?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
md: move assignment of ->utime so that it never gets skipped.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blobd9cce54cc2d5d34aa9a3dfe1151a32ec1f6c4b0c
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/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
77 * idle IO detection.
78 *
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
81 */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100 {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
107 },
108 {
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
115 },
116 { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120 {
121 .ctl_name = DEV_RAID,
122 .procname = "raid",
123 .maxlen = 0,
124 .mode = S_IRUGO|S_IXUGO,
125 .child = raid_table,
126 },
127 { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131 {
132 .ctl_name = CTL_DEV,
133 .procname = "dev",
134 .maxlen = 0,
135 .mode = 0555,
136 .child = raid_dir_table,
137 },
138 { .ctl_name = 0 }
139 };
140
141 static struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
149 * count increases.
150 *
151 * Events are:
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
154 */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
166 */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171 }
172
173 /*
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
176 */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
187 */
188 #define for_each_mddev(mddev,tmp) \
189 \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
192 mddev = NULL;}); \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
200 tmp = tmp->next;}) \
201 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
210 */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213 mddev_t *mddev = q->queuedata;
214 int rv;
215 if (mddev == NULL || mddev->pers == NULL) {
216 bio_io_error(bio);
217 return 0;
218 }
219 rcu_read_lock();
220 if (mddev->suspended) {
221 DEFINE_WAIT(__wait);
222 for (;;) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
226 break;
227 rcu_read_unlock();
228 schedule();
229 rcu_read_lock();
230 }
231 finish_wait(&mddev->sb_wait, &__wait);
232 }
233 atomic_inc(&mddev->active_io);
234 rcu_read_unlock();
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
238
239 return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
246 synchronize_rcu();
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
254 * unloaded
255 */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
263 }
264
265
266 static inline mddev_t *mddev_get(mddev_t *mddev)
267 {
268 atomic_inc(&mddev->active);
269 return mddev;
270 }
271
272 static void mddev_delayed_delete(struct work_struct *ws);
273
274 static void mddev_put(mddev_t *mddev)
275 {
276 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277 return;
278 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279 !mddev->hold_active) {
280 list_del(&mddev->all_mddevs);
281 if (mddev->gendisk) {
282 /* we did a probe so need to clean up.
283 * Call schedule_work inside the spinlock
284 * so that flush_scheduled_work() after
285 * mddev_find will succeed in waiting for the
286 * work to be done.
287 */
288 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289 schedule_work(&mddev->del_work);
290 } else
291 kfree(mddev);
292 }
293 spin_unlock(&all_mddevs_lock);
294 }
295
296 static mddev_t * mddev_find(dev_t unit)
297 {
298 mddev_t *mddev, *new = NULL;
299
300 retry:
301 spin_lock(&all_mddevs_lock);
302
303 if (unit) {
304 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305 if (mddev->unit == unit) {
306 mddev_get(mddev);
307 spin_unlock(&all_mddevs_lock);
308 kfree(new);
309 return mddev;
310 }
311
312 if (new) {
313 list_add(&new->all_mddevs, &all_mddevs);
314 spin_unlock(&all_mddevs_lock);
315 new->hold_active = UNTIL_IOCTL;
316 return new;
317 }
318 } else if (new) {
319 /* find an unused unit number */
320 static int next_minor = 512;
321 int start = next_minor;
322 int is_free = 0;
323 int dev = 0;
324 while (!is_free) {
325 dev = MKDEV(MD_MAJOR, next_minor);
326 next_minor++;
327 if (next_minor > MINORMASK)
328 next_minor = 0;
329 if (next_minor == start) {
330 /* Oh dear, all in use. */
331 spin_unlock(&all_mddevs_lock);
332 kfree(new);
333 return NULL;
334 }
335
336 is_free = 1;
337 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338 if (mddev->unit == dev) {
339 is_free = 0;
340 break;
341 }
342 }
343 new->unit = dev;
344 new->md_minor = MINOR(dev);
345 new->hold_active = UNTIL_STOP;
346 list_add(&new->all_mddevs, &all_mddevs);
347 spin_unlock(&all_mddevs_lock);
348 return new;
349 }
350 spin_unlock(&all_mddevs_lock);
351
352 new = kzalloc(sizeof(*new), GFP_KERNEL);
353 if (!new)
354 return NULL;
355
356 new->unit = unit;
357 if (MAJOR(unit) == MD_MAJOR)
358 new->md_minor = MINOR(unit);
359 else
360 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
362 mutex_init(&new->reconfig_mutex);
363 INIT_LIST_HEAD(&new->disks);
364 INIT_LIST_HEAD(&new->all_mddevs);
365 init_timer(&new->safemode_timer);
366 atomic_set(&new->active, 1);
367 atomic_set(&new->openers, 0);
368 atomic_set(&new->active_io, 0);
369 spin_lock_init(&new->write_lock);
370 init_waitqueue_head(&new->sb_wait);
371 init_waitqueue_head(&new->recovery_wait);
372 new->reshape_position = MaxSector;
373 new->resync_min = 0;
374 new->resync_max = MaxSector;
375 new->level = LEVEL_NONE;
376
377 goto retry;
378 }
379
380 static inline int mddev_lock(mddev_t * mddev)
381 {
382 return mutex_lock_interruptible(&mddev->reconfig_mutex);
383 }
384
385 static inline int mddev_is_locked(mddev_t *mddev)
386 {
387 return mutex_is_locked(&mddev->reconfig_mutex);
388 }
389
390 static inline int mddev_trylock(mddev_t * mddev)
391 {
392 return mutex_trylock(&mddev->reconfig_mutex);
393 }
394
395 static inline void mddev_unlock(mddev_t * mddev)
396 {
397 mutex_unlock(&mddev->reconfig_mutex);
398
399 md_wakeup_thread(mddev->thread);
400 }
401
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
403 {
404 mdk_rdev_t *rdev;
405
406 list_for_each_entry(rdev, &mddev->disks, same_set)
407 if (rdev->desc_nr == nr)
408 return rdev;
409
410 return NULL;
411 }
412
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414 {
415 mdk_rdev_t *rdev;
416
417 list_for_each_entry(rdev, &mddev->disks, same_set)
418 if (rdev->bdev->bd_dev == dev)
419 return rdev;
420
421 return NULL;
422 }
423
424 static struct mdk_personality *find_pers(int level, char *clevel)
425 {
426 struct mdk_personality *pers;
427 list_for_each_entry(pers, &pers_list, list) {
428 if (level != LEVEL_NONE && pers->level == level)
429 return pers;
430 if (strcmp(pers->name, clevel)==0)
431 return pers;
432 }
433 return NULL;
434 }
435
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
438 {
439 sector_t num_sectors = bdev->bd_inode->i_size / 512;
440 return MD_NEW_SIZE_SECTORS(num_sectors);
441 }
442
443 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
444 {
445 sector_t num_sectors = rdev->sb_start;
446
447 if (chunk_size) {
448 unsigned chunk_sects = chunk_size>>9;
449 sector_div(num_sectors, chunk_sects);
450 num_sectors *= chunk_sects;
451 }
452 return num_sectors;
453 }
454
455 static int alloc_disk_sb(mdk_rdev_t * rdev)
456 {
457 if (rdev->sb_page)
458 MD_BUG();
459
460 rdev->sb_page = alloc_page(GFP_KERNEL);
461 if (!rdev->sb_page) {
462 printk(KERN_ALERT "md: out of memory.\n");
463 return -ENOMEM;
464 }
465
466 return 0;
467 }
468
469 static void free_disk_sb(mdk_rdev_t * rdev)
470 {
471 if (rdev->sb_page) {
472 put_page(rdev->sb_page);
473 rdev->sb_loaded = 0;
474 rdev->sb_page = NULL;
475 rdev->sb_start = 0;
476 rdev->sectors = 0;
477 }
478 }
479
480
481 static void super_written(struct bio *bio, int error)
482 {
483 mdk_rdev_t *rdev = bio->bi_private;
484 mddev_t *mddev = rdev->mddev;
485
486 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
487 printk("md: super_written gets error=%d, uptodate=%d\n",
488 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
489 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
490 md_error(mddev, rdev);
491 }
492
493 if (atomic_dec_and_test(&mddev->pending_writes))
494 wake_up(&mddev->sb_wait);
495 bio_put(bio);
496 }
497
498 static void super_written_barrier(struct bio *bio, int error)
499 {
500 struct bio *bio2 = bio->bi_private;
501 mdk_rdev_t *rdev = bio2->bi_private;
502 mddev_t *mddev = rdev->mddev;
503
504 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
505 error == -EOPNOTSUPP) {
506 unsigned long flags;
507 /* barriers don't appear to be supported :-( */
508 set_bit(BarriersNotsupp, &rdev->flags);
509 mddev->barriers_work = 0;
510 spin_lock_irqsave(&mddev->write_lock, flags);
511 bio2->bi_next = mddev->biolist;
512 mddev->biolist = bio2;
513 spin_unlock_irqrestore(&mddev->write_lock, flags);
514 wake_up(&mddev->sb_wait);
515 bio_put(bio);
516 } else {
517 bio_put(bio2);
518 bio->bi_private = rdev;
519 super_written(bio, error);
520 }
521 }
522
523 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
524 sector_t sector, int size, struct page *page)
525 {
526 /* write first size bytes of page to sector of rdev
527 * Increment mddev->pending_writes before returning
528 * and decrement it on completion, waking up sb_wait
529 * if zero is reached.
530 * If an error occurred, call md_error
531 *
532 * As we might need to resubmit the request if BIO_RW_BARRIER
533 * causes ENOTSUPP, we allocate a spare bio...
534 */
535 struct bio *bio = bio_alloc(GFP_NOIO, 1);
536 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
537
538 bio->bi_bdev = rdev->bdev;
539 bio->bi_sector = sector;
540 bio_add_page(bio, page, size, 0);
541 bio->bi_private = rdev;
542 bio->bi_end_io = super_written;
543 bio->bi_rw = rw;
544
545 atomic_inc(&mddev->pending_writes);
546 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
547 struct bio *rbio;
548 rw |= (1<<BIO_RW_BARRIER);
549 rbio = bio_clone(bio, GFP_NOIO);
550 rbio->bi_private = bio;
551 rbio->bi_end_io = super_written_barrier;
552 submit_bio(rw, rbio);
553 } else
554 submit_bio(rw, bio);
555 }
556
557 void md_super_wait(mddev_t *mddev)
558 {
559 /* wait for all superblock writes that were scheduled to complete.
560 * if any had to be retried (due to BARRIER problems), retry them
561 */
562 DEFINE_WAIT(wq);
563 for(;;) {
564 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
565 if (atomic_read(&mddev->pending_writes)==0)
566 break;
567 while (mddev->biolist) {
568 struct bio *bio;
569 spin_lock_irq(&mddev->write_lock);
570 bio = mddev->biolist;
571 mddev->biolist = bio->bi_next ;
572 bio->bi_next = NULL;
573 spin_unlock_irq(&mddev->write_lock);
574 submit_bio(bio->bi_rw, bio);
575 }
576 schedule();
577 }
578 finish_wait(&mddev->sb_wait, &wq);
579 }
580
581 static void bi_complete(struct bio *bio, int error)
582 {
583 complete((struct completion*)bio->bi_private);
584 }
585
586 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
587 struct page *page, int rw)
588 {
589 struct bio *bio = bio_alloc(GFP_NOIO, 1);
590 struct completion event;
591 int ret;
592
593 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
594
595 bio->bi_bdev = bdev;
596 bio->bi_sector = sector;
597 bio_add_page(bio, page, size, 0);
598 init_completion(&event);
599 bio->bi_private = &event;
600 bio->bi_end_io = bi_complete;
601 submit_bio(rw, bio);
602 wait_for_completion(&event);
603
604 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
605 bio_put(bio);
606 return ret;
607 }
608 EXPORT_SYMBOL_GPL(sync_page_io);
609
610 static int read_disk_sb(mdk_rdev_t * rdev, int size)
611 {
612 char b[BDEVNAME_SIZE];
613 if (!rdev->sb_page) {
614 MD_BUG();
615 return -EINVAL;
616 }
617 if (rdev->sb_loaded)
618 return 0;
619
620
621 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
622 goto fail;
623 rdev->sb_loaded = 1;
624 return 0;
625
626 fail:
627 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
628 bdevname(rdev->bdev,b));
629 return -EINVAL;
630 }
631
632 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
633 {
634 return sb1->set_uuid0 == sb2->set_uuid0 &&
635 sb1->set_uuid1 == sb2->set_uuid1 &&
636 sb1->set_uuid2 == sb2->set_uuid2 &&
637 sb1->set_uuid3 == sb2->set_uuid3;
638 }
639
640 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
641 {
642 int ret;
643 mdp_super_t *tmp1, *tmp2;
644
645 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
646 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
647
648 if (!tmp1 || !tmp2) {
649 ret = 0;
650 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
651 goto abort;
652 }
653
654 *tmp1 = *sb1;
655 *tmp2 = *sb2;
656
657 /*
658 * nr_disks is not constant
659 */
660 tmp1->nr_disks = 0;
661 tmp2->nr_disks = 0;
662
663 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
664 abort:
665 kfree(tmp1);
666 kfree(tmp2);
667 return ret;
668 }
669
670
671 static u32 md_csum_fold(u32 csum)
672 {
673 csum = (csum & 0xffff) + (csum >> 16);
674 return (csum & 0xffff) + (csum >> 16);
675 }
676
677 static unsigned int calc_sb_csum(mdp_super_t * sb)
678 {
679 u64 newcsum = 0;
680 u32 *sb32 = (u32*)sb;
681 int i;
682 unsigned int disk_csum, csum;
683
684 disk_csum = sb->sb_csum;
685 sb->sb_csum = 0;
686
687 for (i = 0; i < MD_SB_BYTES/4 ; i++)
688 newcsum += sb32[i];
689 csum = (newcsum & 0xffffffff) + (newcsum>>32);
690
691
692 #ifdef CONFIG_ALPHA
693 /* This used to use csum_partial, which was wrong for several
694 * reasons including that different results are returned on
695 * different architectures. It isn't critical that we get exactly
696 * the same return value as before (we always csum_fold before
697 * testing, and that removes any differences). However as we
698 * know that csum_partial always returned a 16bit value on
699 * alphas, do a fold to maximise conformity to previous behaviour.
700 */
701 sb->sb_csum = md_csum_fold(disk_csum);
702 #else
703 sb->sb_csum = disk_csum;
704 #endif
705 return csum;
706 }
707
708
709 /*
710 * Handle superblock details.
711 * We want to be able to handle multiple superblock formats
712 * so we have a common interface to them all, and an array of
713 * different handlers.
714 * We rely on user-space to write the initial superblock, and support
715 * reading and updating of superblocks.
716 * Interface methods are:
717 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
718 * loads and validates a superblock on dev.
719 * if refdev != NULL, compare superblocks on both devices
720 * Return:
721 * 0 - dev has a superblock that is compatible with refdev
722 * 1 - dev has a superblock that is compatible and newer than refdev
723 * so dev should be used as the refdev in future
724 * -EINVAL superblock incompatible or invalid
725 * -othererror e.g. -EIO
726 *
727 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
728 * Verify that dev is acceptable into mddev.
729 * The first time, mddev->raid_disks will be 0, and data from
730 * dev should be merged in. Subsequent calls check that dev
731 * is new enough. Return 0 or -EINVAL
732 *
733 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
734 * Update the superblock for rdev with data in mddev
735 * This does not write to disc.
736 *
737 */
738
739 struct super_type {
740 char *name;
741 struct module *owner;
742 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
743 int minor_version);
744 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
745 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
746 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
747 sector_t num_sectors);
748 };
749
750 /*
751 * load_super for 0.90.0
752 */
753 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
754 {
755 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
756 mdp_super_t *sb;
757 int ret;
758
759 /*
760 * Calculate the position of the superblock (512byte sectors),
761 * it's at the end of the disk.
762 *
763 * It also happens to be a multiple of 4Kb.
764 */
765 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
766
767 ret = read_disk_sb(rdev, MD_SB_BYTES);
768 if (ret) return ret;
769
770 ret = -EINVAL;
771
772 bdevname(rdev->bdev, b);
773 sb = (mdp_super_t*)page_address(rdev->sb_page);
774
775 if (sb->md_magic != MD_SB_MAGIC) {
776 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
777 b);
778 goto abort;
779 }
780
781 if (sb->major_version != 0 ||
782 sb->minor_version < 90 ||
783 sb->minor_version > 91) {
784 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
785 sb->major_version, sb->minor_version,
786 b);
787 goto abort;
788 }
789
790 if (sb->raid_disks <= 0)
791 goto abort;
792
793 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
794 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
795 b);
796 goto abort;
797 }
798
799 rdev->preferred_minor = sb->md_minor;
800 rdev->data_offset = 0;
801 rdev->sb_size = MD_SB_BYTES;
802
803 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
804 if (sb->level != 1 && sb->level != 4
805 && sb->level != 5 && sb->level != 6
806 && sb->level != 10) {
807 /* FIXME use a better test */
808 printk(KERN_WARNING
809 "md: bitmaps not supported for this level.\n");
810 goto abort;
811 }
812 }
813
814 if (sb->level == LEVEL_MULTIPATH)
815 rdev->desc_nr = -1;
816 else
817 rdev->desc_nr = sb->this_disk.number;
818
819 if (!refdev) {
820 ret = 1;
821 } else {
822 __u64 ev1, ev2;
823 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
824 if (!uuid_equal(refsb, sb)) {
825 printk(KERN_WARNING "md: %s has different UUID to %s\n",
826 b, bdevname(refdev->bdev,b2));
827 goto abort;
828 }
829 if (!sb_equal(refsb, sb)) {
830 printk(KERN_WARNING "md: %s has same UUID"
831 " but different superblock to %s\n",
832 b, bdevname(refdev->bdev, b2));
833 goto abort;
834 }
835 ev1 = md_event(sb);
836 ev2 = md_event(refsb);
837 if (ev1 > ev2)
838 ret = 1;
839 else
840 ret = 0;
841 }
842 rdev->sectors = calc_num_sectors(rdev, sb->chunk_size);
843
844 if (rdev->sectors < sb->size * 2 && sb->level > 1)
845 /* "this cannot possibly happen" ... */
846 ret = -EINVAL;
847
848 abort:
849 return ret;
850 }
851
852 /*
853 * validate_super for 0.90.0
854 */
855 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
856 {
857 mdp_disk_t *desc;
858 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
859 __u64 ev1 = md_event(sb);
860
861 rdev->raid_disk = -1;
862 clear_bit(Faulty, &rdev->flags);
863 clear_bit(In_sync, &rdev->flags);
864 clear_bit(WriteMostly, &rdev->flags);
865 clear_bit(BarriersNotsupp, &rdev->flags);
866
867 if (mddev->raid_disks == 0) {
868 mddev->major_version = 0;
869 mddev->minor_version = sb->minor_version;
870 mddev->patch_version = sb->patch_version;
871 mddev->external = 0;
872 mddev->chunk_sectors = sb->chunk_size >> 9;
873 mddev->ctime = sb->ctime;
874 mddev->utime = sb->utime;
875 mddev->level = sb->level;
876 mddev->clevel[0] = 0;
877 mddev->layout = sb->layout;
878 mddev->raid_disks = sb->raid_disks;
879 mddev->dev_sectors = sb->size * 2;
880 mddev->events = ev1;
881 mddev->bitmap_offset = 0;
882 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
883
884 if (mddev->minor_version >= 91) {
885 mddev->reshape_position = sb->reshape_position;
886 mddev->delta_disks = sb->delta_disks;
887 mddev->new_level = sb->new_level;
888 mddev->new_layout = sb->new_layout;
889 mddev->new_chunk_sectors = sb->new_chunk >> 9;
890 } else {
891 mddev->reshape_position = MaxSector;
892 mddev->delta_disks = 0;
893 mddev->new_level = mddev->level;
894 mddev->new_layout = mddev->layout;
895 mddev->new_chunk_sectors = mddev->chunk_sectors;
896 }
897
898 if (sb->state & (1<<MD_SB_CLEAN))
899 mddev->recovery_cp = MaxSector;
900 else {
901 if (sb->events_hi == sb->cp_events_hi &&
902 sb->events_lo == sb->cp_events_lo) {
903 mddev->recovery_cp = sb->recovery_cp;
904 } else
905 mddev->recovery_cp = 0;
906 }
907
908 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
909 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
910 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
911 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
912
913 mddev->max_disks = MD_SB_DISKS;
914
915 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
916 mddev->bitmap_file == NULL)
917 mddev->bitmap_offset = mddev->default_bitmap_offset;
918
919 } else if (mddev->pers == NULL) {
920 /* Insist on good event counter while assembling */
921 ++ev1;
922 if (ev1 < mddev->events)
923 return -EINVAL;
924 } else if (mddev->bitmap) {
925 /* if adding to array with a bitmap, then we can accept an
926 * older device ... but not too old.
927 */
928 if (ev1 < mddev->bitmap->events_cleared)
929 return 0;
930 } else {
931 if (ev1 < mddev->events)
932 /* just a hot-add of a new device, leave raid_disk at -1 */
933 return 0;
934 }
935
936 if (mddev->level != LEVEL_MULTIPATH) {
937 desc = sb->disks + rdev->desc_nr;
938
939 if (desc->state & (1<<MD_DISK_FAULTY))
940 set_bit(Faulty, &rdev->flags);
941 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
942 desc->raid_disk < mddev->raid_disks */) {
943 set_bit(In_sync, &rdev->flags);
944 rdev->raid_disk = desc->raid_disk;
945 }
946 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
947 set_bit(WriteMostly, &rdev->flags);
948 } else /* MULTIPATH are always insync */
949 set_bit(In_sync, &rdev->flags);
950 return 0;
951 }
952
953 /*
954 * sync_super for 0.90.0
955 */
956 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
957 {
958 mdp_super_t *sb;
959 mdk_rdev_t *rdev2;
960 int next_spare = mddev->raid_disks;
961
962
963 /* make rdev->sb match mddev data..
964 *
965 * 1/ zero out disks
966 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
967 * 3/ any empty disks < next_spare become removed
968 *
969 * disks[0] gets initialised to REMOVED because
970 * we cannot be sure from other fields if it has
971 * been initialised or not.
972 */
973 int i;
974 int active=0, working=0,failed=0,spare=0,nr_disks=0;
975
976 rdev->sb_size = MD_SB_BYTES;
977
978 sb = (mdp_super_t*)page_address(rdev->sb_page);
979
980 memset(sb, 0, sizeof(*sb));
981
982 sb->md_magic = MD_SB_MAGIC;
983 sb->major_version = mddev->major_version;
984 sb->patch_version = mddev->patch_version;
985 sb->gvalid_words = 0; /* ignored */
986 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
987 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
988 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
989 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
990
991 sb->ctime = mddev->ctime;
992 sb->level = mddev->level;
993 sb->size = mddev->dev_sectors / 2;
994 sb->raid_disks = mddev->raid_disks;
995 sb->md_minor = mddev->md_minor;
996 sb->not_persistent = 0;
997 sb->utime = mddev->utime;
998 sb->state = 0;
999 sb->events_hi = (mddev->events>>32);
1000 sb->events_lo = (u32)mddev->events;
1001
1002 if (mddev->reshape_position == MaxSector)
1003 sb->minor_version = 90;
1004 else {
1005 sb->minor_version = 91;
1006 sb->reshape_position = mddev->reshape_position;
1007 sb->new_level = mddev->new_level;
1008 sb->delta_disks = mddev->delta_disks;
1009 sb->new_layout = mddev->new_layout;
1010 sb->new_chunk = mddev->new_chunk_sectors << 9;
1011 }
1012 mddev->minor_version = sb->minor_version;
1013 if (mddev->in_sync)
1014 {
1015 sb->recovery_cp = mddev->recovery_cp;
1016 sb->cp_events_hi = (mddev->events>>32);
1017 sb->cp_events_lo = (u32)mddev->events;
1018 if (mddev->recovery_cp == MaxSector)
1019 sb->state = (1<< MD_SB_CLEAN);
1020 } else
1021 sb->recovery_cp = 0;
1022
1023 sb->layout = mddev->layout;
1024 sb->chunk_size = mddev->chunk_sectors << 9;
1025
1026 if (mddev->bitmap && mddev->bitmap_file == NULL)
1027 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1028
1029 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1030 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1031 mdp_disk_t *d;
1032 int desc_nr;
1033 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1034 && !test_bit(Faulty, &rdev2->flags))
1035 desc_nr = rdev2->raid_disk;
1036 else
1037 desc_nr = next_spare++;
1038 rdev2->desc_nr = desc_nr;
1039 d = &sb->disks[rdev2->desc_nr];
1040 nr_disks++;
1041 d->number = rdev2->desc_nr;
1042 d->major = MAJOR(rdev2->bdev->bd_dev);
1043 d->minor = MINOR(rdev2->bdev->bd_dev);
1044 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1045 && !test_bit(Faulty, &rdev2->flags))
1046 d->raid_disk = rdev2->raid_disk;
1047 else
1048 d->raid_disk = rdev2->desc_nr; /* compatibility */
1049 if (test_bit(Faulty, &rdev2->flags))
1050 d->state = (1<<MD_DISK_FAULTY);
1051 else if (test_bit(In_sync, &rdev2->flags)) {
1052 d->state = (1<<MD_DISK_ACTIVE);
1053 d->state |= (1<<MD_DISK_SYNC);
1054 active++;
1055 working++;
1056 } else {
1057 d->state = 0;
1058 spare++;
1059 working++;
1060 }
1061 if (test_bit(WriteMostly, &rdev2->flags))
1062 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1063 }
1064 /* now set the "removed" and "faulty" bits on any missing devices */
1065 for (i=0 ; i < mddev->raid_disks ; i++) {
1066 mdp_disk_t *d = &sb->disks[i];
1067 if (d->state == 0 && d->number == 0) {
1068 d->number = i;
1069 d->raid_disk = i;
1070 d->state = (1<<MD_DISK_REMOVED);
1071 d->state |= (1<<MD_DISK_FAULTY);
1072 failed++;
1073 }
1074 }
1075 sb->nr_disks = nr_disks;
1076 sb->active_disks = active;
1077 sb->working_disks = working;
1078 sb->failed_disks = failed;
1079 sb->spare_disks = spare;
1080
1081 sb->this_disk = sb->disks[rdev->desc_nr];
1082 sb->sb_csum = calc_sb_csum(sb);
1083 }
1084
1085 /*
1086 * rdev_size_change for 0.90.0
1087 */
1088 static unsigned long long
1089 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1090 {
1091 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1092 return 0; /* component must fit device */
1093 if (rdev->mddev->bitmap_offset)
1094 return 0; /* can't move bitmap */
1095 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1096 if (!num_sectors || num_sectors > rdev->sb_start)
1097 num_sectors = rdev->sb_start;
1098 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1099 rdev->sb_page);
1100 md_super_wait(rdev->mddev);
1101 return num_sectors / 2; /* kB for sysfs */
1102 }
1103
1104
1105 /*
1106 * version 1 superblock
1107 */
1108
1109 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1110 {
1111 __le32 disk_csum;
1112 u32 csum;
1113 unsigned long long newcsum;
1114 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1115 __le32 *isuper = (__le32*)sb;
1116 int i;
1117
1118 disk_csum = sb->sb_csum;
1119 sb->sb_csum = 0;
1120 newcsum = 0;
1121 for (i=0; size>=4; size -= 4 )
1122 newcsum += le32_to_cpu(*isuper++);
1123
1124 if (size == 2)
1125 newcsum += le16_to_cpu(*(__le16*) isuper);
1126
1127 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1128 sb->sb_csum = disk_csum;
1129 return cpu_to_le32(csum);
1130 }
1131
1132 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1133 {
1134 struct mdp_superblock_1 *sb;
1135 int ret;
1136 sector_t sb_start;
1137 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1138 int bmask;
1139
1140 /*
1141 * Calculate the position of the superblock in 512byte sectors.
1142 * It is always aligned to a 4K boundary and
1143 * depeding on minor_version, it can be:
1144 * 0: At least 8K, but less than 12K, from end of device
1145 * 1: At start of device
1146 * 2: 4K from start of device.
1147 */
1148 switch(minor_version) {
1149 case 0:
1150 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1151 sb_start -= 8*2;
1152 sb_start &= ~(sector_t)(4*2-1);
1153 break;
1154 case 1:
1155 sb_start = 0;
1156 break;
1157 case 2:
1158 sb_start = 8;
1159 break;
1160 default:
1161 return -EINVAL;
1162 }
1163 rdev->sb_start = sb_start;
1164
1165 /* superblock is rarely larger than 1K, but it can be larger,
1166 * and it is safe to read 4k, so we do that
1167 */
1168 ret = read_disk_sb(rdev, 4096);
1169 if (ret) return ret;
1170
1171
1172 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1173
1174 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1175 sb->major_version != cpu_to_le32(1) ||
1176 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1177 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1178 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1179 return -EINVAL;
1180
1181 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1182 printk("md: invalid superblock checksum on %s\n",
1183 bdevname(rdev->bdev,b));
1184 return -EINVAL;
1185 }
1186 if (le64_to_cpu(sb->data_size) < 10) {
1187 printk("md: data_size too small on %s\n",
1188 bdevname(rdev->bdev,b));
1189 return -EINVAL;
1190 }
1191 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1192 if (sb->level != cpu_to_le32(1) &&
1193 sb->level != cpu_to_le32(4) &&
1194 sb->level != cpu_to_le32(5) &&
1195 sb->level != cpu_to_le32(6) &&
1196 sb->level != cpu_to_le32(10)) {
1197 printk(KERN_WARNING
1198 "md: bitmaps not supported for this level.\n");
1199 return -EINVAL;
1200 }
1201 }
1202
1203 rdev->preferred_minor = 0xffff;
1204 rdev->data_offset = le64_to_cpu(sb->data_offset);
1205 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1206
1207 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1208 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1209 if (rdev->sb_size & bmask)
1210 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1211
1212 if (minor_version
1213 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1214 return -EINVAL;
1215
1216 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1217 rdev->desc_nr = -1;
1218 else
1219 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1220
1221 if (!refdev) {
1222 ret = 1;
1223 } else {
1224 __u64 ev1, ev2;
1225 struct mdp_superblock_1 *refsb =
1226 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1227
1228 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1229 sb->level != refsb->level ||
1230 sb->layout != refsb->layout ||
1231 sb->chunksize != refsb->chunksize) {
1232 printk(KERN_WARNING "md: %s has strangely different"
1233 " superblock to %s\n",
1234 bdevname(rdev->bdev,b),
1235 bdevname(refdev->bdev,b2));
1236 return -EINVAL;
1237 }
1238 ev1 = le64_to_cpu(sb->events);
1239 ev2 = le64_to_cpu(refsb->events);
1240
1241 if (ev1 > ev2)
1242 ret = 1;
1243 else
1244 ret = 0;
1245 }
1246 if (minor_version)
1247 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1248 le64_to_cpu(sb->data_offset);
1249 else
1250 rdev->sectors = rdev->sb_start;
1251 if (rdev->sectors < le64_to_cpu(sb->data_size))
1252 return -EINVAL;
1253 rdev->sectors = le64_to_cpu(sb->data_size);
1254 if (le32_to_cpu(sb->chunksize)) {
1255 int chunk_sects = le32_to_cpu(sb->chunksize);
1256 sector_t chunks = rdev->sectors;
1257 sector_div(chunks, chunk_sects);
1258 rdev->sectors = chunks * chunk_sects;
1259 }
1260
1261 if (le64_to_cpu(sb->size) > rdev->sectors)
1262 return -EINVAL;
1263 return ret;
1264 }
1265
1266 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1267 {
1268 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1269 __u64 ev1 = le64_to_cpu(sb->events);
1270
1271 rdev->raid_disk = -1;
1272 clear_bit(Faulty, &rdev->flags);
1273 clear_bit(In_sync, &rdev->flags);
1274 clear_bit(WriteMostly, &rdev->flags);
1275 clear_bit(BarriersNotsupp, &rdev->flags);
1276
1277 if (mddev->raid_disks == 0) {
1278 mddev->major_version = 1;
1279 mddev->patch_version = 0;
1280 mddev->external = 0;
1281 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1282 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1283 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1284 mddev->level = le32_to_cpu(sb->level);
1285 mddev->clevel[0] = 0;
1286 mddev->layout = le32_to_cpu(sb->layout);
1287 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1288 mddev->dev_sectors = le64_to_cpu(sb->size);
1289 mddev->events = ev1;
1290 mddev->bitmap_offset = 0;
1291 mddev->default_bitmap_offset = 1024 >> 9;
1292
1293 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1294 memcpy(mddev->uuid, sb->set_uuid, 16);
1295
1296 mddev->max_disks = (4096-256)/2;
1297
1298 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1299 mddev->bitmap_file == NULL )
1300 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1301
1302 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1303 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1304 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1305 mddev->new_level = le32_to_cpu(sb->new_level);
1306 mddev->new_layout = le32_to_cpu(sb->new_layout);
1307 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1308 } else {
1309 mddev->reshape_position = MaxSector;
1310 mddev->delta_disks = 0;
1311 mddev->new_level = mddev->level;
1312 mddev->new_layout = mddev->layout;
1313 mddev->new_chunk_sectors = mddev->chunk_sectors;
1314 }
1315
1316 } else if (mddev->pers == NULL) {
1317 /* Insist of good event counter while assembling */
1318 ++ev1;
1319 if (ev1 < mddev->events)
1320 return -EINVAL;
1321 } else if (mddev->bitmap) {
1322 /* If adding to array with a bitmap, then we can accept an
1323 * older device, but not too old.
1324 */
1325 if (ev1 < mddev->bitmap->events_cleared)
1326 return 0;
1327 } else {
1328 if (ev1 < mddev->events)
1329 /* just a hot-add of a new device, leave raid_disk at -1 */
1330 return 0;
1331 }
1332 if (mddev->level != LEVEL_MULTIPATH) {
1333 int role;
1334 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1335 switch(role) {
1336 case 0xffff: /* spare */
1337 break;
1338 case 0xfffe: /* faulty */
1339 set_bit(Faulty, &rdev->flags);
1340 break;
1341 default:
1342 if ((le32_to_cpu(sb->feature_map) &
1343 MD_FEATURE_RECOVERY_OFFSET))
1344 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1345 else
1346 set_bit(In_sync, &rdev->flags);
1347 rdev->raid_disk = role;
1348 break;
1349 }
1350 if (sb->devflags & WriteMostly1)
1351 set_bit(WriteMostly, &rdev->flags);
1352 } else /* MULTIPATH are always insync */
1353 set_bit(In_sync, &rdev->flags);
1354
1355 return 0;
1356 }
1357
1358 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1359 {
1360 struct mdp_superblock_1 *sb;
1361 mdk_rdev_t *rdev2;
1362 int max_dev, i;
1363 /* make rdev->sb match mddev and rdev data. */
1364
1365 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1366
1367 sb->feature_map = 0;
1368 sb->pad0 = 0;
1369 sb->recovery_offset = cpu_to_le64(0);
1370 memset(sb->pad1, 0, sizeof(sb->pad1));
1371 memset(sb->pad2, 0, sizeof(sb->pad2));
1372 memset(sb->pad3, 0, sizeof(sb->pad3));
1373
1374 sb->utime = cpu_to_le64((__u64)mddev->utime);
1375 sb->events = cpu_to_le64(mddev->events);
1376 if (mddev->in_sync)
1377 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1378 else
1379 sb->resync_offset = cpu_to_le64(0);
1380
1381 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1382
1383 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1384 sb->size = cpu_to_le64(mddev->dev_sectors);
1385 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1386 sb->level = cpu_to_le32(mddev->level);
1387 sb->layout = cpu_to_le32(mddev->layout);
1388
1389 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1390 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1391 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1392 }
1393
1394 if (rdev->raid_disk >= 0 &&
1395 !test_bit(In_sync, &rdev->flags)) {
1396 if (mddev->curr_resync_completed > rdev->recovery_offset)
1397 rdev->recovery_offset = mddev->curr_resync_completed;
1398 if (rdev->recovery_offset > 0) {
1399 sb->feature_map |=
1400 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1401 sb->recovery_offset =
1402 cpu_to_le64(rdev->recovery_offset);
1403 }
1404 }
1405
1406 if (mddev->reshape_position != MaxSector) {
1407 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1408 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1409 sb->new_layout = cpu_to_le32(mddev->new_layout);
1410 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1411 sb->new_level = cpu_to_le32(mddev->new_level);
1412 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1413 }
1414
1415 max_dev = 0;
1416 list_for_each_entry(rdev2, &mddev->disks, same_set)
1417 if (rdev2->desc_nr+1 > max_dev)
1418 max_dev = rdev2->desc_nr+1;
1419
1420 if (max_dev > le32_to_cpu(sb->max_dev))
1421 sb->max_dev = cpu_to_le32(max_dev);
1422 for (i=0; i<max_dev;i++)
1423 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1424
1425 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1426 i = rdev2->desc_nr;
1427 if (test_bit(Faulty, &rdev2->flags))
1428 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1429 else if (test_bit(In_sync, &rdev2->flags))
1430 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1431 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1432 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1433 else
1434 sb->dev_roles[i] = cpu_to_le16(0xffff);
1435 }
1436
1437 sb->sb_csum = calc_sb_1_csum(sb);
1438 }
1439
1440 static unsigned long long
1441 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1442 {
1443 struct mdp_superblock_1 *sb;
1444 sector_t max_sectors;
1445 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1446 return 0; /* component must fit device */
1447 if (rdev->sb_start < rdev->data_offset) {
1448 /* minor versions 1 and 2; superblock before data */
1449 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1450 max_sectors -= rdev->data_offset;
1451 if (!num_sectors || num_sectors > max_sectors)
1452 num_sectors = max_sectors;
1453 } else if (rdev->mddev->bitmap_offset) {
1454 /* minor version 0 with bitmap we can't move */
1455 return 0;
1456 } else {
1457 /* minor version 0; superblock after data */
1458 sector_t sb_start;
1459 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1460 sb_start &= ~(sector_t)(4*2 - 1);
1461 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1462 if (!num_sectors || num_sectors > max_sectors)
1463 num_sectors = max_sectors;
1464 rdev->sb_start = sb_start;
1465 }
1466 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1467 sb->data_size = cpu_to_le64(num_sectors);
1468 sb->super_offset = rdev->sb_start;
1469 sb->sb_csum = calc_sb_1_csum(sb);
1470 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1471 rdev->sb_page);
1472 md_super_wait(rdev->mddev);
1473 return num_sectors / 2; /* kB for sysfs */
1474 }
1475
1476 static struct super_type super_types[] = {
1477 [0] = {
1478 .name = "0.90.0",
1479 .owner = THIS_MODULE,
1480 .load_super = super_90_load,
1481 .validate_super = super_90_validate,
1482 .sync_super = super_90_sync,
1483 .rdev_size_change = super_90_rdev_size_change,
1484 },
1485 [1] = {
1486 .name = "md-1",
1487 .owner = THIS_MODULE,
1488 .load_super = super_1_load,
1489 .validate_super = super_1_validate,
1490 .sync_super = super_1_sync,
1491 .rdev_size_change = super_1_rdev_size_change,
1492 },
1493 };
1494
1495 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1496 {
1497 mdk_rdev_t *rdev, *rdev2;
1498
1499 rcu_read_lock();
1500 rdev_for_each_rcu(rdev, mddev1)
1501 rdev_for_each_rcu(rdev2, mddev2)
1502 if (rdev->bdev->bd_contains ==
1503 rdev2->bdev->bd_contains) {
1504 rcu_read_unlock();
1505 return 1;
1506 }
1507 rcu_read_unlock();
1508 return 0;
1509 }
1510
1511 static LIST_HEAD(pending_raid_disks);
1512
1513 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1514 {
1515 struct mdk_personality *pers = mddev->pers;
1516 struct gendisk *disk = mddev->gendisk;
1517 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1518 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1519
1520 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1521 if (pers && pers->level >= 4 && pers->level <= 6)
1522 return;
1523
1524 /* If rdev is integrity capable, register profile for mddev */
1525 if (!bi_mddev && bi_rdev) {
1526 if (blk_integrity_register(disk, bi_rdev))
1527 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1528 __func__, disk->disk_name);
1529 else
1530 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1531 disk->disk_name);
1532 return;
1533 }
1534
1535 /* Check that mddev and rdev have matching profiles */
1536 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1537 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1538 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1539 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1540 disk->disk_name);
1541 blk_integrity_unregister(disk);
1542 }
1543 }
1544
1545 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1546 {
1547 char b[BDEVNAME_SIZE];
1548 struct kobject *ko;
1549 char *s;
1550 int err;
1551
1552 if (rdev->mddev) {
1553 MD_BUG();
1554 return -EINVAL;
1555 }
1556
1557 /* prevent duplicates */
1558 if (find_rdev(mddev, rdev->bdev->bd_dev))
1559 return -EEXIST;
1560
1561 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1562 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1563 rdev->sectors < mddev->dev_sectors)) {
1564 if (mddev->pers) {
1565 /* Cannot change size, so fail
1566 * If mddev->level <= 0, then we don't care
1567 * about aligning sizes (e.g. linear)
1568 */
1569 if (mddev->level > 0)
1570 return -ENOSPC;
1571 } else
1572 mddev->dev_sectors = rdev->sectors;
1573 }
1574
1575 /* Verify rdev->desc_nr is unique.
1576 * If it is -1, assign a free number, else
1577 * check number is not in use
1578 */
1579 if (rdev->desc_nr < 0) {
1580 int choice = 0;
1581 if (mddev->pers) choice = mddev->raid_disks;
1582 while (find_rdev_nr(mddev, choice))
1583 choice++;
1584 rdev->desc_nr = choice;
1585 } else {
1586 if (find_rdev_nr(mddev, rdev->desc_nr))
1587 return -EBUSY;
1588 }
1589 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1590 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1591 mdname(mddev), mddev->max_disks);
1592 return -EBUSY;
1593 }
1594 bdevname(rdev->bdev,b);
1595 while ( (s=strchr(b, '/')) != NULL)
1596 *s = '!';
1597
1598 rdev->mddev = mddev;
1599 printk(KERN_INFO "md: bind<%s>\n", b);
1600
1601 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1602 goto fail;
1603
1604 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1605 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1606 kobject_del(&rdev->kobj);
1607 goto fail;
1608 }
1609 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1610
1611 list_add_rcu(&rdev->same_set, &mddev->disks);
1612 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1613
1614 /* May as well allow recovery to be retried once */
1615 mddev->recovery_disabled = 0;
1616
1617 md_integrity_check(rdev, mddev);
1618 return 0;
1619
1620 fail:
1621 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1622 b, mdname(mddev));
1623 return err;
1624 }
1625
1626 static void md_delayed_delete(struct work_struct *ws)
1627 {
1628 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1629 kobject_del(&rdev->kobj);
1630 kobject_put(&rdev->kobj);
1631 }
1632
1633 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1634 {
1635 char b[BDEVNAME_SIZE];
1636 if (!rdev->mddev) {
1637 MD_BUG();
1638 return;
1639 }
1640 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1641 list_del_rcu(&rdev->same_set);
1642 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1643 rdev->mddev = NULL;
1644 sysfs_remove_link(&rdev->kobj, "block");
1645 sysfs_put(rdev->sysfs_state);
1646 rdev->sysfs_state = NULL;
1647 /* We need to delay this, otherwise we can deadlock when
1648 * writing to 'remove' to "dev/state". We also need
1649 * to delay it due to rcu usage.
1650 */
1651 synchronize_rcu();
1652 INIT_WORK(&rdev->del_work, md_delayed_delete);
1653 kobject_get(&rdev->kobj);
1654 schedule_work(&rdev->del_work);
1655 }
1656
1657 /*
1658 * prevent the device from being mounted, repartitioned or
1659 * otherwise reused by a RAID array (or any other kernel
1660 * subsystem), by bd_claiming the device.
1661 */
1662 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1663 {
1664 int err = 0;
1665 struct block_device *bdev;
1666 char b[BDEVNAME_SIZE];
1667
1668 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1669 if (IS_ERR(bdev)) {
1670 printk(KERN_ERR "md: could not open %s.\n",
1671 __bdevname(dev, b));
1672 return PTR_ERR(bdev);
1673 }
1674 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1675 if (err) {
1676 printk(KERN_ERR "md: could not bd_claim %s.\n",
1677 bdevname(bdev, b));
1678 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1679 return err;
1680 }
1681 if (!shared)
1682 set_bit(AllReserved, &rdev->flags);
1683 rdev->bdev = bdev;
1684 return err;
1685 }
1686
1687 static void unlock_rdev(mdk_rdev_t *rdev)
1688 {
1689 struct block_device *bdev = rdev->bdev;
1690 rdev->bdev = NULL;
1691 if (!bdev)
1692 MD_BUG();
1693 bd_release(bdev);
1694 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1695 }
1696
1697 void md_autodetect_dev(dev_t dev);
1698
1699 static void export_rdev(mdk_rdev_t * rdev)
1700 {
1701 char b[BDEVNAME_SIZE];
1702 printk(KERN_INFO "md: export_rdev(%s)\n",
1703 bdevname(rdev->bdev,b));
1704 if (rdev->mddev)
1705 MD_BUG();
1706 free_disk_sb(rdev);
1707 #ifndef MODULE
1708 if (test_bit(AutoDetected, &rdev->flags))
1709 md_autodetect_dev(rdev->bdev->bd_dev);
1710 #endif
1711 unlock_rdev(rdev);
1712 kobject_put(&rdev->kobj);
1713 }
1714
1715 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1716 {
1717 unbind_rdev_from_array(rdev);
1718 export_rdev(rdev);
1719 }
1720
1721 static void export_array(mddev_t *mddev)
1722 {
1723 mdk_rdev_t *rdev, *tmp;
1724
1725 rdev_for_each(rdev, tmp, mddev) {
1726 if (!rdev->mddev) {
1727 MD_BUG();
1728 continue;
1729 }
1730 kick_rdev_from_array(rdev);
1731 }
1732 if (!list_empty(&mddev->disks))
1733 MD_BUG();
1734 mddev->raid_disks = 0;
1735 mddev->major_version = 0;
1736 }
1737
1738 static void print_desc(mdp_disk_t *desc)
1739 {
1740 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1741 desc->major,desc->minor,desc->raid_disk,desc->state);
1742 }
1743
1744 static void print_sb_90(mdp_super_t *sb)
1745 {
1746 int i;
1747
1748 printk(KERN_INFO
1749 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1750 sb->major_version, sb->minor_version, sb->patch_version,
1751 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1752 sb->ctime);
1753 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1754 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1755 sb->md_minor, sb->layout, sb->chunk_size);
1756 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1757 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1758 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1759 sb->failed_disks, sb->spare_disks,
1760 sb->sb_csum, (unsigned long)sb->events_lo);
1761
1762 printk(KERN_INFO);
1763 for (i = 0; i < MD_SB_DISKS; i++) {
1764 mdp_disk_t *desc;
1765
1766 desc = sb->disks + i;
1767 if (desc->number || desc->major || desc->minor ||
1768 desc->raid_disk || (desc->state && (desc->state != 4))) {
1769 printk(" D %2d: ", i);
1770 print_desc(desc);
1771 }
1772 }
1773 printk(KERN_INFO "md: THIS: ");
1774 print_desc(&sb->this_disk);
1775 }
1776
1777 static void print_sb_1(struct mdp_superblock_1 *sb)
1778 {
1779 __u8 *uuid;
1780
1781 uuid = sb->set_uuid;
1782 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1783 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1784 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1785 le32_to_cpu(sb->major_version),
1786 le32_to_cpu(sb->feature_map),
1787 uuid[0], uuid[1], uuid[2], uuid[3],
1788 uuid[4], uuid[5], uuid[6], uuid[7],
1789 uuid[8], uuid[9], uuid[10], uuid[11],
1790 uuid[12], uuid[13], uuid[14], uuid[15],
1791 sb->set_name,
1792 (unsigned long long)le64_to_cpu(sb->ctime)
1793 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1794
1795 uuid = sb->device_uuid;
1796 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1797 " RO:%llu\n"
1798 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1799 ":%02x%02x%02x%02x%02x%02x\n"
1800 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1801 KERN_INFO "md: (MaxDev:%u) \n",
1802 le32_to_cpu(sb->level),
1803 (unsigned long long)le64_to_cpu(sb->size),
1804 le32_to_cpu(sb->raid_disks),
1805 le32_to_cpu(sb->layout),
1806 le32_to_cpu(sb->chunksize),
1807 (unsigned long long)le64_to_cpu(sb->data_offset),
1808 (unsigned long long)le64_to_cpu(sb->data_size),
1809 (unsigned long long)le64_to_cpu(sb->super_offset),
1810 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1811 le32_to_cpu(sb->dev_number),
1812 uuid[0], uuid[1], uuid[2], uuid[3],
1813 uuid[4], uuid[5], uuid[6], uuid[7],
1814 uuid[8], uuid[9], uuid[10], uuid[11],
1815 uuid[12], uuid[13], uuid[14], uuid[15],
1816 sb->devflags,
1817 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1818 (unsigned long long)le64_to_cpu(sb->events),
1819 (unsigned long long)le64_to_cpu(sb->resync_offset),
1820 le32_to_cpu(sb->sb_csum),
1821 le32_to_cpu(sb->max_dev)
1822 );
1823 }
1824
1825 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1826 {
1827 char b[BDEVNAME_SIZE];
1828 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1829 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1830 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1831 rdev->desc_nr);
1832 if (rdev->sb_loaded) {
1833 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1834 switch (major_version) {
1835 case 0:
1836 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1837 break;
1838 case 1:
1839 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1840 break;
1841 }
1842 } else
1843 printk(KERN_INFO "md: no rdev superblock!\n");
1844 }
1845
1846 static void md_print_devices(void)
1847 {
1848 struct list_head *tmp;
1849 mdk_rdev_t *rdev;
1850 mddev_t *mddev;
1851 char b[BDEVNAME_SIZE];
1852
1853 printk("\n");
1854 printk("md: **********************************\n");
1855 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1856 printk("md: **********************************\n");
1857 for_each_mddev(mddev, tmp) {
1858
1859 if (mddev->bitmap)
1860 bitmap_print_sb(mddev->bitmap);
1861 else
1862 printk("%s: ", mdname(mddev));
1863 list_for_each_entry(rdev, &mddev->disks, same_set)
1864 printk("<%s>", bdevname(rdev->bdev,b));
1865 printk("\n");
1866
1867 list_for_each_entry(rdev, &mddev->disks, same_set)
1868 print_rdev(rdev, mddev->major_version);
1869 }
1870 printk("md: **********************************\n");
1871 printk("\n");
1872 }
1873
1874
1875 static void sync_sbs(mddev_t * mddev, int nospares)
1876 {
1877 /* Update each superblock (in-memory image), but
1878 * if we are allowed to, skip spares which already
1879 * have the right event counter, or have one earlier
1880 * (which would mean they aren't being marked as dirty
1881 * with the rest of the array)
1882 */
1883 mdk_rdev_t *rdev;
1884
1885 list_for_each_entry(rdev, &mddev->disks, same_set) {
1886 if (rdev->sb_events == mddev->events ||
1887 (nospares &&
1888 rdev->raid_disk < 0 &&
1889 (rdev->sb_events&1)==0 &&
1890 rdev->sb_events+1 == mddev->events)) {
1891 /* Don't update this superblock */
1892 rdev->sb_loaded = 2;
1893 } else {
1894 super_types[mddev->major_version].
1895 sync_super(mddev, rdev);
1896 rdev->sb_loaded = 1;
1897 }
1898 }
1899 }
1900
1901 static void md_update_sb(mddev_t * mddev, int force_change)
1902 {
1903 mdk_rdev_t *rdev;
1904 int sync_req;
1905 int nospares = 0;
1906
1907 mddev->utime = get_seconds();
1908 if (mddev->external)
1909 return;
1910 repeat:
1911 spin_lock_irq(&mddev->write_lock);
1912
1913 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1914 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1915 force_change = 1;
1916 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1917 /* just a clean<-> dirty transition, possibly leave spares alone,
1918 * though if events isn't the right even/odd, we will have to do
1919 * spares after all
1920 */
1921 nospares = 1;
1922 if (force_change)
1923 nospares = 0;
1924 if (mddev->degraded)
1925 /* If the array is degraded, then skipping spares is both
1926 * dangerous and fairly pointless.
1927 * Dangerous because a device that was removed from the array
1928 * might have a event_count that still looks up-to-date,
1929 * so it can be re-added without a resync.
1930 * Pointless because if there are any spares to skip,
1931 * then a recovery will happen and soon that array won't
1932 * be degraded any more and the spare can go back to sleep then.
1933 */
1934 nospares = 0;
1935
1936 sync_req = mddev->in_sync;
1937
1938 /* If this is just a dirty<->clean transition, and the array is clean
1939 * and 'events' is odd, we can roll back to the previous clean state */
1940 if (nospares
1941 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1942 && (mddev->events & 1)
1943 && mddev->events != 1)
1944 mddev->events--;
1945 else {
1946 /* otherwise we have to go forward and ... */
1947 mddev->events ++;
1948 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1949 /* .. if the array isn't clean, insist on an odd 'events' */
1950 if ((mddev->events&1)==0) {
1951 mddev->events++;
1952 nospares = 0;
1953 }
1954 } else {
1955 /* otherwise insist on an even 'events' (for clean states) */
1956 if ((mddev->events&1)) {
1957 mddev->events++;
1958 nospares = 0;
1959 }
1960 }
1961 }
1962
1963 if (!mddev->events) {
1964 /*
1965 * oops, this 64-bit counter should never wrap.
1966 * Either we are in around ~1 trillion A.C., assuming
1967 * 1 reboot per second, or we have a bug:
1968 */
1969 MD_BUG();
1970 mddev->events --;
1971 }
1972
1973 /*
1974 * do not write anything to disk if using
1975 * nonpersistent superblocks
1976 */
1977 if (!mddev->persistent) {
1978 if (!mddev->external)
1979 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1980
1981 spin_unlock_irq(&mddev->write_lock);
1982 wake_up(&mddev->sb_wait);
1983 return;
1984 }
1985 sync_sbs(mddev, nospares);
1986 spin_unlock_irq(&mddev->write_lock);
1987
1988 dprintk(KERN_INFO
1989 "md: updating %s RAID superblock on device (in sync %d)\n",
1990 mdname(mddev),mddev->in_sync);
1991
1992 bitmap_update_sb(mddev->bitmap);
1993 list_for_each_entry(rdev, &mddev->disks, same_set) {
1994 char b[BDEVNAME_SIZE];
1995 dprintk(KERN_INFO "md: ");
1996 if (rdev->sb_loaded != 1)
1997 continue; /* no noise on spare devices */
1998 if (test_bit(Faulty, &rdev->flags))
1999 dprintk("(skipping faulty ");
2000
2001 dprintk("%s ", bdevname(rdev->bdev,b));
2002 if (!test_bit(Faulty, &rdev->flags)) {
2003 md_super_write(mddev,rdev,
2004 rdev->sb_start, rdev->sb_size,
2005 rdev->sb_page);
2006 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2007 bdevname(rdev->bdev,b),
2008 (unsigned long long)rdev->sb_start);
2009 rdev->sb_events = mddev->events;
2010
2011 } else
2012 dprintk(")\n");
2013 if (mddev->level == LEVEL_MULTIPATH)
2014 /* only need to write one superblock... */
2015 break;
2016 }
2017 md_super_wait(mddev);
2018 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2019
2020 spin_lock_irq(&mddev->write_lock);
2021 if (mddev->in_sync != sync_req ||
2022 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2023 /* have to write it out again */
2024 spin_unlock_irq(&mddev->write_lock);
2025 goto repeat;
2026 }
2027 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2028 spin_unlock_irq(&mddev->write_lock);
2029 wake_up(&mddev->sb_wait);
2030 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2031 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2032
2033 }
2034
2035 /* words written to sysfs files may, or may not, be \n terminated.
2036 * We want to accept with case. For this we use cmd_match.
2037 */
2038 static int cmd_match(const char *cmd, const char *str)
2039 {
2040 /* See if cmd, written into a sysfs file, matches
2041 * str. They must either be the same, or cmd can
2042 * have a trailing newline
2043 */
2044 while (*cmd && *str && *cmd == *str) {
2045 cmd++;
2046 str++;
2047 }
2048 if (*cmd == '\n')
2049 cmd++;
2050 if (*str || *cmd)
2051 return 0;
2052 return 1;
2053 }
2054
2055 struct rdev_sysfs_entry {
2056 struct attribute attr;
2057 ssize_t (*show)(mdk_rdev_t *, char *);
2058 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2059 };
2060
2061 static ssize_t
2062 state_show(mdk_rdev_t *rdev, char *page)
2063 {
2064 char *sep = "";
2065 size_t len = 0;
2066
2067 if (test_bit(Faulty, &rdev->flags)) {
2068 len+= sprintf(page+len, "%sfaulty",sep);
2069 sep = ",";
2070 }
2071 if (test_bit(In_sync, &rdev->flags)) {
2072 len += sprintf(page+len, "%sin_sync",sep);
2073 sep = ",";
2074 }
2075 if (test_bit(WriteMostly, &rdev->flags)) {
2076 len += sprintf(page+len, "%swrite_mostly",sep);
2077 sep = ",";
2078 }
2079 if (test_bit(Blocked, &rdev->flags)) {
2080 len += sprintf(page+len, "%sblocked", sep);
2081 sep = ",";
2082 }
2083 if (!test_bit(Faulty, &rdev->flags) &&
2084 !test_bit(In_sync, &rdev->flags)) {
2085 len += sprintf(page+len, "%sspare", sep);
2086 sep = ",";
2087 }
2088 return len+sprintf(page+len, "\n");
2089 }
2090
2091 static ssize_t
2092 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2093 {
2094 /* can write
2095 * faulty - simulates and error
2096 * remove - disconnects the device
2097 * writemostly - sets write_mostly
2098 * -writemostly - clears write_mostly
2099 * blocked - sets the Blocked flag
2100 * -blocked - clears the Blocked flag
2101 * insync - sets Insync providing device isn't active
2102 */
2103 int err = -EINVAL;
2104 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2105 md_error(rdev->mddev, rdev);
2106 err = 0;
2107 } else if (cmd_match(buf, "remove")) {
2108 if (rdev->raid_disk >= 0)
2109 err = -EBUSY;
2110 else {
2111 mddev_t *mddev = rdev->mddev;
2112 kick_rdev_from_array(rdev);
2113 if (mddev->pers)
2114 md_update_sb(mddev, 1);
2115 md_new_event(mddev);
2116 err = 0;
2117 }
2118 } else if (cmd_match(buf, "writemostly")) {
2119 set_bit(WriteMostly, &rdev->flags);
2120 err = 0;
2121 } else if (cmd_match(buf, "-writemostly")) {
2122 clear_bit(WriteMostly, &rdev->flags);
2123 err = 0;
2124 } else if (cmd_match(buf, "blocked")) {
2125 set_bit(Blocked, &rdev->flags);
2126 err = 0;
2127 } else if (cmd_match(buf, "-blocked")) {
2128 clear_bit(Blocked, &rdev->flags);
2129 wake_up(&rdev->blocked_wait);
2130 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2131 md_wakeup_thread(rdev->mddev->thread);
2132
2133 err = 0;
2134 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2135 set_bit(In_sync, &rdev->flags);
2136 err = 0;
2137 }
2138 if (!err && rdev->sysfs_state)
2139 sysfs_notify_dirent(rdev->sysfs_state);
2140 return err ? err : len;
2141 }
2142 static struct rdev_sysfs_entry rdev_state =
2143 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2144
2145 static ssize_t
2146 errors_show(mdk_rdev_t *rdev, char *page)
2147 {
2148 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2149 }
2150
2151 static ssize_t
2152 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2153 {
2154 char *e;
2155 unsigned long n = simple_strtoul(buf, &e, 10);
2156 if (*buf && (*e == 0 || *e == '\n')) {
2157 atomic_set(&rdev->corrected_errors, n);
2158 return len;
2159 }
2160 return -EINVAL;
2161 }
2162 static struct rdev_sysfs_entry rdev_errors =
2163 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2164
2165 static ssize_t
2166 slot_show(mdk_rdev_t *rdev, char *page)
2167 {
2168 if (rdev->raid_disk < 0)
2169 return sprintf(page, "none\n");
2170 else
2171 return sprintf(page, "%d\n", rdev->raid_disk);
2172 }
2173
2174 static ssize_t
2175 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2176 {
2177 char *e;
2178 int err;
2179 char nm[20];
2180 int slot = simple_strtoul(buf, &e, 10);
2181 if (strncmp(buf, "none", 4)==0)
2182 slot = -1;
2183 else if (e==buf || (*e && *e!= '\n'))
2184 return -EINVAL;
2185 if (rdev->mddev->pers && slot == -1) {
2186 /* Setting 'slot' on an active array requires also
2187 * updating the 'rd%d' link, and communicating
2188 * with the personality with ->hot_*_disk.
2189 * For now we only support removing
2190 * failed/spare devices. This normally happens automatically,
2191 * but not when the metadata is externally managed.
2192 */
2193 if (rdev->raid_disk == -1)
2194 return -EEXIST;
2195 /* personality does all needed checks */
2196 if (rdev->mddev->pers->hot_add_disk == NULL)
2197 return -EINVAL;
2198 err = rdev->mddev->pers->
2199 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2200 if (err)
2201 return err;
2202 sprintf(nm, "rd%d", rdev->raid_disk);
2203 sysfs_remove_link(&rdev->mddev->kobj, nm);
2204 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2205 md_wakeup_thread(rdev->mddev->thread);
2206 } else if (rdev->mddev->pers) {
2207 mdk_rdev_t *rdev2;
2208 /* Activating a spare .. or possibly reactivating
2209 * if we ever get bitmaps working here.
2210 */
2211
2212 if (rdev->raid_disk != -1)
2213 return -EBUSY;
2214
2215 if (rdev->mddev->pers->hot_add_disk == NULL)
2216 return -EINVAL;
2217
2218 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2219 if (rdev2->raid_disk == slot)
2220 return -EEXIST;
2221
2222 rdev->raid_disk = slot;
2223 if (test_bit(In_sync, &rdev->flags))
2224 rdev->saved_raid_disk = slot;
2225 else
2226 rdev->saved_raid_disk = -1;
2227 err = rdev->mddev->pers->
2228 hot_add_disk(rdev->mddev, rdev);
2229 if (err) {
2230 rdev->raid_disk = -1;
2231 return err;
2232 } else
2233 sysfs_notify_dirent(rdev->sysfs_state);
2234 sprintf(nm, "rd%d", rdev->raid_disk);
2235 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2236 printk(KERN_WARNING
2237 "md: cannot register "
2238 "%s for %s\n",
2239 nm, mdname(rdev->mddev));
2240
2241 /* don't wakeup anyone, leave that to userspace. */
2242 } else {
2243 if (slot >= rdev->mddev->raid_disks)
2244 return -ENOSPC;
2245 rdev->raid_disk = slot;
2246 /* assume it is working */
2247 clear_bit(Faulty, &rdev->flags);
2248 clear_bit(WriteMostly, &rdev->flags);
2249 set_bit(In_sync, &rdev->flags);
2250 sysfs_notify_dirent(rdev->sysfs_state);
2251 }
2252 return len;
2253 }
2254
2255
2256 static struct rdev_sysfs_entry rdev_slot =
2257 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2258
2259 static ssize_t
2260 offset_show(mdk_rdev_t *rdev, char *page)
2261 {
2262 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2263 }
2264
2265 static ssize_t
2266 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2267 {
2268 char *e;
2269 unsigned long long offset = simple_strtoull(buf, &e, 10);
2270 if (e==buf || (*e && *e != '\n'))
2271 return -EINVAL;
2272 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2273 return -EBUSY;
2274 if (rdev->sectors && rdev->mddev->external)
2275 /* Must set offset before size, so overlap checks
2276 * can be sane */
2277 return -EBUSY;
2278 rdev->data_offset = offset;
2279 return len;
2280 }
2281
2282 static struct rdev_sysfs_entry rdev_offset =
2283 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2284
2285 static ssize_t
2286 rdev_size_show(mdk_rdev_t *rdev, char *page)
2287 {
2288 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2289 }
2290
2291 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2292 {
2293 /* check if two start/length pairs overlap */
2294 if (s1+l1 <= s2)
2295 return 0;
2296 if (s2+l2 <= s1)
2297 return 0;
2298 return 1;
2299 }
2300
2301 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2302 {
2303 unsigned long long blocks;
2304 sector_t new;
2305
2306 if (strict_strtoull(buf, 10, &blocks) < 0)
2307 return -EINVAL;
2308
2309 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2310 return -EINVAL; /* sector conversion overflow */
2311
2312 new = blocks * 2;
2313 if (new != blocks * 2)
2314 return -EINVAL; /* unsigned long long to sector_t overflow */
2315
2316 *sectors = new;
2317 return 0;
2318 }
2319
2320 static ssize_t
2321 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2322 {
2323 mddev_t *my_mddev = rdev->mddev;
2324 sector_t oldsectors = rdev->sectors;
2325 sector_t sectors;
2326
2327 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2328 return -EINVAL;
2329 if (my_mddev->pers && rdev->raid_disk >= 0) {
2330 if (my_mddev->persistent) {
2331 sectors = super_types[my_mddev->major_version].
2332 rdev_size_change(rdev, sectors);
2333 if (!sectors)
2334 return -EBUSY;
2335 } else if (!sectors)
2336 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2337 rdev->data_offset;
2338 }
2339 if (sectors < my_mddev->dev_sectors)
2340 return -EINVAL; /* component must fit device */
2341
2342 rdev->sectors = sectors;
2343 if (sectors > oldsectors && my_mddev->external) {
2344 /* need to check that all other rdevs with the same ->bdev
2345 * do not overlap. We need to unlock the mddev to avoid
2346 * a deadlock. We have already changed rdev->sectors, and if
2347 * we have to change it back, we will have the lock again.
2348 */
2349 mddev_t *mddev;
2350 int overlap = 0;
2351 struct list_head *tmp;
2352
2353 mddev_unlock(my_mddev);
2354 for_each_mddev(mddev, tmp) {
2355 mdk_rdev_t *rdev2;
2356
2357 mddev_lock(mddev);
2358 list_for_each_entry(rdev2, &mddev->disks, same_set)
2359 if (test_bit(AllReserved, &rdev2->flags) ||
2360 (rdev->bdev == rdev2->bdev &&
2361 rdev != rdev2 &&
2362 overlaps(rdev->data_offset, rdev->sectors,
2363 rdev2->data_offset,
2364 rdev2->sectors))) {
2365 overlap = 1;
2366 break;
2367 }
2368 mddev_unlock(mddev);
2369 if (overlap) {
2370 mddev_put(mddev);
2371 break;
2372 }
2373 }
2374 mddev_lock(my_mddev);
2375 if (overlap) {
2376 /* Someone else could have slipped in a size
2377 * change here, but doing so is just silly.
2378 * We put oldsectors back because we *know* it is
2379 * safe, and trust userspace not to race with
2380 * itself
2381 */
2382 rdev->sectors = oldsectors;
2383 return -EBUSY;
2384 }
2385 }
2386 return len;
2387 }
2388
2389 static struct rdev_sysfs_entry rdev_size =
2390 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2391
2392 static struct attribute *rdev_default_attrs[] = {
2393 &rdev_state.attr,
2394 &rdev_errors.attr,
2395 &rdev_slot.attr,
2396 &rdev_offset.attr,
2397 &rdev_size.attr,
2398 NULL,
2399 };
2400 static ssize_t
2401 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2402 {
2403 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2404 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2405 mddev_t *mddev = rdev->mddev;
2406 ssize_t rv;
2407
2408 if (!entry->show)
2409 return -EIO;
2410
2411 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2412 if (!rv) {
2413 if (rdev->mddev == NULL)
2414 rv = -EBUSY;
2415 else
2416 rv = entry->show(rdev, page);
2417 mddev_unlock(mddev);
2418 }
2419 return rv;
2420 }
2421
2422 static ssize_t
2423 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2424 const char *page, size_t length)
2425 {
2426 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2427 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2428 ssize_t rv;
2429 mddev_t *mddev = rdev->mddev;
2430
2431 if (!entry->store)
2432 return -EIO;
2433 if (!capable(CAP_SYS_ADMIN))
2434 return -EACCES;
2435 rv = mddev ? mddev_lock(mddev): -EBUSY;
2436 if (!rv) {
2437 if (rdev->mddev == NULL)
2438 rv = -EBUSY;
2439 else
2440 rv = entry->store(rdev, page, length);
2441 mddev_unlock(mddev);
2442 }
2443 return rv;
2444 }
2445
2446 static void rdev_free(struct kobject *ko)
2447 {
2448 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2449 kfree(rdev);
2450 }
2451 static struct sysfs_ops rdev_sysfs_ops = {
2452 .show = rdev_attr_show,
2453 .store = rdev_attr_store,
2454 };
2455 static struct kobj_type rdev_ktype = {
2456 .release = rdev_free,
2457 .sysfs_ops = &rdev_sysfs_ops,
2458 .default_attrs = rdev_default_attrs,
2459 };
2460
2461 /*
2462 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2463 *
2464 * mark the device faulty if:
2465 *
2466 * - the device is nonexistent (zero size)
2467 * - the device has no valid superblock
2468 *
2469 * a faulty rdev _never_ has rdev->sb set.
2470 */
2471 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2472 {
2473 char b[BDEVNAME_SIZE];
2474 int err;
2475 mdk_rdev_t *rdev;
2476 sector_t size;
2477
2478 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2479 if (!rdev) {
2480 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2481 return ERR_PTR(-ENOMEM);
2482 }
2483
2484 if ((err = alloc_disk_sb(rdev)))
2485 goto abort_free;
2486
2487 err = lock_rdev(rdev, newdev, super_format == -2);
2488 if (err)
2489 goto abort_free;
2490
2491 kobject_init(&rdev->kobj, &rdev_ktype);
2492
2493 rdev->desc_nr = -1;
2494 rdev->saved_raid_disk = -1;
2495 rdev->raid_disk = -1;
2496 rdev->flags = 0;
2497 rdev->data_offset = 0;
2498 rdev->sb_events = 0;
2499 atomic_set(&rdev->nr_pending, 0);
2500 atomic_set(&rdev->read_errors, 0);
2501 atomic_set(&rdev->corrected_errors, 0);
2502
2503 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2504 if (!size) {
2505 printk(KERN_WARNING
2506 "md: %s has zero or unknown size, marking faulty!\n",
2507 bdevname(rdev->bdev,b));
2508 err = -EINVAL;
2509 goto abort_free;
2510 }
2511
2512 if (super_format >= 0) {
2513 err = super_types[super_format].
2514 load_super(rdev, NULL, super_minor);
2515 if (err == -EINVAL) {
2516 printk(KERN_WARNING
2517 "md: %s does not have a valid v%d.%d "
2518 "superblock, not importing!\n",
2519 bdevname(rdev->bdev,b),
2520 super_format, super_minor);
2521 goto abort_free;
2522 }
2523 if (err < 0) {
2524 printk(KERN_WARNING
2525 "md: could not read %s's sb, not importing!\n",
2526 bdevname(rdev->bdev,b));
2527 goto abort_free;
2528 }
2529 }
2530
2531 INIT_LIST_HEAD(&rdev->same_set);
2532 init_waitqueue_head(&rdev->blocked_wait);
2533
2534 return rdev;
2535
2536 abort_free:
2537 if (rdev->sb_page) {
2538 if (rdev->bdev)
2539 unlock_rdev(rdev);
2540 free_disk_sb(rdev);
2541 }
2542 kfree(rdev);
2543 return ERR_PTR(err);
2544 }
2545
2546 /*
2547 * Check a full RAID array for plausibility
2548 */
2549
2550
2551 static void analyze_sbs(mddev_t * mddev)
2552 {
2553 int i;
2554 mdk_rdev_t *rdev, *freshest, *tmp;
2555 char b[BDEVNAME_SIZE];
2556
2557 freshest = NULL;
2558 rdev_for_each(rdev, tmp, mddev)
2559 switch (super_types[mddev->major_version].
2560 load_super(rdev, freshest, mddev->minor_version)) {
2561 case 1:
2562 freshest = rdev;
2563 break;
2564 case 0:
2565 break;
2566 default:
2567 printk( KERN_ERR \
2568 "md: fatal superblock inconsistency in %s"
2569 " -- removing from array\n",
2570 bdevname(rdev->bdev,b));
2571 kick_rdev_from_array(rdev);
2572 }
2573
2574
2575 super_types[mddev->major_version].
2576 validate_super(mddev, freshest);
2577
2578 i = 0;
2579 rdev_for_each(rdev, tmp, mddev) {
2580 if (rdev->desc_nr >= mddev->max_disks ||
2581 i > mddev->max_disks) {
2582 printk(KERN_WARNING
2583 "md: %s: %s: only %d devices permitted\n",
2584 mdname(mddev), bdevname(rdev->bdev, b),
2585 mddev->max_disks);
2586 kick_rdev_from_array(rdev);
2587 continue;
2588 }
2589 if (rdev != freshest)
2590 if (super_types[mddev->major_version].
2591 validate_super(mddev, rdev)) {
2592 printk(KERN_WARNING "md: kicking non-fresh %s"
2593 " from array!\n",
2594 bdevname(rdev->bdev,b));
2595 kick_rdev_from_array(rdev);
2596 continue;
2597 }
2598 if (mddev->level == LEVEL_MULTIPATH) {
2599 rdev->desc_nr = i++;
2600 rdev->raid_disk = rdev->desc_nr;
2601 set_bit(In_sync, &rdev->flags);
2602 } else if (rdev->raid_disk >= mddev->raid_disks) {
2603 rdev->raid_disk = -1;
2604 clear_bit(In_sync, &rdev->flags);
2605 }
2606 }
2607 }
2608
2609 static void md_safemode_timeout(unsigned long data);
2610
2611 static ssize_t
2612 safe_delay_show(mddev_t *mddev, char *page)
2613 {
2614 int msec = (mddev->safemode_delay*1000)/HZ;
2615 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2616 }
2617 static ssize_t
2618 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2619 {
2620 int scale=1;
2621 int dot=0;
2622 int i;
2623 unsigned long msec;
2624 char buf[30];
2625
2626 /* remove a period, and count digits after it */
2627 if (len >= sizeof(buf))
2628 return -EINVAL;
2629 strlcpy(buf, cbuf, sizeof(buf));
2630 for (i=0; i<len; i++) {
2631 if (dot) {
2632 if (isdigit(buf[i])) {
2633 buf[i-1] = buf[i];
2634 scale *= 10;
2635 }
2636 buf[i] = 0;
2637 } else if (buf[i] == '.') {
2638 dot=1;
2639 buf[i] = 0;
2640 }
2641 }
2642 if (strict_strtoul(buf, 10, &msec) < 0)
2643 return -EINVAL;
2644 msec = (msec * 1000) / scale;
2645 if (msec == 0)
2646 mddev->safemode_delay = 0;
2647 else {
2648 unsigned long old_delay = mddev->safemode_delay;
2649 mddev->safemode_delay = (msec*HZ)/1000;
2650 if (mddev->safemode_delay == 0)
2651 mddev->safemode_delay = 1;
2652 if (mddev->safemode_delay < old_delay)
2653 md_safemode_timeout((unsigned long)mddev);
2654 }
2655 return len;
2656 }
2657 static struct md_sysfs_entry md_safe_delay =
2658 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2659
2660 static ssize_t
2661 level_show(mddev_t *mddev, char *page)
2662 {
2663 struct mdk_personality *p = mddev->pers;
2664 if (p)
2665 return sprintf(page, "%s\n", p->name);
2666 else if (mddev->clevel[0])
2667 return sprintf(page, "%s\n", mddev->clevel);
2668 else if (mddev->level != LEVEL_NONE)
2669 return sprintf(page, "%d\n", mddev->level);
2670 else
2671 return 0;
2672 }
2673
2674 static ssize_t
2675 level_store(mddev_t *mddev, const char *buf, size_t len)
2676 {
2677 char level[16];
2678 ssize_t rv = len;
2679 struct mdk_personality *pers;
2680 void *priv;
2681
2682 if (mddev->pers == NULL) {
2683 if (len == 0)
2684 return 0;
2685 if (len >= sizeof(mddev->clevel))
2686 return -ENOSPC;
2687 strncpy(mddev->clevel, buf, len);
2688 if (mddev->clevel[len-1] == '\n')
2689 len--;
2690 mddev->clevel[len] = 0;
2691 mddev->level = LEVEL_NONE;
2692 return rv;
2693 }
2694
2695 /* request to change the personality. Need to ensure:
2696 * - array is not engaged in resync/recovery/reshape
2697 * - old personality can be suspended
2698 * - new personality will access other array.
2699 */
2700
2701 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2702 return -EBUSY;
2703
2704 if (!mddev->pers->quiesce) {
2705 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2706 mdname(mddev), mddev->pers->name);
2707 return -EINVAL;
2708 }
2709
2710 /* Now find the new personality */
2711 if (len == 0 || len >= sizeof(level))
2712 return -EINVAL;
2713 strncpy(level, buf, len);
2714 if (level[len-1] == '\n')
2715 len--;
2716 level[len] = 0;
2717
2718 request_module("md-%s", level);
2719 spin_lock(&pers_lock);
2720 pers = find_pers(LEVEL_NONE, level);
2721 if (!pers || !try_module_get(pers->owner)) {
2722 spin_unlock(&pers_lock);
2723 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2724 return -EINVAL;
2725 }
2726 spin_unlock(&pers_lock);
2727
2728 if (pers == mddev->pers) {
2729 /* Nothing to do! */
2730 module_put(pers->owner);
2731 return rv;
2732 }
2733 if (!pers->takeover) {
2734 module_put(pers->owner);
2735 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2736 mdname(mddev), level);
2737 return -EINVAL;
2738 }
2739
2740 /* ->takeover must set new_* and/or delta_disks
2741 * if it succeeds, and may set them when it fails.
2742 */
2743 priv = pers->takeover(mddev);
2744 if (IS_ERR(priv)) {
2745 mddev->new_level = mddev->level;
2746 mddev->new_layout = mddev->layout;
2747 mddev->new_chunk_sectors = mddev->chunk_sectors;
2748 mddev->raid_disks -= mddev->delta_disks;
2749 mddev->delta_disks = 0;
2750 module_put(pers->owner);
2751 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2752 mdname(mddev), level);
2753 return PTR_ERR(priv);
2754 }
2755
2756 /* Looks like we have a winner */
2757 mddev_suspend(mddev);
2758 mddev->pers->stop(mddev);
2759 module_put(mddev->pers->owner);
2760 mddev->pers = pers;
2761 mddev->private = priv;
2762 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2763 mddev->level = mddev->new_level;
2764 mddev->layout = mddev->new_layout;
2765 mddev->chunk_sectors = mddev->new_chunk_sectors;
2766 mddev->delta_disks = 0;
2767 pers->run(mddev);
2768 mddev_resume(mddev);
2769 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2770 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2771 md_wakeup_thread(mddev->thread);
2772 return rv;
2773 }
2774
2775 static struct md_sysfs_entry md_level =
2776 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2777
2778
2779 static ssize_t
2780 layout_show(mddev_t *mddev, char *page)
2781 {
2782 /* just a number, not meaningful for all levels */
2783 if (mddev->reshape_position != MaxSector &&
2784 mddev->layout != mddev->new_layout)
2785 return sprintf(page, "%d (%d)\n",
2786 mddev->new_layout, mddev->layout);
2787 return sprintf(page, "%d\n", mddev->layout);
2788 }
2789
2790 static ssize_t
2791 layout_store(mddev_t *mddev, const char *buf, size_t len)
2792 {
2793 char *e;
2794 unsigned long n = simple_strtoul(buf, &e, 10);
2795
2796 if (!*buf || (*e && *e != '\n'))
2797 return -EINVAL;
2798
2799 if (mddev->pers) {
2800 int err;
2801 if (mddev->pers->check_reshape == NULL)
2802 return -EBUSY;
2803 mddev->new_layout = n;
2804 err = mddev->pers->check_reshape(mddev);
2805 if (err) {
2806 mddev->new_layout = mddev->layout;
2807 return err;
2808 }
2809 } else {
2810 mddev->new_layout = n;
2811 if (mddev->reshape_position == MaxSector)
2812 mddev->layout = n;
2813 }
2814 return len;
2815 }
2816 static struct md_sysfs_entry md_layout =
2817 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2818
2819
2820 static ssize_t
2821 raid_disks_show(mddev_t *mddev, char *page)
2822 {
2823 if (mddev->raid_disks == 0)
2824 return 0;
2825 if (mddev->reshape_position != MaxSector &&
2826 mddev->delta_disks != 0)
2827 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2828 mddev->raid_disks - mddev->delta_disks);
2829 return sprintf(page, "%d\n", mddev->raid_disks);
2830 }
2831
2832 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2833
2834 static ssize_t
2835 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2836 {
2837 char *e;
2838 int rv = 0;
2839 unsigned long n = simple_strtoul(buf, &e, 10);
2840
2841 if (!*buf || (*e && *e != '\n'))
2842 return -EINVAL;
2843
2844 if (mddev->pers)
2845 rv = update_raid_disks(mddev, n);
2846 else if (mddev->reshape_position != MaxSector) {
2847 int olddisks = mddev->raid_disks - mddev->delta_disks;
2848 mddev->delta_disks = n - olddisks;
2849 mddev->raid_disks = n;
2850 } else
2851 mddev->raid_disks = n;
2852 return rv ? rv : len;
2853 }
2854 static struct md_sysfs_entry md_raid_disks =
2855 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2856
2857 static ssize_t
2858 chunk_size_show(mddev_t *mddev, char *page)
2859 {
2860 if (mddev->reshape_position != MaxSector &&
2861 mddev->chunk_sectors != mddev->new_chunk_sectors)
2862 return sprintf(page, "%d (%d)\n",
2863 mddev->new_chunk_sectors << 9,
2864 mddev->chunk_sectors << 9);
2865 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2866 }
2867
2868 static ssize_t
2869 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2870 {
2871 char *e;
2872 unsigned long n = simple_strtoul(buf, &e, 10);
2873
2874 if (!*buf || (*e && *e != '\n'))
2875 return -EINVAL;
2876
2877 if (mddev->pers) {
2878 int err;
2879 if (mddev->pers->check_reshape == NULL)
2880 return -EBUSY;
2881 mddev->new_chunk_sectors = n >> 9;
2882 err = mddev->pers->check_reshape(mddev);
2883 if (err) {
2884 mddev->new_chunk_sectors = mddev->chunk_sectors;
2885 return err;
2886 }
2887 } else {
2888 mddev->new_chunk_sectors = n >> 9;
2889 if (mddev->reshape_position == MaxSector)
2890 mddev->chunk_sectors = n >> 9;
2891 }
2892 return len;
2893 }
2894 static struct md_sysfs_entry md_chunk_size =
2895 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2896
2897 static ssize_t
2898 resync_start_show(mddev_t *mddev, char *page)
2899 {
2900 if (mddev->recovery_cp == MaxSector)
2901 return sprintf(page, "none\n");
2902 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2903 }
2904
2905 static ssize_t
2906 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2907 {
2908 char *e;
2909 unsigned long long n = simple_strtoull(buf, &e, 10);
2910
2911 if (mddev->pers)
2912 return -EBUSY;
2913 if (!*buf || (*e && *e != '\n'))
2914 return -EINVAL;
2915
2916 mddev->recovery_cp = n;
2917 return len;
2918 }
2919 static struct md_sysfs_entry md_resync_start =
2920 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2921
2922 /*
2923 * The array state can be:
2924 *
2925 * clear
2926 * No devices, no size, no level
2927 * Equivalent to STOP_ARRAY ioctl
2928 * inactive
2929 * May have some settings, but array is not active
2930 * all IO results in error
2931 * When written, doesn't tear down array, but just stops it
2932 * suspended (not supported yet)
2933 * All IO requests will block. The array can be reconfigured.
2934 * Writing this, if accepted, will block until array is quiescent
2935 * readonly
2936 * no resync can happen. no superblocks get written.
2937 * write requests fail
2938 * read-auto
2939 * like readonly, but behaves like 'clean' on a write request.
2940 *
2941 * clean - no pending writes, but otherwise active.
2942 * When written to inactive array, starts without resync
2943 * If a write request arrives then
2944 * if metadata is known, mark 'dirty' and switch to 'active'.
2945 * if not known, block and switch to write-pending
2946 * If written to an active array that has pending writes, then fails.
2947 * active
2948 * fully active: IO and resync can be happening.
2949 * When written to inactive array, starts with resync
2950 *
2951 * write-pending
2952 * clean, but writes are blocked waiting for 'active' to be written.
2953 *
2954 * active-idle
2955 * like active, but no writes have been seen for a while (100msec).
2956 *
2957 */
2958 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2959 write_pending, active_idle, bad_word};
2960 static char *array_states[] = {
2961 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2962 "write-pending", "active-idle", NULL };
2963
2964 static int match_word(const char *word, char **list)
2965 {
2966 int n;
2967 for (n=0; list[n]; n++)
2968 if (cmd_match(word, list[n]))
2969 break;
2970 return n;
2971 }
2972
2973 static ssize_t
2974 array_state_show(mddev_t *mddev, char *page)
2975 {
2976 enum array_state st = inactive;
2977
2978 if (mddev->pers)
2979 switch(mddev->ro) {
2980 case 1:
2981 st = readonly;
2982 break;
2983 case 2:
2984 st = read_auto;
2985 break;
2986 case 0:
2987 if (mddev->in_sync)
2988 st = clean;
2989 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2990 st = write_pending;
2991 else if (mddev->safemode)
2992 st = active_idle;
2993 else
2994 st = active;
2995 }
2996 else {
2997 if (list_empty(&mddev->disks) &&
2998 mddev->raid_disks == 0 &&
2999 mddev->dev_sectors == 0)
3000 st = clear;
3001 else
3002 st = inactive;
3003 }
3004 return sprintf(page, "%s\n", array_states[st]);
3005 }
3006
3007 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3008 static int do_md_run(mddev_t * mddev);
3009 static int restart_array(mddev_t *mddev);
3010
3011 static ssize_t
3012 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3013 {
3014 int err = -EINVAL;
3015 enum array_state st = match_word(buf, array_states);
3016 switch(st) {
3017 case bad_word:
3018 break;
3019 case clear:
3020 /* stopping an active array */
3021 if (atomic_read(&mddev->openers) > 0)
3022 return -EBUSY;
3023 err = do_md_stop(mddev, 0, 0);
3024 break;
3025 case inactive:
3026 /* stopping an active array */
3027 if (mddev->pers) {
3028 if (atomic_read(&mddev->openers) > 0)
3029 return -EBUSY;
3030 err = do_md_stop(mddev, 2, 0);
3031 } else
3032 err = 0; /* already inactive */
3033 break;
3034 case suspended:
3035 break; /* not supported yet */
3036 case readonly:
3037 if (mddev->pers)
3038 err = do_md_stop(mddev, 1, 0);
3039 else {
3040 mddev->ro = 1;
3041 set_disk_ro(mddev->gendisk, 1);
3042 err = do_md_run(mddev);
3043 }
3044 break;
3045 case read_auto:
3046 if (mddev->pers) {
3047 if (mddev->ro == 0)
3048 err = do_md_stop(mddev, 1, 0);
3049 else if (mddev->ro == 1)
3050 err = restart_array(mddev);
3051 if (err == 0) {
3052 mddev->ro = 2;
3053 set_disk_ro(mddev->gendisk, 0);
3054 }
3055 } else {
3056 mddev->ro = 2;
3057 err = do_md_run(mddev);
3058 }
3059 break;
3060 case clean:
3061 if (mddev->pers) {
3062 restart_array(mddev);
3063 spin_lock_irq(&mddev->write_lock);
3064 if (atomic_read(&mddev->writes_pending) == 0) {
3065 if (mddev->in_sync == 0) {
3066 mddev->in_sync = 1;
3067 if (mddev->safemode == 1)
3068 mddev->safemode = 0;
3069 if (mddev->persistent)
3070 set_bit(MD_CHANGE_CLEAN,
3071 &mddev->flags);
3072 }
3073 err = 0;
3074 } else
3075 err = -EBUSY;
3076 spin_unlock_irq(&mddev->write_lock);
3077 } else
3078 err = -EINVAL;
3079 break;
3080 case active:
3081 if (mddev->pers) {
3082 restart_array(mddev);
3083 if (mddev->external)
3084 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3085 wake_up(&mddev->sb_wait);
3086 err = 0;
3087 } else {
3088 mddev->ro = 0;
3089 set_disk_ro(mddev->gendisk, 0);
3090 err = do_md_run(mddev);
3091 }
3092 break;
3093 case write_pending:
3094 case active_idle:
3095 /* these cannot be set */
3096 break;
3097 }
3098 if (err)
3099 return err;
3100 else {
3101 sysfs_notify_dirent(mddev->sysfs_state);
3102 return len;
3103 }
3104 }
3105 static struct md_sysfs_entry md_array_state =
3106 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3107
3108 static ssize_t
3109 null_show(mddev_t *mddev, char *page)
3110 {
3111 return -EINVAL;
3112 }
3113
3114 static ssize_t
3115 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3116 {
3117 /* buf must be %d:%d\n? giving major and minor numbers */
3118 /* The new device is added to the array.
3119 * If the array has a persistent superblock, we read the
3120 * superblock to initialise info and check validity.
3121 * Otherwise, only checking done is that in bind_rdev_to_array,
3122 * which mainly checks size.
3123 */
3124 char *e;
3125 int major = simple_strtoul(buf, &e, 10);
3126 int minor;
3127 dev_t dev;
3128 mdk_rdev_t *rdev;
3129 int err;
3130
3131 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3132 return -EINVAL;
3133 minor = simple_strtoul(e+1, &e, 10);
3134 if (*e && *e != '\n')
3135 return -EINVAL;
3136 dev = MKDEV(major, minor);
3137 if (major != MAJOR(dev) ||
3138 minor != MINOR(dev))
3139 return -EOVERFLOW;
3140
3141
3142 if (mddev->persistent) {
3143 rdev = md_import_device(dev, mddev->major_version,
3144 mddev->minor_version);
3145 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3146 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3147 mdk_rdev_t, same_set);
3148 err = super_types[mddev->major_version]
3149 .load_super(rdev, rdev0, mddev->minor_version);
3150 if (err < 0)
3151 goto out;
3152 }
3153 } else if (mddev->external)
3154 rdev = md_import_device(dev, -2, -1);
3155 else
3156 rdev = md_import_device(dev, -1, -1);
3157
3158 if (IS_ERR(rdev))
3159 return PTR_ERR(rdev);
3160 err = bind_rdev_to_array(rdev, mddev);
3161 out:
3162 if (err)
3163 export_rdev(rdev);
3164 return err ? err : len;
3165 }
3166
3167 static struct md_sysfs_entry md_new_device =
3168 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3169
3170 static ssize_t
3171 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3172 {
3173 char *end;
3174 unsigned long chunk, end_chunk;
3175
3176 if (!mddev->bitmap)
3177 goto out;
3178 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3179 while (*buf) {
3180 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3181 if (buf == end) break;
3182 if (*end == '-') { /* range */
3183 buf = end + 1;
3184 end_chunk = simple_strtoul(buf, &end, 0);
3185 if (buf == end) break;
3186 }
3187 if (*end && !isspace(*end)) break;
3188 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3189 buf = end;
3190 while (isspace(*buf)) buf++;
3191 }
3192 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3193 out:
3194 return len;
3195 }
3196
3197 static struct md_sysfs_entry md_bitmap =
3198 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3199
3200 static ssize_t
3201 size_show(mddev_t *mddev, char *page)
3202 {
3203 return sprintf(page, "%llu\n",
3204 (unsigned long long)mddev->dev_sectors / 2);
3205 }
3206
3207 static int update_size(mddev_t *mddev, sector_t num_sectors);
3208
3209 static ssize_t
3210 size_store(mddev_t *mddev, const char *buf, size_t len)
3211 {
3212 /* If array is inactive, we can reduce the component size, but
3213 * not increase it (except from 0).
3214 * If array is active, we can try an on-line resize
3215 */
3216 sector_t sectors;
3217 int err = strict_blocks_to_sectors(buf, &sectors);
3218
3219 if (err < 0)
3220 return err;
3221 if (mddev->pers) {
3222 err = update_size(mddev, sectors);
3223 md_update_sb(mddev, 1);
3224 } else {
3225 if (mddev->dev_sectors == 0 ||
3226 mddev->dev_sectors > sectors)
3227 mddev->dev_sectors = sectors;
3228 else
3229 err = -ENOSPC;
3230 }
3231 return err ? err : len;
3232 }
3233
3234 static struct md_sysfs_entry md_size =
3235 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3236
3237
3238 /* Metdata version.
3239 * This is one of
3240 * 'none' for arrays with no metadata (good luck...)
3241 * 'external' for arrays with externally managed metadata,
3242 * or N.M for internally known formats
3243 */
3244 static ssize_t
3245 metadata_show(mddev_t *mddev, char *page)
3246 {
3247 if (mddev->persistent)
3248 return sprintf(page, "%d.%d\n",
3249 mddev->major_version, mddev->minor_version);
3250 else if (mddev->external)
3251 return sprintf(page, "external:%s\n", mddev->metadata_type);
3252 else
3253 return sprintf(page, "none\n");
3254 }
3255
3256 static ssize_t
3257 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3258 {
3259 int major, minor;
3260 char *e;
3261 /* Changing the details of 'external' metadata is
3262 * always permitted. Otherwise there must be
3263 * no devices attached to the array.
3264 */
3265 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3266 ;
3267 else if (!list_empty(&mddev->disks))
3268 return -EBUSY;
3269
3270 if (cmd_match(buf, "none")) {
3271 mddev->persistent = 0;
3272 mddev->external = 0;
3273 mddev->major_version = 0;
3274 mddev->minor_version = 90;
3275 return len;
3276 }
3277 if (strncmp(buf, "external:", 9) == 0) {
3278 size_t namelen = len-9;
3279 if (namelen >= sizeof(mddev->metadata_type))
3280 namelen = sizeof(mddev->metadata_type)-1;
3281 strncpy(mddev->metadata_type, buf+9, namelen);
3282 mddev->metadata_type[namelen] = 0;
3283 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3284 mddev->metadata_type[--namelen] = 0;
3285 mddev->persistent = 0;
3286 mddev->external = 1;
3287 mddev->major_version = 0;
3288 mddev->minor_version = 90;
3289 return len;
3290 }
3291 major = simple_strtoul(buf, &e, 10);
3292 if (e==buf || *e != '.')
3293 return -EINVAL;
3294 buf = e+1;
3295 minor = simple_strtoul(buf, &e, 10);
3296 if (e==buf || (*e && *e != '\n') )
3297 return -EINVAL;
3298 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3299 return -ENOENT;
3300 mddev->major_version = major;
3301 mddev->minor_version = minor;
3302 mddev->persistent = 1;
3303 mddev->external = 0;
3304 return len;
3305 }
3306
3307 static struct md_sysfs_entry md_metadata =
3308 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3309
3310 static ssize_t
3311 action_show(mddev_t *mddev, char *page)
3312 {
3313 char *type = "idle";
3314 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3315 type = "frozen";
3316 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3317 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3318 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3319 type = "reshape";
3320 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3321 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3322 type = "resync";
3323 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3324 type = "check";
3325 else
3326 type = "repair";
3327 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3328 type = "recover";
3329 }
3330 return sprintf(page, "%s\n", type);
3331 }
3332
3333 static ssize_t
3334 action_store(mddev_t *mddev, const char *page, size_t len)
3335 {
3336 if (!mddev->pers || !mddev->pers->sync_request)
3337 return -EINVAL;
3338
3339 if (cmd_match(page, "frozen"))
3340 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3341 else
3342 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3343
3344 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3345 if (mddev->sync_thread) {
3346 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3347 md_unregister_thread(mddev->sync_thread);
3348 mddev->sync_thread = NULL;
3349 mddev->recovery = 0;
3350 }
3351 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3352 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3353 return -EBUSY;
3354 else if (cmd_match(page, "resync"))
3355 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3356 else if (cmd_match(page, "recover")) {
3357 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3358 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3359 } else if (cmd_match(page, "reshape")) {
3360 int err;
3361 if (mddev->pers->start_reshape == NULL)
3362 return -EINVAL;
3363 err = mddev->pers->start_reshape(mddev);
3364 if (err)
3365 return err;
3366 sysfs_notify(&mddev->kobj, NULL, "degraded");
3367 } else {
3368 if (cmd_match(page, "check"))
3369 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3370 else if (!cmd_match(page, "repair"))
3371 return -EINVAL;
3372 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3373 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3374 }
3375 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3376 md_wakeup_thread(mddev->thread);
3377 sysfs_notify_dirent(mddev->sysfs_action);
3378 return len;
3379 }
3380
3381 static ssize_t
3382 mismatch_cnt_show(mddev_t *mddev, char *page)
3383 {
3384 return sprintf(page, "%llu\n",
3385 (unsigned long long) mddev->resync_mismatches);
3386 }
3387
3388 static struct md_sysfs_entry md_scan_mode =
3389 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3390
3391
3392 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3393
3394 static ssize_t
3395 sync_min_show(mddev_t *mddev, char *page)
3396 {
3397 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3398 mddev->sync_speed_min ? "local": "system");
3399 }
3400
3401 static ssize_t
3402 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3403 {
3404 int min;
3405 char *e;
3406 if (strncmp(buf, "system", 6)==0) {
3407 mddev->sync_speed_min = 0;
3408 return len;
3409 }
3410 min = simple_strtoul(buf, &e, 10);
3411 if (buf == e || (*e && *e != '\n') || min <= 0)
3412 return -EINVAL;
3413 mddev->sync_speed_min = min;
3414 return len;
3415 }
3416
3417 static struct md_sysfs_entry md_sync_min =
3418 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3419
3420 static ssize_t
3421 sync_max_show(mddev_t *mddev, char *page)
3422 {
3423 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3424 mddev->sync_speed_max ? "local": "system");
3425 }
3426
3427 static ssize_t
3428 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3429 {
3430 int max;
3431 char *e;
3432 if (strncmp(buf, "system", 6)==0) {
3433 mddev->sync_speed_max = 0;
3434 return len;
3435 }
3436 max = simple_strtoul(buf, &e, 10);
3437 if (buf == e || (*e && *e != '\n') || max <= 0)
3438 return -EINVAL;
3439 mddev->sync_speed_max = max;
3440 return len;
3441 }
3442
3443 static struct md_sysfs_entry md_sync_max =
3444 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3445
3446 static ssize_t
3447 degraded_show(mddev_t *mddev, char *page)
3448 {
3449 return sprintf(page, "%d\n", mddev->degraded);
3450 }
3451 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3452
3453 static ssize_t
3454 sync_force_parallel_show(mddev_t *mddev, char *page)
3455 {
3456 return sprintf(page, "%d\n", mddev->parallel_resync);
3457 }
3458
3459 static ssize_t
3460 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3461 {
3462 long n;
3463
3464 if (strict_strtol(buf, 10, &n))
3465 return -EINVAL;
3466
3467 if (n != 0 && n != 1)
3468 return -EINVAL;
3469
3470 mddev->parallel_resync = n;
3471
3472 if (mddev->sync_thread)
3473 wake_up(&resync_wait);
3474
3475 return len;
3476 }
3477
3478 /* force parallel resync, even with shared block devices */
3479 static struct md_sysfs_entry md_sync_force_parallel =
3480 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3481 sync_force_parallel_show, sync_force_parallel_store);
3482
3483 static ssize_t
3484 sync_speed_show(mddev_t *mddev, char *page)
3485 {
3486 unsigned long resync, dt, db;
3487 if (mddev->curr_resync == 0)
3488 return sprintf(page, "none\n");
3489 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3490 dt = (jiffies - mddev->resync_mark) / HZ;
3491 if (!dt) dt++;
3492 db = resync - mddev->resync_mark_cnt;
3493 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3494 }
3495
3496 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3497
3498 static ssize_t
3499 sync_completed_show(mddev_t *mddev, char *page)
3500 {
3501 unsigned long max_sectors, resync;
3502
3503 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3504 return sprintf(page, "none\n");
3505
3506 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3507 max_sectors = mddev->resync_max_sectors;
3508 else
3509 max_sectors = mddev->dev_sectors;
3510
3511 resync = mddev->curr_resync_completed;
3512 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3513 }
3514
3515 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3516
3517 static ssize_t
3518 min_sync_show(mddev_t *mddev, char *page)
3519 {
3520 return sprintf(page, "%llu\n",
3521 (unsigned long long)mddev->resync_min);
3522 }
3523 static ssize_t
3524 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3525 {
3526 unsigned long long min;
3527 if (strict_strtoull(buf, 10, &min))
3528 return -EINVAL;
3529 if (min > mddev->resync_max)
3530 return -EINVAL;
3531 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3532 return -EBUSY;
3533
3534 /* Must be a multiple of chunk_size */
3535 if (mddev->chunk_sectors) {
3536 sector_t temp = min;
3537 if (sector_div(temp, mddev->chunk_sectors))
3538 return -EINVAL;
3539 }
3540 mddev->resync_min = min;
3541
3542 return len;
3543 }
3544
3545 static struct md_sysfs_entry md_min_sync =
3546 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3547
3548 static ssize_t
3549 max_sync_show(mddev_t *mddev, char *page)
3550 {
3551 if (mddev->resync_max == MaxSector)
3552 return sprintf(page, "max\n");
3553 else
3554 return sprintf(page, "%llu\n",
3555 (unsigned long long)mddev->resync_max);
3556 }
3557 static ssize_t
3558 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3559 {
3560 if (strncmp(buf, "max", 3) == 0)
3561 mddev->resync_max = MaxSector;
3562 else {
3563 unsigned long long max;
3564 if (strict_strtoull(buf, 10, &max))
3565 return -EINVAL;
3566 if (max < mddev->resync_min)
3567 return -EINVAL;
3568 if (max < mddev->resync_max &&
3569 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3570 return -EBUSY;
3571
3572 /* Must be a multiple of chunk_size */
3573 if (mddev->chunk_sectors) {
3574 sector_t temp = max;
3575 if (sector_div(temp, mddev->chunk_sectors))
3576 return -EINVAL;
3577 }
3578 mddev->resync_max = max;
3579 }
3580 wake_up(&mddev->recovery_wait);
3581 return len;
3582 }
3583
3584 static struct md_sysfs_entry md_max_sync =
3585 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3586
3587 static ssize_t
3588 suspend_lo_show(mddev_t *mddev, char *page)
3589 {
3590 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3591 }
3592
3593 static ssize_t
3594 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3595 {
3596 char *e;
3597 unsigned long long new = simple_strtoull(buf, &e, 10);
3598
3599 if (mddev->pers->quiesce == NULL)
3600 return -EINVAL;
3601 if (buf == e || (*e && *e != '\n'))
3602 return -EINVAL;
3603 if (new >= mddev->suspend_hi ||
3604 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3605 mddev->suspend_lo = new;
3606 mddev->pers->quiesce(mddev, 2);
3607 return len;
3608 } else
3609 return -EINVAL;
3610 }
3611 static struct md_sysfs_entry md_suspend_lo =
3612 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3613
3614
3615 static ssize_t
3616 suspend_hi_show(mddev_t *mddev, char *page)
3617 {
3618 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3619 }
3620
3621 static ssize_t
3622 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3623 {
3624 char *e;
3625 unsigned long long new = simple_strtoull(buf, &e, 10);
3626
3627 if (mddev->pers->quiesce == NULL)
3628 return -EINVAL;
3629 if (buf == e || (*e && *e != '\n'))
3630 return -EINVAL;
3631 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3632 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3633 mddev->suspend_hi = new;
3634 mddev->pers->quiesce(mddev, 1);
3635 mddev->pers->quiesce(mddev, 0);
3636 return len;
3637 } else
3638 return -EINVAL;
3639 }
3640 static struct md_sysfs_entry md_suspend_hi =
3641 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3642
3643 static ssize_t
3644 reshape_position_show(mddev_t *mddev, char *page)
3645 {
3646 if (mddev->reshape_position != MaxSector)
3647 return sprintf(page, "%llu\n",
3648 (unsigned long long)mddev->reshape_position);
3649 strcpy(page, "none\n");
3650 return 5;
3651 }
3652
3653 static ssize_t
3654 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3655 {
3656 char *e;
3657 unsigned long long new = simple_strtoull(buf, &e, 10);
3658 if (mddev->pers)
3659 return -EBUSY;
3660 if (buf == e || (*e && *e != '\n'))
3661 return -EINVAL;
3662 mddev->reshape_position = new;
3663 mddev->delta_disks = 0;
3664 mddev->new_level = mddev->level;
3665 mddev->new_layout = mddev->layout;
3666 mddev->new_chunk_sectors = mddev->chunk_sectors;
3667 return len;
3668 }
3669
3670 static struct md_sysfs_entry md_reshape_position =
3671 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3672 reshape_position_store);
3673
3674 static ssize_t
3675 array_size_show(mddev_t *mddev, char *page)
3676 {
3677 if (mddev->external_size)
3678 return sprintf(page, "%llu\n",
3679 (unsigned long long)mddev->array_sectors/2);
3680 else
3681 return sprintf(page, "default\n");
3682 }
3683
3684 static ssize_t
3685 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3686 {
3687 sector_t sectors;
3688
3689 if (strncmp(buf, "default", 7) == 0) {
3690 if (mddev->pers)
3691 sectors = mddev->pers->size(mddev, 0, 0);
3692 else
3693 sectors = mddev->array_sectors;
3694
3695 mddev->external_size = 0;
3696 } else {
3697 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3698 return -EINVAL;
3699 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3700 return -E2BIG;
3701
3702 mddev->external_size = 1;
3703 }
3704
3705 mddev->array_sectors = sectors;
3706 set_capacity(mddev->gendisk, mddev->array_sectors);
3707 if (mddev->pers) {
3708 struct block_device *bdev = bdget_disk(mddev->gendisk, 0);
3709
3710 if (bdev) {
3711 mutex_lock(&bdev->bd_inode->i_mutex);
3712 i_size_write(bdev->bd_inode,
3713 (loff_t)mddev->array_sectors << 9);
3714 mutex_unlock(&bdev->bd_inode->i_mutex);
3715 bdput(bdev);
3716 }
3717 }
3718
3719 return len;
3720 }
3721
3722 static struct md_sysfs_entry md_array_size =
3723 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3724 array_size_store);
3725
3726 static struct attribute *md_default_attrs[] = {
3727 &md_level.attr,
3728 &md_layout.attr,
3729 &md_raid_disks.attr,
3730 &md_chunk_size.attr,
3731 &md_size.attr,
3732 &md_resync_start.attr,
3733 &md_metadata.attr,
3734 &md_new_device.attr,
3735 &md_safe_delay.attr,
3736 &md_array_state.attr,
3737 &md_reshape_position.attr,
3738 &md_array_size.attr,
3739 NULL,
3740 };
3741
3742 static struct attribute *md_redundancy_attrs[] = {
3743 &md_scan_mode.attr,
3744 &md_mismatches.attr,
3745 &md_sync_min.attr,
3746 &md_sync_max.attr,
3747 &md_sync_speed.attr,
3748 &md_sync_force_parallel.attr,
3749 &md_sync_completed.attr,
3750 &md_min_sync.attr,
3751 &md_max_sync.attr,
3752 &md_suspend_lo.attr,
3753 &md_suspend_hi.attr,
3754 &md_bitmap.attr,
3755 &md_degraded.attr,
3756 NULL,
3757 };
3758 static struct attribute_group md_redundancy_group = {
3759 .name = NULL,
3760 .attrs = md_redundancy_attrs,
3761 };
3762
3763
3764 static ssize_t
3765 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3766 {
3767 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3768 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3769 ssize_t rv;
3770
3771 if (!entry->show)
3772 return -EIO;
3773 rv = mddev_lock(mddev);
3774 if (!rv) {
3775 rv = entry->show(mddev, page);
3776 mddev_unlock(mddev);
3777 }
3778 return rv;
3779 }
3780
3781 static ssize_t
3782 md_attr_store(struct kobject *kobj, struct attribute *attr,
3783 const char *page, size_t length)
3784 {
3785 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3786 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3787 ssize_t rv;
3788
3789 if (!entry->store)
3790 return -EIO;
3791 if (!capable(CAP_SYS_ADMIN))
3792 return -EACCES;
3793 rv = mddev_lock(mddev);
3794 if (mddev->hold_active == UNTIL_IOCTL)
3795 mddev->hold_active = 0;
3796 if (!rv) {
3797 rv = entry->store(mddev, page, length);
3798 mddev_unlock(mddev);
3799 }
3800 return rv;
3801 }
3802
3803 static void md_free(struct kobject *ko)
3804 {
3805 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3806
3807 if (mddev->sysfs_state)
3808 sysfs_put(mddev->sysfs_state);
3809
3810 if (mddev->gendisk) {
3811 del_gendisk(mddev->gendisk);
3812 put_disk(mddev->gendisk);
3813 }
3814 if (mddev->queue)
3815 blk_cleanup_queue(mddev->queue);
3816
3817 kfree(mddev);
3818 }
3819
3820 static struct sysfs_ops md_sysfs_ops = {
3821 .show = md_attr_show,
3822 .store = md_attr_store,
3823 };
3824 static struct kobj_type md_ktype = {
3825 .release = md_free,
3826 .sysfs_ops = &md_sysfs_ops,
3827 .default_attrs = md_default_attrs,
3828 };
3829
3830 int mdp_major = 0;
3831
3832 static void mddev_delayed_delete(struct work_struct *ws)
3833 {
3834 mddev_t *mddev = container_of(ws, mddev_t, del_work);
3835
3836 if (mddev->private == &md_redundancy_group) {
3837 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3838 if (mddev->sysfs_action)
3839 sysfs_put(mddev->sysfs_action);
3840 mddev->sysfs_action = NULL;
3841 mddev->private = NULL;
3842 }
3843 kobject_del(&mddev->kobj);
3844 kobject_put(&mddev->kobj);
3845 }
3846
3847 static int md_alloc(dev_t dev, char *name)
3848 {
3849 static DEFINE_MUTEX(disks_mutex);
3850 mddev_t *mddev = mddev_find(dev);
3851 struct gendisk *disk;
3852 int partitioned;
3853 int shift;
3854 int unit;
3855 int error;
3856
3857 if (!mddev)
3858 return -ENODEV;
3859
3860 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3861 shift = partitioned ? MdpMinorShift : 0;
3862 unit = MINOR(mddev->unit) >> shift;
3863
3864 /* wait for any previous instance if this device
3865 * to be completed removed (mddev_delayed_delete).
3866 */
3867 flush_scheduled_work();
3868
3869 mutex_lock(&disks_mutex);
3870 if (mddev->gendisk) {
3871 mutex_unlock(&disks_mutex);
3872 mddev_put(mddev);
3873 return -EEXIST;
3874 }
3875
3876 if (name) {
3877 /* Need to ensure that 'name' is not a duplicate.
3878 */
3879 mddev_t *mddev2;
3880 spin_lock(&all_mddevs_lock);
3881
3882 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3883 if (mddev2->gendisk &&
3884 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3885 spin_unlock(&all_mddevs_lock);
3886 return -EEXIST;
3887 }
3888 spin_unlock(&all_mddevs_lock);
3889 }
3890
3891 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3892 if (!mddev->queue) {
3893 mutex_unlock(&disks_mutex);
3894 mddev_put(mddev);
3895 return -ENOMEM;
3896 }
3897 mddev->queue->queuedata = mddev;
3898
3899 /* Can be unlocked because the queue is new: no concurrency */
3900 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3901
3902 blk_queue_make_request(mddev->queue, md_make_request);
3903
3904 disk = alloc_disk(1 << shift);
3905 if (!disk) {
3906 mutex_unlock(&disks_mutex);
3907 blk_cleanup_queue(mddev->queue);
3908 mddev->queue = NULL;
3909 mddev_put(mddev);
3910 return -ENOMEM;
3911 }
3912 disk->major = MAJOR(mddev->unit);
3913 disk->first_minor = unit << shift;
3914 if (name)
3915 strcpy(disk->disk_name, name);
3916 else if (partitioned)
3917 sprintf(disk->disk_name, "md_d%d", unit);
3918 else
3919 sprintf(disk->disk_name, "md%d", unit);
3920 disk->fops = &md_fops;
3921 disk->private_data = mddev;
3922 disk->queue = mddev->queue;
3923 /* Allow extended partitions. This makes the
3924 * 'mdp' device redundant, but we can't really
3925 * remove it now.
3926 */
3927 disk->flags |= GENHD_FL_EXT_DEVT;
3928 add_disk(disk);
3929 mddev->gendisk = disk;
3930 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3931 &disk_to_dev(disk)->kobj, "%s", "md");
3932 mutex_unlock(&disks_mutex);
3933 if (error)
3934 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3935 disk->disk_name);
3936 else {
3937 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3938 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3939 }
3940 mddev_put(mddev);
3941 return 0;
3942 }
3943
3944 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3945 {
3946 md_alloc(dev, NULL);
3947 return NULL;
3948 }
3949
3950 static int add_named_array(const char *val, struct kernel_param *kp)
3951 {
3952 /* val must be "md_*" where * is not all digits.
3953 * We allocate an array with a large free minor number, and
3954 * set the name to val. val must not already be an active name.
3955 */
3956 int len = strlen(val);
3957 char buf[DISK_NAME_LEN];
3958
3959 while (len && val[len-1] == '\n')
3960 len--;
3961 if (len >= DISK_NAME_LEN)
3962 return -E2BIG;
3963 strlcpy(buf, val, len+1);
3964 if (strncmp(buf, "md_", 3) != 0)
3965 return -EINVAL;
3966 return md_alloc(0, buf);
3967 }
3968
3969 static void md_safemode_timeout(unsigned long data)
3970 {
3971 mddev_t *mddev = (mddev_t *) data;
3972
3973 if (!atomic_read(&mddev->writes_pending)) {
3974 mddev->safemode = 1;
3975 if (mddev->external)
3976 sysfs_notify_dirent(mddev->sysfs_state);
3977 }
3978 md_wakeup_thread(mddev->thread);
3979 }
3980
3981 static int start_dirty_degraded;
3982
3983 static int do_md_run(mddev_t * mddev)
3984 {
3985 int err;
3986 int chunk_size;
3987 mdk_rdev_t *rdev;
3988 struct gendisk *disk;
3989 struct mdk_personality *pers;
3990 char b[BDEVNAME_SIZE];
3991
3992 if (list_empty(&mddev->disks))
3993 /* cannot run an array with no devices.. */
3994 return -EINVAL;
3995
3996 if (mddev->pers)
3997 return -EBUSY;
3998
3999 /*
4000 * Analyze all RAID superblock(s)
4001 */
4002 if (!mddev->raid_disks) {
4003 if (!mddev->persistent)
4004 return -EINVAL;
4005 analyze_sbs(mddev);
4006 }
4007
4008 chunk_size = mddev->chunk_sectors << 9;
4009
4010 if (chunk_size) {
4011 if (chunk_size > MAX_CHUNK_SIZE) {
4012 printk(KERN_ERR "too big chunk_size: %d > %d\n",
4013 chunk_size, MAX_CHUNK_SIZE);
4014 return -EINVAL;
4015 }
4016 /* devices must have minimum size of one chunk */
4017 list_for_each_entry(rdev, &mddev->disks, same_set) {
4018 if (test_bit(Faulty, &rdev->flags))
4019 continue;
4020 if (rdev->sectors < chunk_size / 512) {
4021 printk(KERN_WARNING
4022 "md: Dev %s smaller than chunk_size:"
4023 " %llu < %d\n",
4024 bdevname(rdev->bdev,b),
4025 (unsigned long long)rdev->sectors,
4026 chunk_size / 512);
4027 return -EINVAL;
4028 }
4029 }
4030 }
4031
4032 if (mddev->level != LEVEL_NONE)
4033 request_module("md-level-%d", mddev->level);
4034 else if (mddev->clevel[0])
4035 request_module("md-%s", mddev->clevel);
4036
4037 /*
4038 * Drop all container device buffers, from now on
4039 * the only valid external interface is through the md
4040 * device.
4041 */
4042 list_for_each_entry(rdev, &mddev->disks, same_set) {
4043 if (test_bit(Faulty, &rdev->flags))
4044 continue;
4045 sync_blockdev(rdev->bdev);
4046 invalidate_bdev(rdev->bdev);
4047
4048 /* perform some consistency tests on the device.
4049 * We don't want the data to overlap the metadata,
4050 * Internal Bitmap issues have been handled elsewhere.
4051 */
4052 if (rdev->data_offset < rdev->sb_start) {
4053 if (mddev->dev_sectors &&
4054 rdev->data_offset + mddev->dev_sectors
4055 > rdev->sb_start) {
4056 printk("md: %s: data overlaps metadata\n",
4057 mdname(mddev));
4058 return -EINVAL;
4059 }
4060 } else {
4061 if (rdev->sb_start + rdev->sb_size/512
4062 > rdev->data_offset) {
4063 printk("md: %s: metadata overlaps data\n",
4064 mdname(mddev));
4065 return -EINVAL;
4066 }
4067 }
4068 sysfs_notify_dirent(rdev->sysfs_state);
4069 }
4070
4071 md_probe(mddev->unit, NULL, NULL);
4072 disk = mddev->gendisk;
4073 if (!disk)
4074 return -ENOMEM;
4075
4076 spin_lock(&pers_lock);
4077 pers = find_pers(mddev->level, mddev->clevel);
4078 if (!pers || !try_module_get(pers->owner)) {
4079 spin_unlock(&pers_lock);
4080 if (mddev->level != LEVEL_NONE)
4081 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4082 mddev->level);
4083 else
4084 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4085 mddev->clevel);
4086 return -EINVAL;
4087 }
4088 mddev->pers = pers;
4089 spin_unlock(&pers_lock);
4090 if (mddev->level != pers->level) {
4091 mddev->level = pers->level;
4092 mddev->new_level = pers->level;
4093 }
4094 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4095
4096 if (pers->level >= 4 && pers->level <= 6)
4097 /* Cannot support integrity (yet) */
4098 blk_integrity_unregister(mddev->gendisk);
4099
4100 if (mddev->reshape_position != MaxSector &&
4101 pers->start_reshape == NULL) {
4102 /* This personality cannot handle reshaping... */
4103 mddev->pers = NULL;
4104 module_put(pers->owner);
4105 return -EINVAL;
4106 }
4107
4108 if (pers->sync_request) {
4109 /* Warn if this is a potentially silly
4110 * configuration.
4111 */
4112 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4113 mdk_rdev_t *rdev2;
4114 int warned = 0;
4115
4116 list_for_each_entry(rdev, &mddev->disks, same_set)
4117 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4118 if (rdev < rdev2 &&
4119 rdev->bdev->bd_contains ==
4120 rdev2->bdev->bd_contains) {
4121 printk(KERN_WARNING
4122 "%s: WARNING: %s appears to be"
4123 " on the same physical disk as"
4124 " %s.\n",
4125 mdname(mddev),
4126 bdevname(rdev->bdev,b),
4127 bdevname(rdev2->bdev,b2));
4128 warned = 1;
4129 }
4130 }
4131
4132 if (warned)
4133 printk(KERN_WARNING
4134 "True protection against single-disk"
4135 " failure might be compromised.\n");
4136 }
4137
4138 mddev->recovery = 0;
4139 /* may be over-ridden by personality */
4140 mddev->resync_max_sectors = mddev->dev_sectors;
4141
4142 mddev->barriers_work = 1;
4143 mddev->ok_start_degraded = start_dirty_degraded;
4144
4145 if (start_readonly)
4146 mddev->ro = 2; /* read-only, but switch on first write */
4147
4148 err = mddev->pers->run(mddev);
4149 if (err)
4150 printk(KERN_ERR "md: pers->run() failed ...\n");
4151 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4152 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4153 " but 'external_size' not in effect?\n", __func__);
4154 printk(KERN_ERR
4155 "md: invalid array_size %llu > default size %llu\n",
4156 (unsigned long long)mddev->array_sectors / 2,
4157 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4158 err = -EINVAL;
4159 mddev->pers->stop(mddev);
4160 }
4161 if (err == 0 && mddev->pers->sync_request) {
4162 err = bitmap_create(mddev);
4163 if (err) {
4164 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4165 mdname(mddev), err);
4166 mddev->pers->stop(mddev);
4167 }
4168 }
4169 if (err) {
4170 module_put(mddev->pers->owner);
4171 mddev->pers = NULL;
4172 bitmap_destroy(mddev);
4173 return err;
4174 }
4175 if (mddev->pers->sync_request) {
4176 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4177 printk(KERN_WARNING
4178 "md: cannot register extra attributes for %s\n",
4179 mdname(mddev));
4180 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4181 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4182 mddev->ro = 0;
4183
4184 atomic_set(&mddev->writes_pending,0);
4185 mddev->safemode = 0;
4186 mddev->safemode_timer.function = md_safemode_timeout;
4187 mddev->safemode_timer.data = (unsigned long) mddev;
4188 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4189 mddev->in_sync = 1;
4190
4191 list_for_each_entry(rdev, &mddev->disks, same_set)
4192 if (rdev->raid_disk >= 0) {
4193 char nm[20];
4194 sprintf(nm, "rd%d", rdev->raid_disk);
4195 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4196 printk("md: cannot register %s for %s\n",
4197 nm, mdname(mddev));
4198 }
4199
4200 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4201
4202 if (mddev->flags)
4203 md_update_sb(mddev, 0);
4204
4205 set_capacity(disk, mddev->array_sectors);
4206
4207 /* If there is a partially-recovered drive we need to
4208 * start recovery here. If we leave it to md_check_recovery,
4209 * it will remove the drives and not do the right thing
4210 */
4211 if (mddev->degraded && !mddev->sync_thread) {
4212 int spares = 0;
4213 list_for_each_entry(rdev, &mddev->disks, same_set)
4214 if (rdev->raid_disk >= 0 &&
4215 !test_bit(In_sync, &rdev->flags) &&
4216 !test_bit(Faulty, &rdev->flags))
4217 /* complete an interrupted recovery */
4218 spares++;
4219 if (spares && mddev->pers->sync_request) {
4220 mddev->recovery = 0;
4221 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4222 mddev->sync_thread = md_register_thread(md_do_sync,
4223 mddev,
4224 "%s_resync");
4225 if (!mddev->sync_thread) {
4226 printk(KERN_ERR "%s: could not start resync"
4227 " thread...\n",
4228 mdname(mddev));
4229 /* leave the spares where they are, it shouldn't hurt */
4230 mddev->recovery = 0;
4231 }
4232 }
4233 }
4234 md_wakeup_thread(mddev->thread);
4235 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4236
4237 mddev->changed = 1;
4238 md_new_event(mddev);
4239 sysfs_notify_dirent(mddev->sysfs_state);
4240 if (mddev->sysfs_action)
4241 sysfs_notify_dirent(mddev->sysfs_action);
4242 sysfs_notify(&mddev->kobj, NULL, "degraded");
4243 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4244 return 0;
4245 }
4246
4247 static int restart_array(mddev_t *mddev)
4248 {
4249 struct gendisk *disk = mddev->gendisk;
4250
4251 /* Complain if it has no devices */
4252 if (list_empty(&mddev->disks))
4253 return -ENXIO;
4254 if (!mddev->pers)
4255 return -EINVAL;
4256 if (!mddev->ro)
4257 return -EBUSY;
4258 mddev->safemode = 0;
4259 mddev->ro = 0;
4260 set_disk_ro(disk, 0);
4261 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4262 mdname(mddev));
4263 /* Kick recovery or resync if necessary */
4264 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4265 md_wakeup_thread(mddev->thread);
4266 md_wakeup_thread(mddev->sync_thread);
4267 sysfs_notify_dirent(mddev->sysfs_state);
4268 return 0;
4269 }
4270
4271 /* similar to deny_write_access, but accounts for our holding a reference
4272 * to the file ourselves */
4273 static int deny_bitmap_write_access(struct file * file)
4274 {
4275 struct inode *inode = file->f_mapping->host;
4276
4277 spin_lock(&inode->i_lock);
4278 if (atomic_read(&inode->i_writecount) > 1) {
4279 spin_unlock(&inode->i_lock);
4280 return -ETXTBSY;
4281 }
4282 atomic_set(&inode->i_writecount, -1);
4283 spin_unlock(&inode->i_lock);
4284
4285 return 0;
4286 }
4287
4288 static void restore_bitmap_write_access(struct file *file)
4289 {
4290 struct inode *inode = file->f_mapping->host;
4291
4292 spin_lock(&inode->i_lock);
4293 atomic_set(&inode->i_writecount, 1);
4294 spin_unlock(&inode->i_lock);
4295 }
4296
4297 /* mode:
4298 * 0 - completely stop and dis-assemble array
4299 * 1 - switch to readonly
4300 * 2 - stop but do not disassemble array
4301 */
4302 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4303 {
4304 int err = 0;
4305 struct gendisk *disk = mddev->gendisk;
4306 mdk_rdev_t *rdev;
4307
4308 if (atomic_read(&mddev->openers) > is_open) {
4309 printk("md: %s still in use.\n",mdname(mddev));
4310 return -EBUSY;
4311 }
4312
4313 if (mddev->pers) {
4314
4315 if (mddev->sync_thread) {
4316 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4318 md_unregister_thread(mddev->sync_thread);
4319 mddev->sync_thread = NULL;
4320 }
4321
4322 del_timer_sync(&mddev->safemode_timer);
4323
4324 switch(mode) {
4325 case 1: /* readonly */
4326 err = -ENXIO;
4327 if (mddev->ro==1)
4328 goto out;
4329 mddev->ro = 1;
4330 break;
4331 case 0: /* disassemble */
4332 case 2: /* stop */
4333 bitmap_flush(mddev);
4334 md_super_wait(mddev);
4335 if (mddev->ro)
4336 set_disk_ro(disk, 0);
4337
4338 mddev->pers->stop(mddev);
4339 mddev->queue->merge_bvec_fn = NULL;
4340 mddev->queue->unplug_fn = NULL;
4341 mddev->queue->backing_dev_info.congested_fn = NULL;
4342 module_put(mddev->pers->owner);
4343 if (mddev->pers->sync_request)
4344 mddev->private = &md_redundancy_group;
4345 mddev->pers = NULL;
4346 /* tell userspace to handle 'inactive' */
4347 sysfs_notify_dirent(mddev->sysfs_state);
4348
4349 list_for_each_entry(rdev, &mddev->disks, same_set)
4350 if (rdev->raid_disk >= 0) {
4351 char nm[20];
4352 sprintf(nm, "rd%d", rdev->raid_disk);
4353 sysfs_remove_link(&mddev->kobj, nm);
4354 }
4355
4356 set_capacity(disk, 0);
4357 mddev->changed = 1;
4358
4359 if (mddev->ro)
4360 mddev->ro = 0;
4361 }
4362 if (!mddev->in_sync || mddev->flags) {
4363 /* mark array as shutdown cleanly */
4364 mddev->in_sync = 1;
4365 md_update_sb(mddev, 1);
4366 }
4367 if (mode == 1)
4368 set_disk_ro(disk, 1);
4369 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4370 }
4371
4372 /*
4373 * Free resources if final stop
4374 */
4375 if (mode == 0) {
4376
4377 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4378
4379 bitmap_destroy(mddev);
4380 if (mddev->bitmap_file) {
4381 restore_bitmap_write_access(mddev->bitmap_file);
4382 fput(mddev->bitmap_file);
4383 mddev->bitmap_file = NULL;
4384 }
4385 mddev->bitmap_offset = 0;
4386
4387 /* make sure all md_delayed_delete calls have finished */
4388 flush_scheduled_work();
4389
4390 export_array(mddev);
4391
4392 mddev->array_sectors = 0;
4393 mddev->external_size = 0;
4394 mddev->dev_sectors = 0;
4395 mddev->raid_disks = 0;
4396 mddev->recovery_cp = 0;
4397 mddev->resync_min = 0;
4398 mddev->resync_max = MaxSector;
4399 mddev->reshape_position = MaxSector;
4400 mddev->external = 0;
4401 mddev->persistent = 0;
4402 mddev->level = LEVEL_NONE;
4403 mddev->clevel[0] = 0;
4404 mddev->flags = 0;
4405 mddev->ro = 0;
4406 mddev->metadata_type[0] = 0;
4407 mddev->chunk_sectors = 0;
4408 mddev->ctime = mddev->utime = 0;
4409 mddev->layout = 0;
4410 mddev->max_disks = 0;
4411 mddev->events = 0;
4412 mddev->delta_disks = 0;
4413 mddev->new_level = LEVEL_NONE;
4414 mddev->new_layout = 0;
4415 mddev->new_chunk_sectors = 0;
4416 mddev->curr_resync = 0;
4417 mddev->resync_mismatches = 0;
4418 mddev->suspend_lo = mddev->suspend_hi = 0;
4419 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4420 mddev->recovery = 0;
4421 mddev->in_sync = 0;
4422 mddev->changed = 0;
4423 mddev->degraded = 0;
4424 mddev->barriers_work = 0;
4425 mddev->safemode = 0;
4426 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4427 if (mddev->hold_active == UNTIL_STOP)
4428 mddev->hold_active = 0;
4429
4430 } else if (mddev->pers)
4431 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4432 mdname(mddev));
4433 err = 0;
4434 blk_integrity_unregister(disk);
4435 md_new_event(mddev);
4436 sysfs_notify_dirent(mddev->sysfs_state);
4437 out:
4438 return err;
4439 }
4440
4441 #ifndef MODULE
4442 static void autorun_array(mddev_t *mddev)
4443 {
4444 mdk_rdev_t *rdev;
4445 int err;
4446
4447 if (list_empty(&mddev->disks))
4448 return;
4449
4450 printk(KERN_INFO "md: running: ");
4451
4452 list_for_each_entry(rdev, &mddev->disks, same_set) {
4453 char b[BDEVNAME_SIZE];
4454 printk("<%s>", bdevname(rdev->bdev,b));
4455 }
4456 printk("\n");
4457
4458 err = do_md_run(mddev);
4459 if (err) {
4460 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4461 do_md_stop(mddev, 0, 0);
4462 }
4463 }
4464
4465 /*
4466 * lets try to run arrays based on all disks that have arrived
4467 * until now. (those are in pending_raid_disks)
4468 *
4469 * the method: pick the first pending disk, collect all disks with
4470 * the same UUID, remove all from the pending list and put them into
4471 * the 'same_array' list. Then order this list based on superblock
4472 * update time (freshest comes first), kick out 'old' disks and
4473 * compare superblocks. If everything's fine then run it.
4474 *
4475 * If "unit" is allocated, then bump its reference count
4476 */
4477 static void autorun_devices(int part)
4478 {
4479 mdk_rdev_t *rdev0, *rdev, *tmp;
4480 mddev_t *mddev;
4481 char b[BDEVNAME_SIZE];
4482
4483 printk(KERN_INFO "md: autorun ...\n");
4484 while (!list_empty(&pending_raid_disks)) {
4485 int unit;
4486 dev_t dev;
4487 LIST_HEAD(candidates);
4488 rdev0 = list_entry(pending_raid_disks.next,
4489 mdk_rdev_t, same_set);
4490
4491 printk(KERN_INFO "md: considering %s ...\n",
4492 bdevname(rdev0->bdev,b));
4493 INIT_LIST_HEAD(&candidates);
4494 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4495 if (super_90_load(rdev, rdev0, 0) >= 0) {
4496 printk(KERN_INFO "md: adding %s ...\n",
4497 bdevname(rdev->bdev,b));
4498 list_move(&rdev->same_set, &candidates);
4499 }
4500 /*
4501 * now we have a set of devices, with all of them having
4502 * mostly sane superblocks. It's time to allocate the
4503 * mddev.
4504 */
4505 if (part) {
4506 dev = MKDEV(mdp_major,
4507 rdev0->preferred_minor << MdpMinorShift);
4508 unit = MINOR(dev) >> MdpMinorShift;
4509 } else {
4510 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4511 unit = MINOR(dev);
4512 }
4513 if (rdev0->preferred_minor != unit) {
4514 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4515 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4516 break;
4517 }
4518
4519 md_probe(dev, NULL, NULL);
4520 mddev = mddev_find(dev);
4521 if (!mddev || !mddev->gendisk) {
4522 if (mddev)
4523 mddev_put(mddev);
4524 printk(KERN_ERR
4525 "md: cannot allocate memory for md drive.\n");
4526 break;
4527 }
4528 if (mddev_lock(mddev))
4529 printk(KERN_WARNING "md: %s locked, cannot run\n",
4530 mdname(mddev));
4531 else if (mddev->raid_disks || mddev->major_version
4532 || !list_empty(&mddev->disks)) {
4533 printk(KERN_WARNING
4534 "md: %s already running, cannot run %s\n",
4535 mdname(mddev), bdevname(rdev0->bdev,b));
4536 mddev_unlock(mddev);
4537 } else {
4538 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4539 mddev->persistent = 1;
4540 rdev_for_each_list(rdev, tmp, &candidates) {
4541 list_del_init(&rdev->same_set);
4542 if (bind_rdev_to_array(rdev, mddev))
4543 export_rdev(rdev);
4544 }
4545 autorun_array(mddev);
4546 mddev_unlock(mddev);
4547 }
4548 /* on success, candidates will be empty, on error
4549 * it won't...
4550 */
4551 rdev_for_each_list(rdev, tmp, &candidates) {
4552 list_del_init(&rdev->same_set);
4553 export_rdev(rdev);
4554 }
4555 mddev_put(mddev);
4556 }
4557 printk(KERN_INFO "md: ... autorun DONE.\n");
4558 }
4559 #endif /* !MODULE */
4560
4561 static int get_version(void __user * arg)
4562 {
4563 mdu_version_t ver;
4564
4565 ver.major = MD_MAJOR_VERSION;
4566 ver.minor = MD_MINOR_VERSION;
4567 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4568
4569 if (copy_to_user(arg, &ver, sizeof(ver)))
4570 return -EFAULT;
4571
4572 return 0;
4573 }
4574
4575 static int get_array_info(mddev_t * mddev, void __user * arg)
4576 {
4577 mdu_array_info_t info;
4578 int nr,working,active,failed,spare;
4579 mdk_rdev_t *rdev;
4580
4581 nr=working=active=failed=spare=0;
4582 list_for_each_entry(rdev, &mddev->disks, same_set) {
4583 nr++;
4584 if (test_bit(Faulty, &rdev->flags))
4585 failed++;
4586 else {
4587 working++;
4588 if (test_bit(In_sync, &rdev->flags))
4589 active++;
4590 else
4591 spare++;
4592 }
4593 }
4594
4595 info.major_version = mddev->major_version;
4596 info.minor_version = mddev->minor_version;
4597 info.patch_version = MD_PATCHLEVEL_VERSION;
4598 info.ctime = mddev->ctime;
4599 info.level = mddev->level;
4600 info.size = mddev->dev_sectors / 2;
4601 if (info.size != mddev->dev_sectors / 2) /* overflow */
4602 info.size = -1;
4603 info.nr_disks = nr;
4604 info.raid_disks = mddev->raid_disks;
4605 info.md_minor = mddev->md_minor;
4606 info.not_persistent= !mddev->persistent;
4607
4608 info.utime = mddev->utime;
4609 info.state = 0;
4610 if (mddev->in_sync)
4611 info.state = (1<<MD_SB_CLEAN);
4612 if (mddev->bitmap && mddev->bitmap_offset)
4613 info.state = (1<<MD_SB_BITMAP_PRESENT);
4614 info.active_disks = active;
4615 info.working_disks = working;
4616 info.failed_disks = failed;
4617 info.spare_disks = spare;
4618
4619 info.layout = mddev->layout;
4620 info.chunk_size = mddev->chunk_sectors << 9;
4621
4622 if (copy_to_user(arg, &info, sizeof(info)))
4623 return -EFAULT;
4624
4625 return 0;
4626 }
4627
4628 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4629 {
4630 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4631 char *ptr, *buf = NULL;
4632 int err = -ENOMEM;
4633
4634 if (md_allow_write(mddev))
4635 file = kmalloc(sizeof(*file), GFP_NOIO);
4636 else
4637 file = kmalloc(sizeof(*file), GFP_KERNEL);
4638
4639 if (!file)
4640 goto out;
4641
4642 /* bitmap disabled, zero the first byte and copy out */
4643 if (!mddev->bitmap || !mddev->bitmap->file) {
4644 file->pathname[0] = '\0';
4645 goto copy_out;
4646 }
4647
4648 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4649 if (!buf)
4650 goto out;
4651
4652 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4653 if (IS_ERR(ptr))
4654 goto out;
4655
4656 strcpy(file->pathname, ptr);
4657
4658 copy_out:
4659 err = 0;
4660 if (copy_to_user(arg, file, sizeof(*file)))
4661 err = -EFAULT;
4662 out:
4663 kfree(buf);
4664 kfree(file);
4665 return err;
4666 }
4667
4668 static int get_disk_info(mddev_t * mddev, void __user * arg)
4669 {
4670 mdu_disk_info_t info;
4671 mdk_rdev_t *rdev;
4672
4673 if (copy_from_user(&info, arg, sizeof(info)))
4674 return -EFAULT;
4675
4676 rdev = find_rdev_nr(mddev, info.number);
4677 if (rdev) {
4678 info.major = MAJOR(rdev->bdev->bd_dev);
4679 info.minor = MINOR(rdev->bdev->bd_dev);
4680 info.raid_disk = rdev->raid_disk;
4681 info.state = 0;
4682 if (test_bit(Faulty, &rdev->flags))
4683 info.state |= (1<<MD_DISK_FAULTY);
4684 else if (test_bit(In_sync, &rdev->flags)) {
4685 info.state |= (1<<MD_DISK_ACTIVE);
4686 info.state |= (1<<MD_DISK_SYNC);
4687 }
4688 if (test_bit(WriteMostly, &rdev->flags))
4689 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4690 } else {
4691 info.major = info.minor = 0;
4692 info.raid_disk = -1;
4693 info.state = (1<<MD_DISK_REMOVED);
4694 }
4695
4696 if (copy_to_user(arg, &info, sizeof(info)))
4697 return -EFAULT;
4698
4699 return 0;
4700 }
4701
4702 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4703 {
4704 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4705 mdk_rdev_t *rdev;
4706 dev_t dev = MKDEV(info->major,info->minor);
4707
4708 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4709 return -EOVERFLOW;
4710
4711 if (!mddev->raid_disks) {
4712 int err;
4713 /* expecting a device which has a superblock */
4714 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4715 if (IS_ERR(rdev)) {
4716 printk(KERN_WARNING
4717 "md: md_import_device returned %ld\n",
4718 PTR_ERR(rdev));
4719 return PTR_ERR(rdev);
4720 }
4721 if (!list_empty(&mddev->disks)) {
4722 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4723 mdk_rdev_t, same_set);
4724 int err = super_types[mddev->major_version]
4725 .load_super(rdev, rdev0, mddev->minor_version);
4726 if (err < 0) {
4727 printk(KERN_WARNING
4728 "md: %s has different UUID to %s\n",
4729 bdevname(rdev->bdev,b),
4730 bdevname(rdev0->bdev,b2));
4731 export_rdev(rdev);
4732 return -EINVAL;
4733 }
4734 }
4735 err = bind_rdev_to_array(rdev, mddev);
4736 if (err)
4737 export_rdev(rdev);
4738 return err;
4739 }
4740
4741 /*
4742 * add_new_disk can be used once the array is assembled
4743 * to add "hot spares". They must already have a superblock
4744 * written
4745 */
4746 if (mddev->pers) {
4747 int err;
4748 if (!mddev->pers->hot_add_disk) {
4749 printk(KERN_WARNING
4750 "%s: personality does not support diskops!\n",
4751 mdname(mddev));
4752 return -EINVAL;
4753 }
4754 if (mddev->persistent)
4755 rdev = md_import_device(dev, mddev->major_version,
4756 mddev->minor_version);
4757 else
4758 rdev = md_import_device(dev, -1, -1);
4759 if (IS_ERR(rdev)) {
4760 printk(KERN_WARNING
4761 "md: md_import_device returned %ld\n",
4762 PTR_ERR(rdev));
4763 return PTR_ERR(rdev);
4764 }
4765 /* set save_raid_disk if appropriate */
4766 if (!mddev->persistent) {
4767 if (info->state & (1<<MD_DISK_SYNC) &&
4768 info->raid_disk < mddev->raid_disks)
4769 rdev->raid_disk = info->raid_disk;
4770 else
4771 rdev->raid_disk = -1;
4772 } else
4773 super_types[mddev->major_version].
4774 validate_super(mddev, rdev);
4775 rdev->saved_raid_disk = rdev->raid_disk;
4776
4777 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4778 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4779 set_bit(WriteMostly, &rdev->flags);
4780 else
4781 clear_bit(WriteMostly, &rdev->flags);
4782
4783 rdev->raid_disk = -1;
4784 err = bind_rdev_to_array(rdev, mddev);
4785 if (!err && !mddev->pers->hot_remove_disk) {
4786 /* If there is hot_add_disk but no hot_remove_disk
4787 * then added disks for geometry changes,
4788 * and should be added immediately.
4789 */
4790 super_types[mddev->major_version].
4791 validate_super(mddev, rdev);
4792 err = mddev->pers->hot_add_disk(mddev, rdev);
4793 if (err)
4794 unbind_rdev_from_array(rdev);
4795 }
4796 if (err)
4797 export_rdev(rdev);
4798 else
4799 sysfs_notify_dirent(rdev->sysfs_state);
4800
4801 md_update_sb(mddev, 1);
4802 if (mddev->degraded)
4803 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4804 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4805 md_wakeup_thread(mddev->thread);
4806 return err;
4807 }
4808
4809 /* otherwise, add_new_disk is only allowed
4810 * for major_version==0 superblocks
4811 */
4812 if (mddev->major_version != 0) {
4813 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4814 mdname(mddev));
4815 return -EINVAL;
4816 }
4817
4818 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4819 int err;
4820 rdev = md_import_device(dev, -1, 0);
4821 if (IS_ERR(rdev)) {
4822 printk(KERN_WARNING
4823 "md: error, md_import_device() returned %ld\n",
4824 PTR_ERR(rdev));
4825 return PTR_ERR(rdev);
4826 }
4827 rdev->desc_nr = info->number;
4828 if (info->raid_disk < mddev->raid_disks)
4829 rdev->raid_disk = info->raid_disk;
4830 else
4831 rdev->raid_disk = -1;
4832
4833 if (rdev->raid_disk < mddev->raid_disks)
4834 if (info->state & (1<<MD_DISK_SYNC))
4835 set_bit(In_sync, &rdev->flags);
4836
4837 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4838 set_bit(WriteMostly, &rdev->flags);
4839
4840 if (!mddev->persistent) {
4841 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4842 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4843 } else
4844 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4845 rdev->sectors = calc_num_sectors(rdev,
4846 mddev->chunk_sectors << 9);
4847
4848 err = bind_rdev_to_array(rdev, mddev);
4849 if (err) {
4850 export_rdev(rdev);
4851 return err;
4852 }
4853 }
4854
4855 return 0;
4856 }
4857
4858 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4859 {
4860 char b[BDEVNAME_SIZE];
4861 mdk_rdev_t *rdev;
4862
4863 rdev = find_rdev(mddev, dev);
4864 if (!rdev)
4865 return -ENXIO;
4866
4867 if (rdev->raid_disk >= 0)
4868 goto busy;
4869
4870 kick_rdev_from_array(rdev);
4871 md_update_sb(mddev, 1);
4872 md_new_event(mddev);
4873
4874 return 0;
4875 busy:
4876 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4877 bdevname(rdev->bdev,b), mdname(mddev));
4878 return -EBUSY;
4879 }
4880
4881 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4882 {
4883 char b[BDEVNAME_SIZE];
4884 int err;
4885 mdk_rdev_t *rdev;
4886
4887 if (!mddev->pers)
4888 return -ENODEV;
4889
4890 if (mddev->major_version != 0) {
4891 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4892 " version-0 superblocks.\n",
4893 mdname(mddev));
4894 return -EINVAL;
4895 }
4896 if (!mddev->pers->hot_add_disk) {
4897 printk(KERN_WARNING
4898 "%s: personality does not support diskops!\n",
4899 mdname(mddev));
4900 return -EINVAL;
4901 }
4902
4903 rdev = md_import_device(dev, -1, 0);
4904 if (IS_ERR(rdev)) {
4905 printk(KERN_WARNING
4906 "md: error, md_import_device() returned %ld\n",
4907 PTR_ERR(rdev));
4908 return -EINVAL;
4909 }
4910
4911 if (mddev->persistent)
4912 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4913 else
4914 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4915
4916 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_sectors << 9);
4917
4918 if (test_bit(Faulty, &rdev->flags)) {
4919 printk(KERN_WARNING
4920 "md: can not hot-add faulty %s disk to %s!\n",
4921 bdevname(rdev->bdev,b), mdname(mddev));
4922 err = -EINVAL;
4923 goto abort_export;
4924 }
4925 clear_bit(In_sync, &rdev->flags);
4926 rdev->desc_nr = -1;
4927 rdev->saved_raid_disk = -1;
4928 err = bind_rdev_to_array(rdev, mddev);
4929 if (err)
4930 goto abort_export;
4931
4932 /*
4933 * The rest should better be atomic, we can have disk failures
4934 * noticed in interrupt contexts ...
4935 */
4936
4937 rdev->raid_disk = -1;
4938
4939 md_update_sb(mddev, 1);
4940
4941 /*
4942 * Kick recovery, maybe this spare has to be added to the
4943 * array immediately.
4944 */
4945 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4946 md_wakeup_thread(mddev->thread);
4947 md_new_event(mddev);
4948 return 0;
4949
4950 abort_export:
4951 export_rdev(rdev);
4952 return err;
4953 }
4954
4955 static int set_bitmap_file(mddev_t *mddev, int fd)
4956 {
4957 int err;
4958
4959 if (mddev->pers) {
4960 if (!mddev->pers->quiesce)
4961 return -EBUSY;
4962 if (mddev->recovery || mddev->sync_thread)
4963 return -EBUSY;
4964 /* we should be able to change the bitmap.. */
4965 }
4966
4967
4968 if (fd >= 0) {
4969 if (mddev->bitmap)
4970 return -EEXIST; /* cannot add when bitmap is present */
4971 mddev->bitmap_file = fget(fd);
4972
4973 if (mddev->bitmap_file == NULL) {
4974 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4975 mdname(mddev));
4976 return -EBADF;
4977 }
4978
4979 err = deny_bitmap_write_access(mddev->bitmap_file);
4980 if (err) {
4981 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4982 mdname(mddev));
4983 fput(mddev->bitmap_file);
4984 mddev->bitmap_file = NULL;
4985 return err;
4986 }
4987 mddev->bitmap_offset = 0; /* file overrides offset */
4988 } else if (mddev->bitmap == NULL)
4989 return -ENOENT; /* cannot remove what isn't there */
4990 err = 0;
4991 if (mddev->pers) {
4992 mddev->pers->quiesce(mddev, 1);
4993 if (fd >= 0)
4994 err = bitmap_create(mddev);
4995 if (fd < 0 || err) {
4996 bitmap_destroy(mddev);
4997 fd = -1; /* make sure to put the file */
4998 }
4999 mddev->pers->quiesce(mddev, 0);
5000 }
5001 if (fd < 0) {
5002 if (mddev->bitmap_file) {
5003 restore_bitmap_write_access(mddev->bitmap_file);
5004 fput(mddev->bitmap_file);
5005 }
5006 mddev->bitmap_file = NULL;
5007 }
5008
5009 return err;
5010 }
5011
5012 /*
5013 * set_array_info is used two different ways
5014 * The original usage is when creating a new array.
5015 * In this usage, raid_disks is > 0 and it together with
5016 * level, size, not_persistent,layout,chunksize determine the
5017 * shape of the array.
5018 * This will always create an array with a type-0.90.0 superblock.
5019 * The newer usage is when assembling an array.
5020 * In this case raid_disks will be 0, and the major_version field is
5021 * use to determine which style super-blocks are to be found on the devices.
5022 * The minor and patch _version numbers are also kept incase the
5023 * super_block handler wishes to interpret them.
5024 */
5025 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5026 {
5027
5028 if (info->raid_disks == 0) {
5029 /* just setting version number for superblock loading */
5030 if (info->major_version < 0 ||
5031 info->major_version >= ARRAY_SIZE(super_types) ||
5032 super_types[info->major_version].name == NULL) {
5033 /* maybe try to auto-load a module? */
5034 printk(KERN_INFO
5035 "md: superblock version %d not known\n",
5036 info->major_version);
5037 return -EINVAL;
5038 }
5039 mddev->major_version = info->major_version;
5040 mddev->minor_version = info->minor_version;
5041 mddev->patch_version = info->patch_version;
5042 mddev->persistent = !info->not_persistent;
5043 return 0;
5044 }
5045 mddev->major_version = MD_MAJOR_VERSION;
5046 mddev->minor_version = MD_MINOR_VERSION;
5047 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5048 mddev->ctime = get_seconds();
5049
5050 mddev->level = info->level;
5051 mddev->clevel[0] = 0;
5052 mddev->dev_sectors = 2 * (sector_t)info->size;
5053 mddev->raid_disks = info->raid_disks;
5054 /* don't set md_minor, it is determined by which /dev/md* was
5055 * openned
5056 */
5057 if (info->state & (1<<MD_SB_CLEAN))
5058 mddev->recovery_cp = MaxSector;
5059 else
5060 mddev->recovery_cp = 0;
5061 mddev->persistent = ! info->not_persistent;
5062 mddev->external = 0;
5063
5064 mddev->layout = info->layout;
5065 mddev->chunk_sectors = info->chunk_size >> 9;
5066
5067 mddev->max_disks = MD_SB_DISKS;
5068
5069 if (mddev->persistent)
5070 mddev->flags = 0;
5071 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5072
5073 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5074 mddev->bitmap_offset = 0;
5075
5076 mddev->reshape_position = MaxSector;
5077
5078 /*
5079 * Generate a 128 bit UUID
5080 */
5081 get_random_bytes(mddev->uuid, 16);
5082
5083 mddev->new_level = mddev->level;
5084 mddev->new_chunk_sectors = mddev->chunk_sectors;
5085 mddev->new_layout = mddev->layout;
5086 mddev->delta_disks = 0;
5087
5088 return 0;
5089 }
5090
5091 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5092 {
5093 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5094
5095 if (mddev->external_size)
5096 return;
5097
5098 mddev->array_sectors = array_sectors;
5099 }
5100 EXPORT_SYMBOL(md_set_array_sectors);
5101
5102 static int update_size(mddev_t *mddev, sector_t num_sectors)
5103 {
5104 mdk_rdev_t *rdev;
5105 int rv;
5106 int fit = (num_sectors == 0);
5107
5108 if (mddev->pers->resize == NULL)
5109 return -EINVAL;
5110 /* The "num_sectors" is the number of sectors of each device that
5111 * is used. This can only make sense for arrays with redundancy.
5112 * linear and raid0 always use whatever space is available. We can only
5113 * consider changing this number if no resync or reconstruction is
5114 * happening, and if the new size is acceptable. It must fit before the
5115 * sb_start or, if that is <data_offset, it must fit before the size
5116 * of each device. If num_sectors is zero, we find the largest size
5117 * that fits.
5118
5119 */
5120 if (mddev->sync_thread)
5121 return -EBUSY;
5122 if (mddev->bitmap)
5123 /* Sorry, cannot grow a bitmap yet, just remove it,
5124 * grow, and re-add.
5125 */
5126 return -EBUSY;
5127 list_for_each_entry(rdev, &mddev->disks, same_set) {
5128 sector_t avail = rdev->sectors;
5129
5130 if (fit && (num_sectors == 0 || num_sectors > avail))
5131 num_sectors = avail;
5132 if (avail < num_sectors)
5133 return -ENOSPC;
5134 }
5135 rv = mddev->pers->resize(mddev, num_sectors);
5136 if (!rv) {
5137 struct block_device *bdev;
5138
5139 bdev = bdget_disk(mddev->gendisk, 0);
5140 if (bdev) {
5141 mutex_lock(&bdev->bd_inode->i_mutex);
5142 i_size_write(bdev->bd_inode,
5143 (loff_t)mddev->array_sectors << 9);
5144 mutex_unlock(&bdev->bd_inode->i_mutex);
5145 bdput(bdev);
5146 }
5147 }
5148 return rv;
5149 }
5150
5151 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5152 {
5153 int rv;
5154 /* change the number of raid disks */
5155 if (mddev->pers->check_reshape == NULL)
5156 return -EINVAL;
5157 if (raid_disks <= 0 ||
5158 raid_disks >= mddev->max_disks)
5159 return -EINVAL;
5160 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5161 return -EBUSY;
5162 mddev->delta_disks = raid_disks - mddev->raid_disks;
5163
5164 rv = mddev->pers->check_reshape(mddev);
5165 return rv;
5166 }
5167
5168
5169 /*
5170 * update_array_info is used to change the configuration of an
5171 * on-line array.
5172 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5173 * fields in the info are checked against the array.
5174 * Any differences that cannot be handled will cause an error.
5175 * Normally, only one change can be managed at a time.
5176 */
5177 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5178 {
5179 int rv = 0;
5180 int cnt = 0;
5181 int state = 0;
5182
5183 /* calculate expected state,ignoring low bits */
5184 if (mddev->bitmap && mddev->bitmap_offset)
5185 state |= (1 << MD_SB_BITMAP_PRESENT);
5186
5187 if (mddev->major_version != info->major_version ||
5188 mddev->minor_version != info->minor_version ||
5189 /* mddev->patch_version != info->patch_version || */
5190 mddev->ctime != info->ctime ||
5191 mddev->level != info->level ||
5192 /* mddev->layout != info->layout || */
5193 !mddev->persistent != info->not_persistent||
5194 mddev->chunk_sectors != info->chunk_size >> 9 ||
5195 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5196 ((state^info->state) & 0xfffffe00)
5197 )
5198 return -EINVAL;
5199 /* Check there is only one change */
5200 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5201 cnt++;
5202 if (mddev->raid_disks != info->raid_disks)
5203 cnt++;
5204 if (mddev->layout != info->layout)
5205 cnt++;
5206 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5207 cnt++;
5208 if (cnt == 0)
5209 return 0;
5210 if (cnt > 1)
5211 return -EINVAL;
5212
5213 if (mddev->layout != info->layout) {
5214 /* Change layout
5215 * we don't need to do anything at the md level, the
5216 * personality will take care of it all.
5217 */
5218 if (mddev->pers->check_reshape == NULL)
5219 return -EINVAL;
5220 else {
5221 mddev->new_layout = info->layout;
5222 rv = mddev->pers->check_reshape(mddev);
5223 if (rv)
5224 mddev->new_layout = mddev->layout;
5225 return rv;
5226 }
5227 }
5228 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5229 rv = update_size(mddev, (sector_t)info->size * 2);
5230
5231 if (mddev->raid_disks != info->raid_disks)
5232 rv = update_raid_disks(mddev, info->raid_disks);
5233
5234 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5235 if (mddev->pers->quiesce == NULL)
5236 return -EINVAL;
5237 if (mddev->recovery || mddev->sync_thread)
5238 return -EBUSY;
5239 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5240 /* add the bitmap */
5241 if (mddev->bitmap)
5242 return -EEXIST;
5243 if (mddev->default_bitmap_offset == 0)
5244 return -EINVAL;
5245 mddev->bitmap_offset = mddev->default_bitmap_offset;
5246 mddev->pers->quiesce(mddev, 1);
5247 rv = bitmap_create(mddev);
5248 if (rv)
5249 bitmap_destroy(mddev);
5250 mddev->pers->quiesce(mddev, 0);
5251 } else {
5252 /* remove the bitmap */
5253 if (!mddev->bitmap)
5254 return -ENOENT;
5255 if (mddev->bitmap->file)
5256 return -EINVAL;
5257 mddev->pers->quiesce(mddev, 1);
5258 bitmap_destroy(mddev);
5259 mddev->pers->quiesce(mddev, 0);
5260 mddev->bitmap_offset = 0;
5261 }
5262 }
5263 md_update_sb(mddev, 1);
5264 return rv;
5265 }
5266
5267 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5268 {
5269 mdk_rdev_t *rdev;
5270
5271 if (mddev->pers == NULL)
5272 return -ENODEV;
5273
5274 rdev = find_rdev(mddev, dev);
5275 if (!rdev)
5276 return -ENODEV;
5277
5278 md_error(mddev, rdev);
5279 return 0;
5280 }
5281
5282 /*
5283 * We have a problem here : there is no easy way to give a CHS
5284 * virtual geometry. We currently pretend that we have a 2 heads
5285 * 4 sectors (with a BIG number of cylinders...). This drives
5286 * dosfs just mad... ;-)
5287 */
5288 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5289 {
5290 mddev_t *mddev = bdev->bd_disk->private_data;
5291
5292 geo->heads = 2;
5293 geo->sectors = 4;
5294 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5295 return 0;
5296 }
5297
5298 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5299 unsigned int cmd, unsigned long arg)
5300 {
5301 int err = 0;
5302 void __user *argp = (void __user *)arg;
5303 mddev_t *mddev = NULL;
5304
5305 if (!capable(CAP_SYS_ADMIN))
5306 return -EACCES;
5307
5308 /*
5309 * Commands dealing with the RAID driver but not any
5310 * particular array:
5311 */
5312 switch (cmd)
5313 {
5314 case RAID_VERSION:
5315 err = get_version(argp);
5316 goto done;
5317
5318 case PRINT_RAID_DEBUG:
5319 err = 0;
5320 md_print_devices();
5321 goto done;
5322
5323 #ifndef MODULE
5324 case RAID_AUTORUN:
5325 err = 0;
5326 autostart_arrays(arg);
5327 goto done;
5328 #endif
5329 default:;
5330 }
5331
5332 /*
5333 * Commands creating/starting a new array:
5334 */
5335
5336 mddev = bdev->bd_disk->private_data;
5337
5338 if (!mddev) {
5339 BUG();
5340 goto abort;
5341 }
5342
5343 err = mddev_lock(mddev);
5344 if (err) {
5345 printk(KERN_INFO
5346 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5347 err, cmd);
5348 goto abort;
5349 }
5350
5351 switch (cmd)
5352 {
5353 case SET_ARRAY_INFO:
5354 {
5355 mdu_array_info_t info;
5356 if (!arg)
5357 memset(&info, 0, sizeof(info));
5358 else if (copy_from_user(&info, argp, sizeof(info))) {
5359 err = -EFAULT;
5360 goto abort_unlock;
5361 }
5362 if (mddev->pers) {
5363 err = update_array_info(mddev, &info);
5364 if (err) {
5365 printk(KERN_WARNING "md: couldn't update"
5366 " array info. %d\n", err);
5367 goto abort_unlock;
5368 }
5369 goto done_unlock;
5370 }
5371 if (!list_empty(&mddev->disks)) {
5372 printk(KERN_WARNING
5373 "md: array %s already has disks!\n",
5374 mdname(mddev));
5375 err = -EBUSY;
5376 goto abort_unlock;
5377 }
5378 if (mddev->raid_disks) {
5379 printk(KERN_WARNING
5380 "md: array %s already initialised!\n",
5381 mdname(mddev));
5382 err = -EBUSY;
5383 goto abort_unlock;
5384 }
5385 err = set_array_info(mddev, &info);
5386 if (err) {
5387 printk(KERN_WARNING "md: couldn't set"
5388 " array info. %d\n", err);
5389 goto abort_unlock;
5390 }
5391 }
5392 goto done_unlock;
5393
5394 default:;
5395 }
5396
5397 /*
5398 * Commands querying/configuring an existing array:
5399 */
5400 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5401 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5402 if ((!mddev->raid_disks && !mddev->external)
5403 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5404 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5405 && cmd != GET_BITMAP_FILE) {
5406 err = -ENODEV;
5407 goto abort_unlock;
5408 }
5409
5410 /*
5411 * Commands even a read-only array can execute:
5412 */
5413 switch (cmd)
5414 {
5415 case GET_ARRAY_INFO:
5416 err = get_array_info(mddev, argp);
5417 goto done_unlock;
5418
5419 case GET_BITMAP_FILE:
5420 err = get_bitmap_file(mddev, argp);
5421 goto done_unlock;
5422
5423 case GET_DISK_INFO:
5424 err = get_disk_info(mddev, argp);
5425 goto done_unlock;
5426
5427 case RESTART_ARRAY_RW:
5428 err = restart_array(mddev);
5429 goto done_unlock;
5430
5431 case STOP_ARRAY:
5432 err = do_md_stop(mddev, 0, 1);
5433 goto done_unlock;
5434
5435 case STOP_ARRAY_RO:
5436 err = do_md_stop(mddev, 1, 1);
5437 goto done_unlock;
5438
5439 }
5440
5441 /*
5442 * The remaining ioctls are changing the state of the
5443 * superblock, so we do not allow them on read-only arrays.
5444 * However non-MD ioctls (e.g. get-size) will still come through
5445 * here and hit the 'default' below, so only disallow
5446 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5447 */
5448 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5449 if (mddev->ro == 2) {
5450 mddev->ro = 0;
5451 sysfs_notify_dirent(mddev->sysfs_state);
5452 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5453 md_wakeup_thread(mddev->thread);
5454 } else {
5455 err = -EROFS;
5456 goto abort_unlock;
5457 }
5458 }
5459
5460 switch (cmd)
5461 {
5462 case ADD_NEW_DISK:
5463 {
5464 mdu_disk_info_t info;
5465 if (copy_from_user(&info, argp, sizeof(info)))
5466 err = -EFAULT;
5467 else
5468 err = add_new_disk(mddev, &info);
5469 goto done_unlock;
5470 }
5471
5472 case HOT_REMOVE_DISK:
5473 err = hot_remove_disk(mddev, new_decode_dev(arg));
5474 goto done_unlock;
5475
5476 case HOT_ADD_DISK:
5477 err = hot_add_disk(mddev, new_decode_dev(arg));
5478 goto done_unlock;
5479
5480 case SET_DISK_FAULTY:
5481 err = set_disk_faulty(mddev, new_decode_dev(arg));
5482 goto done_unlock;
5483
5484 case RUN_ARRAY:
5485 err = do_md_run(mddev);
5486 goto done_unlock;
5487
5488 case SET_BITMAP_FILE:
5489 err = set_bitmap_file(mddev, (int)arg);
5490 goto done_unlock;
5491
5492 default:
5493 err = -EINVAL;
5494 goto abort_unlock;
5495 }
5496
5497 done_unlock:
5498 abort_unlock:
5499 if (mddev->hold_active == UNTIL_IOCTL &&
5500 err != -EINVAL)
5501 mddev->hold_active = 0;
5502 mddev_unlock(mddev);
5503
5504 return err;
5505 done:
5506 if (err)
5507 MD_BUG();
5508 abort:
5509 return err;
5510 }
5511
5512 static int md_open(struct block_device *bdev, fmode_t mode)
5513 {
5514 /*
5515 * Succeed if we can lock the mddev, which confirms that
5516 * it isn't being stopped right now.
5517 */
5518 mddev_t *mddev = mddev_find(bdev->bd_dev);
5519 int err;
5520
5521 if (mddev->gendisk != bdev->bd_disk) {
5522 /* we are racing with mddev_put which is discarding this
5523 * bd_disk.
5524 */
5525 mddev_put(mddev);
5526 /* Wait until bdev->bd_disk is definitely gone */
5527 flush_scheduled_work();
5528 /* Then retry the open from the top */
5529 return -ERESTARTSYS;
5530 }
5531 BUG_ON(mddev != bdev->bd_disk->private_data);
5532
5533 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5534 goto out;
5535
5536 err = 0;
5537 atomic_inc(&mddev->openers);
5538 mddev_unlock(mddev);
5539
5540 check_disk_change(bdev);
5541 out:
5542 return err;
5543 }
5544
5545 static int md_release(struct gendisk *disk, fmode_t mode)
5546 {
5547 mddev_t *mddev = disk->private_data;
5548
5549 BUG_ON(!mddev);
5550 atomic_dec(&mddev->openers);
5551 mddev_put(mddev);
5552
5553 return 0;
5554 }
5555
5556 static int md_media_changed(struct gendisk *disk)
5557 {
5558 mddev_t *mddev = disk->private_data;
5559
5560 return mddev->changed;
5561 }
5562
5563 static int md_revalidate(struct gendisk *disk)
5564 {
5565 mddev_t *mddev = disk->private_data;
5566
5567 mddev->changed = 0;
5568 return 0;
5569 }
5570 static struct block_device_operations md_fops =
5571 {
5572 .owner = THIS_MODULE,
5573 .open = md_open,
5574 .release = md_release,
5575 .ioctl = md_ioctl,
5576 .getgeo = md_getgeo,
5577 .media_changed = md_media_changed,
5578 .revalidate_disk= md_revalidate,
5579 };
5580
5581 static int md_thread(void * arg)
5582 {
5583 mdk_thread_t *thread = arg;
5584
5585 /*
5586 * md_thread is a 'system-thread', it's priority should be very
5587 * high. We avoid resource deadlocks individually in each
5588 * raid personality. (RAID5 does preallocation) We also use RR and
5589 * the very same RT priority as kswapd, thus we will never get
5590 * into a priority inversion deadlock.
5591 *
5592 * we definitely have to have equal or higher priority than
5593 * bdflush, otherwise bdflush will deadlock if there are too
5594 * many dirty RAID5 blocks.
5595 */
5596
5597 allow_signal(SIGKILL);
5598 while (!kthread_should_stop()) {
5599
5600 /* We need to wait INTERRUPTIBLE so that
5601 * we don't add to the load-average.
5602 * That means we need to be sure no signals are
5603 * pending
5604 */
5605 if (signal_pending(current))
5606 flush_signals(current);
5607
5608 wait_event_interruptible_timeout
5609 (thread->wqueue,
5610 test_bit(THREAD_WAKEUP, &thread->flags)
5611 || kthread_should_stop(),
5612 thread->timeout);
5613
5614 clear_bit(THREAD_WAKEUP, &thread->flags);
5615
5616 thread->run(thread->mddev);
5617 }
5618
5619 return 0;
5620 }
5621
5622 void md_wakeup_thread(mdk_thread_t *thread)
5623 {
5624 if (thread) {
5625 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5626 set_bit(THREAD_WAKEUP, &thread->flags);
5627 wake_up(&thread->wqueue);
5628 }
5629 }
5630
5631 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5632 const char *name)
5633 {
5634 mdk_thread_t *thread;
5635
5636 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5637 if (!thread)
5638 return NULL;
5639
5640 init_waitqueue_head(&thread->wqueue);
5641
5642 thread->run = run;
5643 thread->mddev = mddev;
5644 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5645 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5646 if (IS_ERR(thread->tsk)) {
5647 kfree(thread);
5648 return NULL;
5649 }
5650 return thread;
5651 }
5652
5653 void md_unregister_thread(mdk_thread_t *thread)
5654 {
5655 if (!thread)
5656 return;
5657 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5658
5659 kthread_stop(thread->tsk);
5660 kfree(thread);
5661 }
5662
5663 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5664 {
5665 if (!mddev) {
5666 MD_BUG();
5667 return;
5668 }
5669
5670 if (!rdev || test_bit(Faulty, &rdev->flags))
5671 return;
5672
5673 if (mddev->external)
5674 set_bit(Blocked, &rdev->flags);
5675 /*
5676 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5677 mdname(mddev),
5678 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5679 __builtin_return_address(0),__builtin_return_address(1),
5680 __builtin_return_address(2),__builtin_return_address(3));
5681 */
5682 if (!mddev->pers)
5683 return;
5684 if (!mddev->pers->error_handler)
5685 return;
5686 mddev->pers->error_handler(mddev,rdev);
5687 if (mddev->degraded)
5688 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5689 set_bit(StateChanged, &rdev->flags);
5690 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5691 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5692 md_wakeup_thread(mddev->thread);
5693 md_new_event_inintr(mddev);
5694 }
5695
5696 /* seq_file implementation /proc/mdstat */
5697
5698 static void status_unused(struct seq_file *seq)
5699 {
5700 int i = 0;
5701 mdk_rdev_t *rdev;
5702
5703 seq_printf(seq, "unused devices: ");
5704
5705 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5706 char b[BDEVNAME_SIZE];
5707 i++;
5708 seq_printf(seq, "%s ",
5709 bdevname(rdev->bdev,b));
5710 }
5711 if (!i)
5712 seq_printf(seq, "<none>");
5713
5714 seq_printf(seq, "\n");
5715 }
5716
5717
5718 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5719 {
5720 sector_t max_sectors, resync, res;
5721 unsigned long dt, db;
5722 sector_t rt;
5723 int scale;
5724 unsigned int per_milli;
5725
5726 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5727
5728 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5729 max_sectors = mddev->resync_max_sectors;
5730 else
5731 max_sectors = mddev->dev_sectors;
5732
5733 /*
5734 * Should not happen.
5735 */
5736 if (!max_sectors) {
5737 MD_BUG();
5738 return;
5739 }
5740 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5741 * in a sector_t, and (max_sectors>>scale) will fit in a
5742 * u32, as those are the requirements for sector_div.
5743 * Thus 'scale' must be at least 10
5744 */
5745 scale = 10;
5746 if (sizeof(sector_t) > sizeof(unsigned long)) {
5747 while ( max_sectors/2 > (1ULL<<(scale+32)))
5748 scale++;
5749 }
5750 res = (resync>>scale)*1000;
5751 sector_div(res, (u32)((max_sectors>>scale)+1));
5752
5753 per_milli = res;
5754 {
5755 int i, x = per_milli/50, y = 20-x;
5756 seq_printf(seq, "[");
5757 for (i = 0; i < x; i++)
5758 seq_printf(seq, "=");
5759 seq_printf(seq, ">");
5760 for (i = 0; i < y; i++)
5761 seq_printf(seq, ".");
5762 seq_printf(seq, "] ");
5763 }
5764 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5765 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5766 "reshape" :
5767 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5768 "check" :
5769 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5770 "resync" : "recovery"))),
5771 per_milli/10, per_milli % 10,
5772 (unsigned long long) resync/2,
5773 (unsigned long long) max_sectors/2);
5774
5775 /*
5776 * dt: time from mark until now
5777 * db: blocks written from mark until now
5778 * rt: remaining time
5779 *
5780 * rt is a sector_t, so could be 32bit or 64bit.
5781 * So we divide before multiply in case it is 32bit and close
5782 * to the limit.
5783 * We scale the divisor (db) by 32 to avoid loosing precision
5784 * near the end of resync when the number of remaining sectors
5785 * is close to 'db'.
5786 * We then divide rt by 32 after multiplying by db to compensate.
5787 * The '+1' avoids division by zero if db is very small.
5788 */
5789 dt = ((jiffies - mddev->resync_mark) / HZ);
5790 if (!dt) dt++;
5791 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5792 - mddev->resync_mark_cnt;
5793
5794 rt = max_sectors - resync; /* number of remaining sectors */
5795 sector_div(rt, db/32+1);
5796 rt *= dt;
5797 rt >>= 5;
5798
5799 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5800 ((unsigned long)rt % 60)/6);
5801
5802 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5803 }
5804
5805 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5806 {
5807 struct list_head *tmp;
5808 loff_t l = *pos;
5809 mddev_t *mddev;
5810
5811 if (l >= 0x10000)
5812 return NULL;
5813 if (!l--)
5814 /* header */
5815 return (void*)1;
5816
5817 spin_lock(&all_mddevs_lock);
5818 list_for_each(tmp,&all_mddevs)
5819 if (!l--) {
5820 mddev = list_entry(tmp, mddev_t, all_mddevs);
5821 mddev_get(mddev);
5822 spin_unlock(&all_mddevs_lock);
5823 return mddev;
5824 }
5825 spin_unlock(&all_mddevs_lock);
5826 if (!l--)
5827 return (void*)2;/* tail */
5828 return NULL;
5829 }
5830
5831 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5832 {
5833 struct list_head *tmp;
5834 mddev_t *next_mddev, *mddev = v;
5835
5836 ++*pos;
5837 if (v == (void*)2)
5838 return NULL;
5839
5840 spin_lock(&all_mddevs_lock);
5841 if (v == (void*)1)
5842 tmp = all_mddevs.next;
5843 else
5844 tmp = mddev->all_mddevs.next;
5845 if (tmp != &all_mddevs)
5846 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5847 else {
5848 next_mddev = (void*)2;
5849 *pos = 0x10000;
5850 }
5851 spin_unlock(&all_mddevs_lock);
5852
5853 if (v != (void*)1)
5854 mddev_put(mddev);
5855 return next_mddev;
5856
5857 }
5858
5859 static void md_seq_stop(struct seq_file *seq, void *v)
5860 {
5861 mddev_t *mddev = v;
5862
5863 if (mddev && v != (void*)1 && v != (void*)2)
5864 mddev_put(mddev);
5865 }
5866
5867 struct mdstat_info {
5868 int event;
5869 };
5870
5871 static int md_seq_show(struct seq_file *seq, void *v)
5872 {
5873 mddev_t *mddev = v;
5874 sector_t sectors;
5875 mdk_rdev_t *rdev;
5876 struct mdstat_info *mi = seq->private;
5877 struct bitmap *bitmap;
5878
5879 if (v == (void*)1) {
5880 struct mdk_personality *pers;
5881 seq_printf(seq, "Personalities : ");
5882 spin_lock(&pers_lock);
5883 list_for_each_entry(pers, &pers_list, list)
5884 seq_printf(seq, "[%s] ", pers->name);
5885
5886 spin_unlock(&pers_lock);
5887 seq_printf(seq, "\n");
5888 mi->event = atomic_read(&md_event_count);
5889 return 0;
5890 }
5891 if (v == (void*)2) {
5892 status_unused(seq);
5893 return 0;
5894 }
5895
5896 if (mddev_lock(mddev) < 0)
5897 return -EINTR;
5898
5899 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5900 seq_printf(seq, "%s : %sactive", mdname(mddev),
5901 mddev->pers ? "" : "in");
5902 if (mddev->pers) {
5903 if (mddev->ro==1)
5904 seq_printf(seq, " (read-only)");
5905 if (mddev->ro==2)
5906 seq_printf(seq, " (auto-read-only)");
5907 seq_printf(seq, " %s", mddev->pers->name);
5908 }
5909
5910 sectors = 0;
5911 list_for_each_entry(rdev, &mddev->disks, same_set) {
5912 char b[BDEVNAME_SIZE];
5913 seq_printf(seq, " %s[%d]",
5914 bdevname(rdev->bdev,b), rdev->desc_nr);
5915 if (test_bit(WriteMostly, &rdev->flags))
5916 seq_printf(seq, "(W)");
5917 if (test_bit(Faulty, &rdev->flags)) {
5918 seq_printf(seq, "(F)");
5919 continue;
5920 } else if (rdev->raid_disk < 0)
5921 seq_printf(seq, "(S)"); /* spare */
5922 sectors += rdev->sectors;
5923 }
5924
5925 if (!list_empty(&mddev->disks)) {
5926 if (mddev->pers)
5927 seq_printf(seq, "\n %llu blocks",
5928 (unsigned long long)
5929 mddev->array_sectors / 2);
5930 else
5931 seq_printf(seq, "\n %llu blocks",
5932 (unsigned long long)sectors / 2);
5933 }
5934 if (mddev->persistent) {
5935 if (mddev->major_version != 0 ||
5936 mddev->minor_version != 90) {
5937 seq_printf(seq," super %d.%d",
5938 mddev->major_version,
5939 mddev->minor_version);
5940 }
5941 } else if (mddev->external)
5942 seq_printf(seq, " super external:%s",
5943 mddev->metadata_type);
5944 else
5945 seq_printf(seq, " super non-persistent");
5946
5947 if (mddev->pers) {
5948 mddev->pers->status(seq, mddev);
5949 seq_printf(seq, "\n ");
5950 if (mddev->pers->sync_request) {
5951 if (mddev->curr_resync > 2) {
5952 status_resync(seq, mddev);
5953 seq_printf(seq, "\n ");
5954 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5955 seq_printf(seq, "\tresync=DELAYED\n ");
5956 else if (mddev->recovery_cp < MaxSector)
5957 seq_printf(seq, "\tresync=PENDING\n ");
5958 }
5959 } else
5960 seq_printf(seq, "\n ");
5961
5962 if ((bitmap = mddev->bitmap)) {
5963 unsigned long chunk_kb;
5964 unsigned long flags;
5965 spin_lock_irqsave(&bitmap->lock, flags);
5966 chunk_kb = bitmap->chunksize >> 10;
5967 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5968 "%lu%s chunk",
5969 bitmap->pages - bitmap->missing_pages,
5970 bitmap->pages,
5971 (bitmap->pages - bitmap->missing_pages)
5972 << (PAGE_SHIFT - 10),
5973 chunk_kb ? chunk_kb : bitmap->chunksize,
5974 chunk_kb ? "KB" : "B");
5975 if (bitmap->file) {
5976 seq_printf(seq, ", file: ");
5977 seq_path(seq, &bitmap->file->f_path, " \t\n");
5978 }
5979
5980 seq_printf(seq, "\n");
5981 spin_unlock_irqrestore(&bitmap->lock, flags);
5982 }
5983
5984 seq_printf(seq, "\n");
5985 }
5986 mddev_unlock(mddev);
5987
5988 return 0;
5989 }
5990
5991 static const struct seq_operations md_seq_ops = {
5992 .start = md_seq_start,
5993 .next = md_seq_next,
5994 .stop = md_seq_stop,
5995 .show = md_seq_show,
5996 };
5997
5998 static int md_seq_open(struct inode *inode, struct file *file)
5999 {
6000 int error;
6001 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6002 if (mi == NULL)
6003 return -ENOMEM;
6004
6005 error = seq_open(file, &md_seq_ops);
6006 if (error)
6007 kfree(mi);
6008 else {
6009 struct seq_file *p = file->private_data;
6010 p->private = mi;
6011 mi->event = atomic_read(&md_event_count);
6012 }
6013 return error;
6014 }
6015
6016 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6017 {
6018 struct seq_file *m = filp->private_data;
6019 struct mdstat_info *mi = m->private;
6020 int mask;
6021
6022 poll_wait(filp, &md_event_waiters, wait);
6023
6024 /* always allow read */
6025 mask = POLLIN | POLLRDNORM;
6026
6027 if (mi->event != atomic_read(&md_event_count))
6028 mask |= POLLERR | POLLPRI;
6029 return mask;
6030 }
6031
6032 static const struct file_operations md_seq_fops = {
6033 .owner = THIS_MODULE,
6034 .open = md_seq_open,
6035 .read = seq_read,
6036 .llseek = seq_lseek,
6037 .release = seq_release_private,
6038 .poll = mdstat_poll,
6039 };
6040
6041 int register_md_personality(struct mdk_personality *p)
6042 {
6043 spin_lock(&pers_lock);
6044 list_add_tail(&p->list, &pers_list);
6045 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6046 spin_unlock(&pers_lock);
6047 return 0;
6048 }
6049
6050 int unregister_md_personality(struct mdk_personality *p)
6051 {
6052 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6053 spin_lock(&pers_lock);
6054 list_del_init(&p->list);
6055 spin_unlock(&pers_lock);
6056 return 0;
6057 }
6058
6059 static int is_mddev_idle(mddev_t *mddev, int init)
6060 {
6061 mdk_rdev_t * rdev;
6062 int idle;
6063 int curr_events;
6064
6065 idle = 1;
6066 rcu_read_lock();
6067 rdev_for_each_rcu(rdev, mddev) {
6068 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6069 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6070 (int)part_stat_read(&disk->part0, sectors[1]) -
6071 atomic_read(&disk->sync_io);
6072 /* sync IO will cause sync_io to increase before the disk_stats
6073 * as sync_io is counted when a request starts, and
6074 * disk_stats is counted when it completes.
6075 * So resync activity will cause curr_events to be smaller than
6076 * when there was no such activity.
6077 * non-sync IO will cause disk_stat to increase without
6078 * increasing sync_io so curr_events will (eventually)
6079 * be larger than it was before. Once it becomes
6080 * substantially larger, the test below will cause
6081 * the array to appear non-idle, and resync will slow
6082 * down.
6083 * If there is a lot of outstanding resync activity when
6084 * we set last_event to curr_events, then all that activity
6085 * completing might cause the array to appear non-idle
6086 * and resync will be slowed down even though there might
6087 * not have been non-resync activity. This will only
6088 * happen once though. 'last_events' will soon reflect
6089 * the state where there is little or no outstanding
6090 * resync requests, and further resync activity will
6091 * always make curr_events less than last_events.
6092 *
6093 */
6094 if (init || curr_events - rdev->last_events > 64) {
6095 rdev->last_events = curr_events;
6096 idle = 0;
6097 }
6098 }
6099 rcu_read_unlock();
6100 return idle;
6101 }
6102
6103 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6104 {
6105 /* another "blocks" (512byte) blocks have been synced */
6106 atomic_sub(blocks, &mddev->recovery_active);
6107 wake_up(&mddev->recovery_wait);
6108 if (!ok) {
6109 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6110 md_wakeup_thread(mddev->thread);
6111 // stop recovery, signal do_sync ....
6112 }
6113 }
6114
6115
6116 /* md_write_start(mddev, bi)
6117 * If we need to update some array metadata (e.g. 'active' flag
6118 * in superblock) before writing, schedule a superblock update
6119 * and wait for it to complete.
6120 */
6121 void md_write_start(mddev_t *mddev, struct bio *bi)
6122 {
6123 int did_change = 0;
6124 if (bio_data_dir(bi) != WRITE)
6125 return;
6126
6127 BUG_ON(mddev->ro == 1);
6128 if (mddev->ro == 2) {
6129 /* need to switch to read/write */
6130 mddev->ro = 0;
6131 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6132 md_wakeup_thread(mddev->thread);
6133 md_wakeup_thread(mddev->sync_thread);
6134 did_change = 1;
6135 }
6136 atomic_inc(&mddev->writes_pending);
6137 if (mddev->safemode == 1)
6138 mddev->safemode = 0;
6139 if (mddev->in_sync) {
6140 spin_lock_irq(&mddev->write_lock);
6141 if (mddev->in_sync) {
6142 mddev->in_sync = 0;
6143 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6144 md_wakeup_thread(mddev->thread);
6145 did_change = 1;
6146 }
6147 spin_unlock_irq(&mddev->write_lock);
6148 }
6149 if (did_change)
6150 sysfs_notify_dirent(mddev->sysfs_state);
6151 wait_event(mddev->sb_wait,
6152 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6153 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6154 }
6155
6156 void md_write_end(mddev_t *mddev)
6157 {
6158 if (atomic_dec_and_test(&mddev->writes_pending)) {
6159 if (mddev->safemode == 2)
6160 md_wakeup_thread(mddev->thread);
6161 else if (mddev->safemode_delay)
6162 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6163 }
6164 }
6165
6166 /* md_allow_write(mddev)
6167 * Calling this ensures that the array is marked 'active' so that writes
6168 * may proceed without blocking. It is important to call this before
6169 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6170 * Must be called with mddev_lock held.
6171 *
6172 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6173 * is dropped, so return -EAGAIN after notifying userspace.
6174 */
6175 int md_allow_write(mddev_t *mddev)
6176 {
6177 if (!mddev->pers)
6178 return 0;
6179 if (mddev->ro)
6180 return 0;
6181 if (!mddev->pers->sync_request)
6182 return 0;
6183
6184 spin_lock_irq(&mddev->write_lock);
6185 if (mddev->in_sync) {
6186 mddev->in_sync = 0;
6187 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6188 if (mddev->safemode_delay &&
6189 mddev->safemode == 0)
6190 mddev->safemode = 1;
6191 spin_unlock_irq(&mddev->write_lock);
6192 md_update_sb(mddev, 0);
6193 sysfs_notify_dirent(mddev->sysfs_state);
6194 } else
6195 spin_unlock_irq(&mddev->write_lock);
6196
6197 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6198 return -EAGAIN;
6199 else
6200 return 0;
6201 }
6202 EXPORT_SYMBOL_GPL(md_allow_write);
6203
6204 #define SYNC_MARKS 10
6205 #define SYNC_MARK_STEP (3*HZ)
6206 void md_do_sync(mddev_t *mddev)
6207 {
6208 mddev_t *mddev2;
6209 unsigned int currspeed = 0,
6210 window;
6211 sector_t max_sectors,j, io_sectors;
6212 unsigned long mark[SYNC_MARKS];
6213 sector_t mark_cnt[SYNC_MARKS];
6214 int last_mark,m;
6215 struct list_head *tmp;
6216 sector_t last_check;
6217 int skipped = 0;
6218 mdk_rdev_t *rdev;
6219 char *desc;
6220
6221 /* just incase thread restarts... */
6222 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6223 return;
6224 if (mddev->ro) /* never try to sync a read-only array */
6225 return;
6226
6227 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6228 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6229 desc = "data-check";
6230 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6231 desc = "requested-resync";
6232 else
6233 desc = "resync";
6234 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6235 desc = "reshape";
6236 else
6237 desc = "recovery";
6238
6239 /* we overload curr_resync somewhat here.
6240 * 0 == not engaged in resync at all
6241 * 2 == checking that there is no conflict with another sync
6242 * 1 == like 2, but have yielded to allow conflicting resync to
6243 * commense
6244 * other == active in resync - this many blocks
6245 *
6246 * Before starting a resync we must have set curr_resync to
6247 * 2, and then checked that every "conflicting" array has curr_resync
6248 * less than ours. When we find one that is the same or higher
6249 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6250 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6251 * This will mean we have to start checking from the beginning again.
6252 *
6253 */
6254
6255 do {
6256 mddev->curr_resync = 2;
6257
6258 try_again:
6259 if (kthread_should_stop()) {
6260 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6261 goto skip;
6262 }
6263 for_each_mddev(mddev2, tmp) {
6264 if (mddev2 == mddev)
6265 continue;
6266 if (!mddev->parallel_resync
6267 && mddev2->curr_resync
6268 && match_mddev_units(mddev, mddev2)) {
6269 DEFINE_WAIT(wq);
6270 if (mddev < mddev2 && mddev->curr_resync == 2) {
6271 /* arbitrarily yield */
6272 mddev->curr_resync = 1;
6273 wake_up(&resync_wait);
6274 }
6275 if (mddev > mddev2 && mddev->curr_resync == 1)
6276 /* no need to wait here, we can wait the next
6277 * time 'round when curr_resync == 2
6278 */
6279 continue;
6280 /* We need to wait 'interruptible' so as not to
6281 * contribute to the load average, and not to
6282 * be caught by 'softlockup'
6283 */
6284 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6285 if (!kthread_should_stop() &&
6286 mddev2->curr_resync >= mddev->curr_resync) {
6287 printk(KERN_INFO "md: delaying %s of %s"
6288 " until %s has finished (they"
6289 " share one or more physical units)\n",
6290 desc, mdname(mddev), mdname(mddev2));
6291 mddev_put(mddev2);
6292 if (signal_pending(current))
6293 flush_signals(current);
6294 schedule();
6295 finish_wait(&resync_wait, &wq);
6296 goto try_again;
6297 }
6298 finish_wait(&resync_wait, &wq);
6299 }
6300 }
6301 } while (mddev->curr_resync < 2);
6302
6303 j = 0;
6304 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6305 /* resync follows the size requested by the personality,
6306 * which defaults to physical size, but can be virtual size
6307 */
6308 max_sectors = mddev->resync_max_sectors;
6309 mddev->resync_mismatches = 0;
6310 /* we don't use the checkpoint if there's a bitmap */
6311 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6312 j = mddev->resync_min;
6313 else if (!mddev->bitmap)
6314 j = mddev->recovery_cp;
6315
6316 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6317 max_sectors = mddev->dev_sectors;
6318 else {
6319 /* recovery follows the physical size of devices */
6320 max_sectors = mddev->dev_sectors;
6321 j = MaxSector;
6322 list_for_each_entry(rdev, &mddev->disks, same_set)
6323 if (rdev->raid_disk >= 0 &&
6324 !test_bit(Faulty, &rdev->flags) &&
6325 !test_bit(In_sync, &rdev->flags) &&
6326 rdev->recovery_offset < j)
6327 j = rdev->recovery_offset;
6328 }
6329
6330 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6331 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6332 " %d KB/sec/disk.\n", speed_min(mddev));
6333 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6334 "(but not more than %d KB/sec) for %s.\n",
6335 speed_max(mddev), desc);
6336
6337 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6338
6339 io_sectors = 0;
6340 for (m = 0; m < SYNC_MARKS; m++) {
6341 mark[m] = jiffies;
6342 mark_cnt[m] = io_sectors;
6343 }
6344 last_mark = 0;
6345 mddev->resync_mark = mark[last_mark];
6346 mddev->resync_mark_cnt = mark_cnt[last_mark];
6347
6348 /*
6349 * Tune reconstruction:
6350 */
6351 window = 32*(PAGE_SIZE/512);
6352 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6353 window/2,(unsigned long long) max_sectors/2);
6354
6355 atomic_set(&mddev->recovery_active, 0);
6356 last_check = 0;
6357
6358 if (j>2) {
6359 printk(KERN_INFO
6360 "md: resuming %s of %s from checkpoint.\n",
6361 desc, mdname(mddev));
6362 mddev->curr_resync = j;
6363 }
6364
6365 while (j < max_sectors) {
6366 sector_t sectors;
6367
6368 skipped = 0;
6369
6370 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6371 ((mddev->curr_resync > mddev->curr_resync_completed &&
6372 (mddev->curr_resync - mddev->curr_resync_completed)
6373 > (max_sectors >> 4)) ||
6374 (j - mddev->curr_resync_completed)*2
6375 >= mddev->resync_max - mddev->curr_resync_completed
6376 )) {
6377 /* time to update curr_resync_completed */
6378 blk_unplug(mddev->queue);
6379 wait_event(mddev->recovery_wait,
6380 atomic_read(&mddev->recovery_active) == 0);
6381 mddev->curr_resync_completed =
6382 mddev->curr_resync;
6383 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6384 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6385 }
6386
6387 if (j >= mddev->resync_max)
6388 wait_event(mddev->recovery_wait,
6389 mddev->resync_max > j
6390 || kthread_should_stop());
6391
6392 if (kthread_should_stop())
6393 goto interrupted;
6394
6395 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6396 currspeed < speed_min(mddev));
6397 if (sectors == 0) {
6398 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6399 goto out;
6400 }
6401
6402 if (!skipped) { /* actual IO requested */
6403 io_sectors += sectors;
6404 atomic_add(sectors, &mddev->recovery_active);
6405 }
6406
6407 j += sectors;
6408 if (j>1) mddev->curr_resync = j;
6409 mddev->curr_mark_cnt = io_sectors;
6410 if (last_check == 0)
6411 /* this is the earliers that rebuilt will be
6412 * visible in /proc/mdstat
6413 */
6414 md_new_event(mddev);
6415
6416 if (last_check + window > io_sectors || j == max_sectors)
6417 continue;
6418
6419 last_check = io_sectors;
6420
6421 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6422 break;
6423
6424 repeat:
6425 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6426 /* step marks */
6427 int next = (last_mark+1) % SYNC_MARKS;
6428
6429 mddev->resync_mark = mark[next];
6430 mddev->resync_mark_cnt = mark_cnt[next];
6431 mark[next] = jiffies;
6432 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6433 last_mark = next;
6434 }
6435
6436
6437 if (kthread_should_stop())
6438 goto interrupted;
6439
6440
6441 /*
6442 * this loop exits only if either when we are slower than
6443 * the 'hard' speed limit, or the system was IO-idle for
6444 * a jiffy.
6445 * the system might be non-idle CPU-wise, but we only care
6446 * about not overloading the IO subsystem. (things like an
6447 * e2fsck being done on the RAID array should execute fast)
6448 */
6449 blk_unplug(mddev->queue);
6450 cond_resched();
6451
6452 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6453 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6454
6455 if (currspeed > speed_min(mddev)) {
6456 if ((currspeed > speed_max(mddev)) ||
6457 !is_mddev_idle(mddev, 0)) {
6458 msleep(500);
6459 goto repeat;
6460 }
6461 }
6462 }
6463 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6464 /*
6465 * this also signals 'finished resyncing' to md_stop
6466 */
6467 out:
6468 blk_unplug(mddev->queue);
6469
6470 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6471
6472 /* tell personality that we are finished */
6473 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6474
6475 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6476 mddev->curr_resync > 2) {
6477 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6478 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6479 if (mddev->curr_resync >= mddev->recovery_cp) {
6480 printk(KERN_INFO
6481 "md: checkpointing %s of %s.\n",
6482 desc, mdname(mddev));
6483 mddev->recovery_cp = mddev->curr_resync;
6484 }
6485 } else
6486 mddev->recovery_cp = MaxSector;
6487 } else {
6488 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6489 mddev->curr_resync = MaxSector;
6490 list_for_each_entry(rdev, &mddev->disks, same_set)
6491 if (rdev->raid_disk >= 0 &&
6492 !test_bit(Faulty, &rdev->flags) &&
6493 !test_bit(In_sync, &rdev->flags) &&
6494 rdev->recovery_offset < mddev->curr_resync)
6495 rdev->recovery_offset = mddev->curr_resync;
6496 }
6497 }
6498 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6499
6500 skip:
6501 mddev->curr_resync = 0;
6502 mddev->curr_resync_completed = 0;
6503 mddev->resync_min = 0;
6504 mddev->resync_max = MaxSector;
6505 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6506 wake_up(&resync_wait);
6507 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6508 md_wakeup_thread(mddev->thread);
6509 return;
6510
6511 interrupted:
6512 /*
6513 * got a signal, exit.
6514 */
6515 printk(KERN_INFO
6516 "md: md_do_sync() got signal ... exiting\n");
6517 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6518 goto out;
6519
6520 }
6521 EXPORT_SYMBOL_GPL(md_do_sync);
6522
6523
6524 static int remove_and_add_spares(mddev_t *mddev)
6525 {
6526 mdk_rdev_t *rdev;
6527 int spares = 0;
6528
6529 mddev->curr_resync_completed = 0;
6530
6531 list_for_each_entry(rdev, &mddev->disks, same_set)
6532 if (rdev->raid_disk >= 0 &&
6533 !test_bit(Blocked, &rdev->flags) &&
6534 (test_bit(Faulty, &rdev->flags) ||
6535 ! test_bit(In_sync, &rdev->flags)) &&
6536 atomic_read(&rdev->nr_pending)==0) {
6537 if (mddev->pers->hot_remove_disk(
6538 mddev, rdev->raid_disk)==0) {
6539 char nm[20];
6540 sprintf(nm,"rd%d", rdev->raid_disk);
6541 sysfs_remove_link(&mddev->kobj, nm);
6542 rdev->raid_disk = -1;
6543 }
6544 }
6545
6546 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6547 list_for_each_entry(rdev, &mddev->disks, same_set) {
6548 if (rdev->raid_disk >= 0 &&
6549 !test_bit(In_sync, &rdev->flags) &&
6550 !test_bit(Blocked, &rdev->flags))
6551 spares++;
6552 if (rdev->raid_disk < 0
6553 && !test_bit(Faulty, &rdev->flags)) {
6554 rdev->recovery_offset = 0;
6555 if (mddev->pers->
6556 hot_add_disk(mddev, rdev) == 0) {
6557 char nm[20];
6558 sprintf(nm, "rd%d", rdev->raid_disk);
6559 if (sysfs_create_link(&mddev->kobj,
6560 &rdev->kobj, nm))
6561 printk(KERN_WARNING
6562 "md: cannot register "
6563 "%s for %s\n",
6564 nm, mdname(mddev));
6565 spares++;
6566 md_new_event(mddev);
6567 } else
6568 break;
6569 }
6570 }
6571 }
6572 return spares;
6573 }
6574 /*
6575 * This routine is regularly called by all per-raid-array threads to
6576 * deal with generic issues like resync and super-block update.
6577 * Raid personalities that don't have a thread (linear/raid0) do not
6578 * need this as they never do any recovery or update the superblock.
6579 *
6580 * It does not do any resync itself, but rather "forks" off other threads
6581 * to do that as needed.
6582 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6583 * "->recovery" and create a thread at ->sync_thread.
6584 * When the thread finishes it sets MD_RECOVERY_DONE
6585 * and wakeups up this thread which will reap the thread and finish up.
6586 * This thread also removes any faulty devices (with nr_pending == 0).
6587 *
6588 * The overall approach is:
6589 * 1/ if the superblock needs updating, update it.
6590 * 2/ If a recovery thread is running, don't do anything else.
6591 * 3/ If recovery has finished, clean up, possibly marking spares active.
6592 * 4/ If there are any faulty devices, remove them.
6593 * 5/ If array is degraded, try to add spares devices
6594 * 6/ If array has spares or is not in-sync, start a resync thread.
6595 */
6596 void md_check_recovery(mddev_t *mddev)
6597 {
6598 mdk_rdev_t *rdev;
6599
6600
6601 if (mddev->bitmap)
6602 bitmap_daemon_work(mddev->bitmap);
6603
6604 if (mddev->ro)
6605 return;
6606
6607 if (signal_pending(current)) {
6608 if (mddev->pers->sync_request && !mddev->external) {
6609 printk(KERN_INFO "md: %s in immediate safe mode\n",
6610 mdname(mddev));
6611 mddev->safemode = 2;
6612 }
6613 flush_signals(current);
6614 }
6615
6616 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6617 return;
6618 if ( ! (
6619 (mddev->flags && !mddev->external) ||
6620 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6621 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6622 (mddev->external == 0 && mddev->safemode == 1) ||
6623 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6624 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6625 ))
6626 return;
6627
6628 if (mddev_trylock(mddev)) {
6629 int spares = 0;
6630
6631 if (mddev->ro) {
6632 /* Only thing we do on a ro array is remove
6633 * failed devices.
6634 */
6635 remove_and_add_spares(mddev);
6636 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6637 goto unlock;
6638 }
6639
6640 if (!mddev->external) {
6641 int did_change = 0;
6642 spin_lock_irq(&mddev->write_lock);
6643 if (mddev->safemode &&
6644 !atomic_read(&mddev->writes_pending) &&
6645 !mddev->in_sync &&
6646 mddev->recovery_cp == MaxSector) {
6647 mddev->in_sync = 1;
6648 did_change = 1;
6649 if (mddev->persistent)
6650 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6651 }
6652 if (mddev->safemode == 1)
6653 mddev->safemode = 0;
6654 spin_unlock_irq(&mddev->write_lock);
6655 if (did_change)
6656 sysfs_notify_dirent(mddev->sysfs_state);
6657 }
6658
6659 if (mddev->flags)
6660 md_update_sb(mddev, 0);
6661
6662 list_for_each_entry(rdev, &mddev->disks, same_set)
6663 if (test_and_clear_bit(StateChanged, &rdev->flags))
6664 sysfs_notify_dirent(rdev->sysfs_state);
6665
6666
6667 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6668 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6669 /* resync/recovery still happening */
6670 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6671 goto unlock;
6672 }
6673 if (mddev->sync_thread) {
6674 /* resync has finished, collect result */
6675 md_unregister_thread(mddev->sync_thread);
6676 mddev->sync_thread = NULL;
6677 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6678 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6679 /* success...*/
6680 /* activate any spares */
6681 if (mddev->pers->spare_active(mddev))
6682 sysfs_notify(&mddev->kobj, NULL,
6683 "degraded");
6684 }
6685 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6686 mddev->pers->finish_reshape)
6687 mddev->pers->finish_reshape(mddev);
6688 md_update_sb(mddev, 1);
6689
6690 /* if array is no-longer degraded, then any saved_raid_disk
6691 * information must be scrapped
6692 */
6693 if (!mddev->degraded)
6694 list_for_each_entry(rdev, &mddev->disks, same_set)
6695 rdev->saved_raid_disk = -1;
6696
6697 mddev->recovery = 0;
6698 /* flag recovery needed just to double check */
6699 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6700 sysfs_notify_dirent(mddev->sysfs_action);
6701 md_new_event(mddev);
6702 goto unlock;
6703 }
6704 /* Set RUNNING before clearing NEEDED to avoid
6705 * any transients in the value of "sync_action".
6706 */
6707 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6708 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6709 /* Clear some bits that don't mean anything, but
6710 * might be left set
6711 */
6712 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6713 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6714
6715 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6716 goto unlock;
6717 /* no recovery is running.
6718 * remove any failed drives, then
6719 * add spares if possible.
6720 * Spare are also removed and re-added, to allow
6721 * the personality to fail the re-add.
6722 */
6723
6724 if (mddev->reshape_position != MaxSector) {
6725 if (mddev->pers->check_reshape == NULL ||
6726 mddev->pers->check_reshape(mddev) != 0)
6727 /* Cannot proceed */
6728 goto unlock;
6729 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6730 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6731 } else if ((spares = remove_and_add_spares(mddev))) {
6732 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6733 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6734 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6735 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6736 } else if (mddev->recovery_cp < MaxSector) {
6737 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6738 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6739 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6740 /* nothing to be done ... */
6741 goto unlock;
6742
6743 if (mddev->pers->sync_request) {
6744 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6745 /* We are adding a device or devices to an array
6746 * which has the bitmap stored on all devices.
6747 * So make sure all bitmap pages get written
6748 */
6749 bitmap_write_all(mddev->bitmap);
6750 }
6751 mddev->sync_thread = md_register_thread(md_do_sync,
6752 mddev,
6753 "%s_resync");
6754 if (!mddev->sync_thread) {
6755 printk(KERN_ERR "%s: could not start resync"
6756 " thread...\n",
6757 mdname(mddev));
6758 /* leave the spares where they are, it shouldn't hurt */
6759 mddev->recovery = 0;
6760 } else
6761 md_wakeup_thread(mddev->sync_thread);
6762 sysfs_notify_dirent(mddev->sysfs_action);
6763 md_new_event(mddev);
6764 }
6765 unlock:
6766 if (!mddev->sync_thread) {
6767 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6768 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6769 &mddev->recovery))
6770 if (mddev->sysfs_action)
6771 sysfs_notify_dirent(mddev->sysfs_action);
6772 }
6773 mddev_unlock(mddev);
6774 }
6775 }
6776
6777 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6778 {
6779 sysfs_notify_dirent(rdev->sysfs_state);
6780 wait_event_timeout(rdev->blocked_wait,
6781 !test_bit(Blocked, &rdev->flags),
6782 msecs_to_jiffies(5000));
6783 rdev_dec_pending(rdev, mddev);
6784 }
6785 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6786
6787 static int md_notify_reboot(struct notifier_block *this,
6788 unsigned long code, void *x)
6789 {
6790 struct list_head *tmp;
6791 mddev_t *mddev;
6792
6793 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6794
6795 printk(KERN_INFO "md: stopping all md devices.\n");
6796
6797 for_each_mddev(mddev, tmp)
6798 if (mddev_trylock(mddev)) {
6799 /* Force a switch to readonly even array
6800 * appears to still be in use. Hence
6801 * the '100'.
6802 */
6803 do_md_stop(mddev, 1, 100);
6804 mddev_unlock(mddev);
6805 }
6806 /*
6807 * certain more exotic SCSI devices are known to be
6808 * volatile wrt too early system reboots. While the
6809 * right place to handle this issue is the given
6810 * driver, we do want to have a safe RAID driver ...
6811 */
6812 mdelay(1000*1);
6813 }
6814 return NOTIFY_DONE;
6815 }
6816
6817 static struct notifier_block md_notifier = {
6818 .notifier_call = md_notify_reboot,
6819 .next = NULL,
6820 .priority = INT_MAX, /* before any real devices */
6821 };
6822
6823 static void md_geninit(void)
6824 {
6825 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6826
6827 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6828 }
6829
6830 static int __init md_init(void)
6831 {
6832 if (register_blkdev(MD_MAJOR, "md"))
6833 return -1;
6834 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6835 unregister_blkdev(MD_MAJOR, "md");
6836 return -1;
6837 }
6838 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6839 md_probe, NULL, NULL);
6840 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6841 md_probe, NULL, NULL);
6842
6843 register_reboot_notifier(&md_notifier);
6844 raid_table_header = register_sysctl_table(raid_root_table);
6845
6846 md_geninit();
6847 return 0;
6848 }
6849
6850
6851 #ifndef MODULE
6852
6853 /*
6854 * Searches all registered partitions for autorun RAID arrays
6855 * at boot time.
6856 */
6857
6858 static LIST_HEAD(all_detected_devices);
6859 struct detected_devices_node {
6860 struct list_head list;
6861 dev_t dev;
6862 };
6863
6864 void md_autodetect_dev(dev_t dev)
6865 {
6866 struct detected_devices_node *node_detected_dev;
6867
6868 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6869 if (node_detected_dev) {
6870 node_detected_dev->dev = dev;
6871 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6872 } else {
6873 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6874 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6875 }
6876 }
6877
6878
6879 static void autostart_arrays(int part)
6880 {
6881 mdk_rdev_t *rdev;
6882 struct detected_devices_node *node_detected_dev;
6883 dev_t dev;
6884 int i_scanned, i_passed;
6885
6886 i_scanned = 0;
6887 i_passed = 0;
6888
6889 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6890
6891 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6892 i_scanned++;
6893 node_detected_dev = list_entry(all_detected_devices.next,
6894 struct detected_devices_node, list);
6895 list_del(&node_detected_dev->list);
6896 dev = node_detected_dev->dev;
6897 kfree(node_detected_dev);
6898 rdev = md_import_device(dev,0, 90);
6899 if (IS_ERR(rdev))
6900 continue;
6901
6902 if (test_bit(Faulty, &rdev->flags)) {
6903 MD_BUG();
6904 continue;
6905 }
6906 set_bit(AutoDetected, &rdev->flags);
6907 list_add(&rdev->same_set, &pending_raid_disks);
6908 i_passed++;
6909 }
6910
6911 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6912 i_scanned, i_passed);
6913
6914 autorun_devices(part);
6915 }
6916
6917 #endif /* !MODULE */
6918
6919 static __exit void md_exit(void)
6920 {
6921 mddev_t *mddev;
6922 struct list_head *tmp;
6923
6924 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6925 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6926
6927 unregister_blkdev(MD_MAJOR,"md");
6928 unregister_blkdev(mdp_major, "mdp");
6929 unregister_reboot_notifier(&md_notifier);
6930 unregister_sysctl_table(raid_table_header);
6931 remove_proc_entry("mdstat", NULL);
6932 for_each_mddev(mddev, tmp) {
6933 export_array(mddev);
6934 mddev->hold_active = 0;
6935 }
6936 }
6937
6938 subsys_initcall(md_init);
6939 module_exit(md_exit)
6940
6941 static int get_ro(char *buffer, struct kernel_param *kp)
6942 {
6943 return sprintf(buffer, "%d", start_readonly);
6944 }
6945 static int set_ro(const char *val, struct kernel_param *kp)
6946 {
6947 char *e;
6948 int num = simple_strtoul(val, &e, 10);
6949 if (*val && (*e == '\0' || *e == '\n')) {
6950 start_readonly = num;
6951 return 0;
6952 }
6953 return -EINVAL;
6954 }
6955
6956 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6957 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6958
6959 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6960
6961 EXPORT_SYMBOL(register_md_personality);
6962 EXPORT_SYMBOL(unregister_md_personality);
6963 EXPORT_SYMBOL(md_error);
6964 EXPORT_SYMBOL(md_done_sync);
6965 EXPORT_SYMBOL(md_write_start);
6966 EXPORT_SYMBOL(md_write_end);
6967 EXPORT_SYMBOL(md_register_thread);
6968 EXPORT_SYMBOL(md_unregister_thread);
6969 EXPORT_SYMBOL(md_wakeup_thread);
6970 EXPORT_SYMBOL(md_check_recovery);
6971 MODULE_LICENSE("GPL");
6972 MODULE_ALIAS("md");
6973 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree