blob: c402f6cc7047f6308747448e7142dc15b60e88c2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
NeilBrown32a76272005-06-21 17:17:14 -070022 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 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>
NeilBrown32a76272005-06-21 17:17:14 -070039#include <linux/raid/bitmap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#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
NeilBrown32a76272005-06-21 17:17:14 -070047#include <linux/file.h>
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#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);
NeilBrown06d91a52005-06-21 17:17:12 -0700227 bio_list_init(&new->write_list);
228 spin_lock_init(&new->write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
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
331static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
332{
333 if (bio->bi_size)
334 return 1;
335
336 complete((struct completion*)bio->bi_private);
337 return 0;
338}
339
340static int sync_page_io(struct block_device *bdev, sector_t sector, int size,
341 struct page *page, int rw)
342{
Neil Brownbaaa2c52005-04-16 15:23:54 -0700343 struct bio *bio = bio_alloc(GFP_NOIO, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 struct completion event;
345 int ret;
346
347 rw |= (1 << BIO_RW_SYNC);
348
349 bio->bi_bdev = bdev;
350 bio->bi_sector = sector;
351 bio_add_page(bio, page, size, 0);
352 init_completion(&event);
353 bio->bi_private = &event;
354 bio->bi_end_io = bi_complete;
355 submit_bio(rw, bio);
356 wait_for_completion(&event);
357
358 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
359 bio_put(bio);
360 return ret;
361}
362
363static int read_disk_sb(mdk_rdev_t * rdev)
364{
365 char b[BDEVNAME_SIZE];
366 if (!rdev->sb_page) {
367 MD_BUG();
368 return -EINVAL;
369 }
370 if (rdev->sb_loaded)
371 return 0;
372
373
374 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
375 goto fail;
376 rdev->sb_loaded = 1;
377 return 0;
378
379fail:
380 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
381 bdevname(rdev->bdev,b));
382 return -EINVAL;
383}
384
385static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
386{
387 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
388 (sb1->set_uuid1 == sb2->set_uuid1) &&
389 (sb1->set_uuid2 == sb2->set_uuid2) &&
390 (sb1->set_uuid3 == sb2->set_uuid3))
391
392 return 1;
393
394 return 0;
395}
396
397
398static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
399{
400 int ret;
401 mdp_super_t *tmp1, *tmp2;
402
403 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
404 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
405
406 if (!tmp1 || !tmp2) {
407 ret = 0;
408 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
409 goto abort;
410 }
411
412 *tmp1 = *sb1;
413 *tmp2 = *sb2;
414
415 /*
416 * nr_disks is not constant
417 */
418 tmp1->nr_disks = 0;
419 tmp2->nr_disks = 0;
420
421 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
422 ret = 0;
423 else
424 ret = 1;
425
426abort:
427 if (tmp1)
428 kfree(tmp1);
429 if (tmp2)
430 kfree(tmp2);
431
432 return ret;
433}
434
435static unsigned int calc_sb_csum(mdp_super_t * sb)
436{
437 unsigned int disk_csum, csum;
438
439 disk_csum = sb->sb_csum;
440 sb->sb_csum = 0;
441 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
442 sb->sb_csum = disk_csum;
443 return csum;
444}
445
446
447/*
448 * Handle superblock details.
449 * We want to be able to handle multiple superblock formats
450 * so we have a common interface to them all, and an array of
451 * different handlers.
452 * We rely on user-space to write the initial superblock, and support
453 * reading and updating of superblocks.
454 * Interface methods are:
455 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
456 * loads and validates a superblock on dev.
457 * if refdev != NULL, compare superblocks on both devices
458 * Return:
459 * 0 - dev has a superblock that is compatible with refdev
460 * 1 - dev has a superblock that is compatible and newer than refdev
461 * so dev should be used as the refdev in future
462 * -EINVAL superblock incompatible or invalid
463 * -othererror e.g. -EIO
464 *
465 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
466 * Verify that dev is acceptable into mddev.
467 * The first time, mddev->raid_disks will be 0, and data from
468 * dev should be merged in. Subsequent calls check that dev
469 * is new enough. Return 0 or -EINVAL
470 *
471 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
472 * Update the superblock for rdev with data in mddev
473 * This does not write to disc.
474 *
475 */
476
477struct super_type {
478 char *name;
479 struct module *owner;
480 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
481 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
482 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
483};
484
485/*
486 * load_super for 0.90.0
487 */
488static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
489{
490 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
491 mdp_super_t *sb;
492 int ret;
493 sector_t sb_offset;
494
495 /*
496 * Calculate the position of the superblock,
497 * it's at the end of the disk.
498 *
499 * It also happens to be a multiple of 4Kb.
500 */
501 sb_offset = calc_dev_sboffset(rdev->bdev);
502 rdev->sb_offset = sb_offset;
503
504 ret = read_disk_sb(rdev);
505 if (ret) return ret;
506
507 ret = -EINVAL;
508
509 bdevname(rdev->bdev, b);
510 sb = (mdp_super_t*)page_address(rdev->sb_page);
511
512 if (sb->md_magic != MD_SB_MAGIC) {
513 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
514 b);
515 goto abort;
516 }
517
518 if (sb->major_version != 0 ||
519 sb->minor_version != 90) {
520 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
521 sb->major_version, sb->minor_version,
522 b);
523 goto abort;
524 }
525
526 if (sb->raid_disks <= 0)
527 goto abort;
528
529 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
530 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
531 b);
532 goto abort;
533 }
534
535 rdev->preferred_minor = sb->md_minor;
536 rdev->data_offset = 0;
537
538 if (sb->level == LEVEL_MULTIPATH)
539 rdev->desc_nr = -1;
540 else
541 rdev->desc_nr = sb->this_disk.number;
542
543 if (refdev == 0)
544 ret = 1;
545 else {
546 __u64 ev1, ev2;
547 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
548 if (!uuid_equal(refsb, sb)) {
549 printk(KERN_WARNING "md: %s has different UUID to %s\n",
550 b, bdevname(refdev->bdev,b2));
551 goto abort;
552 }
553 if (!sb_equal(refsb, sb)) {
554 printk(KERN_WARNING "md: %s has same UUID"
555 " but different superblock to %s\n",
556 b, bdevname(refdev->bdev, b2));
557 goto abort;
558 }
559 ev1 = md_event(sb);
560 ev2 = md_event(refsb);
561 if (ev1 > ev2)
562 ret = 1;
563 else
564 ret = 0;
565 }
566 rdev->size = calc_dev_size(rdev, sb->chunk_size);
567
568 abort:
569 return ret;
570}
571
572/*
573 * validate_super for 0.90.0
574 */
575static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
576{
577 mdp_disk_t *desc;
578 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
579
580 if (mddev->raid_disks == 0) {
581 mddev->major_version = 0;
582 mddev->minor_version = sb->minor_version;
583 mddev->patch_version = sb->patch_version;
584 mddev->persistent = ! sb->not_persistent;
585 mddev->chunk_size = sb->chunk_size;
586 mddev->ctime = sb->ctime;
587 mddev->utime = sb->utime;
588 mddev->level = sb->level;
589 mddev->layout = sb->layout;
590 mddev->raid_disks = sb->raid_disks;
591 mddev->size = sb->size;
592 mddev->events = md_event(sb);
593
594 if (sb->state & (1<<MD_SB_CLEAN))
595 mddev->recovery_cp = MaxSector;
596 else {
597 if (sb->events_hi == sb->cp_events_hi &&
598 sb->events_lo == sb->cp_events_lo) {
599 mddev->recovery_cp = sb->recovery_cp;
600 } else
601 mddev->recovery_cp = 0;
602 }
603
604 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
605 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
606 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
607 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
608
609 mddev->max_disks = MD_SB_DISKS;
610 } else {
611 __u64 ev1;
612 ev1 = md_event(sb);
613 ++ev1;
614 if (ev1 < mddev->events)
615 return -EINVAL;
616 }
617 if (mddev->level != LEVEL_MULTIPATH) {
618 rdev->raid_disk = -1;
619 rdev->in_sync = rdev->faulty = 0;
620 desc = sb->disks + rdev->desc_nr;
621
622 if (desc->state & (1<<MD_DISK_FAULTY))
623 rdev->faulty = 1;
624 else if (desc->state & (1<<MD_DISK_SYNC) &&
625 desc->raid_disk < mddev->raid_disks) {
626 rdev->in_sync = 1;
627 rdev->raid_disk = desc->raid_disk;
628 }
629 }
630 return 0;
631}
632
633/*
634 * sync_super for 0.90.0
635 */
636static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
637{
638 mdp_super_t *sb;
639 struct list_head *tmp;
640 mdk_rdev_t *rdev2;
641 int next_spare = mddev->raid_disks;
642
643 /* make rdev->sb match mddev data..
644 *
645 * 1/ zero out disks
646 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
647 * 3/ any empty disks < next_spare become removed
648 *
649 * disks[0] gets initialised to REMOVED because
650 * we cannot be sure from other fields if it has
651 * been initialised or not.
652 */
653 int i;
654 int active=0, working=0,failed=0,spare=0,nr_disks=0;
655
656 sb = (mdp_super_t*)page_address(rdev->sb_page);
657
658 memset(sb, 0, sizeof(*sb));
659
660 sb->md_magic = MD_SB_MAGIC;
661 sb->major_version = mddev->major_version;
662 sb->minor_version = mddev->minor_version;
663 sb->patch_version = mddev->patch_version;
664 sb->gvalid_words = 0; /* ignored */
665 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
666 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
667 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
668 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
669
670 sb->ctime = mddev->ctime;
671 sb->level = mddev->level;
672 sb->size = mddev->size;
673 sb->raid_disks = mddev->raid_disks;
674 sb->md_minor = mddev->md_minor;
675 sb->not_persistent = !mddev->persistent;
676 sb->utime = mddev->utime;
677 sb->state = 0;
678 sb->events_hi = (mddev->events>>32);
679 sb->events_lo = (u32)mddev->events;
680
681 if (mddev->in_sync)
682 {
683 sb->recovery_cp = mddev->recovery_cp;
684 sb->cp_events_hi = (mddev->events>>32);
685 sb->cp_events_lo = (u32)mddev->events;
686 if (mddev->recovery_cp == MaxSector)
687 sb->state = (1<< MD_SB_CLEAN);
688 } else
689 sb->recovery_cp = 0;
690
691 sb->layout = mddev->layout;
692 sb->chunk_size = mddev->chunk_size;
693
694 sb->disks[0].state = (1<<MD_DISK_REMOVED);
695 ITERATE_RDEV(mddev,rdev2,tmp) {
696 mdp_disk_t *d;
697 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
698 rdev2->desc_nr = rdev2->raid_disk;
699 else
700 rdev2->desc_nr = next_spare++;
701 d = &sb->disks[rdev2->desc_nr];
702 nr_disks++;
703 d->number = rdev2->desc_nr;
704 d->major = MAJOR(rdev2->bdev->bd_dev);
705 d->minor = MINOR(rdev2->bdev->bd_dev);
706 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
707 d->raid_disk = rdev2->raid_disk;
708 else
709 d->raid_disk = rdev2->desc_nr; /* compatibility */
710 if (rdev2->faulty) {
711 d->state = (1<<MD_DISK_FAULTY);
712 failed++;
713 } else if (rdev2->in_sync) {
714 d->state = (1<<MD_DISK_ACTIVE);
715 d->state |= (1<<MD_DISK_SYNC);
716 active++;
717 working++;
718 } else {
719 d->state = 0;
720 spare++;
721 working++;
722 }
723 }
724
725 /* now set the "removed" and "faulty" bits on any missing devices */
726 for (i=0 ; i < mddev->raid_disks ; i++) {
727 mdp_disk_t *d = &sb->disks[i];
728 if (d->state == 0 && d->number == 0) {
729 d->number = i;
730 d->raid_disk = i;
731 d->state = (1<<MD_DISK_REMOVED);
732 d->state |= (1<<MD_DISK_FAULTY);
733 failed++;
734 }
735 }
736 sb->nr_disks = nr_disks;
737 sb->active_disks = active;
738 sb->working_disks = working;
739 sb->failed_disks = failed;
740 sb->spare_disks = spare;
741
742 sb->this_disk = sb->disks[rdev->desc_nr];
743 sb->sb_csum = calc_sb_csum(sb);
744}
745
746/*
747 * version 1 superblock
748 */
749
750static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
751{
752 unsigned int disk_csum, csum;
753 unsigned long long newcsum;
754 int size = 256 + le32_to_cpu(sb->max_dev)*2;
755 unsigned int *isuper = (unsigned int*)sb;
756 int i;
757
758 disk_csum = sb->sb_csum;
759 sb->sb_csum = 0;
760 newcsum = 0;
761 for (i=0; size>=4; size -= 4 )
762 newcsum += le32_to_cpu(*isuper++);
763
764 if (size == 2)
765 newcsum += le16_to_cpu(*(unsigned short*) isuper);
766
767 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
768 sb->sb_csum = disk_csum;
769 return cpu_to_le32(csum);
770}
771
772static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
773{
774 struct mdp_superblock_1 *sb;
775 int ret;
776 sector_t sb_offset;
777 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
778
779 /*
780 * Calculate the position of the superblock.
781 * It is always aligned to a 4K boundary and
782 * depeding on minor_version, it can be:
783 * 0: At least 8K, but less than 12K, from end of device
784 * 1: At start of device
785 * 2: 4K from start of device.
786 */
787 switch(minor_version) {
788 case 0:
789 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
790 sb_offset -= 8*2;
791 sb_offset &= ~(4*2-1);
792 /* convert from sectors to K */
793 sb_offset /= 2;
794 break;
795 case 1:
796 sb_offset = 0;
797 break;
798 case 2:
799 sb_offset = 4;
800 break;
801 default:
802 return -EINVAL;
803 }
804 rdev->sb_offset = sb_offset;
805
806 ret = read_disk_sb(rdev);
807 if (ret) return ret;
808
809
810 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
811
812 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
813 sb->major_version != cpu_to_le32(1) ||
814 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
815 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
816 sb->feature_map != 0)
817 return -EINVAL;
818
819 if (calc_sb_1_csum(sb) != sb->sb_csum) {
820 printk("md: invalid superblock checksum on %s\n",
821 bdevname(rdev->bdev,b));
822 return -EINVAL;
823 }
824 if (le64_to_cpu(sb->data_size) < 10) {
825 printk("md: data_size too small on %s\n",
826 bdevname(rdev->bdev,b));
827 return -EINVAL;
828 }
829 rdev->preferred_minor = 0xffff;
830 rdev->data_offset = le64_to_cpu(sb->data_offset);
831
832 if (refdev == 0)
833 return 1;
834 else {
835 __u64 ev1, ev2;
836 struct mdp_superblock_1 *refsb =
837 (struct mdp_superblock_1*)page_address(refdev->sb_page);
838
839 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
840 sb->level != refsb->level ||
841 sb->layout != refsb->layout ||
842 sb->chunksize != refsb->chunksize) {
843 printk(KERN_WARNING "md: %s has strangely different"
844 " superblock to %s\n",
845 bdevname(rdev->bdev,b),
846 bdevname(refdev->bdev,b2));
847 return -EINVAL;
848 }
849 ev1 = le64_to_cpu(sb->events);
850 ev2 = le64_to_cpu(refsb->events);
851
852 if (ev1 > ev2)
853 return 1;
854 }
855 if (minor_version)
856 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
857 else
858 rdev->size = rdev->sb_offset;
859 if (rdev->size < le64_to_cpu(sb->data_size)/2)
860 return -EINVAL;
861 rdev->size = le64_to_cpu(sb->data_size)/2;
862 if (le32_to_cpu(sb->chunksize))
863 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
864 return 0;
865}
866
867static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
868{
869 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
870
871 if (mddev->raid_disks == 0) {
872 mddev->major_version = 1;
873 mddev->patch_version = 0;
874 mddev->persistent = 1;
875 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
876 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
877 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
878 mddev->level = le32_to_cpu(sb->level);
879 mddev->layout = le32_to_cpu(sb->layout);
880 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
881 mddev->size = le64_to_cpu(sb->size)/2;
882 mddev->events = le64_to_cpu(sb->events);
883
884 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
885 memcpy(mddev->uuid, sb->set_uuid, 16);
886
887 mddev->max_disks = (4096-256)/2;
888 } else {
889 __u64 ev1;
890 ev1 = le64_to_cpu(sb->events);
891 ++ev1;
892 if (ev1 < mddev->events)
893 return -EINVAL;
894 }
895
896 if (mddev->level != LEVEL_MULTIPATH) {
897 int role;
898 rdev->desc_nr = le32_to_cpu(sb->dev_number);
899 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
900 switch(role) {
901 case 0xffff: /* spare */
902 rdev->in_sync = 0;
903 rdev->faulty = 0;
904 rdev->raid_disk = -1;
905 break;
906 case 0xfffe: /* faulty */
907 rdev->in_sync = 0;
908 rdev->faulty = 1;
909 rdev->raid_disk = -1;
910 break;
911 default:
912 rdev->in_sync = 1;
913 rdev->faulty = 0;
914 rdev->raid_disk = role;
915 break;
916 }
917 }
918 return 0;
919}
920
921static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
922{
923 struct mdp_superblock_1 *sb;
924 struct list_head *tmp;
925 mdk_rdev_t *rdev2;
926 int max_dev, i;
927 /* make rdev->sb match mddev and rdev data. */
928
929 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
930
931 sb->feature_map = 0;
932 sb->pad0 = 0;
933 memset(sb->pad1, 0, sizeof(sb->pad1));
934 memset(sb->pad2, 0, sizeof(sb->pad2));
935 memset(sb->pad3, 0, sizeof(sb->pad3));
936
937 sb->utime = cpu_to_le64((__u64)mddev->utime);
938 sb->events = cpu_to_le64(mddev->events);
939 if (mddev->in_sync)
940 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
941 else
942 sb->resync_offset = cpu_to_le64(0);
943
944 max_dev = 0;
945 ITERATE_RDEV(mddev,rdev2,tmp)
946 if (rdev2->desc_nr+1 > max_dev)
947 max_dev = rdev2->desc_nr+1;
948
949 sb->max_dev = cpu_to_le32(max_dev);
950 for (i=0; i<max_dev;i++)
951 sb->dev_roles[i] = cpu_to_le16(0xfffe);
952
953 ITERATE_RDEV(mddev,rdev2,tmp) {
954 i = rdev2->desc_nr;
955 if (rdev2->faulty)
956 sb->dev_roles[i] = cpu_to_le16(0xfffe);
957 else if (rdev2->in_sync)
958 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
959 else
960 sb->dev_roles[i] = cpu_to_le16(0xffff);
961 }
962
963 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
964 sb->sb_csum = calc_sb_1_csum(sb);
965}
966
967
Adrian Bunk75c96f82005-05-05 16:16:09 -0700968static struct super_type super_types[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 [0] = {
970 .name = "0.90.0",
971 .owner = THIS_MODULE,
972 .load_super = super_90_load,
973 .validate_super = super_90_validate,
974 .sync_super = super_90_sync,
975 },
976 [1] = {
977 .name = "md-1",
978 .owner = THIS_MODULE,
979 .load_super = super_1_load,
980 .validate_super = super_1_validate,
981 .sync_super = super_1_sync,
982 },
983};
984
985static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
986{
987 struct list_head *tmp;
988 mdk_rdev_t *rdev;
989
990 ITERATE_RDEV(mddev,rdev,tmp)
991 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
992 return rdev;
993
994 return NULL;
995}
996
997static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
998{
999 struct list_head *tmp;
1000 mdk_rdev_t *rdev;
1001
1002 ITERATE_RDEV(mddev1,rdev,tmp)
1003 if (match_dev_unit(mddev2, rdev))
1004 return 1;
1005
1006 return 0;
1007}
1008
1009static LIST_HEAD(pending_raid_disks);
1010
1011static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1012{
1013 mdk_rdev_t *same_pdev;
1014 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1015
1016 if (rdev->mddev) {
1017 MD_BUG();
1018 return -EINVAL;
1019 }
1020 same_pdev = match_dev_unit(mddev, rdev);
1021 if (same_pdev)
1022 printk(KERN_WARNING
1023 "%s: WARNING: %s appears to be on the same physical"
1024 " disk as %s. True\n protection against single-disk"
1025 " failure might be compromised.\n",
1026 mdname(mddev), bdevname(rdev->bdev,b),
1027 bdevname(same_pdev->bdev,b2));
1028
1029 /* Verify rdev->desc_nr is unique.
1030 * If it is -1, assign a free number, else
1031 * check number is not in use
1032 */
1033 if (rdev->desc_nr < 0) {
1034 int choice = 0;
1035 if (mddev->pers) choice = mddev->raid_disks;
1036 while (find_rdev_nr(mddev, choice))
1037 choice++;
1038 rdev->desc_nr = choice;
1039 } else {
1040 if (find_rdev_nr(mddev, rdev->desc_nr))
1041 return -EBUSY;
1042 }
1043
1044 list_add(&rdev->same_set, &mddev->disks);
1045 rdev->mddev = mddev;
1046 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1047 return 0;
1048}
1049
1050static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1051{
1052 char b[BDEVNAME_SIZE];
1053 if (!rdev->mddev) {
1054 MD_BUG();
1055 return;
1056 }
1057 list_del_init(&rdev->same_set);
1058 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1059 rdev->mddev = NULL;
1060}
1061
1062/*
1063 * prevent the device from being mounted, repartitioned or
1064 * otherwise reused by a RAID array (or any other kernel
1065 * subsystem), by bd_claiming the device.
1066 */
1067static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1068{
1069 int err = 0;
1070 struct block_device *bdev;
1071 char b[BDEVNAME_SIZE];
1072
1073 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1074 if (IS_ERR(bdev)) {
1075 printk(KERN_ERR "md: could not open %s.\n",
1076 __bdevname(dev, b));
1077 return PTR_ERR(bdev);
1078 }
1079 err = bd_claim(bdev, rdev);
1080 if (err) {
1081 printk(KERN_ERR "md: could not bd_claim %s.\n",
1082 bdevname(bdev, b));
1083 blkdev_put(bdev);
1084 return err;
1085 }
1086 rdev->bdev = bdev;
1087 return err;
1088}
1089
1090static void unlock_rdev(mdk_rdev_t *rdev)
1091{
1092 struct block_device *bdev = rdev->bdev;
1093 rdev->bdev = NULL;
1094 if (!bdev)
1095 MD_BUG();
1096 bd_release(bdev);
1097 blkdev_put(bdev);
1098}
1099
1100void md_autodetect_dev(dev_t dev);
1101
1102static void export_rdev(mdk_rdev_t * rdev)
1103{
1104 char b[BDEVNAME_SIZE];
1105 printk(KERN_INFO "md: export_rdev(%s)\n",
1106 bdevname(rdev->bdev,b));
1107 if (rdev->mddev)
1108 MD_BUG();
1109 free_disk_sb(rdev);
1110 list_del_init(&rdev->same_set);
1111#ifndef MODULE
1112 md_autodetect_dev(rdev->bdev->bd_dev);
1113#endif
1114 unlock_rdev(rdev);
1115 kfree(rdev);
1116}
1117
1118static void kick_rdev_from_array(mdk_rdev_t * rdev)
1119{
1120 unbind_rdev_from_array(rdev);
1121 export_rdev(rdev);
1122}
1123
1124static void export_array(mddev_t *mddev)
1125{
1126 struct list_head *tmp;
1127 mdk_rdev_t *rdev;
1128
1129 ITERATE_RDEV(mddev,rdev,tmp) {
1130 if (!rdev->mddev) {
1131 MD_BUG();
1132 continue;
1133 }
1134 kick_rdev_from_array(rdev);
1135 }
1136 if (!list_empty(&mddev->disks))
1137 MD_BUG();
1138 mddev->raid_disks = 0;
1139 mddev->major_version = 0;
1140}
1141
1142static void print_desc(mdp_disk_t *desc)
1143{
1144 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1145 desc->major,desc->minor,desc->raid_disk,desc->state);
1146}
1147
1148static void print_sb(mdp_super_t *sb)
1149{
1150 int i;
1151
1152 printk(KERN_INFO
1153 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1154 sb->major_version, sb->minor_version, sb->patch_version,
1155 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1156 sb->ctime);
1157 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1158 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1159 sb->md_minor, sb->layout, sb->chunk_size);
1160 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1161 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1162 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1163 sb->failed_disks, sb->spare_disks,
1164 sb->sb_csum, (unsigned long)sb->events_lo);
1165
1166 printk(KERN_INFO);
1167 for (i = 0; i < MD_SB_DISKS; i++) {
1168 mdp_disk_t *desc;
1169
1170 desc = sb->disks + i;
1171 if (desc->number || desc->major || desc->minor ||
1172 desc->raid_disk || (desc->state && (desc->state != 4))) {
1173 printk(" D %2d: ", i);
1174 print_desc(desc);
1175 }
1176 }
1177 printk(KERN_INFO "md: THIS: ");
1178 print_desc(&sb->this_disk);
1179
1180}
1181
1182static void print_rdev(mdk_rdev_t *rdev)
1183{
1184 char b[BDEVNAME_SIZE];
1185 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1186 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1187 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1188 if (rdev->sb_loaded) {
1189 printk(KERN_INFO "md: rdev superblock:\n");
1190 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1191 } else
1192 printk(KERN_INFO "md: no rdev superblock!\n");
1193}
1194
1195void md_print_devices(void)
1196{
1197 struct list_head *tmp, *tmp2;
1198 mdk_rdev_t *rdev;
1199 mddev_t *mddev;
1200 char b[BDEVNAME_SIZE];
1201
1202 printk("\n");
1203 printk("md: **********************************\n");
1204 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1205 printk("md: **********************************\n");
1206 ITERATE_MDDEV(mddev,tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
NeilBrown32a76272005-06-21 17:17:14 -07001208 if (mddev->bitmap)
1209 bitmap_print_sb(mddev->bitmap);
1210 else
1211 printk("%s: ", mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 ITERATE_RDEV(mddev,rdev,tmp2)
1213 printk("<%s>", bdevname(rdev->bdev,b));
1214 printk("\n");
1215
1216 ITERATE_RDEV(mddev,rdev,tmp2)
1217 print_rdev(rdev);
1218 }
1219 printk("md: **********************************\n");
1220 printk("\n");
1221}
1222
1223
1224static int write_disk_sb(mdk_rdev_t * rdev)
1225{
1226 char b[BDEVNAME_SIZE];
1227 if (!rdev->sb_loaded) {
1228 MD_BUG();
1229 return 1;
1230 }
1231 if (rdev->faulty) {
1232 MD_BUG();
1233 return 1;
1234 }
1235
1236 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1237 bdevname(rdev->bdev,b),
1238 (unsigned long long)rdev->sb_offset);
1239
1240 if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE))
1241 return 0;
1242
1243 printk("md: write_disk_sb failed for device %s\n",
1244 bdevname(rdev->bdev,b));
1245 return 1;
1246}
1247
1248static void sync_sbs(mddev_t * mddev)
1249{
1250 mdk_rdev_t *rdev;
1251 struct list_head *tmp;
1252
1253 ITERATE_RDEV(mddev,rdev,tmp) {
1254 super_types[mddev->major_version].
1255 sync_super(mddev, rdev);
1256 rdev->sb_loaded = 1;
1257 }
1258}
1259
1260static void md_update_sb(mddev_t * mddev)
1261{
1262 int err, count = 100;
1263 struct list_head *tmp;
1264 mdk_rdev_t *rdev;
NeilBrown06d91a52005-06-21 17:17:12 -07001265 int sync_req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267repeat:
NeilBrown06d91a52005-06-21 17:17:12 -07001268 spin_lock(&mddev->write_lock);
1269 sync_req = mddev->in_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 mddev->utime = get_seconds();
1271 mddev->events ++;
1272
1273 if (!mddev->events) {
1274 /*
1275 * oops, this 64-bit counter should never wrap.
1276 * Either we are in around ~1 trillion A.C., assuming
1277 * 1 reboot per second, or we have a bug:
1278 */
1279 MD_BUG();
1280 mddev->events --;
1281 }
1282 sync_sbs(mddev);
1283
1284 /*
1285 * do not write anything to disk if using
1286 * nonpersistent superblocks
1287 */
NeilBrown06d91a52005-06-21 17:17:12 -07001288 if (!mddev->persistent) {
1289 mddev->sb_dirty = 0;
1290 spin_unlock(&mddev->write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 return;
NeilBrown06d91a52005-06-21 17:17:12 -07001292 }
1293 spin_unlock(&mddev->write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
1295 dprintk(KERN_INFO
1296 "md: updating %s RAID superblock on device (in sync %d)\n",
1297 mdname(mddev),mddev->in_sync);
1298
NeilBrown32a76272005-06-21 17:17:14 -07001299 err = bitmap_update_sb(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 ITERATE_RDEV(mddev,rdev,tmp) {
1301 char b[BDEVNAME_SIZE];
1302 dprintk(KERN_INFO "md: ");
1303 if (rdev->faulty)
1304 dprintk("(skipping faulty ");
1305
1306 dprintk("%s ", bdevname(rdev->bdev,b));
1307 if (!rdev->faulty) {
1308 err += write_disk_sb(rdev);
1309 } else
1310 dprintk(")\n");
1311 if (!err && mddev->level == LEVEL_MULTIPATH)
1312 /* only need to write one superblock... */
1313 break;
1314 }
1315 if (err) {
1316 if (--count) {
1317 printk(KERN_ERR "md: errors occurred during superblock"
1318 " update, repeating\n");
1319 goto repeat;
1320 }
1321 printk(KERN_ERR \
1322 "md: excessive errors occurred during superblock update, exiting\n");
1323 }
NeilBrown06d91a52005-06-21 17:17:12 -07001324 spin_lock(&mddev->write_lock);
1325 if (mddev->in_sync != sync_req) {
1326 /* have to write it out again */
1327 spin_unlock(&mddev->write_lock);
1328 goto repeat;
1329 }
1330 mddev->sb_dirty = 0;
1331 spin_unlock(&mddev->write_lock);
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333}
1334
1335/*
1336 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1337 *
1338 * mark the device faulty if:
1339 *
1340 * - the device is nonexistent (zero size)
1341 * - the device has no valid superblock
1342 *
1343 * a faulty rdev _never_ has rdev->sb set.
1344 */
1345static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1346{
1347 char b[BDEVNAME_SIZE];
1348 int err;
1349 mdk_rdev_t *rdev;
1350 sector_t size;
1351
1352 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1353 if (!rdev) {
1354 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1355 return ERR_PTR(-ENOMEM);
1356 }
1357 memset(rdev, 0, sizeof(*rdev));
1358
1359 if ((err = alloc_disk_sb(rdev)))
1360 goto abort_free;
1361
1362 err = lock_rdev(rdev, newdev);
1363 if (err)
1364 goto abort_free;
1365
1366 rdev->desc_nr = -1;
1367 rdev->faulty = 0;
1368 rdev->in_sync = 0;
1369 rdev->data_offset = 0;
1370 atomic_set(&rdev->nr_pending, 0);
1371
1372 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1373 if (!size) {
1374 printk(KERN_WARNING
1375 "md: %s has zero or unknown size, marking faulty!\n",
1376 bdevname(rdev->bdev,b));
1377 err = -EINVAL;
1378 goto abort_free;
1379 }
1380
1381 if (super_format >= 0) {
1382 err = super_types[super_format].
1383 load_super(rdev, NULL, super_minor);
1384 if (err == -EINVAL) {
1385 printk(KERN_WARNING
1386 "md: %s has invalid sb, not importing!\n",
1387 bdevname(rdev->bdev,b));
1388 goto abort_free;
1389 }
1390 if (err < 0) {
1391 printk(KERN_WARNING
1392 "md: could not read %s's sb, not importing!\n",
1393 bdevname(rdev->bdev,b));
1394 goto abort_free;
1395 }
1396 }
1397 INIT_LIST_HEAD(&rdev->same_set);
1398
1399 return rdev;
1400
1401abort_free:
1402 if (rdev->sb_page) {
1403 if (rdev->bdev)
1404 unlock_rdev(rdev);
1405 free_disk_sb(rdev);
1406 }
1407 kfree(rdev);
1408 return ERR_PTR(err);
1409}
1410
1411/*
1412 * Check a full RAID array for plausibility
1413 */
1414
1415
NeilBrowna757e642005-04-16 15:26:42 -07001416static void analyze_sbs(mddev_t * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
1418 int i;
1419 struct list_head *tmp;
1420 mdk_rdev_t *rdev, *freshest;
1421 char b[BDEVNAME_SIZE];
1422
1423 freshest = NULL;
1424 ITERATE_RDEV(mddev,rdev,tmp)
1425 switch (super_types[mddev->major_version].
1426 load_super(rdev, freshest, mddev->minor_version)) {
1427 case 1:
1428 freshest = rdev;
1429 break;
1430 case 0:
1431 break;
1432 default:
1433 printk( KERN_ERR \
1434 "md: fatal superblock inconsistency in %s"
1435 " -- removing from array\n",
1436 bdevname(rdev->bdev,b));
1437 kick_rdev_from_array(rdev);
1438 }
1439
1440
1441 super_types[mddev->major_version].
1442 validate_super(mddev, freshest);
1443
1444 i = 0;
1445 ITERATE_RDEV(mddev,rdev,tmp) {
1446 if (rdev != freshest)
1447 if (super_types[mddev->major_version].
1448 validate_super(mddev, rdev)) {
1449 printk(KERN_WARNING "md: kicking non-fresh %s"
1450 " from array!\n",
1451 bdevname(rdev->bdev,b));
1452 kick_rdev_from_array(rdev);
1453 continue;
1454 }
1455 if (mddev->level == LEVEL_MULTIPATH) {
1456 rdev->desc_nr = i++;
1457 rdev->raid_disk = rdev->desc_nr;
1458 rdev->in_sync = 1;
1459 }
1460 }
1461
1462
1463
1464 if (mddev->recovery_cp != MaxSector &&
1465 mddev->level >= 1)
1466 printk(KERN_ERR "md: %s: raid array is not clean"
1467 " -- starting background reconstruction\n",
1468 mdname(mddev));
1469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
1472int mdp_major = 0;
1473
1474static struct kobject *md_probe(dev_t dev, int *part, void *data)
1475{
1476 static DECLARE_MUTEX(disks_sem);
1477 mddev_t *mddev = mddev_find(dev);
1478 struct gendisk *disk;
1479 int partitioned = (MAJOR(dev) != MD_MAJOR);
1480 int shift = partitioned ? MdpMinorShift : 0;
1481 int unit = MINOR(dev) >> shift;
1482
1483 if (!mddev)
1484 return NULL;
1485
1486 down(&disks_sem);
1487 if (mddev->gendisk) {
1488 up(&disks_sem);
1489 mddev_put(mddev);
1490 return NULL;
1491 }
1492 disk = alloc_disk(1 << shift);
1493 if (!disk) {
1494 up(&disks_sem);
1495 mddev_put(mddev);
1496 return NULL;
1497 }
1498 disk->major = MAJOR(dev);
1499 disk->first_minor = unit << shift;
1500 if (partitioned) {
1501 sprintf(disk->disk_name, "md_d%d", unit);
1502 sprintf(disk->devfs_name, "md/d%d", unit);
1503 } else {
1504 sprintf(disk->disk_name, "md%d", unit);
1505 sprintf(disk->devfs_name, "md/%d", unit);
1506 }
1507 disk->fops = &md_fops;
1508 disk->private_data = mddev;
1509 disk->queue = mddev->queue;
1510 add_disk(disk);
1511 mddev->gendisk = disk;
1512 up(&disks_sem);
1513 return NULL;
1514}
1515
1516void md_wakeup_thread(mdk_thread_t *thread);
1517
1518static void md_safemode_timeout(unsigned long data)
1519{
1520 mddev_t *mddev = (mddev_t *) data;
1521
1522 mddev->safemode = 1;
1523 md_wakeup_thread(mddev->thread);
1524}
1525
1526
1527static int do_md_run(mddev_t * mddev)
1528{
1529 int pnum, err;
1530 int chunk_size;
1531 struct list_head *tmp;
1532 mdk_rdev_t *rdev;
1533 struct gendisk *disk;
1534 char b[BDEVNAME_SIZE];
1535
NeilBrowna757e642005-04-16 15:26:42 -07001536 if (list_empty(&mddev->disks))
1537 /* cannot run an array with no devices.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
1540 if (mddev->pers)
1541 return -EBUSY;
1542
1543 /*
1544 * Analyze all RAID superblock(s)
1545 */
NeilBrowna757e642005-04-16 15:26:42 -07001546 if (!mddev->raid_disks)
1547 analyze_sbs(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 chunk_size = mddev->chunk_size;
1550 pnum = level_to_pers(mddev->level);
1551
1552 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1553 if (!chunk_size) {
1554 /*
1555 * 'default chunksize' in the old md code used to
1556 * be PAGE_SIZE, baaad.
1557 * we abort here to be on the safe side. We don't
1558 * want to continue the bad practice.
1559 */
1560 printk(KERN_ERR
1561 "no chunksize specified, see 'man raidtab'\n");
1562 return -EINVAL;
1563 }
1564 if (chunk_size > MAX_CHUNK_SIZE) {
1565 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1566 chunk_size, MAX_CHUNK_SIZE);
1567 return -EINVAL;
1568 }
1569 /*
1570 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1571 */
1572 if ( (1 << ffz(~chunk_size)) != chunk_size) {
NeilBrowna757e642005-04-16 15:26:42 -07001573 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 return -EINVAL;
1575 }
1576 if (chunk_size < PAGE_SIZE) {
1577 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1578 chunk_size, PAGE_SIZE);
1579 return -EINVAL;
1580 }
1581
1582 /* devices must have minimum size of one chunk */
1583 ITERATE_RDEV(mddev,rdev,tmp) {
1584 if (rdev->faulty)
1585 continue;
1586 if (rdev->size < chunk_size / 1024) {
1587 printk(KERN_WARNING
1588 "md: Dev %s smaller than chunk_size:"
1589 " %lluk < %dk\n",
1590 bdevname(rdev->bdev,b),
1591 (unsigned long long)rdev->size,
1592 chunk_size / 1024);
1593 return -EINVAL;
1594 }
1595 }
1596 }
1597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598#ifdef CONFIG_KMOD
1599 if (!pers[pnum])
1600 {
1601 request_module("md-personality-%d", pnum);
1602 }
1603#endif
1604
1605 /*
1606 * Drop all container device buffers, from now on
1607 * the only valid external interface is through the md
1608 * device.
1609 * Also find largest hardsector size
1610 */
1611 ITERATE_RDEV(mddev,rdev,tmp) {
1612 if (rdev->faulty)
1613 continue;
1614 sync_blockdev(rdev->bdev);
1615 invalidate_bdev(rdev->bdev, 0);
1616 }
1617
1618 md_probe(mddev->unit, NULL, NULL);
1619 disk = mddev->gendisk;
1620 if (!disk)
1621 return -ENOMEM;
1622
1623 spin_lock(&pers_lock);
1624 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1625 spin_unlock(&pers_lock);
1626 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1627 pnum);
1628 return -EINVAL;
1629 }
1630
1631 mddev->pers = pers[pnum];
1632 spin_unlock(&pers_lock);
1633
1634 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1635
NeilBrown32a76272005-06-21 17:17:14 -07001636 /* before we start the array running, initialise the bitmap */
1637 err = bitmap_create(mddev);
1638 if (err)
1639 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1640 mdname(mddev), err);
1641 else
1642 err = mddev->pers->run(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 if (err) {
1644 printk(KERN_ERR "md: pers->run() failed ...\n");
1645 module_put(mddev->pers->owner);
1646 mddev->pers = NULL;
NeilBrown32a76272005-06-21 17:17:14 -07001647 bitmap_destroy(mddev);
1648 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 }
1650 atomic_set(&mddev->writes_pending,0);
1651 mddev->safemode = 0;
1652 mddev->safemode_timer.function = md_safemode_timeout;
1653 mddev->safemode_timer.data = (unsigned long) mddev;
1654 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1655 mddev->in_sync = 1;
1656
1657 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1658
1659 if (mddev->sb_dirty)
1660 md_update_sb(mddev);
1661
1662 set_capacity(disk, mddev->array_size<<1);
1663
1664 /* If we call blk_queue_make_request here, it will
1665 * re-initialise max_sectors etc which may have been
1666 * refined inside -> run. So just set the bits we need to set.
1667 * Most initialisation happended when we called
1668 * blk_queue_make_request(..., md_fail_request)
1669 * earlier.
1670 */
1671 mddev->queue->queuedata = mddev;
1672 mddev->queue->make_request_fn = mddev->pers->make_request;
1673
1674 mddev->changed = 1;
1675 return 0;
1676}
1677
1678static int restart_array(mddev_t *mddev)
1679{
1680 struct gendisk *disk = mddev->gendisk;
1681 int err;
1682
1683 /*
1684 * Complain if it has no devices
1685 */
1686 err = -ENXIO;
1687 if (list_empty(&mddev->disks))
1688 goto out;
1689
1690 if (mddev->pers) {
1691 err = -EBUSY;
1692 if (!mddev->ro)
1693 goto out;
1694
1695 mddev->safemode = 0;
1696 mddev->ro = 0;
1697 set_disk_ro(disk, 0);
1698
1699 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1700 mdname(mddev));
1701 /*
1702 * Kick recovery or resync if necessary
1703 */
1704 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1705 md_wakeup_thread(mddev->thread);
1706 err = 0;
1707 } else {
1708 printk(KERN_ERR "md: %s has no personality assigned.\n",
1709 mdname(mddev));
1710 err = -EINVAL;
1711 }
1712
1713out:
1714 return err;
1715}
1716
1717static int do_md_stop(mddev_t * mddev, int ro)
1718{
1719 int err = 0;
1720 struct gendisk *disk = mddev->gendisk;
1721
1722 if (mddev->pers) {
1723 if (atomic_read(&mddev->active)>2) {
1724 printk("md: %s still in use.\n",mdname(mddev));
1725 return -EBUSY;
1726 }
1727
1728 if (mddev->sync_thread) {
1729 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1730 md_unregister_thread(mddev->sync_thread);
1731 mddev->sync_thread = NULL;
1732 }
1733
1734 del_timer_sync(&mddev->safemode_timer);
1735
1736 invalidate_partition(disk, 0);
1737
1738 if (ro) {
1739 err = -ENXIO;
1740 if (mddev->ro)
1741 goto out;
1742 mddev->ro = 1;
1743 } else {
1744 if (mddev->ro)
1745 set_disk_ro(disk, 0);
1746 blk_queue_make_request(mddev->queue, md_fail_request);
1747 mddev->pers->stop(mddev);
1748 module_put(mddev->pers->owner);
1749 mddev->pers = NULL;
1750 if (mddev->ro)
1751 mddev->ro = 0;
1752 }
1753 if (!mddev->in_sync) {
1754 /* mark array as shutdown cleanly */
1755 mddev->in_sync = 1;
1756 md_update_sb(mddev);
1757 }
1758 if (ro)
1759 set_disk_ro(disk, 1);
1760 }
NeilBrown32a76272005-06-21 17:17:14 -07001761
1762 bitmap_destroy(mddev);
1763 if (mddev->bitmap_file) {
1764 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1765 fput(mddev->bitmap_file);
1766 mddev->bitmap_file = NULL;
1767 }
1768
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 /*
1770 * Free resources if final stop
1771 */
1772 if (!ro) {
1773 struct gendisk *disk;
1774 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1775
1776 export_array(mddev);
1777
1778 mddev->array_size = 0;
1779 disk = mddev->gendisk;
1780 if (disk)
1781 set_capacity(disk, 0);
1782 mddev->changed = 1;
1783 } else
1784 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1785 mdname(mddev));
1786 err = 0;
1787out:
1788 return err;
1789}
1790
1791static void autorun_array(mddev_t *mddev)
1792{
1793 mdk_rdev_t *rdev;
1794 struct list_head *tmp;
1795 int err;
1796
NeilBrowna757e642005-04-16 15:26:42 -07001797 if (list_empty(&mddev->disks))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
1800 printk(KERN_INFO "md: running: ");
1801
1802 ITERATE_RDEV(mddev,rdev,tmp) {
1803 char b[BDEVNAME_SIZE];
1804 printk("<%s>", bdevname(rdev->bdev,b));
1805 }
1806 printk("\n");
1807
1808 err = do_md_run (mddev);
1809 if (err) {
1810 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1811 do_md_stop (mddev, 0);
1812 }
1813}
1814
1815/*
1816 * lets try to run arrays based on all disks that have arrived
1817 * until now. (those are in pending_raid_disks)
1818 *
1819 * the method: pick the first pending disk, collect all disks with
1820 * the same UUID, remove all from the pending list and put them into
1821 * the 'same_array' list. Then order this list based on superblock
1822 * update time (freshest comes first), kick out 'old' disks and
1823 * compare superblocks. If everything's fine then run it.
1824 *
1825 * If "unit" is allocated, then bump its reference count
1826 */
1827static void autorun_devices(int part)
1828{
1829 struct list_head candidates;
1830 struct list_head *tmp;
1831 mdk_rdev_t *rdev0, *rdev;
1832 mddev_t *mddev;
1833 char b[BDEVNAME_SIZE];
1834
1835 printk(KERN_INFO "md: autorun ...\n");
1836 while (!list_empty(&pending_raid_disks)) {
1837 dev_t dev;
1838 rdev0 = list_entry(pending_raid_disks.next,
1839 mdk_rdev_t, same_set);
1840
1841 printk(KERN_INFO "md: considering %s ...\n",
1842 bdevname(rdev0->bdev,b));
1843 INIT_LIST_HEAD(&candidates);
1844 ITERATE_RDEV_PENDING(rdev,tmp)
1845 if (super_90_load(rdev, rdev0, 0) >= 0) {
1846 printk(KERN_INFO "md: adding %s ...\n",
1847 bdevname(rdev->bdev,b));
1848 list_move(&rdev->same_set, &candidates);
1849 }
1850 /*
1851 * now we have a set of devices, with all of them having
1852 * mostly sane superblocks. It's time to allocate the
1853 * mddev.
1854 */
1855 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1856 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1857 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1858 break;
1859 }
1860 if (part)
1861 dev = MKDEV(mdp_major,
1862 rdev0->preferred_minor << MdpMinorShift);
1863 else
1864 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1865
1866 md_probe(dev, NULL, NULL);
1867 mddev = mddev_find(dev);
1868 if (!mddev) {
1869 printk(KERN_ERR
1870 "md: cannot allocate memory for md drive.\n");
1871 break;
1872 }
1873 if (mddev_lock(mddev))
1874 printk(KERN_WARNING "md: %s locked, cannot run\n",
1875 mdname(mddev));
1876 else if (mddev->raid_disks || mddev->major_version
1877 || !list_empty(&mddev->disks)) {
1878 printk(KERN_WARNING
1879 "md: %s already running, cannot run %s\n",
1880 mdname(mddev), bdevname(rdev0->bdev,b));
1881 mddev_unlock(mddev);
1882 } else {
1883 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1884 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1885 list_del_init(&rdev->same_set);
1886 if (bind_rdev_to_array(rdev, mddev))
1887 export_rdev(rdev);
1888 }
1889 autorun_array(mddev);
1890 mddev_unlock(mddev);
1891 }
1892 /* on success, candidates will be empty, on error
1893 * it won't...
1894 */
1895 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1896 export_rdev(rdev);
1897 mddev_put(mddev);
1898 }
1899 printk(KERN_INFO "md: ... autorun DONE.\n");
1900}
1901
1902/*
1903 * import RAID devices based on one partition
1904 * if possible, the array gets run as well.
1905 */
1906
1907static int autostart_array(dev_t startdev)
1908{
1909 char b[BDEVNAME_SIZE];
1910 int err = -EINVAL, i;
1911 mdp_super_t *sb = NULL;
1912 mdk_rdev_t *start_rdev = NULL, *rdev;
1913
1914 start_rdev = md_import_device(startdev, 0, 0);
1915 if (IS_ERR(start_rdev))
1916 return err;
1917
1918
1919 /* NOTE: this can only work for 0.90.0 superblocks */
1920 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1921 if (sb->major_version != 0 ||
1922 sb->minor_version != 90 ) {
1923 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1924 export_rdev(start_rdev);
1925 return err;
1926 }
1927
1928 if (start_rdev->faulty) {
1929 printk(KERN_WARNING
1930 "md: can not autostart based on faulty %s!\n",
1931 bdevname(start_rdev->bdev,b));
1932 export_rdev(start_rdev);
1933 return err;
1934 }
1935 list_add(&start_rdev->same_set, &pending_raid_disks);
1936
1937 for (i = 0; i < MD_SB_DISKS; i++) {
1938 mdp_disk_t *desc = sb->disks + i;
1939 dev_t dev = MKDEV(desc->major, desc->minor);
1940
1941 if (!dev)
1942 continue;
1943 if (dev == startdev)
1944 continue;
1945 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
1946 continue;
1947 rdev = md_import_device(dev, 0, 0);
1948 if (IS_ERR(rdev))
1949 continue;
1950
1951 list_add(&rdev->same_set, &pending_raid_disks);
1952 }
1953
1954 /*
1955 * possibly return codes
1956 */
1957 autorun_devices(0);
1958 return 0;
1959
1960}
1961
1962
1963static int get_version(void __user * arg)
1964{
1965 mdu_version_t ver;
1966
1967 ver.major = MD_MAJOR_VERSION;
1968 ver.minor = MD_MINOR_VERSION;
1969 ver.patchlevel = MD_PATCHLEVEL_VERSION;
1970
1971 if (copy_to_user(arg, &ver, sizeof(ver)))
1972 return -EFAULT;
1973
1974 return 0;
1975}
1976
1977static int get_array_info(mddev_t * mddev, void __user * arg)
1978{
1979 mdu_array_info_t info;
1980 int nr,working,active,failed,spare;
1981 mdk_rdev_t *rdev;
1982 struct list_head *tmp;
1983
1984 nr=working=active=failed=spare=0;
1985 ITERATE_RDEV(mddev,rdev,tmp) {
1986 nr++;
1987 if (rdev->faulty)
1988 failed++;
1989 else {
1990 working++;
1991 if (rdev->in_sync)
1992 active++;
1993 else
1994 spare++;
1995 }
1996 }
1997
1998 info.major_version = mddev->major_version;
1999 info.minor_version = mddev->minor_version;
2000 info.patch_version = MD_PATCHLEVEL_VERSION;
2001 info.ctime = mddev->ctime;
2002 info.level = mddev->level;
2003 info.size = mddev->size;
2004 info.nr_disks = nr;
2005 info.raid_disks = mddev->raid_disks;
2006 info.md_minor = mddev->md_minor;
2007 info.not_persistent= !mddev->persistent;
2008
2009 info.utime = mddev->utime;
2010 info.state = 0;
2011 if (mddev->in_sync)
2012 info.state = (1<<MD_SB_CLEAN);
2013 info.active_disks = active;
2014 info.working_disks = working;
2015 info.failed_disks = failed;
2016 info.spare_disks = spare;
2017
2018 info.layout = mddev->layout;
2019 info.chunk_size = mddev->chunk_size;
2020
2021 if (copy_to_user(arg, &info, sizeof(info)))
2022 return -EFAULT;
2023
2024 return 0;
2025}
2026
NeilBrown32a76272005-06-21 17:17:14 -07002027static int get_bitmap_file(mddev_t * mddev, void * arg)
2028{
2029 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2030 char *ptr, *buf = NULL;
2031 int err = -ENOMEM;
2032
2033 file = kmalloc(sizeof(*file), GFP_KERNEL);
2034 if (!file)
2035 goto out;
2036
2037 /* bitmap disabled, zero the first byte and copy out */
2038 if (!mddev->bitmap || !mddev->bitmap->file) {
2039 file->pathname[0] = '\0';
2040 goto copy_out;
2041 }
2042
2043 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2044 if (!buf)
2045 goto out;
2046
2047 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2048 if (!ptr)
2049 goto out;
2050
2051 strcpy(file->pathname, ptr);
2052
2053copy_out:
2054 err = 0;
2055 if (copy_to_user(arg, file, sizeof(*file)))
2056 err = -EFAULT;
2057out:
2058 kfree(buf);
2059 kfree(file);
2060 return err;
2061}
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063static int get_disk_info(mddev_t * mddev, void __user * arg)
2064{
2065 mdu_disk_info_t info;
2066 unsigned int nr;
2067 mdk_rdev_t *rdev;
2068
2069 if (copy_from_user(&info, arg, sizeof(info)))
2070 return -EFAULT;
2071
2072 nr = info.number;
2073
2074 rdev = find_rdev_nr(mddev, nr);
2075 if (rdev) {
2076 info.major = MAJOR(rdev->bdev->bd_dev);
2077 info.minor = MINOR(rdev->bdev->bd_dev);
2078 info.raid_disk = rdev->raid_disk;
2079 info.state = 0;
2080 if (rdev->faulty)
2081 info.state |= (1<<MD_DISK_FAULTY);
2082 else if (rdev->in_sync) {
2083 info.state |= (1<<MD_DISK_ACTIVE);
2084 info.state |= (1<<MD_DISK_SYNC);
2085 }
2086 } else {
2087 info.major = info.minor = 0;
2088 info.raid_disk = -1;
2089 info.state = (1<<MD_DISK_REMOVED);
2090 }
2091
2092 if (copy_to_user(arg, &info, sizeof(info)))
2093 return -EFAULT;
2094
2095 return 0;
2096}
2097
2098static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2099{
2100 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2101 mdk_rdev_t *rdev;
2102 dev_t dev = MKDEV(info->major,info->minor);
2103
2104 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2105 return -EOVERFLOW;
2106
2107 if (!mddev->raid_disks) {
2108 int err;
2109 /* expecting a device which has a superblock */
2110 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2111 if (IS_ERR(rdev)) {
2112 printk(KERN_WARNING
2113 "md: md_import_device returned %ld\n",
2114 PTR_ERR(rdev));
2115 return PTR_ERR(rdev);
2116 }
2117 if (!list_empty(&mddev->disks)) {
2118 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2119 mdk_rdev_t, same_set);
2120 int err = super_types[mddev->major_version]
2121 .load_super(rdev, rdev0, mddev->minor_version);
2122 if (err < 0) {
2123 printk(KERN_WARNING
2124 "md: %s has different UUID to %s\n",
2125 bdevname(rdev->bdev,b),
2126 bdevname(rdev0->bdev,b2));
2127 export_rdev(rdev);
2128 return -EINVAL;
2129 }
2130 }
2131 err = bind_rdev_to_array(rdev, mddev);
2132 if (err)
2133 export_rdev(rdev);
2134 return err;
2135 }
2136
2137 /*
2138 * add_new_disk can be used once the array is assembled
2139 * to add "hot spares". They must already have a superblock
2140 * written
2141 */
2142 if (mddev->pers) {
2143 int err;
2144 if (!mddev->pers->hot_add_disk) {
2145 printk(KERN_WARNING
2146 "%s: personality does not support diskops!\n",
2147 mdname(mddev));
2148 return -EINVAL;
2149 }
2150 rdev = md_import_device(dev, mddev->major_version,
2151 mddev->minor_version);
2152 if (IS_ERR(rdev)) {
2153 printk(KERN_WARNING
2154 "md: md_import_device returned %ld\n",
2155 PTR_ERR(rdev));
2156 return PTR_ERR(rdev);
2157 }
2158 rdev->in_sync = 0; /* just to be sure */
2159 rdev->raid_disk = -1;
2160 err = bind_rdev_to_array(rdev, mddev);
2161 if (err)
2162 export_rdev(rdev);
NeilBrownc3617772005-06-21 17:17:10 -07002163
2164 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 if (mddev->thread)
2166 md_wakeup_thread(mddev->thread);
2167 return err;
2168 }
2169
2170 /* otherwise, add_new_disk is only allowed
2171 * for major_version==0 superblocks
2172 */
2173 if (mddev->major_version != 0) {
2174 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2175 mdname(mddev));
2176 return -EINVAL;
2177 }
2178
2179 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2180 int err;
2181 rdev = md_import_device (dev, -1, 0);
2182 if (IS_ERR(rdev)) {
2183 printk(KERN_WARNING
2184 "md: error, md_import_device() returned %ld\n",
2185 PTR_ERR(rdev));
2186 return PTR_ERR(rdev);
2187 }
2188 rdev->desc_nr = info->number;
2189 if (info->raid_disk < mddev->raid_disks)
2190 rdev->raid_disk = info->raid_disk;
2191 else
2192 rdev->raid_disk = -1;
2193
2194 rdev->faulty = 0;
2195 if (rdev->raid_disk < mddev->raid_disks)
2196 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2197 else
2198 rdev->in_sync = 0;
2199
2200 err = bind_rdev_to_array(rdev, mddev);
2201 if (err) {
2202 export_rdev(rdev);
2203 return err;
2204 }
2205
2206 if (!mddev->persistent) {
2207 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2208 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2209 } else
2210 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2211 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2212
2213 if (!mddev->size || (mddev->size > rdev->size))
2214 mddev->size = rdev->size;
2215 }
2216
2217 return 0;
2218}
2219
2220static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2221{
2222 char b[BDEVNAME_SIZE];
2223 mdk_rdev_t *rdev;
2224
2225 if (!mddev->pers)
2226 return -ENODEV;
2227
2228 rdev = find_rdev(mddev, dev);
2229 if (!rdev)
2230 return -ENXIO;
2231
2232 if (rdev->raid_disk >= 0)
2233 goto busy;
2234
2235 kick_rdev_from_array(rdev);
2236 md_update_sb(mddev);
2237
2238 return 0;
2239busy:
2240 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2241 bdevname(rdev->bdev,b), mdname(mddev));
2242 return -EBUSY;
2243}
2244
2245static int hot_add_disk(mddev_t * mddev, dev_t dev)
2246{
2247 char b[BDEVNAME_SIZE];
2248 int err;
2249 unsigned int size;
2250 mdk_rdev_t *rdev;
2251
2252 if (!mddev->pers)
2253 return -ENODEV;
2254
2255 if (mddev->major_version != 0) {
2256 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2257 " version-0 superblocks.\n",
2258 mdname(mddev));
2259 return -EINVAL;
2260 }
2261 if (!mddev->pers->hot_add_disk) {
2262 printk(KERN_WARNING
2263 "%s: personality does not support diskops!\n",
2264 mdname(mddev));
2265 return -EINVAL;
2266 }
2267
2268 rdev = md_import_device (dev, -1, 0);
2269 if (IS_ERR(rdev)) {
2270 printk(KERN_WARNING
2271 "md: error, md_import_device() returned %ld\n",
2272 PTR_ERR(rdev));
2273 return -EINVAL;
2274 }
2275
2276 if (mddev->persistent)
2277 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2278 else
2279 rdev->sb_offset =
2280 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2281
2282 size = calc_dev_size(rdev, mddev->chunk_size);
2283 rdev->size = size;
2284
2285 if (size < mddev->size) {
2286 printk(KERN_WARNING
2287 "%s: disk size %llu blocks < array size %llu\n",
2288 mdname(mddev), (unsigned long long)size,
2289 (unsigned long long)mddev->size);
2290 err = -ENOSPC;
2291 goto abort_export;
2292 }
2293
2294 if (rdev->faulty) {
2295 printk(KERN_WARNING
2296 "md: can not hot-add faulty %s disk to %s!\n",
2297 bdevname(rdev->bdev,b), mdname(mddev));
2298 err = -EINVAL;
2299 goto abort_export;
2300 }
2301 rdev->in_sync = 0;
2302 rdev->desc_nr = -1;
2303 bind_rdev_to_array(rdev, mddev);
2304
2305 /*
2306 * The rest should better be atomic, we can have disk failures
2307 * noticed in interrupt contexts ...
2308 */
2309
2310 if (rdev->desc_nr == mddev->max_disks) {
2311 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2312 mdname(mddev));
2313 err = -EBUSY;
2314 goto abort_unbind_export;
2315 }
2316
2317 rdev->raid_disk = -1;
2318
2319 md_update_sb(mddev);
2320
2321 /*
2322 * Kick recovery, maybe this spare has to be added to the
2323 * array immediately.
2324 */
2325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2326 md_wakeup_thread(mddev->thread);
2327
2328 return 0;
2329
2330abort_unbind_export:
2331 unbind_rdev_from_array(rdev);
2332
2333abort_export:
2334 export_rdev(rdev);
2335 return err;
2336}
2337
NeilBrown32a76272005-06-21 17:17:14 -07002338/* similar to deny_write_access, but accounts for our holding a reference
2339 * to the file ourselves */
2340static int deny_bitmap_write_access(struct file * file)
2341{
2342 struct inode *inode = file->f_mapping->host;
2343
2344 spin_lock(&inode->i_lock);
2345 if (atomic_read(&inode->i_writecount) > 1) {
2346 spin_unlock(&inode->i_lock);
2347 return -ETXTBSY;
2348 }
2349 atomic_set(&inode->i_writecount, -1);
2350 spin_unlock(&inode->i_lock);
2351
2352 return 0;
2353}
2354
2355static int set_bitmap_file(mddev_t *mddev, int fd)
2356{
2357 int err;
2358
2359 if (mddev->pers)
2360 return -EBUSY;
2361
2362 mddev->bitmap_file = fget(fd);
2363
2364 if (mddev->bitmap_file == NULL) {
2365 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2366 mdname(mddev));
2367 return -EBADF;
2368 }
2369
2370 err = deny_bitmap_write_access(mddev->bitmap_file);
2371 if (err) {
2372 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2373 mdname(mddev));
2374 fput(mddev->bitmap_file);
2375 mddev->bitmap_file = NULL;
2376 }
2377 return err;
2378}
2379
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380/*
2381 * set_array_info is used two different ways
2382 * The original usage is when creating a new array.
2383 * In this usage, raid_disks is > 0 and it together with
2384 * level, size, not_persistent,layout,chunksize determine the
2385 * shape of the array.
2386 * This will always create an array with a type-0.90.0 superblock.
2387 * The newer usage is when assembling an array.
2388 * In this case raid_disks will be 0, and the major_version field is
2389 * use to determine which style super-blocks are to be found on the devices.
2390 * The minor and patch _version numbers are also kept incase the
2391 * super_block handler wishes to interpret them.
2392 */
2393static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2394{
2395
2396 if (info->raid_disks == 0) {
2397 /* just setting version number for superblock loading */
2398 if (info->major_version < 0 ||
2399 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2400 super_types[info->major_version].name == NULL) {
2401 /* maybe try to auto-load a module? */
2402 printk(KERN_INFO
2403 "md: superblock version %d not known\n",
2404 info->major_version);
2405 return -EINVAL;
2406 }
2407 mddev->major_version = info->major_version;
2408 mddev->minor_version = info->minor_version;
2409 mddev->patch_version = info->patch_version;
2410 return 0;
2411 }
2412 mddev->major_version = MD_MAJOR_VERSION;
2413 mddev->minor_version = MD_MINOR_VERSION;
2414 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2415 mddev->ctime = get_seconds();
2416
2417 mddev->level = info->level;
2418 mddev->size = info->size;
2419 mddev->raid_disks = info->raid_disks;
2420 /* don't set md_minor, it is determined by which /dev/md* was
2421 * openned
2422 */
2423 if (info->state & (1<<MD_SB_CLEAN))
2424 mddev->recovery_cp = MaxSector;
2425 else
2426 mddev->recovery_cp = 0;
2427 mddev->persistent = ! info->not_persistent;
2428
2429 mddev->layout = info->layout;
2430 mddev->chunk_size = info->chunk_size;
2431
2432 mddev->max_disks = MD_SB_DISKS;
2433
2434 mddev->sb_dirty = 1;
2435
2436 /*
2437 * Generate a 128 bit UUID
2438 */
2439 get_random_bytes(mddev->uuid, 16);
2440
2441 return 0;
2442}
2443
2444/*
2445 * update_array_info is used to change the configuration of an
2446 * on-line array.
2447 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2448 * fields in the info are checked against the array.
2449 * Any differences that cannot be handled will cause an error.
2450 * Normally, only one change can be managed at a time.
2451 */
2452static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2453{
2454 int rv = 0;
2455 int cnt = 0;
2456
2457 if (mddev->major_version != info->major_version ||
2458 mddev->minor_version != info->minor_version ||
2459/* mddev->patch_version != info->patch_version || */
2460 mddev->ctime != info->ctime ||
2461 mddev->level != info->level ||
2462/* mddev->layout != info->layout || */
2463 !mddev->persistent != info->not_persistent||
2464 mddev->chunk_size != info->chunk_size )
2465 return -EINVAL;
2466 /* Check there is only one change */
2467 if (mddev->size != info->size) cnt++;
2468 if (mddev->raid_disks != info->raid_disks) cnt++;
2469 if (mddev->layout != info->layout) cnt++;
2470 if (cnt == 0) return 0;
2471 if (cnt > 1) return -EINVAL;
2472
2473 if (mddev->layout != info->layout) {
2474 /* Change layout
2475 * we don't need to do anything at the md level, the
2476 * personality will take care of it all.
2477 */
2478 if (mddev->pers->reconfig == NULL)
2479 return -EINVAL;
2480 else
2481 return mddev->pers->reconfig(mddev, info->layout, -1);
2482 }
2483 if (mddev->size != info->size) {
2484 mdk_rdev_t * rdev;
2485 struct list_head *tmp;
2486 if (mddev->pers->resize == NULL)
2487 return -EINVAL;
2488 /* The "size" is the amount of each device that is used.
2489 * This can only make sense for arrays with redundancy.
2490 * linear and raid0 always use whatever space is available
2491 * We can only consider changing the size if no resync
2492 * or reconstruction is happening, and if the new size
2493 * is acceptable. It must fit before the sb_offset or,
2494 * if that is <data_offset, it must fit before the
2495 * size of each device.
2496 * If size is zero, we find the largest size that fits.
2497 */
2498 if (mddev->sync_thread)
2499 return -EBUSY;
2500 ITERATE_RDEV(mddev,rdev,tmp) {
2501 sector_t avail;
2502 int fit = (info->size == 0);
2503 if (rdev->sb_offset > rdev->data_offset)
2504 avail = (rdev->sb_offset*2) - rdev->data_offset;
2505 else
2506 avail = get_capacity(rdev->bdev->bd_disk)
2507 - rdev->data_offset;
2508 if (fit && (info->size == 0 || info->size > avail/2))
2509 info->size = avail/2;
2510 if (avail < ((sector_t)info->size << 1))
2511 return -ENOSPC;
2512 }
2513 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2514 if (!rv) {
2515 struct block_device *bdev;
2516
2517 bdev = bdget_disk(mddev->gendisk, 0);
2518 if (bdev) {
2519 down(&bdev->bd_inode->i_sem);
2520 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2521 up(&bdev->bd_inode->i_sem);
2522 bdput(bdev);
2523 }
2524 }
2525 }
2526 if (mddev->raid_disks != info->raid_disks) {
2527 /* change the number of raid disks */
2528 if (mddev->pers->reshape == NULL)
2529 return -EINVAL;
2530 if (info->raid_disks <= 0 ||
2531 info->raid_disks >= mddev->max_disks)
2532 return -EINVAL;
2533 if (mddev->sync_thread)
2534 return -EBUSY;
2535 rv = mddev->pers->reshape(mddev, info->raid_disks);
2536 if (!rv) {
2537 struct block_device *bdev;
2538
2539 bdev = bdget_disk(mddev->gendisk, 0);
2540 if (bdev) {
2541 down(&bdev->bd_inode->i_sem);
2542 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2543 up(&bdev->bd_inode->i_sem);
2544 bdput(bdev);
2545 }
2546 }
2547 }
2548 md_update_sb(mddev);
2549 return rv;
2550}
2551
2552static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2553{
2554 mdk_rdev_t *rdev;
2555
2556 if (mddev->pers == NULL)
2557 return -ENODEV;
2558
2559 rdev = find_rdev(mddev, dev);
2560 if (!rdev)
2561 return -ENODEV;
2562
2563 md_error(mddev, rdev);
2564 return 0;
2565}
2566
2567static int md_ioctl(struct inode *inode, struct file *file,
2568 unsigned int cmd, unsigned long arg)
2569{
2570 int err = 0;
2571 void __user *argp = (void __user *)arg;
2572 struct hd_geometry __user *loc = argp;
2573 mddev_t *mddev = NULL;
2574
2575 if (!capable(CAP_SYS_ADMIN))
2576 return -EACCES;
2577
2578 /*
2579 * Commands dealing with the RAID driver but not any
2580 * particular array:
2581 */
2582 switch (cmd)
2583 {
2584 case RAID_VERSION:
2585 err = get_version(argp);
2586 goto done;
2587
2588 case PRINT_RAID_DEBUG:
2589 err = 0;
2590 md_print_devices();
2591 goto done;
2592
2593#ifndef MODULE
2594 case RAID_AUTORUN:
2595 err = 0;
2596 autostart_arrays(arg);
2597 goto done;
2598#endif
2599 default:;
2600 }
2601
2602 /*
2603 * Commands creating/starting a new array:
2604 */
2605
2606 mddev = inode->i_bdev->bd_disk->private_data;
2607
2608 if (!mddev) {
2609 BUG();
2610 goto abort;
2611 }
2612
2613
2614 if (cmd == START_ARRAY) {
2615 /* START_ARRAY doesn't need to lock the array as autostart_array
2616 * does the locking, and it could even be a different array
2617 */
2618 static int cnt = 3;
2619 if (cnt > 0 ) {
2620 printk(KERN_WARNING
2621 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2622 "This will not be supported beyond 2.6\n",
2623 current->comm, current->pid);
2624 cnt--;
2625 }
2626 err = autostart_array(new_decode_dev(arg));
2627 if (err) {
2628 printk(KERN_WARNING "md: autostart failed!\n");
2629 goto abort;
2630 }
2631 goto done;
2632 }
2633
2634 err = mddev_lock(mddev);
2635 if (err) {
2636 printk(KERN_INFO
2637 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2638 err, cmd);
2639 goto abort;
2640 }
2641
2642 switch (cmd)
2643 {
2644 case SET_ARRAY_INFO:
2645 {
2646 mdu_array_info_t info;
2647 if (!arg)
2648 memset(&info, 0, sizeof(info));
2649 else if (copy_from_user(&info, argp, sizeof(info))) {
2650 err = -EFAULT;
2651 goto abort_unlock;
2652 }
2653 if (mddev->pers) {
2654 err = update_array_info(mddev, &info);
2655 if (err) {
2656 printk(KERN_WARNING "md: couldn't update"
2657 " array info. %d\n", err);
2658 goto abort_unlock;
2659 }
2660 goto done_unlock;
2661 }
2662 if (!list_empty(&mddev->disks)) {
2663 printk(KERN_WARNING
2664 "md: array %s already has disks!\n",
2665 mdname(mddev));
2666 err = -EBUSY;
2667 goto abort_unlock;
2668 }
2669 if (mddev->raid_disks) {
2670 printk(KERN_WARNING
2671 "md: array %s already initialised!\n",
2672 mdname(mddev));
2673 err = -EBUSY;
2674 goto abort_unlock;
2675 }
2676 err = set_array_info(mddev, &info);
2677 if (err) {
2678 printk(KERN_WARNING "md: couldn't set"
2679 " array info. %d\n", err);
2680 goto abort_unlock;
2681 }
2682 }
2683 goto done_unlock;
2684
2685 default:;
2686 }
2687
2688 /*
2689 * Commands querying/configuring an existing array:
2690 */
NeilBrown32a76272005-06-21 17:17:14 -07002691 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2692 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2693 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2694 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 err = -ENODEV;
2696 goto abort_unlock;
2697 }
2698
2699 /*
2700 * Commands even a read-only array can execute:
2701 */
2702 switch (cmd)
2703 {
2704 case GET_ARRAY_INFO:
2705 err = get_array_info(mddev, argp);
2706 goto done_unlock;
2707
NeilBrown32a76272005-06-21 17:17:14 -07002708 case GET_BITMAP_FILE:
2709 err = get_bitmap_file(mddev, (void *)arg);
2710 goto done_unlock;
2711
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712 case GET_DISK_INFO:
2713 err = get_disk_info(mddev, argp);
2714 goto done_unlock;
2715
2716 case RESTART_ARRAY_RW:
2717 err = restart_array(mddev);
2718 goto done_unlock;
2719
2720 case STOP_ARRAY:
2721 err = do_md_stop (mddev, 0);
2722 goto done_unlock;
2723
2724 case STOP_ARRAY_RO:
2725 err = do_md_stop (mddev, 1);
2726 goto done_unlock;
2727
2728 /*
2729 * We have a problem here : there is no easy way to give a CHS
2730 * virtual geometry. We currently pretend that we have a 2 heads
2731 * 4 sectors (with a BIG number of cylinders...). This drives
2732 * dosfs just mad... ;-)
2733 */
2734 case HDIO_GETGEO:
2735 if (!loc) {
2736 err = -EINVAL;
2737 goto abort_unlock;
2738 }
2739 err = put_user (2, (char __user *) &loc->heads);
2740 if (err)
2741 goto abort_unlock;
2742 err = put_user (4, (char __user *) &loc->sectors);
2743 if (err)
2744 goto abort_unlock;
2745 err = put_user(get_capacity(mddev->gendisk)/8,
2746 (short __user *) &loc->cylinders);
2747 if (err)
2748 goto abort_unlock;
2749 err = put_user (get_start_sect(inode->i_bdev),
2750 (long __user *) &loc->start);
2751 goto done_unlock;
2752 }
2753
2754 /*
2755 * The remaining ioctls are changing the state of the
2756 * superblock, so we do not allow read-only arrays
2757 * here:
2758 */
2759 if (mddev->ro) {
2760 err = -EROFS;
2761 goto abort_unlock;
2762 }
2763
2764 switch (cmd)
2765 {
2766 case ADD_NEW_DISK:
2767 {
2768 mdu_disk_info_t info;
2769 if (copy_from_user(&info, argp, sizeof(info)))
2770 err = -EFAULT;
2771 else
2772 err = add_new_disk(mddev, &info);
2773 goto done_unlock;
2774 }
2775
2776 case HOT_REMOVE_DISK:
2777 err = hot_remove_disk(mddev, new_decode_dev(arg));
2778 goto done_unlock;
2779
2780 case HOT_ADD_DISK:
2781 err = hot_add_disk(mddev, new_decode_dev(arg));
2782 goto done_unlock;
2783
2784 case SET_DISK_FAULTY:
2785 err = set_disk_faulty(mddev, new_decode_dev(arg));
2786 goto done_unlock;
2787
2788 case RUN_ARRAY:
2789 err = do_md_run (mddev);
2790 goto done_unlock;
2791
NeilBrown32a76272005-06-21 17:17:14 -07002792 case SET_BITMAP_FILE:
2793 err = set_bitmap_file(mddev, (int)arg);
2794 goto done_unlock;
2795
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 default:
2797 if (_IOC_TYPE(cmd) == MD_MAJOR)
2798 printk(KERN_WARNING "md: %s(pid %d) used"
2799 " obsolete MD ioctl, upgrade your"
2800 " software to use new ictls.\n",
2801 current->comm, current->pid);
2802 err = -EINVAL;
2803 goto abort_unlock;
2804 }
2805
2806done_unlock:
2807abort_unlock:
2808 mddev_unlock(mddev);
2809
2810 return err;
2811done:
2812 if (err)
2813 MD_BUG();
2814abort:
2815 return err;
2816}
2817
2818static int md_open(struct inode *inode, struct file *file)
2819{
2820 /*
2821 * Succeed if we can lock the mddev, which confirms that
2822 * it isn't being stopped right now.
2823 */
2824 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2825 int err;
2826
2827 if ((err = mddev_lock(mddev)))
2828 goto out;
2829
2830 err = 0;
2831 mddev_get(mddev);
2832 mddev_unlock(mddev);
2833
2834 check_disk_change(inode->i_bdev);
2835 out:
2836 return err;
2837}
2838
2839static int md_release(struct inode *inode, struct file * file)
2840{
2841 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2842
2843 if (!mddev)
2844 BUG();
2845 mddev_put(mddev);
2846
2847 return 0;
2848}
2849
2850static int md_media_changed(struct gendisk *disk)
2851{
2852 mddev_t *mddev = disk->private_data;
2853
2854 return mddev->changed;
2855}
2856
2857static int md_revalidate(struct gendisk *disk)
2858{
2859 mddev_t *mddev = disk->private_data;
2860
2861 mddev->changed = 0;
2862 return 0;
2863}
2864static struct block_device_operations md_fops =
2865{
2866 .owner = THIS_MODULE,
2867 .open = md_open,
2868 .release = md_release,
2869 .ioctl = md_ioctl,
2870 .media_changed = md_media_changed,
2871 .revalidate_disk= md_revalidate,
2872};
2873
Adrian Bunk75c96f82005-05-05 16:16:09 -07002874static int md_thread(void * arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875{
2876 mdk_thread_t *thread = arg;
2877
2878 lock_kernel();
2879
2880 /*
2881 * Detach thread
2882 */
2883
2884 daemonize(thread->name, mdname(thread->mddev));
2885
2886 current->exit_signal = SIGCHLD;
2887 allow_signal(SIGKILL);
2888 thread->tsk = current;
2889
2890 /*
2891 * md_thread is a 'system-thread', it's priority should be very
2892 * high. We avoid resource deadlocks individually in each
2893 * raid personality. (RAID5 does preallocation) We also use RR and
2894 * the very same RT priority as kswapd, thus we will never get
2895 * into a priority inversion deadlock.
2896 *
2897 * we definitely have to have equal or higher priority than
2898 * bdflush, otherwise bdflush will deadlock if there are too
2899 * many dirty RAID5 blocks.
2900 */
2901 unlock_kernel();
2902
2903 complete(thread->event);
2904 while (thread->run) {
2905 void (*run)(mddev_t *);
2906
NeilBrown32a76272005-06-21 17:17:14 -07002907 wait_event_interruptible_timeout(thread->wqueue,
2908 test_bit(THREAD_WAKEUP, &thread->flags),
2909 thread->timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 if (current->flags & PF_FREEZE)
2911 refrigerator(PF_FREEZE);
2912
2913 clear_bit(THREAD_WAKEUP, &thread->flags);
2914
2915 run = thread->run;
2916 if (run)
2917 run(thread->mddev);
2918
2919 if (signal_pending(current))
2920 flush_signals(current);
2921 }
2922 complete(thread->event);
2923 return 0;
2924}
2925
2926void md_wakeup_thread(mdk_thread_t *thread)
2927{
2928 if (thread) {
2929 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2930 set_bit(THREAD_WAKEUP, &thread->flags);
2931 wake_up(&thread->wqueue);
2932 }
2933}
2934
2935mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
2936 const char *name)
2937{
2938 mdk_thread_t *thread;
2939 int ret;
2940 struct completion event;
2941
2942 thread = (mdk_thread_t *) kmalloc
2943 (sizeof(mdk_thread_t), GFP_KERNEL);
2944 if (!thread)
2945 return NULL;
2946
2947 memset(thread, 0, sizeof(mdk_thread_t));
2948 init_waitqueue_head(&thread->wqueue);
2949
2950 init_completion(&event);
2951 thread->event = &event;
2952 thread->run = run;
2953 thread->mddev = mddev;
2954 thread->name = name;
NeilBrown32a76272005-06-21 17:17:14 -07002955 thread->timeout = MAX_SCHEDULE_TIMEOUT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 ret = kernel_thread(md_thread, thread, 0);
2957 if (ret < 0) {
2958 kfree(thread);
2959 return NULL;
2960 }
2961 wait_for_completion(&event);
2962 return thread;
2963}
2964
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965void md_unregister_thread(mdk_thread_t *thread)
2966{
2967 struct completion event;
2968
2969 init_completion(&event);
2970
2971 thread->event = &event;
NeilBrownd28446f2005-04-16 15:26:41 -07002972
2973 /* As soon as ->run is set to NULL, the task could disappear,
2974 * so we need to hold tasklist_lock until we have sent the signal
2975 */
2976 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
2977 read_lock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 thread->run = NULL;
NeilBrownd28446f2005-04-16 15:26:41 -07002979 send_sig(SIGKILL, thread->tsk, 1);
2980 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 wait_for_completion(&event);
2982 kfree(thread);
2983}
2984
2985void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
2986{
2987 if (!mddev) {
2988 MD_BUG();
2989 return;
2990 }
2991
2992 if (!rdev || rdev->faulty)
2993 return;
NeilBrown32a76272005-06-21 17:17:14 -07002994/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
2996 mdname(mddev),
2997 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
2998 __builtin_return_address(0),__builtin_return_address(1),
2999 __builtin_return_address(2),__builtin_return_address(3));
NeilBrown32a76272005-06-21 17:17:14 -07003000*/
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 if (!mddev->pers->error_handler)
3002 return;
3003 mddev->pers->error_handler(mddev,rdev);
3004 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3005 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3006 md_wakeup_thread(mddev->thread);
3007}
3008
3009/* seq_file implementation /proc/mdstat */
3010
3011static void status_unused(struct seq_file *seq)
3012{
3013 int i = 0;
3014 mdk_rdev_t *rdev;
3015 struct list_head *tmp;
3016
3017 seq_printf(seq, "unused devices: ");
3018
3019 ITERATE_RDEV_PENDING(rdev,tmp) {
3020 char b[BDEVNAME_SIZE];
3021 i++;
3022 seq_printf(seq, "%s ",
3023 bdevname(rdev->bdev,b));
3024 }
3025 if (!i)
3026 seq_printf(seq, "<none>");
3027
3028 seq_printf(seq, "\n");
3029}
3030
3031
3032static void status_resync(struct seq_file *seq, mddev_t * mddev)
3033{
3034 unsigned long max_blocks, resync, res, dt, db, rt;
3035
3036 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3037
3038 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3039 max_blocks = mddev->resync_max_sectors >> 1;
3040 else
3041 max_blocks = mddev->size;
3042
3043 /*
3044 * Should not happen.
3045 */
3046 if (!max_blocks) {
3047 MD_BUG();
3048 return;
3049 }
3050 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3051 {
3052 int i, x = res/50, y = 20-x;
3053 seq_printf(seq, "[");
3054 for (i = 0; i < x; i++)
3055 seq_printf(seq, "=");
3056 seq_printf(seq, ">");
3057 for (i = 0; i < y; i++)
3058 seq_printf(seq, ".");
3059 seq_printf(seq, "] ");
3060 }
3061 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3062 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3063 "resync" : "recovery"),
3064 res/10, res % 10, resync, max_blocks);
3065
3066 /*
3067 * We do not want to overflow, so the order of operands and
3068 * the * 100 / 100 trick are important. We do a +1 to be
3069 * safe against division by zero. We only estimate anyway.
3070 *
3071 * dt: time from mark until now
3072 * db: blocks written from mark until now
3073 * rt: remaining time
3074 */
3075 dt = ((jiffies - mddev->resync_mark) / HZ);
3076 if (!dt) dt++;
3077 db = resync - (mddev->resync_mark_cnt/2);
3078 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3079
3080 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3081
3082 seq_printf(seq, " speed=%ldK/sec", db/dt);
3083}
3084
3085static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3086{
3087 struct list_head *tmp;
3088 loff_t l = *pos;
3089 mddev_t *mddev;
3090
3091 if (l >= 0x10000)
3092 return NULL;
3093 if (!l--)
3094 /* header */
3095 return (void*)1;
3096
3097 spin_lock(&all_mddevs_lock);
3098 list_for_each(tmp,&all_mddevs)
3099 if (!l--) {
3100 mddev = list_entry(tmp, mddev_t, all_mddevs);
3101 mddev_get(mddev);
3102 spin_unlock(&all_mddevs_lock);
3103 return mddev;
3104 }
3105 spin_unlock(&all_mddevs_lock);
3106 if (!l--)
3107 return (void*)2;/* tail */
3108 return NULL;
3109}
3110
3111static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3112{
3113 struct list_head *tmp;
3114 mddev_t *next_mddev, *mddev = v;
3115
3116 ++*pos;
3117 if (v == (void*)2)
3118 return NULL;
3119
3120 spin_lock(&all_mddevs_lock);
3121 if (v == (void*)1)
3122 tmp = all_mddevs.next;
3123 else
3124 tmp = mddev->all_mddevs.next;
3125 if (tmp != &all_mddevs)
3126 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3127 else {
3128 next_mddev = (void*)2;
3129 *pos = 0x10000;
3130 }
3131 spin_unlock(&all_mddevs_lock);
3132
3133 if (v != (void*)1)
3134 mddev_put(mddev);
3135 return next_mddev;
3136
3137}
3138
3139static void md_seq_stop(struct seq_file *seq, void *v)
3140{
3141 mddev_t *mddev = v;
3142
3143 if (mddev && v != (void*)1 && v != (void*)2)
3144 mddev_put(mddev);
3145}
3146
3147static int md_seq_show(struct seq_file *seq, void *v)
3148{
3149 mddev_t *mddev = v;
3150 sector_t size;
3151 struct list_head *tmp2;
3152 mdk_rdev_t *rdev;
3153 int i;
NeilBrown32a76272005-06-21 17:17:14 -07003154 struct bitmap *bitmap;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155
3156 if (v == (void*)1) {
3157 seq_printf(seq, "Personalities : ");
3158 spin_lock(&pers_lock);
3159 for (i = 0; i < MAX_PERSONALITY; i++)
3160 if (pers[i])
3161 seq_printf(seq, "[%s] ", pers[i]->name);
3162
3163 spin_unlock(&pers_lock);
3164 seq_printf(seq, "\n");
3165 return 0;
3166 }
3167 if (v == (void*)2) {
3168 status_unused(seq);
3169 return 0;
3170 }
3171
3172 if (mddev_lock(mddev)!=0)
3173 return -EINTR;
3174 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3175 seq_printf(seq, "%s : %sactive", mdname(mddev),
3176 mddev->pers ? "" : "in");
3177 if (mddev->pers) {
3178 if (mddev->ro)
3179 seq_printf(seq, " (read-only)");
3180 seq_printf(seq, " %s", mddev->pers->name);
3181 }
3182
3183 size = 0;
3184 ITERATE_RDEV(mddev,rdev,tmp2) {
3185 char b[BDEVNAME_SIZE];
3186 seq_printf(seq, " %s[%d]",
3187 bdevname(rdev->bdev,b), rdev->desc_nr);
3188 if (rdev->faulty) {
3189 seq_printf(seq, "(F)");
3190 continue;
3191 }
3192 size += rdev->size;
3193 }
3194
3195 if (!list_empty(&mddev->disks)) {
3196 if (mddev->pers)
3197 seq_printf(seq, "\n %llu blocks",
3198 (unsigned long long)mddev->array_size);
3199 else
3200 seq_printf(seq, "\n %llu blocks",
3201 (unsigned long long)size);
3202 }
3203
3204 if (mddev->pers) {
3205 mddev->pers->status (seq, mddev);
3206 seq_printf(seq, "\n ");
NeilBrown32a76272005-06-21 17:17:14 -07003207 if (mddev->curr_resync > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003208 status_resync (seq, mddev);
NeilBrown32a76272005-06-21 17:17:14 -07003209 seq_printf(seq, "\n ");
3210 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3211 seq_printf(seq, " resync=DELAYED\n ");
3212 } else
3213 seq_printf(seq, "\n ");
3214
3215 if ((bitmap = mddev->bitmap)) {
3216 char *buf, *path;
3217 unsigned long chunk_kb;
3218 unsigned long flags;
3219 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
3220 spin_lock_irqsave(&bitmap->lock, flags);
3221 chunk_kb = bitmap->chunksize >> 10;
3222 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3223 "%lu%s chunk",
3224 bitmap->pages - bitmap->missing_pages,
3225 bitmap->pages,
3226 (bitmap->pages - bitmap->missing_pages)
3227 << (PAGE_SHIFT - 10),
3228 chunk_kb ? chunk_kb : bitmap->chunksize,
3229 chunk_kb ? "KB" : "B");
3230 if (bitmap->file && buf) {
3231 path = file_path(bitmap->file, buf, PAGE_SIZE);
3232 seq_printf(seq, ", file: %s", path ? path : "");
3233 }
3234 seq_printf(seq, "\n");
3235 spin_unlock_irqrestore(&bitmap->lock, flags);
3236 kfree(buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 }
3238
3239 seq_printf(seq, "\n");
3240 }
3241 mddev_unlock(mddev);
3242
3243 return 0;
3244}
3245
3246static struct seq_operations md_seq_ops = {
3247 .start = md_seq_start,
3248 .next = md_seq_next,
3249 .stop = md_seq_stop,
3250 .show = md_seq_show,
3251};
3252
3253static int md_seq_open(struct inode *inode, struct file *file)
3254{
3255 int error;
3256
3257 error = seq_open(file, &md_seq_ops);
3258 return error;
3259}
3260
3261static struct file_operations md_seq_fops = {
3262 .open = md_seq_open,
3263 .read = seq_read,
3264 .llseek = seq_lseek,
3265 .release = seq_release,
3266};
3267
3268int register_md_personality(int pnum, mdk_personality_t *p)
3269{
3270 if (pnum >= MAX_PERSONALITY) {
3271 printk(KERN_ERR
3272 "md: tried to install personality %s as nr %d, but max is %lu\n",
3273 p->name, pnum, MAX_PERSONALITY-1);
3274 return -EINVAL;
3275 }
3276
3277 spin_lock(&pers_lock);
3278 if (pers[pnum]) {
3279 spin_unlock(&pers_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 return -EBUSY;
3281 }
3282
3283 pers[pnum] = p;
3284 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3285 spin_unlock(&pers_lock);
3286 return 0;
3287}
3288
3289int unregister_md_personality(int pnum)
3290{
NeilBrowna757e642005-04-16 15:26:42 -07003291 if (pnum >= MAX_PERSONALITY)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293
3294 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3295 spin_lock(&pers_lock);
3296 pers[pnum] = NULL;
3297 spin_unlock(&pers_lock);
3298 return 0;
3299}
3300
3301static int is_mddev_idle(mddev_t *mddev)
3302{
3303 mdk_rdev_t * rdev;
3304 struct list_head *tmp;
3305 int idle;
3306 unsigned long curr_events;
3307
3308 idle = 1;
3309 ITERATE_RDEV(mddev,rdev,tmp) {
3310 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3311 curr_events = disk_stat_read(disk, read_sectors) +
3312 disk_stat_read(disk, write_sectors) -
3313 atomic_read(&disk->sync_io);
3314 /* Allow some slack between valud of curr_events and last_events,
3315 * as there are some uninteresting races.
3316 * Note: the following is an unsigned comparison.
3317 */
3318 if ((curr_events - rdev->last_events + 32) > 64) {
3319 rdev->last_events = curr_events;
3320 idle = 0;
3321 }
3322 }
3323 return idle;
3324}
3325
3326void md_done_sync(mddev_t *mddev, int blocks, int ok)
3327{
3328 /* another "blocks" (512byte) blocks have been synced */
3329 atomic_sub(blocks, &mddev->recovery_active);
3330 wake_up(&mddev->recovery_wait);
3331 if (!ok) {
3332 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3333 md_wakeup_thread(mddev->thread);
3334 // stop recovery, signal do_sync ....
3335 }
3336}
3337
3338
NeilBrown06d91a52005-06-21 17:17:12 -07003339/* md_write_start(mddev, bi)
3340 * If we need to update some array metadata (e.g. 'active' flag
3341 * in superblock) before writing, queue bi for later writing
3342 * and return 0, else return 1 and it will be written now
3343 */
3344int md_write_start(mddev_t *mddev, struct bio *bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003345{
NeilBrown06d91a52005-06-21 17:17:12 -07003346 if (bio_data_dir(bi) != WRITE)
3347 return 1;
3348
3349 atomic_inc(&mddev->writes_pending);
3350 spin_lock(&mddev->write_lock);
3351 if (mddev->in_sync == 0 && mddev->sb_dirty == 0) {
3352 spin_unlock(&mddev->write_lock);
3353 return 1;
3354 }
3355 bio_list_add(&mddev->write_list, bi);
3356
3357 if (mddev->in_sync) {
3358 mddev->in_sync = 0;
3359 mddev->sb_dirty = 1;
3360 }
3361 spin_unlock(&mddev->write_lock);
3362 md_wakeup_thread(mddev->thread);
3363 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364}
3365
3366void md_write_end(mddev_t *mddev)
3367{
3368 if (atomic_dec_and_test(&mddev->writes_pending)) {
3369 if (mddev->safemode == 2)
3370 md_wakeup_thread(mddev->thread);
3371 else
3372 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3373 }
3374}
3375
Adrian Bunk75c96f82005-05-05 16:16:09 -07003376static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377
3378#define SYNC_MARKS 10
3379#define SYNC_MARK_STEP (3*HZ)
3380static void md_do_sync(mddev_t *mddev)
3381{
3382 mddev_t *mddev2;
3383 unsigned int currspeed = 0,
3384 window;
NeilBrown57afd892005-06-21 17:17:13 -07003385 sector_t max_sectors,j, io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 unsigned long mark[SYNC_MARKS];
3387 sector_t mark_cnt[SYNC_MARKS];
3388 int last_mark,m;
3389 struct list_head *tmp;
3390 sector_t last_check;
NeilBrown57afd892005-06-21 17:17:13 -07003391 int skipped = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392
3393 /* just incase thread restarts... */
3394 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3395 return;
3396
3397 /* we overload curr_resync somewhat here.
3398 * 0 == not engaged in resync at all
3399 * 2 == checking that there is no conflict with another sync
3400 * 1 == like 2, but have yielded to allow conflicting resync to
3401 * commense
3402 * other == active in resync - this many blocks
3403 *
3404 * Before starting a resync we must have set curr_resync to
3405 * 2, and then checked that every "conflicting" array has curr_resync
3406 * less than ours. When we find one that is the same or higher
3407 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3408 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3409 * This will mean we have to start checking from the beginning again.
3410 *
3411 */
3412
3413 do {
3414 mddev->curr_resync = 2;
3415
3416 try_again:
3417 if (signal_pending(current)) {
3418 flush_signals(current);
3419 goto skip;
3420 }
3421 ITERATE_MDDEV(mddev2,tmp) {
3422 printk(".");
3423 if (mddev2 == mddev)
3424 continue;
3425 if (mddev2->curr_resync &&
3426 match_mddev_units(mddev,mddev2)) {
3427 DEFINE_WAIT(wq);
3428 if (mddev < mddev2 && mddev->curr_resync == 2) {
3429 /* arbitrarily yield */
3430 mddev->curr_resync = 1;
3431 wake_up(&resync_wait);
3432 }
3433 if (mddev > mddev2 && mddev->curr_resync == 1)
3434 /* no need to wait here, we can wait the next
3435 * time 'round when curr_resync == 2
3436 */
3437 continue;
3438 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3439 if (!signal_pending(current)
3440 && mddev2->curr_resync >= mddev->curr_resync) {
3441 printk(KERN_INFO "md: delaying resync of %s"
3442 " until %s has finished resync (they"
3443 " share one or more physical units)\n",
3444 mdname(mddev), mdname(mddev2));
3445 mddev_put(mddev2);
3446 schedule();
3447 finish_wait(&resync_wait, &wq);
3448 goto try_again;
3449 }
3450 finish_wait(&resync_wait, &wq);
3451 }
3452 }
3453 } while (mddev->curr_resync < 2);
3454
3455 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3456 /* resync follows the size requested by the personality,
NeilBrown57afd892005-06-21 17:17:13 -07003457 * which defaults to physical size, but can be virtual size
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458 */
3459 max_sectors = mddev->resync_max_sectors;
3460 else
3461 /* recovery follows the physical size of devices */
3462 max_sectors = mddev->size << 1;
3463
3464 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3465 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3466 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3467 printk(KERN_INFO "md: using maximum available idle IO bandwith "
3468 "(but not more than %d KB/sec) for reconstruction.\n",
3469 sysctl_speed_limit_max);
3470
3471 is_mddev_idle(mddev); /* this also initializes IO event counters */
NeilBrown32a76272005-06-21 17:17:14 -07003472 /* we don't use the checkpoint if there's a bitmap */
3473 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003474 j = mddev->recovery_cp;
3475 else
3476 j = 0;
NeilBrown57afd892005-06-21 17:17:13 -07003477 io_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003478 for (m = 0; m < SYNC_MARKS; m++) {
3479 mark[m] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07003480 mark_cnt[m] = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481 }
3482 last_mark = 0;
3483 mddev->resync_mark = mark[last_mark];
3484 mddev->resync_mark_cnt = mark_cnt[last_mark];
3485
3486 /*
3487 * Tune reconstruction:
3488 */
3489 window = 32*(PAGE_SIZE/512);
3490 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3491 window/2,(unsigned long long) max_sectors/2);
3492
3493 atomic_set(&mddev->recovery_active, 0);
3494 init_waitqueue_head(&mddev->recovery_wait);
3495 last_check = 0;
3496
3497 if (j>2) {
3498 printk(KERN_INFO
3499 "md: resuming recovery of %s from checkpoint.\n",
3500 mdname(mddev));
3501 mddev->curr_resync = j;
3502 }
3503
3504 while (j < max_sectors) {
NeilBrown57afd892005-06-21 17:17:13 -07003505 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506
NeilBrown57afd892005-06-21 17:17:13 -07003507 skipped = 0;
3508 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3509 currspeed < sysctl_speed_limit_min);
3510 if (sectors == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3512 goto out;
3513 }
NeilBrown57afd892005-06-21 17:17:13 -07003514
3515 if (!skipped) { /* actual IO requested */
3516 io_sectors += sectors;
3517 atomic_add(sectors, &mddev->recovery_active);
3518 }
3519
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520 j += sectors;
3521 if (j>1) mddev->curr_resync = j;
3522
NeilBrown57afd892005-06-21 17:17:13 -07003523
3524 if (last_check + window > io_sectors || j == max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 continue;
3526
NeilBrown57afd892005-06-21 17:17:13 -07003527 last_check = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528
3529 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3530 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3531 break;
3532
3533 repeat:
3534 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3535 /* step marks */
3536 int next = (last_mark+1) % SYNC_MARKS;
3537
3538 mddev->resync_mark = mark[next];
3539 mddev->resync_mark_cnt = mark_cnt[next];
3540 mark[next] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07003541 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 last_mark = next;
3543 }
3544
3545
3546 if (signal_pending(current)) {
3547 /*
3548 * got a signal, exit.
3549 */
3550 printk(KERN_INFO
3551 "md: md_do_sync() got signal ... exiting\n");
3552 flush_signals(current);
3553 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3554 goto out;
3555 }
3556
3557 /*
3558 * this loop exits only if either when we are slower than
3559 * the 'hard' speed limit, or the system was IO-idle for
3560 * a jiffy.
3561 * the system might be non-idle CPU-wise, but we only care
3562 * about not overloading the IO subsystem. (things like an
3563 * e2fsck being done on the RAID array should execute fast)
3564 */
3565 mddev->queue->unplug_fn(mddev->queue);
3566 cond_resched();
3567
NeilBrown57afd892005-06-21 17:17:13 -07003568 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3569 /((jiffies-mddev->resync_mark)/HZ +1) +1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003570
3571 if (currspeed > sysctl_speed_limit_min) {
3572 if ((currspeed > sysctl_speed_limit_max) ||
3573 !is_mddev_idle(mddev)) {
3574 msleep_interruptible(250);
3575 goto repeat;
3576 }
3577 }
3578 }
3579 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3580 /*
3581 * this also signals 'finished resyncing' to md_stop
3582 */
3583 out:
3584 mddev->queue->unplug_fn(mddev->queue);
3585
3586 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3587
3588 /* tell personality that we are finished */
NeilBrown57afd892005-06-21 17:17:13 -07003589 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003590
3591 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3592 mddev->curr_resync > 2 &&
3593 mddev->curr_resync >= mddev->recovery_cp) {
3594 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3595 printk(KERN_INFO
3596 "md: checkpointing recovery of %s.\n",
3597 mdname(mddev));
3598 mddev->recovery_cp = mddev->curr_resync;
3599 } else
3600 mddev->recovery_cp = MaxSector;
3601 }
3602
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 skip:
3604 mddev->curr_resync = 0;
3605 wake_up(&resync_wait);
3606 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3607 md_wakeup_thread(mddev->thread);
3608}
3609
3610
3611/*
3612 * This routine is regularly called by all per-raid-array threads to
3613 * deal with generic issues like resync and super-block update.
3614 * Raid personalities that don't have a thread (linear/raid0) do not
3615 * need this as they never do any recovery or update the superblock.
3616 *
3617 * It does not do any resync itself, but rather "forks" off other threads
3618 * to do that as needed.
3619 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3620 * "->recovery" and create a thread at ->sync_thread.
3621 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3622 * and wakeups up this thread which will reap the thread and finish up.
3623 * This thread also removes any faulty devices (with nr_pending == 0).
3624 *
3625 * The overall approach is:
3626 * 1/ if the superblock needs updating, update it.
3627 * 2/ If a recovery thread is running, don't do anything else.
3628 * 3/ If recovery has finished, clean up, possibly marking spares active.
3629 * 4/ If there are any faulty devices, remove them.
3630 * 5/ If array is degraded, try to add spares devices
3631 * 6/ If array has spares or is not in-sync, start a resync thread.
3632 */
3633void md_check_recovery(mddev_t *mddev)
3634{
3635 mdk_rdev_t *rdev;
3636 struct list_head *rtmp;
3637
3638
3639 dprintk(KERN_INFO "md: recovery thread got woken up ...\n");
3640
3641 if (mddev->ro)
3642 return;
NeilBrownfca4d842005-06-21 17:17:11 -07003643
3644 if (signal_pending(current)) {
3645 if (mddev->pers->sync_request) {
3646 printk(KERN_INFO "md: %s in immediate safe mode\n",
3647 mdname(mddev));
3648 mddev->safemode = 2;
3649 }
3650 flush_signals(current);
3651 }
3652
Linus Torvalds1da177e2005-04-16 15:20:36 -07003653 if ( ! (
3654 mddev->sb_dirty ||
3655 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
NeilBrownfca4d842005-06-21 17:17:11 -07003656 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
NeilBrown06d91a52005-06-21 17:17:12 -07003657 mddev->write_list.head ||
NeilBrownfca4d842005-06-21 17:17:11 -07003658 (mddev->safemode == 1) ||
3659 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3660 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 ))
3662 return;
NeilBrownfca4d842005-06-21 17:17:11 -07003663
Linus Torvalds1da177e2005-04-16 15:20:36 -07003664 if (mddev_trylock(mddev)==0) {
3665 int spares =0;
NeilBrown06d91a52005-06-21 17:17:12 -07003666 struct bio *blist;
NeilBrownfca4d842005-06-21 17:17:11 -07003667
NeilBrown06d91a52005-06-21 17:17:12 -07003668 spin_lock(&mddev->write_lock);
NeilBrownfca4d842005-06-21 17:17:11 -07003669 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3670 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3671 mddev->in_sync = 1;
3672 mddev->sb_dirty = 1;
3673 }
3674 if (mddev->safemode == 1)
3675 mddev->safemode = 0;
NeilBrown06d91a52005-06-21 17:17:12 -07003676 blist = bio_list_get(&mddev->write_list);
3677 spin_unlock(&mddev->write_lock);
NeilBrownfca4d842005-06-21 17:17:11 -07003678
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 if (mddev->sb_dirty)
3680 md_update_sb(mddev);
NeilBrown06d91a52005-06-21 17:17:12 -07003681
3682 while (blist) {
3683 struct bio *b = blist;
3684 blist = blist->bi_next;
3685 b->bi_next = NULL;
3686 generic_make_request(b);
3687 /* we already counted this, so need to un-count */
3688 md_write_end(mddev);
3689 }
3690
3691
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3693 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3694 /* resync/recovery still happening */
3695 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3696 goto unlock;
3697 }
3698 if (mddev->sync_thread) {
3699 /* resync has finished, collect result */
3700 md_unregister_thread(mddev->sync_thread);
3701 mddev->sync_thread = NULL;
3702 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3703 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3704 /* success...*/
3705 /* activate any spares */
3706 mddev->pers->spare_active(mddev);
3707 }
3708 md_update_sb(mddev);
3709 mddev->recovery = 0;
3710 /* flag recovery needed just to double check */
3711 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3712 goto unlock;
3713 }
3714 if (mddev->recovery)
3715 /* probably just the RECOVERY_NEEDED flag */
3716 mddev->recovery = 0;
3717
3718 /* no recovery is running.
3719 * remove any failed drives, then
3720 * add spares if possible.
3721 * Spare are also removed and re-added, to allow
3722 * the personality to fail the re-add.
3723 */
3724 ITERATE_RDEV(mddev,rdev,rtmp)
3725 if (rdev->raid_disk >= 0 &&
3726 (rdev->faulty || ! rdev->in_sync) &&
3727 atomic_read(&rdev->nr_pending)==0) {
3728 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3729 rdev->raid_disk = -1;
3730 }
3731
3732 if (mddev->degraded) {
3733 ITERATE_RDEV(mddev,rdev,rtmp)
3734 if (rdev->raid_disk < 0
3735 && !rdev->faulty) {
3736 if (mddev->pers->hot_add_disk(mddev,rdev))
3737 spares++;
3738 else
3739 break;
3740 }
3741 }
3742
3743 if (!spares && (mddev->recovery_cp == MaxSector )) {
3744 /* nothing we can do ... */
3745 goto unlock;
3746 }
3747 if (mddev->pers->sync_request) {
3748 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3749 if (!spares)
3750 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3751 mddev->sync_thread = md_register_thread(md_do_sync,
3752 mddev,
3753 "%s_resync");
3754 if (!mddev->sync_thread) {
3755 printk(KERN_ERR "%s: could not start resync"
3756 " thread...\n",
3757 mdname(mddev));
3758 /* leave the spares where they are, it shouldn't hurt */
3759 mddev->recovery = 0;
3760 } else {
3761 md_wakeup_thread(mddev->sync_thread);
3762 }
3763 }
3764 unlock:
3765 mddev_unlock(mddev);
3766 }
3767}
3768
Adrian Bunk75c96f82005-05-05 16:16:09 -07003769static int md_notify_reboot(struct notifier_block *this,
3770 unsigned long code, void *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003771{
3772 struct list_head *tmp;
3773 mddev_t *mddev;
3774
3775 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3776
3777 printk(KERN_INFO "md: stopping all md devices.\n");
3778
3779 ITERATE_MDDEV(mddev,tmp)
3780 if (mddev_trylock(mddev)==0)
3781 do_md_stop (mddev, 1);
3782 /*
3783 * certain more exotic SCSI devices are known to be
3784 * volatile wrt too early system reboots. While the
3785 * right place to handle this issue is the given
3786 * driver, we do want to have a safe RAID driver ...
3787 */
3788 mdelay(1000*1);
3789 }
3790 return NOTIFY_DONE;
3791}
3792
Adrian Bunk75c96f82005-05-05 16:16:09 -07003793static struct notifier_block md_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003794 .notifier_call = md_notify_reboot,
3795 .next = NULL,
3796 .priority = INT_MAX, /* before any real devices */
3797};
3798
3799static void md_geninit(void)
3800{
3801 struct proc_dir_entry *p;
3802
3803 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3804
3805 p = create_proc_entry("mdstat", S_IRUGO, NULL);
3806 if (p)
3807 p->proc_fops = &md_seq_fops;
3808}
3809
Adrian Bunk75c96f82005-05-05 16:16:09 -07003810static int __init md_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811{
3812 int minor;
3813
3814 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3815 " MD_SB_DISKS=%d\n",
3816 MD_MAJOR_VERSION, MD_MINOR_VERSION,
3817 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
NeilBrown32a76272005-06-21 17:17:14 -07003818 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3819 BITMAP_MINOR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003820
3821 if (register_blkdev(MAJOR_NR, "md"))
3822 return -1;
3823 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3824 unregister_blkdev(MAJOR_NR, "md");
3825 return -1;
3826 }
3827 devfs_mk_dir("md");
3828 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3829 md_probe, NULL, NULL);
3830 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3831 md_probe, NULL, NULL);
3832
3833 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3834 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3835 S_IFBLK|S_IRUSR|S_IWUSR,
3836 "md/%d", minor);
3837
3838 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3839 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3840 S_IFBLK|S_IRUSR|S_IWUSR,
3841 "md/mdp%d", minor);
3842
3843
3844 register_reboot_notifier(&md_notifier);
3845 raid_table_header = register_sysctl_table(raid_root_table, 1);
3846
3847 md_geninit();
3848 return (0);
3849}
3850
3851
3852#ifndef MODULE
3853
3854/*
3855 * Searches all registered partitions for autorun RAID arrays
3856 * at boot time.
3857 */
3858static dev_t detected_devices[128];
3859static int dev_cnt;
3860
3861void md_autodetect_dev(dev_t dev)
3862{
3863 if (dev_cnt >= 0 && dev_cnt < 127)
3864 detected_devices[dev_cnt++] = dev;
3865}
3866
3867
3868static void autostart_arrays(int part)
3869{
3870 mdk_rdev_t *rdev;
3871 int i;
3872
3873 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3874
3875 for (i = 0; i < dev_cnt; i++) {
3876 dev_t dev = detected_devices[i];
3877
3878 rdev = md_import_device(dev,0, 0);
3879 if (IS_ERR(rdev))
3880 continue;
3881
3882 if (rdev->faulty) {
3883 MD_BUG();
3884 continue;
3885 }
3886 list_add(&rdev->same_set, &pending_raid_disks);
3887 }
3888 dev_cnt = 0;
3889
3890 autorun_devices(part);
3891}
3892
3893#endif
3894
3895static __exit void md_exit(void)
3896{
3897 mddev_t *mddev;
3898 struct list_head *tmp;
3899 int i;
3900 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3901 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3902 for (i=0; i < MAX_MD_DEVS; i++)
3903 devfs_remove("md/%d", i);
3904 for (i=0; i < MAX_MD_DEVS; i++)
3905 devfs_remove("md/d%d", i);
3906
3907 devfs_remove("md");
3908
3909 unregister_blkdev(MAJOR_NR,"md");
3910 unregister_blkdev(mdp_major, "mdp");
3911 unregister_reboot_notifier(&md_notifier);
3912 unregister_sysctl_table(raid_table_header);
3913 remove_proc_entry("mdstat", NULL);
3914 ITERATE_MDDEV(mddev,tmp) {
3915 struct gendisk *disk = mddev->gendisk;
3916 if (!disk)
3917 continue;
3918 export_array(mddev);
3919 del_gendisk(disk);
3920 put_disk(disk);
3921 mddev->gendisk = NULL;
3922 mddev_put(mddev);
3923 }
3924}
3925
3926module_init(md_init)
3927module_exit(md_exit)
3928
3929EXPORT_SYMBOL(register_md_personality);
3930EXPORT_SYMBOL(unregister_md_personality);
3931EXPORT_SYMBOL(md_error);
3932EXPORT_SYMBOL(md_done_sync);
3933EXPORT_SYMBOL(md_write_start);
3934EXPORT_SYMBOL(md_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003935EXPORT_SYMBOL(md_register_thread);
3936EXPORT_SYMBOL(md_unregister_thread);
3937EXPORT_SYMBOL(md_wakeup_thread);
3938EXPORT_SYMBOL(md_print_devices);
3939EXPORT_SYMBOL(md_check_recovery);
3940MODULE_LICENSE("GPL");