?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] svcsock timestamp fix
[linux-2.6/kvm.git] / drivers / md / md.c
blob2a8a5696bf8aec3ad18647347500c9a519f43e17
1 /*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45
46 #include <linux/init.h>
47
48 #include <linux/file.h>
49
50 #ifdef CONFIG_KMOD
51 #include <linux/kmod.h>
52 #endif
53
54 #include <asm/unaligned.h>
55
56 #define MAJOR_NR MD_MAJOR
57 #define MD_DRIVER
58
59 /* 63 partitions with the alternate major number (mdp) */
60 #define MdpMinorShift 6
61
62 #define DEBUG 0
63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
64
65
66 #ifndef MODULE
67 static void autostart_arrays (int part);
68 #endif
69
70 static mdk_personality_t *pers[MAX_PERSONALITY];
71 static DEFINE_SPINLOCK(pers_lock);
72
73 /*
74 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
75 * is 1000 KB/sec, so the extra system load does not show up that much.
76 * Increase it if you want to have more _guaranteed_ speed. Note that
77 * the RAID driver will use the maximum available bandwidth if the IO
78 * subsystem is idle. There is also an 'absolute maximum' reconstruction
79 * speed limit - in case reconstruction slows down your system despite
80 * idle IO detection.
81 *
82 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
83 */
84
85 static int sysctl_speed_limit_min = 1000;
86 static int sysctl_speed_limit_max = 200000;
87
88 static struct ctl_table_header *raid_table_header;
89
90 static ctl_table raid_table[] = {
91 {
92 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
93 .procname = "speed_limit_min",
94 .data = &sysctl_speed_limit_min,
95 .maxlen = sizeof(int),
96 .mode = 0644,
97 .proc_handler = &proc_dointvec,
98 },
99 {
100 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
101 .procname = "speed_limit_max",
102 .data = &sysctl_speed_limit_max,
103 .maxlen = sizeof(int),
104 .mode = 0644,
105 .proc_handler = &proc_dointvec,
106 },
107 { .ctl_name = 0 }
108 };
109
110 static ctl_table raid_dir_table[] = {
111 {
112 .ctl_name = DEV_RAID,
113 .procname = "raid",
114 .maxlen = 0,
115 .mode = 0555,
116 .child = raid_table,
117 },
118 { .ctl_name = 0 }
119 };
120
121 static ctl_table raid_root_table[] = {
122 {
123 .ctl_name = CTL_DEV,
124 .procname = "dev",
125 .maxlen = 0,
126 .mode = 0555,
127 .child = raid_dir_table,
128 },
129 { .ctl_name = 0 }
130 };
131
132 static struct block_device_operations md_fops;
133
134 /*
135 * Enables to iterate over all existing md arrays
136 * all_mddevs_lock protects this list.
137 */
138 static LIST_HEAD(all_mddevs);
139 static DEFINE_SPINLOCK(all_mddevs_lock);
140
141
142 /*
143 * iterates through all used mddevs in the system.
144 * We take care to grab the all_mddevs_lock whenever navigating
145 * the list, and to always hold a refcount when unlocked.
146 * Any code which breaks out of this loop while own
147 * a reference to the current mddev and must mddev_put it.
148 */
149 #define ITERATE_MDDEV(mddev,tmp) \
150 \
151 for (({ spin_lock(&all_mddevs_lock); \
152 tmp = all_mddevs.next; \
153 mddev = NULL;}); \
154 ({ if (tmp != &all_mddevs) \
155 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
156 spin_unlock(&all_mddevs_lock); \
157 if (mddev) mddev_put(mddev); \
158 mddev = list_entry(tmp, mddev_t, all_mddevs); \
159 tmp != &all_mddevs;}); \
160 ({ spin_lock(&all_mddevs_lock); \
161 tmp = tmp->next;}) \
162 )
163
164
165 static int md_fail_request (request_queue_t *q, struct bio *bio)
166 {
167 bio_io_error(bio, bio->bi_size);
168 return 0;
169 }
170
171 static inline mddev_t *mddev_get(mddev_t *mddev)
172 {
173 atomic_inc(&mddev->active);
174 return mddev;
175 }
176
177 static void mddev_put(mddev_t *mddev)
178 {
179 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
180 return;
181 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
182 list_del(&mddev->all_mddevs);
183 blk_put_queue(mddev->queue);
184 kfree(mddev);
185 }
186 spin_unlock(&all_mddevs_lock);
187 }
188
189 static mddev_t * mddev_find(dev_t unit)
190 {
191 mddev_t *mddev, *new = NULL;
192
193 retry:
194 spin_lock(&all_mddevs_lock);
195 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
196 if (mddev->unit == unit) {
197 mddev_get(mddev);
198 spin_unlock(&all_mddevs_lock);
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);
227 spin_lock_init(&new->write_lock);
228 init_waitqueue_head(&new->sb_wait);
229
230 new->queue = blk_alloc_queue(GFP_KERNEL);
231 if (!new->queue) {
232 kfree(new);
233 return NULL;
234 }
235
236 blk_queue_make_request(new->queue, md_fail_request);
237
238 goto retry;
239 }
240
241 static inline int mddev_lock(mddev_t * mddev)
242 {
243 return down_interruptible(&mddev->reconfig_sem);
244 }
245
246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
247 {
248 down(&mddev->reconfig_sem);
249 }
250
251 static inline int mddev_trylock(mddev_t * mddev)
252 {
253 return down_trylock(&mddev->reconfig_sem);
254 }
255
256 static inline void mddev_unlock(mddev_t * mddev)
257 {
258 up(&mddev->reconfig_sem);
259
260 md_wakeup_thread(mddev->thread);
261 }
262
263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
264 {
265 mdk_rdev_t * rdev;
266 struct list_head *tmp;
267
268 ITERATE_RDEV(mddev,rdev,tmp) {
269 if (rdev->desc_nr == nr)
270 return rdev;
271 }
272 return NULL;
273 }
274
275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
276 {
277 struct list_head *tmp;
278 mdk_rdev_t *rdev;
279
280 ITERATE_RDEV(mddev,rdev,tmp) {
281 if (rdev->bdev->bd_dev == dev)
282 return rdev;
283 }
284 return NULL;
285 }
286
287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
288 {
289 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
290 return MD_NEW_SIZE_BLOCKS(size);
291 }
292
293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
294 {
295 sector_t size;
296
297 size = rdev->sb_offset;
298
299 if (chunk_size)
300 size &= ~((sector_t)chunk_size/1024 - 1);
301 return size;
302 }
303
304 static int alloc_disk_sb(mdk_rdev_t * rdev)
305 {
306 if (rdev->sb_page)
307 MD_BUG();
308
309 rdev->sb_page = alloc_page(GFP_KERNEL);
310 if (!rdev->sb_page) {
311 printk(KERN_ALERT "md: out of memory.\n");
312 return -EINVAL;
313 }
314
315 return 0;
316 }
317
318 static void free_disk_sb(mdk_rdev_t * rdev)
319 {
320 if (rdev->sb_page) {
321 page_cache_release(rdev->sb_page);
322 rdev->sb_loaded = 0;
323 rdev->sb_page = NULL;
324 rdev->sb_offset = 0;
325 rdev->size = 0;
326 }
327 }
328
329
330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
331 {
332 mdk_rdev_t *rdev = bio->bi_private;
333 if (bio->bi_size)
334 return 1;
335
336 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
337 md_error(rdev->mddev, rdev);
338
339 if (atomic_dec_and_test(&rdev->mddev->pending_writes))
340 wake_up(&rdev->mddev->sb_wait);
341 bio_put(bio);
342 return 0;
343 }
344
345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
346 sector_t sector, int size, struct page *page)
347 {
348 /* write first size bytes of page to sector of rdev
349 * Increment mddev->pending_writes before returning
350 * and decrement it on completion, waking up sb_wait
351 * if zero is reached.
352 * If an error occurred, call md_error
353 */
354 struct bio *bio = bio_alloc(GFP_NOIO, 1);
355
356 bio->bi_bdev = rdev->bdev;
357 bio->bi_sector = sector;
358 bio_add_page(bio, page, size, 0);
359 bio->bi_private = rdev;
360 bio->bi_end_io = super_written;
361 atomic_inc(&mddev->pending_writes);
362 submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
363 }
364
365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
366 {
367 if (bio->bi_size)
368 return 1;
369
370 complete((struct completion*)bio->bi_private);
371 return 0;
372 }
373
374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
375 struct page *page, int rw)
376 {
377 struct bio *bio = bio_alloc(GFP_NOIO, 1);
378 struct completion event;
379 int ret;
380
381 rw |= (1 << BIO_RW_SYNC);
382
383 bio->bi_bdev = bdev;
384 bio->bi_sector = sector;
385 bio_add_page(bio, page, size, 0);
386 init_completion(&event);
387 bio->bi_private = &event;
388 bio->bi_end_io = bi_complete;
389 submit_bio(rw, bio);
390 wait_for_completion(&event);
391
392 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
393 bio_put(bio);
394 return ret;
395 }
396
397 static int read_disk_sb(mdk_rdev_t * rdev, int size)
398 {
399 char b[BDEVNAME_SIZE];
400 if (!rdev->sb_page) {
401 MD_BUG();
402 return -EINVAL;
403 }
404 if (rdev->sb_loaded)
405 return 0;
406
407
408 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
409 goto fail;
410 rdev->sb_loaded = 1;
411 return 0;
412
413 fail:
414 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
415 bdevname(rdev->bdev,b));
416 return -EINVAL;
417 }
418
419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
420 {
421 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
422 (sb1->set_uuid1 == sb2->set_uuid1) &&
423 (sb1->set_uuid2 == sb2->set_uuid2) &&
424 (sb1->set_uuid3 == sb2->set_uuid3))
425
426 return 1;
427
428 return 0;
429 }
430
431
432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
433 {
434 int ret;
435 mdp_super_t *tmp1, *tmp2;
436
437 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
438 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
439
440 if (!tmp1 || !tmp2) {
441 ret = 0;
442 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
443 goto abort;
444 }
445
446 *tmp1 = *sb1;
447 *tmp2 = *sb2;
448
449 /*
450 * nr_disks is not constant
451 */
452 tmp1->nr_disks = 0;
453 tmp2->nr_disks = 0;
454
455 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
456 ret = 0;
457 else
458 ret = 1;
459
460 abort:
461 kfree(tmp1);
462 kfree(tmp2);
463 return ret;
464 }
465
466 static unsigned int calc_sb_csum(mdp_super_t * sb)
467 {
468 unsigned int disk_csum, csum;
469
470 disk_csum = sb->sb_csum;
471 sb->sb_csum = 0;
472 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
473 sb->sb_csum = disk_csum;
474 return csum;
475 }
476
477
478 /*
479 * Handle superblock details.
480 * We want to be able to handle multiple superblock formats
481 * so we have a common interface to them all, and an array of
482 * different handlers.
483 * We rely on user-space to write the initial superblock, and support
484 * reading and updating of superblocks.
485 * Interface methods are:
486 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
487 * loads and validates a superblock on dev.
488 * if refdev != NULL, compare superblocks on both devices
489 * Return:
490 * 0 - dev has a superblock that is compatible with refdev
491 * 1 - dev has a superblock that is compatible and newer than refdev
492 * so dev should be used as the refdev in future
493 * -EINVAL superblock incompatible or invalid
494 * -othererror e.g. -EIO
495 *
496 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
497 * Verify that dev is acceptable into mddev.
498 * The first time, mddev->raid_disks will be 0, and data from
499 * dev should be merged in. Subsequent calls check that dev
500 * is new enough. Return 0 or -EINVAL
501 *
502 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
503 * Update the superblock for rdev with data in mddev
504 * This does not write to disc.
505 *
506 */
507
508 struct super_type {
509 char *name;
510 struct module *owner;
511 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
512 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
513 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
514 };
515
516 /*
517 * load_super for 0.90.0
518 */
519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
520 {
521 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
522 mdp_super_t *sb;
523 int ret;
524 sector_t sb_offset;
525
526 /*
527 * Calculate the position of the superblock,
528 * it's at the end of the disk.
529 *
530 * It also happens to be a multiple of 4Kb.
531 */
532 sb_offset = calc_dev_sboffset(rdev->bdev);
533 rdev->sb_offset = sb_offset;
534
535 ret = read_disk_sb(rdev, MD_SB_BYTES);
536 if (ret) return ret;
537
538 ret = -EINVAL;
539
540 bdevname(rdev->bdev, b);
541 sb = (mdp_super_t*)page_address(rdev->sb_page);
542
543 if (sb->md_magic != MD_SB_MAGIC) {
544 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
545 b);
546 goto abort;
547 }
548
549 if (sb->major_version != 0 ||
550 sb->minor_version != 90) {
551 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
552 sb->major_version, sb->minor_version,
553 b);
554 goto abort;
555 }
556
557 if (sb->raid_disks <= 0)
558 goto abort;
559
560 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
561 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
562 b);
563 goto abort;
564 }
565
566 rdev->preferred_minor = sb->md_minor;
567 rdev->data_offset = 0;
568 rdev->sb_size = MD_SB_BYTES;
569
570 if (sb->level == LEVEL_MULTIPATH)
571 rdev->desc_nr = -1;
572 else
573 rdev->desc_nr = sb->this_disk.number;
574
575 if (refdev == 0)
576 ret = 1;
577 else {
578 __u64 ev1, ev2;
579 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
580 if (!uuid_equal(refsb, sb)) {
581 printk(KERN_WARNING "md: %s has different UUID to %s\n",
582 b, bdevname(refdev->bdev,b2));
583 goto abort;
584 }
585 if (!sb_equal(refsb, sb)) {
586 printk(KERN_WARNING "md: %s has same UUID"
587 " but different superblock to %s\n",
588 b, bdevname(refdev->bdev, b2));
589 goto abort;
590 }
591 ev1 = md_event(sb);
592 ev2 = md_event(refsb);
593 if (ev1 > ev2)
594 ret = 1;
595 else
596 ret = 0;
597 }
598 rdev->size = calc_dev_size(rdev, sb->chunk_size);
599
600 abort:
601 return ret;
602 }
603
604 /*
605 * validate_super for 0.90.0
606 */
607 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
608 {
609 mdp_disk_t *desc;
610 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
611
612 rdev->raid_disk = -1;
613 rdev->in_sync = 0;
614 if (mddev->raid_disks == 0) {
615 mddev->major_version = 0;
616 mddev->minor_version = sb->minor_version;
617 mddev->patch_version = sb->patch_version;
618 mddev->persistent = ! sb->not_persistent;
619 mddev->chunk_size = sb->chunk_size;
620 mddev->ctime = sb->ctime;
621 mddev->utime = sb->utime;
622 mddev->level = sb->level;
623 mddev->layout = sb->layout;
624 mddev->raid_disks = sb->raid_disks;
625 mddev->size = sb->size;
626 mddev->events = md_event(sb);
627 mddev->bitmap_offset = 0;
628 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
629
630 if (sb->state & (1<<MD_SB_CLEAN))
631 mddev->recovery_cp = MaxSector;
632 else {
633 if (sb->events_hi == sb->cp_events_hi &&
634 sb->events_lo == sb->cp_events_lo) {
635 mddev->recovery_cp = sb->recovery_cp;
636 } else
637 mddev->recovery_cp = 0;
638 }
639
640 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
641 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
642 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
643 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
644
645 mddev->max_disks = MD_SB_DISKS;
646
647 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
648 mddev->bitmap_file == NULL) {
649 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
650 /* FIXME use a better test */
651 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
652 return -EINVAL;
653 }
654 mddev->bitmap_offset = mddev->default_bitmap_offset;
655 }
656
657 } else if (mddev->pers == NULL) {
658 /* Insist on good event counter while assembling */
659 __u64 ev1 = md_event(sb);
660 ++ev1;
661 if (ev1 < mddev->events)
662 return -EINVAL;
663 } else if (mddev->bitmap) {
664 /* if adding to array with a bitmap, then we can accept an
665 * older device ... but not too old.
666 */
667 __u64 ev1 = md_event(sb);
668 if (ev1 < mddev->bitmap->events_cleared)
669 return 0;
670 } else /* just a hot-add of a new device, leave raid_disk at -1 */
671 return 0;
672
673 if (mddev->level != LEVEL_MULTIPATH) {
674 rdev->faulty = 0;
675 rdev->flags = 0;
676 desc = sb->disks + rdev->desc_nr;
677
678 if (desc->state & (1<<MD_DISK_FAULTY))
679 rdev->faulty = 1;
680 else if (desc->state & (1<<MD_DISK_SYNC) &&
681 desc->raid_disk < mddev->raid_disks) {
682 rdev->in_sync = 1;
683 rdev->raid_disk = desc->raid_disk;
684 }
685 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
686 set_bit(WriteMostly, &rdev->flags);
687 } else /* MULTIPATH are always insync */
688 rdev->in_sync = 1;
689 return 0;
690 }
691
692 /*
693 * sync_super for 0.90.0
694 */
695 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
696 {
697 mdp_super_t *sb;
698 struct list_head *tmp;
699 mdk_rdev_t *rdev2;
700 int next_spare = mddev->raid_disks;
701
702 /* make rdev->sb match mddev data..
703 *
704 * 1/ zero out disks
705 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
706 * 3/ any empty disks < next_spare become removed
707 *
708 * disks[0] gets initialised to REMOVED because
709 * we cannot be sure from other fields if it has
710 * been initialised or not.
711 */
712 int i;
713 int active=0, working=0,failed=0,spare=0,nr_disks=0;
714
715 rdev->sb_size = MD_SB_BYTES;
716
717 sb = (mdp_super_t*)page_address(rdev->sb_page);
718
719 memset(sb, 0, sizeof(*sb));
720
721 sb->md_magic = MD_SB_MAGIC;
722 sb->major_version = mddev->major_version;
723 sb->minor_version = mddev->minor_version;
724 sb->patch_version = mddev->patch_version;
725 sb->gvalid_words = 0; /* ignored */
726 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
727 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
728 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
729 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
730
731 sb->ctime = mddev->ctime;
732 sb->level = mddev->level;
733 sb->size = mddev->size;
734 sb->raid_disks = mddev->raid_disks;
735 sb->md_minor = mddev->md_minor;
736 sb->not_persistent = !mddev->persistent;
737 sb->utime = mddev->utime;
738 sb->state = 0;
739 sb->events_hi = (mddev->events>>32);
740 sb->events_lo = (u32)mddev->events;
741
742 if (mddev->in_sync)
743 {
744 sb->recovery_cp = mddev->recovery_cp;
745 sb->cp_events_hi = (mddev->events>>32);
746 sb->cp_events_lo = (u32)mddev->events;
747 if (mddev->recovery_cp == MaxSector)
748 sb->state = (1<< MD_SB_CLEAN);
749 } else
750 sb->recovery_cp = 0;
751
752 sb->layout = mddev->layout;
753 sb->chunk_size = mddev->chunk_size;
754
755 if (mddev->bitmap && mddev->bitmap_file == NULL)
756 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
757
758 sb->disks[0].state = (1<<MD_DISK_REMOVED);
759 ITERATE_RDEV(mddev,rdev2,tmp) {
760 mdp_disk_t *d;
761 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
762 rdev2->desc_nr = rdev2->raid_disk;
763 else
764 rdev2->desc_nr = next_spare++;
765 d = &sb->disks[rdev2->desc_nr];
766 nr_disks++;
767 d->number = rdev2->desc_nr;
768 d->major = MAJOR(rdev2->bdev->bd_dev);
769 d->minor = MINOR(rdev2->bdev->bd_dev);
770 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
771 d->raid_disk = rdev2->raid_disk;
772 else
773 d->raid_disk = rdev2->desc_nr; /* compatibility */
774 if (rdev2->faulty) {
775 d->state = (1<<MD_DISK_FAULTY);
776 failed++;
777 } else if (rdev2->in_sync) {
778 d->state = (1<<MD_DISK_ACTIVE);
779 d->state |= (1<<MD_DISK_SYNC);
780 active++;
781 working++;
782 } else {
783 d->state = 0;
784 spare++;
785 working++;
786 }
787 if (test_bit(WriteMostly, &rdev2->flags))
788 d->state |= (1<<MD_DISK_WRITEMOSTLY);
789 }
790
791 /* now set the "removed" and "faulty" bits on any missing devices */
792 for (i=0 ; i < mddev->raid_disks ; i++) {
793 mdp_disk_t *d = &sb->disks[i];
794 if (d->state == 0 && d->number == 0) {
795 d->number = i;
796 d->raid_disk = i;
797 d->state = (1<<MD_DISK_REMOVED);
798 d->state |= (1<<MD_DISK_FAULTY);
799 failed++;
800 }
801 }
802 sb->nr_disks = nr_disks;
803 sb->active_disks = active;
804 sb->working_disks = working;
805 sb->failed_disks = failed;
806 sb->spare_disks = spare;
807
808 sb->this_disk = sb->disks[rdev->desc_nr];
809 sb->sb_csum = calc_sb_csum(sb);
810 }
811
812 /*
813 * version 1 superblock
814 */
815
816 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
817 {
818 unsigned int disk_csum, csum;
819 unsigned long long newcsum;
820 int size = 256 + le32_to_cpu(sb->max_dev)*2;
821 unsigned int *isuper = (unsigned int*)sb;
822 int i;
823
824 disk_csum = sb->sb_csum;
825 sb->sb_csum = 0;
826 newcsum = 0;
827 for (i=0; size>=4; size -= 4 )
828 newcsum += le32_to_cpu(*isuper++);
829
830 if (size == 2)
831 newcsum += le16_to_cpu(*(unsigned short*) isuper);
832
833 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
834 sb->sb_csum = disk_csum;
835 return cpu_to_le32(csum);
836 }
837
838 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
839 {
840 struct mdp_superblock_1 *sb;
841 int ret;
842 sector_t sb_offset;
843 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
844 int bmask;
845
846 /*
847 * Calculate the position of the superblock.
848 * It is always aligned to a 4K boundary and
849 * depeding on minor_version, it can be:
850 * 0: At least 8K, but less than 12K, from end of device
851 * 1: At start of device
852 * 2: 4K from start of device.
853 */
854 switch(minor_version) {
855 case 0:
856 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
857 sb_offset -= 8*2;
858 sb_offset &= ~(sector_t)(4*2-1);
859 /* convert from sectors to K */
860 sb_offset /= 2;
861 break;
862 case 1:
863 sb_offset = 0;
864 break;
865 case 2:
866 sb_offset = 4;
867 break;
868 default:
869 return -EINVAL;
870 }
871 rdev->sb_offset = sb_offset;
872
873 /* superblock is rarely larger than 1K, but it can be larger,
874 * and it is safe to read 4k, so we do that
875 */
876 ret = read_disk_sb(rdev, 4096);
877 if (ret) return ret;
878
879
880 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
881
882 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
883 sb->major_version != cpu_to_le32(1) ||
884 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
885 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
886 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
887 return -EINVAL;
888
889 if (calc_sb_1_csum(sb) != sb->sb_csum) {
890 printk("md: invalid superblock checksum on %s\n",
891 bdevname(rdev->bdev,b));
892 return -EINVAL;
893 }
894 if (le64_to_cpu(sb->data_size) < 10) {
895 printk("md: data_size too small on %s\n",
896 bdevname(rdev->bdev,b));
897 return -EINVAL;
898 }
899 rdev->preferred_minor = 0xffff;
900 rdev->data_offset = le64_to_cpu(sb->data_offset);
901
902 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
903 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
904 if (rdev->sb_size & bmask)
905 rdev-> sb_size = (rdev->sb_size | bmask)+1;
906
907 if (refdev == 0)
908 return 1;
909 else {
910 __u64 ev1, ev2;
911 struct mdp_superblock_1 *refsb =
912 (struct mdp_superblock_1*)page_address(refdev->sb_page);
913
914 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
915 sb->level != refsb->level ||
916 sb->layout != refsb->layout ||
917 sb->chunksize != refsb->chunksize) {
918 printk(KERN_WARNING "md: %s has strangely different"
919 " superblock to %s\n",
920 bdevname(rdev->bdev,b),
921 bdevname(refdev->bdev,b2));
922 return -EINVAL;
923 }
924 ev1 = le64_to_cpu(sb->events);
925 ev2 = le64_to_cpu(refsb->events);
926
927 if (ev1 > ev2)
928 return 1;
929 }
930 if (minor_version)
931 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
932 else
933 rdev->size = rdev->sb_offset;
934 if (rdev->size < le64_to_cpu(sb->data_size)/2)
935 return -EINVAL;
936 rdev->size = le64_to_cpu(sb->data_size)/2;
937 if (le32_to_cpu(sb->chunksize))
938 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
939 return 0;
940 }
941
942 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
943 {
944 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
945
946 rdev->raid_disk = -1;
947 rdev->in_sync = 0;
948 if (mddev->raid_disks == 0) {
949 mddev->major_version = 1;
950 mddev->patch_version = 0;
951 mddev->persistent = 1;
952 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
953 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
954 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
955 mddev->level = le32_to_cpu(sb->level);
956 mddev->layout = le32_to_cpu(sb->layout);
957 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
958 mddev->size = le64_to_cpu(sb->size)/2;
959 mddev->events = le64_to_cpu(sb->events);
960 mddev->bitmap_offset = 0;
961 mddev->default_bitmap_offset = 0;
962 mddev->default_bitmap_offset = 1024;
963
964 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
965 memcpy(mddev->uuid, sb->set_uuid, 16);
966
967 mddev->max_disks = (4096-256)/2;
968
969 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
970 mddev->bitmap_file == NULL ) {
971 if (mddev->level != 1) {
972 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
973 return -EINVAL;
974 }
975 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
976 }
977 } else if (mddev->pers == NULL) {
978 /* Insist of good event counter while assembling */
979 __u64 ev1 = le64_to_cpu(sb->events);
980 ++ev1;
981 if (ev1 < mddev->events)
982 return -EINVAL;
983 } else if (mddev->bitmap) {
984 /* If adding to array with a bitmap, then we can accept an
985 * older device, but not too old.
986 */
987 __u64 ev1 = le64_to_cpu(sb->events);
988 if (ev1 < mddev->bitmap->events_cleared)
989 return 0;
990 } else /* just a hot-add of a new device, leave raid_disk at -1 */
991 return 0;
992
993 if (mddev->level != LEVEL_MULTIPATH) {
994 int role;
995 rdev->desc_nr = le32_to_cpu(sb->dev_number);
996 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
997 switch(role) {
998 case 0xffff: /* spare */
999 rdev->faulty = 0;
1000 break;
1001 case 0xfffe: /* faulty */
1002 rdev->faulty = 1;
1003 break;
1004 default:
1005 rdev->in_sync = 1;
1006 rdev->faulty = 0;
1007 rdev->raid_disk = role;
1008 break;
1009 }
1010 rdev->flags = 0;
1011 if (sb->devflags & WriteMostly1)
1012 set_bit(WriteMostly, &rdev->flags);
1013 } else /* MULTIPATH are always insync */
1014 rdev->in_sync = 1;
1015
1016 return 0;
1017 }
1018
1019 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1020 {
1021 struct mdp_superblock_1 *sb;
1022 struct list_head *tmp;
1023 mdk_rdev_t *rdev2;
1024 int max_dev, i;
1025 /* make rdev->sb match mddev and rdev data. */
1026
1027 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1028
1029 sb->feature_map = 0;
1030 sb->pad0 = 0;
1031 memset(sb->pad1, 0, sizeof(sb->pad1));
1032 memset(sb->pad2, 0, sizeof(sb->pad2));
1033 memset(sb->pad3, 0, sizeof(sb->pad3));
1034
1035 sb->utime = cpu_to_le64((__u64)mddev->utime);
1036 sb->events = cpu_to_le64(mddev->events);
1037 if (mddev->in_sync)
1038 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1039 else
1040 sb->resync_offset = cpu_to_le64(0);
1041
1042 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1043 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1044 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1045 }
1046
1047 max_dev = 0;
1048 ITERATE_RDEV(mddev,rdev2,tmp)
1049 if (rdev2->desc_nr+1 > max_dev)
1050 max_dev = rdev2->desc_nr+1;
1051
1052 sb->max_dev = cpu_to_le32(max_dev);
1053 for (i=0; i<max_dev;i++)
1054 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1055
1056 ITERATE_RDEV(mddev,rdev2,tmp) {
1057 i = rdev2->desc_nr;
1058 if (rdev2->faulty)
1059 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1060 else if (rdev2->in_sync)
1061 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1062 else
1063 sb->dev_roles[i] = cpu_to_le16(0xffff);
1064 }
1065
1066 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1067 sb->sb_csum = calc_sb_1_csum(sb);
1068 }
1069
1070
1071 static struct super_type super_types[] = {
1072 [0] = {
1073 .name = "0.90.0",
1074 .owner = THIS_MODULE,
1075 .load_super = super_90_load,
1076 .validate_super = super_90_validate,
1077 .sync_super = super_90_sync,
1078 },
1079 [1] = {
1080 .name = "md-1",
1081 .owner = THIS_MODULE,
1082 .load_super = super_1_load,
1083 .validate_super = super_1_validate,
1084 .sync_super = super_1_sync,
1085 },
1086 };
1087
1088 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1089 {
1090 struct list_head *tmp;
1091 mdk_rdev_t *rdev;
1092
1093 ITERATE_RDEV(mddev,rdev,tmp)
1094 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1095 return rdev;
1096
1097 return NULL;
1098 }
1099
1100 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1101 {
1102 struct list_head *tmp;
1103 mdk_rdev_t *rdev;
1104
1105 ITERATE_RDEV(mddev1,rdev,tmp)
1106 if (match_dev_unit(mddev2, rdev))
1107 return 1;
1108
1109 return 0;
1110 }
1111
1112 static LIST_HEAD(pending_raid_disks);
1113
1114 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1115 {
1116 mdk_rdev_t *same_pdev;
1117 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1118
1119 if (rdev->mddev) {
1120 MD_BUG();
1121 return -EINVAL;
1122 }
1123 same_pdev = match_dev_unit(mddev, rdev);
1124 if (same_pdev)
1125 printk(KERN_WARNING
1126 "%s: WARNING: %s appears to be on the same physical"
1127 " disk as %s. True\n protection against single-disk"
1128 " failure might be compromised.\n",
1129 mdname(mddev), bdevname(rdev->bdev,b),
1130 bdevname(same_pdev->bdev,b2));
1131
1132 /* Verify rdev->desc_nr is unique.
1133 * If it is -1, assign a free number, else
1134 * check number is not in use
1135 */
1136 if (rdev->desc_nr < 0) {
1137 int choice = 0;
1138 if (mddev->pers) choice = mddev->raid_disks;
1139 while (find_rdev_nr(mddev, choice))
1140 choice++;
1141 rdev->desc_nr = choice;
1142 } else {
1143 if (find_rdev_nr(mddev, rdev->desc_nr))
1144 return -EBUSY;
1145 }
1146
1147 list_add(&rdev->same_set, &mddev->disks);
1148 rdev->mddev = mddev;
1149 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1150 return 0;
1151 }
1152
1153 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1154 {
1155 char b[BDEVNAME_SIZE];
1156 if (!rdev->mddev) {
1157 MD_BUG();
1158 return;
1159 }
1160 list_del_init(&rdev->same_set);
1161 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1162 rdev->mddev = NULL;
1163 }
1164
1165 /*
1166 * prevent the device from being mounted, repartitioned or
1167 * otherwise reused by a RAID array (or any other kernel
1168 * subsystem), by bd_claiming the device.
1169 */
1170 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1171 {
1172 int err = 0;
1173 struct block_device *bdev;
1174 char b[BDEVNAME_SIZE];
1175
1176 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1177 if (IS_ERR(bdev)) {
1178 printk(KERN_ERR "md: could not open %s.\n",
1179 __bdevname(dev, b));
1180 return PTR_ERR(bdev);
1181 }
1182 err = bd_claim(bdev, rdev);
1183 if (err) {
1184 printk(KERN_ERR "md: could not bd_claim %s.\n",
1185 bdevname(bdev, b));
1186 blkdev_put(bdev);
1187 return err;
1188 }
1189 rdev->bdev = bdev;
1190 return err;
1191 }
1192
1193 static void unlock_rdev(mdk_rdev_t *rdev)
1194 {
1195 struct block_device *bdev = rdev->bdev;
1196 rdev->bdev = NULL;
1197 if (!bdev)
1198 MD_BUG();
1199 bd_release(bdev);
1200 blkdev_put(bdev);
1201 }
1202
1203 void md_autodetect_dev(dev_t dev);
1204
1205 static void export_rdev(mdk_rdev_t * rdev)
1206 {
1207 char b[BDEVNAME_SIZE];
1208 printk(KERN_INFO "md: export_rdev(%s)\n",
1209 bdevname(rdev->bdev,b));
1210 if (rdev->mddev)
1211 MD_BUG();
1212 free_disk_sb(rdev);
1213 list_del_init(&rdev->same_set);
1214 #ifndef MODULE
1215 md_autodetect_dev(rdev->bdev->bd_dev);
1216 #endif
1217 unlock_rdev(rdev);
1218 kfree(rdev);
1219 }
1220
1221 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1222 {
1223 unbind_rdev_from_array(rdev);
1224 export_rdev(rdev);
1225 }
1226
1227 static void export_array(mddev_t *mddev)
1228 {
1229 struct list_head *tmp;
1230 mdk_rdev_t *rdev;
1231
1232 ITERATE_RDEV(mddev,rdev,tmp) {
1233 if (!rdev->mddev) {
1234 MD_BUG();
1235 continue;
1236 }
1237 kick_rdev_from_array(rdev);
1238 }
1239 if (!list_empty(&mddev->disks))
1240 MD_BUG();
1241 mddev->raid_disks = 0;
1242 mddev->major_version = 0;
1243 }
1244
1245 static void print_desc(mdp_disk_t *desc)
1246 {
1247 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1248 desc->major,desc->minor,desc->raid_disk,desc->state);
1249 }
1250
1251 static void print_sb(mdp_super_t *sb)
1252 {
1253 int i;
1254
1255 printk(KERN_INFO
1256 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1257 sb->major_version, sb->minor_version, sb->patch_version,
1258 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1259 sb->ctime);
1260 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1261 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1262 sb->md_minor, sb->layout, sb->chunk_size);
1263 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1264 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1265 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1266 sb->failed_disks, sb->spare_disks,
1267 sb->sb_csum, (unsigned long)sb->events_lo);
1268
1269 printk(KERN_INFO);
1270 for (i = 0; i < MD_SB_DISKS; i++) {
1271 mdp_disk_t *desc;
1272
1273 desc = sb->disks + i;
1274 if (desc->number || desc->major || desc->minor ||
1275 desc->raid_disk || (desc->state && (desc->state != 4))) {
1276 printk(" D %2d: ", i);
1277 print_desc(desc);
1278 }
1279 }
1280 printk(KERN_INFO "md: THIS: ");
1281 print_desc(&sb->this_disk);
1282
1283 }
1284
1285 static void print_rdev(mdk_rdev_t *rdev)
1286 {
1287 char b[BDEVNAME_SIZE];
1288 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1289 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1290 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1291 if (rdev->sb_loaded) {
1292 printk(KERN_INFO "md: rdev superblock:\n");
1293 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1294 } else
1295 printk(KERN_INFO "md: no rdev superblock!\n");
1296 }
1297
1298 void md_print_devices(void)
1299 {
1300 struct list_head *tmp, *tmp2;
1301 mdk_rdev_t *rdev;
1302 mddev_t *mddev;
1303 char b[BDEVNAME_SIZE];
1304
1305 printk("\n");
1306 printk("md: **********************************\n");
1307 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1308 printk("md: **********************************\n");
1309 ITERATE_MDDEV(mddev,tmp) {
1310
1311 if (mddev->bitmap)
1312 bitmap_print_sb(mddev->bitmap);
1313 else
1314 printk("%s: ", mdname(mddev));
1315 ITERATE_RDEV(mddev,rdev,tmp2)
1316 printk("<%s>", bdevname(rdev->bdev,b));
1317 printk("\n");
1318
1319 ITERATE_RDEV(mddev,rdev,tmp2)
1320 print_rdev(rdev);
1321 }
1322 printk("md: **********************************\n");
1323 printk("\n");
1324 }
1325
1326
1327 static void sync_sbs(mddev_t * mddev)
1328 {
1329 mdk_rdev_t *rdev;
1330 struct list_head *tmp;
1331
1332 ITERATE_RDEV(mddev,rdev,tmp) {
1333 super_types[mddev->major_version].
1334 sync_super(mddev, rdev);
1335 rdev->sb_loaded = 1;
1336 }
1337 }
1338
1339 static void md_update_sb(mddev_t * mddev)
1340 {
1341 int err;
1342 struct list_head *tmp;
1343 mdk_rdev_t *rdev;
1344 int sync_req;
1345
1346 repeat:
1347 spin_lock(&mddev->write_lock);
1348 sync_req = mddev->in_sync;
1349 mddev->utime = get_seconds();
1350 mddev->events ++;
1351
1352 if (!mddev->events) {
1353 /*
1354 * oops, this 64-bit counter should never wrap.
1355 * Either we are in around ~1 trillion A.C., assuming
1356 * 1 reboot per second, or we have a bug:
1357 */
1358 MD_BUG();
1359 mddev->events --;
1360 }
1361 mddev->sb_dirty = 2;
1362 sync_sbs(mddev);
1363
1364 /*
1365 * do not write anything to disk if using
1366 * nonpersistent superblocks
1367 */
1368 if (!mddev->persistent) {
1369 mddev->sb_dirty = 0;
1370 spin_unlock(&mddev->write_lock);
1371 wake_up(&mddev->sb_wait);
1372 return;
1373 }
1374 spin_unlock(&mddev->write_lock);
1375
1376 dprintk(KERN_INFO
1377 "md: updating %s RAID superblock on device (in sync %d)\n",
1378 mdname(mddev),mddev->in_sync);
1379
1380 err = bitmap_update_sb(mddev->bitmap);
1381 ITERATE_RDEV(mddev,rdev,tmp) {
1382 char b[BDEVNAME_SIZE];
1383 dprintk(KERN_INFO "md: ");
1384 if (rdev->faulty)
1385 dprintk("(skipping faulty ");
1386
1387 dprintk("%s ", bdevname(rdev->bdev,b));
1388 if (!rdev->faulty) {
1389 md_super_write(mddev,rdev,
1390 rdev->sb_offset<<1, rdev->sb_size,
1391 rdev->sb_page);
1392 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1393 bdevname(rdev->bdev,b),
1394 (unsigned long long)rdev->sb_offset);
1395
1396 } else
1397 dprintk(")\n");
1398 if (mddev->level == LEVEL_MULTIPATH)
1399 /* only need to write one superblock... */
1400 break;
1401 }
1402 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1403 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1404
1405 spin_lock(&mddev->write_lock);
1406 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1407 /* have to write it out again */
1408 spin_unlock(&mddev->write_lock);
1409 goto repeat;
1410 }
1411 mddev->sb_dirty = 0;
1412 spin_unlock(&mddev->write_lock);
1413 wake_up(&mddev->sb_wait);
1414
1415 }
1416
1417 /*
1418 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1419 *
1420 * mark the device faulty if:
1421 *
1422 * - the device is nonexistent (zero size)
1423 * - the device has no valid superblock
1424 *
1425 * a faulty rdev _never_ has rdev->sb set.
1426 */
1427 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1428 {
1429 char b[BDEVNAME_SIZE];
1430 int err;
1431 mdk_rdev_t *rdev;
1432 sector_t size;
1433
1434 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1435 if (!rdev) {
1436 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1437 return ERR_PTR(-ENOMEM);
1438 }
1439 memset(rdev, 0, sizeof(*rdev));
1440
1441 if ((err = alloc_disk_sb(rdev)))
1442 goto abort_free;
1443
1444 err = lock_rdev(rdev, newdev);
1445 if (err)
1446 goto abort_free;
1447
1448 rdev->desc_nr = -1;
1449 rdev->faulty = 0;
1450 rdev->in_sync = 0;
1451 rdev->data_offset = 0;
1452 atomic_set(&rdev->nr_pending, 0);
1453
1454 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1455 if (!size) {
1456 printk(KERN_WARNING
1457 "md: %s has zero or unknown size, marking faulty!\n",
1458 bdevname(rdev->bdev,b));
1459 err = -EINVAL;
1460 goto abort_free;
1461 }
1462
1463 if (super_format >= 0) {
1464 err = super_types[super_format].
1465 load_super(rdev, NULL, super_minor);
1466 if (err == -EINVAL) {
1467 printk(KERN_WARNING
1468 "md: %s has invalid sb, not importing!\n",
1469 bdevname(rdev->bdev,b));
1470 goto abort_free;
1471 }
1472 if (err < 0) {
1473 printk(KERN_WARNING
1474 "md: could not read %s's sb, not importing!\n",
1475 bdevname(rdev->bdev,b));
1476 goto abort_free;
1477 }
1478 }
1479 INIT_LIST_HEAD(&rdev->same_set);
1480
1481 return rdev;
1482
1483 abort_free:
1484 if (rdev->sb_page) {
1485 if (rdev->bdev)
1486 unlock_rdev(rdev);
1487 free_disk_sb(rdev);
1488 }
1489 kfree(rdev);
1490 return ERR_PTR(err);
1491 }
1492
1493 /*
1494 * Check a full RAID array for plausibility
1495 */
1496
1497
1498 static void analyze_sbs(mddev_t * mddev)
1499 {
1500 int i;
1501 struct list_head *tmp;
1502 mdk_rdev_t *rdev, *freshest;
1503 char b[BDEVNAME_SIZE];
1504
1505 freshest = NULL;
1506 ITERATE_RDEV(mddev,rdev,tmp)
1507 switch (super_types[mddev->major_version].
1508 load_super(rdev, freshest, mddev->minor_version)) {
1509 case 1:
1510 freshest = rdev;
1511 break;
1512 case 0:
1513 break;
1514 default:
1515 printk( KERN_ERR \
1516 "md: fatal superblock inconsistency in %s"
1517 " -- removing from array\n",
1518 bdevname(rdev->bdev,b));
1519 kick_rdev_from_array(rdev);
1520 }
1521
1522
1523 super_types[mddev->major_version].
1524 validate_super(mddev, freshest);
1525
1526 i = 0;
1527 ITERATE_RDEV(mddev,rdev,tmp) {
1528 if (rdev != freshest)
1529 if (super_types[mddev->major_version].
1530 validate_super(mddev, rdev)) {
1531 printk(KERN_WARNING "md: kicking non-fresh %s"
1532 " from array!\n",
1533 bdevname(rdev->bdev,b));
1534 kick_rdev_from_array(rdev);
1535 continue;
1536 }
1537 if (mddev->level == LEVEL_MULTIPATH) {
1538 rdev->desc_nr = i++;
1539 rdev->raid_disk = rdev->desc_nr;
1540 rdev->in_sync = 1;
1541 }
1542 }
1543
1544
1545
1546 if (mddev->recovery_cp != MaxSector &&
1547 mddev->level >= 1)
1548 printk(KERN_ERR "md: %s: raid array is not clean"
1549 " -- starting background reconstruction\n",
1550 mdname(mddev));
1551
1552 }
1553
1554 int mdp_major = 0;
1555
1556 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1557 {
1558 static DECLARE_MUTEX(disks_sem);
1559 mddev_t *mddev = mddev_find(dev);
1560 struct gendisk *disk;
1561 int partitioned = (MAJOR(dev) != MD_MAJOR);
1562 int shift = partitioned ? MdpMinorShift : 0;
1563 int unit = MINOR(dev) >> shift;
1564
1565 if (!mddev)
1566 return NULL;
1567
1568 down(&disks_sem);
1569 if (mddev->gendisk) {
1570 up(&disks_sem);
1571 mddev_put(mddev);
1572 return NULL;
1573 }
1574 disk = alloc_disk(1 << shift);
1575 if (!disk) {
1576 up(&disks_sem);
1577 mddev_put(mddev);
1578 return NULL;
1579 }
1580 disk->major = MAJOR(dev);
1581 disk->first_minor = unit << shift;
1582 if (partitioned) {
1583 sprintf(disk->disk_name, "md_d%d", unit);
1584 sprintf(disk->devfs_name, "md/d%d", unit);
1585 } else {
1586 sprintf(disk->disk_name, "md%d", unit);
1587 sprintf(disk->devfs_name, "md/%d", unit);
1588 }
1589 disk->fops = &md_fops;
1590 disk->private_data = mddev;
1591 disk->queue = mddev->queue;
1592 add_disk(disk);
1593 mddev->gendisk = disk;
1594 up(&disks_sem);
1595 return NULL;
1596 }
1597
1598 void md_wakeup_thread(mdk_thread_t *thread);
1599
1600 static void md_safemode_timeout(unsigned long data)
1601 {
1602 mddev_t *mddev = (mddev_t *) data;
1603
1604 mddev->safemode = 1;
1605 md_wakeup_thread(mddev->thread);
1606 }
1607
1608
1609 static int do_md_run(mddev_t * mddev)
1610 {
1611 int pnum, err;
1612 int chunk_size;
1613 struct list_head *tmp;
1614 mdk_rdev_t *rdev;
1615 struct gendisk *disk;
1616 char b[BDEVNAME_SIZE];
1617
1618 if (list_empty(&mddev->disks))
1619 /* cannot run an array with no devices.. */
1620 return -EINVAL;
1621
1622 if (mddev->pers)
1623 return -EBUSY;
1624
1625 /*
1626 * Analyze all RAID superblock(s)
1627 */
1628 if (!mddev->raid_disks)
1629 analyze_sbs(mddev);
1630
1631 chunk_size = mddev->chunk_size;
1632 pnum = level_to_pers(mddev->level);
1633
1634 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1635 if (!chunk_size) {
1636 /*
1637 * 'default chunksize' in the old md code used to
1638 * be PAGE_SIZE, baaad.
1639 * we abort here to be on the safe side. We don't
1640 * want to continue the bad practice.
1641 */
1642 printk(KERN_ERR
1643 "no chunksize specified, see 'man raidtab'\n");
1644 return -EINVAL;
1645 }
1646 if (chunk_size > MAX_CHUNK_SIZE) {
1647 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1648 chunk_size, MAX_CHUNK_SIZE);
1649 return -EINVAL;
1650 }
1651 /*
1652 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1653 */
1654 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1655 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1656 return -EINVAL;
1657 }
1658 if (chunk_size < PAGE_SIZE) {
1659 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1660 chunk_size, PAGE_SIZE);
1661 return -EINVAL;
1662 }
1663
1664 /* devices must have minimum size of one chunk */
1665 ITERATE_RDEV(mddev,rdev,tmp) {
1666 if (rdev->faulty)
1667 continue;
1668 if (rdev->size < chunk_size / 1024) {
1669 printk(KERN_WARNING
1670 "md: Dev %s smaller than chunk_size:"
1671 " %lluk < %dk\n",
1672 bdevname(rdev->bdev,b),
1673 (unsigned long long)rdev->size,
1674 chunk_size / 1024);
1675 return -EINVAL;
1676 }
1677 }
1678 }
1679
1680 #ifdef CONFIG_KMOD
1681 if (!pers[pnum])
1682 {
1683 request_module("md-personality-%d", pnum);
1684 }
1685 #endif
1686
1687 /*
1688 * Drop all container device buffers, from now on
1689 * the only valid external interface is through the md
1690 * device.
1691 * Also find largest hardsector size
1692 */
1693 ITERATE_RDEV(mddev,rdev,tmp) {
1694 if (rdev->faulty)
1695 continue;
1696 sync_blockdev(rdev->bdev);
1697 invalidate_bdev(rdev->bdev, 0);
1698 }
1699
1700 md_probe(mddev->unit, NULL, NULL);
1701 disk = mddev->gendisk;
1702 if (!disk)
1703 return -ENOMEM;
1704
1705 spin_lock(&pers_lock);
1706 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1707 spin_unlock(&pers_lock);
1708 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1709 pnum);
1710 return -EINVAL;
1711 }
1712
1713 mddev->pers = pers[pnum];
1714 spin_unlock(&pers_lock);
1715
1716 mddev->recovery = 0;
1717 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1718
1719 /* before we start the array running, initialise the bitmap */
1720 err = bitmap_create(mddev);
1721 if (err)
1722 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1723 mdname(mddev), err);
1724 else
1725 err = mddev->pers->run(mddev);
1726 if (err) {
1727 printk(KERN_ERR "md: pers->run() failed ...\n");
1728 module_put(mddev->pers->owner);
1729 mddev->pers = NULL;
1730 bitmap_destroy(mddev);
1731 return err;
1732 }
1733 atomic_set(&mddev->writes_pending,0);
1734 mddev->safemode = 0;
1735 mddev->safemode_timer.function = md_safemode_timeout;
1736 mddev->safemode_timer.data = (unsigned long) mddev;
1737 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1738 mddev->in_sync = 1;
1739
1740 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1741 md_wakeup_thread(mddev->thread);
1742
1743 if (mddev->sb_dirty)
1744 md_update_sb(mddev);
1745
1746 set_capacity(disk, mddev->array_size<<1);
1747
1748 /* If we call blk_queue_make_request here, it will
1749 * re-initialise max_sectors etc which may have been
1750 * refined inside -> run. So just set the bits we need to set.
1751 * Most initialisation happended when we called
1752 * blk_queue_make_request(..., md_fail_request)
1753 * earlier.
1754 */
1755 mddev->queue->queuedata = mddev;
1756 mddev->queue->make_request_fn = mddev->pers->make_request;
1757
1758 mddev->changed = 1;
1759 return 0;
1760 }
1761
1762 static int restart_array(mddev_t *mddev)
1763 {
1764 struct gendisk *disk = mddev->gendisk;
1765 int err;
1766
1767 /*
1768 * Complain if it has no devices
1769 */
1770 err = -ENXIO;
1771 if (list_empty(&mddev->disks))
1772 goto out;
1773
1774 if (mddev->pers) {
1775 err = -EBUSY;
1776 if (!mddev->ro)
1777 goto out;
1778
1779 mddev->safemode = 0;
1780 mddev->ro = 0;
1781 set_disk_ro(disk, 0);
1782
1783 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1784 mdname(mddev));
1785 /*
1786 * Kick recovery or resync if necessary
1787 */
1788 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1789 md_wakeup_thread(mddev->thread);
1790 err = 0;
1791 } else {
1792 printk(KERN_ERR "md: %s has no personality assigned.\n",
1793 mdname(mddev));
1794 err = -EINVAL;
1795 }
1796
1797 out:
1798 return err;
1799 }
1800
1801 static int do_md_stop(mddev_t * mddev, int ro)
1802 {
1803 int err = 0;
1804 struct gendisk *disk = mddev->gendisk;
1805
1806 if (mddev->pers) {
1807 if (atomic_read(&mddev->active)>2) {
1808 printk("md: %s still in use.\n",mdname(mddev));
1809 return -EBUSY;
1810 }
1811
1812 if (mddev->sync_thread) {
1813 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1814 md_unregister_thread(mddev->sync_thread);
1815 mddev->sync_thread = NULL;
1816 }
1817
1818 del_timer_sync(&mddev->safemode_timer);
1819
1820 invalidate_partition(disk, 0);
1821
1822 if (ro) {
1823 err = -ENXIO;
1824 if (mddev->ro)
1825 goto out;
1826 mddev->ro = 1;
1827 } else {
1828 bitmap_flush(mddev);
1829 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1830 if (mddev->ro)
1831 set_disk_ro(disk, 0);
1832 blk_queue_make_request(mddev->queue, md_fail_request);
1833 mddev->pers->stop(mddev);
1834 module_put(mddev->pers->owner);
1835 mddev->pers = NULL;
1836 if (mddev->ro)
1837 mddev->ro = 0;
1838 }
1839 if (!mddev->in_sync) {
1840 /* mark array as shutdown cleanly */
1841 mddev->in_sync = 1;
1842 md_update_sb(mddev);
1843 }
1844 if (ro)
1845 set_disk_ro(disk, 1);
1846 }
1847
1848 bitmap_destroy(mddev);
1849 if (mddev->bitmap_file) {
1850 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1851 fput(mddev->bitmap_file);
1852 mddev->bitmap_file = NULL;
1853 }
1854 mddev->bitmap_offset = 0;
1855
1856 /*
1857 * Free resources if final stop
1858 */
1859 if (!ro) {
1860 struct gendisk *disk;
1861 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1862
1863 export_array(mddev);
1864
1865 mddev->array_size = 0;
1866 disk = mddev->gendisk;
1867 if (disk)
1868 set_capacity(disk, 0);
1869 mddev->changed = 1;
1870 } else
1871 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1872 mdname(mddev));
1873 err = 0;
1874 out:
1875 return err;
1876 }
1877
1878 static void autorun_array(mddev_t *mddev)
1879 {
1880 mdk_rdev_t *rdev;
1881 struct list_head *tmp;
1882 int err;
1883
1884 if (list_empty(&mddev->disks))
1885 return;
1886
1887 printk(KERN_INFO "md: running: ");
1888
1889 ITERATE_RDEV(mddev,rdev,tmp) {
1890 char b[BDEVNAME_SIZE];
1891 printk("<%s>", bdevname(rdev->bdev,b));
1892 }
1893 printk("\n");
1894
1895 err = do_md_run (mddev);
1896 if (err) {
1897 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1898 do_md_stop (mddev, 0);
1899 }
1900 }
1901
1902 /*
1903 * lets try to run arrays based on all disks that have arrived
1904 * until now. (those are in pending_raid_disks)
1905 *
1906 * the method: pick the first pending disk, collect all disks with
1907 * the same UUID, remove all from the pending list and put them into
1908 * the 'same_array' list. Then order this list based on superblock
1909 * update time (freshest comes first), kick out 'old' disks and
1910 * compare superblocks. If everything's fine then run it.
1911 *
1912 * If "unit" is allocated, then bump its reference count
1913 */
1914 static void autorun_devices(int part)
1915 {
1916 struct list_head candidates;
1917 struct list_head *tmp;
1918 mdk_rdev_t *rdev0, *rdev;
1919 mddev_t *mddev;
1920 char b[BDEVNAME_SIZE];
1921
1922 printk(KERN_INFO "md: autorun ...\n");
1923 while (!list_empty(&pending_raid_disks)) {
1924 dev_t dev;
1925 rdev0 = list_entry(pending_raid_disks.next,
1926 mdk_rdev_t, same_set);
1927
1928 printk(KERN_INFO "md: considering %s ...\n",
1929 bdevname(rdev0->bdev,b));
1930 INIT_LIST_HEAD(&candidates);
1931 ITERATE_RDEV_PENDING(rdev,tmp)
1932 if (super_90_load(rdev, rdev0, 0) >= 0) {
1933 printk(KERN_INFO "md: adding %s ...\n",
1934 bdevname(rdev->bdev,b));
1935 list_move(&rdev->same_set, &candidates);
1936 }
1937 /*
1938 * now we have a set of devices, with all of them having
1939 * mostly sane superblocks. It's time to allocate the
1940 * mddev.
1941 */
1942 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1943 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1944 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1945 break;
1946 }
1947 if (part)
1948 dev = MKDEV(mdp_major,
1949 rdev0->preferred_minor << MdpMinorShift);
1950 else
1951 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1952
1953 md_probe(dev, NULL, NULL);
1954 mddev = mddev_find(dev);
1955 if (!mddev) {
1956 printk(KERN_ERR
1957 "md: cannot allocate memory for md drive.\n");
1958 break;
1959 }
1960 if (mddev_lock(mddev))
1961 printk(KERN_WARNING "md: %s locked, cannot run\n",
1962 mdname(mddev));
1963 else if (mddev->raid_disks || mddev->major_version
1964 || !list_empty(&mddev->disks)) {
1965 printk(KERN_WARNING
1966 "md: %s already running, cannot run %s\n",
1967 mdname(mddev), bdevname(rdev0->bdev,b));
1968 mddev_unlock(mddev);
1969 } else {
1970 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1971 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1972 list_del_init(&rdev->same_set);
1973 if (bind_rdev_to_array(rdev, mddev))
1974 export_rdev(rdev);
1975 }
1976 autorun_array(mddev);
1977 mddev_unlock(mddev);
1978 }
1979 /* on success, candidates will be empty, on error
1980 * it won't...
1981 */
1982 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1983 export_rdev(rdev);
1984 mddev_put(mddev);
1985 }
1986 printk(KERN_INFO "md: ... autorun DONE.\n");
1987 }
1988
1989 /*
1990 * import RAID devices based on one partition
1991 * if possible, the array gets run as well.
1992 */
1993
1994 static int autostart_array(dev_t startdev)
1995 {
1996 char b[BDEVNAME_SIZE];
1997 int err = -EINVAL, i;
1998 mdp_super_t *sb = NULL;
1999 mdk_rdev_t *start_rdev = NULL, *rdev;
2000
2001 start_rdev = md_import_device(startdev, 0, 0);
2002 if (IS_ERR(start_rdev))
2003 return err;
2004
2005
2006 /* NOTE: this can only work for 0.90.0 superblocks */
2007 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2008 if (sb->major_version != 0 ||
2009 sb->minor_version != 90 ) {
2010 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2011 export_rdev(start_rdev);
2012 return err;
2013 }
2014
2015 if (start_rdev->faulty) {
2016 printk(KERN_WARNING
2017 "md: can not autostart based on faulty %s!\n",
2018 bdevname(start_rdev->bdev,b));
2019 export_rdev(start_rdev);
2020 return err;
2021 }
2022 list_add(&start_rdev->same_set, &pending_raid_disks);
2023
2024 for (i = 0; i < MD_SB_DISKS; i++) {
2025 mdp_disk_t *desc = sb->disks + i;
2026 dev_t dev = MKDEV(desc->major, desc->minor);
2027
2028 if (!dev)
2029 continue;
2030 if (dev == startdev)
2031 continue;
2032 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2033 continue;
2034 rdev = md_import_device(dev, 0, 0);
2035 if (IS_ERR(rdev))
2036 continue;
2037
2038 list_add(&rdev->same_set, &pending_raid_disks);
2039 }
2040
2041 /*
2042 * possibly return codes
2043 */
2044 autorun_devices(0);
2045 return 0;
2046
2047 }
2048
2049
2050 static int get_version(void __user * arg)
2051 {
2052 mdu_version_t ver;
2053
2054 ver.major = MD_MAJOR_VERSION;
2055 ver.minor = MD_MINOR_VERSION;
2056 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2057
2058 if (copy_to_user(arg, &ver, sizeof(ver)))
2059 return -EFAULT;
2060
2061 return 0;
2062 }
2063
2064 static int get_array_info(mddev_t * mddev, void __user * arg)
2065 {
2066 mdu_array_info_t info;
2067 int nr,working,active,failed,spare;
2068 mdk_rdev_t *rdev;
2069 struct list_head *tmp;
2070
2071 nr=working=active=failed=spare=0;
2072 ITERATE_RDEV(mddev,rdev,tmp) {
2073 nr++;
2074 if (rdev->faulty)
2075 failed++;
2076 else {
2077 working++;
2078 if (rdev->in_sync)
2079 active++;
2080 else
2081 spare++;
2082 }
2083 }
2084
2085 info.major_version = mddev->major_version;
2086 info.minor_version = mddev->minor_version;
2087 info.patch_version = MD_PATCHLEVEL_VERSION;
2088 info.ctime = mddev->ctime;
2089 info.level = mddev->level;
2090 info.size = mddev->size;
2091 info.nr_disks = nr;
2092 info.raid_disks = mddev->raid_disks;
2093 info.md_minor = mddev->md_minor;
2094 info.not_persistent= !mddev->persistent;
2095
2096 info.utime = mddev->utime;
2097 info.state = 0;
2098 if (mddev->in_sync)
2099 info.state = (1<<MD_SB_CLEAN);
2100 if (mddev->bitmap && mddev->bitmap_offset)
2101 info.state = (1<<MD_SB_BITMAP_PRESENT);
2102 info.active_disks = active;
2103 info.working_disks = working;
2104 info.failed_disks = failed;
2105 info.spare_disks = spare;
2106
2107 info.layout = mddev->layout;
2108 info.chunk_size = mddev->chunk_size;
2109
2110 if (copy_to_user(arg, &info, sizeof(info)))
2111 return -EFAULT;
2112
2113 return 0;
2114 }
2115
2116 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2117 {
2118 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2119 char *ptr, *buf = NULL;
2120 int err = -ENOMEM;
2121
2122 file = kmalloc(sizeof(*file), GFP_KERNEL);
2123 if (!file)
2124 goto out;
2125
2126 /* bitmap disabled, zero the first byte and copy out */
2127 if (!mddev->bitmap || !mddev->bitmap->file) {
2128 file->pathname[0] = '\0';
2129 goto copy_out;
2130 }
2131
2132 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2133 if (!buf)
2134 goto out;
2135
2136 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2137 if (!ptr)
2138 goto out;
2139
2140 strcpy(file->pathname, ptr);
2141
2142 copy_out:
2143 err = 0;
2144 if (copy_to_user(arg, file, sizeof(*file)))
2145 err = -EFAULT;
2146 out:
2147 kfree(buf);
2148 kfree(file);
2149 return err;
2150 }
2151
2152 static int get_disk_info(mddev_t * mddev, void __user * arg)
2153 {
2154 mdu_disk_info_t info;
2155 unsigned int nr;
2156 mdk_rdev_t *rdev;
2157
2158 if (copy_from_user(&info, arg, sizeof(info)))
2159 return -EFAULT;
2160
2161 nr = info.number;
2162
2163 rdev = find_rdev_nr(mddev, nr);
2164 if (rdev) {
2165 info.major = MAJOR(rdev->bdev->bd_dev);
2166 info.minor = MINOR(rdev->bdev->bd_dev);
2167 info.raid_disk = rdev->raid_disk;
2168 info.state = 0;
2169 if (rdev->faulty)
2170 info.state |= (1<<MD_DISK_FAULTY);
2171 else if (rdev->in_sync) {
2172 info.state |= (1<<MD_DISK_ACTIVE);
2173 info.state |= (1<<MD_DISK_SYNC);
2174 }
2175 if (test_bit(WriteMostly, &rdev->flags))
2176 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2177 } else {
2178 info.major = info.minor = 0;
2179 info.raid_disk = -1;
2180 info.state = (1<<MD_DISK_REMOVED);
2181 }
2182
2183 if (copy_to_user(arg, &info, sizeof(info)))
2184 return -EFAULT;
2185
2186 return 0;
2187 }
2188
2189 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2190 {
2191 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2192 mdk_rdev_t *rdev;
2193 dev_t dev = MKDEV(info->major,info->minor);
2194
2195 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2196 return -EOVERFLOW;
2197
2198 if (!mddev->raid_disks) {
2199 int err;
2200 /* expecting a device which has a superblock */
2201 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2202 if (IS_ERR(rdev)) {
2203 printk(KERN_WARNING
2204 "md: md_import_device returned %ld\n",
2205 PTR_ERR(rdev));
2206 return PTR_ERR(rdev);
2207 }
2208 if (!list_empty(&mddev->disks)) {
2209 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2210 mdk_rdev_t, same_set);
2211 int err = super_types[mddev->major_version]
2212 .load_super(rdev, rdev0, mddev->minor_version);
2213 if (err < 0) {
2214 printk(KERN_WARNING
2215 "md: %s has different UUID to %s\n",
2216 bdevname(rdev->bdev,b),
2217 bdevname(rdev0->bdev,b2));
2218 export_rdev(rdev);
2219 return -EINVAL;
2220 }
2221 }
2222 err = bind_rdev_to_array(rdev, mddev);
2223 if (err)
2224 export_rdev(rdev);
2225 return err;
2226 }
2227
2228 /*
2229 * add_new_disk can be used once the array is assembled
2230 * to add "hot spares". They must already have a superblock
2231 * written
2232 */
2233 if (mddev->pers) {
2234 int err;
2235 if (!mddev->pers->hot_add_disk) {
2236 printk(KERN_WARNING
2237 "%s: personality does not support diskops!\n",
2238 mdname(mddev));
2239 return -EINVAL;
2240 }
2241 if (mddev->persistent)
2242 rdev = md_import_device(dev, mddev->major_version,
2243 mddev->minor_version);
2244 else
2245 rdev = md_import_device(dev, -1, -1);
2246 if (IS_ERR(rdev)) {
2247 printk(KERN_WARNING
2248 "md: md_import_device returned %ld\n",
2249 PTR_ERR(rdev));
2250 return PTR_ERR(rdev);
2251 }
2252 /* set save_raid_disk if appropriate */
2253 if (!mddev->persistent) {
2254 if (info->state & (1<<MD_DISK_SYNC) &&
2255 info->raid_disk < mddev->raid_disks)
2256 rdev->raid_disk = info->raid_disk;
2257 else
2258 rdev->raid_disk = -1;
2259 } else
2260 super_types[mddev->major_version].
2261 validate_super(mddev, rdev);
2262 rdev->saved_raid_disk = rdev->raid_disk;
2263
2264 rdev->in_sync = 0; /* just to be sure */
2265 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2266 set_bit(WriteMostly, &rdev->flags);
2267
2268 rdev->raid_disk = -1;
2269 err = bind_rdev_to_array(rdev, mddev);
2270 if (err)
2271 export_rdev(rdev);
2272
2273 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2274 md_wakeup_thread(mddev->thread);
2275 return err;
2276 }
2277
2278 /* otherwise, add_new_disk is only allowed
2279 * for major_version==0 superblocks
2280 */
2281 if (mddev->major_version != 0) {
2282 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2283 mdname(mddev));
2284 return -EINVAL;
2285 }
2286
2287 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2288 int err;
2289 rdev = md_import_device (dev, -1, 0);
2290 if (IS_ERR(rdev)) {
2291 printk(KERN_WARNING
2292 "md: error, md_import_device() returned %ld\n",
2293 PTR_ERR(rdev));
2294 return PTR_ERR(rdev);
2295 }
2296 rdev->desc_nr = info->number;
2297 if (info->raid_disk < mddev->raid_disks)
2298 rdev->raid_disk = info->raid_disk;
2299 else
2300 rdev->raid_disk = -1;
2301
2302 rdev->faulty = 0;
2303 if (rdev->raid_disk < mddev->raid_disks)
2304 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2305 else
2306 rdev->in_sync = 0;
2307
2308 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2309 set_bit(WriteMostly, &rdev->flags);
2310
2311 err = bind_rdev_to_array(rdev, mddev);
2312 if (err) {
2313 export_rdev(rdev);
2314 return err;
2315 }
2316
2317 if (!mddev->persistent) {
2318 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2319 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2320 } else
2321 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2322 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2323
2324 if (!mddev->size || (mddev->size > rdev->size))
2325 mddev->size = rdev->size;
2326 }
2327
2328 return 0;
2329 }
2330
2331 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2332 {
2333 char b[BDEVNAME_SIZE];
2334 mdk_rdev_t *rdev;
2335
2336 if (!mddev->pers)
2337 return -ENODEV;
2338
2339 rdev = find_rdev(mddev, dev);
2340 if (!rdev)
2341 return -ENXIO;
2342
2343 if (rdev->raid_disk >= 0)
2344 goto busy;
2345
2346 kick_rdev_from_array(rdev);
2347 md_update_sb(mddev);
2348
2349 return 0;
2350 busy:
2351 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2352 bdevname(rdev->bdev,b), mdname(mddev));
2353 return -EBUSY;
2354 }
2355
2356 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2357 {
2358 char b[BDEVNAME_SIZE];
2359 int err;
2360 unsigned int size;
2361 mdk_rdev_t *rdev;
2362
2363 if (!mddev->pers)
2364 return -ENODEV;
2365
2366 if (mddev->major_version != 0) {
2367 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2368 " version-0 superblocks.\n",
2369 mdname(mddev));
2370 return -EINVAL;
2371 }
2372 if (!mddev->pers->hot_add_disk) {
2373 printk(KERN_WARNING
2374 "%s: personality does not support diskops!\n",
2375 mdname(mddev));
2376 return -EINVAL;
2377 }
2378
2379 rdev = md_import_device (dev, -1, 0);
2380 if (IS_ERR(rdev)) {
2381 printk(KERN_WARNING
2382 "md: error, md_import_device() returned %ld\n",
2383 PTR_ERR(rdev));
2384 return -EINVAL;
2385 }
2386
2387 if (mddev->persistent)
2388 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2389 else
2390 rdev->sb_offset =
2391 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2392
2393 size = calc_dev_size(rdev, mddev->chunk_size);
2394 rdev->size = size;
2395
2396 if (size < mddev->size) {
2397 printk(KERN_WARNING
2398 "%s: disk size %llu blocks < array size %llu\n",
2399 mdname(mddev), (unsigned long long)size,
2400 (unsigned long long)mddev->size);
2401 err = -ENOSPC;
2402 goto abort_export;
2403 }
2404
2405 if (rdev->faulty) {
2406 printk(KERN_WARNING
2407 "md: can not hot-add faulty %s disk to %s!\n",
2408 bdevname(rdev->bdev,b), mdname(mddev));
2409 err = -EINVAL;
2410 goto abort_export;
2411 }
2412 rdev->in_sync = 0;
2413 rdev->desc_nr = -1;
2414 bind_rdev_to_array(rdev, mddev);
2415
2416 /*
2417 * The rest should better be atomic, we can have disk failures
2418 * noticed in interrupt contexts ...
2419 */
2420
2421 if (rdev->desc_nr == mddev->max_disks) {
2422 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2423 mdname(mddev));
2424 err = -EBUSY;
2425 goto abort_unbind_export;
2426 }
2427
2428 rdev->raid_disk = -1;
2429
2430 md_update_sb(mddev);
2431
2432 /*
2433 * Kick recovery, maybe this spare has to be added to the
2434 * array immediately.
2435 */
2436 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2437 md_wakeup_thread(mddev->thread);
2438
2439 return 0;
2440
2441 abort_unbind_export:
2442 unbind_rdev_from_array(rdev);
2443
2444 abort_export:
2445 export_rdev(rdev);
2446 return err;
2447 }
2448
2449 /* similar to deny_write_access, but accounts for our holding a reference
2450 * to the file ourselves */
2451 static int deny_bitmap_write_access(struct file * file)
2452 {
2453 struct inode *inode = file->f_mapping->host;
2454
2455 spin_lock(&inode->i_lock);
2456 if (atomic_read(&inode->i_writecount) > 1) {
2457 spin_unlock(&inode->i_lock);
2458 return -ETXTBSY;
2459 }
2460 atomic_set(&inode->i_writecount, -1);
2461 spin_unlock(&inode->i_lock);
2462
2463 return 0;
2464 }
2465
2466 static int set_bitmap_file(mddev_t *mddev, int fd)
2467 {
2468 int err;
2469
2470 if (mddev->pers) {
2471 if (!mddev->pers->quiesce)
2472 return -EBUSY;
2473 if (mddev->recovery || mddev->sync_thread)
2474 return -EBUSY;
2475 /* we should be able to change the bitmap.. */
2476 }
2477
2478
2479 if (fd >= 0) {
2480 if (mddev->bitmap)
2481 return -EEXIST; /* cannot add when bitmap is present */
2482 mddev->bitmap_file = fget(fd);
2483
2484 if (mddev->bitmap_file == NULL) {
2485 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2486 mdname(mddev));
2487 return -EBADF;
2488 }
2489
2490 err = deny_bitmap_write_access(mddev->bitmap_file);
2491 if (err) {
2492 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2493 mdname(mddev));
2494 fput(mddev->bitmap_file);
2495 mddev->bitmap_file = NULL;
2496 return err;
2497 }
2498 mddev->bitmap_offset = 0; /* file overrides offset */
2499 } else if (mddev->bitmap == NULL)
2500 return -ENOENT; /* cannot remove what isn't there */
2501 err = 0;
2502 if (mddev->pers) {
2503 mddev->pers->quiesce(mddev, 1);
2504 if (fd >= 0)
2505 err = bitmap_create(mddev);
2506 if (fd < 0 || err)
2507 bitmap_destroy(mddev);
2508 mddev->pers->quiesce(mddev, 0);
2509 } else if (fd < 0) {
2510 if (mddev->bitmap_file)
2511 fput(mddev->bitmap_file);
2512 mddev->bitmap_file = NULL;
2513 }
2514
2515 return err;
2516 }
2517
2518 /*
2519 * set_array_info is used two different ways
2520 * The original usage is when creating a new array.
2521 * In this usage, raid_disks is > 0 and it together with
2522 * level, size, not_persistent,layout,chunksize determine the
2523 * shape of the array.
2524 * This will always create an array with a type-0.90.0 superblock.
2525 * The newer usage is when assembling an array.
2526 * In this case raid_disks will be 0, and the major_version field is
2527 * use to determine which style super-blocks are to be found on the devices.
2528 * The minor and patch _version numbers are also kept incase the
2529 * super_block handler wishes to interpret them.
2530 */
2531 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2532 {
2533
2534 if (info->raid_disks == 0) {
2535 /* just setting version number for superblock loading */
2536 if (info->major_version < 0 ||
2537 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2538 super_types[info->major_version].name == NULL) {
2539 /* maybe try to auto-load a module? */
2540 printk(KERN_INFO
2541 "md: superblock version %d not known\n",
2542 info->major_version);
2543 return -EINVAL;
2544 }
2545 mddev->major_version = info->major_version;
2546 mddev->minor_version = info->minor_version;
2547 mddev->patch_version = info->patch_version;
2548 return 0;
2549 }
2550 mddev->major_version = MD_MAJOR_VERSION;
2551 mddev->minor_version = MD_MINOR_VERSION;
2552 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2553 mddev->ctime = get_seconds();
2554
2555 mddev->level = info->level;
2556 mddev->size = info->size;
2557 mddev->raid_disks = info->raid_disks;
2558 /* don't set md_minor, it is determined by which /dev/md* was
2559 * openned
2560 */
2561 if (info->state & (1<<MD_SB_CLEAN))
2562 mddev->recovery_cp = MaxSector;
2563 else
2564 mddev->recovery_cp = 0;
2565 mddev->persistent = ! info->not_persistent;
2566
2567 mddev->layout = info->layout;
2568 mddev->chunk_size = info->chunk_size;
2569
2570 mddev->max_disks = MD_SB_DISKS;
2571
2572 mddev->sb_dirty = 1;
2573
2574 /*
2575 * Generate a 128 bit UUID
2576 */
2577 get_random_bytes(mddev->uuid, 16);
2578
2579 return 0;
2580 }
2581
2582 /*
2583 * update_array_info is used to change the configuration of an
2584 * on-line array.
2585 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2586 * fields in the info are checked against the array.
2587 * Any differences that cannot be handled will cause an error.
2588 * Normally, only one change can be managed at a time.
2589 */
2590 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2591 {
2592 int rv = 0;
2593 int cnt = 0;
2594 int state = 0;
2595
2596 /* calculate expected state,ignoring low bits */
2597 if (mddev->bitmap && mddev->bitmap_offset)
2598 state |= (1 << MD_SB_BITMAP_PRESENT);
2599
2600 if (mddev->major_version != info->major_version ||
2601 mddev->minor_version != info->minor_version ||
2602 /* mddev->patch_version != info->patch_version || */
2603 mddev->ctime != info->ctime ||
2604 mddev->level != info->level ||
2605 /* mddev->layout != info->layout || */
2606 !mddev->persistent != info->not_persistent||
2607 mddev->chunk_size != info->chunk_size ||
2608 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2609 ((state^info->state) & 0xfffffe00)
2610 )
2611 return -EINVAL;
2612 /* Check there is only one change */
2613 if (mddev->size != info->size) cnt++;
2614 if (mddev->raid_disks != info->raid_disks) cnt++;
2615 if (mddev->layout != info->layout) cnt++;
2616 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2617 if (cnt == 0) return 0;
2618 if (cnt > 1) return -EINVAL;
2619
2620 if (mddev->layout != info->layout) {
2621 /* Change layout
2622 * we don't need to do anything at the md level, the
2623 * personality will take care of it all.
2624 */
2625 if (mddev->pers->reconfig == NULL)
2626 return -EINVAL;
2627 else
2628 return mddev->pers->reconfig(mddev, info->layout, -1);
2629 }
2630 if (mddev->size != info->size) {
2631 mdk_rdev_t * rdev;
2632 struct list_head *tmp;
2633 if (mddev->pers->resize == NULL)
2634 return -EINVAL;
2635 /* The "size" is the amount of each device that is used.
2636 * This can only make sense for arrays with redundancy.
2637 * linear and raid0 always use whatever space is available
2638 * We can only consider changing the size if no resync
2639 * or reconstruction is happening, and if the new size
2640 * is acceptable. It must fit before the sb_offset or,
2641 * if that is <data_offset, it must fit before the
2642 * size of each device.
2643 * If size is zero, we find the largest size that fits.
2644 */
2645 if (mddev->sync_thread)
2646 return -EBUSY;
2647 ITERATE_RDEV(mddev,rdev,tmp) {
2648 sector_t avail;
2649 int fit = (info->size == 0);
2650 if (rdev->sb_offset > rdev->data_offset)
2651 avail = (rdev->sb_offset*2) - rdev->data_offset;
2652 else
2653 avail = get_capacity(rdev->bdev->bd_disk)
2654 - rdev->data_offset;
2655 if (fit && (info->size == 0 || info->size > avail/2))
2656 info->size = avail/2;
2657 if (avail < ((sector_t)info->size << 1))
2658 return -ENOSPC;
2659 }
2660 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2661 if (!rv) {
2662 struct block_device *bdev;
2663
2664 bdev = bdget_disk(mddev->gendisk, 0);
2665 if (bdev) {
2666 down(&bdev->bd_inode->i_sem);
2667 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2668 up(&bdev->bd_inode->i_sem);
2669 bdput(bdev);
2670 }
2671 }
2672 }
2673 if (mddev->raid_disks != info->raid_disks) {
2674 /* change the number of raid disks */
2675 if (mddev->pers->reshape == NULL)
2676 return -EINVAL;
2677 if (info->raid_disks <= 0 ||
2678 info->raid_disks >= mddev->max_disks)
2679 return -EINVAL;
2680 if (mddev->sync_thread)
2681 return -EBUSY;
2682 rv = mddev->pers->reshape(mddev, info->raid_disks);
2683 if (!rv) {
2684 struct block_device *bdev;
2685
2686 bdev = bdget_disk(mddev->gendisk, 0);
2687 if (bdev) {
2688 down(&bdev->bd_inode->i_sem);
2689 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2690 up(&bdev->bd_inode->i_sem);
2691 bdput(bdev);
2692 }
2693 }
2694 }
2695 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2696 if (mddev->pers->quiesce == NULL)
2697 return -EINVAL;
2698 if (mddev->recovery || mddev->sync_thread)
2699 return -EBUSY;
2700 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2701 /* add the bitmap */
2702 if (mddev->bitmap)
2703 return -EEXIST;
2704 if (mddev->default_bitmap_offset == 0)
2705 return -EINVAL;
2706 mddev->bitmap_offset = mddev->default_bitmap_offset;
2707 mddev->pers->quiesce(mddev, 1);
2708 rv = bitmap_create(mddev);
2709 if (rv)
2710 bitmap_destroy(mddev);
2711 mddev->pers->quiesce(mddev, 0);
2712 } else {
2713 /* remove the bitmap */
2714 if (!mddev->bitmap)
2715 return -ENOENT;
2716 if (mddev->bitmap->file)
2717 return -EINVAL;
2718 mddev->pers->quiesce(mddev, 1);
2719 bitmap_destroy(mddev);
2720 mddev->pers->quiesce(mddev, 0);
2721 mddev->bitmap_offset = 0;
2722 }
2723 }
2724 md_update_sb(mddev);
2725 return rv;
2726 }
2727
2728 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2729 {
2730 mdk_rdev_t *rdev;
2731
2732 if (mddev->pers == NULL)
2733 return -ENODEV;
2734
2735 rdev = find_rdev(mddev, dev);
2736 if (!rdev)
2737 return -ENODEV;
2738
2739 md_error(mddev, rdev);
2740 return 0;
2741 }
2742
2743 static int md_ioctl(struct inode *inode, struct file *file,
2744 unsigned int cmd, unsigned long arg)
2745 {
2746 int err = 0;
2747 void __user *argp = (void __user *)arg;
2748 struct hd_geometry __user *loc = argp;
2749 mddev_t *mddev = NULL;
2750
2751 if (!capable(CAP_SYS_ADMIN))
2752 return -EACCES;
2753
2754 /*
2755 * Commands dealing with the RAID driver but not any
2756 * particular array:
2757 */
2758 switch (cmd)
2759 {
2760 case RAID_VERSION:
2761 err = get_version(argp);
2762 goto done;
2763
2764 case PRINT_RAID_DEBUG:
2765 err = 0;
2766 md_print_devices();
2767 goto done;
2768
2769 #ifndef MODULE
2770 case RAID_AUTORUN:
2771 err = 0;
2772 autostart_arrays(arg);
2773 goto done;
2774 #endif
2775 default:;
2776 }
2777
2778 /*
2779 * Commands creating/starting a new array:
2780 */
2781
2782 mddev = inode->i_bdev->bd_disk->private_data;
2783
2784 if (!mddev) {
2785 BUG();
2786 goto abort;
2787 }
2788
2789
2790 if (cmd == START_ARRAY) {
2791 /* START_ARRAY doesn't need to lock the array as autostart_array
2792 * does the locking, and it could even be a different array
2793 */
2794 static int cnt = 3;
2795 if (cnt > 0 ) {
2796 printk(KERN_WARNING
2797 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2798 "This will not be supported beyond 2.6\n",
2799 current->comm, current->pid);
2800 cnt--;
2801 }
2802 err = autostart_array(new_decode_dev(arg));
2803 if (err) {
2804 printk(KERN_WARNING "md: autostart failed!\n");
2805 goto abort;
2806 }
2807 goto done;
2808 }
2809
2810 err = mddev_lock(mddev);
2811 if (err) {
2812 printk(KERN_INFO
2813 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2814 err, cmd);
2815 goto abort;
2816 }
2817
2818 switch (cmd)
2819 {
2820 case SET_ARRAY_INFO:
2821 {
2822 mdu_array_info_t info;
2823 if (!arg)
2824 memset(&info, 0, sizeof(info));
2825 else if (copy_from_user(&info, argp, sizeof(info))) {
2826 err = -EFAULT;
2827 goto abort_unlock;
2828 }
2829 if (mddev->pers) {
2830 err = update_array_info(mddev, &info);
2831 if (err) {
2832 printk(KERN_WARNING "md: couldn't update"
2833 " array info. %d\n", err);
2834 goto abort_unlock;
2835 }
2836 goto done_unlock;
2837 }
2838 if (!list_empty(&mddev->disks)) {
2839 printk(KERN_WARNING
2840 "md: array %s already has disks!\n",
2841 mdname(mddev));
2842 err = -EBUSY;
2843 goto abort_unlock;
2844 }
2845 if (mddev->raid_disks) {
2846 printk(KERN_WARNING
2847 "md: array %s already initialised!\n",
2848 mdname(mddev));
2849 err = -EBUSY;
2850 goto abort_unlock;
2851 }
2852 err = set_array_info(mddev, &info);
2853 if (err) {
2854 printk(KERN_WARNING "md: couldn't set"
2855 " array info. %d\n", err);
2856 goto abort_unlock;
2857 }
2858 }
2859 goto done_unlock;
2860
2861 default:;
2862 }
2863
2864 /*
2865 * Commands querying/configuring an existing array:
2866 */
2867 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2868 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2869 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2870 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2871 err = -ENODEV;
2872 goto abort_unlock;
2873 }
2874
2875 /*
2876 * Commands even a read-only array can execute:
2877 */
2878 switch (cmd)
2879 {
2880 case GET_ARRAY_INFO:
2881 err = get_array_info(mddev, argp);
2882 goto done_unlock;
2883
2884 case GET_BITMAP_FILE:
2885 err = get_bitmap_file(mddev, argp);
2886 goto done_unlock;
2887
2888 case GET_DISK_INFO:
2889 err = get_disk_info(mddev, argp);
2890 goto done_unlock;
2891
2892 case RESTART_ARRAY_RW:
2893 err = restart_array(mddev);
2894 goto done_unlock;
2895
2896 case STOP_ARRAY:
2897 err = do_md_stop (mddev, 0);
2898 goto done_unlock;
2899
2900 case STOP_ARRAY_RO:
2901 err = do_md_stop (mddev, 1);
2902 goto done_unlock;
2903
2904 /*
2905 * We have a problem here : there is no easy way to give a CHS
2906 * virtual geometry. We currently pretend that we have a 2 heads
2907 * 4 sectors (with a BIG number of cylinders...). This drives
2908 * dosfs just mad... ;-)
2909 */
2910 case HDIO_GETGEO:
2911 if (!loc) {
2912 err = -EINVAL;
2913 goto abort_unlock;
2914 }
2915 err = put_user (2, (char __user *) &loc->heads);
2916 if (err)
2917 goto abort_unlock;
2918 err = put_user (4, (char __user *) &loc->sectors);
2919 if (err)
2920 goto abort_unlock;
2921 err = put_user(get_capacity(mddev->gendisk)/8,
2922 (short __user *) &loc->cylinders);
2923 if (err)
2924 goto abort_unlock;
2925 err = put_user (get_start_sect(inode->i_bdev),
2926 (long __user *) &loc->start);
2927 goto done_unlock;
2928 }
2929
2930 /*
2931 * The remaining ioctls are changing the state of the
2932 * superblock, so we do not allow read-only arrays
2933 * here:
2934 */
2935 if (mddev->ro) {
2936 err = -EROFS;
2937 goto abort_unlock;
2938 }
2939
2940 switch (cmd)
2941 {
2942 case ADD_NEW_DISK:
2943 {
2944 mdu_disk_info_t info;
2945 if (copy_from_user(&info, argp, sizeof(info)))
2946 err = -EFAULT;
2947 else
2948 err = add_new_disk(mddev, &info);
2949 goto done_unlock;
2950 }
2951
2952 case HOT_REMOVE_DISK:
2953 err = hot_remove_disk(mddev, new_decode_dev(arg));
2954 goto done_unlock;
2955
2956 case HOT_ADD_DISK:
2957 err = hot_add_disk(mddev, new_decode_dev(arg));
2958 goto done_unlock;
2959
2960 case SET_DISK_FAULTY:
2961 err = set_disk_faulty(mddev, new_decode_dev(arg));
2962 goto done_unlock;
2963
2964 case RUN_ARRAY:
2965 err = do_md_run (mddev);
2966 goto done_unlock;
2967
2968 case SET_BITMAP_FILE:
2969 err = set_bitmap_file(mddev, (int)arg);
2970 goto done_unlock;
2971
2972 default:
2973 if (_IOC_TYPE(cmd) == MD_MAJOR)
2974 printk(KERN_WARNING "md: %s(pid %d) used"
2975 " obsolete MD ioctl, upgrade your"
2976 " software to use new ictls.\n",
2977 current->comm, current->pid);
2978 err = -EINVAL;
2979 goto abort_unlock;
2980 }
2981
2982 done_unlock:
2983 abort_unlock:
2984 mddev_unlock(mddev);
2985
2986 return err;
2987 done:
2988 if (err)
2989 MD_BUG();
2990 abort:
2991 return err;
2992 }
2993
2994 static int md_open(struct inode *inode, struct file *file)
2995 {
2996 /*
2997 * Succeed if we can lock the mddev, which confirms that
2998 * it isn't being stopped right now.
2999 */
3000 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3001 int err;
3002
3003 if ((err = mddev_lock(mddev)))
3004 goto out;
3005
3006 err = 0;
3007 mddev_get(mddev);
3008 mddev_unlock(mddev);
3009
3010 check_disk_change(inode->i_bdev);
3011 out:
3012 return err;
3013 }
3014
3015 static int md_release(struct inode *inode, struct file * file)
3016 {
3017 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3018
3019 if (!mddev)
3020 BUG();
3021 mddev_put(mddev);
3022
3023 return 0;
3024 }
3025
3026 static int md_media_changed(struct gendisk *disk)
3027 {
3028 mddev_t *mddev = disk->private_data;
3029
3030 return mddev->changed;
3031 }
3032
3033 static int md_revalidate(struct gendisk *disk)
3034 {
3035 mddev_t *mddev = disk->private_data;
3036
3037 mddev->changed = 0;
3038 return 0;
3039 }
3040 static struct block_device_operations md_fops =
3041 {
3042 .owner = THIS_MODULE,
3043 .open = md_open,
3044 .release = md_release,
3045 .ioctl = md_ioctl,
3046 .media_changed = md_media_changed,
3047 .revalidate_disk= md_revalidate,
3048 };
3049
3050 static int md_thread(void * arg)
3051 {
3052 mdk_thread_t *thread = arg;
3053
3054 /*
3055 * md_thread is a 'system-thread', it's priority should be very
3056 * high. We avoid resource deadlocks individually in each
3057 * raid personality. (RAID5 does preallocation) We also use RR and
3058 * the very same RT priority as kswapd, thus we will never get
3059 * into a priority inversion deadlock.
3060 *
3061 * we definitely have to have equal or higher priority than
3062 * bdflush, otherwise bdflush will deadlock if there are too
3063 * many dirty RAID5 blocks.
3064 */
3065
3066 allow_signal(SIGKILL);
3067 complete(thread->event);
3068 while (!kthread_should_stop()) {
3069 void (*run)(mddev_t *);
3070
3071 wait_event_interruptible_timeout(thread->wqueue,
3072 test_bit(THREAD_WAKEUP, &thread->flags)
3073 || kthread_should_stop(),
3074 thread->timeout);
3075 try_to_freeze();
3076
3077 clear_bit(THREAD_WAKEUP, &thread->flags);
3078
3079 run = thread->run;
3080 if (run)
3081 run(thread->mddev);
3082 }
3083
3084 return 0;
3085 }
3086
3087 void md_wakeup_thread(mdk_thread_t *thread)
3088 {
3089 if (thread) {
3090 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3091 set_bit(THREAD_WAKEUP, &thread->flags);
3092 wake_up(&thread->wqueue);
3093 }
3094 }
3095
3096 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3097 const char *name)
3098 {
3099 mdk_thread_t *thread;
3100 struct completion event;
3101
3102 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3103 if (!thread)
3104 return NULL;
3105
3106 memset(thread, 0, sizeof(mdk_thread_t));
3107 init_waitqueue_head(&thread->wqueue);
3108
3109 init_completion(&event);
3110 thread->event = &event;
3111 thread->run = run;
3112 thread->mddev = mddev;
3113 thread->name = name;
3114 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3115 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3116 if (IS_ERR(thread->tsk)) {
3117 kfree(thread);
3118 return NULL;
3119 }
3120 wait_for_completion(&event);
3121 return thread;
3122 }
3123
3124 void md_unregister_thread(mdk_thread_t *thread)
3125 {
3126 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3127
3128 kthread_stop(thread->tsk);
3129 kfree(thread);
3130 }
3131
3132 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3133 {
3134 if (!mddev) {
3135 MD_BUG();
3136 return;
3137 }
3138
3139 if (!rdev || rdev->faulty)
3140 return;
3141 /*
3142 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3143 mdname(mddev),
3144 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3145 __builtin_return_address(0),__builtin_return_address(1),
3146 __builtin_return_address(2),__builtin_return_address(3));
3147 */
3148 if (!mddev->pers->error_handler)
3149 return;
3150 mddev->pers->error_handler(mddev,rdev);
3151 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3152 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3153 md_wakeup_thread(mddev->thread);
3154 }
3155
3156 /* seq_file implementation /proc/mdstat */
3157
3158 static void status_unused(struct seq_file *seq)
3159 {
3160 int i = 0;
3161 mdk_rdev_t *rdev;
3162 struct list_head *tmp;
3163
3164 seq_printf(seq, "unused devices: ");
3165
3166 ITERATE_RDEV_PENDING(rdev,tmp) {
3167 char b[BDEVNAME_SIZE];
3168 i++;
3169 seq_printf(seq, "%s ",
3170 bdevname(rdev->bdev,b));
3171 }
3172 if (!i)
3173 seq_printf(seq, "<none>");
3174
3175 seq_printf(seq, "\n");
3176 }
3177
3178
3179 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3180 {
3181 unsigned long max_blocks, resync, res, dt, db, rt;
3182
3183 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3184
3185 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3186 max_blocks = mddev->resync_max_sectors >> 1;
3187 else
3188 max_blocks = mddev->size;
3189
3190 /*
3191 * Should not happen.
3192 */
3193 if (!max_blocks) {
3194 MD_BUG();
3195 return;
3196 }
3197 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3198 {
3199 int i, x = res/50, y = 20-x;
3200 seq_printf(seq, "[");
3201 for (i = 0; i < x; i++)
3202 seq_printf(seq, "=");
3203 seq_printf(seq, ">");
3204 for (i = 0; i < y; i++)
3205 seq_printf(seq, ".");
3206 seq_printf(seq, "] ");
3207 }
3208 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3209 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3210 "resync" : "recovery"),
3211 res/10, res % 10, resync, max_blocks);
3212
3213 /*
3214 * We do not want to overflow, so the order of operands and
3215 * the * 100 / 100 trick are important. We do a +1 to be
3216 * safe against division by zero. We only estimate anyway.
3217 *
3218 * dt: time from mark until now
3219 * db: blocks written from mark until now
3220 * rt: remaining time
3221 */
3222 dt = ((jiffies - mddev->resync_mark) / HZ);
3223 if (!dt) dt++;
3224 db = resync - (mddev->resync_mark_cnt/2);
3225 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3226
3227 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3228
3229 seq_printf(seq, " speed=%ldK/sec", db/dt);
3230 }
3231
3232 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3233 {
3234 struct list_head *tmp;
3235 loff_t l = *pos;
3236 mddev_t *mddev;
3237
3238 if (l >= 0x10000)
3239 return NULL;
3240 if (!l--)
3241 /* header */
3242 return (void*)1;
3243
3244 spin_lock(&all_mddevs_lock);
3245 list_for_each(tmp,&all_mddevs)
3246 if (!l--) {
3247 mddev = list_entry(tmp, mddev_t, all_mddevs);
3248 mddev_get(mddev);
3249 spin_unlock(&all_mddevs_lock);
3250 return mddev;
3251 }
3252 spin_unlock(&all_mddevs_lock);
3253 if (!l--)
3254 return (void*)2;/* tail */
3255 return NULL;
3256 }
3257
3258 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3259 {
3260 struct list_head *tmp;
3261 mddev_t *next_mddev, *mddev = v;
3262
3263 ++*pos;
3264 if (v == (void*)2)
3265 return NULL;
3266
3267 spin_lock(&all_mddevs_lock);
3268 if (v == (void*)1)
3269 tmp = all_mddevs.next;
3270 else
3271 tmp = mddev->all_mddevs.next;
3272 if (tmp != &all_mddevs)
3273 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3274 else {
3275 next_mddev = (void*)2;
3276 *pos = 0x10000;
3277 }
3278 spin_unlock(&all_mddevs_lock);
3279
3280 if (v != (void*)1)
3281 mddev_put(mddev);
3282 return next_mddev;
3283
3284 }
3285
3286 static void md_seq_stop(struct seq_file *seq, void *v)
3287 {
3288 mddev_t *mddev = v;
3289
3290 if (mddev && v != (void*)1 && v != (void*)2)
3291 mddev_put(mddev);
3292 }
3293
3294 static int md_seq_show(struct seq_file *seq, void *v)
3295 {
3296 mddev_t *mddev = v;
3297 sector_t size;
3298 struct list_head *tmp2;
3299 mdk_rdev_t *rdev;
3300 int i;
3301 struct bitmap *bitmap;
3302
3303 if (v == (void*)1) {
3304 seq_printf(seq, "Personalities : ");
3305 spin_lock(&pers_lock);
3306 for (i = 0; i < MAX_PERSONALITY; i++)
3307 if (pers[i])
3308 seq_printf(seq, "[%s] ", pers[i]->name);
3309
3310 spin_unlock(&pers_lock);
3311 seq_printf(seq, "\n");
3312 return 0;
3313 }
3314 if (v == (void*)2) {
3315 status_unused(seq);
3316 return 0;
3317 }
3318
3319 if (mddev_lock(mddev)!=0)
3320 return -EINTR;
3321 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3322 seq_printf(seq, "%s : %sactive", mdname(mddev),
3323 mddev->pers ? "" : "in");
3324 if (mddev->pers) {
3325 if (mddev->ro)
3326 seq_printf(seq, " (read-only)");
3327 seq_printf(seq, " %s", mddev->pers->name);
3328 }
3329
3330 size = 0;
3331 ITERATE_RDEV(mddev,rdev,tmp2) {
3332 char b[BDEVNAME_SIZE];
3333 seq_printf(seq, " %s[%d]",
3334 bdevname(rdev->bdev,b), rdev->desc_nr);
3335 if (test_bit(WriteMostly, &rdev->flags))
3336 seq_printf(seq, "(W)");
3337 if (rdev->faulty) {
3338 seq_printf(seq, "(F)");
3339 continue;
3340 } else if (rdev->raid_disk < 0)
3341 seq_printf(seq, "(S)"); /* spare */
3342 size += rdev->size;
3343 }
3344
3345 if (!list_empty(&mddev->disks)) {
3346 if (mddev->pers)
3347 seq_printf(seq, "\n %llu blocks",
3348 (unsigned long long)mddev->array_size);
3349 else
3350 seq_printf(seq, "\n %llu blocks",
3351 (unsigned long long)size);
3352 }
3353 if (mddev->persistent) {
3354 if (mddev->major_version != 0 ||
3355 mddev->minor_version != 90) {
3356 seq_printf(seq," super %d.%d",
3357 mddev->major_version,
3358 mddev->minor_version);
3359 }
3360 } else
3361 seq_printf(seq, " super non-persistent");
3362
3363 if (mddev->pers) {
3364 mddev->pers->status (seq, mddev);
3365 seq_printf(seq, "\n ");
3366 if (mddev->curr_resync > 2) {
3367 status_resync (seq, mddev);
3368 seq_printf(seq, "\n ");
3369 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3370 seq_printf(seq, " resync=DELAYED\n ");
3371 } else
3372 seq_printf(seq, "\n ");
3373
3374 if ((bitmap = mddev->bitmap)) {
3375 unsigned long chunk_kb;
3376 unsigned long flags;
3377 spin_lock_irqsave(&bitmap->lock, flags);
3378 chunk_kb = bitmap->chunksize >> 10;
3379 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3380 "%lu%s chunk",
3381 bitmap->pages - bitmap->missing_pages,
3382 bitmap->pages,
3383 (bitmap->pages - bitmap->missing_pages)
3384 << (PAGE_SHIFT - 10),
3385 chunk_kb ? chunk_kb : bitmap->chunksize,
3386 chunk_kb ? "KB" : "B");
3387 if (bitmap->file) {
3388 seq_printf(seq, ", file: ");
3389 seq_path(seq, bitmap->file->f_vfsmnt,
3390 bitmap->file->f_dentry," \t\n");
3391 }
3392
3393 seq_printf(seq, "\n");
3394 spin_unlock_irqrestore(&bitmap->lock, flags);
3395 }
3396
3397 seq_printf(seq, "\n");
3398 }
3399 mddev_unlock(mddev);
3400
3401 return 0;
3402 }
3403
3404 static struct seq_operations md_seq_ops = {
3405 .start = md_seq_start,
3406 .next = md_seq_next,
3407 .stop = md_seq_stop,
3408 .show = md_seq_show,
3409 };
3410
3411 static int md_seq_open(struct inode *inode, struct file *file)
3412 {
3413 int error;
3414
3415 error = seq_open(file, &md_seq_ops);
3416 return error;
3417 }
3418
3419 static struct file_operations md_seq_fops = {
3420 .open = md_seq_open,
3421 .read = seq_read,
3422 .llseek = seq_lseek,
3423 .release = seq_release,
3424 };
3425
3426 int register_md_personality(int pnum, mdk_personality_t *p)
3427 {
3428 if (pnum >= MAX_PERSONALITY) {
3429 printk(KERN_ERR
3430 "md: tried to install personality %s as nr %d, but max is %lu\n",
3431 p->name, pnum, MAX_PERSONALITY-1);
3432 return -EINVAL;
3433 }
3434
3435 spin_lock(&pers_lock);
3436 if (pers[pnum]) {
3437 spin_unlock(&pers_lock);
3438 return -EBUSY;
3439 }
3440
3441 pers[pnum] = p;
3442 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3443 spin_unlock(&pers_lock);
3444 return 0;
3445 }
3446
3447 int unregister_md_personality(int pnum)
3448 {
3449 if (pnum >= MAX_PERSONALITY)
3450 return -EINVAL;
3451
3452 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3453 spin_lock(&pers_lock);
3454 pers[pnum] = NULL;
3455 spin_unlock(&pers_lock);
3456 return 0;
3457 }
3458
3459 static int is_mddev_idle(mddev_t *mddev)
3460 {
3461 mdk_rdev_t * rdev;
3462 struct list_head *tmp;
3463 int idle;
3464 unsigned long curr_events;
3465
3466 idle = 1;
3467 ITERATE_RDEV(mddev,rdev,tmp) {
3468 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3469 curr_events = disk_stat_read(disk, read_sectors) +
3470 disk_stat_read(disk, write_sectors) -
3471 atomic_read(&disk->sync_io);
3472 /* Allow some slack between valud of curr_events and last_events,
3473 * as there are some uninteresting races.
3474 * Note: the following is an unsigned comparison.
3475 */
3476 if ((curr_events - rdev->last_events + 32) > 64) {
3477 rdev->last_events = curr_events;
3478 idle = 0;
3479 }
3480 }
3481 return idle;
3482 }
3483
3484 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3485 {
3486 /* another "blocks" (512byte) blocks have been synced */
3487 atomic_sub(blocks, &mddev->recovery_active);
3488 wake_up(&mddev->recovery_wait);
3489 if (!ok) {
3490 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3491 md_wakeup_thread(mddev->thread);
3492 // stop recovery, signal do_sync ....
3493 }
3494 }
3495
3496
3497 /* md_write_start(mddev, bi)
3498 * If we need to update some array metadata (e.g. 'active' flag
3499 * in superblock) before writing, schedule a superblock update
3500 * and wait for it to complete.
3501 */
3502 void md_write_start(mddev_t *mddev, struct bio *bi)
3503 {
3504 if (bio_data_dir(bi) != WRITE)
3505 return;
3506
3507 atomic_inc(&mddev->writes_pending);
3508 if (mddev->in_sync) {
3509 spin_lock(&mddev->write_lock);
3510 if (mddev->in_sync) {
3511 mddev->in_sync = 0;
3512 mddev->sb_dirty = 1;
3513 md_wakeup_thread(mddev->thread);
3514 }
3515 spin_unlock(&mddev->write_lock);
3516 }
3517 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3518 }
3519
3520 void md_write_end(mddev_t *mddev)
3521 {
3522 if (atomic_dec_and_test(&mddev->writes_pending)) {
3523 if (mddev->safemode == 2)
3524 md_wakeup_thread(mddev->thread);
3525 else
3526 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3527 }
3528 }
3529
3530 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3531
3532 #define SYNC_MARKS 10
3533 #define SYNC_MARK_STEP (3*HZ)
3534 static void md_do_sync(mddev_t *mddev)
3535 {
3536 mddev_t *mddev2;
3537 unsigned int currspeed = 0,
3538 window;
3539 sector_t max_sectors,j, io_sectors;
3540 unsigned long mark[SYNC_MARKS];
3541 sector_t mark_cnt[SYNC_MARKS];
3542 int last_mark,m;
3543 struct list_head *tmp;
3544 sector_t last_check;
3545 int skipped = 0;
3546
3547 /* just incase thread restarts... */
3548 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3549 return;
3550
3551 /* we overload curr_resync somewhat here.
3552 * 0 == not engaged in resync at all
3553 * 2 == checking that there is no conflict with another sync
3554 * 1 == like 2, but have yielded to allow conflicting resync to
3555 * commense
3556 * other == active in resync - this many blocks
3557 *
3558 * Before starting a resync we must have set curr_resync to
3559 * 2, and then checked that every "conflicting" array has curr_resync
3560 * less than ours. When we find one that is the same or higher
3561 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3562 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3563 * This will mean we have to start checking from the beginning again.
3564 *
3565 */
3566
3567 do {
3568 mddev->curr_resync = 2;
3569
3570 try_again:
3571 if (signal_pending(current) ||
3572 kthread_should_stop()) {
3573 flush_signals(current);
3574 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3575 goto skip;
3576 }
3577 ITERATE_MDDEV(mddev2,tmp) {
3578 if (mddev2 == mddev)
3579 continue;
3580 if (mddev2->curr_resync &&
3581 match_mddev_units(mddev,mddev2)) {
3582 DEFINE_WAIT(wq);
3583 if (mddev < mddev2 && mddev->curr_resync == 2) {
3584 /* arbitrarily yield */
3585 mddev->curr_resync = 1;
3586 wake_up(&resync_wait);
3587 }
3588 if (mddev > mddev2 && mddev->curr_resync == 1)
3589 /* no need to wait here, we can wait the next
3590 * time 'round when curr_resync == 2
3591 */
3592 continue;
3593 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3594 if (!signal_pending(current) &&
3595 !kthread_should_stop() &&
3596 mddev2->curr_resync >= mddev->curr_resync) {
3597 printk(KERN_INFO "md: delaying resync of %s"
3598 " until %s has finished resync (they"
3599 " share one or more physical units)\n",
3600 mdname(mddev), mdname(mddev2));
3601 mddev_put(mddev2);
3602 schedule();
3603 finish_wait(&resync_wait, &wq);
3604 goto try_again;
3605 }
3606 finish_wait(&resync_wait, &wq);
3607 }
3608 }
3609 } while (mddev->curr_resync < 2);
3610
3611 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3612 /* resync follows the size requested by the personality,
3613 * which defaults to physical size, but can be virtual size
3614 */
3615 max_sectors = mddev->resync_max_sectors;
3616 else
3617 /* recovery follows the physical size of devices */
3618 max_sectors = mddev->size << 1;
3619
3620 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3621 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3622 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3623 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3624 "(but not more than %d KB/sec) for reconstruction.\n",
3625 sysctl_speed_limit_max);
3626
3627 is_mddev_idle(mddev); /* this also initializes IO event counters */
3628 /* we don't use the checkpoint if there's a bitmap */
3629 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3630 j = mddev->recovery_cp;
3631 else
3632 j = 0;
3633 io_sectors = 0;
3634 for (m = 0; m < SYNC_MARKS; m++) {
3635 mark[m] = jiffies;
3636 mark_cnt[m] = io_sectors;
3637 }
3638 last_mark = 0;
3639 mddev->resync_mark = mark[last_mark];
3640 mddev->resync_mark_cnt = mark_cnt[last_mark];
3641
3642 /*
3643 * Tune reconstruction:
3644 */
3645 window = 32*(PAGE_SIZE/512);
3646 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3647 window/2,(unsigned long long) max_sectors/2);
3648
3649 atomic_set(&mddev->recovery_active, 0);
3650 init_waitqueue_head(&mddev->recovery_wait);
3651 last_check = 0;
3652
3653 if (j>2) {
3654 printk(KERN_INFO
3655 "md: resuming recovery of %s from checkpoint.\n",
3656 mdname(mddev));
3657 mddev->curr_resync = j;
3658 }
3659
3660 while (j < max_sectors) {
3661 sector_t sectors;
3662
3663 skipped = 0;
3664 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3665 currspeed < sysctl_speed_limit_min);
3666 if (sectors == 0) {
3667 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3668 goto out;
3669 }
3670
3671 if (!skipped) { /* actual IO requested */
3672 io_sectors += sectors;
3673 atomic_add(sectors, &mddev->recovery_active);
3674 }
3675
3676 j += sectors;
3677 if (j>1) mddev->curr_resync = j;
3678
3679
3680 if (last_check + window > io_sectors || j == max_sectors)
3681 continue;
3682
3683 last_check = io_sectors;
3684
3685 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3686 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3687 break;
3688
3689 repeat:
3690 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3691 /* step marks */
3692 int next = (last_mark+1) % SYNC_MARKS;
3693
3694 mddev->resync_mark = mark[next];
3695 mddev->resync_mark_cnt = mark_cnt[next];
3696 mark[next] = jiffies;
3697 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3698 last_mark = next;
3699 }
3700
3701
3702 if (signal_pending(current) || kthread_should_stop()) {
3703 /*
3704 * got a signal, exit.
3705 */
3706 printk(KERN_INFO
3707 "md: md_do_sync() got signal ... exiting\n");
3708 flush_signals(current);
3709 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3710 goto out;
3711 }
3712
3713 /*
3714 * this loop exits only if either when we are slower than
3715 * the 'hard' speed limit, or the system was IO-idle for
3716 * a jiffy.
3717 * the system might be non-idle CPU-wise, but we only care
3718 * about not overloading the IO subsystem. (things like an
3719 * e2fsck being done on the RAID array should execute fast)
3720 */
3721 mddev->queue->unplug_fn(mddev->queue);
3722 cond_resched();
3723
3724 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3725 /((jiffies-mddev->resync_mark)/HZ +1) +1;
3726
3727 if (currspeed > sysctl_speed_limit_min) {
3728 if ((currspeed > sysctl_speed_limit_max) ||
3729 !is_mddev_idle(mddev)) {
3730 msleep_interruptible(250);
3731 goto repeat;
3732 }
3733 }
3734 }
3735 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3736 /*
3737 * this also signals 'finished resyncing' to md_stop
3738 */
3739 out:
3740 mddev->queue->unplug_fn(mddev->queue);
3741
3742 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3743
3744 /* tell personality that we are finished */
3745 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3746
3747 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3748 mddev->curr_resync > 2 &&
3749 mddev->curr_resync >= mddev->recovery_cp) {
3750 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3751 printk(KERN_INFO
3752 "md: checkpointing recovery of %s.\n",
3753 mdname(mddev));
3754 mddev->recovery_cp = mddev->curr_resync;
3755 } else
3756 mddev->recovery_cp = MaxSector;
3757 }
3758
3759 skip:
3760 mddev->curr_resync = 0;
3761 wake_up(&resync_wait);
3762 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3763 md_wakeup_thread(mddev->thread);
3764 }
3765
3766
3767 /*
3768 * This routine is regularly called by all per-raid-array threads to
3769 * deal with generic issues like resync and super-block update.
3770 * Raid personalities that don't have a thread (linear/raid0) do not
3771 * need this as they never do any recovery or update the superblock.
3772 *
3773 * It does not do any resync itself, but rather "forks" off other threads
3774 * to do that as needed.
3775 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3776 * "->recovery" and create a thread at ->sync_thread.
3777 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3778 * and wakeups up this thread which will reap the thread and finish up.
3779 * This thread also removes any faulty devices (with nr_pending == 0).
3780 *
3781 * The overall approach is:
3782 * 1/ if the superblock needs updating, update it.
3783 * 2/ If a recovery thread is running, don't do anything else.
3784 * 3/ If recovery has finished, clean up, possibly marking spares active.
3785 * 4/ If there are any faulty devices, remove them.
3786 * 5/ If array is degraded, try to add spares devices
3787 * 6/ If array has spares or is not in-sync, start a resync thread.
3788 */
3789 void md_check_recovery(mddev_t *mddev)
3790 {
3791 mdk_rdev_t *rdev;
3792 struct list_head *rtmp;
3793
3794
3795 if (mddev->bitmap)
3796 bitmap_daemon_work(mddev->bitmap);
3797
3798 if (mddev->ro)
3799 return;
3800
3801 if (signal_pending(current)) {
3802 if (mddev->pers->sync_request) {
3803 printk(KERN_INFO "md: %s in immediate safe mode\n",
3804 mdname(mddev));
3805 mddev->safemode = 2;
3806 }
3807 flush_signals(current);
3808 }
3809
3810 if ( ! (
3811 mddev->sb_dirty ||
3812 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3813 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3814 (mddev->safemode == 1) ||
3815 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3816 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3817 ))
3818 return;
3819
3820 if (mddev_trylock(mddev)==0) {
3821 int spares =0;
3822
3823 spin_lock(&mddev->write_lock);
3824 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3825 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3826 mddev->in_sync = 1;
3827 mddev->sb_dirty = 1;
3828 }
3829 if (mddev->safemode == 1)
3830 mddev->safemode = 0;
3831 spin_unlock(&mddev->write_lock);
3832
3833 if (mddev->sb_dirty)
3834 md_update_sb(mddev);
3835
3836
3837 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3838 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3839 /* resync/recovery still happening */
3840 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3841 goto unlock;
3842 }
3843 if (mddev->sync_thread) {
3844 /* resync has finished, collect result */
3845 md_unregister_thread(mddev->sync_thread);
3846 mddev->sync_thread = NULL;
3847 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3848 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3849 /* success...*/
3850 /* activate any spares */
3851 mddev->pers->spare_active(mddev);
3852 }
3853 md_update_sb(mddev);
3854
3855 /* if array is no-longer degraded, then any saved_raid_disk
3856 * information must be scrapped
3857 */
3858 if (!mddev->degraded)
3859 ITERATE_RDEV(mddev,rdev,rtmp)
3860 rdev->saved_raid_disk = -1;
3861
3862 mddev->recovery = 0;
3863 /* flag recovery needed just to double check */
3864 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3865 goto unlock;
3866 }
3867 if (mddev->recovery)
3868 /* probably just the RECOVERY_NEEDED flag */
3869 mddev->recovery = 0;
3870
3871 /* no recovery is running.
3872 * remove any failed drives, then
3873 * add spares if possible.
3874 * Spare are also removed and re-added, to allow
3875 * the personality to fail the re-add.
3876 */
3877 ITERATE_RDEV(mddev,rdev,rtmp)
3878 if (rdev->raid_disk >= 0 &&
3879 (rdev->faulty || ! rdev->in_sync) &&
3880 atomic_read(&rdev->nr_pending)==0) {
3881 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3882 rdev->raid_disk = -1;
3883 }
3884
3885 if (mddev->degraded) {
3886 ITERATE_RDEV(mddev,rdev,rtmp)
3887 if (rdev->raid_disk < 0
3888 && !rdev->faulty) {
3889 if (mddev->pers->hot_add_disk(mddev,rdev))
3890 spares++;
3891 else
3892 break;
3893 }
3894 }
3895
3896 if (!spares && (mddev->recovery_cp == MaxSector )) {
3897 /* nothing we can do ... */
3898 goto unlock;
3899 }
3900 if (mddev->pers->sync_request) {
3901 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3902 if (!spares)
3903 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3904 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3905 /* We are adding a device or devices to an array
3906 * which has the bitmap stored on all devices.
3907 * So make sure all bitmap pages get written
3908 */
3909 bitmap_write_all(mddev->bitmap);
3910 }
3911 mddev->sync_thread = md_register_thread(md_do_sync,
3912 mddev,
3913 "%s_resync");
3914 if (!mddev->sync_thread) {
3915 printk(KERN_ERR "%s: could not start resync"
3916 " thread...\n",
3917 mdname(mddev));
3918 /* leave the spares where they are, it shouldn't hurt */
3919 mddev->recovery = 0;
3920 } else {
3921 md_wakeup_thread(mddev->sync_thread);
3922 }
3923 }
3924 unlock:
3925 mddev_unlock(mddev);
3926 }
3927 }
3928
3929 static int md_notify_reboot(struct notifier_block *this,
3930 unsigned long code, void *x)
3931 {
3932 struct list_head *tmp;
3933 mddev_t *mddev;
3934
3935 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3936
3937 printk(KERN_INFO "md: stopping all md devices.\n");
3938
3939 ITERATE_MDDEV(mddev,tmp)
3940 if (mddev_trylock(mddev)==0)
3941 do_md_stop (mddev, 1);
3942 /*
3943 * certain more exotic SCSI devices are known to be
3944 * volatile wrt too early system reboots. While the
3945 * right place to handle this issue is the given
3946 * driver, we do want to have a safe RAID driver ...
3947 */
3948 mdelay(1000*1);
3949 }
3950 return NOTIFY_DONE;
3951 }
3952
3953 static struct notifier_block md_notifier = {
3954 .notifier_call = md_notify_reboot,
3955 .next = NULL,
3956 .priority = INT_MAX, /* before any real devices */
3957 };
3958
3959 static void md_geninit(void)
3960 {
3961 struct proc_dir_entry *p;
3962
3963 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3964
3965 p = create_proc_entry("mdstat", S_IRUGO, NULL);
3966 if (p)
3967 p->proc_fops = &md_seq_fops;
3968 }
3969
3970 static int __init md_init(void)
3971 {
3972 int minor;
3973
3974 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3975 " MD_SB_DISKS=%d\n",
3976 MD_MAJOR_VERSION, MD_MINOR_VERSION,
3977 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3978 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3979 BITMAP_MINOR);
3980
3981 if (register_blkdev(MAJOR_NR, "md"))
3982 return -1;
3983 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3984 unregister_blkdev(MAJOR_NR, "md");
3985 return -1;
3986 }
3987 devfs_mk_dir("md");
3988 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3989 md_probe, NULL, NULL);
3990 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3991 md_probe, NULL, NULL);
3992
3993 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3994 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3995 S_IFBLK|S_IRUSR|S_IWUSR,
3996 "md/%d", minor);
3997
3998 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3999 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4000 S_IFBLK|S_IRUSR|S_IWUSR,
4001 "md/mdp%d", minor);
4002
4003
4004 register_reboot_notifier(&md_notifier);
4005 raid_table_header = register_sysctl_table(raid_root_table, 1);
4006
4007 md_geninit();
4008 return (0);
4009 }
4010
4011
4012 #ifndef MODULE
4013
4014 /*
4015 * Searches all registered partitions for autorun RAID arrays
4016 * at boot time.
4017 */
4018 static dev_t detected_devices[128];
4019 static int dev_cnt;
4020
4021 void md_autodetect_dev(dev_t dev)
4022 {
4023 if (dev_cnt >= 0 && dev_cnt < 127)
4024 detected_devices[dev_cnt++] = dev;
4025 }
4026
4027
4028 static void autostart_arrays(int part)
4029 {
4030 mdk_rdev_t *rdev;
4031 int i;
4032
4033 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4034
4035 for (i = 0; i < dev_cnt; i++) {
4036 dev_t dev = detected_devices[i];
4037
4038 rdev = md_import_device(dev,0, 0);
4039 if (IS_ERR(rdev))
4040 continue;
4041
4042 if (rdev->faulty) {
4043 MD_BUG();
4044 continue;
4045 }
4046 list_add(&rdev->same_set, &pending_raid_disks);
4047 }
4048 dev_cnt = 0;
4049
4050 autorun_devices(part);
4051 }
4052
4053 #endif
4054
4055 static __exit void md_exit(void)
4056 {
4057 mddev_t *mddev;
4058 struct list_head *tmp;
4059 int i;
4060 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4061 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4062 for (i=0; i < MAX_MD_DEVS; i++)
4063 devfs_remove("md/%d", i);
4064 for (i=0; i < MAX_MD_DEVS; i++)
4065 devfs_remove("md/d%d", i);
4066
4067 devfs_remove("md");
4068
4069 unregister_blkdev(MAJOR_NR,"md");
4070 unregister_blkdev(mdp_major, "mdp");
4071 unregister_reboot_notifier(&md_notifier);
4072 unregister_sysctl_table(raid_table_header);
4073 remove_proc_entry("mdstat", NULL);
4074 ITERATE_MDDEV(mddev,tmp) {
4075 struct gendisk *disk = mddev->gendisk;
4076 if (!disk)
4077 continue;
4078 export_array(mddev);
4079 del_gendisk(disk);
4080 put_disk(disk);
4081 mddev->gendisk = NULL;
4082 mddev_put(mddev);
4083 }
4084 }
4085
4086 module_init(md_init)
4087 module_exit(md_exit)
4088
4089 EXPORT_SYMBOL(register_md_personality);
4090 EXPORT_SYMBOL(unregister_md_personality);
4091 EXPORT_SYMBOL(md_error);
4092 EXPORT_SYMBOL(md_done_sync);
4093 EXPORT_SYMBOL(md_write_start);
4094 EXPORT_SYMBOL(md_write_end);
4095 EXPORT_SYMBOL(md_register_thread);
4096 EXPORT_SYMBOL(md_unregister_thread);
4097 EXPORT_SYMBOL(md_wakeup_thread);
4098 EXPORT_SYMBOL(md_print_devices);
4099 EXPORT_SYMBOL(md_check_recovery);
4100 MODULE_LICENSE("GPL");
4101 MODULE_ALIAS("md");
4102 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel-based virtual machine