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