?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] md: allow chunk_size to be settable through sysfs
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blob9e57e97bd530f0193b48559470a0707fce0a703a
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 #include <linux/poll.h>
46
47 #include <linux/init.h>
48
49 #include <linux/file.h>
50
51 #ifdef CONFIG_KMOD
52 #include <linux/kmod.h>
53 #endif
54
55 #include <asm/unaligned.h>
56
57 #define MAJOR_NR MD_MAJOR
58 #define MD_DRIVER
59
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
62
63 #define DEBUG 0
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67 #ifndef MODULE
68 static void autostart_arrays (int part);
69 #endif
70
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
73
74 /*
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
81 * idle IO detection.
82 *
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84 */
85
86 static int sysctl_speed_limit_min = 1000;
87 static int sysctl_speed_limit_max = 200000;
88
89 static struct ctl_table_header *raid_table_header;
90
91 static ctl_table raid_table[] = {
92 {
93 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
94 .procname = "speed_limit_min",
95 .data = &sysctl_speed_limit_min,
96 .maxlen = sizeof(int),
97 .mode = 0644,
98 .proc_handler = &proc_dointvec,
99 },
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
102 .procname = "speed_limit_max",
103 .data = &sysctl_speed_limit_max,
104 .maxlen = sizeof(int),
105 .mode = 0644,
106 .proc_handler = &proc_dointvec,
107 },
108 { .ctl_name = 0 }
109 };
110
111 static ctl_table raid_dir_table[] = {
112 {
113 .ctl_name = DEV_RAID,
114 .procname = "raid",
115 .maxlen = 0,
116 .mode = 0555,
117 .child = raid_table,
118 },
119 { .ctl_name = 0 }
120 };
121
122 static ctl_table raid_root_table[] = {
123 {
124 .ctl_name = CTL_DEV,
125 .procname = "dev",
126 .maxlen = 0,
127 .mode = 0555,
128 .child = raid_dir_table,
129 },
130 { .ctl_name = 0 }
131 };
132
133 static struct block_device_operations md_fops;
134
135 static int start_readonly;
136
137 /*
138 * We have a system wide 'event count' that is incremented
139 * on any 'interesting' event, and readers of /proc/mdstat
140 * can use 'poll' or 'select' to find out when the event
141 * count increases.
142 *
143 * Events are:
144 * start array, stop array, error, add device, remove device,
145 * start build, activate spare
146 */
147 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
148 static atomic_t md_event_count;
149 static void md_new_event(mddev_t *mddev)
150 {
151 atomic_inc(&md_event_count);
152 wake_up(&md_event_waiters);
153 }
154
155 /*
156 * Enables to iterate over all existing md arrays
157 * all_mddevs_lock protects this list.
158 */
159 static LIST_HEAD(all_mddevs);
160 static DEFINE_SPINLOCK(all_mddevs_lock);
161
162
163 /*
164 * iterates through all used mddevs in the system.
165 * We take care to grab the all_mddevs_lock whenever navigating
166 * the list, and to always hold a refcount when unlocked.
167 * Any code which breaks out of this loop while own
168 * a reference to the current mddev and must mddev_put it.
169 */
170 #define ITERATE_MDDEV(mddev,tmp) \
171 \
172 for (({ spin_lock(&all_mddevs_lock); \
173 tmp = all_mddevs.next; \
174 mddev = NULL;}); \
175 ({ if (tmp != &all_mddevs) \
176 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
177 spin_unlock(&all_mddevs_lock); \
178 if (mddev) mddev_put(mddev); \
179 mddev = list_entry(tmp, mddev_t, all_mddevs); \
180 tmp != &all_mddevs;}); \
181 ({ spin_lock(&all_mddevs_lock); \
182 tmp = tmp->next;}) \
183 )
184
185
186 static int md_fail_request (request_queue_t *q, struct bio *bio)
187 {
188 bio_io_error(bio, bio->bi_size);
189 return 0;
190 }
191
192 static inline mddev_t *mddev_get(mddev_t *mddev)
193 {
194 atomic_inc(&mddev->active);
195 return mddev;
196 }
197
198 static void mddev_put(mddev_t *mddev)
199 {
200 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
201 return;
202 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
203 list_del(&mddev->all_mddevs);
204 blk_put_queue(mddev->queue);
205 kobject_unregister(&mddev->kobj);
206 }
207 spin_unlock(&all_mddevs_lock);
208 }
209
210 static mddev_t * mddev_find(dev_t unit)
211 {
212 mddev_t *mddev, *new = NULL;
213
214 retry:
215 spin_lock(&all_mddevs_lock);
216 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
217 if (mddev->unit == unit) {
218 mddev_get(mddev);
219 spin_unlock(&all_mddevs_lock);
220 kfree(new);
221 return mddev;
222 }
223
224 if (new) {
225 list_add(&new->all_mddevs, &all_mddevs);
226 spin_unlock(&all_mddevs_lock);
227 return new;
228 }
229 spin_unlock(&all_mddevs_lock);
230
231 new = kzalloc(sizeof(*new), GFP_KERNEL);
232 if (!new)
233 return NULL;
234
235 new->unit = unit;
236 if (MAJOR(unit) == MD_MAJOR)
237 new->md_minor = MINOR(unit);
238 else
239 new->md_minor = MINOR(unit) >> MdpMinorShift;
240
241 init_MUTEX(&new->reconfig_sem);
242 INIT_LIST_HEAD(&new->disks);
243 INIT_LIST_HEAD(&new->all_mddevs);
244 init_timer(&new->safemode_timer);
245 atomic_set(&new->active, 1);
246 spin_lock_init(&new->write_lock);
247 init_waitqueue_head(&new->sb_wait);
248
249 new->queue = blk_alloc_queue(GFP_KERNEL);
250 if (!new->queue) {
251 kfree(new);
252 return NULL;
253 }
254
255 blk_queue_make_request(new->queue, md_fail_request);
256
257 goto retry;
258 }
259
260 static inline int mddev_lock(mddev_t * mddev)
261 {
262 return down_interruptible(&mddev->reconfig_sem);
263 }
264
265 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
266 {
267 down(&mddev->reconfig_sem);
268 }
269
270 static inline int mddev_trylock(mddev_t * mddev)
271 {
272 return down_trylock(&mddev->reconfig_sem);
273 }
274
275 static inline void mddev_unlock(mddev_t * mddev)
276 {
277 up(&mddev->reconfig_sem);
278
279 md_wakeup_thread(mddev->thread);
280 }
281
282 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
283 {
284 mdk_rdev_t * rdev;
285 struct list_head *tmp;
286
287 ITERATE_RDEV(mddev,rdev,tmp) {
288 if (rdev->desc_nr == nr)
289 return rdev;
290 }
291 return NULL;
292 }
293
294 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
295 {
296 struct list_head *tmp;
297 mdk_rdev_t *rdev;
298
299 ITERATE_RDEV(mddev,rdev,tmp) {
300 if (rdev->bdev->bd_dev == dev)
301 return rdev;
302 }
303 return NULL;
304 }
305
306 static struct mdk_personality *find_pers(int level)
307 {
308 struct mdk_personality *pers;
309 list_for_each_entry(pers, &pers_list, list)
310 if (pers->level == level)
311 return pers;
312 return NULL;
313 }
314
315 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
316 {
317 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
318 return MD_NEW_SIZE_BLOCKS(size);
319 }
320
321 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
322 {
323 sector_t size;
324
325 size = rdev->sb_offset;
326
327 if (chunk_size)
328 size &= ~((sector_t)chunk_size/1024 - 1);
329 return size;
330 }
331
332 static int alloc_disk_sb(mdk_rdev_t * rdev)
333 {
334 if (rdev->sb_page)
335 MD_BUG();
336
337 rdev->sb_page = alloc_page(GFP_KERNEL);
338 if (!rdev->sb_page) {
339 printk(KERN_ALERT "md: out of memory.\n");
340 return -EINVAL;
341 }
342
343 return 0;
344 }
345
346 static void free_disk_sb(mdk_rdev_t * rdev)
347 {
348 if (rdev->sb_page) {
349 put_page(rdev->sb_page);
350 rdev->sb_loaded = 0;
351 rdev->sb_page = NULL;
352 rdev->sb_offset = 0;
353 rdev->size = 0;
354 }
355 }
356
357
358 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
359 {
360 mdk_rdev_t *rdev = bio->bi_private;
361 mddev_t *mddev = rdev->mddev;
362 if (bio->bi_size)
363 return 1;
364
365 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
366 md_error(mddev, rdev);
367
368 if (atomic_dec_and_test(&mddev->pending_writes))
369 wake_up(&mddev->sb_wait);
370 bio_put(bio);
371 return 0;
372 }
373
374 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
375 {
376 struct bio *bio2 = bio->bi_private;
377 mdk_rdev_t *rdev = bio2->bi_private;
378 mddev_t *mddev = rdev->mddev;
379 if (bio->bi_size)
380 return 1;
381
382 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
383 error == -EOPNOTSUPP) {
384 unsigned long flags;
385 /* barriers don't appear to be supported :-( */
386 set_bit(BarriersNotsupp, &rdev->flags);
387 mddev->barriers_work = 0;
388 spin_lock_irqsave(&mddev->write_lock, flags);
389 bio2->bi_next = mddev->biolist;
390 mddev->biolist = bio2;
391 spin_unlock_irqrestore(&mddev->write_lock, flags);
392 wake_up(&mddev->sb_wait);
393 bio_put(bio);
394 return 0;
395 }
396 bio_put(bio2);
397 bio->bi_private = rdev;
398 return super_written(bio, bytes_done, error);
399 }
400
401 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
402 sector_t sector, int size, struct page *page)
403 {
404 /* write first size bytes of page to sector of rdev
405 * Increment mddev->pending_writes before returning
406 * and decrement it on completion, waking up sb_wait
407 * if zero is reached.
408 * If an error occurred, call md_error
409 *
410 * As we might need to resubmit the request if BIO_RW_BARRIER
411 * causes ENOTSUPP, we allocate a spare bio...
412 */
413 struct bio *bio = bio_alloc(GFP_NOIO, 1);
414 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
415
416 bio->bi_bdev = rdev->bdev;
417 bio->bi_sector = sector;
418 bio_add_page(bio, page, size, 0);
419 bio->bi_private = rdev;
420 bio->bi_end_io = super_written;
421 bio->bi_rw = rw;
422
423 atomic_inc(&mddev->pending_writes);
424 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
425 struct bio *rbio;
426 rw |= (1<<BIO_RW_BARRIER);
427 rbio = bio_clone(bio, GFP_NOIO);
428 rbio->bi_private = bio;
429 rbio->bi_end_io = super_written_barrier;
430 submit_bio(rw, rbio);
431 } else
432 submit_bio(rw, bio);
433 }
434
435 void md_super_wait(mddev_t *mddev)
436 {
437 /* wait for all superblock writes that were scheduled to complete.
438 * if any had to be retried (due to BARRIER problems), retry them
439 */
440 DEFINE_WAIT(wq);
441 for(;;) {
442 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
443 if (atomic_read(&mddev->pending_writes)==0)
444 break;
445 while (mddev->biolist) {
446 struct bio *bio;
447 spin_lock_irq(&mddev->write_lock);
448 bio = mddev->biolist;
449 mddev->biolist = bio->bi_next ;
450 bio->bi_next = NULL;
451 spin_unlock_irq(&mddev->write_lock);
452 submit_bio(bio->bi_rw, bio);
453 }
454 schedule();
455 }
456 finish_wait(&mddev->sb_wait, &wq);
457 }
458
459 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
460 {
461 if (bio->bi_size)
462 return 1;
463
464 complete((struct completion*)bio->bi_private);
465 return 0;
466 }
467
468 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
469 struct page *page, int rw)
470 {
471 struct bio *bio = bio_alloc(GFP_NOIO, 1);
472 struct completion event;
473 int ret;
474
475 rw |= (1 << BIO_RW_SYNC);
476
477 bio->bi_bdev = bdev;
478 bio->bi_sector = sector;
479 bio_add_page(bio, page, size, 0);
480 init_completion(&event);
481 bio->bi_private = &event;
482 bio->bi_end_io = bi_complete;
483 submit_bio(rw, bio);
484 wait_for_completion(&event);
485
486 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
487 bio_put(bio);
488 return ret;
489 }
490 EXPORT_SYMBOL_GPL(sync_page_io);
491
492 static int read_disk_sb(mdk_rdev_t * rdev, int size)
493 {
494 char b[BDEVNAME_SIZE];
495 if (!rdev->sb_page) {
496 MD_BUG();
497 return -EINVAL;
498 }
499 if (rdev->sb_loaded)
500 return 0;
501
502
503 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
504 goto fail;
505 rdev->sb_loaded = 1;
506 return 0;
507
508 fail:
509 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
510 bdevname(rdev->bdev,b));
511 return -EINVAL;
512 }
513
514 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
515 {
516 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
517 (sb1->set_uuid1 == sb2->set_uuid1) &&
518 (sb1->set_uuid2 == sb2->set_uuid2) &&
519 (sb1->set_uuid3 == sb2->set_uuid3))
520
521 return 1;
522
523 return 0;
524 }
525
526
527 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
528 {
529 int ret;
530 mdp_super_t *tmp1, *tmp2;
531
532 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
533 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
534
535 if (!tmp1 || !tmp2) {
536 ret = 0;
537 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
538 goto abort;
539 }
540
541 *tmp1 = *sb1;
542 *tmp2 = *sb2;
543
544 /*
545 * nr_disks is not constant
546 */
547 tmp1->nr_disks = 0;
548 tmp2->nr_disks = 0;
549
550 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
551 ret = 0;
552 else
553 ret = 1;
554
555 abort:
556 kfree(tmp1);
557 kfree(tmp2);
558 return ret;
559 }
560
561 static unsigned int calc_sb_csum(mdp_super_t * sb)
562 {
563 unsigned int disk_csum, csum;
564
565 disk_csum = sb->sb_csum;
566 sb->sb_csum = 0;
567 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
568 sb->sb_csum = disk_csum;
569 return csum;
570 }
571
572
573 /*
574 * Handle superblock details.
575 * We want to be able to handle multiple superblock formats
576 * so we have a common interface to them all, and an array of
577 * different handlers.
578 * We rely on user-space to write the initial superblock, and support
579 * reading and updating of superblocks.
580 * Interface methods are:
581 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
582 * loads and validates a superblock on dev.
583 * if refdev != NULL, compare superblocks on both devices
584 * Return:
585 * 0 - dev has a superblock that is compatible with refdev
586 * 1 - dev has a superblock that is compatible and newer than refdev
587 * so dev should be used as the refdev in future
588 * -EINVAL superblock incompatible or invalid
589 * -othererror e.g. -EIO
590 *
591 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
592 * Verify that dev is acceptable into mddev.
593 * The first time, mddev->raid_disks will be 0, and data from
594 * dev should be merged in. Subsequent calls check that dev
595 * is new enough. Return 0 or -EINVAL
596 *
597 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
598 * Update the superblock for rdev with data in mddev
599 * This does not write to disc.
600 *
601 */
602
603 struct super_type {
604 char *name;
605 struct module *owner;
606 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
607 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
608 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
609 };
610
611 /*
612 * load_super for 0.90.0
613 */
614 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
615 {
616 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
617 mdp_super_t *sb;
618 int ret;
619 sector_t sb_offset;
620
621 /*
622 * Calculate the position of the superblock,
623 * it's at the end of the disk.
624 *
625 * It also happens to be a multiple of 4Kb.
626 */
627 sb_offset = calc_dev_sboffset(rdev->bdev);
628 rdev->sb_offset = sb_offset;
629
630 ret = read_disk_sb(rdev, MD_SB_BYTES);
631 if (ret) return ret;
632
633 ret = -EINVAL;
634
635 bdevname(rdev->bdev, b);
636 sb = (mdp_super_t*)page_address(rdev->sb_page);
637
638 if (sb->md_magic != MD_SB_MAGIC) {
639 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
640 b);
641 goto abort;
642 }
643
644 if (sb->major_version != 0 ||
645 sb->minor_version != 90) {
646 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
647 sb->major_version, sb->minor_version,
648 b);
649 goto abort;
650 }
651
652 if (sb->raid_disks <= 0)
653 goto abort;
654
655 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
656 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
657 b);
658 goto abort;
659 }
660
661 rdev->preferred_minor = sb->md_minor;
662 rdev->data_offset = 0;
663 rdev->sb_size = MD_SB_BYTES;
664
665 if (sb->level == LEVEL_MULTIPATH)
666 rdev->desc_nr = -1;
667 else
668 rdev->desc_nr = sb->this_disk.number;
669
670 if (refdev == 0)
671 ret = 1;
672 else {
673 __u64 ev1, ev2;
674 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
675 if (!uuid_equal(refsb, sb)) {
676 printk(KERN_WARNING "md: %s has different UUID to %s\n",
677 b, bdevname(refdev->bdev,b2));
678 goto abort;
679 }
680 if (!sb_equal(refsb, sb)) {
681 printk(KERN_WARNING "md: %s has same UUID"
682 " but different superblock to %s\n",
683 b, bdevname(refdev->bdev, b2));
684 goto abort;
685 }
686 ev1 = md_event(sb);
687 ev2 = md_event(refsb);
688 if (ev1 > ev2)
689 ret = 1;
690 else
691 ret = 0;
692 }
693 rdev->size = calc_dev_size(rdev, sb->chunk_size);
694
695 abort:
696 return ret;
697 }
698
699 /*
700 * validate_super for 0.90.0
701 */
702 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
703 {
704 mdp_disk_t *desc;
705 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
706
707 rdev->raid_disk = -1;
708 rdev->flags = 0;
709 if (mddev->raid_disks == 0) {
710 mddev->major_version = 0;
711 mddev->minor_version = sb->minor_version;
712 mddev->patch_version = sb->patch_version;
713 mddev->persistent = ! sb->not_persistent;
714 mddev->chunk_size = sb->chunk_size;
715 mddev->ctime = sb->ctime;
716 mddev->utime = sb->utime;
717 mddev->level = sb->level;
718 mddev->layout = sb->layout;
719 mddev->raid_disks = sb->raid_disks;
720 mddev->size = sb->size;
721 mddev->events = md_event(sb);
722 mddev->bitmap_offset = 0;
723 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
724
725 if (sb->state & (1<<MD_SB_CLEAN))
726 mddev->recovery_cp = MaxSector;
727 else {
728 if (sb->events_hi == sb->cp_events_hi &&
729 sb->events_lo == sb->cp_events_lo) {
730 mddev->recovery_cp = sb->recovery_cp;
731 } else
732 mddev->recovery_cp = 0;
733 }
734
735 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
736 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
737 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
738 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
739
740 mddev->max_disks = MD_SB_DISKS;
741
742 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
743 mddev->bitmap_file == NULL) {
744 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
745 && mddev->level != 10) {
746 /* FIXME use a better test */
747 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
748 return -EINVAL;
749 }
750 mddev->bitmap_offset = mddev->default_bitmap_offset;
751 }
752
753 } else if (mddev->pers == NULL) {
754 /* Insist on good event counter while assembling */
755 __u64 ev1 = md_event(sb);
756 ++ev1;
757 if (ev1 < mddev->events)
758 return -EINVAL;
759 } else if (mddev->bitmap) {
760 /* if adding to array with a bitmap, then we can accept an
761 * older device ... but not too old.
762 */
763 __u64 ev1 = md_event(sb);
764 if (ev1 < mddev->bitmap->events_cleared)
765 return 0;
766 } else /* just a hot-add of a new device, leave raid_disk at -1 */
767 return 0;
768
769 if (mddev->level != LEVEL_MULTIPATH) {
770 desc = sb->disks + rdev->desc_nr;
771
772 if (desc->state & (1<<MD_DISK_FAULTY))
773 set_bit(Faulty, &rdev->flags);
774 else if (desc->state & (1<<MD_DISK_SYNC) &&
775 desc->raid_disk < mddev->raid_disks) {
776 set_bit(In_sync, &rdev->flags);
777 rdev->raid_disk = desc->raid_disk;
778 }
779 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
780 set_bit(WriteMostly, &rdev->flags);
781 } else /* MULTIPATH are always insync */
782 set_bit(In_sync, &rdev->flags);
783 return 0;
784 }
785
786 /*
787 * sync_super for 0.90.0
788 */
789 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
790 {
791 mdp_super_t *sb;
792 struct list_head *tmp;
793 mdk_rdev_t *rdev2;
794 int next_spare = mddev->raid_disks;
795
796
797 /* make rdev->sb match mddev data..
798 *
799 * 1/ zero out disks
800 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
801 * 3/ any empty disks < next_spare become removed
802 *
803 * disks[0] gets initialised to REMOVED because
804 * we cannot be sure from other fields if it has
805 * been initialised or not.
806 */
807 int i;
808 int active=0, working=0,failed=0,spare=0,nr_disks=0;
809
810 rdev->sb_size = MD_SB_BYTES;
811
812 sb = (mdp_super_t*)page_address(rdev->sb_page);
813
814 memset(sb, 0, sizeof(*sb));
815
816 sb->md_magic = MD_SB_MAGIC;
817 sb->major_version = mddev->major_version;
818 sb->minor_version = mddev->minor_version;
819 sb->patch_version = mddev->patch_version;
820 sb->gvalid_words = 0; /* ignored */
821 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
822 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
823 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
824 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
825
826 sb->ctime = mddev->ctime;
827 sb->level = mddev->level;
828 sb->size = mddev->size;
829 sb->raid_disks = mddev->raid_disks;
830 sb->md_minor = mddev->md_minor;
831 sb->not_persistent = !mddev->persistent;
832 sb->utime = mddev->utime;
833 sb->state = 0;
834 sb->events_hi = (mddev->events>>32);
835 sb->events_lo = (u32)mddev->events;
836
837 if (mddev->in_sync)
838 {
839 sb->recovery_cp = mddev->recovery_cp;
840 sb->cp_events_hi = (mddev->events>>32);
841 sb->cp_events_lo = (u32)mddev->events;
842 if (mddev->recovery_cp == MaxSector)
843 sb->state = (1<< MD_SB_CLEAN);
844 } else
845 sb->recovery_cp = 0;
846
847 sb->layout = mddev->layout;
848 sb->chunk_size = mddev->chunk_size;
849
850 if (mddev->bitmap && mddev->bitmap_file == NULL)
851 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
852
853 sb->disks[0].state = (1<<MD_DISK_REMOVED);
854 ITERATE_RDEV(mddev,rdev2,tmp) {
855 mdp_disk_t *d;
856 int desc_nr;
857 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
858 && !test_bit(Faulty, &rdev2->flags))
859 desc_nr = rdev2->raid_disk;
860 else
861 desc_nr = next_spare++;
862 rdev2->desc_nr = desc_nr;
863 d = &sb->disks[rdev2->desc_nr];
864 nr_disks++;
865 d->number = rdev2->desc_nr;
866 d->major = MAJOR(rdev2->bdev->bd_dev);
867 d->minor = MINOR(rdev2->bdev->bd_dev);
868 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
869 && !test_bit(Faulty, &rdev2->flags))
870 d->raid_disk = rdev2->raid_disk;
871 else
872 d->raid_disk = rdev2->desc_nr; /* compatibility */
873 if (test_bit(Faulty, &rdev2->flags)) {
874 d->state = (1<<MD_DISK_FAULTY);
875 failed++;
876 } else if (test_bit(In_sync, &rdev2->flags)) {
877 d->state = (1<<MD_DISK_ACTIVE);
878 d->state |= (1<<MD_DISK_SYNC);
879 active++;
880 working++;
881 } else {
882 d->state = 0;
883 spare++;
884 working++;
885 }
886 if (test_bit(WriteMostly, &rdev2->flags))
887 d->state |= (1<<MD_DISK_WRITEMOSTLY);
888 }
889 /* now set the "removed" and "faulty" bits on any missing devices */
890 for (i=0 ; i < mddev->raid_disks ; i++) {
891 mdp_disk_t *d = &sb->disks[i];
892 if (d->state == 0 && d->number == 0) {
893 d->number = i;
894 d->raid_disk = i;
895 d->state = (1<<MD_DISK_REMOVED);
896 d->state |= (1<<MD_DISK_FAULTY);
897 failed++;
898 }
899 }
900 sb->nr_disks = nr_disks;
901 sb->active_disks = active;
902 sb->working_disks = working;
903 sb->failed_disks = failed;
904 sb->spare_disks = spare;
905
906 sb->this_disk = sb->disks[rdev->desc_nr];
907 sb->sb_csum = calc_sb_csum(sb);
908 }
909
910 /*
911 * version 1 superblock
912 */
913
914 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
915 {
916 unsigned int disk_csum, csum;
917 unsigned long long newcsum;
918 int size = 256 + le32_to_cpu(sb->max_dev)*2;
919 unsigned int *isuper = (unsigned int*)sb;
920 int i;
921
922 disk_csum = sb->sb_csum;
923 sb->sb_csum = 0;
924 newcsum = 0;
925 for (i=0; size>=4; size -= 4 )
926 newcsum += le32_to_cpu(*isuper++);
927
928 if (size == 2)
929 newcsum += le16_to_cpu(*(unsigned short*) isuper);
930
931 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
932 sb->sb_csum = disk_csum;
933 return cpu_to_le32(csum);
934 }
935
936 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
937 {
938 struct mdp_superblock_1 *sb;
939 int ret;
940 sector_t sb_offset;
941 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
942 int bmask;
943
944 /*
945 * Calculate the position of the superblock.
946 * It is always aligned to a 4K boundary and
947 * depeding on minor_version, it can be:
948 * 0: At least 8K, but less than 12K, from end of device
949 * 1: At start of device
950 * 2: 4K from start of device.
951 */
952 switch(minor_version) {
953 case 0:
954 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
955 sb_offset -= 8*2;
956 sb_offset &= ~(sector_t)(4*2-1);
957 /* convert from sectors to K */
958 sb_offset /= 2;
959 break;
960 case 1:
961 sb_offset = 0;
962 break;
963 case 2:
964 sb_offset = 4;
965 break;
966 default:
967 return -EINVAL;
968 }
969 rdev->sb_offset = sb_offset;
970
971 /* superblock is rarely larger than 1K, but it can be larger,
972 * and it is safe to read 4k, so we do that
973 */
974 ret = read_disk_sb(rdev, 4096);
975 if (ret) return ret;
976
977
978 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
979
980 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
981 sb->major_version != cpu_to_le32(1) ||
982 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
983 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
984 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
985 return -EINVAL;
986
987 if (calc_sb_1_csum(sb) != sb->sb_csum) {
988 printk("md: invalid superblock checksum on %s\n",
989 bdevname(rdev->bdev,b));
990 return -EINVAL;
991 }
992 if (le64_to_cpu(sb->data_size) < 10) {
993 printk("md: data_size too small on %s\n",
994 bdevname(rdev->bdev,b));
995 return -EINVAL;
996 }
997 rdev->preferred_minor = 0xffff;
998 rdev->data_offset = le64_to_cpu(sb->data_offset);
999
1000 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1001 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1002 if (rdev->sb_size & bmask)
1003 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1004
1005 if (refdev == 0)
1006 return 1;
1007 else {
1008 __u64 ev1, ev2;
1009 struct mdp_superblock_1 *refsb =
1010 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1011
1012 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1013 sb->level != refsb->level ||
1014 sb->layout != refsb->layout ||
1015 sb->chunksize != refsb->chunksize) {
1016 printk(KERN_WARNING "md: %s has strangely different"
1017 " superblock to %s\n",
1018 bdevname(rdev->bdev,b),
1019 bdevname(refdev->bdev,b2));
1020 return -EINVAL;
1021 }
1022 ev1 = le64_to_cpu(sb->events);
1023 ev2 = le64_to_cpu(refsb->events);
1024
1025 if (ev1 > ev2)
1026 return 1;
1027 }
1028 if (minor_version)
1029 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1030 else
1031 rdev->size = rdev->sb_offset;
1032 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1033 return -EINVAL;
1034 rdev->size = le64_to_cpu(sb->data_size)/2;
1035 if (le32_to_cpu(sb->chunksize))
1036 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1037 return 0;
1038 }
1039
1040 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1041 {
1042 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1043
1044 rdev->raid_disk = -1;
1045 rdev->flags = 0;
1046 if (mddev->raid_disks == 0) {
1047 mddev->major_version = 1;
1048 mddev->patch_version = 0;
1049 mddev->persistent = 1;
1050 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1051 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1052 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1053 mddev->level = le32_to_cpu(sb->level);
1054 mddev->layout = le32_to_cpu(sb->layout);
1055 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1056 mddev->size = le64_to_cpu(sb->size)/2;
1057 mddev->events = le64_to_cpu(sb->events);
1058 mddev->bitmap_offset = 0;
1059 mddev->default_bitmap_offset = 1024;
1060
1061 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1062 memcpy(mddev->uuid, sb->set_uuid, 16);
1063
1064 mddev->max_disks = (4096-256)/2;
1065
1066 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1067 mddev->bitmap_file == NULL ) {
1068 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1069 && mddev->level != 10) {
1070 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1071 return -EINVAL;
1072 }
1073 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1074 }
1075 } else if (mddev->pers == NULL) {
1076 /* Insist of good event counter while assembling */
1077 __u64 ev1 = le64_to_cpu(sb->events);
1078 ++ev1;
1079 if (ev1 < mddev->events)
1080 return -EINVAL;
1081 } else if (mddev->bitmap) {
1082 /* If adding to array with a bitmap, then we can accept an
1083 * older device, but not too old.
1084 */
1085 __u64 ev1 = le64_to_cpu(sb->events);
1086 if (ev1 < mddev->bitmap->events_cleared)
1087 return 0;
1088 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1089 return 0;
1090
1091 if (mddev->level != LEVEL_MULTIPATH) {
1092 int role;
1093 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1094 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1095 switch(role) {
1096 case 0xffff: /* spare */
1097 break;
1098 case 0xfffe: /* faulty */
1099 set_bit(Faulty, &rdev->flags);
1100 break;
1101 default:
1102 set_bit(In_sync, &rdev->flags);
1103 rdev->raid_disk = role;
1104 break;
1105 }
1106 if (sb->devflags & WriteMostly1)
1107 set_bit(WriteMostly, &rdev->flags);
1108 } else /* MULTIPATH are always insync */
1109 set_bit(In_sync, &rdev->flags);
1110
1111 return 0;
1112 }
1113
1114 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1115 {
1116 struct mdp_superblock_1 *sb;
1117 struct list_head *tmp;
1118 mdk_rdev_t *rdev2;
1119 int max_dev, i;
1120 /* make rdev->sb match mddev and rdev data. */
1121
1122 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1123
1124 sb->feature_map = 0;
1125 sb->pad0 = 0;
1126 memset(sb->pad1, 0, sizeof(sb->pad1));
1127 memset(sb->pad2, 0, sizeof(sb->pad2));
1128 memset(sb->pad3, 0, sizeof(sb->pad3));
1129
1130 sb->utime = cpu_to_le64((__u64)mddev->utime);
1131 sb->events = cpu_to_le64(mddev->events);
1132 if (mddev->in_sync)
1133 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1134 else
1135 sb->resync_offset = cpu_to_le64(0);
1136
1137 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1138 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1139 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1140 }
1141
1142 max_dev = 0;
1143 ITERATE_RDEV(mddev,rdev2,tmp)
1144 if (rdev2->desc_nr+1 > max_dev)
1145 max_dev = rdev2->desc_nr+1;
1146
1147 sb->max_dev = cpu_to_le32(max_dev);
1148 for (i=0; i<max_dev;i++)
1149 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1150
1151 ITERATE_RDEV(mddev,rdev2,tmp) {
1152 i = rdev2->desc_nr;
1153 if (test_bit(Faulty, &rdev2->flags))
1154 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1155 else if (test_bit(In_sync, &rdev2->flags))
1156 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1157 else
1158 sb->dev_roles[i] = cpu_to_le16(0xffff);
1159 }
1160
1161 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1162 sb->sb_csum = calc_sb_1_csum(sb);
1163 }
1164
1165
1166 static struct super_type super_types[] = {
1167 [0] = {
1168 .name = "0.90.0",
1169 .owner = THIS_MODULE,
1170 .load_super = super_90_load,
1171 .validate_super = super_90_validate,
1172 .sync_super = super_90_sync,
1173 },
1174 [1] = {
1175 .name = "md-1",
1176 .owner = THIS_MODULE,
1177 .load_super = super_1_load,
1178 .validate_super = super_1_validate,
1179 .sync_super = super_1_sync,
1180 },
1181 };
1182
1183 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1184 {
1185 struct list_head *tmp;
1186 mdk_rdev_t *rdev;
1187
1188 ITERATE_RDEV(mddev,rdev,tmp)
1189 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1190 return rdev;
1191
1192 return NULL;
1193 }
1194
1195 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1196 {
1197 struct list_head *tmp;
1198 mdk_rdev_t *rdev;
1199
1200 ITERATE_RDEV(mddev1,rdev,tmp)
1201 if (match_dev_unit(mddev2, rdev))
1202 return 1;
1203
1204 return 0;
1205 }
1206
1207 static LIST_HEAD(pending_raid_disks);
1208
1209 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1210 {
1211 mdk_rdev_t *same_pdev;
1212 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1213 struct kobject *ko;
1214
1215 if (rdev->mddev) {
1216 MD_BUG();
1217 return -EINVAL;
1218 }
1219 same_pdev = match_dev_unit(mddev, rdev);
1220 if (same_pdev)
1221 printk(KERN_WARNING
1222 "%s: WARNING: %s appears to be on the same physical"
1223 " disk as %s. True\n protection against single-disk"
1224 " failure might be compromised.\n",
1225 mdname(mddev), bdevname(rdev->bdev,b),
1226 bdevname(same_pdev->bdev,b2));
1227
1228 /* Verify rdev->desc_nr is unique.
1229 * If it is -1, assign a free number, else
1230 * check number is not in use
1231 */
1232 if (rdev->desc_nr < 0) {
1233 int choice = 0;
1234 if (mddev->pers) choice = mddev->raid_disks;
1235 while (find_rdev_nr(mddev, choice))
1236 choice++;
1237 rdev->desc_nr = choice;
1238 } else {
1239 if (find_rdev_nr(mddev, rdev->desc_nr))
1240 return -EBUSY;
1241 }
1242 bdevname(rdev->bdev,b);
1243 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1244 return -ENOMEM;
1245
1246 list_add(&rdev->same_set, &mddev->disks);
1247 rdev->mddev = mddev;
1248 printk(KERN_INFO "md: bind<%s>\n", b);
1249
1250 rdev->kobj.parent = &mddev->kobj;
1251 kobject_add(&rdev->kobj);
1252
1253 if (rdev->bdev->bd_part)
1254 ko = &rdev->bdev->bd_part->kobj;
1255 else
1256 ko = &rdev->bdev->bd_disk->kobj;
1257 sysfs_create_link(&rdev->kobj, ko, "block");
1258 return 0;
1259 }
1260
1261 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1262 {
1263 char b[BDEVNAME_SIZE];
1264 if (!rdev->mddev) {
1265 MD_BUG();
1266 return;
1267 }
1268 list_del_init(&rdev->same_set);
1269 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1270 rdev->mddev = NULL;
1271 sysfs_remove_link(&rdev->kobj, "block");
1272 kobject_del(&rdev->kobj);
1273 }
1274
1275 /*
1276 * prevent the device from being mounted, repartitioned or
1277 * otherwise reused by a RAID array (or any other kernel
1278 * subsystem), by bd_claiming the device.
1279 */
1280 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1281 {
1282 int err = 0;
1283 struct block_device *bdev;
1284 char b[BDEVNAME_SIZE];
1285
1286 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1287 if (IS_ERR(bdev)) {
1288 printk(KERN_ERR "md: could not open %s.\n",
1289 __bdevname(dev, b));
1290 return PTR_ERR(bdev);
1291 }
1292 err = bd_claim(bdev, rdev);
1293 if (err) {
1294 printk(KERN_ERR "md: could not bd_claim %s.\n",
1295 bdevname(bdev, b));
1296 blkdev_put(bdev);
1297 return err;
1298 }
1299 rdev->bdev = bdev;
1300 return err;
1301 }
1302
1303 static void unlock_rdev(mdk_rdev_t *rdev)
1304 {
1305 struct block_device *bdev = rdev->bdev;
1306 rdev->bdev = NULL;
1307 if (!bdev)
1308 MD_BUG();
1309 bd_release(bdev);
1310 blkdev_put(bdev);
1311 }
1312
1313 void md_autodetect_dev(dev_t dev);
1314
1315 static void export_rdev(mdk_rdev_t * rdev)
1316 {
1317 char b[BDEVNAME_SIZE];
1318 printk(KERN_INFO "md: export_rdev(%s)\n",
1319 bdevname(rdev->bdev,b));
1320 if (rdev->mddev)
1321 MD_BUG();
1322 free_disk_sb(rdev);
1323 list_del_init(&rdev->same_set);
1324 #ifndef MODULE
1325 md_autodetect_dev(rdev->bdev->bd_dev);
1326 #endif
1327 unlock_rdev(rdev);
1328 kobject_put(&rdev->kobj);
1329 }
1330
1331 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1332 {
1333 unbind_rdev_from_array(rdev);
1334 export_rdev(rdev);
1335 }
1336
1337 static void export_array(mddev_t *mddev)
1338 {
1339 struct list_head *tmp;
1340 mdk_rdev_t *rdev;
1341
1342 ITERATE_RDEV(mddev,rdev,tmp) {
1343 if (!rdev->mddev) {
1344 MD_BUG();
1345 continue;
1346 }
1347 kick_rdev_from_array(rdev);
1348 }
1349 if (!list_empty(&mddev->disks))
1350 MD_BUG();
1351 mddev->raid_disks = 0;
1352 mddev->major_version = 0;
1353 }
1354
1355 static void print_desc(mdp_disk_t *desc)
1356 {
1357 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1358 desc->major,desc->minor,desc->raid_disk,desc->state);
1359 }
1360
1361 static void print_sb(mdp_super_t *sb)
1362 {
1363 int i;
1364
1365 printk(KERN_INFO
1366 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1367 sb->major_version, sb->minor_version, sb->patch_version,
1368 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1369 sb->ctime);
1370 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1371 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1372 sb->md_minor, sb->layout, sb->chunk_size);
1373 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1374 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1375 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1376 sb->failed_disks, sb->spare_disks,
1377 sb->sb_csum, (unsigned long)sb->events_lo);
1378
1379 printk(KERN_INFO);
1380 for (i = 0; i < MD_SB_DISKS; i++) {
1381 mdp_disk_t *desc;
1382
1383 desc = sb->disks + i;
1384 if (desc->number || desc->major || desc->minor ||
1385 desc->raid_disk || (desc->state && (desc->state != 4))) {
1386 printk(" D %2d: ", i);
1387 print_desc(desc);
1388 }
1389 }
1390 printk(KERN_INFO "md: THIS: ");
1391 print_desc(&sb->this_disk);
1392
1393 }
1394
1395 static void print_rdev(mdk_rdev_t *rdev)
1396 {
1397 char b[BDEVNAME_SIZE];
1398 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1399 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1400 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1401 rdev->desc_nr);
1402 if (rdev->sb_loaded) {
1403 printk(KERN_INFO "md: rdev superblock:\n");
1404 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1405 } else
1406 printk(KERN_INFO "md: no rdev superblock!\n");
1407 }
1408
1409 void md_print_devices(void)
1410 {
1411 struct list_head *tmp, *tmp2;
1412 mdk_rdev_t *rdev;
1413 mddev_t *mddev;
1414 char b[BDEVNAME_SIZE];
1415
1416 printk("\n");
1417 printk("md: **********************************\n");
1418 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1419 printk("md: **********************************\n");
1420 ITERATE_MDDEV(mddev,tmp) {
1421
1422 if (mddev->bitmap)
1423 bitmap_print_sb(mddev->bitmap);
1424 else
1425 printk("%s: ", mdname(mddev));
1426 ITERATE_RDEV(mddev,rdev,tmp2)
1427 printk("<%s>", bdevname(rdev->bdev,b));
1428 printk("\n");
1429
1430 ITERATE_RDEV(mddev,rdev,tmp2)
1431 print_rdev(rdev);
1432 }
1433 printk("md: **********************************\n");
1434 printk("\n");
1435 }
1436
1437
1438 static void sync_sbs(mddev_t * mddev)
1439 {
1440 mdk_rdev_t *rdev;
1441 struct list_head *tmp;
1442
1443 ITERATE_RDEV(mddev,rdev,tmp) {
1444 super_types[mddev->major_version].
1445 sync_super(mddev, rdev);
1446 rdev->sb_loaded = 1;
1447 }
1448 }
1449
1450 static void md_update_sb(mddev_t * mddev)
1451 {
1452 int err;
1453 struct list_head *tmp;
1454 mdk_rdev_t *rdev;
1455 int sync_req;
1456
1457 repeat:
1458 spin_lock_irq(&mddev->write_lock);
1459 sync_req = mddev->in_sync;
1460 mddev->utime = get_seconds();
1461 mddev->events ++;
1462
1463 if (!mddev->events) {
1464 /*
1465 * oops, this 64-bit counter should never wrap.
1466 * Either we are in around ~1 trillion A.C., assuming
1467 * 1 reboot per second, or we have a bug:
1468 */
1469 MD_BUG();
1470 mddev->events --;
1471 }
1472 mddev->sb_dirty = 2;
1473 sync_sbs(mddev);
1474
1475 /*
1476 * do not write anything to disk if using
1477 * nonpersistent superblocks
1478 */
1479 if (!mddev->persistent) {
1480 mddev->sb_dirty = 0;
1481 spin_unlock_irq(&mddev->write_lock);
1482 wake_up(&mddev->sb_wait);
1483 return;
1484 }
1485 spin_unlock_irq(&mddev->write_lock);
1486
1487 dprintk(KERN_INFO
1488 "md: updating %s RAID superblock on device (in sync %d)\n",
1489 mdname(mddev),mddev->in_sync);
1490
1491 err = bitmap_update_sb(mddev->bitmap);
1492 ITERATE_RDEV(mddev,rdev,tmp) {
1493 char b[BDEVNAME_SIZE];
1494 dprintk(KERN_INFO "md: ");
1495 if (test_bit(Faulty, &rdev->flags))
1496 dprintk("(skipping faulty ");
1497
1498 dprintk("%s ", bdevname(rdev->bdev,b));
1499 if (!test_bit(Faulty, &rdev->flags)) {
1500 md_super_write(mddev,rdev,
1501 rdev->sb_offset<<1, rdev->sb_size,
1502 rdev->sb_page);
1503 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1504 bdevname(rdev->bdev,b),
1505 (unsigned long long)rdev->sb_offset);
1506
1507 } else
1508 dprintk(")\n");
1509 if (mddev->level == LEVEL_MULTIPATH)
1510 /* only need to write one superblock... */
1511 break;
1512 }
1513 md_super_wait(mddev);
1514 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1515
1516 spin_lock_irq(&mddev->write_lock);
1517 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1518 /* have to write it out again */
1519 spin_unlock_irq(&mddev->write_lock);
1520 goto repeat;
1521 }
1522 mddev->sb_dirty = 0;
1523 spin_unlock_irq(&mddev->write_lock);
1524 wake_up(&mddev->sb_wait);
1525
1526 }
1527
1528 /* words written to sysfs files may, or my not, be \n terminated.
1529 * We want to accept with case. For this we use cmd_match.
1530 */
1531 static int cmd_match(const char *cmd, const char *str)
1532 {
1533 /* See if cmd, written into a sysfs file, matches
1534 * str. They must either be the same, or cmd can
1535 * have a trailing newline
1536 */
1537 while (*cmd && *str && *cmd == *str) {
1538 cmd++;
1539 str++;
1540 }
1541 if (*cmd == '\n')
1542 cmd++;
1543 if (*str || *cmd)
1544 return 0;
1545 return 1;
1546 }
1547
1548 struct rdev_sysfs_entry {
1549 struct attribute attr;
1550 ssize_t (*show)(mdk_rdev_t *, char *);
1551 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1552 };
1553
1554 static ssize_t
1555 state_show(mdk_rdev_t *rdev, char *page)
1556 {
1557 char *sep = "";
1558 int len=0;
1559
1560 if (test_bit(Faulty, &rdev->flags)) {
1561 len+= sprintf(page+len, "%sfaulty",sep);
1562 sep = ",";
1563 }
1564 if (test_bit(In_sync, &rdev->flags)) {
1565 len += sprintf(page+len, "%sin_sync",sep);
1566 sep = ",";
1567 }
1568 if (!test_bit(Faulty, &rdev->flags) &&
1569 !test_bit(In_sync, &rdev->flags)) {
1570 len += sprintf(page+len, "%sspare", sep);
1571 sep = ",";
1572 }
1573 return len+sprintf(page+len, "\n");
1574 }
1575
1576 static struct rdev_sysfs_entry
1577 rdev_state = __ATTR_RO(state);
1578
1579 static ssize_t
1580 super_show(mdk_rdev_t *rdev, char *page)
1581 {
1582 if (rdev->sb_loaded && rdev->sb_size) {
1583 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1584 return rdev->sb_size;
1585 } else
1586 return 0;
1587 }
1588 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1589
1590 static struct attribute *rdev_default_attrs[] = {
1591 &rdev_state.attr,
1592 &rdev_super.attr,
1593 NULL,
1594 };
1595 static ssize_t
1596 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1597 {
1598 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1599 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1600
1601 if (!entry->show)
1602 return -EIO;
1603 return entry->show(rdev, page);
1604 }
1605
1606 static ssize_t
1607 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1608 const char *page, size_t length)
1609 {
1610 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1611 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1612
1613 if (!entry->store)
1614 return -EIO;
1615 return entry->store(rdev, page, length);
1616 }
1617
1618 static void rdev_free(struct kobject *ko)
1619 {
1620 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1621 kfree(rdev);
1622 }
1623 static struct sysfs_ops rdev_sysfs_ops = {
1624 .show = rdev_attr_show,
1625 .store = rdev_attr_store,
1626 };
1627 static struct kobj_type rdev_ktype = {
1628 .release = rdev_free,
1629 .sysfs_ops = &rdev_sysfs_ops,
1630 .default_attrs = rdev_default_attrs,
1631 };
1632
1633 /*
1634 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1635 *
1636 * mark the device faulty if:
1637 *
1638 * - the device is nonexistent (zero size)
1639 * - the device has no valid superblock
1640 *
1641 * a faulty rdev _never_ has rdev->sb set.
1642 */
1643 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1644 {
1645 char b[BDEVNAME_SIZE];
1646 int err;
1647 mdk_rdev_t *rdev;
1648 sector_t size;
1649
1650 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1651 if (!rdev) {
1652 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1653 return ERR_PTR(-ENOMEM);
1654 }
1655
1656 if ((err = alloc_disk_sb(rdev)))
1657 goto abort_free;
1658
1659 err = lock_rdev(rdev, newdev);
1660 if (err)
1661 goto abort_free;
1662
1663 rdev->kobj.parent = NULL;
1664 rdev->kobj.ktype = &rdev_ktype;
1665 kobject_init(&rdev->kobj);
1666
1667 rdev->desc_nr = -1;
1668 rdev->flags = 0;
1669 rdev->data_offset = 0;
1670 atomic_set(&rdev->nr_pending, 0);
1671 atomic_set(&rdev->read_errors, 0);
1672
1673 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1674 if (!size) {
1675 printk(KERN_WARNING
1676 "md: %s has zero or unknown size, marking faulty!\n",
1677 bdevname(rdev->bdev,b));
1678 err = -EINVAL;
1679 goto abort_free;
1680 }
1681
1682 if (super_format >= 0) {
1683 err = super_types[super_format].
1684 load_super(rdev, NULL, super_minor);
1685 if (err == -EINVAL) {
1686 printk(KERN_WARNING
1687 "md: %s has invalid sb, not importing!\n",
1688 bdevname(rdev->bdev,b));
1689 goto abort_free;
1690 }
1691 if (err < 0) {
1692 printk(KERN_WARNING
1693 "md: could not read %s's sb, not importing!\n",
1694 bdevname(rdev->bdev,b));
1695 goto abort_free;
1696 }
1697 }
1698 INIT_LIST_HEAD(&rdev->same_set);
1699
1700 return rdev;
1701
1702 abort_free:
1703 if (rdev->sb_page) {
1704 if (rdev->bdev)
1705 unlock_rdev(rdev);
1706 free_disk_sb(rdev);
1707 }
1708 kfree(rdev);
1709 return ERR_PTR(err);
1710 }
1711
1712 /*
1713 * Check a full RAID array for plausibility
1714 */
1715
1716
1717 static void analyze_sbs(mddev_t * mddev)
1718 {
1719 int i;
1720 struct list_head *tmp;
1721 mdk_rdev_t *rdev, *freshest;
1722 char b[BDEVNAME_SIZE];
1723
1724 freshest = NULL;
1725 ITERATE_RDEV(mddev,rdev,tmp)
1726 switch (super_types[mddev->major_version].
1727 load_super(rdev, freshest, mddev->minor_version)) {
1728 case 1:
1729 freshest = rdev;
1730 break;
1731 case 0:
1732 break;
1733 default:
1734 printk( KERN_ERR \
1735 "md: fatal superblock inconsistency in %s"
1736 " -- removing from array\n",
1737 bdevname(rdev->bdev,b));
1738 kick_rdev_from_array(rdev);
1739 }
1740
1741
1742 super_types[mddev->major_version].
1743 validate_super(mddev, freshest);
1744
1745 i = 0;
1746 ITERATE_RDEV(mddev,rdev,tmp) {
1747 if (rdev != freshest)
1748 if (super_types[mddev->major_version].
1749 validate_super(mddev, rdev)) {
1750 printk(KERN_WARNING "md: kicking non-fresh %s"
1751 " from array!\n",
1752 bdevname(rdev->bdev,b));
1753 kick_rdev_from_array(rdev);
1754 continue;
1755 }
1756 if (mddev->level == LEVEL_MULTIPATH) {
1757 rdev->desc_nr = i++;
1758 rdev->raid_disk = rdev->desc_nr;
1759 set_bit(In_sync, &rdev->flags);
1760 }
1761 }
1762
1763
1764
1765 if (mddev->recovery_cp != MaxSector &&
1766 mddev->level >= 1)
1767 printk(KERN_ERR "md: %s: raid array is not clean"
1768 " -- starting background reconstruction\n",
1769 mdname(mddev));
1770
1771 }
1772
1773 static ssize_t
1774 level_show(mddev_t *mddev, char *page)
1775 {
1776 struct mdk_personality *p = mddev->pers;
1777 if (p == NULL && mddev->raid_disks == 0)
1778 return 0;
1779 if (mddev->level >= 0)
1780 return sprintf(page, "raid%d\n", mddev->level);
1781 else
1782 return sprintf(page, "%s\n", p->name);
1783 }
1784
1785 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1786
1787 static ssize_t
1788 raid_disks_show(mddev_t *mddev, char *page)
1789 {
1790 if (mddev->raid_disks == 0)
1791 return 0;
1792 return sprintf(page, "%d\n", mddev->raid_disks);
1793 }
1794
1795 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1796
1797 static ssize_t
1798 chunk_size_show(mddev_t *mddev, char *page)
1799 {
1800 return sprintf(page, "%d\n", mddev->chunk_size);
1801 }
1802
1803 static ssize_t
1804 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1805 {
1806 /* can only set chunk_size if array is not yet active */
1807 char *e;
1808 unsigned long n = simple_strtoul(buf, &e, 10);
1809
1810 if (mddev->pers)
1811 return -EBUSY;
1812 if (!*buf || (*e && *e != '\n'))
1813 return -EINVAL;
1814
1815 mddev->chunk_size = n;
1816 return len;
1817 }
1818 static struct md_sysfs_entry md_chunk_size =
1819 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
1820
1821
1822 static ssize_t
1823 action_show(mddev_t *mddev, char *page)
1824 {
1825 char *type = "idle";
1826 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1827 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1828 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1829 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1830 type = "resync";
1831 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1832 type = "check";
1833 else
1834 type = "repair";
1835 } else
1836 type = "recover";
1837 }
1838 return sprintf(page, "%s\n", type);
1839 }
1840
1841 static ssize_t
1842 action_store(mddev_t *mddev, const char *page, size_t len)
1843 {
1844 if (!mddev->pers || !mddev->pers->sync_request)
1845 return -EINVAL;
1846
1847 if (cmd_match(page, "idle")) {
1848 if (mddev->sync_thread) {
1849 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1850 md_unregister_thread(mddev->sync_thread);
1851 mddev->sync_thread = NULL;
1852 mddev->recovery = 0;
1853 }
1854 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1855 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1856 return -EBUSY;
1857 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
1858 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1859 else {
1860 if (cmd_match(page, "check"))
1861 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1862 else if (cmd_match(page, "repair"))
1863 return -EINVAL;
1864 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1865 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1866 }
1867 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1868 md_wakeup_thread(mddev->thread);
1869 return len;
1870 }
1871
1872 static ssize_t
1873 mismatch_cnt_show(mddev_t *mddev, char *page)
1874 {
1875 return sprintf(page, "%llu\n",
1876 (unsigned long long) mddev->resync_mismatches);
1877 }
1878
1879 static struct md_sysfs_entry
1880 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1881
1882
1883 static struct md_sysfs_entry
1884 md_mismatches = __ATTR_RO(mismatch_cnt);
1885
1886 static struct attribute *md_default_attrs[] = {
1887 &md_level.attr,
1888 &md_raid_disks.attr,
1889 &md_chunk_size.attr,
1890 NULL,
1891 };
1892
1893 static struct attribute *md_redundancy_attrs[] = {
1894 &md_scan_mode.attr,
1895 &md_mismatches.attr,
1896 NULL,
1897 };
1898 static struct attribute_group md_redundancy_group = {
1899 .name = NULL,
1900 .attrs = md_redundancy_attrs,
1901 };
1902
1903
1904 static ssize_t
1905 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1906 {
1907 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1908 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1909 ssize_t rv;
1910
1911 if (!entry->show)
1912 return -EIO;
1913 mddev_lock(mddev);
1914 rv = entry->show(mddev, page);
1915 mddev_unlock(mddev);
1916 return rv;
1917 }
1918
1919 static ssize_t
1920 md_attr_store(struct kobject *kobj, struct attribute *attr,
1921 const char *page, size_t length)
1922 {
1923 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1924 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1925 ssize_t rv;
1926
1927 if (!entry->store)
1928 return -EIO;
1929 mddev_lock(mddev);
1930 rv = entry->store(mddev, page, length);
1931 mddev_unlock(mddev);
1932 return rv;
1933 }
1934
1935 static void md_free(struct kobject *ko)
1936 {
1937 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1938 kfree(mddev);
1939 }
1940
1941 static struct sysfs_ops md_sysfs_ops = {
1942 .show = md_attr_show,
1943 .store = md_attr_store,
1944 };
1945 static struct kobj_type md_ktype = {
1946 .release = md_free,
1947 .sysfs_ops = &md_sysfs_ops,
1948 .default_attrs = md_default_attrs,
1949 };
1950
1951 int mdp_major = 0;
1952
1953 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1954 {
1955 static DECLARE_MUTEX(disks_sem);
1956 mddev_t *mddev = mddev_find(dev);
1957 struct gendisk *disk;
1958 int partitioned = (MAJOR(dev) != MD_MAJOR);
1959 int shift = partitioned ? MdpMinorShift : 0;
1960 int unit = MINOR(dev) >> shift;
1961
1962 if (!mddev)
1963 return NULL;
1964
1965 down(&disks_sem);
1966 if (mddev->gendisk) {
1967 up(&disks_sem);
1968 mddev_put(mddev);
1969 return NULL;
1970 }
1971 disk = alloc_disk(1 << shift);
1972 if (!disk) {
1973 up(&disks_sem);
1974 mddev_put(mddev);
1975 return NULL;
1976 }
1977 disk->major = MAJOR(dev);
1978 disk->first_minor = unit << shift;
1979 if (partitioned) {
1980 sprintf(disk->disk_name, "md_d%d", unit);
1981 sprintf(disk->devfs_name, "md/d%d", unit);
1982 } else {
1983 sprintf(disk->disk_name, "md%d", unit);
1984 sprintf(disk->devfs_name, "md/%d", unit);
1985 }
1986 disk->fops = &md_fops;
1987 disk->private_data = mddev;
1988 disk->queue = mddev->queue;
1989 add_disk(disk);
1990 mddev->gendisk = disk;
1991 up(&disks_sem);
1992 mddev->kobj.parent = &disk->kobj;
1993 mddev->kobj.k_name = NULL;
1994 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1995 mddev->kobj.ktype = &md_ktype;
1996 kobject_register(&mddev->kobj);
1997 return NULL;
1998 }
1999
2000 void md_wakeup_thread(mdk_thread_t *thread);
2001
2002 static void md_safemode_timeout(unsigned long data)
2003 {
2004 mddev_t *mddev = (mddev_t *) data;
2005
2006 mddev->safemode = 1;
2007 md_wakeup_thread(mddev->thread);
2008 }
2009
2010 static int start_dirty_degraded;
2011
2012 static int do_md_run(mddev_t * mddev)
2013 {
2014 int err;
2015 int chunk_size;
2016 struct list_head *tmp;
2017 mdk_rdev_t *rdev;
2018 struct gendisk *disk;
2019 struct mdk_personality *pers;
2020 char b[BDEVNAME_SIZE];
2021
2022 if (list_empty(&mddev->disks))
2023 /* cannot run an array with no devices.. */
2024 return -EINVAL;
2025
2026 if (mddev->pers)
2027 return -EBUSY;
2028
2029 /*
2030 * Analyze all RAID superblock(s)
2031 */
2032 if (!mddev->raid_disks)
2033 analyze_sbs(mddev);
2034
2035 chunk_size = mddev->chunk_size;
2036
2037 if (chunk_size) {
2038 if (chunk_size > MAX_CHUNK_SIZE) {
2039 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2040 chunk_size, MAX_CHUNK_SIZE);
2041 return -EINVAL;
2042 }
2043 /*
2044 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2045 */
2046 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2047 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2048 return -EINVAL;
2049 }
2050 if (chunk_size < PAGE_SIZE) {
2051 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2052 chunk_size, PAGE_SIZE);
2053 return -EINVAL;
2054 }
2055
2056 /* devices must have minimum size of one chunk */
2057 ITERATE_RDEV(mddev,rdev,tmp) {
2058 if (test_bit(Faulty, &rdev->flags))
2059 continue;
2060 if (rdev->size < chunk_size / 1024) {
2061 printk(KERN_WARNING
2062 "md: Dev %s smaller than chunk_size:"
2063 " %lluk < %dk\n",
2064 bdevname(rdev->bdev,b),
2065 (unsigned long long)rdev->size,
2066 chunk_size / 1024);
2067 return -EINVAL;
2068 }
2069 }
2070 }
2071
2072 #ifdef CONFIG_KMOD
2073 request_module("md-level-%d", mddev->level);
2074 #endif
2075
2076 /*
2077 * Drop all container device buffers, from now on
2078 * the only valid external interface is through the md
2079 * device.
2080 * Also find largest hardsector size
2081 */
2082 ITERATE_RDEV(mddev,rdev,tmp) {
2083 if (test_bit(Faulty, &rdev->flags))
2084 continue;
2085 sync_blockdev(rdev->bdev);
2086 invalidate_bdev(rdev->bdev, 0);
2087 }
2088
2089 md_probe(mddev->unit, NULL, NULL);
2090 disk = mddev->gendisk;
2091 if (!disk)
2092 return -ENOMEM;
2093
2094 spin_lock(&pers_lock);
2095 pers = find_pers(mddev->level);
2096 if (!pers || !try_module_get(pers->owner)) {
2097 spin_unlock(&pers_lock);
2098 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2099 mddev->level);
2100 return -EINVAL;
2101 }
2102 mddev->pers = pers;
2103 spin_unlock(&pers_lock);
2104
2105 mddev->recovery = 0;
2106 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2107 mddev->barriers_work = 1;
2108 mddev->ok_start_degraded = start_dirty_degraded;
2109
2110 if (start_readonly)
2111 mddev->ro = 2; /* read-only, but switch on first write */
2112
2113 err = mddev->pers->run(mddev);
2114 if (!err && mddev->pers->sync_request) {
2115 err = bitmap_create(mddev);
2116 if (err) {
2117 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2118 mdname(mddev), err);
2119 mddev->pers->stop(mddev);
2120 }
2121 }
2122 if (err) {
2123 printk(KERN_ERR "md: pers->run() failed ...\n");
2124 module_put(mddev->pers->owner);
2125 mddev->pers = NULL;
2126 bitmap_destroy(mddev);
2127 return err;
2128 }
2129 if (mddev->pers->sync_request)
2130 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2131 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2132 mddev->ro = 0;
2133
2134 atomic_set(&mddev->writes_pending,0);
2135 mddev->safemode = 0;
2136 mddev->safemode_timer.function = md_safemode_timeout;
2137 mddev->safemode_timer.data = (unsigned long) mddev;
2138 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2139 mddev->in_sync = 1;
2140
2141 ITERATE_RDEV(mddev,rdev,tmp)
2142 if (rdev->raid_disk >= 0) {
2143 char nm[20];
2144 sprintf(nm, "rd%d", rdev->raid_disk);
2145 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2146 }
2147
2148 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2149 md_wakeup_thread(mddev->thread);
2150
2151 if (mddev->sb_dirty)
2152 md_update_sb(mddev);
2153
2154 set_capacity(disk, mddev->array_size<<1);
2155
2156 /* If we call blk_queue_make_request here, it will
2157 * re-initialise max_sectors etc which may have been
2158 * refined inside -> run. So just set the bits we need to set.
2159 * Most initialisation happended when we called
2160 * blk_queue_make_request(..., md_fail_request)
2161 * earlier.
2162 */
2163 mddev->queue->queuedata = mddev;
2164 mddev->queue->make_request_fn = mddev->pers->make_request;
2165
2166 mddev->changed = 1;
2167 md_new_event(mddev);
2168 return 0;
2169 }
2170
2171 static int restart_array(mddev_t *mddev)
2172 {
2173 struct gendisk *disk = mddev->gendisk;
2174 int err;
2175
2176 /*
2177 * Complain if it has no devices
2178 */
2179 err = -ENXIO;
2180 if (list_empty(&mddev->disks))
2181 goto out;
2182
2183 if (mddev->pers) {
2184 err = -EBUSY;
2185 if (!mddev->ro)
2186 goto out;
2187
2188 mddev->safemode = 0;
2189 mddev->ro = 0;
2190 set_disk_ro(disk, 0);
2191
2192 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2193 mdname(mddev));
2194 /*
2195 * Kick recovery or resync if necessary
2196 */
2197 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2198 md_wakeup_thread(mddev->thread);
2199 err = 0;
2200 } else {
2201 printk(KERN_ERR "md: %s has no personality assigned.\n",
2202 mdname(mddev));
2203 err = -EINVAL;
2204 }
2205
2206 out:
2207 return err;
2208 }
2209
2210 static int do_md_stop(mddev_t * mddev, int ro)
2211 {
2212 int err = 0;
2213 struct gendisk *disk = mddev->gendisk;
2214
2215 if (mddev->pers) {
2216 if (atomic_read(&mddev->active)>2) {
2217 printk("md: %s still in use.\n",mdname(mddev));
2218 return -EBUSY;
2219 }
2220
2221 if (mddev->sync_thread) {
2222 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2223 md_unregister_thread(mddev->sync_thread);
2224 mddev->sync_thread = NULL;
2225 }
2226
2227 del_timer_sync(&mddev->safemode_timer);
2228
2229 invalidate_partition(disk, 0);
2230
2231 if (ro) {
2232 err = -ENXIO;
2233 if (mddev->ro==1)
2234 goto out;
2235 mddev->ro = 1;
2236 } else {
2237 bitmap_flush(mddev);
2238 md_super_wait(mddev);
2239 if (mddev->ro)
2240 set_disk_ro(disk, 0);
2241 blk_queue_make_request(mddev->queue, md_fail_request);
2242 mddev->pers->stop(mddev);
2243 if (mddev->pers->sync_request)
2244 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2245
2246 module_put(mddev->pers->owner);
2247 mddev->pers = NULL;
2248 if (mddev->ro)
2249 mddev->ro = 0;
2250 }
2251 if (!mddev->in_sync) {
2252 /* mark array as shutdown cleanly */
2253 mddev->in_sync = 1;
2254 md_update_sb(mddev);
2255 }
2256 if (ro)
2257 set_disk_ro(disk, 1);
2258 }
2259
2260 bitmap_destroy(mddev);
2261 if (mddev->bitmap_file) {
2262 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2263 fput(mddev->bitmap_file);
2264 mddev->bitmap_file = NULL;
2265 }
2266 mddev->bitmap_offset = 0;
2267
2268 /*
2269 * Free resources if final stop
2270 */
2271 if (!ro) {
2272 mdk_rdev_t *rdev;
2273 struct list_head *tmp;
2274 struct gendisk *disk;
2275 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2276
2277 ITERATE_RDEV(mddev,rdev,tmp)
2278 if (rdev->raid_disk >= 0) {
2279 char nm[20];
2280 sprintf(nm, "rd%d", rdev->raid_disk);
2281 sysfs_remove_link(&mddev->kobj, nm);
2282 }
2283
2284 export_array(mddev);
2285
2286 mddev->array_size = 0;
2287 disk = mddev->gendisk;
2288 if (disk)
2289 set_capacity(disk, 0);
2290 mddev->changed = 1;
2291 } else
2292 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2293 mdname(mddev));
2294 err = 0;
2295 md_new_event(mddev);
2296 out:
2297 return err;
2298 }
2299
2300 static void autorun_array(mddev_t *mddev)
2301 {
2302 mdk_rdev_t *rdev;
2303 struct list_head *tmp;
2304 int err;
2305
2306 if (list_empty(&mddev->disks))
2307 return;
2308
2309 printk(KERN_INFO "md: running: ");
2310
2311 ITERATE_RDEV(mddev,rdev,tmp) {
2312 char b[BDEVNAME_SIZE];
2313 printk("<%s>", bdevname(rdev->bdev,b));
2314 }
2315 printk("\n");
2316
2317 err = do_md_run (mddev);
2318 if (err) {
2319 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2320 do_md_stop (mddev, 0);
2321 }
2322 }
2323
2324 /*
2325 * lets try to run arrays based on all disks that have arrived
2326 * until now. (those are in pending_raid_disks)
2327 *
2328 * the method: pick the first pending disk, collect all disks with
2329 * the same UUID, remove all from the pending list and put them into
2330 * the 'same_array' list. Then order this list based on superblock
2331 * update time (freshest comes first), kick out 'old' disks and
2332 * compare superblocks. If everything's fine then run it.
2333 *
2334 * If "unit" is allocated, then bump its reference count
2335 */
2336 static void autorun_devices(int part)
2337 {
2338 struct list_head candidates;
2339 struct list_head *tmp;
2340 mdk_rdev_t *rdev0, *rdev;
2341 mddev_t *mddev;
2342 char b[BDEVNAME_SIZE];
2343
2344 printk(KERN_INFO "md: autorun ...\n");
2345 while (!list_empty(&pending_raid_disks)) {
2346 dev_t dev;
2347 rdev0 = list_entry(pending_raid_disks.next,
2348 mdk_rdev_t, same_set);
2349
2350 printk(KERN_INFO "md: considering %s ...\n",
2351 bdevname(rdev0->bdev,b));
2352 INIT_LIST_HEAD(&candidates);
2353 ITERATE_RDEV_PENDING(rdev,tmp)
2354 if (super_90_load(rdev, rdev0, 0) >= 0) {
2355 printk(KERN_INFO "md: adding %s ...\n",
2356 bdevname(rdev->bdev,b));
2357 list_move(&rdev->same_set, &candidates);
2358 }
2359 /*
2360 * now we have a set of devices, with all of them having
2361 * mostly sane superblocks. It's time to allocate the
2362 * mddev.
2363 */
2364 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2365 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2366 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2367 break;
2368 }
2369 if (part)
2370 dev = MKDEV(mdp_major,
2371 rdev0->preferred_minor << MdpMinorShift);
2372 else
2373 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2374
2375 md_probe(dev, NULL, NULL);
2376 mddev = mddev_find(dev);
2377 if (!mddev) {
2378 printk(KERN_ERR
2379 "md: cannot allocate memory for md drive.\n");
2380 break;
2381 }
2382 if (mddev_lock(mddev))
2383 printk(KERN_WARNING "md: %s locked, cannot run\n",
2384 mdname(mddev));
2385 else if (mddev->raid_disks || mddev->major_version
2386 || !list_empty(&mddev->disks)) {
2387 printk(KERN_WARNING
2388 "md: %s already running, cannot run %s\n",
2389 mdname(mddev), bdevname(rdev0->bdev,b));
2390 mddev_unlock(mddev);
2391 } else {
2392 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2393 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2394 list_del_init(&rdev->same_set);
2395 if (bind_rdev_to_array(rdev, mddev))
2396 export_rdev(rdev);
2397 }
2398 autorun_array(mddev);
2399 mddev_unlock(mddev);
2400 }
2401 /* on success, candidates will be empty, on error
2402 * it won't...
2403 */
2404 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2405 export_rdev(rdev);
2406 mddev_put(mddev);
2407 }
2408 printk(KERN_INFO "md: ... autorun DONE.\n");
2409 }
2410
2411 /*
2412 * import RAID devices based on one partition
2413 * if possible, the array gets run as well.
2414 */
2415
2416 static int autostart_array(dev_t startdev)
2417 {
2418 char b[BDEVNAME_SIZE];
2419 int err = -EINVAL, i;
2420 mdp_super_t *sb = NULL;
2421 mdk_rdev_t *start_rdev = NULL, *rdev;
2422
2423 start_rdev = md_import_device(startdev, 0, 0);
2424 if (IS_ERR(start_rdev))
2425 return err;
2426
2427
2428 /* NOTE: this can only work for 0.90.0 superblocks */
2429 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2430 if (sb->major_version != 0 ||
2431 sb->minor_version != 90 ) {
2432 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2433 export_rdev(start_rdev);
2434 return err;
2435 }
2436
2437 if (test_bit(Faulty, &start_rdev->flags)) {
2438 printk(KERN_WARNING
2439 "md: can not autostart based on faulty %s!\n",
2440 bdevname(start_rdev->bdev,b));
2441 export_rdev(start_rdev);
2442 return err;
2443 }
2444 list_add(&start_rdev->same_set, &pending_raid_disks);
2445
2446 for (i = 0; i < MD_SB_DISKS; i++) {
2447 mdp_disk_t *desc = sb->disks + i;
2448 dev_t dev = MKDEV(desc->major, desc->minor);
2449
2450 if (!dev)
2451 continue;
2452 if (dev == startdev)
2453 continue;
2454 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2455 continue;
2456 rdev = md_import_device(dev, 0, 0);
2457 if (IS_ERR(rdev))
2458 continue;
2459
2460 list_add(&rdev->same_set, &pending_raid_disks);
2461 }
2462
2463 /*
2464 * possibly return codes
2465 */
2466 autorun_devices(0);
2467 return 0;
2468
2469 }
2470
2471
2472 static int get_version(void __user * arg)
2473 {
2474 mdu_version_t ver;
2475
2476 ver.major = MD_MAJOR_VERSION;
2477 ver.minor = MD_MINOR_VERSION;
2478 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2479
2480 if (copy_to_user(arg, &ver, sizeof(ver)))
2481 return -EFAULT;
2482
2483 return 0;
2484 }
2485
2486 static int get_array_info(mddev_t * mddev, void __user * arg)
2487 {
2488 mdu_array_info_t info;
2489 int nr,working,active,failed,spare;
2490 mdk_rdev_t *rdev;
2491 struct list_head *tmp;
2492
2493 nr=working=active=failed=spare=0;
2494 ITERATE_RDEV(mddev,rdev,tmp) {
2495 nr++;
2496 if (test_bit(Faulty, &rdev->flags))
2497 failed++;
2498 else {
2499 working++;
2500 if (test_bit(In_sync, &rdev->flags))
2501 active++;
2502 else
2503 spare++;
2504 }
2505 }
2506
2507 info.major_version = mddev->major_version;
2508 info.minor_version = mddev->minor_version;
2509 info.patch_version = MD_PATCHLEVEL_VERSION;
2510 info.ctime = mddev->ctime;
2511 info.level = mddev->level;
2512 info.size = mddev->size;
2513 info.nr_disks = nr;
2514 info.raid_disks = mddev->raid_disks;
2515 info.md_minor = mddev->md_minor;
2516 info.not_persistent= !mddev->persistent;
2517
2518 info.utime = mddev->utime;
2519 info.state = 0;
2520 if (mddev->in_sync)
2521 info.state = (1<<MD_SB_CLEAN);
2522 if (mddev->bitmap && mddev->bitmap_offset)
2523 info.state = (1<<MD_SB_BITMAP_PRESENT);
2524 info.active_disks = active;
2525 info.working_disks = working;
2526 info.failed_disks = failed;
2527 info.spare_disks = spare;
2528
2529 info.layout = mddev->layout;
2530 info.chunk_size = mddev->chunk_size;
2531
2532 if (copy_to_user(arg, &info, sizeof(info)))
2533 return -EFAULT;
2534
2535 return 0;
2536 }
2537
2538 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2539 {
2540 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2541 char *ptr, *buf = NULL;
2542 int err = -ENOMEM;
2543
2544 file = kmalloc(sizeof(*file), GFP_KERNEL);
2545 if (!file)
2546 goto out;
2547
2548 /* bitmap disabled, zero the first byte and copy out */
2549 if (!mddev->bitmap || !mddev->bitmap->file) {
2550 file->pathname[0] = '\0';
2551 goto copy_out;
2552 }
2553
2554 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2555 if (!buf)
2556 goto out;
2557
2558 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2559 if (!ptr)
2560 goto out;
2561
2562 strcpy(file->pathname, ptr);
2563
2564 copy_out:
2565 err = 0;
2566 if (copy_to_user(arg, file, sizeof(*file)))
2567 err = -EFAULT;
2568 out:
2569 kfree(buf);
2570 kfree(file);
2571 return err;
2572 }
2573
2574 static int get_disk_info(mddev_t * mddev, void __user * arg)
2575 {
2576 mdu_disk_info_t info;
2577 unsigned int nr;
2578 mdk_rdev_t *rdev;
2579
2580 if (copy_from_user(&info, arg, sizeof(info)))
2581 return -EFAULT;
2582
2583 nr = info.number;
2584
2585 rdev = find_rdev_nr(mddev, nr);
2586 if (rdev) {
2587 info.major = MAJOR(rdev->bdev->bd_dev);
2588 info.minor = MINOR(rdev->bdev->bd_dev);
2589 info.raid_disk = rdev->raid_disk;
2590 info.state = 0;
2591 if (test_bit(Faulty, &rdev->flags))
2592 info.state |= (1<<MD_DISK_FAULTY);
2593 else if (test_bit(In_sync, &rdev->flags)) {
2594 info.state |= (1<<MD_DISK_ACTIVE);
2595 info.state |= (1<<MD_DISK_SYNC);
2596 }
2597 if (test_bit(WriteMostly, &rdev->flags))
2598 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2599 } else {
2600 info.major = info.minor = 0;
2601 info.raid_disk = -1;
2602 info.state = (1<<MD_DISK_REMOVED);
2603 }
2604
2605 if (copy_to_user(arg, &info, sizeof(info)))
2606 return -EFAULT;
2607
2608 return 0;
2609 }
2610
2611 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2612 {
2613 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2614 mdk_rdev_t *rdev;
2615 dev_t dev = MKDEV(info->major,info->minor);
2616
2617 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2618 return -EOVERFLOW;
2619
2620 if (!mddev->raid_disks) {
2621 int err;
2622 /* expecting a device which has a superblock */
2623 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2624 if (IS_ERR(rdev)) {
2625 printk(KERN_WARNING
2626 "md: md_import_device returned %ld\n",
2627 PTR_ERR(rdev));
2628 return PTR_ERR(rdev);
2629 }
2630 if (!list_empty(&mddev->disks)) {
2631 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2632 mdk_rdev_t, same_set);
2633 int err = super_types[mddev->major_version]
2634 .load_super(rdev, rdev0, mddev->minor_version);
2635 if (err < 0) {
2636 printk(KERN_WARNING
2637 "md: %s has different UUID to %s\n",
2638 bdevname(rdev->bdev,b),
2639 bdevname(rdev0->bdev,b2));
2640 export_rdev(rdev);
2641 return -EINVAL;
2642 }
2643 }
2644 err = bind_rdev_to_array(rdev, mddev);
2645 if (err)
2646 export_rdev(rdev);
2647 return err;
2648 }
2649
2650 /*
2651 * add_new_disk can be used once the array is assembled
2652 * to add "hot spares". They must already have a superblock
2653 * written
2654 */
2655 if (mddev->pers) {
2656 int err;
2657 if (!mddev->pers->hot_add_disk) {
2658 printk(KERN_WARNING
2659 "%s: personality does not support diskops!\n",
2660 mdname(mddev));
2661 return -EINVAL;
2662 }
2663 if (mddev->persistent)
2664 rdev = md_import_device(dev, mddev->major_version,
2665 mddev->minor_version);
2666 else
2667 rdev = md_import_device(dev, -1, -1);
2668 if (IS_ERR(rdev)) {
2669 printk(KERN_WARNING
2670 "md: md_import_device returned %ld\n",
2671 PTR_ERR(rdev));
2672 return PTR_ERR(rdev);
2673 }
2674 /* set save_raid_disk if appropriate */
2675 if (!mddev->persistent) {
2676 if (info->state & (1<<MD_DISK_SYNC) &&
2677 info->raid_disk < mddev->raid_disks)
2678 rdev->raid_disk = info->raid_disk;
2679 else
2680 rdev->raid_disk = -1;
2681 } else
2682 super_types[mddev->major_version].
2683 validate_super(mddev, rdev);
2684 rdev->saved_raid_disk = rdev->raid_disk;
2685
2686 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2687 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2688 set_bit(WriteMostly, &rdev->flags);
2689
2690 rdev->raid_disk = -1;
2691 err = bind_rdev_to_array(rdev, mddev);
2692 if (err)
2693 export_rdev(rdev);
2694
2695 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2696 md_wakeup_thread(mddev->thread);
2697 return err;
2698 }
2699
2700 /* otherwise, add_new_disk is only allowed
2701 * for major_version==0 superblocks
2702 */
2703 if (mddev->major_version != 0) {
2704 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2705 mdname(mddev));
2706 return -EINVAL;
2707 }
2708
2709 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2710 int err;
2711 rdev = md_import_device (dev, -1, 0);
2712 if (IS_ERR(rdev)) {
2713 printk(KERN_WARNING
2714 "md: error, md_import_device() returned %ld\n",
2715 PTR_ERR(rdev));
2716 return PTR_ERR(rdev);
2717 }
2718 rdev->desc_nr = info->number;
2719 if (info->raid_disk < mddev->raid_disks)
2720 rdev->raid_disk = info->raid_disk;
2721 else
2722 rdev->raid_disk = -1;
2723
2724 rdev->flags = 0;
2725
2726 if (rdev->raid_disk < mddev->raid_disks)
2727 if (info->state & (1<<MD_DISK_SYNC))
2728 set_bit(In_sync, &rdev->flags);
2729
2730 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2731 set_bit(WriteMostly, &rdev->flags);
2732
2733 err = bind_rdev_to_array(rdev, mddev);
2734 if (err) {
2735 export_rdev(rdev);
2736 return err;
2737 }
2738
2739 if (!mddev->persistent) {
2740 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2741 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2742 } else
2743 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2744 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2745
2746 if (!mddev->size || (mddev->size > rdev->size))
2747 mddev->size = rdev->size;
2748 }
2749
2750 return 0;
2751 }
2752
2753 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2754 {
2755 char b[BDEVNAME_SIZE];
2756 mdk_rdev_t *rdev;
2757
2758 if (!mddev->pers)
2759 return -ENODEV;
2760
2761 rdev = find_rdev(mddev, dev);
2762 if (!rdev)
2763 return -ENXIO;
2764
2765 if (rdev->raid_disk >= 0)
2766 goto busy;
2767
2768 kick_rdev_from_array(rdev);
2769 md_update_sb(mddev);
2770 md_new_event(mddev);
2771
2772 return 0;
2773 busy:
2774 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2775 bdevname(rdev->bdev,b), mdname(mddev));
2776 return -EBUSY;
2777 }
2778
2779 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2780 {
2781 char b[BDEVNAME_SIZE];
2782 int err;
2783 unsigned int size;
2784 mdk_rdev_t *rdev;
2785
2786 if (!mddev->pers)
2787 return -ENODEV;
2788
2789 if (mddev->major_version != 0) {
2790 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2791 " version-0 superblocks.\n",
2792 mdname(mddev));
2793 return -EINVAL;
2794 }
2795 if (!mddev->pers->hot_add_disk) {
2796 printk(KERN_WARNING
2797 "%s: personality does not support diskops!\n",
2798 mdname(mddev));
2799 return -EINVAL;
2800 }
2801
2802 rdev = md_import_device (dev, -1, 0);
2803 if (IS_ERR(rdev)) {
2804 printk(KERN_WARNING
2805 "md: error, md_import_device() returned %ld\n",
2806 PTR_ERR(rdev));
2807 return -EINVAL;
2808 }
2809
2810 if (mddev->persistent)
2811 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2812 else
2813 rdev->sb_offset =
2814 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2815
2816 size = calc_dev_size(rdev, mddev->chunk_size);
2817 rdev->size = size;
2818
2819 if (size < mddev->size) {
2820 printk(KERN_WARNING
2821 "%s: disk size %llu blocks < array size %llu\n",
2822 mdname(mddev), (unsigned long long)size,
2823 (unsigned long long)mddev->size);
2824 err = -ENOSPC;
2825 goto abort_export;
2826 }
2827
2828 if (test_bit(Faulty, &rdev->flags)) {
2829 printk(KERN_WARNING
2830 "md: can not hot-add faulty %s disk to %s!\n",
2831 bdevname(rdev->bdev,b), mdname(mddev));
2832 err = -EINVAL;
2833 goto abort_export;
2834 }
2835 clear_bit(In_sync, &rdev->flags);
2836 rdev->desc_nr = -1;
2837 bind_rdev_to_array(rdev, mddev);
2838
2839 /*
2840 * The rest should better be atomic, we can have disk failures
2841 * noticed in interrupt contexts ...
2842 */
2843
2844 if (rdev->desc_nr == mddev->max_disks) {
2845 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2846 mdname(mddev));
2847 err = -EBUSY;
2848 goto abort_unbind_export;
2849 }
2850
2851 rdev->raid_disk = -1;
2852
2853 md_update_sb(mddev);
2854
2855 /*
2856 * Kick recovery, maybe this spare has to be added to the
2857 * array immediately.
2858 */
2859 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2860 md_wakeup_thread(mddev->thread);
2861 md_new_event(mddev);
2862 return 0;
2863
2864 abort_unbind_export:
2865 unbind_rdev_from_array(rdev);
2866
2867 abort_export:
2868 export_rdev(rdev);
2869 return err;
2870 }
2871
2872 /* similar to deny_write_access, but accounts for our holding a reference
2873 * to the file ourselves */
2874 static int deny_bitmap_write_access(struct file * file)
2875 {
2876 struct inode *inode = file->f_mapping->host;
2877
2878 spin_lock(&inode->i_lock);
2879 if (atomic_read(&inode->i_writecount) > 1) {
2880 spin_unlock(&inode->i_lock);
2881 return -ETXTBSY;
2882 }
2883 atomic_set(&inode->i_writecount, -1);
2884 spin_unlock(&inode->i_lock);
2885
2886 return 0;
2887 }
2888
2889 static int set_bitmap_file(mddev_t *mddev, int fd)
2890 {
2891 int err;
2892
2893 if (mddev->pers) {
2894 if (!mddev->pers->quiesce)
2895 return -EBUSY;
2896 if (mddev->recovery || mddev->sync_thread)
2897 return -EBUSY;
2898 /* we should be able to change the bitmap.. */
2899 }
2900
2901
2902 if (fd >= 0) {
2903 if (mddev->bitmap)
2904 return -EEXIST; /* cannot add when bitmap is present */
2905 mddev->bitmap_file = fget(fd);
2906
2907 if (mddev->bitmap_file == NULL) {
2908 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2909 mdname(mddev));
2910 return -EBADF;
2911 }
2912
2913 err = deny_bitmap_write_access(mddev->bitmap_file);
2914 if (err) {
2915 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2916 mdname(mddev));
2917 fput(mddev->bitmap_file);
2918 mddev->bitmap_file = NULL;
2919 return err;
2920 }
2921 mddev->bitmap_offset = 0; /* file overrides offset */
2922 } else if (mddev->bitmap == NULL)
2923 return -ENOENT; /* cannot remove what isn't there */
2924 err = 0;
2925 if (mddev->pers) {
2926 mddev->pers->quiesce(mddev, 1);
2927 if (fd >= 0)
2928 err = bitmap_create(mddev);
2929 if (fd < 0 || err)
2930 bitmap_destroy(mddev);
2931 mddev->pers->quiesce(mddev, 0);
2932 } else if (fd < 0) {
2933 if (mddev->bitmap_file)
2934 fput(mddev->bitmap_file);
2935 mddev->bitmap_file = NULL;
2936 }
2937
2938 return err;
2939 }
2940
2941 /*
2942 * set_array_info is used two different ways
2943 * The original usage is when creating a new array.
2944 * In this usage, raid_disks is > 0 and it together with
2945 * level, size, not_persistent,layout,chunksize determine the
2946 * shape of the array.
2947 * This will always create an array with a type-0.90.0 superblock.
2948 * The newer usage is when assembling an array.
2949 * In this case raid_disks will be 0, and the major_version field is
2950 * use to determine which style super-blocks are to be found on the devices.
2951 * The minor and patch _version numbers are also kept incase the
2952 * super_block handler wishes to interpret them.
2953 */
2954 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2955 {
2956
2957 if (info->raid_disks == 0) {
2958 /* just setting version number for superblock loading */
2959 if (info->major_version < 0 ||
2960 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2961 super_types[info->major_version].name == NULL) {
2962 /* maybe try to auto-load a module? */
2963 printk(KERN_INFO
2964 "md: superblock version %d not known\n",
2965 info->major_version);
2966 return -EINVAL;
2967 }
2968 mddev->major_version = info->major_version;
2969 mddev->minor_version = info->minor_version;
2970 mddev->patch_version = info->patch_version;
2971 return 0;
2972 }
2973 mddev->major_version = MD_MAJOR_VERSION;
2974 mddev->minor_version = MD_MINOR_VERSION;
2975 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2976 mddev->ctime = get_seconds();
2977
2978 mddev->level = info->level;
2979 mddev->size = info->size;
2980 mddev->raid_disks = info->raid_disks;
2981 /* don't set md_minor, it is determined by which /dev/md* was
2982 * openned
2983 */
2984 if (info->state & (1<<MD_SB_CLEAN))
2985 mddev->recovery_cp = MaxSector;
2986 else
2987 mddev->recovery_cp = 0;
2988 mddev->persistent = ! info->not_persistent;
2989
2990 mddev->layout = info->layout;
2991 mddev->chunk_size = info->chunk_size;
2992
2993 mddev->max_disks = MD_SB_DISKS;
2994
2995 mddev->sb_dirty = 1;
2996
2997 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2998 mddev->bitmap_offset = 0;
2999
3000 /*
3001 * Generate a 128 bit UUID
3002 */
3003 get_random_bytes(mddev->uuid, 16);
3004
3005 return 0;
3006 }
3007
3008 /*
3009 * update_array_info is used to change the configuration of an
3010 * on-line array.
3011 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3012 * fields in the info are checked against the array.
3013 * Any differences that cannot be handled will cause an error.
3014 * Normally, only one change can be managed at a time.
3015 */
3016 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3017 {
3018 int rv = 0;
3019 int cnt = 0;
3020 int state = 0;
3021
3022 /* calculate expected state,ignoring low bits */
3023 if (mddev->bitmap && mddev->bitmap_offset)
3024 state |= (1 << MD_SB_BITMAP_PRESENT);
3025
3026 if (mddev->major_version != info->major_version ||
3027 mddev->minor_version != info->minor_version ||
3028 /* mddev->patch_version != info->patch_version || */
3029 mddev->ctime != info->ctime ||
3030 mddev->level != info->level ||
3031 /* mddev->layout != info->layout || */
3032 !mddev->persistent != info->not_persistent||
3033 mddev->chunk_size != info->chunk_size ||
3034 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3035 ((state^info->state) & 0xfffffe00)
3036 )
3037 return -EINVAL;
3038 /* Check there is only one change */
3039 if (mddev->size != info->size) cnt++;
3040 if (mddev->raid_disks != info->raid_disks) cnt++;
3041 if (mddev->layout != info->layout) cnt++;
3042 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3043 if (cnt == 0) return 0;
3044 if (cnt > 1) return -EINVAL;
3045
3046 if (mddev->layout != info->layout) {
3047 /* Change layout
3048 * we don't need to do anything at the md level, the
3049 * personality will take care of it all.
3050 */
3051 if (mddev->pers->reconfig == NULL)
3052 return -EINVAL;
3053 else
3054 return mddev->pers->reconfig(mddev, info->layout, -1);
3055 }
3056 if (mddev->size != info->size) {
3057 mdk_rdev_t * rdev;
3058 struct list_head *tmp;
3059 if (mddev->pers->resize == NULL)
3060 return -EINVAL;
3061 /* The "size" is the amount of each device that is used.
3062 * This can only make sense for arrays with redundancy.
3063 * linear and raid0 always use whatever space is available
3064 * We can only consider changing the size if no resync
3065 * or reconstruction is happening, and if the new size
3066 * is acceptable. It must fit before the sb_offset or,
3067 * if that is <data_offset, it must fit before the
3068 * size of each device.
3069 * If size is zero, we find the largest size that fits.
3070 */
3071 if (mddev->sync_thread)
3072 return -EBUSY;
3073 ITERATE_RDEV(mddev,rdev,tmp) {
3074 sector_t avail;
3075 int fit = (info->size == 0);
3076 if (rdev->sb_offset > rdev->data_offset)
3077 avail = (rdev->sb_offset*2) - rdev->data_offset;
3078 else
3079 avail = get_capacity(rdev->bdev->bd_disk)
3080 - rdev->data_offset;
3081 if (fit && (info->size == 0 || info->size > avail/2))
3082 info->size = avail/2;
3083 if (avail < ((sector_t)info->size << 1))
3084 return -ENOSPC;
3085 }
3086 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3087 if (!rv) {
3088 struct block_device *bdev;
3089
3090 bdev = bdget_disk(mddev->gendisk, 0);
3091 if (bdev) {
3092 down(&bdev->bd_inode->i_sem);
3093 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3094 up(&bdev->bd_inode->i_sem);
3095 bdput(bdev);
3096 }
3097 }
3098 }
3099 if (mddev->raid_disks != info->raid_disks) {
3100 /* change the number of raid disks */
3101 if (mddev->pers->reshape == NULL)
3102 return -EINVAL;
3103 if (info->raid_disks <= 0 ||
3104 info->raid_disks >= mddev->max_disks)
3105 return -EINVAL;
3106 if (mddev->sync_thread)
3107 return -EBUSY;
3108 rv = mddev->pers->reshape(mddev, info->raid_disks);
3109 if (!rv) {
3110 struct block_device *bdev;
3111
3112 bdev = bdget_disk(mddev->gendisk, 0);
3113 if (bdev) {
3114 down(&bdev->bd_inode->i_sem);
3115 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3116 up(&bdev->bd_inode->i_sem);
3117 bdput(bdev);
3118 }
3119 }
3120 }
3121 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3122 if (mddev->pers->quiesce == NULL)
3123 return -EINVAL;
3124 if (mddev->recovery || mddev->sync_thread)
3125 return -EBUSY;
3126 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3127 /* add the bitmap */
3128 if (mddev->bitmap)
3129 return -EEXIST;
3130 if (mddev->default_bitmap_offset == 0)
3131 return -EINVAL;
3132 mddev->bitmap_offset = mddev->default_bitmap_offset;
3133 mddev->pers->quiesce(mddev, 1);
3134 rv = bitmap_create(mddev);
3135 if (rv)
3136 bitmap_destroy(mddev);
3137 mddev->pers->quiesce(mddev, 0);
3138 } else {
3139 /* remove the bitmap */
3140 if (!mddev->bitmap)
3141 return -ENOENT;
3142 if (mddev->bitmap->file)
3143 return -EINVAL;
3144 mddev->pers->quiesce(mddev, 1);
3145 bitmap_destroy(mddev);
3146 mddev->pers->quiesce(mddev, 0);
3147 mddev->bitmap_offset = 0;
3148 }
3149 }
3150 md_update_sb(mddev);
3151 return rv;
3152 }
3153
3154 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3155 {
3156 mdk_rdev_t *rdev;
3157
3158 if (mddev->pers == NULL)
3159 return -ENODEV;
3160
3161 rdev = find_rdev(mddev, dev);
3162 if (!rdev)
3163 return -ENODEV;
3164
3165 md_error(mddev, rdev);
3166 return 0;
3167 }
3168
3169 static int md_ioctl(struct inode *inode, struct file *file,
3170 unsigned int cmd, unsigned long arg)
3171 {
3172 int err = 0;
3173 void __user *argp = (void __user *)arg;
3174 struct hd_geometry __user *loc = argp;
3175 mddev_t *mddev = NULL;
3176
3177 if (!capable(CAP_SYS_ADMIN))
3178 return -EACCES;
3179
3180 /*
3181 * Commands dealing with the RAID driver but not any
3182 * particular array:
3183 */
3184 switch (cmd)
3185 {
3186 case RAID_VERSION:
3187 err = get_version(argp);
3188 goto done;
3189
3190 case PRINT_RAID_DEBUG:
3191 err = 0;
3192 md_print_devices();
3193 goto done;
3194
3195 #ifndef MODULE
3196 case RAID_AUTORUN:
3197 err = 0;
3198 autostart_arrays(arg);
3199 goto done;
3200 #endif
3201 default:;
3202 }
3203
3204 /*
3205 * Commands creating/starting a new array:
3206 */
3207
3208 mddev = inode->i_bdev->bd_disk->private_data;
3209
3210 if (!mddev) {
3211 BUG();
3212 goto abort;
3213 }
3214
3215
3216 if (cmd == START_ARRAY) {
3217 /* START_ARRAY doesn't need to lock the array as autostart_array
3218 * does the locking, and it could even be a different array
3219 */
3220 static int cnt = 3;
3221 if (cnt > 0 ) {
3222 printk(KERN_WARNING
3223 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3224 "This will not be supported beyond July 2006\n",
3225 current->comm, current->pid);
3226 cnt--;
3227 }
3228 err = autostart_array(new_decode_dev(arg));
3229 if (err) {
3230 printk(KERN_WARNING "md: autostart failed!\n");
3231 goto abort;
3232 }
3233 goto done;
3234 }
3235
3236 err = mddev_lock(mddev);
3237 if (err) {
3238 printk(KERN_INFO
3239 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3240 err, cmd);
3241 goto abort;
3242 }
3243
3244 switch (cmd)
3245 {
3246 case SET_ARRAY_INFO:
3247 {
3248 mdu_array_info_t info;
3249 if (!arg)
3250 memset(&info, 0, sizeof(info));
3251 else if (copy_from_user(&info, argp, sizeof(info))) {
3252 err = -EFAULT;
3253 goto abort_unlock;
3254 }
3255 if (mddev->pers) {
3256 err = update_array_info(mddev, &info);
3257 if (err) {
3258 printk(KERN_WARNING "md: couldn't update"
3259 " array info. %d\n", err);
3260 goto abort_unlock;
3261 }
3262 goto done_unlock;
3263 }
3264 if (!list_empty(&mddev->disks)) {
3265 printk(KERN_WARNING
3266 "md: array %s already has disks!\n",
3267 mdname(mddev));
3268 err = -EBUSY;
3269 goto abort_unlock;
3270 }
3271 if (mddev->raid_disks) {
3272 printk(KERN_WARNING
3273 "md: array %s already initialised!\n",
3274 mdname(mddev));
3275 err = -EBUSY;
3276 goto abort_unlock;
3277 }
3278 err = set_array_info(mddev, &info);
3279 if (err) {
3280 printk(KERN_WARNING "md: couldn't set"
3281 " array info. %d\n", err);
3282 goto abort_unlock;
3283 }
3284 }
3285 goto done_unlock;
3286
3287 default:;
3288 }
3289
3290 /*
3291 * Commands querying/configuring an existing array:
3292 */
3293 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3294 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3295 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3296 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3297 err = -ENODEV;
3298 goto abort_unlock;
3299 }
3300
3301 /*
3302 * Commands even a read-only array can execute:
3303 */
3304 switch (cmd)
3305 {
3306 case GET_ARRAY_INFO:
3307 err = get_array_info(mddev, argp);
3308 goto done_unlock;
3309
3310 case GET_BITMAP_FILE:
3311 err = get_bitmap_file(mddev, argp);
3312 goto done_unlock;
3313
3314 case GET_DISK_INFO:
3315 err = get_disk_info(mddev, argp);
3316 goto done_unlock;
3317
3318 case RESTART_ARRAY_RW:
3319 err = restart_array(mddev);
3320 goto done_unlock;
3321
3322 case STOP_ARRAY:
3323 err = do_md_stop (mddev, 0);
3324 goto done_unlock;
3325
3326 case STOP_ARRAY_RO:
3327 err = do_md_stop (mddev, 1);
3328 goto done_unlock;
3329
3330 /*
3331 * We have a problem here : there is no easy way to give a CHS
3332 * virtual geometry. We currently pretend that we have a 2 heads
3333 * 4 sectors (with a BIG number of cylinders...). This drives
3334 * dosfs just mad... ;-)
3335 */
3336 case HDIO_GETGEO:
3337 if (!loc) {
3338 err = -EINVAL;
3339 goto abort_unlock;
3340 }
3341 err = put_user (2, (char __user *) &loc->heads);
3342 if (err)
3343 goto abort_unlock;
3344 err = put_user (4, (char __user *) &loc->sectors);
3345 if (err)
3346 goto abort_unlock;
3347 err = put_user(get_capacity(mddev->gendisk)/8,
3348 (short __user *) &loc->cylinders);
3349 if (err)
3350 goto abort_unlock;
3351 err = put_user (get_start_sect(inode->i_bdev),
3352 (long __user *) &loc->start);
3353 goto done_unlock;
3354 }
3355
3356 /*
3357 * The remaining ioctls are changing the state of the
3358 * superblock, so we do not allow them on read-only arrays.
3359 * However non-MD ioctls (e.g. get-size) will still come through
3360 * here and hit the 'default' below, so only disallow
3361 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3362 */
3363 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3364 mddev->ro && mddev->pers) {
3365 if (mddev->ro == 2) {
3366 mddev->ro = 0;
3367 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3368 md_wakeup_thread(mddev->thread);
3369
3370 } else {
3371 err = -EROFS;
3372 goto abort_unlock;
3373 }
3374 }
3375
3376 switch (cmd)
3377 {
3378 case ADD_NEW_DISK:
3379 {
3380 mdu_disk_info_t info;
3381 if (copy_from_user(&info, argp, sizeof(info)))
3382 err = -EFAULT;
3383 else
3384 err = add_new_disk(mddev, &info);
3385 goto done_unlock;
3386 }
3387
3388 case HOT_REMOVE_DISK:
3389 err = hot_remove_disk(mddev, new_decode_dev(arg));
3390 goto done_unlock;
3391
3392 case HOT_ADD_DISK:
3393 err = hot_add_disk(mddev, new_decode_dev(arg));
3394 goto done_unlock;
3395
3396 case SET_DISK_FAULTY:
3397 err = set_disk_faulty(mddev, new_decode_dev(arg));
3398 goto done_unlock;
3399
3400 case RUN_ARRAY:
3401 err = do_md_run (mddev);
3402 goto done_unlock;
3403
3404 case SET_BITMAP_FILE:
3405 err = set_bitmap_file(mddev, (int)arg);
3406 goto done_unlock;
3407
3408 default:
3409 if (_IOC_TYPE(cmd) == MD_MAJOR)
3410 printk(KERN_WARNING "md: %s(pid %d) used"
3411 " obsolete MD ioctl, upgrade your"
3412 " software to use new ictls.\n",
3413 current->comm, current->pid);
3414 err = -EINVAL;
3415 goto abort_unlock;
3416 }
3417
3418 done_unlock:
3419 abort_unlock:
3420 mddev_unlock(mddev);
3421
3422 return err;
3423 done:
3424 if (err)
3425 MD_BUG();
3426 abort:
3427 return err;
3428 }
3429
3430 static int md_open(struct inode *inode, struct file *file)
3431 {
3432 /*
3433 * Succeed if we can lock the mddev, which confirms that
3434 * it isn't being stopped right now.
3435 */
3436 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3437 int err;
3438
3439 if ((err = mddev_lock(mddev)))
3440 goto out;
3441
3442 err = 0;
3443 mddev_get(mddev);
3444 mddev_unlock(mddev);
3445
3446 check_disk_change(inode->i_bdev);
3447 out:
3448 return err;
3449 }
3450
3451 static int md_release(struct inode *inode, struct file * file)
3452 {
3453 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3454
3455 if (!mddev)
3456 BUG();
3457 mddev_put(mddev);
3458
3459 return 0;
3460 }
3461
3462 static int md_media_changed(struct gendisk *disk)
3463 {
3464 mddev_t *mddev = disk->private_data;
3465
3466 return mddev->changed;
3467 }
3468
3469 static int md_revalidate(struct gendisk *disk)
3470 {
3471 mddev_t *mddev = disk->private_data;
3472
3473 mddev->changed = 0;
3474 return 0;
3475 }
3476 static struct block_device_operations md_fops =
3477 {
3478 .owner = THIS_MODULE,
3479 .open = md_open,
3480 .release = md_release,
3481 .ioctl = md_ioctl,
3482 .media_changed = md_media_changed,
3483 .revalidate_disk= md_revalidate,
3484 };
3485
3486 static int md_thread(void * arg)
3487 {
3488 mdk_thread_t *thread = arg;
3489
3490 /*
3491 * md_thread is a 'system-thread', it's priority should be very
3492 * high. We avoid resource deadlocks individually in each
3493 * raid personality. (RAID5 does preallocation) We also use RR and
3494 * the very same RT priority as kswapd, thus we will never get
3495 * into a priority inversion deadlock.
3496 *
3497 * we definitely have to have equal or higher priority than
3498 * bdflush, otherwise bdflush will deadlock if there are too
3499 * many dirty RAID5 blocks.
3500 */
3501
3502 allow_signal(SIGKILL);
3503 while (!kthread_should_stop()) {
3504
3505 /* We need to wait INTERRUPTIBLE so that
3506 * we don't add to the load-average.
3507 * That means we need to be sure no signals are
3508 * pending
3509 */
3510 if (signal_pending(current))
3511 flush_signals(current);
3512
3513 wait_event_interruptible_timeout
3514 (thread->wqueue,
3515 test_bit(THREAD_WAKEUP, &thread->flags)
3516 || kthread_should_stop(),
3517 thread->timeout);
3518 try_to_freeze();
3519
3520 clear_bit(THREAD_WAKEUP, &thread->flags);
3521
3522 thread->run(thread->mddev);
3523 }
3524
3525 return 0;
3526 }
3527
3528 void md_wakeup_thread(mdk_thread_t *thread)
3529 {
3530 if (thread) {
3531 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3532 set_bit(THREAD_WAKEUP, &thread->flags);
3533 wake_up(&thread->wqueue);
3534 }
3535 }
3536
3537 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3538 const char *name)
3539 {
3540 mdk_thread_t *thread;
3541
3542 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3543 if (!thread)
3544 return NULL;
3545
3546 init_waitqueue_head(&thread->wqueue);
3547
3548 thread->run = run;
3549 thread->mddev = mddev;
3550 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3551 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3552 if (IS_ERR(thread->tsk)) {
3553 kfree(thread);
3554 return NULL;
3555 }
3556 return thread;
3557 }
3558
3559 void md_unregister_thread(mdk_thread_t *thread)
3560 {
3561 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3562
3563 kthread_stop(thread->tsk);
3564 kfree(thread);
3565 }
3566
3567 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3568 {
3569 if (!mddev) {
3570 MD_BUG();
3571 return;
3572 }
3573
3574 if (!rdev || test_bit(Faulty, &rdev->flags))
3575 return;
3576 /*
3577 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3578 mdname(mddev),
3579 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3580 __builtin_return_address(0),__builtin_return_address(1),
3581 __builtin_return_address(2),__builtin_return_address(3));
3582 */
3583 if (!mddev->pers->error_handler)
3584 return;
3585 mddev->pers->error_handler(mddev,rdev);
3586 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3587 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3588 md_wakeup_thread(mddev->thread);
3589 md_new_event(mddev);
3590 }
3591
3592 /* seq_file implementation /proc/mdstat */
3593
3594 static void status_unused(struct seq_file *seq)
3595 {
3596 int i = 0;
3597 mdk_rdev_t *rdev;
3598 struct list_head *tmp;
3599
3600 seq_printf(seq, "unused devices: ");
3601
3602 ITERATE_RDEV_PENDING(rdev,tmp) {
3603 char b[BDEVNAME_SIZE];
3604 i++;
3605 seq_printf(seq, "%s ",
3606 bdevname(rdev->bdev,b));
3607 }
3608 if (!i)
3609 seq_printf(seq, "<none>");
3610
3611 seq_printf(seq, "\n");
3612 }
3613
3614
3615 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3616 {
3617 unsigned long max_blocks, resync, res, dt, db, rt;
3618
3619 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3620
3621 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3622 max_blocks = mddev->resync_max_sectors >> 1;
3623 else
3624 max_blocks = mddev->size;
3625
3626 /*
3627 * Should not happen.
3628 */
3629 if (!max_blocks) {
3630 MD_BUG();
3631 return;
3632 }
3633 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3634 {
3635 int i, x = res/50, y = 20-x;
3636 seq_printf(seq, "[");
3637 for (i = 0; i < x; i++)
3638 seq_printf(seq, "=");
3639 seq_printf(seq, ">");
3640 for (i = 0; i < y; i++)
3641 seq_printf(seq, ".");
3642 seq_printf(seq, "] ");
3643 }
3644 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3645 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3646 "resync" : "recovery"),
3647 res/10, res % 10, resync, max_blocks);
3648
3649 /*
3650 * We do not want to overflow, so the order of operands and
3651 * the * 100 / 100 trick are important. We do a +1 to be
3652 * safe against division by zero. We only estimate anyway.
3653 *
3654 * dt: time from mark until now
3655 * db: blocks written from mark until now
3656 * rt: remaining time
3657 */
3658 dt = ((jiffies - mddev->resync_mark) / HZ);
3659 if (!dt) dt++;
3660 db = resync - (mddev->resync_mark_cnt/2);
3661 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3662
3663 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3664
3665 seq_printf(seq, " speed=%ldK/sec", db/dt);
3666 }
3667
3668 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3669 {
3670 struct list_head *tmp;
3671 loff_t l = *pos;
3672 mddev_t *mddev;
3673
3674 if (l >= 0x10000)
3675 return NULL;
3676 if (!l--)
3677 /* header */
3678 return (void*)1;
3679
3680 spin_lock(&all_mddevs_lock);
3681 list_for_each(tmp,&all_mddevs)
3682 if (!l--) {
3683 mddev = list_entry(tmp, mddev_t, all_mddevs);
3684 mddev_get(mddev);
3685 spin_unlock(&all_mddevs_lock);
3686 return mddev;
3687 }
3688 spin_unlock(&all_mddevs_lock);
3689 if (!l--)
3690 return (void*)2;/* tail */
3691 return NULL;
3692 }
3693
3694 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3695 {
3696 struct list_head *tmp;
3697 mddev_t *next_mddev, *mddev = v;
3698
3699 ++*pos;
3700 if (v == (void*)2)
3701 return NULL;
3702
3703 spin_lock(&all_mddevs_lock);
3704 if (v == (void*)1)
3705 tmp = all_mddevs.next;
3706 else
3707 tmp = mddev->all_mddevs.next;
3708 if (tmp != &all_mddevs)
3709 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3710 else {
3711 next_mddev = (void*)2;
3712 *pos = 0x10000;
3713 }
3714 spin_unlock(&all_mddevs_lock);
3715
3716 if (v != (void*)1)
3717 mddev_put(mddev);
3718 return next_mddev;
3719
3720 }
3721
3722 static void md_seq_stop(struct seq_file *seq, void *v)
3723 {
3724 mddev_t *mddev = v;
3725
3726 if (mddev && v != (void*)1 && v != (void*)2)
3727 mddev_put(mddev);
3728 }
3729
3730 struct mdstat_info {
3731 int event;
3732 };
3733
3734 static int md_seq_show(struct seq_file *seq, void *v)
3735 {
3736 mddev_t *mddev = v;
3737 sector_t size;
3738 struct list_head *tmp2;
3739 mdk_rdev_t *rdev;
3740 struct mdstat_info *mi = seq->private;
3741 struct bitmap *bitmap;
3742
3743 if (v == (void*)1) {
3744 struct mdk_personality *pers;
3745 seq_printf(seq, "Personalities : ");
3746 spin_lock(&pers_lock);
3747 list_for_each_entry(pers, &pers_list, list)
3748 seq_printf(seq, "[%s] ", pers->name);
3749
3750 spin_unlock(&pers_lock);
3751 seq_printf(seq, "\n");
3752 mi->event = atomic_read(&md_event_count);
3753 return 0;
3754 }
3755 if (v == (void*)2) {
3756 status_unused(seq);
3757 return 0;
3758 }
3759
3760 if (mddev_lock(mddev)!=0)
3761 return -EINTR;
3762 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3763 seq_printf(seq, "%s : %sactive", mdname(mddev),
3764 mddev->pers ? "" : "in");
3765 if (mddev->pers) {
3766 if (mddev->ro==1)
3767 seq_printf(seq, " (read-only)");
3768 if (mddev->ro==2)
3769 seq_printf(seq, "(auto-read-only)");
3770 seq_printf(seq, " %s", mddev->pers->name);
3771 }
3772
3773 size = 0;
3774 ITERATE_RDEV(mddev,rdev,tmp2) {
3775 char b[BDEVNAME_SIZE];
3776 seq_printf(seq, " %s[%d]",
3777 bdevname(rdev->bdev,b), rdev->desc_nr);
3778 if (test_bit(WriteMostly, &rdev->flags))
3779 seq_printf(seq, "(W)");
3780 if (test_bit(Faulty, &rdev->flags)) {
3781 seq_printf(seq, "(F)");
3782 continue;
3783 } else if (rdev->raid_disk < 0)
3784 seq_printf(seq, "(S)"); /* spare */
3785 size += rdev->size;
3786 }
3787
3788 if (!list_empty(&mddev->disks)) {
3789 if (mddev->pers)
3790 seq_printf(seq, "\n %llu blocks",
3791 (unsigned long long)mddev->array_size);
3792 else
3793 seq_printf(seq, "\n %llu blocks",
3794 (unsigned long long)size);
3795 }
3796 if (mddev->persistent) {
3797 if (mddev->major_version != 0 ||
3798 mddev->minor_version != 90) {
3799 seq_printf(seq," super %d.%d",
3800 mddev->major_version,
3801 mddev->minor_version);
3802 }
3803 } else
3804 seq_printf(seq, " super non-persistent");
3805
3806 if (mddev->pers) {
3807 mddev->pers->status (seq, mddev);
3808 seq_printf(seq, "\n ");
3809 if (mddev->pers->sync_request) {
3810 if (mddev->curr_resync > 2) {
3811 status_resync (seq, mddev);
3812 seq_printf(seq, "\n ");
3813 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3814 seq_printf(seq, "\tresync=DELAYED\n ");
3815 else if (mddev->recovery_cp < MaxSector)
3816 seq_printf(seq, "\tresync=PENDING\n ");
3817 }
3818 } else
3819 seq_printf(seq, "\n ");
3820
3821 if ((bitmap = mddev->bitmap)) {
3822 unsigned long chunk_kb;
3823 unsigned long flags;
3824 spin_lock_irqsave(&bitmap->lock, flags);
3825 chunk_kb = bitmap->chunksize >> 10;
3826 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3827 "%lu%s chunk",
3828 bitmap->pages - bitmap->missing_pages,
3829 bitmap->pages,
3830 (bitmap->pages - bitmap->missing_pages)
3831 << (PAGE_SHIFT - 10),
3832 chunk_kb ? chunk_kb : bitmap->chunksize,
3833 chunk_kb ? "KB" : "B");
3834 if (bitmap->file) {
3835 seq_printf(seq, ", file: ");
3836 seq_path(seq, bitmap->file->f_vfsmnt,
3837 bitmap->file->f_dentry," \t\n");
3838 }
3839
3840 seq_printf(seq, "\n");
3841 spin_unlock_irqrestore(&bitmap->lock, flags);
3842 }
3843
3844 seq_printf(seq, "\n");
3845 }
3846 mddev_unlock(mddev);
3847
3848 return 0;
3849 }
3850
3851 static struct seq_operations md_seq_ops = {
3852 .start = md_seq_start,
3853 .next = md_seq_next,
3854 .stop = md_seq_stop,
3855 .show = md_seq_show,
3856 };
3857
3858 static int md_seq_open(struct inode *inode, struct file *file)
3859 {
3860 int error;
3861 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3862 if (mi == NULL)
3863 return -ENOMEM;
3864
3865 error = seq_open(file, &md_seq_ops);
3866 if (error)
3867 kfree(mi);
3868 else {
3869 struct seq_file *p = file->private_data;
3870 p->private = mi;
3871 mi->event = atomic_read(&md_event_count);
3872 }
3873 return error;
3874 }
3875
3876 static int md_seq_release(struct inode *inode, struct file *file)
3877 {
3878 struct seq_file *m = file->private_data;
3879 struct mdstat_info *mi = m->private;
3880 m->private = NULL;
3881 kfree(mi);
3882 return seq_release(inode, file);
3883 }
3884
3885 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
3886 {
3887 struct seq_file *m = filp->private_data;
3888 struct mdstat_info *mi = m->private;
3889 int mask;
3890
3891 poll_wait(filp, &md_event_waiters, wait);
3892
3893 /* always allow read */
3894 mask = POLLIN | POLLRDNORM;
3895
3896 if (mi->event != atomic_read(&md_event_count))
3897 mask |= POLLERR | POLLPRI;
3898 return mask;
3899 }
3900
3901 static struct file_operations md_seq_fops = {
3902 .open = md_seq_open,
3903 .read = seq_read,
3904 .llseek = seq_lseek,
3905 .release = md_seq_release,
3906 .poll = mdstat_poll,
3907 };
3908
3909 int register_md_personality(struct mdk_personality *p)
3910 {
3911 spin_lock(&pers_lock);
3912 list_add_tail(&p->list, &pers_list);
3913 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
3914 spin_unlock(&pers_lock);
3915 return 0;
3916 }
3917
3918 int unregister_md_personality(struct mdk_personality *p)
3919 {
3920 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
3921 spin_lock(&pers_lock);
3922 list_del_init(&p->list);
3923 spin_unlock(&pers_lock);
3924 return 0;
3925 }
3926
3927 static int is_mddev_idle(mddev_t *mddev)
3928 {
3929 mdk_rdev_t * rdev;
3930 struct list_head *tmp;
3931 int idle;
3932 unsigned long curr_events;
3933
3934 idle = 1;
3935 ITERATE_RDEV(mddev,rdev,tmp) {
3936 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3937 curr_events = disk_stat_read(disk, sectors[0]) +
3938 disk_stat_read(disk, sectors[1]) -
3939 atomic_read(&disk->sync_io);
3940 /* The difference between curr_events and last_events
3941 * will be affected by any new non-sync IO (making
3942 * curr_events bigger) and any difference in the amount of
3943 * in-flight syncio (making current_events bigger or smaller)
3944 * The amount in-flight is currently limited to
3945 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3946 * which is at most 4096 sectors.
3947 * These numbers are fairly fragile and should be made
3948 * more robust, probably by enforcing the
3949 * 'window size' that md_do_sync sort-of uses.
3950 *
3951 * Note: the following is an unsigned comparison.
3952 */
3953 if ((curr_events - rdev->last_events + 4096) > 8192) {
3954 rdev->last_events = curr_events;
3955 idle = 0;
3956 }
3957 }
3958 return idle;
3959 }
3960
3961 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3962 {
3963 /* another "blocks" (512byte) blocks have been synced */
3964 atomic_sub(blocks, &mddev->recovery_active);
3965 wake_up(&mddev->recovery_wait);
3966 if (!ok) {
3967 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3968 md_wakeup_thread(mddev->thread);
3969 // stop recovery, signal do_sync ....
3970 }
3971 }
3972
3973
3974 /* md_write_start(mddev, bi)
3975 * If we need to update some array metadata (e.g. 'active' flag
3976 * in superblock) before writing, schedule a superblock update
3977 * and wait for it to complete.
3978 */
3979 void md_write_start(mddev_t *mddev, struct bio *bi)
3980 {
3981 if (bio_data_dir(bi) != WRITE)
3982 return;
3983
3984 BUG_ON(mddev->ro == 1);
3985 if (mddev->ro == 2) {
3986 /* need to switch to read/write */
3987 mddev->ro = 0;
3988 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3989 md_wakeup_thread(mddev->thread);
3990 }
3991 atomic_inc(&mddev->writes_pending);
3992 if (mddev->in_sync) {
3993 spin_lock_irq(&mddev->write_lock);
3994 if (mddev->in_sync) {
3995 mddev->in_sync = 0;
3996 mddev->sb_dirty = 1;
3997 md_wakeup_thread(mddev->thread);
3998 }
3999 spin_unlock_irq(&mddev->write_lock);
4000 }
4001 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4002 }
4003
4004 void md_write_end(mddev_t *mddev)
4005 {
4006 if (atomic_dec_and_test(&mddev->writes_pending)) {
4007 if (mddev->safemode == 2)
4008 md_wakeup_thread(mddev->thread);
4009 else
4010 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4011 }
4012 }
4013
4014 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4015
4016 #define SYNC_MARKS 10
4017 #define SYNC_MARK_STEP (3*HZ)
4018 static void md_do_sync(mddev_t *mddev)
4019 {
4020 mddev_t *mddev2;
4021 unsigned int currspeed = 0,
4022 window;
4023 sector_t max_sectors,j, io_sectors;
4024 unsigned long mark[SYNC_MARKS];
4025 sector_t mark_cnt[SYNC_MARKS];
4026 int last_mark,m;
4027 struct list_head *tmp;
4028 sector_t last_check;
4029 int skipped = 0;
4030
4031 /* just incase thread restarts... */
4032 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4033 return;
4034
4035 /* we overload curr_resync somewhat here.
4036 * 0 == not engaged in resync at all
4037 * 2 == checking that there is no conflict with another sync
4038 * 1 == like 2, but have yielded to allow conflicting resync to
4039 * commense
4040 * other == active in resync - this many blocks
4041 *
4042 * Before starting a resync we must have set curr_resync to
4043 * 2, and then checked that every "conflicting" array has curr_resync
4044 * less than ours. When we find one that is the same or higher
4045 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4046 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4047 * This will mean we have to start checking from the beginning again.
4048 *
4049 */
4050
4051 do {
4052 mddev->curr_resync = 2;
4053
4054 try_again:
4055 if (kthread_should_stop()) {
4056 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4057 goto skip;
4058 }
4059 ITERATE_MDDEV(mddev2,tmp) {
4060 if (mddev2 == mddev)
4061 continue;
4062 if (mddev2->curr_resync &&
4063 match_mddev_units(mddev,mddev2)) {
4064 DEFINE_WAIT(wq);
4065 if (mddev < mddev2 && mddev->curr_resync == 2) {
4066 /* arbitrarily yield */
4067 mddev->curr_resync = 1;
4068 wake_up(&resync_wait);
4069 }
4070 if (mddev > mddev2 && mddev->curr_resync == 1)
4071 /* no need to wait here, we can wait the next
4072 * time 'round when curr_resync == 2
4073 */
4074 continue;
4075 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4076 if (!kthread_should_stop() &&
4077 mddev2->curr_resync >= mddev->curr_resync) {
4078 printk(KERN_INFO "md: delaying resync of %s"
4079 " until %s has finished resync (they"
4080 " share one or more physical units)\n",
4081 mdname(mddev), mdname(mddev2));
4082 mddev_put(mddev2);
4083 schedule();
4084 finish_wait(&resync_wait, &wq);
4085 goto try_again;
4086 }
4087 finish_wait(&resync_wait, &wq);
4088 }
4089 }
4090 } while (mddev->curr_resync < 2);
4091
4092 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4093 /* resync follows the size requested by the personality,
4094 * which defaults to physical size, but can be virtual size
4095 */
4096 max_sectors = mddev->resync_max_sectors;
4097 mddev->resync_mismatches = 0;
4098 } else
4099 /* recovery follows the physical size of devices */
4100 max_sectors = mddev->size << 1;
4101
4102 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4103 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4104 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4105 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4106 "(but not more than %d KB/sec) for reconstruction.\n",
4107 sysctl_speed_limit_max);
4108
4109 is_mddev_idle(mddev); /* this also initializes IO event counters */
4110 /* we don't use the checkpoint if there's a bitmap */
4111 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4112 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4113 j = mddev->recovery_cp;
4114 else
4115 j = 0;
4116 io_sectors = 0;
4117 for (m = 0; m < SYNC_MARKS; m++) {
4118 mark[m] = jiffies;
4119 mark_cnt[m] = io_sectors;
4120 }
4121 last_mark = 0;
4122 mddev->resync_mark = mark[last_mark];
4123 mddev->resync_mark_cnt = mark_cnt[last_mark];
4124
4125 /*
4126 * Tune reconstruction:
4127 */
4128 window = 32*(PAGE_SIZE/512);
4129 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4130 window/2,(unsigned long long) max_sectors/2);
4131
4132 atomic_set(&mddev->recovery_active, 0);
4133 init_waitqueue_head(&mddev->recovery_wait);
4134 last_check = 0;
4135
4136 if (j>2) {
4137 printk(KERN_INFO
4138 "md: resuming recovery of %s from checkpoint.\n",
4139 mdname(mddev));
4140 mddev->curr_resync = j;
4141 }
4142
4143 while (j < max_sectors) {
4144 sector_t sectors;
4145
4146 skipped = 0;
4147 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4148 currspeed < sysctl_speed_limit_min);
4149 if (sectors == 0) {
4150 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4151 goto out;
4152 }
4153
4154 if (!skipped) { /* actual IO requested */
4155 io_sectors += sectors;
4156 atomic_add(sectors, &mddev->recovery_active);
4157 }
4158
4159 j += sectors;
4160 if (j>1) mddev->curr_resync = j;
4161 if (last_check == 0)
4162 /* this is the earliers that rebuilt will be
4163 * visible in /proc/mdstat
4164 */
4165 md_new_event(mddev);
4166
4167 if (last_check + window > io_sectors || j == max_sectors)
4168 continue;
4169
4170 last_check = io_sectors;
4171
4172 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4173 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4174 break;
4175
4176 repeat:
4177 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4178 /* step marks */
4179 int next = (last_mark+1) % SYNC_MARKS;
4180
4181 mddev->resync_mark = mark[next];
4182 mddev->resync_mark_cnt = mark_cnt[next];
4183 mark[next] = jiffies;
4184 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4185 last_mark = next;
4186 }
4187
4188
4189 if (kthread_should_stop()) {
4190 /*
4191 * got a signal, exit.
4192 */
4193 printk(KERN_INFO
4194 "md: md_do_sync() got signal ... exiting\n");
4195 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4196 goto out;
4197 }
4198
4199 /*
4200 * this loop exits only if either when we are slower than
4201 * the 'hard' speed limit, or the system was IO-idle for
4202 * a jiffy.
4203 * the system might be non-idle CPU-wise, but we only care
4204 * about not overloading the IO subsystem. (things like an
4205 * e2fsck being done on the RAID array should execute fast)
4206 */
4207 mddev->queue->unplug_fn(mddev->queue);
4208 cond_resched();
4209
4210 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4211 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4212
4213 if (currspeed > sysctl_speed_limit_min) {
4214 if ((currspeed > sysctl_speed_limit_max) ||
4215 !is_mddev_idle(mddev)) {
4216 msleep(500);
4217 goto repeat;
4218 }
4219 }
4220 }
4221 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4222 /*
4223 * this also signals 'finished resyncing' to md_stop
4224 */
4225 out:
4226 mddev->queue->unplug_fn(mddev->queue);
4227
4228 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4229
4230 /* tell personality that we are finished */
4231 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4232
4233 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4234 mddev->curr_resync > 2 &&
4235 mddev->curr_resync >= mddev->recovery_cp) {
4236 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4237 printk(KERN_INFO
4238 "md: checkpointing recovery of %s.\n",
4239 mdname(mddev));
4240 mddev->recovery_cp = mddev->curr_resync;
4241 } else
4242 mddev->recovery_cp = MaxSector;
4243 }
4244
4245 skip:
4246 mddev->curr_resync = 0;
4247 wake_up(&resync_wait);
4248 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4249 md_wakeup_thread(mddev->thread);
4250 }
4251
4252
4253 /*
4254 * This routine is regularly called by all per-raid-array threads to
4255 * deal with generic issues like resync and super-block update.
4256 * Raid personalities that don't have a thread (linear/raid0) do not
4257 * need this as they never do any recovery or update the superblock.
4258 *
4259 * It does not do any resync itself, but rather "forks" off other threads
4260 * to do that as needed.
4261 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4262 * "->recovery" and create a thread at ->sync_thread.
4263 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4264 * and wakeups up this thread which will reap the thread and finish up.
4265 * This thread also removes any faulty devices (with nr_pending == 0).
4266 *
4267 * The overall approach is:
4268 * 1/ if the superblock needs updating, update it.
4269 * 2/ If a recovery thread is running, don't do anything else.
4270 * 3/ If recovery has finished, clean up, possibly marking spares active.
4271 * 4/ If there are any faulty devices, remove them.
4272 * 5/ If array is degraded, try to add spares devices
4273 * 6/ If array has spares or is not in-sync, start a resync thread.
4274 */
4275 void md_check_recovery(mddev_t *mddev)
4276 {
4277 mdk_rdev_t *rdev;
4278 struct list_head *rtmp;
4279
4280
4281 if (mddev->bitmap)
4282 bitmap_daemon_work(mddev->bitmap);
4283
4284 if (mddev->ro)
4285 return;
4286
4287 if (signal_pending(current)) {
4288 if (mddev->pers->sync_request) {
4289 printk(KERN_INFO "md: %s in immediate safe mode\n",
4290 mdname(mddev));
4291 mddev->safemode = 2;
4292 }
4293 flush_signals(current);
4294 }
4295
4296 if ( ! (
4297 mddev->sb_dirty ||
4298 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4299 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4300 (mddev->safemode == 1) ||
4301 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4302 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4303 ))
4304 return;
4305
4306 if (mddev_trylock(mddev)==0) {
4307 int spares =0;
4308
4309 spin_lock_irq(&mddev->write_lock);
4310 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4311 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4312 mddev->in_sync = 1;
4313 mddev->sb_dirty = 1;
4314 }
4315 if (mddev->safemode == 1)
4316 mddev->safemode = 0;
4317 spin_unlock_irq(&mddev->write_lock);
4318
4319 if (mddev->sb_dirty)
4320 md_update_sb(mddev);
4321
4322
4323 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4324 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4325 /* resync/recovery still happening */
4326 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4327 goto unlock;
4328 }
4329 if (mddev->sync_thread) {
4330 /* resync has finished, collect result */
4331 md_unregister_thread(mddev->sync_thread);
4332 mddev->sync_thread = NULL;
4333 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4334 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4335 /* success...*/
4336 /* activate any spares */
4337 mddev->pers->spare_active(mddev);
4338 }
4339 md_update_sb(mddev);
4340
4341 /* if array is no-longer degraded, then any saved_raid_disk
4342 * information must be scrapped
4343 */
4344 if (!mddev->degraded)
4345 ITERATE_RDEV(mddev,rdev,rtmp)
4346 rdev->saved_raid_disk = -1;
4347
4348 mddev->recovery = 0;
4349 /* flag recovery needed just to double check */
4350 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4351 md_new_event(mddev);
4352 goto unlock;
4353 }
4354 /* Clear some bits that don't mean anything, but
4355 * might be left set
4356 */
4357 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4358 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4359 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4360 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4361
4362 /* no recovery is running.
4363 * remove any failed drives, then
4364 * add spares if possible.
4365 * Spare are also removed and re-added, to allow
4366 * the personality to fail the re-add.
4367 */
4368 ITERATE_RDEV(mddev,rdev,rtmp)
4369 if (rdev->raid_disk >= 0 &&
4370 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4371 atomic_read(&rdev->nr_pending)==0) {
4372 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4373 char nm[20];
4374 sprintf(nm,"rd%d", rdev->raid_disk);
4375 sysfs_remove_link(&mddev->kobj, nm);
4376 rdev->raid_disk = -1;
4377 }
4378 }
4379
4380 if (mddev->degraded) {
4381 ITERATE_RDEV(mddev,rdev,rtmp)
4382 if (rdev->raid_disk < 0
4383 && !test_bit(Faulty, &rdev->flags)) {
4384 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4385 char nm[20];
4386 sprintf(nm, "rd%d", rdev->raid_disk);
4387 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4388 spares++;
4389 md_new_event(mddev);
4390 } else
4391 break;
4392 }
4393 }
4394
4395 if (spares) {
4396 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4397 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4398 } else if (mddev->recovery_cp < MaxSector) {
4399 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4400 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4401 /* nothing to be done ... */
4402 goto unlock;
4403
4404 if (mddev->pers->sync_request) {
4405 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4406 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4407 /* We are adding a device or devices to an array
4408 * which has the bitmap stored on all devices.
4409 * So make sure all bitmap pages get written
4410 */
4411 bitmap_write_all(mddev->bitmap);
4412 }
4413 mddev->sync_thread = md_register_thread(md_do_sync,
4414 mddev,
4415 "%s_resync");
4416 if (!mddev->sync_thread) {
4417 printk(KERN_ERR "%s: could not start resync"
4418 " thread...\n",
4419 mdname(mddev));
4420 /* leave the spares where they are, it shouldn't hurt */
4421 mddev->recovery = 0;
4422 } else
4423 md_wakeup_thread(mddev->sync_thread);
4424 md_new_event(mddev);
4425 }
4426 unlock:
4427 mddev_unlock(mddev);
4428 }
4429 }
4430
4431 static int md_notify_reboot(struct notifier_block *this,
4432 unsigned long code, void *x)
4433 {
4434 struct list_head *tmp;
4435 mddev_t *mddev;
4436
4437 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4438
4439 printk(KERN_INFO "md: stopping all md devices.\n");
4440
4441 ITERATE_MDDEV(mddev,tmp)
4442 if (mddev_trylock(mddev)==0)
4443 do_md_stop (mddev, 1);
4444 /*
4445 * certain more exotic SCSI devices are known to be
4446 * volatile wrt too early system reboots. While the
4447 * right place to handle this issue is the given
4448 * driver, we do want to have a safe RAID driver ...
4449 */
4450 mdelay(1000*1);
4451 }
4452 return NOTIFY_DONE;
4453 }
4454
4455 static struct notifier_block md_notifier = {
4456 .notifier_call = md_notify_reboot,
4457 .next = NULL,
4458 .priority = INT_MAX, /* before any real devices */
4459 };
4460
4461 static void md_geninit(void)
4462 {
4463 struct proc_dir_entry *p;
4464
4465 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4466
4467 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4468 if (p)
4469 p->proc_fops = &md_seq_fops;
4470 }
4471
4472 static int __init md_init(void)
4473 {
4474 int minor;
4475
4476 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4477 " MD_SB_DISKS=%d\n",
4478 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4479 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4480 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4481 BITMAP_MINOR);
4482
4483 if (register_blkdev(MAJOR_NR, "md"))
4484 return -1;
4485 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4486 unregister_blkdev(MAJOR_NR, "md");
4487 return -1;
4488 }
4489 devfs_mk_dir("md");
4490 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4491 md_probe, NULL, NULL);
4492 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4493 md_probe, NULL, NULL);
4494
4495 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4496 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4497 S_IFBLK|S_IRUSR|S_IWUSR,
4498 "md/%d", minor);
4499
4500 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4501 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4502 S_IFBLK|S_IRUSR|S_IWUSR,
4503 "md/mdp%d", minor);
4504
4505
4506 register_reboot_notifier(&md_notifier);
4507 raid_table_header = register_sysctl_table(raid_root_table, 1);
4508
4509 md_geninit();
4510 return (0);
4511 }
4512
4513
4514 #ifndef MODULE
4515
4516 /*
4517 * Searches all registered partitions for autorun RAID arrays
4518 * at boot time.
4519 */
4520 static dev_t detected_devices[128];
4521 static int dev_cnt;
4522
4523 void md_autodetect_dev(dev_t dev)
4524 {
4525 if (dev_cnt >= 0 && dev_cnt < 127)
4526 detected_devices[dev_cnt++] = dev;
4527 }
4528
4529
4530 static void autostart_arrays(int part)
4531 {
4532 mdk_rdev_t *rdev;
4533 int i;
4534
4535 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4536
4537 for (i = 0; i < dev_cnt; i++) {
4538 dev_t dev = detected_devices[i];
4539
4540 rdev = md_import_device(dev,0, 0);
4541 if (IS_ERR(rdev))
4542 continue;
4543
4544 if (test_bit(Faulty, &rdev->flags)) {
4545 MD_BUG();
4546 continue;
4547 }
4548 list_add(&rdev->same_set, &pending_raid_disks);
4549 }
4550 dev_cnt = 0;
4551
4552 autorun_devices(part);
4553 }
4554
4555 #endif
4556
4557 static __exit void md_exit(void)
4558 {
4559 mddev_t *mddev;
4560 struct list_head *tmp;
4561 int i;
4562 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4563 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4564 for (i=0; i < MAX_MD_DEVS; i++)
4565 devfs_remove("md/%d", i);
4566 for (i=0; i < MAX_MD_DEVS; i++)
4567 devfs_remove("md/d%d", i);
4568
4569 devfs_remove("md");
4570
4571 unregister_blkdev(MAJOR_NR,"md");
4572 unregister_blkdev(mdp_major, "mdp");
4573 unregister_reboot_notifier(&md_notifier);
4574 unregister_sysctl_table(raid_table_header);
4575 remove_proc_entry("mdstat", NULL);
4576 ITERATE_MDDEV(mddev,tmp) {
4577 struct gendisk *disk = mddev->gendisk;
4578 if (!disk)
4579 continue;
4580 export_array(mddev);
4581 del_gendisk(disk);
4582 put_disk(disk);
4583 mddev->gendisk = NULL;
4584 mddev_put(mddev);
4585 }
4586 }
4587
4588 module_init(md_init)
4589 module_exit(md_exit)
4590
4591 static int get_ro(char *buffer, struct kernel_param *kp)
4592 {
4593 return sprintf(buffer, "%d", start_readonly);
4594 }
4595 static int set_ro(const char *val, struct kernel_param *kp)
4596 {
4597 char *e;
4598 int num = simple_strtoul(val, &e, 10);
4599 if (*val && (*e == '\0' || *e == '\n')) {
4600 start_readonly = num;
4601 return 0;;
4602 }
4603 return -EINVAL;
4604 }
4605
4606 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4607 module_param(start_dirty_degraded, int, 0644);
4608
4609
4610 EXPORT_SYMBOL(register_md_personality);
4611 EXPORT_SYMBOL(unregister_md_personality);
4612 EXPORT_SYMBOL(md_error);
4613 EXPORT_SYMBOL(md_done_sync);
4614 EXPORT_SYMBOL(md_write_start);
4615 EXPORT_SYMBOL(md_write_end);
4616 EXPORT_SYMBOL(md_register_thread);
4617 EXPORT_SYMBOL(md_unregister_thread);
4618 EXPORT_SYMBOL(md_wakeup_thread);
4619 EXPORT_SYMBOL(md_print_devices);
4620 EXPORT_SYMBOL(md_check_recovery);
4621 MODULE_LICENSE("GPL");
4622 MODULE_ALIAS("md");
4623 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree