddev->pers->quiesce(mddev, 1);
3076 bitmap_destroy(mddev);
3077 mddev->pers->quiesce(mddev, 0);
3078 mddev->bitmap_offset = 0;
3079 }
3080 }
1da177e4
LT		 3081 md_update_sb(mddev);
3082 return rv;
3083}
3084
3085static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3086{
3087 mdk_rdev_t *rdev;
3088
3089 if (mddev->pers == NULL)
3090 return -ENODEV;
3091
3092 rdev = find_rdev(mddev, dev);
3093 if (!rdev)
3094 return -ENODEV;
3095
3096 md_error(mddev, rdev);
3097 return 0;
3098}
3099
3100static int md_ioctl(struct inode *inode, struct file *file,
3101 unsigned int cmd, unsigned long arg)
3102{
3103 int err = 0;
3104 void __user *argp = (void __user *)arg;
3105 struct hd_geometry __user *loc = argp;
3106 mddev_t *mddev = NULL;
3107
3108 if (!capable(CAP_SYS_ADMIN))
3109 return -EACCES;
3110
3111 /*
3112 * Commands dealing with the RAID driver but not any
3113 * particular array:
3114 */
3115 switch (cmd)
3116 {
3117 case RAID_VERSION:
3118 err = get_version(argp);
3119 goto done;
3120
3121 case PRINT_RAID_DEBUG:
3122 err = 0;
3123 md_print_devices();
3124 goto done;
3125
3126#ifndef MODULE
3127 case RAID_AUTORUN:
3128 err = 0;
3129 autostart_arrays(arg);
3130 goto done;
3131#endif
3132 default:;
3133 }
3134
3135 /*
3136 * Commands creating/starting a new array:
3137 */
3138
3139 mddev = inode->i_bdev->bd_disk->private_data;
3140
3141 if (!mddev) {
3142 BUG();
3143 goto abort;
3144 }
3145
3146
3147 if (cmd == START_ARRAY) {
3148 /* START_ARRAY doesn't need to lock the array as autostart_array
3149 * does the locking, and it could even be a different array
3150 */
3151 static int cnt = 3;
3152 if (cnt > 0 ) {
3153 printk(KERN_WARNING
3154 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3155 "This will not be supported beyond 2.6\n",
3156 current->comm, current->pid);
3157 cnt--;
3158 }
3159 err = autostart_array(new_decode_dev(arg));
3160 if (err) {
3161 printk(KERN_WARNING "md: autostart failed!\n");
3162 goto abort;
3163 }
3164 goto done;
3165 }
3166
3167 err = mddev_lock(mddev);
3168 if (err) {
3169 printk(KERN_INFO
3170 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3171 err, cmd);
3172 goto abort;
3173 }
3174
3175 switch (cmd)
3176 {
3177 case SET_ARRAY_INFO:
3178 {
3179 mdu_array_info_t info;
3180 if (!arg)
3181 memset(&info, 0, sizeof(info));
3182 else if (copy_from_user(&info, argp, sizeof(info))) {
3183 err = -EFAULT;
3184 goto abort_unlock;
3185 }
3186 if (mddev->pers) {
3187 err = update_array_info(mddev, &info);
3188 if (err) {
3189 printk(KERN_WARNING "md: couldn't update"
3190 " array info. %d\n", err);
3191 goto abort_unlock;
3192 }
3193 goto done_unlock;
3194 }
3195 if (!list_empty(&mddev->disks)) {
3196 printk(KERN_WARNING
3197 "md: array %s already has disks!\n",
3198 mdname(mddev));
3199 err = -EBUSY;
3200 goto abort_unlock;
3201 }
3202 if (mddev->raid_disks) {
3203 printk(KERN_WARNING
3204 "md: array %s already initialised!\n",
3205 mdname(mddev));
3206 err = -EBUSY;
3207 goto abort_unlock;
3208 }
3209 err = set_array_info(mddev, &info);
3210 if (err) {
3211 printk(KERN_WARNING "md: couldn't set"
3212 " array info. %d\n", err);
3213 goto abort_unlock;
3214 }
3215 }
3216 goto done_unlock;
3217
3218 default:;
3219 }
3220
3221 /*
3222 * Commands querying/configuring an existing array:
3223 */
32a7627c
N		 3224 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3225 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3226 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3227 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
1da177e4
LT		 3228 err = -ENODEV;
3229 goto abort_unlock;
3230 }
3231
3232 /*
3233 * Commands even a read-only array can execute:
3234 */
3235 switch (cmd)
3236 {
3237 case GET_ARRAY_INFO:
3238 err = get_array_info(mddev, argp);
3239 goto done_unlock;
3240
32a7627c 3241 case GET_BITMAP_FILE:
87162a28 3242 err = get_bitmap_file(mddev, argp);
32a7627c
N		 3243 goto done_unlock;
3244
1da177e4
LT		 3245 case GET_DISK_INFO:
3246 err = get_disk_info(mddev, argp);
3247 goto done_unlock;
3248
3249 case RESTART_ARRAY_RW:
3250 err = restart_array(mddev);
3251 goto done_unlock;
3252
3253 case STOP_ARRAY:
3254 err = do_md_stop (mddev, 0);
3255 goto done_unlock;
3256
3257 case STOP_ARRAY_RO:
3258 err = do_md_stop (mddev, 1);
3259 goto done_unlock;
3260
3261 /*
3262 * We have a problem here : there is no easy way to give a CHS
3263 * virtual geometry. We currently pretend that we have a 2 heads
3264 * 4 sectors (with a BIG number of cylinders...). This drives
3265 * dosfs just mad... ;-)
3266 */
3267 case HDIO_GETGEO:
3268 if (!loc) {
3269 err = -EINVAL;
3270 goto abort_unlock;
3271 }
3272 err = put_user (2, (char __user *) &loc->heads);
3273 if (err)
3274 goto abort_unlock;
3275 err = put_user (4, (char __user *) &loc->sectors);
3276 if (err)
3277 goto abort_unlock;
3278 err = put_user(get_capacity(mddev->gendisk)/8,
3279 (short __user *) &loc->cylinders);
3280 if (err)
3281 goto abort_unlock;
3282 err = put_user (get_start_sect(inode->i_bdev),
3283 (long __user *) &loc->start);
3284 goto done_unlock;
3285 }
3286
3287 /*
3288 * The remaining ioctls are changing the state of the
3289 * superblock, so we do not allow read-only arrays
3290 * here:
3291 */
3292 if (mddev->ro) {
3293 err = -EROFS;
3294 goto abort_unlock;
3295 }
3296
3297 switch (cmd)
3298 {
3299 case ADD_NEW_DISK:
3300 {
3301 mdu_disk_info_t info;
3302 if (copy_from_user(&info, argp, sizeof(info)))
3303 err = -EFAULT;
3304 else
3305 err = add_new_disk(mddev, &info);
3306 goto done_unlock;
3307 }
3308
3309 case HOT_REMOVE_DISK:
3310 err = hot_remove_disk(mddev, new_decode_dev(arg));
3311 goto done_unlock;
3312
3313 case HOT_ADD_DISK:
3314 err = hot_add_disk(mddev, new_decode_dev(arg));
3315 goto done_unlock;
3316
3317 case SET_DISK_FAULTY:
3318 err = set_disk_faulty(mddev, new_decode_dev(arg));
3319 goto done_unlock;
3320
3321 case RUN_ARRAY:
3322 err = do_md_run (mddev);
3323 goto done_unlock;
3324
32a7627c
N		 3325 case SET_BITMAP_FILE:
3326 err = set_bitmap_file(mddev, (int)arg);
3327 goto done_unlock;
3328
1da177e4
LT		 3329 default:
3330 if (_IOC_TYPE(cmd) == MD_MAJOR)
3331 printk(KERN_WARNING "md: %s(pid %d) used"
3332 " obsolete MD ioctl, upgrade your"
3333 " software to use new ictls.\n",
3334 current->comm, current->pid);
3335 err = -EINVAL;
3336 goto abort_unlock;
3337 }
3338
3339done_unlock:
3340abort_unlock:
3341 mddev_unlock(mddev);
3342
3343 return err;
3344done:
3345 if (err)
3346 MD_BUG();
3347abort:
3348 return err;
3349}
3350
3351static int md_open(struct inode *inode, struct file *file)
3352{
3353 /*
3354 * Succeed if we can lock the mddev, which confirms that
3355 * it isn't being stopped right now.
3356 */
3357 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3358 int err;
3359
3360 if ((err = mddev_lock(mddev)))
3361 goto out;
3362
3363 err = 0;
3364 mddev_get(mddev);
3365 mddev_unlock(mddev);
3366
3367 check_disk_change(inode->i_bdev);
3368 out:
3369 return err;
3370}
3371
3372static int md_release(struct inode *inode, struct file * file)
3373{
3374 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3375
3376 if (!mddev)
3377 BUG();
3378 mddev_put(mddev);
3379
3380 return 0;
3381}
3382
3383static int md_media_changed(struct gendisk *disk)
3384{
3385 mddev_t *mddev = disk->private_data;
3386
3387 return mddev->changed;
3388}
3389
3390static int md_revalidate(struct gendisk *disk)
3391{
3392 mddev_t *mddev = disk->private_data;
3393
3394 mddev->changed = 0;
3395 return 0;
3396}
3397static struct block_device_operations md_fops =
3398{
3399 .owner = THIS_MODULE,
3400 .open = md_open,
3401 .release = md_release,
3402 .ioctl = md_ioctl,
3403 .media_changed = md_media_changed,
3404 .revalidate_disk= md_revalidate,
3405};
3406
75c96f85 3407static int md_thread(void * arg)
1da177e4
LT		 3408{
3409 mdk_thread_t *thread = arg;
3410
1da177e4
LT		 3411 /*
3412 * md_thread is a 'system-thread', it's priority should be very
3413 * high. We avoid resource deadlocks individually in each
3414 * raid personality. (RAID5 does preallocation) We also use RR and
3415 * the very same RT priority as kswapd, thus we will never get
3416 * into a priority inversion deadlock.
3417 *
3418 * we definitely have to have equal or higher priority than
3419 * bdflush, otherwise bdflush will deadlock if there are too
3420 * many dirty RAID5 blocks.
3421 */
1da177e4 3422
6985c43f 3423 allow_signal(SIGKILL);
1da177e4 3424 complete(thread->event);
a6fb0934 3425 while (!kthread_should_stop()) {
1da177e4
LT		 3426 void (*run)(mddev_t *);
3427
32a7627c 3428 wait_event_interruptible_timeout(thread->wqueue,
a6fb0934
N		 3429 test_bit(THREAD_WAKEUP, &thread->flags)
3430 || kthread_should_stop(),
32a7627c 3431 thread->timeout);
3e1d1d28 3432 try_to_freeze();
1da177e4
LT		 3433
3434 clear_bit(THREAD_WAKEUP, &thread->flags);
3435
3436 run = thread->run;
3437 if (run)
3438 run(thread->mddev);
1da177e4 3439 }
a6fb0934 3440
1da177e4
LT		 3441 return 0;
3442}
3443
3444void md_wakeup_thread(mdk_thread_t *thread)
3445{
3446 if (thread) {
3447 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3448 set_bit(THREAD_WAKEUP, &thread->flags);
3449 wake_up(&thread->wqueue);
3450 }
3451}
3452
3453mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3454 const char *name)
3455{
3456 mdk_thread_t *thread;
1da177e4
LT		 3457 struct completion event;
3458
a6fb0934 3459 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 3460 if (!thread)
3461 return NULL;
3462
3463 memset(thread, 0, sizeof(mdk_thread_t));
3464 init_waitqueue_head(&thread->wqueue);
3465
3466 init_completion(&event);
3467 thread->event = &event;
3468 thread->run = run;
3469 thread->mddev = mddev;
3470 thread->name = name;
32a7627c 3471 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6985c43f 3472 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
a6fb0934 3473 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 3474 kfree(thread);
3475 return NULL;
3476 }
3477 wait_for_completion(&event);
3478 return thread;
3479}
3480
1da177e4
LT		 3481void md_unregister_thread(mdk_thread_t *thread)
3482{
d28446fe 3483 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
a6fb0934
N		 3484
3485 kthread_stop(thread->tsk);
1da177e4
LT		 3486 kfree(thread);
3487}
3488
3489void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3490{
3491 if (!mddev) {
3492 MD_BUG();
3493 return;
3494 }
3495
b2d444d7 3496 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 3497 return;
32a7627c 3498/*
1da177e4
LT		 3499 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3500 mdname(mddev),
3501 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3502 __builtin_return_address(0),__builtin_return_address(1),
3503 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 3504*/
1da177e4
LT		 3505 if (!mddev->pers->error_handler)
3506 return;
3507 mddev->pers->error_handler(mddev,rdev);
3508 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3510 md_wakeup_thread(mddev->thread);
3511}
3512
3513/* seq_file implementation /proc/mdstat */
3514
3515static void status_unused(struct seq_file *seq)
3516{
3517 int i = 0;
3518 mdk_rdev_t *rdev;
3519 struct list_head *tmp;
3520
3521 seq_printf(seq, "unused devices: ");
3522
3523 ITERATE_RDEV_PENDING(rdev,tmp) {
3524 char b[BDEVNAME_SIZE];
3525 i++;
3526 seq_printf(seq, "%s ",
3527 bdevname(rdev->bdev,b));
3528 }
3529 if (!i)
3530 seq_printf(seq, "<none>");
3531
3532 seq_printf(seq, "\n");
3533}
3534
3535
3536static void status_resync(struct seq_file *seq, mddev_t * mddev)
3537{
3538 unsigned long max_blocks, resync, res, dt, db, rt;
3539
3540 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3541
3542 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3543 max_blocks = mddev->resync_max_sectors >> 1;
3544 else
3545 max_blocks = mddev->size;
3546
3547 /*
3548 * Should not happen.
3549 */
3550 if (!max_blocks) {
3551 MD_BUG();
3552 return;
3553 }
3554 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3555 {
3556 int i, x = res/50, y = 20-x;
3557 seq_printf(seq, "[");
3558 for (i = 0; i < x; i++)
3559 seq_printf(seq, "=");
3560 seq_printf(seq, ">");
3561 for (i = 0; i < y; i++)
3562 seq_printf(seq, ".");
3563 seq_printf(seq, "] ");
3564 }
3565 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3566 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3567 "resync" : "recovery"),
3568 res/10, res % 10, resync, max_blocks);
3569
3570 /*
3571 * We do not want to overflow, so the order of operands and
3572 * the * 100 / 100 trick are important. We do a +1 to be
3573 * safe against division by zero. We only estimate anyway.
3574 *
3575 * dt: time from mark until now
3576 * db: blocks written from mark until now
3577 * rt: remaining time
3578 */
3579 dt = ((jiffies - mddev->resync_mark) / HZ);
3580 if (!dt) dt++;
3581 db = resync - (mddev->resync_mark_cnt/2);
3582 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3583
3584 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3585
3586 seq_printf(seq, " speed=%ldK/sec", db/dt);
3587}
3588
3589static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3590{
3591 struct list_head *tmp;
3592 loff_t l = *pos;
3593 mddev_t *mddev;
3594
3595 if (l >= 0x10000)
3596 return NULL;
3597 if (!l--)
3598 /* header */
3599 return (void*)1;
3600
3601 spin_lock(&all_mddevs_lock);
3602 list_for_each(tmp,&all_mddevs)
3603 if (!l--) {
3604 mddev = list_entry(tmp, mddev_t, all_mddevs);
3605 mddev_get(mddev);
3606 spin_unlock(&all_mddevs_lock);
3607 return mddev;
3608 }
3609 spin_unlock(&all_mddevs_lock);
3610 if (!l--)
3611 return (void*)2;/* tail */
3612 return NULL;
3613}
3614
3615static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3616{
3617 struct list_head *tmp;
3618 mddev_t *next_mddev, *mddev = v;
3619
3620 ++*pos;
3621 if (v == (void*)2)
3622 return NULL;
3623
3624 spin_lock(&all_mddevs_lock);
3625 if (v == (void*)1)
3626 tmp = all_mddevs.next;
3627 else
3628 tmp = mddev->all_mddevs.next;
3629 if (tmp != &all_mddevs)
3630 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3631 else {
3632 next_mddev = (void*)2;
3633 *pos = 0x10000;
3634 }
3635 spin_unlock(&all_mddevs_lock);
3636
3637 if (v != (void*)1)
3638 mddev_put(mddev);
3639 return next_mddev;
3640
3641}
3642
3643static void md_seq_stop(struct seq_file *seq, void *v)
3644{
3645 mddev_t *mddev = v;
3646
3647 if (mddev && v != (void*)1 && v != (void*)2)
3648 mddev_put(mddev);
3649}
3650
3651static int md_seq_show(struct seq_file *seq, void *v)
3652{
3653 mddev_t *mddev = v;
3654 sector_t size;
3655 struct list_head *tmp2;
3656 mdk_rdev_t *rdev;
3657 int i;
32a7627c 3658 struct bitmap *bitmap;
1da177e4
LT		 3659
3660 if (v == (void*)1) {
3661 seq_printf(seq, "Personalities : ");
3662 spin_lock(&pers_lock);
3663 for (i = 0; i < MAX_PERSONALITY; i++)
3664 if (pers[i])
3665 seq_printf(seq, "[%s] ", pers[i]->name);
3666
3667 spin_unlock(&pers_lock);
3668 seq_printf(seq, "\n");
3669 return 0;
3670 }
3671 if (v == (void*)2) {
3672 status_unused(seq);
3673 return 0;
3674 }
3675
3676 if (mddev_lock(mddev)!=0)
3677 return -EINTR;
3678 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3679 seq_printf(seq, "%s : %sactive", mdname(mddev),
3680 mddev->pers ? "" : "in");
3681 if (mddev->pers) {
3682 if (mddev->ro)
3683 seq_printf(seq, " (read-only)");
3684 seq_printf(seq, " %s", mddev->pers->name);
3685 }
3686
3687 size = 0;
3688 ITERATE_RDEV(mddev,rdev,tmp2) {
3689 char b[BDEVNAME_SIZE];
3690 seq_printf(seq, " %s[%d]",
3691 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 3692 if (test_bit(WriteMostly, &rdev->flags))
3693 seq_printf(seq, "(W)");
b2d444d7 3694 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 3695 seq_printf(seq, "(F)");
3696 continue;
b325a32e
N		 3697 } else if (rdev->raid_disk < 0)
3698 seq_printf(seq, "(S)"); /* spare */
1da177e4
LT		 3699 size += rdev->size;
3700 }
3701
3702 if (!list_empty(&mddev->disks)) {
3703 if (mddev->pers)
3704 seq_printf(seq, "\n %llu blocks",
3705 (unsigned long long)mddev->array_size);
3706 else
3707 seq_printf(seq, "\n %llu blocks",
3708 (unsigned long long)size);
3709 }
1cd6bf19
N		 3710 if (mddev->persistent) {
3711 if (mddev->major_version != 0 ||
3712 mddev->minor_version != 90) {
3713 seq_printf(seq," super %d.%d",
3714 mddev->major_version,
3715 mddev->minor_version);
3716 }
3717 } else
3718 seq_printf(seq, " super non-persistent");
1da177e4
LT		 3719
3720 if (mddev->pers) {
3721 mddev->pers->status (seq, mddev);
3722 seq_printf(seq, "\n ");
32a7627c 3723 if (mddev->curr_resync > 2) {
1da177e4 3724 status_resync (seq, mddev);
32a7627c
N		 3725 seq_printf(seq, "\n ");
3726 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3727 seq_printf(seq, " resync=DELAYED\n ");
3728 } else
3729 seq_printf(seq, "\n ");
3730
3731 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 3732 unsigned long chunk_kb;
3733 unsigned long flags;
32a7627c
N		 3734 spin_lock_irqsave(&bitmap->lock, flags);
3735 chunk_kb = bitmap->chunksize >> 10;
3736 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3737 "%lu%s chunk",
3738 bitmap->pages - bitmap->missing_pages,
3739 bitmap->pages,
3740 (bitmap->pages - bitmap->missing_pages)
3741 << (PAGE_SHIFT - 10),
3742 chunk_kb ? chunk_kb : bitmap->chunksize,
3743 chunk_kb ? "KB" : "B");
78d742d8
N		 3744 if (bitmap->file) {
3745 seq_printf(seq, ", file: ");
3746 seq_path(seq, bitmap->file->f_vfsmnt,
3747 bitmap->file->f_dentry," \t\n");
32a7627c 3748 }
78d742d8 3749
32a7627c
N		 3750 seq_printf(seq, "\n");
3751 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 3752 }
3753
3754 seq_printf(seq, "\n");
3755 }
3756 mddev_unlock(mddev);
3757
3758 return 0;
3759}
3760
3761static struct seq_operations md_seq_ops = {
3762 .start = md_seq_start,
3763 .next = md_seq_next,
3764 .stop = md_seq_stop,
3765 .show = md_seq_show,
3766};
3767
3768static int md_seq_open(struct inode *inode, struct file *file)
3769{
3770 int error;
3771
3772 error = seq_open(file, &md_seq_ops);
3773 return error;
3774}
3775
3776static struct file_operations md_seq_fops = {
3777 .open = md_seq_open,
3778 .read = seq_read,
3779 .llseek = seq_lseek,
3780 .release = seq_release,
3781};
3782
3783int register_md_personality(int pnum, mdk_personality_t *p)
3784{
3785 if (pnum >= MAX_PERSONALITY) {
3786 printk(KERN_ERR
3787 "md: tried to install personality %s as nr %d, but max is %lu\n",
3788 p->name, pnum, MAX_PERSONALITY-1);
3789 return -EINVAL;
3790 }
3791
3792 spin_lock(&pers_lock);
3793 if (pers[pnum]) {
3794 spin_unlock(&pers_lock);
1da177e4
LT		 3795 return -EBUSY;
3796 }
3797
3798 pers[pnum] = p;
3799 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3800 spin_unlock(&pers_lock);
3801 return 0;
3802}
3803
3804int unregister_md_personality(int pnum)
3805{
a757e64c 3806 if (pnum >= MAX_PERSONALITY)
1da177e4 3807 return -EINVAL;
1da177e4
LT		 3808
3809 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3810 spin_lock(&pers_lock);
3811 pers[pnum] = NULL;
3812 spin_unlock(&pers_lock);
3813 return 0;
3814}
3815
3816static int is_mddev_idle(mddev_t *mddev)
3817{
3818 mdk_rdev_t * rdev;
3819 struct list_head *tmp;
3820 int idle;
3821 unsigned long curr_events;
3822
3823 idle = 1;
3824 ITERATE_RDEV(mddev,rdev,tmp) {
3825 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
a362357b
JA		 3826 curr_events = disk_stat_read(disk, sectors[0]) +
3827 disk_stat_read(disk, sectors[1]) -
1da177e4
LT		 3828 atomic_read(&disk->sync_io);
3829 /* Allow some slack between valud of curr_events and last_events,
3830 * as there are some uninteresting races.
3831 * Note: the following is an unsigned comparison.
3832 */
3833 if ((curr_events - rdev->last_events + 32) > 64) {
3834 rdev->last_events = curr_events;
3835 idle = 0;
3836 }
3837 }
3838 return idle;
3839}
3840
3841void md_done_sync(mddev_t *mddev, int blocks, int ok)
3842{
3843 /* another "blocks" (512byte) blocks have been synced */
3844 atomic_sub(blocks, &mddev->recovery_active);
3845 wake_up(&mddev->recovery_wait);
3846 if (!ok) {
3847 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3848 md_wakeup_thread(mddev->thread);
3849 // stop recovery, signal do_sync ....
3850 }
3851}
3852
3853
06d91a5f
N		 3854/* md_write_start(mddev, bi)
3855 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 3856 * in superblock) before writing, schedule a superblock update
3857 * and wait for it to complete.
06d91a5f 3858 */
3d310eb7 3859void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 3860{
06d91a5f 3861 if (bio_data_dir(bi) != WRITE)
3d310eb7 3862 return;
06d91a5f
N		 3863
3864 atomic_inc(&mddev->writes_pending);
06d91a5f 3865 if (mddev->in_sync) {
a9701a30 3866 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 3867 if (mddev->in_sync) {
3868 mddev->in_sync = 0;
3869 mddev->sb_dirty = 1;
3870 md_wakeup_thread(mddev->thread);
3871 }
a9701a30 3872 spin_unlock_irq(&mddev->write_lock);
06d91a5f 3873 }
3d310eb7 3874 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
1da177e4
LT		 3875}
3876
3877void md_write_end(mddev_t *mddev)
3878{
3879 if (atomic_dec_and_test(&mddev->writes_pending)) {
3880 if (mddev->safemode == 2)
3881 md_wakeup_thread(mddev->thread);
3882 else
3883 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3884 }
3885}
3886
75c96f85 3887static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
1da177e4
LT		 3888
3889#define SYNC_MARKS 10
3890#define SYNC_MARK_STEP (3*HZ)
3891static void md_do_sync(mddev_t *mddev)
3892{
3893 mddev_t *mddev2;
3894 unsigned int currspeed = 0,
3895 window;
57afd89f 3896 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 3897 unsigned long mark[SYNC_MARKS];
3898 sector_t mark_cnt[SYNC_MARKS];
3899 int last_mark,m;
3900 struct list_head *tmp;
3901 sector_t last_check;
57afd89f 3902 int skipped = 0;
1da177e4
LT		 3903
3904 /* just incase thread restarts... */
3905 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3906 return;
3907
3908 /* we overload curr_resync somewhat here.
3909 * 0 == not engaged in resync at all
3910 * 2 == checking that there is no conflict with another sync
3911 * 1 == like 2, but have yielded to allow conflicting resync to
3912 * commense
3913 * other == active in resync - this many blocks
3914 *
3915 * Before starting a resync we must have set curr_resync to
3916 * 2, and then checked that every "conflicting" array has curr_resync
3917 * less than ours. When we find one that is the same or higher
3918 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3919 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3920 * This will mean we have to start checking from the beginning again.
3921 *
3922 */
3923
3924 do {
3925 mddev->curr_resync = 2;
3926
3927 try_again:
8712e553
N		 3928 if (signal_pending(current) ||
3929 kthread_should_stop()) {
1da177e4 3930 flush_signals(current);
6985c43f 3931 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 3932 goto skip;
3933 }
3934 ITERATE_MDDEV(mddev2,tmp) {
1da177e4
LT		 3935 if (mddev2 == mddev)
3936 continue;
3937 if (mddev2->curr_resync &&
3938 match_mddev_units(mddev,mddev2)) {
3939 DEFINE_WAIT(wq);
3940 if (mddev < mddev2 && mddev->curr_resync == 2) {
3941 /* arbitrarily yield */
3942 mddev->curr_resync = 1;
3943 wake_up(&resync_wait);
3944 }
3945 if (mddev > mddev2 && mddev->curr_resync == 1)
3946 /* no need to wait here, we can wait the next
3947 * time 'round when curr_resync == 2
3948 */
3949 continue;
3950 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8712e553
N		 3951 if (!signal_pending(current) &&
3952 !kthread_should_stop() &&
3953 mddev2->curr_resync >= mddev->curr_resync) {
1da177e4
LT		 3954 printk(KERN_INFO "md: delaying resync of %s"
3955 " until %s has finished resync (they"
3956 " share one or more physical units)\n",
3957 mdname(mddev), mdname(mddev2));
3958 mddev_put(mddev2);
3959 schedule();
3960 finish_wait(&resync_wait, &wq);
3961 goto try_again;
3962 }
3963 finish_wait(&resync_wait, &wq);
3964 }
3965 }
3966 } while (mddev->curr_resync < 2);
3967
9d88883e 3968 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 3969 /* resync follows the size requested by the personality,
57afd89f 3970 * which defaults to physical size, but can be virtual size
1da177e4
LT		 3971 */
3972 max_sectors = mddev->resync_max_sectors;
9d88883e
N		 3973 mddev->resync_mismatches = 0;
3974 } else
1da177e4
LT		 3975 /* recovery follows the physical size of devices */
3976 max_sectors = mddev->size << 1;
3977
3978 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3979 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3980 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
338cec32 3981 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
1da177e4
LT		 3982 "(but not more than %d KB/sec) for reconstruction.\n",
3983 sysctl_speed_limit_max);
3984
3985 is_mddev_idle(mddev); /* this also initializes IO event counters */
32a7627c 3986 /* we don't use the checkpoint if there's a bitmap */
24dd469d
N		 3987 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3988 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1da177e4
LT		 3989 j = mddev->recovery_cp;
3990 else
3991 j = 0;
57afd89f 3992 io_sectors = 0;
1da177e4
LT		 3993 for (m = 0; m < SYNC_MARKS; m++) {
3994 mark[m] = jiffies;
57afd89f 3995 mark_cnt[m] = io_sectors;
1da177e4
LT		 3996 }
3997 last_mark = 0;
3998 mddev->resync_mark = mark[last_mark];
3999 mddev->resync_mark_cnt = mark_cnt[last_mark];
4000
4001 /*
4002 * Tune reconstruction:
4003 */
4004 window = 32*(PAGE_SIZE/512);
4005 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4006 window/2,(unsigned long long) max_sectors/2);
4007
4008 atomic_set(&mddev->recovery_active, 0);
4009 init_waitqueue_head(&mddev->recovery_wait);
4010 last_check = 0;
4011
4012 if (j>2) {
4013 printk(KERN_INFO
4014 "md: resuming recovery of %s from checkpoint.\n",
4015 mdname(mddev));
4016 mddev->curr_resync = j;
4017 }
4018
4019 while (j < max_sectors) {
57afd89f 4020 sector_t sectors;
1da177e4 4021
57afd89f
N		 4022 skipped = 0;
4023 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4024 currspeed < sysctl_speed_limit_min);
4025 if (sectors == 0) {
1da177e4
LT		 4026 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4027 goto out;
4028 }
57afd89f
N		 4029
4030 if (!skipped) { /* actual IO requested */
4031 io_sectors += sectors;
4032 atomic_add(sectors, &mddev->recovery_active);
4033 }
4034
1da177e4
LT		 4035 j += sectors;
4036 if (j>1) mddev->curr_resync = j;
4037
57afd89f
N		 4038
4039 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 4040 continue;
4041
57afd89f 4042 last_check = io_sectors;
1da177e4
LT		 4043
4044 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4045 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4046 break;
4047
4048 repeat:
4049 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4050 /* step marks */
4051 int next = (last_mark+1) % SYNC_MARKS;
4052
4053 mddev->resync_mark = mark[next];
4054 mddev->resync_mark_cnt = mark_cnt[next];
4055 mark[next] = jiffies;
57afd89f 4056 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 4057 last_mark = next;
4058 }
4059
4060
8712e553 4061 if (signal_pending(current) || kthread_should_stop()) {
1da177e4
LT		 4062 /*
4063 * got a signal, exit.
4064 */
4065 printk(KERN_INFO
4066 "md: md_do_sync() got signal ... exiting\n");
4067 flush_signals(current);
4068 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4069 goto out;
4070 }
4071
4072 /*
4073 * this loop exits only if either when we are slower than
4074 * the 'hard' speed limit, or the system was IO-idle for
4075 * a jiffy.
4076 * the system might be non-idle CPU-wise, but we only care
4077 * about not overloading the IO subsystem. (things like an
4078 * e2fsck being done on the RAID array should execute fast)
4079 */
4080 mddev->queue->unplug_fn(mddev->queue);
4081 cond_resched();
4082
57afd89f
N		 4083 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4084 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4
LT		 4085
4086 if (currspeed > sysctl_speed_limit_min) {
4087 if ((currspeed > sysctl_speed_limit_max) ||
4088 !is_mddev_idle(mddev)) {
4089 msleep_interruptible(250);
4090 goto repeat;
4091 }
4092 }
4093 }
4094 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4095 /*
4096 * this also signals 'finished resyncing' to md_stop
4097 */
4098 out:
4099 mddev->queue->unplug_fn(mddev->queue);
4100
4101 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4102
4103 /* tell personality that we are finished */
57afd89f 4104 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4
LT		 4105
4106 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4107 mddev->curr_resync > 2 &&
4108 mddev->curr_resync >= mddev->recovery_cp) {
4109 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4110 printk(KERN_INFO
4111 "md: checkpointing recovery of %s.\n",
4112 mdname(mddev));
4113 mddev->recovery_cp = mddev->curr_resync;
4114 } else
4115 mddev->recovery_cp = MaxSector;
4116 }
4117
1da177e4
LT		 4118 skip:
4119 mddev->curr_resync = 0;
4120 wake_up(&resync_wait);
4121 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4122 md_wakeup_thread(mddev->thread);
4123}
4124
4125
4126/*
4127 * This routine is regularly called by all per-raid-array threads to
4128 * deal with generic issues like resync and super-block update.
4129 * Raid personalities that don't have a thread (linear/raid0) do not
4130 * need this as they never do any recovery or update the superblock.
4131 *
4132 * It does not do any resync itself, but rather "forks" off other threads
4133 * to do that as needed.
4134 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4135 * "->recovery" and create a thread at ->sync_thread.
4136 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4137 * and wakeups up this thread which will reap the thread and finish up.
4138 * This thread also removes any faulty devices (with nr_pending == 0).
4139 *
4140 * The overall approach is:
4141 * 1/ if the superblock needs updating, update it.
4142 * 2/ If a recovery thread is running, don't do anything else.
4143 * 3/ If recovery has finished, clean up, possibly marking spares active.
4144 * 4/ If there are any faulty devices, remove them.
4145 * 5/ If array is degraded, try to add spares devices
4146 * 6/ If array has spares or is not in-sync, start a resync thread.
4147 */
4148void md_check_recovery(mddev_t *mddev)
4149{
4150 mdk_rdev_t *rdev;
4151 struct list_head *rtmp;
4152
4153
5f40402d
N		 4154 if (mddev->bitmap)
4155 bitmap_daemon_work(mddev->bitmap);
1da177e4
LT		 4156
4157 if (mddev->ro)
4158 return;
fca4d848
N		 4159
4160 if (signal_pending(current)) {
4161 if (mddev->pers->sync_request) {
4162 printk(KERN_INFO "md: %s in immediate safe mode\n",
4163 mdname(mddev));
4164 mddev->safemode = 2;
4165 }
4166 flush_signals(current);
4167 }
4168
1da177e4
LT		 4169 if ( ! (
4170 mddev->sb_dirty ||
4171 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848
N		 4172 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4173 (mddev->safemode == 1) ||
4174 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4175 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 4176 ))
4177 return;
fca4d848 4178
1da177e4
LT		 4179 if (mddev_trylock(mddev)==0) {
4180 int spares =0;
fca4d848 4181
a9701a30 4182 spin_lock_irq(&mddev->write_lock);
fca4d848
N		 4183 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4184 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4185 mddev->in_sync = 1;
4186 mddev->sb_dirty = 1;
4187 }
4188 if (mddev->safemode == 1)
4189 mddev->safemode = 0;
a9701a30 4190 spin_unlock_irq(&mddev->write_lock);
fca4d848 4191
1da177e4
LT		 4192 if (mddev->sb_dirty)
4193 md_update_sb(mddev);
06d91a5f 4194
06d91a5f 4195
1da177e4
LT		 4196 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4197 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4198 /* resync/recovery still happening */
4199 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4200 goto unlock;
4201 }
4202 if (mddev->sync_thread) {
4203 /* resync has finished, collect result */
4204 md_unregister_thread(mddev->sync_thread);
4205 mddev->sync_thread = NULL;
4206 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4207 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4208 /* success...*/
4209 /* activate any spares */
4210 mddev->pers->spare_active(mddev);
4211 }
4212 md_update_sb(mddev);
41158c7e
N		 4213
4214 /* if array is no-longer degraded, then any saved_raid_disk
4215 * information must be scrapped
4216 */
4217 if (!mddev->degraded)
4218 ITERATE_RDEV(mddev,rdev,rtmp)
4219 rdev->saved_raid_disk = -1;
4220
1da177e4
LT		 4221 mddev->recovery = 0;
4222 /* flag recovery needed just to double check */
4223 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4224 goto unlock;
4225 }
24dd469d
N		 4226 /* Clear some bits that don't mean anything, but
4227 * might be left set
4228 */
4229 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4230 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4231 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4232 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4
LT		 4233
4234 /* no recovery is running.
4235 * remove any failed drives, then
4236 * add spares if possible.
4237 * Spare are also removed and re-added, to allow
4238 * the personality to fail the re-add.
4239 */
4240 ITERATE_RDEV(mddev,rdev,rtmp)
4241 if (rdev->raid_disk >= 0 &&
b2d444d7 4242 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
1da177e4 4243 atomic_read(&rdev->nr_pending)==0) {
86e6ffdd
N		 4244 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4245 char nm[20];
4246 sprintf(nm,"rd%d", rdev->raid_disk);
4247 sysfs_remove_link(&mddev->kobj, nm);
1da177e4 4248 rdev->raid_disk = -1;
86e6ffdd 4249 }
1da177e4
LT		 4250 }
4251
4252 if (mddev->degraded) {
4253 ITERATE_RDEV(mddev,rdev,rtmp)
4254 if (rdev->raid_disk < 0
b2d444d7 4255 && !test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 4256 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4257 char nm[20];
4258 sprintf(nm, "rd%d", rdev->raid_disk);
4259 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
1da177e4 4260 spares++;
86e6ffdd 4261 } else
1da177e4
LT		 4262 break;
4263 }
4264 }
4265
24dd469d
N		 4266 if (spares) {
4267 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4268 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4269 } else if (mddev->recovery_cp < MaxSector) {
4270 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4271 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4272 /* nothing to be done ... */
1da177e4 4273 goto unlock;
24dd469d 4274
1da177e4
LT		 4275 if (mddev->pers->sync_request) {
4276 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
a654b9d8
N		 4277 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4278 /* We are adding a device or devices to an array
4279 * which has the bitmap stored on all devices.
4280 * So make sure all bitmap pages get written
4281 */
4282 bitmap_write_all(mddev->bitmap);
4283 }
1da177e4
LT		 4284 mddev->sync_thread = md_register_thread(md_do_sync,
4285 mddev,
4286 "%s_resync");
4287 if (!mddev->sync_thread) {
4288 printk(KERN_ERR "%s: could not start resync"
4289 " thread...\n",
4290 mdname(mddev));
4291 /* leave the spares where they are, it shouldn't hurt */
4292 mddev->recovery = 0;
4293 } else {
4294 md_wakeup_thread(mddev->sync_thread);
4295 }
4296 }
4297 unlock:
4298 mddev_unlock(mddev);
4299 }
4300}
4301
75c96f85
AB		 4302static int md_notify_reboot(struct notifier_block *this,
4303 unsigned long code, void *x)
1da177e4
LT		 4304{
4305 struct list_head *tmp;
4306 mddev_t *mddev;
4307
4308 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4309
4310 printk(KERN_INFO "md: stopping all md devices.\n");
4311
4312 ITERATE_MDDEV(mddev,tmp)
4313 if (mddev_trylock(mddev)==0)
4314 do_md_stop (mddev, 1);
4315 /*
4316 * certain more exotic SCSI devices are known to be
4317 * volatile wrt too early system reboots. While the
4318 * right place to handle this issue is the given
4319 * driver, we do want to have a safe RAID driver ...
4320 */
4321 mdelay(1000*1);
4322 }
4323 return NOTIFY_DONE;
4324}
4325
75c96f85 4326static struct notifier_block md_notifier = {
1da177e4
LT		 4327 .notifier_call = md_notify_reboot,
4328 .next = NULL,
4329 .priority = INT_MAX, /* before any real devices */
4330};
4331
4332static void md_geninit(void)
4333{
4334 struct proc_dir_entry *p;
4335
4336 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4337
4338 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4339 if (p)
4340 p->proc_fops = &md_seq_fops;
4341}
4342
75c96f85 4343static int __init md_init(void)
1da177e4
LT		 4344{
4345 int minor;
4346
4347 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4348 " MD_SB_DISKS=%d\n",
4349 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4350 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
bd926c63 4351 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
32a7627c 4352 BITMAP_MINOR);
1da177e4
LT		 4353
4354 if (register_blkdev(MAJOR_NR, "md"))
4355 return -1;
4356 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4357 unregister_blkdev(MAJOR_NR, "md");
4358 return -1;
4359 }
4360 devfs_mk_dir("md");
4361 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4362 md_probe, NULL, NULL);
4363 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4364 md_probe, NULL, NULL);
4365
4366 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4367 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4368 S_IFBLK|S_IRUSR|S_IWUSR,
4369 "md/%d", minor);
4370
4371 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4372 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4373 S_IFBLK|S_IRUSR|S_IWUSR,
4374 "md/mdp%d", minor);
4375
4376
4377 register_reboot_notifier(&md_notifier);
4378 raid_table_header = register_sysctl_table(raid_root_table, 1);
4379
4380 md_geninit();
4381 return (0);
4382}
4383
4384
4385#ifndef MODULE
4386
4387/*
4388 * Searches all registered partitions for autorun RAID arrays
4389 * at boot time.
4390 */
4391static dev_t detected_devices[128];
4392static int dev_cnt;
4393
4394void md_autodetect_dev(dev_t dev)
4395{
4396 if (dev_cnt >= 0 && dev_cnt < 127)
4397 detected_devices[dev_cnt++] = dev;
4398}
4399
4400
4401static void autostart_arrays(int part)
4402{
4403 mdk_rdev_t *rdev;
4404 int i;
4405
4406 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4407
4408 for (i = 0; i < dev_cnt; i++) {
4409 dev_t dev = detected_devices[i];
4410
4411 rdev = md_import_device(dev,0, 0);
4412 if (IS_ERR(rdev))
4413 continue;
4414
b2d444d7 4415 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 4416 MD_BUG();
4417 continue;
4418 }
4419 list_add(&rdev->same_set, &pending_raid_disks);
4420 }
4421 dev_cnt = 0;
4422
4423 autorun_devices(part);
4424}
4425
4426#endif
4427
4428static __exit void md_exit(void)
4429{
4430 mddev_t *mddev;
4431 struct list_head *tmp;
4432 int i;
4433 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4434 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4435 for (i=0; i < MAX_MD_DEVS; i++)
4436 devfs_remove("md/%d", i);
4437 for (i=0; i < MAX_MD_DEVS; i++)
4438 devfs_remove("md/d%d", i);
4439
4440 devfs_remove("md");
4441
4442 unregister_blkdev(MAJOR_NR,"md");
4443 unregister_blkdev(mdp_major, "mdp");
4444 unregister_reboot_notifier(&md_notifier);
4445 unregister_sysctl_table(raid_table_header);
4446 remove_proc_entry("mdstat", NULL);
4447 ITERATE_MDDEV(mddev,tmp) {
4448 struct gendisk *disk = mddev->gendisk;
4449 if (!disk)
4450 continue;
4451 export_array(mddev);
4452 del_gendisk(disk);
4453 put_disk(disk);
4454 mddev->gendisk = NULL;
4455 mddev_put(mddev);
4456 }
4457}
4458
4459module_init(md_init)
4460module_exit(md_exit)
4461
4462EXPORT_SYMBOL(register_md_personality);
4463EXPORT_SYMBOL(unregister_md_personality);
4464EXPORT_SYMBOL(md_error);
4465EXPORT_SYMBOL(md_done_sync);
4466EXPORT_SYMBOL(md_write_start);
4467EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 4468EXPORT_SYMBOL(md_register_thread);
4469EXPORT_SYMBOL(md_unregister_thread);
4470EXPORT_SYMBOL(md_wakeup_thread);
4471EXPORT_SYMBOL(md_print_devices);
4472EXPORT_SYMBOL(md_check_recovery);
4473MODULE_LICENSE("GPL");
aa1595e9 4474MODULE_ALIAS("md");
72008652 4475MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)