?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
iwlwifi: add missing rcu_read_lock
[linux-2.6/kvm.git] / drivers / md / md.c
blobcefd63daff3109ee6f54474e1d03f617c61d10f7
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/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/reboot.h>
47 #include <linux/file.h>
48 #include <linux/compat.h>
49 #include <linux/delay.h>
50 #include <linux/raid/md_p.h>
51 #include <linux/raid/md_u.h>
52 #include <linux/slab.h>
53 #include "md.h"
54 #include "bitmap.h"
55
56 #define DEBUG 0
57 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58
59
60 #ifndef MODULE
61 static void autostart_arrays(int part);
62 #endif
63
64 static LIST_HEAD(pers_list);
65 static DEFINE_SPINLOCK(pers_lock);
66
67 static void md_print_devices(void);
68
69 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70
71 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
72
73 /*
74 * Default number of read corrections we'll attempt on an rdev
75 * before ejecting it from the array. We divide the read error
76 * count by 2 for every hour elapsed between read errors.
77 */
78 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
79 /*
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
86 * idle IO detection.
87 *
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
90 */
91
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
95 {
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
98 }
99
100 static inline int speed_max(mddev_t *mddev)
101 {
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
104 }
105
106 static struct ctl_table_header *raid_table_header;
107
108 static ctl_table raid_table[] = {
109 {
110 .procname = "speed_limit_min",
111 .data = &sysctl_speed_limit_min,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = proc_dointvec,
115 },
116 {
117 .procname = "speed_limit_max",
118 .data = &sysctl_speed_limit_max,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
122 },
123 { }
124 };
125
126 static ctl_table raid_dir_table[] = {
127 {
128 .procname = "raid",
129 .maxlen = 0,
130 .mode = S_IRUGO|S_IXUGO,
131 .child = raid_table,
132 },
133 { }
134 };
135
136 static ctl_table raid_root_table[] = {
137 {
138 .procname = "dev",
139 .maxlen = 0,
140 .mode = 0555,
141 .child = raid_dir_table,
142 },
143 { }
144 };
145
146 static const struct block_device_operations md_fops;
147
148 static int start_readonly;
149
150 /*
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
154 * count increases.
155 *
156 * Events are:
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
159 */
160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
161 static atomic_t md_event_count;
162 void md_new_event(mddev_t *mddev)
163 {
164 atomic_inc(&md_event_count);
165 wake_up(&md_event_waiters);
166 }
167 EXPORT_SYMBOL_GPL(md_new_event);
168
169 /* Alternate version that can be called from interrupts
170 * when calling sysfs_notify isn't needed.
171 */
172 static void md_new_event_inintr(mddev_t *mddev)
173 {
174 atomic_inc(&md_event_count);
175 wake_up(&md_event_waiters);
176 }
177
178 /*
179 * Enables to iterate over all existing md arrays
180 * all_mddevs_lock protects this list.
181 */
182 static LIST_HEAD(all_mddevs);
183 static DEFINE_SPINLOCK(all_mddevs_lock);
184
185
186 /*
187 * iterates through all used mddevs in the system.
188 * We take care to grab the all_mddevs_lock whenever navigating
189 * the list, and to always hold a refcount when unlocked.
190 * Any code which breaks out of this loop while own
191 * a reference to the current mddev and must mddev_put it.
192 */
193 #define for_each_mddev(mddev,tmp) \
194 \
195 for (({ spin_lock(&all_mddevs_lock); \
196 tmp = all_mddevs.next; \
197 mddev = NULL;}); \
198 ({ if (tmp != &all_mddevs) \
199 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
200 spin_unlock(&all_mddevs_lock); \
201 if (mddev) mddev_put(mddev); \
202 mddev = list_entry(tmp, mddev_t, all_mddevs); \
203 tmp != &all_mddevs;}); \
204 ({ spin_lock(&all_mddevs_lock); \
205 tmp = tmp->next;}) \
206 )
207
208
209 /* Rather than calling directly into the personality make_request function,
210 * IO requests come here first so that we can check if the device is
211 * being suspended pending a reconfiguration.
212 * We hold a refcount over the call to ->make_request. By the time that
213 * call has finished, the bio has been linked into some internal structure
214 * and so is visible to ->quiesce(), so we don't need the refcount any more.
215 */
216 static int md_make_request(struct request_queue *q, struct bio *bio)
217 {
218 mddev_t *mddev = q->queuedata;
219 int rv;
220 if (mddev == NULL || mddev->pers == NULL) {
221 bio_io_error(bio);
222 return 0;
223 }
224 rcu_read_lock();
225 if (mddev->suspended || mddev->barrier) {
226 DEFINE_WAIT(__wait);
227 for (;;) {
228 prepare_to_wait(&mddev->sb_wait, &__wait,
229 TASK_UNINTERRUPTIBLE);
230 if (!mddev->suspended && !mddev->barrier)
231 break;
232 rcu_read_unlock();
233 schedule();
234 rcu_read_lock();
235 }
236 finish_wait(&mddev->sb_wait, &__wait);
237 }
238 atomic_inc(&mddev->active_io);
239 rcu_read_unlock();
240 rv = mddev->pers->make_request(q, bio);
241 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
242 wake_up(&mddev->sb_wait);
243
244 return rv;
245 }
246
247 static void mddev_suspend(mddev_t *mddev)
248 {
249 BUG_ON(mddev->suspended);
250 mddev->suspended = 1;
251 synchronize_rcu();
252 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
253 mddev->pers->quiesce(mddev, 1);
254 md_unregister_thread(mddev->thread);
255 mddev->thread = NULL;
256 /* we now know that no code is executing in the personality module,
257 * except possibly the tail end of a ->bi_end_io function, but that
258 * is certain to complete before the module has a chance to get
259 * unloaded
260 */
261 }
262
263 static void mddev_resume(mddev_t *mddev)
264 {
265 mddev->suspended = 0;
266 wake_up(&mddev->sb_wait);
267 mddev->pers->quiesce(mddev, 0);
268 }
269
270 int mddev_congested(mddev_t *mddev, int bits)
271 {
272 if (mddev->barrier)
273 return 1;
274 return mddev->suspended;
275 }
276 EXPORT_SYMBOL(mddev_congested);
277
278 /*
279 * Generic barrier handling for md
280 */
281
282 #define POST_REQUEST_BARRIER ((void*)1)
283
284 static void md_end_barrier(struct bio *bio, int err)
285 {
286 mdk_rdev_t *rdev = bio->bi_private;
287 mddev_t *mddev = rdev->mddev;
288 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
289 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
290
291 rdev_dec_pending(rdev, mddev);
292
293 if (atomic_dec_and_test(&mddev->flush_pending)) {
294 if (mddev->barrier == POST_REQUEST_BARRIER) {
295 /* This was a post-request barrier */
296 mddev->barrier = NULL;
297 wake_up(&mddev->sb_wait);
298 } else
299 /* The pre-request barrier has finished */
300 schedule_work(&mddev->barrier_work);
301 }
302 bio_put(bio);
303 }
304
305 static void submit_barriers(mddev_t *mddev)
306 {
307 mdk_rdev_t *rdev;
308
309 rcu_read_lock();
310 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
311 if (rdev->raid_disk >= 0 &&
312 !test_bit(Faulty, &rdev->flags)) {
313 /* Take two references, one is dropped
314 * when request finishes, one after
315 * we reclaim rcu_read_lock
316 */
317 struct bio *bi;
318 atomic_inc(&rdev->nr_pending);
319 atomic_inc(&rdev->nr_pending);
320 rcu_read_unlock();
321 bi = bio_alloc(GFP_KERNEL, 0);
322 bi->bi_end_io = md_end_barrier;
323 bi->bi_private = rdev;
324 bi->bi_bdev = rdev->bdev;
325 atomic_inc(&mddev->flush_pending);
326 submit_bio(WRITE_BARRIER, bi);
327 rcu_read_lock();
328 rdev_dec_pending(rdev, mddev);
329 }
330 rcu_read_unlock();
331 }
332
333 static void md_submit_barrier(struct work_struct *ws)
334 {
335 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
336 struct bio *bio = mddev->barrier;
337
338 atomic_set(&mddev->flush_pending, 1);
339
340 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
341 bio_endio(bio, -EOPNOTSUPP);
342 else if (bio->bi_size == 0)
343 /* an empty barrier - all done */
344 bio_endio(bio, 0);
345 else {
346 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
347 if (mddev->pers->make_request(mddev->queue, bio))
348 generic_make_request(bio);
349 mddev->barrier = POST_REQUEST_BARRIER;
350 submit_barriers(mddev);
351 }
352 if (atomic_dec_and_test(&mddev->flush_pending)) {
353 mddev->barrier = NULL;
354 wake_up(&mddev->sb_wait);
355 }
356 }
357
358 void md_barrier_request(mddev_t *mddev, struct bio *bio)
359 {
360 spin_lock_irq(&mddev->write_lock);
361 wait_event_lock_irq(mddev->sb_wait,
362 !mddev->barrier,
363 mddev->write_lock, /*nothing*/);
364 mddev->barrier = bio;
365 spin_unlock_irq(&mddev->write_lock);
366
367 atomic_set(&mddev->flush_pending, 1);
368 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
369
370 submit_barriers(mddev);
371
372 if (atomic_dec_and_test(&mddev->flush_pending))
373 schedule_work(&mddev->barrier_work);
374 }
375 EXPORT_SYMBOL(md_barrier_request);
376
377 static inline mddev_t *mddev_get(mddev_t *mddev)
378 {
379 atomic_inc(&mddev->active);
380 return mddev;
381 }
382
383 static void mddev_delayed_delete(struct work_struct *ws);
384
385 static void mddev_put(mddev_t *mddev)
386 {
387 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
388 return;
389 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
390 mddev->ctime == 0 && !mddev->hold_active) {
391 /* Array is not configured at all, and not held active,
392 * so destroy it */
393 list_del(&mddev->all_mddevs);
394 if (mddev->gendisk) {
395 /* we did a probe so need to clean up.
396 * Call schedule_work inside the spinlock
397 * so that flush_scheduled_work() after
398 * mddev_find will succeed in waiting for the
399 * work to be done.
400 */
401 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
402 schedule_work(&mddev->del_work);
403 } else
404 kfree(mddev);
405 }
406 spin_unlock(&all_mddevs_lock);
407 }
408
409 static mddev_t * mddev_find(dev_t unit)
410 {
411 mddev_t *mddev, *new = NULL;
412
413 retry:
414 spin_lock(&all_mddevs_lock);
415
416 if (unit) {
417 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
418 if (mddev->unit == unit) {
419 mddev_get(mddev);
420 spin_unlock(&all_mddevs_lock);
421 kfree(new);
422 return mddev;
423 }
424
425 if (new) {
426 list_add(&new->all_mddevs, &all_mddevs);
427 spin_unlock(&all_mddevs_lock);
428 new->hold_active = UNTIL_IOCTL;
429 return new;
430 }
431 } else if (new) {
432 /* find an unused unit number */
433 static int next_minor = 512;
434 int start = next_minor;
435 int is_free = 0;
436 int dev = 0;
437 while (!is_free) {
438 dev = MKDEV(MD_MAJOR, next_minor);
439 next_minor++;
440 if (next_minor > MINORMASK)
441 next_minor = 0;
442 if (next_minor == start) {
443 /* Oh dear, all in use. */
444 spin_unlock(&all_mddevs_lock);
445 kfree(new);
446 return NULL;
447 }
448
449 is_free = 1;
450 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
451 if (mddev->unit == dev) {
452 is_free = 0;
453 break;
454 }
455 }
456 new->unit = dev;
457 new->md_minor = MINOR(dev);
458 new->hold_active = UNTIL_STOP;
459 list_add(&new->all_mddevs, &all_mddevs);
460 spin_unlock(&all_mddevs_lock);
461 return new;
462 }
463 spin_unlock(&all_mddevs_lock);
464
465 new = kzalloc(sizeof(*new), GFP_KERNEL);
466 if (!new)
467 return NULL;
468
469 new->unit = unit;
470 if (MAJOR(unit) == MD_MAJOR)
471 new->md_minor = MINOR(unit);
472 else
473 new->md_minor = MINOR(unit) >> MdpMinorShift;
474
475 mutex_init(&new->open_mutex);
476 mutex_init(&new->reconfig_mutex);
477 mutex_init(&new->bitmap_info.mutex);
478 INIT_LIST_HEAD(&new->disks);
479 INIT_LIST_HEAD(&new->all_mddevs);
480 init_timer(&new->safemode_timer);
481 atomic_set(&new->active, 1);
482 atomic_set(&new->openers, 0);
483 atomic_set(&new->active_io, 0);
484 spin_lock_init(&new->write_lock);
485 atomic_set(&new->flush_pending, 0);
486 init_waitqueue_head(&new->sb_wait);
487 init_waitqueue_head(&new->recovery_wait);
488 new->reshape_position = MaxSector;
489 new->resync_min = 0;
490 new->resync_max = MaxSector;
491 new->level = LEVEL_NONE;
492
493 goto retry;
494 }
495
496 static inline int mddev_lock(mddev_t * mddev)
497 {
498 return mutex_lock_interruptible(&mddev->reconfig_mutex);
499 }
500
501 static inline int mddev_is_locked(mddev_t *mddev)
502 {
503 return mutex_is_locked(&mddev->reconfig_mutex);
504 }
505
506 static inline int mddev_trylock(mddev_t * mddev)
507 {
508 return mutex_trylock(&mddev->reconfig_mutex);
509 }
510
511 static inline void mddev_unlock(mddev_t * mddev)
512 {
513 mutex_unlock(&mddev->reconfig_mutex);
514
515 md_wakeup_thread(mddev->thread);
516 }
517
518 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
519 {
520 mdk_rdev_t *rdev;
521
522 list_for_each_entry(rdev, &mddev->disks, same_set)
523 if (rdev->desc_nr == nr)
524 return rdev;
525
526 return NULL;
527 }
528
529 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
530 {
531 mdk_rdev_t *rdev;
532
533 list_for_each_entry(rdev, &mddev->disks, same_set)
534 if (rdev->bdev->bd_dev == dev)
535 return rdev;
536
537 return NULL;
538 }
539
540 static struct mdk_personality *find_pers(int level, char *clevel)
541 {
542 struct mdk_personality *pers;
543 list_for_each_entry(pers, &pers_list, list) {
544 if (level != LEVEL_NONE && pers->level == level)
545 return pers;
546 if (strcmp(pers->name, clevel)==0)
547 return pers;
548 }
549 return NULL;
550 }
551
552 /* return the offset of the super block in 512byte sectors */
553 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
554 {
555 sector_t num_sectors = bdev->bd_inode->i_size / 512;
556 return MD_NEW_SIZE_SECTORS(num_sectors);
557 }
558
559 static int alloc_disk_sb(mdk_rdev_t * rdev)
560 {
561 if (rdev->sb_page)
562 MD_BUG();
563
564 rdev->sb_page = alloc_page(GFP_KERNEL);
565 if (!rdev->sb_page) {
566 printk(KERN_ALERT "md: out of memory.\n");
567 return -ENOMEM;
568 }
569
570 return 0;
571 }
572
573 static void free_disk_sb(mdk_rdev_t * rdev)
574 {
575 if (rdev->sb_page) {
576 put_page(rdev->sb_page);
577 rdev->sb_loaded = 0;
578 rdev->sb_page = NULL;
579 rdev->sb_start = 0;
580 rdev->sectors = 0;
581 }
582 }
583
584
585 static void super_written(struct bio *bio, int error)
586 {
587 mdk_rdev_t *rdev = bio->bi_private;
588 mddev_t *mddev = rdev->mddev;
589
590 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
591 printk("md: super_written gets error=%d, uptodate=%d\n",
592 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
593 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
594 md_error(mddev, rdev);
595 }
596
597 if (atomic_dec_and_test(&mddev->pending_writes))
598 wake_up(&mddev->sb_wait);
599 bio_put(bio);
600 }
601
602 static void super_written_barrier(struct bio *bio, int error)
603 {
604 struct bio *bio2 = bio->bi_private;
605 mdk_rdev_t *rdev = bio2->bi_private;
606 mddev_t *mddev = rdev->mddev;
607
608 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
609 error == -EOPNOTSUPP) {
610 unsigned long flags;
611 /* barriers don't appear to be supported :-( */
612 set_bit(BarriersNotsupp, &rdev->flags);
613 mddev->barriers_work = 0;
614 spin_lock_irqsave(&mddev->write_lock, flags);
615 bio2->bi_next = mddev->biolist;
616 mddev->biolist = bio2;
617 spin_unlock_irqrestore(&mddev->write_lock, flags);
618 wake_up(&mddev->sb_wait);
619 bio_put(bio);
620 } else {
621 bio_put(bio2);
622 bio->bi_private = rdev;
623 super_written(bio, error);
624 }
625 }
626
627 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
628 sector_t sector, int size, struct page *page)
629 {
630 /* write first size bytes of page to sector of rdev
631 * Increment mddev->pending_writes before returning
632 * and decrement it on completion, waking up sb_wait
633 * if zero is reached.
634 * If an error occurred, call md_error
635 *
636 * As we might need to resubmit the request if BIO_RW_BARRIER
637 * causes ENOTSUPP, we allocate a spare bio...
638 */
639 struct bio *bio = bio_alloc(GFP_NOIO, 1);
640 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
641
642 bio->bi_bdev = rdev->bdev;
643 bio->bi_sector = sector;
644 bio_add_page(bio, page, size, 0);
645 bio->bi_private = rdev;
646 bio->bi_end_io = super_written;
647 bio->bi_rw = rw;
648
649 atomic_inc(&mddev->pending_writes);
650 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
651 struct bio *rbio;
652 rw |= (1<<BIO_RW_BARRIER);
653 rbio = bio_clone(bio, GFP_NOIO);
654 rbio->bi_private = bio;
655 rbio->bi_end_io = super_written_barrier;
656 submit_bio(rw, rbio);
657 } else
658 submit_bio(rw, bio);
659 }
660
661 void md_super_wait(mddev_t *mddev)
662 {
663 /* wait for all superblock writes that were scheduled to complete.
664 * if any had to be retried (due to BARRIER problems), retry them
665 */
666 DEFINE_WAIT(wq);
667 for(;;) {
668 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
669 if (atomic_read(&mddev->pending_writes)==0)
670 break;
671 while (mddev->biolist) {
672 struct bio *bio;
673 spin_lock_irq(&mddev->write_lock);
674 bio = mddev->biolist;
675 mddev->biolist = bio->bi_next ;
676 bio->bi_next = NULL;
677 spin_unlock_irq(&mddev->write_lock);
678 submit_bio(bio->bi_rw, bio);
679 }
680 schedule();
681 }
682 finish_wait(&mddev->sb_wait, &wq);
683 }
684
685 static void bi_complete(struct bio *bio, int error)
686 {
687 complete((struct completion*)bio->bi_private);
688 }
689
690 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
691 struct page *page, int rw)
692 {
693 struct bio *bio = bio_alloc(GFP_NOIO, 1);
694 struct completion event;
695 int ret;
696
697 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
698
699 bio->bi_bdev = bdev;
700 bio->bi_sector = sector;
701 bio_add_page(bio, page, size, 0);
702 init_completion(&event);
703 bio->bi_private = &event;
704 bio->bi_end_io = bi_complete;
705 submit_bio(rw, bio);
706 wait_for_completion(&event);
707
708 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
709 bio_put(bio);
710 return ret;
711 }
712 EXPORT_SYMBOL_GPL(sync_page_io);
713
714 static int read_disk_sb(mdk_rdev_t * rdev, int size)
715 {
716 char b[BDEVNAME_SIZE];
717 if (!rdev->sb_page) {
718 MD_BUG();
719 return -EINVAL;
720 }
721 if (rdev->sb_loaded)
722 return 0;
723
724
725 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
726 goto fail;
727 rdev->sb_loaded = 1;
728 return 0;
729
730 fail:
731 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
732 bdevname(rdev->bdev,b));
733 return -EINVAL;
734 }
735
736 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
737 {
738 return sb1->set_uuid0 == sb2->set_uuid0 &&
739 sb1->set_uuid1 == sb2->set_uuid1 &&
740 sb1->set_uuid2 == sb2->set_uuid2 &&
741 sb1->set_uuid3 == sb2->set_uuid3;
742 }
743
744 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
745 {
746 int ret;
747 mdp_super_t *tmp1, *tmp2;
748
749 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
750 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
751
752 if (!tmp1 || !tmp2) {
753 ret = 0;
754 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
755 goto abort;
756 }
757
758 *tmp1 = *sb1;
759 *tmp2 = *sb2;
760
761 /*
762 * nr_disks is not constant
763 */
764 tmp1->nr_disks = 0;
765 tmp2->nr_disks = 0;
766
767 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
768 abort:
769 kfree(tmp1);
770 kfree(tmp2);
771 return ret;
772 }
773
774
775 static u32 md_csum_fold(u32 csum)
776 {
777 csum = (csum & 0xffff) + (csum >> 16);
778 return (csum & 0xffff) + (csum >> 16);
779 }
780
781 static unsigned int calc_sb_csum(mdp_super_t * sb)
782 {
783 u64 newcsum = 0;
784 u32 *sb32 = (u32*)sb;
785 int i;
786 unsigned int disk_csum, csum;
787
788 disk_csum = sb->sb_csum;
789 sb->sb_csum = 0;
790
791 for (i = 0; i < MD_SB_BYTES/4 ; i++)
792 newcsum += sb32[i];
793 csum = (newcsum & 0xffffffff) + (newcsum>>32);
794
795
796 #ifdef CONFIG_ALPHA
797 /* This used to use csum_partial, which was wrong for several
798 * reasons including that different results are returned on
799 * different architectures. It isn't critical that we get exactly
800 * the same return value as before (we always csum_fold before
801 * testing, and that removes any differences). However as we
802 * know that csum_partial always returned a 16bit value on
803 * alphas, do a fold to maximise conformity to previous behaviour.
804 */
805 sb->sb_csum = md_csum_fold(disk_csum);
806 #else
807 sb->sb_csum = disk_csum;
808 #endif
809 return csum;
810 }
811
812
813 /*
814 * Handle superblock details.
815 * We want to be able to handle multiple superblock formats
816 * so we have a common interface to them all, and an array of
817 * different handlers.
818 * We rely on user-space to write the initial superblock, and support
819 * reading and updating of superblocks.
820 * Interface methods are:
821 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
822 * loads and validates a superblock on dev.
823 * if refdev != NULL, compare superblocks on both devices
824 * Return:
825 * 0 - dev has a superblock that is compatible with refdev
826 * 1 - dev has a superblock that is compatible and newer than refdev
827 * so dev should be used as the refdev in future
828 * -EINVAL superblock incompatible or invalid
829 * -othererror e.g. -EIO
830 *
831 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
832 * Verify that dev is acceptable into mddev.
833 * The first time, mddev->raid_disks will be 0, and data from
834 * dev should be merged in. Subsequent calls check that dev
835 * is new enough. Return 0 or -EINVAL
836 *
837 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
838 * Update the superblock for rdev with data in mddev
839 * This does not write to disc.
840 *
841 */
842
843 struct super_type {
844 char *name;
845 struct module *owner;
846 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
847 int minor_version);
848 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
849 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
850 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
851 sector_t num_sectors);
852 };
853
854 /*
855 * Check that the given mddev has no bitmap.
856 *
857 * This function is called from the run method of all personalities that do not
858 * support bitmaps. It prints an error message and returns non-zero if mddev
859 * has a bitmap. Otherwise, it returns 0.
860 *
861 */
862 int md_check_no_bitmap(mddev_t *mddev)
863 {
864 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
865 return 0;
866 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
867 mdname(mddev), mddev->pers->name);
868 return 1;
869 }
870 EXPORT_SYMBOL(md_check_no_bitmap);
871
872 /*
873 * load_super for 0.90.0
874 */
875 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
876 {
877 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
878 mdp_super_t *sb;
879 int ret;
880
881 /*
882 * Calculate the position of the superblock (512byte sectors),
883 * it's at the end of the disk.
884 *
885 * It also happens to be a multiple of 4Kb.
886 */
887 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
888
889 ret = read_disk_sb(rdev, MD_SB_BYTES);
890 if (ret) return ret;
891
892 ret = -EINVAL;
893
894 bdevname(rdev->bdev, b);
895 sb = (mdp_super_t*)page_address(rdev->sb_page);
896
897 if (sb->md_magic != MD_SB_MAGIC) {
898 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
899 b);
900 goto abort;
901 }
902
903 if (sb->major_version != 0 ||
904 sb->minor_version < 90 ||
905 sb->minor_version > 91) {
906 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
907 sb->major_version, sb->minor_version,
908 b);
909 goto abort;
910 }
911
912 if (sb->raid_disks <= 0)
913 goto abort;
914
915 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
916 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
917 b);
918 goto abort;
919 }
920
921 rdev->preferred_minor = sb->md_minor;
922 rdev->data_offset = 0;
923 rdev->sb_size = MD_SB_BYTES;
924
925 if (sb->level == LEVEL_MULTIPATH)
926 rdev->desc_nr = -1;
927 else
928 rdev->desc_nr = sb->this_disk.number;
929
930 if (!refdev) {
931 ret = 1;
932 } else {
933 __u64 ev1, ev2;
934 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
935 if (!uuid_equal(refsb, sb)) {
936 printk(KERN_WARNING "md: %s has different UUID to %s\n",
937 b, bdevname(refdev->bdev,b2));
938 goto abort;
939 }
940 if (!sb_equal(refsb, sb)) {
941 printk(KERN_WARNING "md: %s has same UUID"
942 " but different superblock to %s\n",
943 b, bdevname(refdev->bdev, b2));
944 goto abort;
945 }
946 ev1 = md_event(sb);
947 ev2 = md_event(refsb);
948 if (ev1 > ev2)
949 ret = 1;
950 else
951 ret = 0;
952 }
953 rdev->sectors = rdev->sb_start;
954
955 if (rdev->sectors < sb->size * 2 && sb->level > 1)
956 /* "this cannot possibly happen" ... */
957 ret = -EINVAL;
958
959 abort:
960 return ret;
961 }
962
963 /*
964 * validate_super for 0.90.0
965 */
966 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
967 {
968 mdp_disk_t *desc;
969 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
970 __u64 ev1 = md_event(sb);
971
972 rdev->raid_disk = -1;
973 clear_bit(Faulty, &rdev->flags);
974 clear_bit(In_sync, &rdev->flags);
975 clear_bit(WriteMostly, &rdev->flags);
976 clear_bit(BarriersNotsupp, &rdev->flags);
977
978 if (mddev->raid_disks == 0) {
979 mddev->major_version = 0;
980 mddev->minor_version = sb->minor_version;
981 mddev->patch_version = sb->patch_version;
982 mddev->external = 0;
983 mddev->chunk_sectors = sb->chunk_size >> 9;
984 mddev->ctime = sb->ctime;
985 mddev->utime = sb->utime;
986 mddev->level = sb->level;
987 mddev->clevel[0] = 0;
988 mddev->layout = sb->layout;
989 mddev->raid_disks = sb->raid_disks;
990 mddev->dev_sectors = sb->size * 2;
991 mddev->events = ev1;
992 mddev->bitmap_info.offset = 0;
993 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
994
995 if (mddev->minor_version >= 91) {
996 mddev->reshape_position = sb->reshape_position;
997 mddev->delta_disks = sb->delta_disks;
998 mddev->new_level = sb->new_level;
999 mddev->new_layout = sb->new_layout;
1000 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1001 } else {
1002 mddev->reshape_position = MaxSector;
1003 mddev->delta_disks = 0;
1004 mddev->new_level = mddev->level;
1005 mddev->new_layout = mddev->layout;
1006 mddev->new_chunk_sectors = mddev->chunk_sectors;
1007 }
1008
1009 if (sb->state & (1<<MD_SB_CLEAN))
1010 mddev->recovery_cp = MaxSector;
1011 else {
1012 if (sb->events_hi == sb->cp_events_hi &&
1013 sb->events_lo == sb->cp_events_lo) {
1014 mddev->recovery_cp = sb->recovery_cp;
1015 } else
1016 mddev->recovery_cp = 0;
1017 }
1018
1019 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1020 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1021 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1022 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1023
1024 mddev->max_disks = MD_SB_DISKS;
1025
1026 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1027 mddev->bitmap_info.file == NULL)
1028 mddev->bitmap_info.offset =
1029 mddev->bitmap_info.default_offset;
1030
1031 } else if (mddev->pers == NULL) {
1032 /* Insist on good event counter while assembling */
1033 ++ev1;
1034 if (ev1 < mddev->events)
1035 return -EINVAL;
1036 } else if (mddev->bitmap) {
1037 /* if adding to array with a bitmap, then we can accept an
1038 * older device ... but not too old.
1039 */
1040 if (ev1 < mddev->bitmap->events_cleared)
1041 return 0;
1042 } else {
1043 if (ev1 < mddev->events)
1044 /* just a hot-add of a new device, leave raid_disk at -1 */
1045 return 0;
1046 }
1047
1048 if (mddev->level != LEVEL_MULTIPATH) {
1049 desc = sb->disks + rdev->desc_nr;
1050
1051 if (desc->state & (1<<MD_DISK_FAULTY))
1052 set_bit(Faulty, &rdev->flags);
1053 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1054 desc->raid_disk < mddev->raid_disks */) {
1055 set_bit(In_sync, &rdev->flags);
1056 rdev->raid_disk = desc->raid_disk;
1057 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1058 /* active but not in sync implies recovery up to
1059 * reshape position. We don't know exactly where
1060 * that is, so set to zero for now */
1061 if (mddev->minor_version >= 91) {
1062 rdev->recovery_offset = 0;
1063 rdev->raid_disk = desc->raid_disk;
1064 }
1065 }
1066 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1067 set_bit(WriteMostly, &rdev->flags);
1068 } else /* MULTIPATH are always insync */
1069 set_bit(In_sync, &rdev->flags);
1070 return 0;
1071 }
1072
1073 /*
1074 * sync_super for 0.90.0
1075 */
1076 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1077 {
1078 mdp_super_t *sb;
1079 mdk_rdev_t *rdev2;
1080 int next_spare = mddev->raid_disks;
1081
1082
1083 /* make rdev->sb match mddev data..
1084 *
1085 * 1/ zero out disks
1086 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1087 * 3/ any empty disks < next_spare become removed
1088 *
1089 * disks[0] gets initialised to REMOVED because
1090 * we cannot be sure from other fields if it has
1091 * been initialised or not.
1092 */
1093 int i;
1094 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1095
1096 rdev->sb_size = MD_SB_BYTES;
1097
1098 sb = (mdp_super_t*)page_address(rdev->sb_page);
1099
1100 memset(sb, 0, sizeof(*sb));
1101
1102 sb->md_magic = MD_SB_MAGIC;
1103 sb->major_version = mddev->major_version;
1104 sb->patch_version = mddev->patch_version;
1105 sb->gvalid_words = 0; /* ignored */
1106 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1107 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1108 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1109 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1110
1111 sb->ctime = mddev->ctime;
1112 sb->level = mddev->level;
1113 sb->size = mddev->dev_sectors / 2;
1114 sb->raid_disks = mddev->raid_disks;
1115 sb->md_minor = mddev->md_minor;
1116 sb->not_persistent = 0;
1117 sb->utime = mddev->utime;
1118 sb->state = 0;
1119 sb->events_hi = (mddev->events>>32);
1120 sb->events_lo = (u32)mddev->events;
1121
1122 if (mddev->reshape_position == MaxSector)
1123 sb->minor_version = 90;
1124 else {
1125 sb->minor_version = 91;
1126 sb->reshape_position = mddev->reshape_position;
1127 sb->new_level = mddev->new_level;
1128 sb->delta_disks = mddev->delta_disks;
1129 sb->new_layout = mddev->new_layout;
1130 sb->new_chunk = mddev->new_chunk_sectors << 9;
1131 }
1132 mddev->minor_version = sb->minor_version;
1133 if (mddev->in_sync)
1134 {
1135 sb->recovery_cp = mddev->recovery_cp;
1136 sb->cp_events_hi = (mddev->events>>32);
1137 sb->cp_events_lo = (u32)mddev->events;
1138 if (mddev->recovery_cp == MaxSector)
1139 sb->state = (1<< MD_SB_CLEAN);
1140 } else
1141 sb->recovery_cp = 0;
1142
1143 sb->layout = mddev->layout;
1144 sb->chunk_size = mddev->chunk_sectors << 9;
1145
1146 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1147 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1148
1149 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1150 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1151 mdp_disk_t *d;
1152 int desc_nr;
1153 int is_active = test_bit(In_sync, &rdev2->flags);
1154
1155 if (rdev2->raid_disk >= 0 &&
1156 sb->minor_version >= 91)
1157 /* we have nowhere to store the recovery_offset,
1158 * but if it is not below the reshape_position,
1159 * we can piggy-back on that.
1160 */
1161 is_active = 1;
1162 if (rdev2->raid_disk < 0 ||
1163 test_bit(Faulty, &rdev2->flags))
1164 is_active = 0;
1165 if (is_active)
1166 desc_nr = rdev2->raid_disk;
1167 else
1168 desc_nr = next_spare++;
1169 rdev2->desc_nr = desc_nr;
1170 d = &sb->disks[rdev2->desc_nr];
1171 nr_disks++;
1172 d->number = rdev2->desc_nr;
1173 d->major = MAJOR(rdev2->bdev->bd_dev);
1174 d->minor = MINOR(rdev2->bdev->bd_dev);
1175 if (is_active)
1176 d->raid_disk = rdev2->raid_disk;
1177 else
1178 d->raid_disk = rdev2->desc_nr; /* compatibility */
1179 if (test_bit(Faulty, &rdev2->flags))
1180 d->state = (1<<MD_DISK_FAULTY);
1181 else if (is_active) {
1182 d->state = (1<<MD_DISK_ACTIVE);
1183 if (test_bit(In_sync, &rdev2->flags))
1184 d->state |= (1<<MD_DISK_SYNC);
1185 active++;
1186 working++;
1187 } else {
1188 d->state = 0;
1189 spare++;
1190 working++;
1191 }
1192 if (test_bit(WriteMostly, &rdev2->flags))
1193 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1194 }
1195 /* now set the "removed" and "faulty" bits on any missing devices */
1196 for (i=0 ; i < mddev->raid_disks ; i++) {
1197 mdp_disk_t *d = &sb->disks[i];
1198 if (d->state == 0 && d->number == 0) {
1199 d->number = i;
1200 d->raid_disk = i;
1201 d->state = (1<<MD_DISK_REMOVED);
1202 d->state |= (1<<MD_DISK_FAULTY);
1203 failed++;
1204 }
1205 }
1206 sb->nr_disks = nr_disks;
1207 sb->active_disks = active;
1208 sb->working_disks = working;
1209 sb->failed_disks = failed;
1210 sb->spare_disks = spare;
1211
1212 sb->this_disk = sb->disks[rdev->desc_nr];
1213 sb->sb_csum = calc_sb_csum(sb);
1214 }
1215
1216 /*
1217 * rdev_size_change for 0.90.0
1218 */
1219 static unsigned long long
1220 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1221 {
1222 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1223 return 0; /* component must fit device */
1224 if (rdev->mddev->bitmap_info.offset)
1225 return 0; /* can't move bitmap */
1226 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1227 if (!num_sectors || num_sectors > rdev->sb_start)
1228 num_sectors = rdev->sb_start;
1229 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1230 rdev->sb_page);
1231 md_super_wait(rdev->mddev);
1232 return num_sectors / 2; /* kB for sysfs */
1233 }
1234
1235
1236 /*
1237 * version 1 superblock
1238 */
1239
1240 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1241 {
1242 __le32 disk_csum;
1243 u32 csum;
1244 unsigned long long newcsum;
1245 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1246 __le32 *isuper = (__le32*)sb;
1247 int i;
1248
1249 disk_csum = sb->sb_csum;
1250 sb->sb_csum = 0;
1251 newcsum = 0;
1252 for (i=0; size>=4; size -= 4 )
1253 newcsum += le32_to_cpu(*isuper++);
1254
1255 if (size == 2)
1256 newcsum += le16_to_cpu(*(__le16*) isuper);
1257
1258 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1259 sb->sb_csum = disk_csum;
1260 return cpu_to_le32(csum);
1261 }
1262
1263 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1264 {
1265 struct mdp_superblock_1 *sb;
1266 int ret;
1267 sector_t sb_start;
1268 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1269 int bmask;
1270
1271 /*
1272 * Calculate the position of the superblock in 512byte sectors.
1273 * It is always aligned to a 4K boundary and
1274 * depeding on minor_version, it can be:
1275 * 0: At least 8K, but less than 12K, from end of device
1276 * 1: At start of device
1277 * 2: 4K from start of device.
1278 */
1279 switch(minor_version) {
1280 case 0:
1281 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1282 sb_start -= 8*2;
1283 sb_start &= ~(sector_t)(4*2-1);
1284 break;
1285 case 1:
1286 sb_start = 0;
1287 break;
1288 case 2:
1289 sb_start = 8;
1290 break;
1291 default:
1292 return -EINVAL;
1293 }
1294 rdev->sb_start = sb_start;
1295
1296 /* superblock is rarely larger than 1K, but it can be larger,
1297 * and it is safe to read 4k, so we do that
1298 */
1299 ret = read_disk_sb(rdev, 4096);
1300 if (ret) return ret;
1301
1302
1303 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1304
1305 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1306 sb->major_version != cpu_to_le32(1) ||
1307 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1308 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1309 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1310 return -EINVAL;
1311
1312 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1313 printk("md: invalid superblock checksum on %s\n",
1314 bdevname(rdev->bdev,b));
1315 return -EINVAL;
1316 }
1317 if (le64_to_cpu(sb->data_size) < 10) {
1318 printk("md: data_size too small on %s\n",
1319 bdevname(rdev->bdev,b));
1320 return -EINVAL;
1321 }
1322
1323 rdev->preferred_minor = 0xffff;
1324 rdev->data_offset = le64_to_cpu(sb->data_offset);
1325 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1326
1327 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1328 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1329 if (rdev->sb_size & bmask)
1330 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1331
1332 if (minor_version
1333 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1334 return -EINVAL;
1335
1336 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1337 rdev->desc_nr = -1;
1338 else
1339 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1340
1341 if (!refdev) {
1342 ret = 1;
1343 } else {
1344 __u64 ev1, ev2;
1345 struct mdp_superblock_1 *refsb =
1346 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1347
1348 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1349 sb->level != refsb->level ||
1350 sb->layout != refsb->layout ||
1351 sb->chunksize != refsb->chunksize) {
1352 printk(KERN_WARNING "md: %s has strangely different"
1353 " superblock to %s\n",
1354 bdevname(rdev->bdev,b),
1355 bdevname(refdev->bdev,b2));
1356 return -EINVAL;
1357 }
1358 ev1 = le64_to_cpu(sb->events);
1359 ev2 = le64_to_cpu(refsb->events);
1360
1361 if (ev1 > ev2)
1362 ret = 1;
1363 else
1364 ret = 0;
1365 }
1366 if (minor_version)
1367 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1368 le64_to_cpu(sb->data_offset);
1369 else
1370 rdev->sectors = rdev->sb_start;
1371 if (rdev->sectors < le64_to_cpu(sb->data_size))
1372 return -EINVAL;
1373 rdev->sectors = le64_to_cpu(sb->data_size);
1374 if (le64_to_cpu(sb->size) > rdev->sectors)
1375 return -EINVAL;
1376 return ret;
1377 }
1378
1379 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1380 {
1381 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1382 __u64 ev1 = le64_to_cpu(sb->events);
1383
1384 rdev->raid_disk = -1;
1385 clear_bit(Faulty, &rdev->flags);
1386 clear_bit(In_sync, &rdev->flags);
1387 clear_bit(WriteMostly, &rdev->flags);
1388 clear_bit(BarriersNotsupp, &rdev->flags);
1389
1390 if (mddev->raid_disks == 0) {
1391 mddev->major_version = 1;
1392 mddev->patch_version = 0;
1393 mddev->external = 0;
1394 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1395 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1396 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1397 mddev->level = le32_to_cpu(sb->level);
1398 mddev->clevel[0] = 0;
1399 mddev->layout = le32_to_cpu(sb->layout);
1400 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1401 mddev->dev_sectors = le64_to_cpu(sb->size);
1402 mddev->events = ev1;
1403 mddev->bitmap_info.offset = 0;
1404 mddev->bitmap_info.default_offset = 1024 >> 9;
1405
1406 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1407 memcpy(mddev->uuid, sb->set_uuid, 16);
1408
1409 mddev->max_disks = (4096-256)/2;
1410
1411 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1412 mddev->bitmap_info.file == NULL )
1413 mddev->bitmap_info.offset =
1414 (__s32)le32_to_cpu(sb->bitmap_offset);
1415
1416 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1417 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1418 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1419 mddev->new_level = le32_to_cpu(sb->new_level);
1420 mddev->new_layout = le32_to_cpu(sb->new_layout);
1421 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1422 } else {
1423 mddev->reshape_position = MaxSector;
1424 mddev->delta_disks = 0;
1425 mddev->new_level = mddev->level;
1426 mddev->new_layout = mddev->layout;
1427 mddev->new_chunk_sectors = mddev->chunk_sectors;
1428 }
1429
1430 } else if (mddev->pers == NULL) {
1431 /* Insist of good event counter while assembling */
1432 ++ev1;
1433 if (ev1 < mddev->events)
1434 return -EINVAL;
1435 } else if (mddev->bitmap) {
1436 /* If adding to array with a bitmap, then we can accept an
1437 * older device, but not too old.
1438 */
1439 if (ev1 < mddev->bitmap->events_cleared)
1440 return 0;
1441 } else {
1442 if (ev1 < mddev->events)
1443 /* just a hot-add of a new device, leave raid_disk at -1 */
1444 return 0;
1445 }
1446 if (mddev->level != LEVEL_MULTIPATH) {
1447 int role;
1448 if (rdev->desc_nr < 0 ||
1449 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1450 role = 0xffff;
1451 rdev->desc_nr = -1;
1452 } else
1453 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1454 switch(role) {
1455 case 0xffff: /* spare */
1456 break;
1457 case 0xfffe: /* faulty */
1458 set_bit(Faulty, &rdev->flags);
1459 break;
1460 default:
1461 if ((le32_to_cpu(sb->feature_map) &
1462 MD_FEATURE_RECOVERY_OFFSET))
1463 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1464 else
1465 set_bit(In_sync, &rdev->flags);
1466 rdev->raid_disk = role;
1467 break;
1468 }
1469 if (sb->devflags & WriteMostly1)
1470 set_bit(WriteMostly, &rdev->flags);
1471 } else /* MULTIPATH are always insync */
1472 set_bit(In_sync, &rdev->flags);
1473
1474 return 0;
1475 }
1476
1477 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1478 {
1479 struct mdp_superblock_1 *sb;
1480 mdk_rdev_t *rdev2;
1481 int max_dev, i;
1482 /* make rdev->sb match mddev and rdev data. */
1483
1484 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1485
1486 sb->feature_map = 0;
1487 sb->pad0 = 0;
1488 sb->recovery_offset = cpu_to_le64(0);
1489 memset(sb->pad1, 0, sizeof(sb->pad1));
1490 memset(sb->pad2, 0, sizeof(sb->pad2));
1491 memset(sb->pad3, 0, sizeof(sb->pad3));
1492
1493 sb->utime = cpu_to_le64((__u64)mddev->utime);
1494 sb->events = cpu_to_le64(mddev->events);
1495 if (mddev->in_sync)
1496 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1497 else
1498 sb->resync_offset = cpu_to_le64(0);
1499
1500 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1501
1502 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1503 sb->size = cpu_to_le64(mddev->dev_sectors);
1504 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1505 sb->level = cpu_to_le32(mddev->level);
1506 sb->layout = cpu_to_le32(mddev->layout);
1507
1508 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1509 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1510 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1511 }
1512
1513 if (rdev->raid_disk >= 0 &&
1514 !test_bit(In_sync, &rdev->flags)) {
1515 sb->feature_map |=
1516 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1517 sb->recovery_offset =
1518 cpu_to_le64(rdev->recovery_offset);
1519 }
1520
1521 if (mddev->reshape_position != MaxSector) {
1522 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1523 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1524 sb->new_layout = cpu_to_le32(mddev->new_layout);
1525 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1526 sb->new_level = cpu_to_le32(mddev->new_level);
1527 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1528 }
1529
1530 max_dev = 0;
1531 list_for_each_entry(rdev2, &mddev->disks, same_set)
1532 if (rdev2->desc_nr+1 > max_dev)
1533 max_dev = rdev2->desc_nr+1;
1534
1535 if (max_dev > le32_to_cpu(sb->max_dev)) {
1536 int bmask;
1537 sb->max_dev = cpu_to_le32(max_dev);
1538 rdev->sb_size = max_dev * 2 + 256;
1539 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1540 if (rdev->sb_size & bmask)
1541 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1542 }
1543 for (i=0; i<max_dev;i++)
1544 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1545
1546 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1547 i = rdev2->desc_nr;
1548 if (test_bit(Faulty, &rdev2->flags))
1549 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1550 else if (test_bit(In_sync, &rdev2->flags))
1551 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1552 else if (rdev2->raid_disk >= 0)
1553 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1554 else
1555 sb->dev_roles[i] = cpu_to_le16(0xffff);
1556 }
1557
1558 sb->sb_csum = calc_sb_1_csum(sb);
1559 }
1560
1561 static unsigned long long
1562 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1563 {
1564 struct mdp_superblock_1 *sb;
1565 sector_t max_sectors;
1566 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1567 return 0; /* component must fit device */
1568 if (rdev->sb_start < rdev->data_offset) {
1569 /* minor versions 1 and 2; superblock before data */
1570 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1571 max_sectors -= rdev->data_offset;
1572 if (!num_sectors || num_sectors > max_sectors)
1573 num_sectors = max_sectors;
1574 } else if (rdev->mddev->bitmap_info.offset) {
1575 /* minor version 0 with bitmap we can't move */
1576 return 0;
1577 } else {
1578 /* minor version 0; superblock after data */
1579 sector_t sb_start;
1580 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1581 sb_start &= ~(sector_t)(4*2 - 1);
1582 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1583 if (!num_sectors || num_sectors > max_sectors)
1584 num_sectors = max_sectors;
1585 rdev->sb_start = sb_start;
1586 }
1587 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1588 sb->data_size = cpu_to_le64(num_sectors);
1589 sb->super_offset = rdev->sb_start;
1590 sb->sb_csum = calc_sb_1_csum(sb);
1591 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1592 rdev->sb_page);
1593 md_super_wait(rdev->mddev);
1594 return num_sectors / 2; /* kB for sysfs */
1595 }
1596
1597 static struct super_type super_types[] = {
1598 [0] = {
1599 .name = "0.90.0",
1600 .owner = THIS_MODULE,
1601 .load_super = super_90_load,
1602 .validate_super = super_90_validate,
1603 .sync_super = super_90_sync,
1604 .rdev_size_change = super_90_rdev_size_change,
1605 },
1606 [1] = {
1607 .name = "md-1",
1608 .owner = THIS_MODULE,
1609 .load_super = super_1_load,
1610 .validate_super = super_1_validate,
1611 .sync_super = super_1_sync,
1612 .rdev_size_change = super_1_rdev_size_change,
1613 },
1614 };
1615
1616 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1617 {
1618 mdk_rdev_t *rdev, *rdev2;
1619
1620 rcu_read_lock();
1621 rdev_for_each_rcu(rdev, mddev1)
1622 rdev_for_each_rcu(rdev2, mddev2)
1623 if (rdev->bdev->bd_contains ==
1624 rdev2->bdev->bd_contains) {
1625 rcu_read_unlock();
1626 return 1;
1627 }
1628 rcu_read_unlock();
1629 return 0;
1630 }
1631
1632 static LIST_HEAD(pending_raid_disks);
1633
1634 /*
1635 * Try to register data integrity profile for an mddev
1636 *
1637 * This is called when an array is started and after a disk has been kicked
1638 * from the array. It only succeeds if all working and active component devices
1639 * are integrity capable with matching profiles.
1640 */
1641 int md_integrity_register(mddev_t *mddev)
1642 {
1643 mdk_rdev_t *rdev, *reference = NULL;
1644
1645 if (list_empty(&mddev->disks))
1646 return 0; /* nothing to do */
1647 if (blk_get_integrity(mddev->gendisk))
1648 return 0; /* already registered */
1649 list_for_each_entry(rdev, &mddev->disks, same_set) {
1650 /* skip spares and non-functional disks */
1651 if (test_bit(Faulty, &rdev->flags))
1652 continue;
1653 if (rdev->raid_disk < 0)
1654 continue;
1655 /*
1656 * If at least one rdev is not integrity capable, we can not
1657 * enable data integrity for the md device.
1658 */
1659 if (!bdev_get_integrity(rdev->bdev))
1660 return -EINVAL;
1661 if (!reference) {
1662 /* Use the first rdev as the reference */
1663 reference = rdev;
1664 continue;
1665 }
1666 /* does this rdev's profile match the reference profile? */
1667 if (blk_integrity_compare(reference->bdev->bd_disk,
1668 rdev->bdev->bd_disk) < 0)
1669 return -EINVAL;
1670 }
1671 /*
1672 * All component devices are integrity capable and have matching
1673 * profiles, register the common profile for the md device.
1674 */
1675 if (blk_integrity_register(mddev->gendisk,
1676 bdev_get_integrity(reference->bdev)) != 0) {
1677 printk(KERN_ERR "md: failed to register integrity for %s\n",
1678 mdname(mddev));
1679 return -EINVAL;
1680 }
1681 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1682 mdname(mddev));
1683 return 0;
1684 }
1685 EXPORT_SYMBOL(md_integrity_register);
1686
1687 /* Disable data integrity if non-capable/non-matching disk is being added */
1688 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1689 {
1690 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1691 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1692
1693 if (!bi_mddev) /* nothing to do */
1694 return;
1695 if (rdev->raid_disk < 0) /* skip spares */
1696 return;
1697 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1698 rdev->bdev->bd_disk) >= 0)
1699 return;
1700 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1701 blk_integrity_unregister(mddev->gendisk);
1702 }
1703 EXPORT_SYMBOL(md_integrity_add_rdev);
1704
1705 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1706 {
1707 char b[BDEVNAME_SIZE];
1708 struct kobject *ko;
1709 char *s;
1710 int err;
1711
1712 if (rdev->mddev) {
1713 MD_BUG();
1714 return -EINVAL;
1715 }
1716
1717 /* prevent duplicates */
1718 if (find_rdev(mddev, rdev->bdev->bd_dev))
1719 return -EEXIST;
1720
1721 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1722 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1723 rdev->sectors < mddev->dev_sectors)) {
1724 if (mddev->pers) {
1725 /* Cannot change size, so fail
1726 * If mddev->level <= 0, then we don't care
1727 * about aligning sizes (e.g. linear)
1728 */
1729 if (mddev->level > 0)
1730 return -ENOSPC;
1731 } else
1732 mddev->dev_sectors = rdev->sectors;
1733 }
1734
1735 /* Verify rdev->desc_nr is unique.
1736 * If it is -1, assign a free number, else
1737 * check number is not in use
1738 */
1739 if (rdev->desc_nr < 0) {
1740 int choice = 0;
1741 if (mddev->pers) choice = mddev->raid_disks;
1742 while (find_rdev_nr(mddev, choice))
1743 choice++;
1744 rdev->desc_nr = choice;
1745 } else {
1746 if (find_rdev_nr(mddev, rdev->desc_nr))
1747 return -EBUSY;
1748 }
1749 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1750 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1751 mdname(mddev), mddev->max_disks);
1752 return -EBUSY;
1753 }
1754 bdevname(rdev->bdev,b);
1755 while ( (s=strchr(b, '/')) != NULL)
1756 *s = '!';
1757
1758 rdev->mddev = mddev;
1759 printk(KERN_INFO "md: bind<%s>\n", b);
1760
1761 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1762 goto fail;
1763
1764 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1765 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1766 kobject_del(&rdev->kobj);
1767 goto fail;
1768 }
1769 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1770
1771 list_add_rcu(&rdev->same_set, &mddev->disks);
1772 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1773
1774 /* May as well allow recovery to be retried once */
1775 mddev->recovery_disabled = 0;
1776
1777 return 0;
1778
1779 fail:
1780 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1781 b, mdname(mddev));
1782 return err;
1783 }
1784
1785 static void md_delayed_delete(struct work_struct *ws)
1786 {
1787 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1788 kobject_del(&rdev->kobj);
1789 kobject_put(&rdev->kobj);
1790 }
1791
1792 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1793 {
1794 char b[BDEVNAME_SIZE];
1795 if (!rdev->mddev) {
1796 MD_BUG();
1797 return;
1798 }
1799 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1800 list_del_rcu(&rdev->same_set);
1801 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1802 rdev->mddev = NULL;
1803 sysfs_remove_link(&rdev->kobj, "block");
1804 sysfs_put(rdev->sysfs_state);
1805 rdev->sysfs_state = NULL;
1806 /* We need to delay this, otherwise we can deadlock when
1807 * writing to 'remove' to "dev/state". We also need
1808 * to delay it due to rcu usage.
1809 */
1810 synchronize_rcu();
1811 INIT_WORK(&rdev->del_work, md_delayed_delete);
1812 kobject_get(&rdev->kobj);
1813 schedule_work(&rdev->del_work);
1814 }
1815
1816 /*
1817 * prevent the device from being mounted, repartitioned or
1818 * otherwise reused by a RAID array (or any other kernel
1819 * subsystem), by bd_claiming the device.
1820 */
1821 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1822 {
1823 int err = 0;
1824 struct block_device *bdev;
1825 char b[BDEVNAME_SIZE];
1826
1827 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1828 if (IS_ERR(bdev)) {
1829 printk(KERN_ERR "md: could not open %s.\n",
1830 __bdevname(dev, b));
1831 return PTR_ERR(bdev);
1832 }
1833 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1834 if (err) {
1835 printk(KERN_ERR "md: could not bd_claim %s.\n",
1836 bdevname(bdev, b));
1837 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1838 return err;
1839 }
1840 if (!shared)
1841 set_bit(AllReserved, &rdev->flags);
1842 rdev->bdev = bdev;
1843 return err;
1844 }
1845
1846 static void unlock_rdev(mdk_rdev_t *rdev)
1847 {
1848 struct block_device *bdev = rdev->bdev;
1849 rdev->bdev = NULL;
1850 if (!bdev)
1851 MD_BUG();
1852 bd_release(bdev);
1853 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1854 }
1855
1856 void md_autodetect_dev(dev_t dev);
1857
1858 static void export_rdev(mdk_rdev_t * rdev)
1859 {
1860 char b[BDEVNAME_SIZE];
1861 printk(KERN_INFO "md: export_rdev(%s)\n",
1862 bdevname(rdev->bdev,b));
1863 if (rdev->mddev)
1864 MD_BUG();
1865 free_disk_sb(rdev);
1866 #ifndef MODULE
1867 if (test_bit(AutoDetected, &rdev->flags))
1868 md_autodetect_dev(rdev->bdev->bd_dev);
1869 #endif
1870 unlock_rdev(rdev);
1871 kobject_put(&rdev->kobj);
1872 }
1873
1874 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1875 {
1876 unbind_rdev_from_array(rdev);
1877 export_rdev(rdev);
1878 }
1879
1880 static void export_array(mddev_t *mddev)
1881 {
1882 mdk_rdev_t *rdev, *tmp;
1883
1884 rdev_for_each(rdev, tmp, mddev) {
1885 if (!rdev->mddev) {
1886 MD_BUG();
1887 continue;
1888 }
1889 kick_rdev_from_array(rdev);
1890 }
1891 if (!list_empty(&mddev->disks))
1892 MD_BUG();
1893 mddev->raid_disks = 0;
1894 mddev->major_version = 0;
1895 }
1896
1897 static void print_desc(mdp_disk_t *desc)
1898 {
1899 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1900 desc->major,desc->minor,desc->raid_disk,desc->state);
1901 }
1902
1903 static void print_sb_90(mdp_super_t *sb)
1904 {
1905 int i;
1906
1907 printk(KERN_INFO
1908 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1909 sb->major_version, sb->minor_version, sb->patch_version,
1910 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1911 sb->ctime);
1912 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1913 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1914 sb->md_minor, sb->layout, sb->chunk_size);
1915 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1916 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1917 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1918 sb->failed_disks, sb->spare_disks,
1919 sb->sb_csum, (unsigned long)sb->events_lo);
1920
1921 printk(KERN_INFO);
1922 for (i = 0; i < MD_SB_DISKS; i++) {
1923 mdp_disk_t *desc;
1924
1925 desc = sb->disks + i;
1926 if (desc->number || desc->major || desc->minor ||
1927 desc->raid_disk || (desc->state && (desc->state != 4))) {
1928 printk(" D %2d: ", i);
1929 print_desc(desc);
1930 }
1931 }
1932 printk(KERN_INFO "md: THIS: ");
1933 print_desc(&sb->this_disk);
1934 }
1935
1936 static void print_sb_1(struct mdp_superblock_1 *sb)
1937 {
1938 __u8 *uuid;
1939
1940 uuid = sb->set_uuid;
1941 printk(KERN_INFO
1942 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1943 "md: Name: \"%s\" CT:%llu\n",
1944 le32_to_cpu(sb->major_version),
1945 le32_to_cpu(sb->feature_map),
1946 uuid,
1947 sb->set_name,
1948 (unsigned long long)le64_to_cpu(sb->ctime)
1949 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1950
1951 uuid = sb->device_uuid;
1952 printk(KERN_INFO
1953 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1954 " RO:%llu\n"
1955 "md: Dev:%08x UUID: %pU\n"
1956 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1957 "md: (MaxDev:%u) \n",
1958 le32_to_cpu(sb->level),
1959 (unsigned long long)le64_to_cpu(sb->size),
1960 le32_to_cpu(sb->raid_disks),
1961 le32_to_cpu(sb->layout),
1962 le32_to_cpu(sb->chunksize),
1963 (unsigned long long)le64_to_cpu(sb->data_offset),
1964 (unsigned long long)le64_to_cpu(sb->data_size),
1965 (unsigned long long)le64_to_cpu(sb->super_offset),
1966 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1967 le32_to_cpu(sb->dev_number),
1968 uuid,
1969 sb->devflags,
1970 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1971 (unsigned long long)le64_to_cpu(sb->events),
1972 (unsigned long long)le64_to_cpu(sb->resync_offset),
1973 le32_to_cpu(sb->sb_csum),
1974 le32_to_cpu(sb->max_dev)
1975 );
1976 }
1977
1978 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1979 {
1980 char b[BDEVNAME_SIZE];
1981 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1982 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1983 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1984 rdev->desc_nr);
1985 if (rdev->sb_loaded) {
1986 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1987 switch (major_version) {
1988 case 0:
1989 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1990 break;
1991 case 1:
1992 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1993 break;
1994 }
1995 } else
1996 printk(KERN_INFO "md: no rdev superblock!\n");
1997 }
1998
1999 static void md_print_devices(void)
2000 {
2001 struct list_head *tmp;
2002 mdk_rdev_t *rdev;
2003 mddev_t *mddev;
2004 char b[BDEVNAME_SIZE];
2005
2006 printk("\n");
2007 printk("md: **********************************\n");
2008 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2009 printk("md: **********************************\n");
2010 for_each_mddev(mddev, tmp) {
2011
2012 if (mddev->bitmap)
2013 bitmap_print_sb(mddev->bitmap);
2014 else
2015 printk("%s: ", mdname(mddev));
2016 list_for_each_entry(rdev, &mddev->disks, same_set)
2017 printk("<%s>", bdevname(rdev->bdev,b));
2018 printk("\n");
2019
2020 list_for_each_entry(rdev, &mddev->disks, same_set)
2021 print_rdev(rdev, mddev->major_version);
2022 }
2023 printk("md: **********************************\n");
2024 printk("\n");
2025 }
2026
2027
2028 static void sync_sbs(mddev_t * mddev, int nospares)
2029 {
2030 /* Update each superblock (in-memory image), but
2031 * if we are allowed to, skip spares which already
2032 * have the right event counter, or have one earlier
2033 * (which would mean they aren't being marked as dirty
2034 * with the rest of the array)
2035 */
2036 mdk_rdev_t *rdev;
2037
2038 /* First make sure individual recovery_offsets are correct */
2039 list_for_each_entry(rdev, &mddev->disks, same_set) {
2040 if (rdev->raid_disk >= 0 &&
2041 !test_bit(In_sync, &rdev->flags) &&
2042 mddev->curr_resync_completed > rdev->recovery_offset)
2043 rdev->recovery_offset = mddev->curr_resync_completed;
2044
2045 }
2046 list_for_each_entry(rdev, &mddev->disks, same_set) {
2047 if (rdev->sb_events == mddev->events ||
2048 (nospares &&
2049 rdev->raid_disk < 0 &&
2050 (rdev->sb_events&1)==0 &&
2051 rdev->sb_events+1 == mddev->events)) {
2052 /* Don't update this superblock */
2053 rdev->sb_loaded = 2;
2054 } else {
2055 super_types[mddev->major_version].
2056 sync_super(mddev, rdev);
2057 rdev->sb_loaded = 1;
2058 }
2059 }
2060 }
2061
2062 static void md_update_sb(mddev_t * mddev, int force_change)
2063 {
2064 mdk_rdev_t *rdev;
2065 int sync_req;
2066 int nospares = 0;
2067
2068 mddev->utime = get_seconds();
2069 if (mddev->external)
2070 return;
2071 repeat:
2072 spin_lock_irq(&mddev->write_lock);
2073
2074 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2075 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2076 force_change = 1;
2077 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2078 /* just a clean<-> dirty transition, possibly leave spares alone,
2079 * though if events isn't the right even/odd, we will have to do
2080 * spares after all
2081 */
2082 nospares = 1;
2083 if (force_change)
2084 nospares = 0;
2085 if (mddev->degraded)
2086 /* If the array is degraded, then skipping spares is both
2087 * dangerous and fairly pointless.
2088 * Dangerous because a device that was removed from the array
2089 * might have a event_count that still looks up-to-date,
2090 * so it can be re-added without a resync.
2091 * Pointless because if there are any spares to skip,
2092 * then a recovery will happen and soon that array won't
2093 * be degraded any more and the spare can go back to sleep then.
2094 */
2095 nospares = 0;
2096
2097 sync_req = mddev->in_sync;
2098
2099 /* If this is just a dirty<->clean transition, and the array is clean
2100 * and 'events' is odd, we can roll back to the previous clean state */
2101 if (nospares
2102 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2103 && (mddev->events & 1)
2104 && mddev->events != 1)
2105 mddev->events--;
2106 else {
2107 /* otherwise we have to go forward and ... */
2108 mddev->events ++;
2109 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2110 /* .. if the array isn't clean, an 'even' event must also go
2111 * to spares. */
2112 if ((mddev->events&1)==0) {
2113 nospares = 0;
2114 sync_req = 2; /* force a second update to get the
2115 * even/odd in sync */
2116 }
2117 } else {
2118 /* otherwise an 'odd' event must go to spares */
2119 if ((mddev->events&1)) {
2120 nospares = 0;
2121 sync_req = 2; /* force a second update to get the
2122 * even/odd in sync */
2123 }
2124 }
2125 }
2126
2127 if (!mddev->events) {
2128 /*
2129 * oops, this 64-bit counter should never wrap.
2130 * Either we are in around ~1 trillion A.C., assuming
2131 * 1 reboot per second, or we have a bug:
2132 */
2133 MD_BUG();
2134 mddev->events --;
2135 }
2136
2137 /*
2138 * do not write anything to disk if using
2139 * nonpersistent superblocks
2140 */
2141 if (!mddev->persistent) {
2142 if (!mddev->external)
2143 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2144
2145 spin_unlock_irq(&mddev->write_lock);
2146 wake_up(&mddev->sb_wait);
2147 return;
2148 }
2149 sync_sbs(mddev, nospares);
2150 spin_unlock_irq(&mddev->write_lock);
2151
2152 dprintk(KERN_INFO
2153 "md: updating %s RAID superblock on device (in sync %d)\n",
2154 mdname(mddev),mddev->in_sync);
2155
2156 bitmap_update_sb(mddev->bitmap);
2157 list_for_each_entry(rdev, &mddev->disks, same_set) {
2158 char b[BDEVNAME_SIZE];
2159 dprintk(KERN_INFO "md: ");
2160 if (rdev->sb_loaded != 1)
2161 continue; /* no noise on spare devices */
2162 if (test_bit(Faulty, &rdev->flags))
2163 dprintk("(skipping faulty ");
2164
2165 dprintk("%s ", bdevname(rdev->bdev,b));
2166 if (!test_bit(Faulty, &rdev->flags)) {
2167 md_super_write(mddev,rdev,
2168 rdev->sb_start, rdev->sb_size,
2169 rdev->sb_page);
2170 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2171 bdevname(rdev->bdev,b),
2172 (unsigned long long)rdev->sb_start);
2173 rdev->sb_events = mddev->events;
2174
2175 } else
2176 dprintk(")\n");
2177 if (mddev->level == LEVEL_MULTIPATH)
2178 /* only need to write one superblock... */
2179 break;
2180 }
2181 md_super_wait(mddev);
2182 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2183
2184 spin_lock_irq(&mddev->write_lock);
2185 if (mddev->in_sync != sync_req ||
2186 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2187 /* have to write it out again */
2188 spin_unlock_irq(&mddev->write_lock);
2189 goto repeat;
2190 }
2191 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2192 spin_unlock_irq(&mddev->write_lock);
2193 wake_up(&mddev->sb_wait);
2194 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2195 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2196
2197 }
2198
2199 /* words written to sysfs files may, or may not, be \n terminated.
2200 * We want to accept with case. For this we use cmd_match.
2201 */
2202 static int cmd_match(const char *cmd, const char *str)
2203 {
2204 /* See if cmd, written into a sysfs file, matches
2205 * str. They must either be the same, or cmd can
2206 * have a trailing newline
2207 */
2208 while (*cmd && *str && *cmd == *str) {
2209 cmd++;
2210 str++;
2211 }
2212 if (*cmd == '\n')
2213 cmd++;
2214 if (*str || *cmd)
2215 return 0;
2216 return 1;
2217 }
2218
2219 struct rdev_sysfs_entry {
2220 struct attribute attr;
2221 ssize_t (*show)(mdk_rdev_t *, char *);
2222 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2223 };
2224
2225 static ssize_t
2226 state_show(mdk_rdev_t *rdev, char *page)
2227 {
2228 char *sep = "";
2229 size_t len = 0;
2230
2231 if (test_bit(Faulty, &rdev->flags)) {
2232 len+= sprintf(page+len, "%sfaulty",sep);
2233 sep = ",";
2234 }
2235 if (test_bit(In_sync, &rdev->flags)) {
2236 len += sprintf(page+len, "%sin_sync",sep);
2237 sep = ",";
2238 }
2239 if (test_bit(WriteMostly, &rdev->flags)) {
2240 len += sprintf(page+len, "%swrite_mostly",sep);
2241 sep = ",";
2242 }
2243 if (test_bit(Blocked, &rdev->flags)) {
2244 len += sprintf(page+len, "%sblocked", sep);
2245 sep = ",";
2246 }
2247 if (!test_bit(Faulty, &rdev->flags) &&
2248 !test_bit(In_sync, &rdev->flags)) {
2249 len += sprintf(page+len, "%sspare", sep);
2250 sep = ",";
2251 }
2252 return len+sprintf(page+len, "\n");
2253 }
2254
2255 static ssize_t
2256 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2257 {
2258 /* can write
2259 * faulty - simulates and error
2260 * remove - disconnects the device
2261 * writemostly - sets write_mostly
2262 * -writemostly - clears write_mostly
2263 * blocked - sets the Blocked flag
2264 * -blocked - clears the Blocked flag
2265 * insync - sets Insync providing device isn't active
2266 */
2267 int err = -EINVAL;
2268 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2269 md_error(rdev->mddev, rdev);
2270 err = 0;
2271 } else if (cmd_match(buf, "remove")) {
2272 if (rdev->raid_disk >= 0)
2273 err = -EBUSY;
2274 else {
2275 mddev_t *mddev = rdev->mddev;
2276 kick_rdev_from_array(rdev);
2277 if (mddev->pers)
2278 md_update_sb(mddev, 1);
2279 md_new_event(mddev);
2280 err = 0;
2281 }
2282 } else if (cmd_match(buf, "writemostly")) {
2283 set_bit(WriteMostly, &rdev->flags);
2284 err = 0;
2285 } else if (cmd_match(buf, "-writemostly")) {
2286 clear_bit(WriteMostly, &rdev->flags);
2287 err = 0;
2288 } else if (cmd_match(buf, "blocked")) {
2289 set_bit(Blocked, &rdev->flags);
2290 err = 0;
2291 } else if (cmd_match(buf, "-blocked")) {
2292 clear_bit(Blocked, &rdev->flags);
2293 wake_up(&rdev->blocked_wait);
2294 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2295 md_wakeup_thread(rdev->mddev->thread);
2296
2297 err = 0;
2298 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2299 set_bit(In_sync, &rdev->flags);
2300 err = 0;
2301 }
2302 if (!err && rdev->sysfs_state)
2303 sysfs_notify_dirent(rdev->sysfs_state);
2304 return err ? err : len;
2305 }
2306 static struct rdev_sysfs_entry rdev_state =
2307 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2308
2309 static ssize_t
2310 errors_show(mdk_rdev_t *rdev, char *page)
2311 {
2312 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2313 }
2314
2315 static ssize_t
2316 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2317 {
2318 char *e;
2319 unsigned long n = simple_strtoul(buf, &e, 10);
2320 if (*buf && (*e == 0 || *e == '\n')) {
2321 atomic_set(&rdev->corrected_errors, n);
2322 return len;
2323 }
2324 return -EINVAL;
2325 }
2326 static struct rdev_sysfs_entry rdev_errors =
2327 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2328
2329 static ssize_t
2330 slot_show(mdk_rdev_t *rdev, char *page)
2331 {
2332 if (rdev->raid_disk < 0)
2333 return sprintf(page, "none\n");
2334 else
2335 return sprintf(page, "%d\n", rdev->raid_disk);
2336 }
2337
2338 static ssize_t
2339 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2340 {
2341 char *e;
2342 int err;
2343 char nm[20];
2344 int slot = simple_strtoul(buf, &e, 10);
2345 if (strncmp(buf, "none", 4)==0)
2346 slot = -1;
2347 else if (e==buf || (*e && *e!= '\n'))
2348 return -EINVAL;
2349 if (rdev->mddev->pers && slot == -1) {
2350 /* Setting 'slot' on an active array requires also
2351 * updating the 'rd%d' link, and communicating
2352 * with the personality with ->hot_*_disk.
2353 * For now we only support removing
2354 * failed/spare devices. This normally happens automatically,
2355 * but not when the metadata is externally managed.
2356 */
2357 if (rdev->raid_disk == -1)
2358 return -EEXIST;
2359 /* personality does all needed checks */
2360 if (rdev->mddev->pers->hot_add_disk == NULL)
2361 return -EINVAL;
2362 err = rdev->mddev->pers->
2363 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2364 if (err)
2365 return err;
2366 sprintf(nm, "rd%d", rdev->raid_disk);
2367 sysfs_remove_link(&rdev->mddev->kobj, nm);
2368 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2369 md_wakeup_thread(rdev->mddev->thread);
2370 } else if (rdev->mddev->pers) {
2371 mdk_rdev_t *rdev2;
2372 /* Activating a spare .. or possibly reactivating
2373 * if we ever get bitmaps working here.
2374 */
2375
2376 if (rdev->raid_disk != -1)
2377 return -EBUSY;
2378
2379 if (rdev->mddev->pers->hot_add_disk == NULL)
2380 return -EINVAL;
2381
2382 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2383 if (rdev2->raid_disk == slot)
2384 return -EEXIST;
2385
2386 rdev->raid_disk = slot;
2387 if (test_bit(In_sync, &rdev->flags))
2388 rdev->saved_raid_disk = slot;
2389 else
2390 rdev->saved_raid_disk = -1;
2391 err = rdev->mddev->pers->
2392 hot_add_disk(rdev->mddev, rdev);
2393 if (err) {
2394 rdev->raid_disk = -1;
2395 return err;
2396 } else
2397 sysfs_notify_dirent(rdev->sysfs_state);
2398 sprintf(nm, "rd%d", rdev->raid_disk);
2399 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2400 printk(KERN_WARNING
2401 "md: cannot register "
2402 "%s for %s\n",
2403 nm, mdname(rdev->mddev));
2404
2405 /* don't wakeup anyone, leave that to userspace. */
2406 } else {
2407 if (slot >= rdev->mddev->raid_disks)
2408 return -ENOSPC;
2409 rdev->raid_disk = slot;
2410 /* assume it is working */
2411 clear_bit(Faulty, &rdev->flags);
2412 clear_bit(WriteMostly, &rdev->flags);
2413 set_bit(In_sync, &rdev->flags);
2414 sysfs_notify_dirent(rdev->sysfs_state);
2415 }
2416 return len;
2417 }
2418
2419
2420 static struct rdev_sysfs_entry rdev_slot =
2421 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2422
2423 static ssize_t
2424 offset_show(mdk_rdev_t *rdev, char *page)
2425 {
2426 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2427 }
2428
2429 static ssize_t
2430 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2431 {
2432 char *e;
2433 unsigned long long offset = simple_strtoull(buf, &e, 10);
2434 if (e==buf || (*e && *e != '\n'))
2435 return -EINVAL;
2436 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2437 return -EBUSY;
2438 if (rdev->sectors && rdev->mddev->external)
2439 /* Must set offset before size, so overlap checks
2440 * can be sane */
2441 return -EBUSY;
2442 rdev->data_offset = offset;
2443 return len;
2444 }
2445
2446 static struct rdev_sysfs_entry rdev_offset =
2447 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2448
2449 static ssize_t
2450 rdev_size_show(mdk_rdev_t *rdev, char *page)
2451 {
2452 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2453 }
2454
2455 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2456 {
2457 /* check if two start/length pairs overlap */
2458 if (s1+l1 <= s2)
2459 return 0;
2460 if (s2+l2 <= s1)
2461 return 0;
2462 return 1;
2463 }
2464
2465 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2466 {
2467 unsigned long long blocks;
2468 sector_t new;
2469
2470 if (strict_strtoull(buf, 10, &blocks) < 0)
2471 return -EINVAL;
2472
2473 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2474 return -EINVAL; /* sector conversion overflow */
2475
2476 new = blocks * 2;
2477 if (new != blocks * 2)
2478 return -EINVAL; /* unsigned long long to sector_t overflow */
2479
2480 *sectors = new;
2481 return 0;
2482 }
2483
2484 static ssize_t
2485 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2486 {
2487 mddev_t *my_mddev = rdev->mddev;
2488 sector_t oldsectors = rdev->sectors;
2489 sector_t sectors;
2490
2491 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2492 return -EINVAL;
2493 if (my_mddev->pers && rdev->raid_disk >= 0) {
2494 if (my_mddev->persistent) {
2495 sectors = super_types[my_mddev->major_version].
2496 rdev_size_change(rdev, sectors);
2497 if (!sectors)
2498 return -EBUSY;
2499 } else if (!sectors)
2500 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2501 rdev->data_offset;
2502 }
2503 if (sectors < my_mddev->dev_sectors)
2504 return -EINVAL; /* component must fit device */
2505
2506 rdev->sectors = sectors;
2507 if (sectors > oldsectors && my_mddev->external) {
2508 /* need to check that all other rdevs with the same ->bdev
2509 * do not overlap. We need to unlock the mddev to avoid
2510 * a deadlock. We have already changed rdev->sectors, and if
2511 * we have to change it back, we will have the lock again.
2512 */
2513 mddev_t *mddev;
2514 int overlap = 0;
2515 struct list_head *tmp;
2516
2517 mddev_unlock(my_mddev);
2518 for_each_mddev(mddev, tmp) {
2519 mdk_rdev_t *rdev2;
2520
2521 mddev_lock(mddev);
2522 list_for_each_entry(rdev2, &mddev->disks, same_set)
2523 if (test_bit(AllReserved, &rdev2->flags) ||
2524 (rdev->bdev == rdev2->bdev &&
2525 rdev != rdev2 &&
2526 overlaps(rdev->data_offset, rdev->sectors,
2527 rdev2->data_offset,
2528 rdev2->sectors))) {
2529 overlap = 1;
2530 break;
2531 }
2532 mddev_unlock(mddev);
2533 if (overlap) {
2534 mddev_put(mddev);
2535 break;
2536 }
2537 }
2538 mddev_lock(my_mddev);
2539 if (overlap) {
2540 /* Someone else could have slipped in a size
2541 * change here, but doing so is just silly.
2542 * We put oldsectors back because we *know* it is
2543 * safe, and trust userspace not to race with
2544 * itself
2545 */
2546 rdev->sectors = oldsectors;
2547 return -EBUSY;
2548 }
2549 }
2550 return len;
2551 }
2552
2553 static struct rdev_sysfs_entry rdev_size =
2554 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2555
2556
2557 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2558 {
2559 unsigned long long recovery_start = rdev->recovery_offset;
2560
2561 if (test_bit(In_sync, &rdev->flags) ||
2562 recovery_start == MaxSector)
2563 return sprintf(page, "none\n");
2564
2565 return sprintf(page, "%llu\n", recovery_start);
2566 }
2567
2568 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2569 {
2570 unsigned long long recovery_start;
2571
2572 if (cmd_match(buf, "none"))
2573 recovery_start = MaxSector;
2574 else if (strict_strtoull(buf, 10, &recovery_start))
2575 return -EINVAL;
2576
2577 if (rdev->mddev->pers &&
2578 rdev->raid_disk >= 0)
2579 return -EBUSY;
2580
2581 rdev->recovery_offset = recovery_start;
2582 if (recovery_start == MaxSector)
2583 set_bit(In_sync, &rdev->flags);
2584 else
2585 clear_bit(In_sync, &rdev->flags);
2586 return len;
2587 }
2588
2589 static struct rdev_sysfs_entry rdev_recovery_start =
2590 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2591
2592 static struct attribute *rdev_default_attrs[] = {
2593 &rdev_state.attr,
2594 &rdev_errors.attr,
2595 &rdev_slot.attr,
2596 &rdev_offset.attr,
2597 &rdev_size.attr,
2598 &rdev_recovery_start.attr,
2599 NULL,
2600 };
2601 static ssize_t
2602 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2603 {
2604 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2605 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2606 mddev_t *mddev = rdev->mddev;
2607 ssize_t rv;
2608
2609 if (!entry->show)
2610 return -EIO;
2611
2612 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2613 if (!rv) {
2614 if (rdev->mddev == NULL)
2615 rv = -EBUSY;
2616 else
2617 rv = entry->show(rdev, page);
2618 mddev_unlock(mddev);
2619 }
2620 return rv;
2621 }
2622
2623 static ssize_t
2624 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2625 const char *page, size_t length)
2626 {
2627 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2628 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2629 ssize_t rv;
2630 mddev_t *mddev = rdev->mddev;
2631
2632 if (!entry->store)
2633 return -EIO;
2634 if (!capable(CAP_SYS_ADMIN))
2635 return -EACCES;
2636 rv = mddev ? mddev_lock(mddev): -EBUSY;
2637 if (!rv) {
2638 if (rdev->mddev == NULL)
2639 rv = -EBUSY;
2640 else
2641 rv = entry->store(rdev, page, length);
2642 mddev_unlock(mddev);
2643 }
2644 return rv;
2645 }
2646
2647 static void rdev_free(struct kobject *ko)
2648 {
2649 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2650 kfree(rdev);
2651 }
2652 static const struct sysfs_ops rdev_sysfs_ops = {
2653 .show = rdev_attr_show,
2654 .store = rdev_attr_store,
2655 };
2656 static struct kobj_type rdev_ktype = {
2657 .release = rdev_free,
2658 .sysfs_ops = &rdev_sysfs_ops,
2659 .default_attrs = rdev_default_attrs,
2660 };
2661
2662 /*
2663 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2664 *
2665 * mark the device faulty if:
2666 *
2667 * - the device is nonexistent (zero size)
2668 * - the device has no valid superblock
2669 *
2670 * a faulty rdev _never_ has rdev->sb set.
2671 */
2672 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2673 {
2674 char b[BDEVNAME_SIZE];
2675 int err;
2676 mdk_rdev_t *rdev;
2677 sector_t size;
2678
2679 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2680 if (!rdev) {
2681 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2682 return ERR_PTR(-ENOMEM);
2683 }
2684
2685 if ((err = alloc_disk_sb(rdev)))
2686 goto abort_free;
2687
2688 err = lock_rdev(rdev, newdev, super_format == -2);
2689 if (err)
2690 goto abort_free;
2691
2692 kobject_init(&rdev->kobj, &rdev_ktype);
2693
2694 rdev->desc_nr = -1;
2695 rdev->saved_raid_disk = -1;
2696 rdev->raid_disk = -1;
2697 rdev->flags = 0;
2698 rdev->data_offset = 0;
2699 rdev->sb_events = 0;
2700 rdev->last_read_error.tv_sec = 0;
2701 rdev->last_read_error.tv_nsec = 0;
2702 atomic_set(&rdev->nr_pending, 0);
2703 atomic_set(&rdev->read_errors, 0);
2704 atomic_set(&rdev->corrected_errors, 0);
2705
2706 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2707 if (!size) {
2708 printk(KERN_WARNING
2709 "md: %s has zero or unknown size, marking faulty!\n",
2710 bdevname(rdev->bdev,b));
2711 err = -EINVAL;
2712 goto abort_free;
2713 }
2714
2715 if (super_format >= 0) {
2716 err = super_types[super_format].
2717 load_super(rdev, NULL, super_minor);
2718 if (err == -EINVAL) {
2719 printk(KERN_WARNING
2720 "md: %s does not have a valid v%d.%d "
2721 "superblock, not importing!\n",
2722 bdevname(rdev->bdev,b),
2723 super_format, super_minor);
2724 goto abort_free;
2725 }
2726 if (err < 0) {
2727 printk(KERN_WARNING
2728 "md: could not read %s's sb, not importing!\n",
2729 bdevname(rdev->bdev,b));
2730 goto abort_free;
2731 }
2732 }
2733
2734 INIT_LIST_HEAD(&rdev->same_set);
2735 init_waitqueue_head(&rdev->blocked_wait);
2736
2737 return rdev;
2738
2739 abort_free:
2740 if (rdev->sb_page) {
2741 if (rdev->bdev)
2742 unlock_rdev(rdev);
2743 free_disk_sb(rdev);
2744 }
2745 kfree(rdev);
2746 return ERR_PTR(err);
2747 }
2748
2749 /*
2750 * Check a full RAID array for plausibility
2751 */
2752
2753
2754 static void analyze_sbs(mddev_t * mddev)
2755 {
2756 int i;
2757 mdk_rdev_t *rdev, *freshest, *tmp;
2758 char b[BDEVNAME_SIZE];
2759
2760 freshest = NULL;
2761 rdev_for_each(rdev, tmp, mddev)
2762 switch (super_types[mddev->major_version].
2763 load_super(rdev, freshest, mddev->minor_version)) {
2764 case 1:
2765 freshest = rdev;
2766 break;
2767 case 0:
2768 break;
2769 default:
2770 printk( KERN_ERR \
2771 "md: fatal superblock inconsistency in %s"
2772 " -- removing from array\n",
2773 bdevname(rdev->bdev,b));
2774 kick_rdev_from_array(rdev);
2775 }
2776
2777
2778 super_types[mddev->major_version].
2779 validate_super(mddev, freshest);
2780
2781 i = 0;
2782 rdev_for_each(rdev, tmp, mddev) {
2783 if (rdev->desc_nr >= mddev->max_disks ||
2784 i > mddev->max_disks) {
2785 printk(KERN_WARNING
2786 "md: %s: %s: only %d devices permitted\n",
2787 mdname(mddev), bdevname(rdev->bdev, b),
2788 mddev->max_disks);
2789 kick_rdev_from_array(rdev);
2790 continue;
2791 }
2792 if (rdev != freshest)
2793 if (super_types[mddev->major_version].
2794 validate_super(mddev, rdev)) {
2795 printk(KERN_WARNING "md: kicking non-fresh %s"
2796 " from array!\n",
2797 bdevname(rdev->bdev,b));
2798 kick_rdev_from_array(rdev);
2799 continue;
2800 }
2801 if (mddev->level == LEVEL_MULTIPATH) {
2802 rdev->desc_nr = i++;
2803 rdev->raid_disk = rdev->desc_nr;
2804 set_bit(In_sync, &rdev->flags);
2805 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2806 rdev->raid_disk = -1;
2807 clear_bit(In_sync, &rdev->flags);
2808 }
2809 }
2810 }
2811
2812 /* Read a fixed-point number.
2813 * Numbers in sysfs attributes should be in "standard" units where
2814 * possible, so time should be in seconds.
2815 * However we internally use a a much smaller unit such as
2816 * milliseconds or jiffies.
2817 * This function takes a decimal number with a possible fractional
2818 * component, and produces an integer which is the result of
2819 * multiplying that number by 10^'scale'.
2820 * all without any floating-point arithmetic.
2821 */
2822 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2823 {
2824 unsigned long result = 0;
2825 long decimals = -1;
2826 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2827 if (*cp == '.')
2828 decimals = 0;
2829 else if (decimals < scale) {
2830 unsigned int value;
2831 value = *cp - '0';
2832 result = result * 10 + value;
2833 if (decimals >= 0)
2834 decimals++;
2835 }
2836 cp++;
2837 }
2838 if (*cp == '\n')
2839 cp++;
2840 if (*cp)
2841 return -EINVAL;
2842 if (decimals < 0)
2843 decimals = 0;
2844 while (decimals < scale) {
2845 result *= 10;
2846 decimals ++;
2847 }
2848 *res = result;
2849 return 0;
2850 }
2851
2852
2853 static void md_safemode_timeout(unsigned long data);
2854
2855 static ssize_t
2856 safe_delay_show(mddev_t *mddev, char *page)
2857 {
2858 int msec = (mddev->safemode_delay*1000)/HZ;
2859 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2860 }
2861 static ssize_t
2862 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2863 {
2864 unsigned long msec;
2865
2866 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2867 return -EINVAL;
2868 if (msec == 0)
2869 mddev->safemode_delay = 0;
2870 else {
2871 unsigned long old_delay = mddev->safemode_delay;
2872 mddev->safemode_delay = (msec*HZ)/1000;
2873 if (mddev->safemode_delay == 0)
2874 mddev->safemode_delay = 1;
2875 if (mddev->safemode_delay < old_delay)
2876 md_safemode_timeout((unsigned long)mddev);
2877 }
2878 return len;
2879 }
2880 static struct md_sysfs_entry md_safe_delay =
2881 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2882
2883 static ssize_t
2884 level_show(mddev_t *mddev, char *page)
2885 {
2886 struct mdk_personality *p = mddev->pers;
2887 if (p)
2888 return sprintf(page, "%s\n", p->name);
2889 else if (mddev->clevel[0])
2890 return sprintf(page, "%s\n", mddev->clevel);
2891 else if (mddev->level != LEVEL_NONE)
2892 return sprintf(page, "%d\n", mddev->level);
2893 else
2894 return 0;
2895 }
2896
2897 static ssize_t
2898 level_store(mddev_t *mddev, const char *buf, size_t len)
2899 {
2900 char level[16];
2901 ssize_t rv = len;
2902 struct mdk_personality *pers;
2903 void *priv;
2904 mdk_rdev_t *rdev;
2905
2906 if (mddev->pers == NULL) {
2907 if (len == 0)
2908 return 0;
2909 if (len >= sizeof(mddev->clevel))
2910 return -ENOSPC;
2911 strncpy(mddev->clevel, buf, len);
2912 if (mddev->clevel[len-1] == '\n')
2913 len--;
2914 mddev->clevel[len] = 0;
2915 mddev->level = LEVEL_NONE;
2916 return rv;
2917 }
2918
2919 /* request to change the personality. Need to ensure:
2920 * - array is not engaged in resync/recovery/reshape
2921 * - old personality can be suspended
2922 * - new personality will access other array.
2923 */
2924
2925 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2926 return -EBUSY;
2927
2928 if (!mddev->pers->quiesce) {
2929 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2930 mdname(mddev), mddev->pers->name);
2931 return -EINVAL;
2932 }
2933
2934 /* Now find the new personality */
2935 if (len == 0 || len >= sizeof(level))
2936 return -EINVAL;
2937 strncpy(level, buf, len);
2938 if (level[len-1] == '\n')
2939 len--;
2940 level[len] = 0;
2941
2942 request_module("md-%s", level);
2943 spin_lock(&pers_lock);
2944 pers = find_pers(LEVEL_NONE, level);
2945 if (!pers || !try_module_get(pers->owner)) {
2946 spin_unlock(&pers_lock);
2947 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2948 return -EINVAL;
2949 }
2950 spin_unlock(&pers_lock);
2951
2952 if (pers == mddev->pers) {
2953 /* Nothing to do! */
2954 module_put(pers->owner);
2955 return rv;
2956 }
2957 if (!pers->takeover) {
2958 module_put(pers->owner);
2959 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2960 mdname(mddev), level);
2961 return -EINVAL;
2962 }
2963
2964 /* ->takeover must set new_* and/or delta_disks
2965 * if it succeeds, and may set them when it fails.
2966 */
2967 priv = pers->takeover(mddev);
2968 if (IS_ERR(priv)) {
2969 mddev->new_level = mddev->level;
2970 mddev->new_layout = mddev->layout;
2971 mddev->new_chunk_sectors = mddev->chunk_sectors;
2972 mddev->raid_disks -= mddev->delta_disks;
2973 mddev->delta_disks = 0;
2974 module_put(pers->owner);
2975 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2976 mdname(mddev), level);
2977 return PTR_ERR(priv);
2978 }
2979
2980 /* Looks like we have a winner */
2981 mddev_suspend(mddev);
2982 mddev->pers->stop(mddev);
2983 module_put(mddev->pers->owner);
2984 /* Invalidate devices that are now superfluous */
2985 list_for_each_entry(rdev, &mddev->disks, same_set)
2986 if (rdev->raid_disk >= mddev->raid_disks) {
2987 rdev->raid_disk = -1;
2988 clear_bit(In_sync, &rdev->flags);
2989 }
2990 mddev->pers = pers;
2991 mddev->private = priv;
2992 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2993 mddev->level = mddev->new_level;
2994 mddev->layout = mddev->new_layout;
2995 mddev->chunk_sectors = mddev->new_chunk_sectors;
2996 mddev->delta_disks = 0;
2997 pers->run(mddev);
2998 mddev_resume(mddev);
2999 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3000 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3001 md_wakeup_thread(mddev->thread);
3002 return rv;
3003 }
3004
3005 static struct md_sysfs_entry md_level =
3006 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3007
3008
3009 static ssize_t
3010 layout_show(mddev_t *mddev, char *page)
3011 {
3012 /* just a number, not meaningful for all levels */
3013 if (mddev->reshape_position != MaxSector &&
3014 mddev->layout != mddev->new_layout)
3015 return sprintf(page, "%d (%d)\n",
3016 mddev->new_layout, mddev->layout);
3017 return sprintf(page, "%d\n", mddev->layout);
3018 }
3019
3020 static ssize_t
3021 layout_store(mddev_t *mddev, const char *buf, size_t len)
3022 {
3023 char *e;
3024 unsigned long n = simple_strtoul(buf, &e, 10);
3025
3026 if (!*buf || (*e && *e != '\n'))
3027 return -EINVAL;
3028
3029 if (mddev->pers) {
3030 int err;
3031 if (mddev->pers->check_reshape == NULL)
3032 return -EBUSY;
3033 mddev->new_layout = n;
3034 err = mddev->pers->check_reshape(mddev);
3035 if (err) {
3036 mddev->new_layout = mddev->layout;
3037 return err;
3038 }
3039 } else {
3040 mddev->new_layout = n;
3041 if (mddev->reshape_position == MaxSector)
3042 mddev->layout = n;
3043 }
3044 return len;
3045 }
3046 static struct md_sysfs_entry md_layout =
3047 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3048
3049
3050 static ssize_t
3051 raid_disks_show(mddev_t *mddev, char *page)
3052 {
3053 if (mddev->raid_disks == 0)
3054 return 0;
3055 if (mddev->reshape_position != MaxSector &&
3056 mddev->delta_disks != 0)
3057 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3058 mddev->raid_disks - mddev->delta_disks);
3059 return sprintf(page, "%d\n", mddev->raid_disks);
3060 }
3061
3062 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3063
3064 static ssize_t
3065 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3066 {
3067 char *e;
3068 int rv = 0;
3069 unsigned long n = simple_strtoul(buf, &e, 10);
3070
3071 if (!*buf || (*e && *e != '\n'))
3072 return -EINVAL;
3073
3074 if (mddev->pers)
3075 rv = update_raid_disks(mddev, n);
3076 else if (mddev->reshape_position != MaxSector) {
3077 int olddisks = mddev->raid_disks - mddev->delta_disks;
3078 mddev->delta_disks = n - olddisks;
3079 mddev->raid_disks = n;
3080 } else
3081 mddev->raid_disks = n;
3082 return rv ? rv : len;
3083 }
3084 static struct md_sysfs_entry md_raid_disks =
3085 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3086
3087 static ssize_t
3088 chunk_size_show(mddev_t *mddev, char *page)
3089 {
3090 if (mddev->reshape_position != MaxSector &&
3091 mddev->chunk_sectors != mddev->new_chunk_sectors)
3092 return sprintf(page, "%d (%d)\n",
3093 mddev->new_chunk_sectors << 9,
3094 mddev->chunk_sectors << 9);
3095 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3096 }
3097
3098 static ssize_t
3099 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3100 {
3101 char *e;
3102 unsigned long n = simple_strtoul(buf, &e, 10);
3103
3104 if (!*buf || (*e && *e != '\n'))
3105 return -EINVAL;
3106
3107 if (mddev->pers) {
3108 int err;
3109 if (mddev->pers->check_reshape == NULL)
3110 return -EBUSY;
3111 mddev->new_chunk_sectors = n >> 9;
3112 err = mddev->pers->check_reshape(mddev);
3113 if (err) {
3114 mddev->new_chunk_sectors = mddev->chunk_sectors;
3115 return err;
3116 }
3117 } else {
3118 mddev->new_chunk_sectors = n >> 9;
3119 if (mddev->reshape_position == MaxSector)
3120 mddev->chunk_sectors = n >> 9;
3121 }
3122 return len;
3123 }
3124 static struct md_sysfs_entry md_chunk_size =
3125 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3126
3127 static ssize_t
3128 resync_start_show(mddev_t *mddev, char *page)
3129 {
3130 if (mddev->recovery_cp == MaxSector)
3131 return sprintf(page, "none\n");
3132 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3133 }
3134
3135 static ssize_t
3136 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3137 {
3138 char *e;
3139 unsigned long long n = simple_strtoull(buf, &e, 10);
3140
3141 if (mddev->pers)
3142 return -EBUSY;
3143 if (cmd_match(buf, "none"))
3144 n = MaxSector;
3145 else if (!*buf || (*e && *e != '\n'))
3146 return -EINVAL;
3147
3148 mddev->recovery_cp = n;
3149 return len;
3150 }
3151 static struct md_sysfs_entry md_resync_start =
3152 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3153
3154 /*
3155 * The array state can be:
3156 *
3157 * clear
3158 * No devices, no size, no level
3159 * Equivalent to STOP_ARRAY ioctl
3160 * inactive
3161 * May have some settings, but array is not active
3162 * all IO results in error
3163 * When written, doesn't tear down array, but just stops it
3164 * suspended (not supported yet)
3165 * All IO requests will block. The array can be reconfigured.
3166 * Writing this, if accepted, will block until array is quiescent
3167 * readonly
3168 * no resync can happen. no superblocks get written.
3169 * write requests fail
3170 * read-auto
3171 * like readonly, but behaves like 'clean' on a write request.
3172 *
3173 * clean - no pending writes, but otherwise active.
3174 * When written to inactive array, starts without resync
3175 * If a write request arrives then
3176 * if metadata is known, mark 'dirty' and switch to 'active'.
3177 * if not known, block and switch to write-pending
3178 * If written to an active array that has pending writes, then fails.
3179 * active
3180 * fully active: IO and resync can be happening.
3181 * When written to inactive array, starts with resync
3182 *
3183 * write-pending
3184 * clean, but writes are blocked waiting for 'active' to be written.
3185 *
3186 * active-idle
3187 * like active, but no writes have been seen for a while (100msec).
3188 *
3189 */
3190 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3191 write_pending, active_idle, bad_word};
3192 static char *array_states[] = {
3193 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3194 "write-pending", "active-idle", NULL };
3195
3196 static int match_word(const char *word, char **list)
3197 {
3198 int n;
3199 for (n=0; list[n]; n++)
3200 if (cmd_match(word, list[n]))
3201 break;
3202 return n;
3203 }
3204
3205 static ssize_t
3206 array_state_show(mddev_t *mddev, char *page)
3207 {
3208 enum array_state st = inactive;
3209
3210 if (mddev->pers)
3211 switch(mddev->ro) {
3212 case 1:
3213 st = readonly;
3214 break;
3215 case 2:
3216 st = read_auto;
3217 break;
3218 case 0:
3219 if (mddev->in_sync)
3220 st = clean;
3221 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3222 st = write_pending;
3223 else if (mddev->safemode)
3224 st = active_idle;
3225 else
3226 st = active;
3227 }
3228 else {
3229 if (list_empty(&mddev->disks) &&
3230 mddev->raid_disks == 0 &&
3231 mddev->dev_sectors == 0)
3232 st = clear;
3233 else
3234 st = inactive;
3235 }
3236 return sprintf(page, "%s\n", array_states[st]);
3237 }
3238
3239 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3240 static int do_md_run(mddev_t * mddev);
3241 static int restart_array(mddev_t *mddev);
3242
3243 static ssize_t
3244 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3245 {
3246 int err = -EINVAL;
3247 enum array_state st = match_word(buf, array_states);
3248 switch(st) {
3249 case bad_word:
3250 break;
3251 case clear:
3252 /* stopping an active array */
3253 if (atomic_read(&mddev->openers) > 0)
3254 return -EBUSY;
3255 err = do_md_stop(mddev, 0, 0);
3256 break;
3257 case inactive:
3258 /* stopping an active array */
3259 if (mddev->pers) {
3260 if (atomic_read(&mddev->openers) > 0)
3261 return -EBUSY;
3262 err = do_md_stop(mddev, 2, 0);
3263 } else
3264 err = 0; /* already inactive */
3265 break;
3266 case suspended:
3267 break; /* not supported yet */
3268 case readonly:
3269 if (mddev->pers)
3270 err = do_md_stop(mddev, 1, 0);
3271 else {
3272 mddev->ro = 1;
3273 set_disk_ro(mddev->gendisk, 1);
3274 err = do_md_run(mddev);
3275 }
3276 break;
3277 case read_auto:
3278 if (mddev->pers) {
3279 if (mddev->ro == 0)
3280 err = do_md_stop(mddev, 1, 0);
3281 else if (mddev->ro == 1)
3282 err = restart_array(mddev);
3283 if (err == 0) {
3284 mddev->ro = 2;
3285 set_disk_ro(mddev->gendisk, 0);
3286 }
3287 } else {
3288 mddev->ro = 2;
3289 err = do_md_run(mddev);
3290 }
3291 break;
3292 case clean:
3293 if (mddev->pers) {
3294 restart_array(mddev);
3295 spin_lock_irq(&mddev->write_lock);
3296 if (atomic_read(&mddev->writes_pending) == 0) {
3297 if (mddev->in_sync == 0) {
3298 mddev->in_sync = 1;
3299 if (mddev->safemode == 1)
3300 mddev->safemode = 0;
3301 if (mddev->persistent)
3302 set_bit(MD_CHANGE_CLEAN,
3303 &mddev->flags);
3304 }
3305 err = 0;
3306 } else
3307 err = -EBUSY;
3308 spin_unlock_irq(&mddev->write_lock);
3309 } else
3310 err = -EINVAL;
3311 break;
3312 case active:
3313 if (mddev->pers) {
3314 restart_array(mddev);
3315 if (mddev->external)
3316 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3317 wake_up(&mddev->sb_wait);
3318 err = 0;
3319 } else {
3320 mddev->ro = 0;
3321 set_disk_ro(mddev->gendisk, 0);
3322 err = do_md_run(mddev);
3323 }
3324 break;
3325 case write_pending:
3326 case active_idle:
3327 /* these cannot be set */
3328 break;
3329 }
3330 if (err)
3331 return err;
3332 else {
3333 sysfs_notify_dirent(mddev->sysfs_state);
3334 return len;
3335 }
3336 }
3337 static struct md_sysfs_entry md_array_state =
3338 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3339
3340 static ssize_t
3341 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3342 return sprintf(page, "%d\n",
3343 atomic_read(&mddev->max_corr_read_errors));
3344 }
3345
3346 static ssize_t
3347 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3348 {
3349 char *e;
3350 unsigned long n = simple_strtoul(buf, &e, 10);
3351
3352 if (*buf && (*e == 0 || *e == '\n')) {
3353 atomic_set(&mddev->max_corr_read_errors, n);
3354 return len;
3355 }
3356 return -EINVAL;
3357 }
3358
3359 static struct md_sysfs_entry max_corr_read_errors =
3360 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3361 max_corrected_read_errors_store);
3362
3363 static ssize_t
3364 null_show(mddev_t *mddev, char *page)
3365 {
3366 return -EINVAL;
3367 }
3368
3369 static ssize_t
3370 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3371 {
3372 /* buf must be %d:%d\n? giving major and minor numbers */
3373 /* The new device is added to the array.
3374 * If the array has a persistent superblock, we read the
3375 * superblock to initialise info and check validity.
3376 * Otherwise, only checking done is that in bind_rdev_to_array,
3377 * which mainly checks size.
3378 */
3379 char *e;
3380 int major = simple_strtoul(buf, &e, 10);
3381 int minor;
3382 dev_t dev;
3383 mdk_rdev_t *rdev;
3384 int err;
3385
3386 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3387 return -EINVAL;
3388 minor = simple_strtoul(e+1, &e, 10);
3389 if (*e && *e != '\n')
3390 return -EINVAL;
3391 dev = MKDEV(major, minor);
3392 if (major != MAJOR(dev) ||
3393 minor != MINOR(dev))
3394 return -EOVERFLOW;
3395
3396
3397 if (mddev->persistent) {
3398 rdev = md_import_device(dev, mddev->major_version,
3399 mddev->minor_version);
3400 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3401 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3402 mdk_rdev_t, same_set);
3403 err = super_types[mddev->major_version]
3404 .load_super(rdev, rdev0, mddev->minor_version);
3405 if (err < 0)
3406 goto out;
3407 }
3408 } else if (mddev->external)
3409 rdev = md_import_device(dev, -2, -1);
3410 else
3411 rdev = md_import_device(dev, -1, -1);
3412
3413 if (IS_ERR(rdev))
3414 return PTR_ERR(rdev);
3415 err = bind_rdev_to_array(rdev, mddev);
3416 out:
3417 if (err)
3418 export_rdev(rdev);
3419 return err ? err : len;
3420 }
3421
3422 static struct md_sysfs_entry md_new_device =
3423 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3424
3425 static ssize_t
3426 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3427 {
3428 char *end;
3429 unsigned long chunk, end_chunk;
3430
3431 if (!mddev->bitmap)
3432 goto out;
3433 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3434 while (*buf) {
3435 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3436 if (buf == end) break;
3437 if (*end == '-') { /* range */
3438 buf = end + 1;
3439 end_chunk = simple_strtoul(buf, &end, 0);
3440 if (buf == end) break;
3441 }
3442 if (*end && !isspace(*end)) break;
3443 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3444 buf = skip_spaces(end);
3445 }
3446 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3447 out:
3448 return len;
3449 }
3450
3451 static struct md_sysfs_entry md_bitmap =
3452 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3453
3454 static ssize_t
3455 size_show(mddev_t *mddev, char *page)
3456 {
3457 return sprintf(page, "%llu\n",
3458 (unsigned long long)mddev->dev_sectors / 2);
3459 }
3460
3461 static int update_size(mddev_t *mddev, sector_t num_sectors);
3462
3463 static ssize_t
3464 size_store(mddev_t *mddev, const char *buf, size_t len)
3465 {
3466 /* If array is inactive, we can reduce the component size, but
3467 * not increase it (except from 0).
3468 * If array is active, we can try an on-line resize
3469 */
3470 sector_t sectors;
3471 int err = strict_blocks_to_sectors(buf, &sectors);
3472
3473 if (err < 0)
3474 return err;
3475 if (mddev->pers) {
3476 err = update_size(mddev, sectors);
3477 md_update_sb(mddev, 1);
3478 } else {
3479 if (mddev->dev_sectors == 0 ||
3480 mddev->dev_sectors > sectors)
3481 mddev->dev_sectors = sectors;
3482 else
3483 err = -ENOSPC;
3484 }
3485 return err ? err : len;
3486 }
3487
3488 static struct md_sysfs_entry md_size =
3489 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3490
3491
3492 /* Metdata version.
3493 * This is one of
3494 * 'none' for arrays with no metadata (good luck...)
3495 * 'external' for arrays with externally managed metadata,
3496 * or N.M for internally known formats
3497 */
3498 static ssize_t
3499 metadata_show(mddev_t *mddev, char *page)
3500 {
3501 if (mddev->persistent)
3502 return sprintf(page, "%d.%d\n",
3503 mddev->major_version, mddev->minor_version);
3504 else if (mddev->external)
3505 return sprintf(page, "external:%s\n", mddev->metadata_type);
3506 else
3507 return sprintf(page, "none\n");
3508 }
3509
3510 static ssize_t
3511 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3512 {
3513 int major, minor;
3514 char *e;
3515 /* Changing the details of 'external' metadata is
3516 * always permitted. Otherwise there must be
3517 * no devices attached to the array.
3518 */
3519 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3520 ;
3521 else if (!list_empty(&mddev->disks))
3522 return -EBUSY;
3523
3524 if (cmd_match(buf, "none")) {
3525 mddev->persistent = 0;
3526 mddev->external = 0;
3527 mddev->major_version = 0;
3528 mddev->minor_version = 90;
3529 return len;
3530 }
3531 if (strncmp(buf, "external:", 9) == 0) {
3532 size_t namelen = len-9;
3533 if (namelen >= sizeof(mddev->metadata_type))
3534 namelen = sizeof(mddev->metadata_type)-1;
3535 strncpy(mddev->metadata_type, buf+9, namelen);
3536 mddev->metadata_type[namelen] = 0;
3537 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3538 mddev->metadata_type[--namelen] = 0;
3539 mddev->persistent = 0;
3540 mddev->external = 1;
3541 mddev->major_version = 0;
3542 mddev->minor_version = 90;
3543 return len;
3544 }
3545 major = simple_strtoul(buf, &e, 10);
3546 if (e==buf || *e != '.')
3547 return -EINVAL;
3548 buf = e+1;
3549 minor = simple_strtoul(buf, &e, 10);
3550 if (e==buf || (*e && *e != '\n') )
3551 return -EINVAL;
3552 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3553 return -ENOENT;
3554 mddev->major_version = major;
3555 mddev->minor_version = minor;
3556 mddev->persistent = 1;
3557 mddev->external = 0;
3558 return len;
3559 }
3560
3561 static struct md_sysfs_entry md_metadata =
3562 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3563
3564 static ssize_t
3565 action_show(mddev_t *mddev, char *page)
3566 {
3567 char *type = "idle";
3568 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3569 type = "frozen";
3570 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3571 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3572 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3573 type = "reshape";
3574 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3575 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3576 type = "resync";
3577 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3578 type = "check";
3579 else
3580 type = "repair";
3581 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3582 type = "recover";
3583 }
3584 return sprintf(page, "%s\n", type);
3585 }
3586
3587 static ssize_t
3588 action_store(mddev_t *mddev, const char *page, size_t len)
3589 {
3590 if (!mddev->pers || !mddev->pers->sync_request)
3591 return -EINVAL;
3592
3593 if (cmd_match(page, "frozen"))
3594 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3595 else
3596 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3597
3598 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3599 if (mddev->sync_thread) {
3600 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3601 md_unregister_thread(mddev->sync_thread);
3602 mddev->sync_thread = NULL;
3603 mddev->recovery = 0;
3604 }
3605 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3606 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3607 return -EBUSY;
3608 else if (cmd_match(page, "resync"))
3609 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3610 else if (cmd_match(page, "recover")) {
3611 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3612 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3613 } else if (cmd_match(page, "reshape")) {
3614 int err;
3615 if (mddev->pers->start_reshape == NULL)
3616 return -EINVAL;
3617 err = mddev->pers->start_reshape(mddev);
3618 if (err)
3619 return err;
3620 sysfs_notify(&mddev->kobj, NULL, "degraded");
3621 } else {
3622 if (cmd_match(page, "check"))
3623 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3624 else if (!cmd_match(page, "repair"))
3625 return -EINVAL;
3626 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3627 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3628 }
3629 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3630 md_wakeup_thread(mddev->thread);
3631 sysfs_notify_dirent(mddev->sysfs_action);
3632 return len;
3633 }
3634
3635 static ssize_t
3636 mismatch_cnt_show(mddev_t *mddev, char *page)
3637 {
3638 return sprintf(page, "%llu\n",
3639 (unsigned long long) mddev->resync_mismatches);
3640 }
3641
3642 static struct md_sysfs_entry md_scan_mode =
3643 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3644
3645
3646 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3647
3648 static ssize_t
3649 sync_min_show(mddev_t *mddev, char *page)
3650 {
3651 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3652 mddev->sync_speed_min ? "local": "system");
3653 }
3654
3655 static ssize_t
3656 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3657 {
3658 int min;
3659 char *e;
3660 if (strncmp(buf, "system", 6)==0) {
3661 mddev->sync_speed_min = 0;
3662 return len;
3663 }
3664 min = simple_strtoul(buf, &e, 10);
3665 if (buf == e || (*e && *e != '\n') || min <= 0)
3666 return -EINVAL;
3667 mddev->sync_speed_min = min;
3668 return len;
3669 }
3670
3671 static struct md_sysfs_entry md_sync_min =
3672 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3673
3674 static ssize_t
3675 sync_max_show(mddev_t *mddev, char *page)
3676 {
3677 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3678 mddev->sync_speed_max ? "local": "system");
3679 }
3680
3681 static ssize_t
3682 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3683 {
3684 int max;
3685 char *e;
3686 if (strncmp(buf, "system", 6)==0) {
3687 mddev->sync_speed_max = 0;
3688 return len;
3689 }
3690 max = simple_strtoul(buf, &e, 10);
3691 if (buf == e || (*e && *e != '\n') || max <= 0)
3692 return -EINVAL;
3693 mddev->sync_speed_max = max;
3694 return len;
3695 }
3696
3697 static struct md_sysfs_entry md_sync_max =
3698 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3699
3700 static ssize_t
3701 degraded_show(mddev_t *mddev, char *page)
3702 {
3703 return sprintf(page, "%d\n", mddev->degraded);
3704 }
3705 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3706
3707 static ssize_t
3708 sync_force_parallel_show(mddev_t *mddev, char *page)
3709 {
3710 return sprintf(page, "%d\n", mddev->parallel_resync);
3711 }
3712
3713 static ssize_t
3714 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3715 {
3716 long n;
3717
3718 if (strict_strtol(buf, 10, &n))
3719 return -EINVAL;
3720
3721 if (n != 0 && n != 1)
3722 return -EINVAL;
3723
3724 mddev->parallel_resync = n;
3725
3726 if (mddev->sync_thread)
3727 wake_up(&resync_wait);
3728
3729 return len;
3730 }
3731
3732 /* force parallel resync, even with shared block devices */
3733 static struct md_sysfs_entry md_sync_force_parallel =
3734 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3735 sync_force_parallel_show, sync_force_parallel_store);
3736
3737 static ssize_t
3738 sync_speed_show(mddev_t *mddev, char *page)
3739 {
3740 unsigned long resync, dt, db;
3741 if (mddev->curr_resync == 0)
3742 return sprintf(page, "none\n");
3743 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3744 dt = (jiffies - mddev->resync_mark) / HZ;
3745 if (!dt) dt++;
3746 db = resync - mddev->resync_mark_cnt;
3747 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3748 }
3749
3750 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3751
3752 static ssize_t
3753 sync_completed_show(mddev_t *mddev, char *page)
3754 {
3755 unsigned long max_sectors, resync;
3756
3757 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3758 return sprintf(page, "none\n");
3759
3760 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3761 max_sectors = mddev->resync_max_sectors;
3762 else
3763 max_sectors = mddev->dev_sectors;
3764
3765 resync = mddev->curr_resync_completed;
3766 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3767 }
3768
3769 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3770
3771 static ssize_t
3772 min_sync_show(mddev_t *mddev, char *page)
3773 {
3774 return sprintf(page, "%llu\n",
3775 (unsigned long long)mddev->resync_min);
3776 }
3777 static ssize_t
3778 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3779 {
3780 unsigned long long min;
3781 if (strict_strtoull(buf, 10, &min))
3782 return -EINVAL;
3783 if (min > mddev->resync_max)
3784 return -EINVAL;
3785 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3786 return -EBUSY;
3787
3788 /* Must be a multiple of chunk_size */
3789 if (mddev->chunk_sectors) {
3790 sector_t temp = min;
3791 if (sector_div(temp, mddev->chunk_sectors))
3792 return -EINVAL;
3793 }
3794 mddev->resync_min = min;
3795
3796 return len;
3797 }
3798
3799 static struct md_sysfs_entry md_min_sync =
3800 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3801
3802 static ssize_t
3803 max_sync_show(mddev_t *mddev, char *page)
3804 {
3805 if (mddev->resync_max == MaxSector)
3806 return sprintf(page, "max\n");
3807 else
3808 return sprintf(page, "%llu\n",
3809 (unsigned long long)mddev->resync_max);
3810 }
3811 static ssize_t
3812 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3813 {
3814 if (strncmp(buf, "max", 3) == 0)
3815 mddev->resync_max = MaxSector;
3816 else {
3817 unsigned long long max;
3818 if (strict_strtoull(buf, 10, &max))
3819 return -EINVAL;
3820 if (max < mddev->resync_min)
3821 return -EINVAL;
3822 if (max < mddev->resync_max &&
3823 mddev->ro == 0 &&
3824 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3825 return -EBUSY;
3826
3827 /* Must be a multiple of chunk_size */
3828 if (mddev->chunk_sectors) {
3829 sector_t temp = max;
3830 if (sector_div(temp, mddev->chunk_sectors))
3831 return -EINVAL;
3832 }
3833 mddev->resync_max = max;
3834 }
3835 wake_up(&mddev->recovery_wait);
3836 return len;
3837 }
3838
3839 static struct md_sysfs_entry md_max_sync =
3840 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3841
3842 static ssize_t
3843 suspend_lo_show(mddev_t *mddev, char *page)
3844 {
3845 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3846 }
3847
3848 static ssize_t
3849 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3850 {
3851 char *e;
3852 unsigned long long new = simple_strtoull(buf, &e, 10);
3853
3854 if (mddev->pers == NULL ||
3855 mddev->pers->quiesce == NULL)
3856 return -EINVAL;
3857 if (buf == e || (*e && *e != '\n'))
3858 return -EINVAL;
3859 if (new >= mddev->suspend_hi ||
3860 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3861 mddev->suspend_lo = new;
3862 mddev->pers->quiesce(mddev, 2);
3863 return len;
3864 } else
3865 return -EINVAL;
3866 }
3867 static struct md_sysfs_entry md_suspend_lo =
3868 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3869
3870
3871 static ssize_t
3872 suspend_hi_show(mddev_t *mddev, char *page)
3873 {
3874 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3875 }
3876
3877 static ssize_t
3878 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3879 {
3880 char *e;
3881 unsigned long long new = simple_strtoull(buf, &e, 10);
3882
3883 if (mddev->pers == NULL ||
3884 mddev->pers->quiesce == NULL)
3885 return -EINVAL;
3886 if (buf == e || (*e && *e != '\n'))
3887 return -EINVAL;
3888 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3889 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3890 mddev->suspend_hi = new;
3891 mddev->pers->quiesce(mddev, 1);
3892 mddev->pers->quiesce(mddev, 0);
3893 return len;
3894 } else
3895 return -EINVAL;
3896 }
3897 static struct md_sysfs_entry md_suspend_hi =
3898 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3899
3900 static ssize_t
3901 reshape_position_show(mddev_t *mddev, char *page)
3902 {
3903 if (mddev->reshape_position != MaxSector)
3904 return sprintf(page, "%llu\n",
3905 (unsigned long long)mddev->reshape_position);
3906 strcpy(page, "none\n");
3907 return 5;
3908 }
3909
3910 static ssize_t
3911 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3912 {
3913 char *e;
3914 unsigned long long new = simple_strtoull(buf, &e, 10);
3915 if (mddev->pers)
3916 return -EBUSY;
3917 if (buf == e || (*e && *e != '\n'))
3918 return -EINVAL;
3919 mddev->reshape_position = new;
3920 mddev->delta_disks = 0;
3921 mddev->new_level = mddev->level;
3922 mddev->new_layout = mddev->layout;
3923 mddev->new_chunk_sectors = mddev->chunk_sectors;
3924 return len;
3925 }
3926
3927 static struct md_sysfs_entry md_reshape_position =
3928 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3929 reshape_position_store);
3930
3931 static ssize_t
3932 array_size_show(mddev_t *mddev, char *page)
3933 {
3934 if (mddev->external_size)
3935 return sprintf(page, "%llu\n",
3936 (unsigned long long)mddev->array_sectors/2);
3937 else
3938 return sprintf(page, "default\n");
3939 }
3940
3941 static ssize_t
3942 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3943 {
3944 sector_t sectors;
3945
3946 if (strncmp(buf, "default", 7) == 0) {
3947 if (mddev->pers)
3948 sectors = mddev->pers->size(mddev, 0, 0);
3949 else
3950 sectors = mddev->array_sectors;
3951
3952 mddev->external_size = 0;
3953 } else {
3954 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3955 return -EINVAL;
3956 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3957 return -E2BIG;
3958
3959 mddev->external_size = 1;
3960 }
3961
3962 mddev->array_sectors = sectors;
3963 set_capacity(mddev->gendisk, mddev->array_sectors);
3964 if (mddev->pers)
3965 revalidate_disk(mddev->gendisk);
3966
3967 return len;
3968 }
3969
3970 static struct md_sysfs_entry md_array_size =
3971 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3972 array_size_store);
3973
3974 static struct attribute *md_default_attrs[] = {
3975 &md_level.attr,
3976 &md_layout.attr,
3977 &md_raid_disks.attr,
3978 &md_chunk_size.attr,
3979 &md_size.attr,
3980 &md_resync_start.attr,
3981 &md_metadata.attr,
3982 &md_new_device.attr,
3983 &md_safe_delay.attr,
3984 &md_array_state.attr,
3985 &md_reshape_position.attr,
3986 &md_array_size.attr,
3987 &max_corr_read_errors.attr,
3988 NULL,
3989 };
3990
3991 static struct attribute *md_redundancy_attrs[] = {
3992 &md_scan_mode.attr,
3993 &md_mismatches.attr,
3994 &md_sync_min.attr,
3995 &md_sync_max.attr,
3996 &md_sync_speed.attr,
3997 &md_sync_force_parallel.attr,
3998 &md_sync_completed.attr,
3999 &md_min_sync.attr,
4000 &md_max_sync.attr,
4001 &md_suspend_lo.attr,
4002 &md_suspend_hi.attr,
4003 &md_bitmap.attr,
4004 &md_degraded.attr,
4005 NULL,
4006 };
4007 static struct attribute_group md_redundancy_group = {
4008 .name = NULL,
4009 .attrs = md_redundancy_attrs,
4010 };
4011
4012
4013 static ssize_t
4014 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4015 {
4016 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4017 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4018 ssize_t rv;
4019
4020 if (!entry->show)
4021 return -EIO;
4022 rv = mddev_lock(mddev);
4023 if (!rv) {
4024 rv = entry->show(mddev, page);
4025 mddev_unlock(mddev);
4026 }
4027 return rv;
4028 }
4029
4030 static ssize_t
4031 md_attr_store(struct kobject *kobj, struct attribute *attr,
4032 const char *page, size_t length)
4033 {
4034 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4035 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4036 ssize_t rv;
4037
4038 if (!entry->store)
4039 return -EIO;
4040 if (!capable(CAP_SYS_ADMIN))
4041 return -EACCES;
4042 rv = mddev_lock(mddev);
4043 if (mddev->hold_active == UNTIL_IOCTL)
4044 mddev->hold_active = 0;
4045 if (!rv) {
4046 rv = entry->store(mddev, page, length);
4047 mddev_unlock(mddev);
4048 }
4049 return rv;
4050 }
4051
4052 static void md_free(struct kobject *ko)
4053 {
4054 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4055
4056 if (mddev->sysfs_state)
4057 sysfs_put(mddev->sysfs_state);
4058
4059 if (mddev->gendisk) {
4060 del_gendisk(mddev->gendisk);
4061 put_disk(mddev->gendisk);
4062 }
4063 if (mddev->queue)
4064 blk_cleanup_queue(mddev->queue);
4065
4066 kfree(mddev);
4067 }
4068
4069 static const struct sysfs_ops md_sysfs_ops = {
4070 .show = md_attr_show,
4071 .store = md_attr_store,
4072 };
4073 static struct kobj_type md_ktype = {
4074 .release = md_free,
4075 .sysfs_ops = &md_sysfs_ops,
4076 .default_attrs = md_default_attrs,
4077 };
4078
4079 int mdp_major = 0;
4080
4081 static void mddev_delayed_delete(struct work_struct *ws)
4082 {
4083 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4084
4085 if (mddev->private) {
4086 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4087 if (mddev->private != (void*)1)
4088 sysfs_remove_group(&mddev->kobj, mddev->private);
4089 if (mddev->sysfs_action)
4090 sysfs_put(mddev->sysfs_action);
4091 mddev->sysfs_action = NULL;
4092 mddev->private = NULL;
4093 }
4094 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4095 kobject_del(&mddev->kobj);
4096 kobject_put(&mddev->kobj);
4097 }
4098
4099 static int md_alloc(dev_t dev, char *name)
4100 {
4101 static DEFINE_MUTEX(disks_mutex);
4102 mddev_t *mddev = mddev_find(dev);
4103 struct gendisk *disk;
4104 int partitioned;
4105 int shift;
4106 int unit;
4107 int error;
4108
4109 if (!mddev)
4110 return -ENODEV;
4111
4112 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4113 shift = partitioned ? MdpMinorShift : 0;
4114 unit = MINOR(mddev->unit) >> shift;
4115
4116 /* wait for any previous instance if this device
4117 * to be completed removed (mddev_delayed_delete).
4118 */
4119 flush_scheduled_work();
4120
4121 mutex_lock(&disks_mutex);
4122 error = -EEXIST;
4123 if (mddev->gendisk)
4124 goto abort;
4125
4126 if (name) {
4127 /* Need to ensure that 'name' is not a duplicate.
4128 */
4129 mddev_t *mddev2;
4130 spin_lock(&all_mddevs_lock);
4131
4132 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4133 if (mddev2->gendisk &&
4134 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4135 spin_unlock(&all_mddevs_lock);
4136 goto abort;
4137 }
4138 spin_unlock(&all_mddevs_lock);
4139 }
4140
4141 error = -ENOMEM;
4142 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4143 if (!mddev->queue)
4144 goto abort;
4145 mddev->queue->queuedata = mddev;
4146
4147 /* Can be unlocked because the queue is new: no concurrency */
4148 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4149
4150 blk_queue_make_request(mddev->queue, md_make_request);
4151
4152 disk = alloc_disk(1 << shift);
4153 if (!disk) {
4154 blk_cleanup_queue(mddev->queue);
4155 mddev->queue = NULL;
4156 goto abort;
4157 }
4158 disk->major = MAJOR(mddev->unit);
4159 disk->first_minor = unit << shift;
4160 if (name)
4161 strcpy(disk->disk_name, name);
4162 else if (partitioned)
4163 sprintf(disk->disk_name, "md_d%d", unit);
4164 else
4165 sprintf(disk->disk_name, "md%d", unit);
4166 disk->fops = &md_fops;
4167 disk->private_data = mddev;
4168 disk->queue = mddev->queue;
4169 /* Allow extended partitions. This makes the
4170 * 'mdp' device redundant, but we can't really
4171 * remove it now.
4172 */
4173 disk->flags |= GENHD_FL_EXT_DEVT;
4174 add_disk(disk);
4175 mddev->gendisk = disk;
4176 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4177 &disk_to_dev(disk)->kobj, "%s", "md");
4178 if (error) {
4179 /* This isn't possible, but as kobject_init_and_add is marked
4180 * __must_check, we must do something with the result
4181 */
4182 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4183 disk->disk_name);
4184 error = 0;
4185 }
4186 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4187 printk(KERN_DEBUG "pointless warning\n");
4188 abort:
4189 mutex_unlock(&disks_mutex);
4190 if (!error) {
4191 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4192 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4193 }
4194 mddev_put(mddev);
4195 return error;
4196 }
4197
4198 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4199 {
4200 md_alloc(dev, NULL);
4201 return NULL;
4202 }
4203
4204 static int add_named_array(const char *val, struct kernel_param *kp)
4205 {
4206 /* val must be "md_*" where * is not all digits.
4207 * We allocate an array with a large free minor number, and
4208 * set the name to val. val must not already be an active name.
4209 */
4210 int len = strlen(val);
4211 char buf[DISK_NAME_LEN];
4212
4213 while (len && val[len-1] == '\n')
4214 len--;
4215 if (len >= DISK_NAME_LEN)
4216 return -E2BIG;
4217 strlcpy(buf, val, len+1);
4218 if (strncmp(buf, "md_", 3) != 0)
4219 return -EINVAL;
4220 return md_alloc(0, buf);
4221 }
4222
4223 static void md_safemode_timeout(unsigned long data)
4224 {
4225 mddev_t *mddev = (mddev_t *) data;
4226
4227 if (!atomic_read(&mddev->writes_pending)) {
4228 mddev->safemode = 1;
4229 if (mddev->external)
4230 sysfs_notify_dirent(mddev->sysfs_state);
4231 }
4232 md_wakeup_thread(mddev->thread);
4233 }
4234
4235 static int start_dirty_degraded;
4236
4237 static int do_md_run(mddev_t * mddev)
4238 {
4239 int err;
4240 mdk_rdev_t *rdev;
4241 struct gendisk *disk;
4242 struct mdk_personality *pers;
4243
4244 if (list_empty(&mddev->disks))
4245 /* cannot run an array with no devices.. */
4246 return -EINVAL;
4247
4248 if (mddev->pers)
4249 return -EBUSY;
4250
4251 /*
4252 * Analyze all RAID superblock(s)
4253 */
4254 if (!mddev->raid_disks) {
4255 if (!mddev->persistent)
4256 return -EINVAL;
4257 analyze_sbs(mddev);
4258 }
4259
4260 if (mddev->level != LEVEL_NONE)
4261 request_module("md-level-%d", mddev->level);
4262 else if (mddev->clevel[0])
4263 request_module("md-%s", mddev->clevel);
4264
4265 /*
4266 * Drop all container device buffers, from now on
4267 * the only valid external interface is through the md
4268 * device.
4269 */
4270 list_for_each_entry(rdev, &mddev->disks, same_set) {
4271 if (test_bit(Faulty, &rdev->flags))
4272 continue;
4273 sync_blockdev(rdev->bdev);
4274 invalidate_bdev(rdev->bdev);
4275
4276 /* perform some consistency tests on the device.
4277 * We don't want the data to overlap the metadata,
4278 * Internal Bitmap issues have been handled elsewhere.
4279 */
4280 if (rdev->data_offset < rdev->sb_start) {
4281 if (mddev->dev_sectors &&
4282 rdev->data_offset + mddev->dev_sectors
4283 > rdev->sb_start) {
4284 printk("md: %s: data overlaps metadata\n",
4285 mdname(mddev));
4286 return -EINVAL;
4287 }
4288 } else {
4289 if (rdev->sb_start + rdev->sb_size/512
4290 > rdev->data_offset) {
4291 printk("md: %s: metadata overlaps data\n",
4292 mdname(mddev));
4293 return -EINVAL;
4294 }
4295 }
4296 sysfs_notify_dirent(rdev->sysfs_state);
4297 }
4298
4299 disk = mddev->gendisk;
4300
4301 spin_lock(&pers_lock);
4302 pers = find_pers(mddev->level, mddev->clevel);
4303 if (!pers || !try_module_get(pers->owner)) {
4304 spin_unlock(&pers_lock);
4305 if (mddev->level != LEVEL_NONE)
4306 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4307 mddev->level);
4308 else
4309 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4310 mddev->clevel);
4311 return -EINVAL;
4312 }
4313 mddev->pers = pers;
4314 spin_unlock(&pers_lock);
4315 if (mddev->level != pers->level) {
4316 mddev->level = pers->level;
4317 mddev->new_level = pers->level;
4318 }
4319 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4320
4321 if (mddev->reshape_position != MaxSector &&
4322 pers->start_reshape == NULL) {
4323 /* This personality cannot handle reshaping... */
4324 mddev->pers = NULL;
4325 module_put(pers->owner);
4326 return -EINVAL;
4327 }
4328
4329 if (pers->sync_request) {
4330 /* Warn if this is a potentially silly
4331 * configuration.
4332 */
4333 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4334 mdk_rdev_t *rdev2;
4335 int warned = 0;
4336
4337 list_for_each_entry(rdev, &mddev->disks, same_set)
4338 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4339 if (rdev < rdev2 &&
4340 rdev->bdev->bd_contains ==
4341 rdev2->bdev->bd_contains) {
4342 printk(KERN_WARNING
4343 "%s: WARNING: %s appears to be"
4344 " on the same physical disk as"
4345 " %s.\n",
4346 mdname(mddev),
4347 bdevname(rdev->bdev,b),
4348 bdevname(rdev2->bdev,b2));
4349 warned = 1;
4350 }
4351 }
4352
4353 if (warned)
4354 printk(KERN_WARNING
4355 "True protection against single-disk"
4356 " failure might be compromised.\n");
4357 }
4358
4359 mddev->recovery = 0;
4360 /* may be over-ridden by personality */
4361 mddev->resync_max_sectors = mddev->dev_sectors;
4362
4363 mddev->barriers_work = 1;
4364 mddev->ok_start_degraded = start_dirty_degraded;
4365
4366 if (start_readonly && mddev->ro == 0)
4367 mddev->ro = 2; /* read-only, but switch on first write */
4368
4369 err = mddev->pers->run(mddev);
4370 if (err)
4371 printk(KERN_ERR "md: pers->run() failed ...\n");
4372 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4373 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4374 " but 'external_size' not in effect?\n", __func__);
4375 printk(KERN_ERR
4376 "md: invalid array_size %llu > default size %llu\n",
4377 (unsigned long long)mddev->array_sectors / 2,
4378 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4379 err = -EINVAL;
4380 mddev->pers->stop(mddev);
4381 }
4382 if (err == 0 && mddev->pers->sync_request) {
4383 err = bitmap_create(mddev);
4384 if (err) {
4385 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4386 mdname(mddev), err);
4387 mddev->pers->stop(mddev);
4388 }
4389 }
4390 if (err) {
4391 module_put(mddev->pers->owner);
4392 mddev->pers = NULL;
4393 bitmap_destroy(mddev);
4394 return err;
4395 }
4396 if (mddev->pers->sync_request) {
4397 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4398 printk(KERN_WARNING
4399 "md: cannot register extra attributes for %s\n",
4400 mdname(mddev));
4401 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4402 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4403 mddev->ro = 0;
4404
4405 atomic_set(&mddev->writes_pending,0);
4406 atomic_set(&mddev->max_corr_read_errors,
4407 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4408 mddev->safemode = 0;
4409 mddev->safemode_timer.function = md_safemode_timeout;
4410 mddev->safemode_timer.data = (unsigned long) mddev;
4411 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4412 mddev->in_sync = 1;
4413
4414 list_for_each_entry(rdev, &mddev->disks, same_set)
4415 if (rdev->raid_disk >= 0) {
4416 char nm[20];
4417 sprintf(nm, "rd%d", rdev->raid_disk);
4418 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4419 printk("md: cannot register %s for %s\n",
4420 nm, mdname(mddev));
4421 }
4422
4423 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4424
4425 if (mddev->flags)
4426 md_update_sb(mddev, 0);
4427
4428 set_capacity(disk, mddev->array_sectors);
4429
4430 md_wakeup_thread(mddev->thread);
4431 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4432
4433 revalidate_disk(mddev->gendisk);
4434 mddev->changed = 1;
4435 md_new_event(mddev);
4436 sysfs_notify_dirent(mddev->sysfs_state);
4437 if (mddev->sysfs_action)
4438 sysfs_notify_dirent(mddev->sysfs_action);
4439 sysfs_notify(&mddev->kobj, NULL, "degraded");
4440 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4441 return 0;
4442 }
4443
4444 static int restart_array(mddev_t *mddev)
4445 {
4446 struct gendisk *disk = mddev->gendisk;
4447
4448 /* Complain if it has no devices */
4449 if (list_empty(&mddev->disks))
4450 return -ENXIO;
4451 if (!mddev->pers)
4452 return -EINVAL;
4453 if (!mddev->ro)
4454 return -EBUSY;
4455 mddev->safemode = 0;
4456 mddev->ro = 0;
4457 set_disk_ro(disk, 0);
4458 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4459 mdname(mddev));
4460 /* Kick recovery or resync if necessary */
4461 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4462 md_wakeup_thread(mddev->thread);
4463 md_wakeup_thread(mddev->sync_thread);
4464 sysfs_notify_dirent(mddev->sysfs_state);
4465 return 0;
4466 }
4467
4468 /* similar to deny_write_access, but accounts for our holding a reference
4469 * to the file ourselves */
4470 static int deny_bitmap_write_access(struct file * file)
4471 {
4472 struct inode *inode = file->f_mapping->host;
4473
4474 spin_lock(&inode->i_lock);
4475 if (atomic_read(&inode->i_writecount) > 1) {
4476 spin_unlock(&inode->i_lock);
4477 return -ETXTBSY;
4478 }
4479 atomic_set(&inode->i_writecount, -1);
4480 spin_unlock(&inode->i_lock);
4481
4482 return 0;
4483 }
4484
4485 void restore_bitmap_write_access(struct file *file)
4486 {
4487 struct inode *inode = file->f_mapping->host;
4488
4489 spin_lock(&inode->i_lock);
4490 atomic_set(&inode->i_writecount, 1);
4491 spin_unlock(&inode->i_lock);
4492 }
4493
4494 /* mode:
4495 * 0 - completely stop and dis-assemble array
4496 * 1 - switch to readonly
4497 * 2 - stop but do not disassemble array
4498 */
4499 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4500 {
4501 int err = 0;
4502 struct gendisk *disk = mddev->gendisk;
4503 mdk_rdev_t *rdev;
4504
4505 mutex_lock(&mddev->open_mutex);
4506 if (atomic_read(&mddev->openers) > is_open) {
4507 printk("md: %s still in use.\n",mdname(mddev));
4508 err = -EBUSY;
4509 } else if (mddev->pers) {
4510
4511 if (mddev->sync_thread) {
4512 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4513 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4514 md_unregister_thread(mddev->sync_thread);
4515 mddev->sync_thread = NULL;
4516 }
4517
4518 del_timer_sync(&mddev->safemode_timer);
4519
4520 switch(mode) {
4521 case 1: /* readonly */
4522 err = -ENXIO;
4523 if (mddev->ro==1)
4524 goto out;
4525 mddev->ro = 1;
4526 break;
4527 case 0: /* disassemble */
4528 case 2: /* stop */
4529 bitmap_flush(mddev);
4530 md_super_wait(mddev);
4531 if (mddev->ro)
4532 set_disk_ro(disk, 0);
4533
4534 mddev->pers->stop(mddev);
4535 mddev->queue->merge_bvec_fn = NULL;
4536 mddev->queue->unplug_fn = NULL;
4537 mddev->queue->backing_dev_info.congested_fn = NULL;
4538 module_put(mddev->pers->owner);
4539 if (mddev->pers->sync_request && mddev->private == NULL)
4540 mddev->private = (void*)1;
4541 mddev->pers = NULL;
4542 /* tell userspace to handle 'inactive' */
4543 sysfs_notify_dirent(mddev->sysfs_state);
4544
4545 list_for_each_entry(rdev, &mddev->disks, same_set)
4546 if (rdev->raid_disk >= 0) {
4547 char nm[20];
4548 sprintf(nm, "rd%d", rdev->raid_disk);
4549 sysfs_remove_link(&mddev->kobj, nm);
4550 }
4551
4552 set_capacity(disk, 0);
4553 mddev->changed = 1;
4554
4555 if (mddev->ro)
4556 mddev->ro = 0;
4557 }
4558 if (!mddev->in_sync || mddev->flags) {
4559 /* mark array as shutdown cleanly */
4560 mddev->in_sync = 1;
4561 md_update_sb(mddev, 1);
4562 }
4563 if (mode == 1)
4564 set_disk_ro(disk, 1);
4565 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4566 err = 0;
4567 }
4568 out:
4569 mutex_unlock(&mddev->open_mutex);
4570 if (err)
4571 return err;
4572 /*
4573 * Free resources if final stop
4574 */
4575 if (mode == 0) {
4576
4577 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4578
4579 bitmap_destroy(mddev);
4580 if (mddev->bitmap_info.file) {
4581 restore_bitmap_write_access(mddev->bitmap_info.file);
4582 fput(mddev->bitmap_info.file);
4583 mddev->bitmap_info.file = NULL;
4584 }
4585 mddev->bitmap_info.offset = 0;
4586
4587 export_array(mddev);
4588
4589 mddev->array_sectors = 0;
4590 mddev->external_size = 0;
4591 mddev->dev_sectors = 0;
4592 mddev->raid_disks = 0;
4593 mddev->recovery_cp = 0;
4594 mddev->resync_min = 0;
4595 mddev->resync_max = MaxSector;
4596 mddev->reshape_position = MaxSector;
4597 mddev->external = 0;
4598 mddev->persistent = 0;
4599 mddev->level = LEVEL_NONE;
4600 mddev->clevel[0] = 0;
4601 mddev->flags = 0;
4602 mddev->ro = 0;
4603 mddev->metadata_type[0] = 0;
4604 mddev->chunk_sectors = 0;
4605 mddev->ctime = mddev->utime = 0;
4606 mddev->layout = 0;
4607 mddev->max_disks = 0;
4608 mddev->events = 0;
4609 mddev->delta_disks = 0;
4610 mddev->new_level = LEVEL_NONE;
4611 mddev->new_layout = 0;
4612 mddev->new_chunk_sectors = 0;
4613 mddev->curr_resync = 0;
4614 mddev->resync_mismatches = 0;
4615 mddev->suspend_lo = mddev->suspend_hi = 0;
4616 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4617 mddev->recovery = 0;
4618 mddev->in_sync = 0;
4619 mddev->changed = 0;
4620 mddev->degraded = 0;
4621 mddev->barriers_work = 0;
4622 mddev->safemode = 0;
4623 mddev->bitmap_info.offset = 0;
4624 mddev->bitmap_info.default_offset = 0;
4625 mddev->bitmap_info.chunksize = 0;
4626 mddev->bitmap_info.daemon_sleep = 0;
4627 mddev->bitmap_info.max_write_behind = 0;
4628 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4629 if (mddev->hold_active == UNTIL_STOP)
4630 mddev->hold_active = 0;
4631
4632 } else if (mddev->pers)
4633 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4634 mdname(mddev));
4635 err = 0;
4636 blk_integrity_unregister(disk);
4637 md_new_event(mddev);
4638 sysfs_notify_dirent(mddev->sysfs_state);
4639 return err;
4640 }
4641
4642 #ifndef MODULE
4643 static void autorun_array(mddev_t *mddev)
4644 {
4645 mdk_rdev_t *rdev;
4646 int err;
4647
4648 if (list_empty(&mddev->disks))
4649 return;
4650
4651 printk(KERN_INFO "md: running: ");
4652
4653 list_for_each_entry(rdev, &mddev->disks, same_set) {
4654 char b[BDEVNAME_SIZE];
4655 printk("<%s>", bdevname(rdev->bdev,b));
4656 }
4657 printk("\n");
4658
4659 err = do_md_run(mddev);
4660 if (err) {
4661 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4662 do_md_stop(mddev, 0, 0);
4663 }
4664 }
4665
4666 /*
4667 * lets try to run arrays based on all disks that have arrived
4668 * until now. (those are in pending_raid_disks)
4669 *
4670 * the method: pick the first pending disk, collect all disks with
4671 * the same UUID, remove all from the pending list and put them into
4672 * the 'same_array' list. Then order this list based on superblock
4673 * update time (freshest comes first), kick out 'old' disks and
4674 * compare superblocks. If everything's fine then run it.
4675 *
4676 * If "unit" is allocated, then bump its reference count
4677 */
4678 static void autorun_devices(int part)
4679 {
4680 mdk_rdev_t *rdev0, *rdev, *tmp;
4681 mddev_t *mddev;
4682 char b[BDEVNAME_SIZE];
4683
4684 printk(KERN_INFO "md: autorun ...\n");
4685 while (!list_empty(&pending_raid_disks)) {
4686 int unit;
4687 dev_t dev;
4688 LIST_HEAD(candidates);
4689 rdev0 = list_entry(pending_raid_disks.next,
4690 mdk_rdev_t, same_set);
4691
4692 printk(KERN_INFO "md: considering %s ...\n",
4693 bdevname(rdev0->bdev,b));
4694 INIT_LIST_HEAD(&candidates);
4695 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4696 if (super_90_load(rdev, rdev0, 0) >= 0) {
4697 printk(KERN_INFO "md: adding %s ...\n",
4698 bdevname(rdev->bdev,b));
4699 list_move(&rdev->same_set, &candidates);
4700 }
4701 /*
4702 * now we have a set of devices, with all of them having
4703 * mostly sane superblocks. It's time to allocate the
4704 * mddev.
4705 */
4706 if (part) {
4707 dev = MKDEV(mdp_major,
4708 rdev0->preferred_minor << MdpMinorShift);
4709 unit = MINOR(dev) >> MdpMinorShift;
4710 } else {
4711 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4712 unit = MINOR(dev);
4713 }
4714 if (rdev0->preferred_minor != unit) {
4715 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4716 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4717 break;
4718 }
4719
4720 md_probe(dev, NULL, NULL);
4721 mddev = mddev_find(dev);
4722 if (!mddev || !mddev->gendisk) {
4723 if (mddev)
4724 mddev_put(mddev);
4725 printk(KERN_ERR
4726 "md: cannot allocate memory for md drive.\n");
4727 break;
4728 }
4729 if (mddev_lock(mddev))
4730 printk(KERN_WARNING "md: %s locked, cannot run\n",
4731 mdname(mddev));
4732 else if (mddev->raid_disks || mddev->major_version
4733 || !list_empty(&mddev->disks)) {
4734 printk(KERN_WARNING
4735 "md: %s already running, cannot run %s\n",
4736 mdname(mddev), bdevname(rdev0->bdev,b));
4737 mddev_unlock(mddev);
4738 } else {
4739 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4740 mddev->persistent = 1;
4741 rdev_for_each_list(rdev, tmp, &candidates) {
4742 list_del_init(&rdev->same_set);
4743 if (bind_rdev_to_array(rdev, mddev))
4744 export_rdev(rdev);
4745 }
4746 autorun_array(mddev);
4747 mddev_unlock(mddev);
4748 }
4749 /* on success, candidates will be empty, on error
4750 * it won't...
4751 */
4752 rdev_for_each_list(rdev, tmp, &candidates) {
4753 list_del_init(&rdev->same_set);
4754 export_rdev(rdev);
4755 }
4756 mddev_put(mddev);
4757 }
4758 printk(KERN_INFO "md: ... autorun DONE.\n");
4759 }
4760 #endif /* !MODULE */
4761
4762 static int get_version(void __user * arg)
4763 {
4764 mdu_version_t ver;
4765
4766 ver.major = MD_MAJOR_VERSION;
4767 ver.minor = MD_MINOR_VERSION;
4768 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4769
4770 if (copy_to_user(arg, &ver, sizeof(ver)))
4771 return -EFAULT;
4772
4773 return 0;
4774 }
4775
4776 static int get_array_info(mddev_t * mddev, void __user * arg)
4777 {
4778 mdu_array_info_t info;
4779 int nr,working,insync,failed,spare;
4780 mdk_rdev_t *rdev;
4781
4782 nr=working=insync=failed=spare=0;
4783 list_for_each_entry(rdev, &mddev->disks, same_set) {
4784 nr++;
4785 if (test_bit(Faulty, &rdev->flags))
4786 failed++;
4787 else {
4788 working++;
4789 if (test_bit(In_sync, &rdev->flags))
4790 insync++;
4791 else
4792 spare++;
4793 }
4794 }
4795
4796 info.major_version = mddev->major_version;
4797 info.minor_version = mddev->minor_version;
4798 info.patch_version = MD_PATCHLEVEL_VERSION;
4799 info.ctime = mddev->ctime;
4800 info.level = mddev->level;
4801 info.size = mddev->dev_sectors / 2;
4802 if (info.size != mddev->dev_sectors / 2) /* overflow */
4803 info.size = -1;
4804 info.nr_disks = nr;
4805 info.raid_disks = mddev->raid_disks;
4806 info.md_minor = mddev->md_minor;
4807 info.not_persistent= !mddev->persistent;
4808
4809 info.utime = mddev->utime;
4810 info.state = 0;
4811 if (mddev->in_sync)
4812 info.state = (1<<MD_SB_CLEAN);
4813 if (mddev->bitmap && mddev->bitmap_info.offset)
4814 info.state = (1<<MD_SB_BITMAP_PRESENT);
4815 info.active_disks = insync;
4816 info.working_disks = working;
4817 info.failed_disks = failed;
4818 info.spare_disks = spare;
4819
4820 info.layout = mddev->layout;
4821 info.chunk_size = mddev->chunk_sectors << 9;
4822
4823 if (copy_to_user(arg, &info, sizeof(info)))
4824 return -EFAULT;
4825
4826 return 0;
4827 }
4828
4829 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4830 {
4831 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4832 char *ptr, *buf = NULL;
4833 int err = -ENOMEM;
4834
4835 if (md_allow_write(mddev))
4836 file = kmalloc(sizeof(*file), GFP_NOIO);
4837 else
4838 file = kmalloc(sizeof(*file), GFP_KERNEL);
4839
4840 if (!file)
4841 goto out;
4842
4843 /* bitmap disabled, zero the first byte and copy out */
4844 if (!mddev->bitmap || !mddev->bitmap->file) {
4845 file->pathname[0] = '\0';
4846 goto copy_out;
4847 }
4848
4849 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4850 if (!buf)
4851 goto out;
4852
4853 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4854 if (IS_ERR(ptr))
4855 goto out;
4856
4857 strcpy(file->pathname, ptr);
4858
4859 copy_out:
4860 err = 0;
4861 if (copy_to_user(arg, file, sizeof(*file)))
4862 err = -EFAULT;
4863 out:
4864 kfree(buf);
4865 kfree(file);
4866 return err;
4867 }
4868
4869 static int get_disk_info(mddev_t * mddev, void __user * arg)
4870 {
4871 mdu_disk_info_t info;
4872 mdk_rdev_t *rdev;
4873
4874 if (copy_from_user(&info, arg, sizeof(info)))
4875 return -EFAULT;
4876
4877 rdev = find_rdev_nr(mddev, info.number);
4878 if (rdev) {
4879 info.major = MAJOR(rdev->bdev->bd_dev);
4880 info.minor = MINOR(rdev->bdev->bd_dev);
4881 info.raid_disk = rdev->raid_disk;
4882 info.state = 0;
4883 if (test_bit(Faulty, &rdev->flags))
4884 info.state |= (1<<MD_DISK_FAULTY);
4885 else if (test_bit(In_sync, &rdev->flags)) {
4886 info.state |= (1<<MD_DISK_ACTIVE);
4887 info.state |= (1<<MD_DISK_SYNC);
4888 }
4889 if (test_bit(WriteMostly, &rdev->flags))
4890 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4891 } else {
4892 info.major = info.minor = 0;
4893 info.raid_disk = -1;
4894 info.state = (1<<MD_DISK_REMOVED);
4895 }
4896
4897 if (copy_to_user(arg, &info, sizeof(info)))
4898 return -EFAULT;
4899
4900 return 0;
4901 }
4902
4903 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4904 {
4905 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4906 mdk_rdev_t *rdev;
4907 dev_t dev = MKDEV(info->major,info->minor);
4908
4909 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4910 return -EOVERFLOW;
4911
4912 if (!mddev->raid_disks) {
4913 int err;
4914 /* expecting a device which has a superblock */
4915 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4916 if (IS_ERR(rdev)) {
4917 printk(KERN_WARNING
4918 "md: md_import_device returned %ld\n",
4919 PTR_ERR(rdev));
4920 return PTR_ERR(rdev);
4921 }
4922 if (!list_empty(&mddev->disks)) {
4923 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4924 mdk_rdev_t, same_set);
4925 err = super_types[mddev->major_version]
4926 .load_super(rdev, rdev0, mddev->minor_version);
4927 if (err < 0) {
4928 printk(KERN_WARNING
4929 "md: %s has different UUID to %s\n",
4930 bdevname(rdev->bdev,b),
4931 bdevname(rdev0->bdev,b2));
4932 export_rdev(rdev);
4933 return -EINVAL;
4934 }
4935 }
4936 err = bind_rdev_to_array(rdev, mddev);
4937 if (err)
4938 export_rdev(rdev);
4939 return err;
4940 }
4941
4942 /*
4943 * add_new_disk can be used once the array is assembled
4944 * to add "hot spares". They must already have a superblock
4945 * written
4946 */
4947 if (mddev->pers) {
4948 int err;
4949 if (!mddev->pers->hot_add_disk) {
4950 printk(KERN_WARNING
4951 "%s: personality does not support diskops!\n",
4952 mdname(mddev));
4953 return -EINVAL;
4954 }
4955 if (mddev->persistent)
4956 rdev = md_import_device(dev, mddev->major_version,
4957 mddev->minor_version);
4958 else
4959 rdev = md_import_device(dev, -1, -1);
4960 if (IS_ERR(rdev)) {
4961 printk(KERN_WARNING
4962 "md: md_import_device returned %ld\n",
4963 PTR_ERR(rdev));
4964 return PTR_ERR(rdev);
4965 }
4966 /* set save_raid_disk if appropriate */
4967 if (!mddev->persistent) {
4968 if (info->state & (1<<MD_DISK_SYNC) &&
4969 info->raid_disk < mddev->raid_disks)
4970 rdev->raid_disk = info->raid_disk;
4971 else
4972 rdev->raid_disk = -1;
4973 } else
4974 super_types[mddev->major_version].
4975 validate_super(mddev, rdev);
4976 rdev->saved_raid_disk = rdev->raid_disk;
4977
4978 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4979 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4980 set_bit(WriteMostly, &rdev->flags);
4981 else
4982 clear_bit(WriteMostly, &rdev->flags);
4983
4984 rdev->raid_disk = -1;
4985 err = bind_rdev_to_array(rdev, mddev);
4986 if (!err && !mddev->pers->hot_remove_disk) {
4987 /* If there is hot_add_disk but no hot_remove_disk
4988 * then added disks for geometry changes,
4989 * and should be added immediately.
4990 */
4991 super_types[mddev->major_version].
4992 validate_super(mddev, rdev);
4993 err = mddev->pers->hot_add_disk(mddev, rdev);
4994 if (err)
4995 unbind_rdev_from_array(rdev);
4996 }
4997 if (err)
4998 export_rdev(rdev);
4999 else
5000 sysfs_notify_dirent(rdev->sysfs_state);
5001
5002 md_update_sb(mddev, 1);
5003 if (mddev->degraded)
5004 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5005 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5006 md_wakeup_thread(mddev->thread);
5007 return err;
5008 }
5009
5010 /* otherwise, add_new_disk is only allowed
5011 * for major_version==0 superblocks
5012 */
5013 if (mddev->major_version != 0) {
5014 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5015 mdname(mddev));
5016 return -EINVAL;
5017 }
5018
5019 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5020 int err;
5021 rdev = md_import_device(dev, -1, 0);
5022 if (IS_ERR(rdev)) {
5023 printk(KERN_WARNING
5024 "md: error, md_import_device() returned %ld\n",
5025 PTR_ERR(rdev));
5026 return PTR_ERR(rdev);
5027 }
5028 rdev->desc_nr = info->number;
5029 if (info->raid_disk < mddev->raid_disks)
5030 rdev->raid_disk = info->raid_disk;
5031 else
5032 rdev->raid_disk = -1;
5033
5034 if (rdev->raid_disk < mddev->raid_disks)
5035 if (info->state & (1<<MD_DISK_SYNC))
5036 set_bit(In_sync, &rdev->flags);
5037
5038 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5039 set_bit(WriteMostly, &rdev->flags);
5040
5041 if (!mddev->persistent) {
5042 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5043 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5044 } else
5045 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5046 rdev->sectors = rdev->sb_start;
5047
5048 err = bind_rdev_to_array(rdev, mddev);
5049 if (err) {
5050 export_rdev(rdev);
5051 return err;
5052 }
5053 }
5054
5055 return 0;
5056 }
5057
5058 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5059 {
5060 char b[BDEVNAME_SIZE];
5061 mdk_rdev_t *rdev;
5062
5063 rdev = find_rdev(mddev, dev);
5064 if (!rdev)
5065 return -ENXIO;
5066
5067 if (rdev->raid_disk >= 0)
5068 goto busy;
5069
5070 kick_rdev_from_array(rdev);
5071 md_update_sb(mddev, 1);
5072 md_new_event(mddev);
5073
5074 return 0;
5075 busy:
5076 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5077 bdevname(rdev->bdev,b), mdname(mddev));
5078 return -EBUSY;
5079 }
5080
5081 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5082 {
5083 char b[BDEVNAME_SIZE];
5084 int err;
5085 mdk_rdev_t *rdev;
5086
5087 if (!mddev->pers)
5088 return -ENODEV;
5089
5090 if (mddev->major_version != 0) {
5091 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5092 " version-0 superblocks.\n",
5093 mdname(mddev));
5094 return -EINVAL;
5095 }
5096 if (!mddev->pers->hot_add_disk) {
5097 printk(KERN_WARNING
5098 "%s: personality does not support diskops!\n",
5099 mdname(mddev));
5100 return -EINVAL;
5101 }
5102
5103 rdev = md_import_device(dev, -1, 0);
5104 if (IS_ERR(rdev)) {
5105 printk(KERN_WARNING
5106 "md: error, md_import_device() returned %ld\n",
5107 PTR_ERR(rdev));
5108 return -EINVAL;
5109 }
5110
5111 if (mddev->persistent)
5112 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5113 else
5114 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5115
5116 rdev->sectors = rdev->sb_start;
5117
5118 if (test_bit(Faulty, &rdev->flags)) {
5119 printk(KERN_WARNING
5120 "md: can not hot-add faulty %s disk to %s!\n",
5121 bdevname(rdev->bdev,b), mdname(mddev));
5122 err = -EINVAL;
5123 goto abort_export;
5124 }
5125 clear_bit(In_sync, &rdev->flags);
5126 rdev->desc_nr = -1;
5127 rdev->saved_raid_disk = -1;
5128 err = bind_rdev_to_array(rdev, mddev);
5129 if (err)
5130 goto abort_export;
5131
5132 /*
5133 * The rest should better be atomic, we can have disk failures
5134 * noticed in interrupt contexts ...
5135 */
5136
5137 rdev->raid_disk = -1;
5138
5139 md_update_sb(mddev, 1);
5140
5141 /*
5142 * Kick recovery, maybe this spare has to be added to the
5143 * array immediately.
5144 */
5145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5146 md_wakeup_thread(mddev->thread);
5147 md_new_event(mddev);
5148 return 0;
5149
5150 abort_export:
5151 export_rdev(rdev);
5152 return err;
5153 }
5154
5155 static int set_bitmap_file(mddev_t *mddev, int fd)
5156 {
5157 int err;
5158
5159 if (mddev->pers) {
5160 if (!mddev->pers->quiesce)
5161 return -EBUSY;
5162 if (mddev->recovery || mddev->sync_thread)
5163 return -EBUSY;
5164 /* we should be able to change the bitmap.. */
5165 }
5166
5167
5168 if (fd >= 0) {
5169 if (mddev->bitmap)
5170 return -EEXIST; /* cannot add when bitmap is present */
5171 mddev->bitmap_info.file = fget(fd);
5172
5173 if (mddev->bitmap_info.file == NULL) {
5174 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5175 mdname(mddev));
5176 return -EBADF;
5177 }
5178
5179 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5180 if (err) {
5181 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5182 mdname(mddev));
5183 fput(mddev->bitmap_info.file);
5184 mddev->bitmap_info.file = NULL;
5185 return err;
5186 }
5187 mddev->bitmap_info.offset = 0; /* file overrides offset */
5188 } else if (mddev->bitmap == NULL)
5189 return -ENOENT; /* cannot remove what isn't there */
5190 err = 0;
5191 if (mddev->pers) {
5192 mddev->pers->quiesce(mddev, 1);
5193 if (fd >= 0)
5194 err = bitmap_create(mddev);
5195 if (fd < 0 || err) {
5196 bitmap_destroy(mddev);
5197 fd = -1; /* make sure to put the file */
5198 }
5199 mddev->pers->quiesce(mddev, 0);
5200 }
5201 if (fd < 0) {
5202 if (mddev->bitmap_info.file) {
5203 restore_bitmap_write_access(mddev->bitmap_info.file);
5204 fput(mddev->bitmap_info.file);
5205 }
5206 mddev->bitmap_info.file = NULL;
5207 }
5208
5209 return err;
5210 }
5211
5212 /*
5213 * set_array_info is used two different ways
5214 * The original usage is when creating a new array.
5215 * In this usage, raid_disks is > 0 and it together with
5216 * level, size, not_persistent,layout,chunksize determine the
5217 * shape of the array.
5218 * This will always create an array with a type-0.90.0 superblock.
5219 * The newer usage is when assembling an array.
5220 * In this case raid_disks will be 0, and the major_version field is
5221 * use to determine which style super-blocks are to be found on the devices.
5222 * The minor and patch _version numbers are also kept incase the
5223 * super_block handler wishes to interpret them.
5224 */
5225 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5226 {
5227
5228 if (info->raid_disks == 0) {
5229 /* just setting version number for superblock loading */
5230 if (info->major_version < 0 ||
5231 info->major_version >= ARRAY_SIZE(super_types) ||
5232 super_types[info->major_version].name == NULL) {
5233 /* maybe try to auto-load a module? */
5234 printk(KERN_INFO
5235 "md: superblock version %d not known\n",
5236 info->major_version);
5237 return -EINVAL;
5238 }
5239 mddev->major_version = info->major_version;
5240 mddev->minor_version = info->minor_version;
5241 mddev->patch_version = info->patch_version;
5242 mddev->persistent = !info->not_persistent;
5243 /* ensure mddev_put doesn't delete this now that there
5244 * is some minimal configuration.
5245 */
5246 mddev->ctime = get_seconds();
5247 return 0;
5248 }
5249 mddev->major_version = MD_MAJOR_VERSION;
5250 mddev->minor_version = MD_MINOR_VERSION;
5251 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5252 mddev->ctime = get_seconds();
5253
5254 mddev->level = info->level;
5255 mddev->clevel[0] = 0;
5256 mddev->dev_sectors = 2 * (sector_t)info->size;
5257 mddev->raid_disks = info->raid_disks;
5258 /* don't set md_minor, it is determined by which /dev/md* was
5259 * openned
5260 */
5261 if (info->state & (1<<MD_SB_CLEAN))
5262 mddev->recovery_cp = MaxSector;
5263 else
5264 mddev->recovery_cp = 0;
5265 mddev->persistent = ! info->not_persistent;
5266 mddev->external = 0;
5267
5268 mddev->layout = info->layout;
5269 mddev->chunk_sectors = info->chunk_size >> 9;
5270
5271 mddev->max_disks = MD_SB_DISKS;
5272
5273 if (mddev->persistent)
5274 mddev->flags = 0;
5275 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5276
5277 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5278 mddev->bitmap_info.offset = 0;
5279
5280 mddev->reshape_position = MaxSector;
5281
5282 /*
5283 * Generate a 128 bit UUID
5284 */
5285 get_random_bytes(mddev->uuid, 16);
5286
5287 mddev->new_level = mddev->level;
5288 mddev->new_chunk_sectors = mddev->chunk_sectors;
5289 mddev->new_layout = mddev->layout;
5290 mddev->delta_disks = 0;
5291
5292 return 0;
5293 }
5294
5295 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5296 {
5297 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5298
5299 if (mddev->external_size)
5300 return;
5301
5302 mddev->array_sectors = array_sectors;
5303 }
5304 EXPORT_SYMBOL(md_set_array_sectors);
5305
5306 static int update_size(mddev_t *mddev, sector_t num_sectors)
5307 {
5308 mdk_rdev_t *rdev;
5309 int rv;
5310 int fit = (num_sectors == 0);
5311
5312 if (mddev->pers->resize == NULL)
5313 return -EINVAL;
5314 /* The "num_sectors" is the number of sectors of each device that
5315 * is used. This can only make sense for arrays with redundancy.
5316 * linear and raid0 always use whatever space is available. We can only
5317 * consider changing this number if no resync or reconstruction is
5318 * happening, and if the new size is acceptable. It must fit before the
5319 * sb_start or, if that is <data_offset, it must fit before the size
5320 * of each device. If num_sectors is zero, we find the largest size
5321 * that fits.
5322
5323 */
5324 if (mddev->sync_thread)
5325 return -EBUSY;
5326 if (mddev->bitmap)
5327 /* Sorry, cannot grow a bitmap yet, just remove it,
5328 * grow, and re-add.
5329 */
5330 return -EBUSY;
5331 list_for_each_entry(rdev, &mddev->disks, same_set) {
5332 sector_t avail = rdev->sectors;
5333
5334 if (fit && (num_sectors == 0 || num_sectors > avail))
5335 num_sectors = avail;
5336 if (avail < num_sectors)
5337 return -ENOSPC;
5338 }
5339 rv = mddev->pers->resize(mddev, num_sectors);
5340 if (!rv)
5341 revalidate_disk(mddev->gendisk);
5342 return rv;
5343 }
5344
5345 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5346 {
5347 int rv;
5348 /* change the number of raid disks */
5349 if (mddev->pers->check_reshape == NULL)
5350 return -EINVAL;
5351 if (raid_disks <= 0 ||
5352 raid_disks >= mddev->max_disks)
5353 return -EINVAL;
5354 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5355 return -EBUSY;
5356 mddev->delta_disks = raid_disks - mddev->raid_disks;
5357
5358 rv = mddev->pers->check_reshape(mddev);
5359 return rv;
5360 }
5361
5362
5363 /*
5364 * update_array_info is used to change the configuration of an
5365 * on-line array.
5366 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5367 * fields in the info are checked against the array.
5368 * Any differences that cannot be handled will cause an error.
5369 * Normally, only one change can be managed at a time.
5370 */
5371 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5372 {
5373 int rv = 0;
5374 int cnt = 0;
5375 int state = 0;
5376
5377 /* calculate expected state,ignoring low bits */
5378 if (mddev->bitmap && mddev->bitmap_info.offset)
5379 state |= (1 << MD_SB_BITMAP_PRESENT);
5380
5381 if (mddev->major_version != info->major_version ||
5382 mddev->minor_version != info->minor_version ||
5383 /* mddev->patch_version != info->patch_version || */
5384 mddev->ctime != info->ctime ||
5385 mddev->level != info->level ||
5386 /* mddev->layout != info->layout || */
5387 !mddev->persistent != info->not_persistent||
5388 mddev->chunk_sectors != info->chunk_size >> 9 ||
5389 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5390 ((state^info->state) & 0xfffffe00)
5391 )
5392 return -EINVAL;
5393 /* Check there is only one change */
5394 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5395 cnt++;
5396 if (mddev->raid_disks != info->raid_disks)
5397 cnt++;
5398 if (mddev->layout != info->layout)
5399 cnt++;
5400 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5401 cnt++;
5402 if (cnt == 0)
5403 return 0;
5404 if (cnt > 1)
5405 return -EINVAL;
5406
5407 if (mddev->layout != info->layout) {
5408 /* Change layout
5409 * we don't need to do anything at the md level, the
5410 * personality will take care of it all.
5411 */
5412 if (mddev->pers->check_reshape == NULL)
5413 return -EINVAL;
5414 else {
5415 mddev->new_layout = info->layout;
5416 rv = mddev->pers->check_reshape(mddev);
5417 if (rv)
5418 mddev->new_layout = mddev->layout;
5419 return rv;
5420 }
5421 }
5422 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5423 rv = update_size(mddev, (sector_t)info->size * 2);
5424
5425 if (mddev->raid_disks != info->raid_disks)
5426 rv = update_raid_disks(mddev, info->raid_disks);
5427
5428 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5429 if (mddev->pers->quiesce == NULL)
5430 return -EINVAL;
5431 if (mddev->recovery || mddev->sync_thread)
5432 return -EBUSY;
5433 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5434 /* add the bitmap */
5435 if (mddev->bitmap)
5436 return -EEXIST;
5437 if (mddev->bitmap_info.default_offset == 0)
5438 return -EINVAL;
5439 mddev->bitmap_info.offset =
5440 mddev->bitmap_info.default_offset;
5441 mddev->pers->quiesce(mddev, 1);
5442 rv = bitmap_create(mddev);
5443 if (rv)
5444 bitmap_destroy(mddev);
5445 mddev->pers->quiesce(mddev, 0);
5446 } else {
5447 /* remove the bitmap */
5448 if (!mddev->bitmap)
5449 return -ENOENT;
5450 if (mddev->bitmap->file)
5451 return -EINVAL;
5452 mddev->pers->quiesce(mddev, 1);
5453 bitmap_destroy(mddev);
5454 mddev->pers->quiesce(mddev, 0);
5455 mddev->bitmap_info.offset = 0;
5456 }
5457 }
5458 md_update_sb(mddev, 1);
5459 return rv;
5460 }
5461
5462 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5463 {
5464 mdk_rdev_t *rdev;
5465
5466 if (mddev->pers == NULL)
5467 return -ENODEV;
5468
5469 rdev = find_rdev(mddev, dev);
5470 if (!rdev)
5471 return -ENODEV;
5472
5473 md_error(mddev, rdev);
5474 return 0;
5475 }
5476
5477 /*
5478 * We have a problem here : there is no easy way to give a CHS
5479 * virtual geometry. We currently pretend that we have a 2 heads
5480 * 4 sectors (with a BIG number of cylinders...). This drives
5481 * dosfs just mad... ;-)
5482 */
5483 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5484 {
5485 mddev_t *mddev = bdev->bd_disk->private_data;
5486
5487 geo->heads = 2;
5488 geo->sectors = 4;
5489 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5490 return 0;
5491 }
5492
5493 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5494 unsigned int cmd, unsigned long arg)
5495 {
5496 int err = 0;
5497 void __user *argp = (void __user *)arg;
5498 mddev_t *mddev = NULL;
5499
5500 if (!capable(CAP_SYS_ADMIN))
5501 return -EACCES;
5502
5503 /*
5504 * Commands dealing with the RAID driver but not any
5505 * particular array:
5506 */
5507 switch (cmd)
5508 {
5509 case RAID_VERSION:
5510 err = get_version(argp);
5511 goto done;
5512
5513 case PRINT_RAID_DEBUG:
5514 err = 0;
5515 md_print_devices();
5516 goto done;
5517
5518 #ifndef MODULE
5519 case RAID_AUTORUN:
5520 err = 0;
5521 autostart_arrays(arg);
5522 goto done;
5523 #endif
5524 default:;
5525 }
5526
5527 /*
5528 * Commands creating/starting a new array:
5529 */
5530
5531 mddev = bdev->bd_disk->private_data;
5532
5533 if (!mddev) {
5534 BUG();
5535 goto abort;
5536 }
5537
5538 err = mddev_lock(mddev);
5539 if (err) {
5540 printk(KERN_INFO
5541 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5542 err, cmd);
5543 goto abort;
5544 }
5545
5546 switch (cmd)
5547 {
5548 case SET_ARRAY_INFO:
5549 {
5550 mdu_array_info_t info;
5551 if (!arg)
5552 memset(&info, 0, sizeof(info));
5553 else if (copy_from_user(&info, argp, sizeof(info))) {
5554 err = -EFAULT;
5555 goto abort_unlock;
5556 }
5557 if (mddev->pers) {
5558 err = update_array_info(mddev, &info);
5559 if (err) {
5560 printk(KERN_WARNING "md: couldn't update"
5561 " array info. %d\n", err);
5562 goto abort_unlock;
5563 }
5564 goto done_unlock;
5565 }
5566 if (!list_empty(&mddev->disks)) {
5567 printk(KERN_WARNING
5568 "md: array %s already has disks!\n",
5569 mdname(mddev));
5570 err = -EBUSY;
5571 goto abort_unlock;
5572 }
5573 if (mddev->raid_disks) {
5574 printk(KERN_WARNING
5575 "md: array %s already initialised!\n",
5576 mdname(mddev));
5577 err = -EBUSY;
5578 goto abort_unlock;
5579 }
5580 err = set_array_info(mddev, &info);
5581 if (err) {
5582 printk(KERN_WARNING "md: couldn't set"
5583 " array info. %d\n", err);
5584 goto abort_unlock;
5585 }
5586 }
5587 goto done_unlock;
5588
5589 default:;
5590 }
5591
5592 /*
5593 * Commands querying/configuring an existing array:
5594 */
5595 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5596 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5597 if ((!mddev->raid_disks && !mddev->external)
5598 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5599 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5600 && cmd != GET_BITMAP_FILE) {
5601 err = -ENODEV;
5602 goto abort_unlock;
5603 }
5604
5605 /*
5606 * Commands even a read-only array can execute:
5607 */
5608 switch (cmd)
5609 {
5610 case GET_ARRAY_INFO:
5611 err = get_array_info(mddev, argp);
5612 goto done_unlock;
5613
5614 case GET_BITMAP_FILE:
5615 err = get_bitmap_file(mddev, argp);
5616 goto done_unlock;
5617
5618 case GET_DISK_INFO:
5619 err = get_disk_info(mddev, argp);
5620 goto done_unlock;
5621
5622 case RESTART_ARRAY_RW:
5623 err = restart_array(mddev);
5624 goto done_unlock;
5625
5626 case STOP_ARRAY:
5627 err = do_md_stop(mddev, 0, 1);
5628 goto done_unlock;
5629
5630 case STOP_ARRAY_RO:
5631 err = do_md_stop(mddev, 1, 1);
5632 goto done_unlock;
5633
5634 }
5635
5636 /*
5637 * The remaining ioctls are changing the state of the
5638 * superblock, so we do not allow them on read-only arrays.
5639 * However non-MD ioctls (e.g. get-size) will still come through
5640 * here and hit the 'default' below, so only disallow
5641 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5642 */
5643 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5644 if (mddev->ro == 2) {
5645 mddev->ro = 0;
5646 sysfs_notify_dirent(mddev->sysfs_state);
5647 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5648 md_wakeup_thread(mddev->thread);
5649 } else {
5650 err = -EROFS;
5651 goto abort_unlock;
5652 }
5653 }
5654
5655 switch (cmd)
5656 {
5657 case ADD_NEW_DISK:
5658 {
5659 mdu_disk_info_t info;
5660 if (copy_from_user(&info, argp, sizeof(info)))
5661 err = -EFAULT;
5662 else
5663 err = add_new_disk(mddev, &info);
5664 goto done_unlock;
5665 }
5666
5667 case HOT_REMOVE_DISK:
5668 err = hot_remove_disk(mddev, new_decode_dev(arg));
5669 goto done_unlock;
5670
5671 case HOT_ADD_DISK:
5672 err = hot_add_disk(mddev, new_decode_dev(arg));
5673 goto done_unlock;
5674
5675 case SET_DISK_FAULTY:
5676 err = set_disk_faulty(mddev, new_decode_dev(arg));
5677 goto done_unlock;
5678
5679 case RUN_ARRAY:
5680 err = do_md_run(mddev);
5681 goto done_unlock;
5682
5683 case SET_BITMAP_FILE:
5684 err = set_bitmap_file(mddev, (int)arg);
5685 goto done_unlock;
5686
5687 default:
5688 err = -EINVAL;
5689 goto abort_unlock;
5690 }
5691
5692 done_unlock:
5693 abort_unlock:
5694 if (mddev->hold_active == UNTIL_IOCTL &&
5695 err != -EINVAL)
5696 mddev->hold_active = 0;
5697 mddev_unlock(mddev);
5698
5699 return err;
5700 done:
5701 if (err)
5702 MD_BUG();
5703 abort:
5704 return err;
5705 }
5706 #ifdef CONFIG_COMPAT
5707 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5708 unsigned int cmd, unsigned long arg)
5709 {
5710 switch (cmd) {
5711 case HOT_REMOVE_DISK:
5712 case HOT_ADD_DISK:
5713 case SET_DISK_FAULTY:
5714 case SET_BITMAP_FILE:
5715 /* These take in integer arg, do not convert */
5716 break;
5717 default:
5718 arg = (unsigned long)compat_ptr(arg);
5719 break;
5720 }
5721
5722 return md_ioctl(bdev, mode, cmd, arg);
5723 }
5724 #endif /* CONFIG_COMPAT */
5725
5726 static int md_open(struct block_device *bdev, fmode_t mode)
5727 {
5728 /*
5729 * Succeed if we can lock the mddev, which confirms that
5730 * it isn't being stopped right now.
5731 */
5732 mddev_t *mddev = mddev_find(bdev->bd_dev);
5733 int err;
5734
5735 if (mddev->gendisk != bdev->bd_disk) {
5736 /* we are racing with mddev_put which is discarding this
5737 * bd_disk.
5738 */
5739 mddev_put(mddev);
5740 /* Wait until bdev->bd_disk is definitely gone */
5741 flush_scheduled_work();
5742 /* Then retry the open from the top */
5743 return -ERESTARTSYS;
5744 }
5745 BUG_ON(mddev != bdev->bd_disk->private_data);
5746
5747 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5748 goto out;
5749
5750 err = 0;
5751 atomic_inc(&mddev->openers);
5752 mutex_unlock(&mddev->open_mutex);
5753
5754 check_disk_change(bdev);
5755 out:
5756 return err;
5757 }
5758
5759 static int md_release(struct gendisk *disk, fmode_t mode)
5760 {
5761 mddev_t *mddev = disk->private_data;
5762
5763 BUG_ON(!mddev);
5764 atomic_dec(&mddev->openers);
5765 mddev_put(mddev);
5766
5767 return 0;
5768 }
5769
5770 static int md_media_changed(struct gendisk *disk)
5771 {
5772 mddev_t *mddev = disk->private_data;
5773
5774 return mddev->changed;
5775 }
5776
5777 static int md_revalidate(struct gendisk *disk)
5778 {
5779 mddev_t *mddev = disk->private_data;
5780
5781 mddev->changed = 0;
5782 return 0;
5783 }
5784 static const struct block_device_operations md_fops =
5785 {
5786 .owner = THIS_MODULE,
5787 .open = md_open,
5788 .release = md_release,
5789 .ioctl = md_ioctl,
5790 #ifdef CONFIG_COMPAT
5791 .compat_ioctl = md_compat_ioctl,
5792 #endif
5793 .getgeo = md_getgeo,
5794 .media_changed = md_media_changed,
5795 .revalidate_disk= md_revalidate,
5796 };
5797
5798 static int md_thread(void * arg)
5799 {
5800 mdk_thread_t *thread = arg;
5801
5802 /*
5803 * md_thread is a 'system-thread', it's priority should be very
5804 * high. We avoid resource deadlocks individually in each
5805 * raid personality. (RAID5 does preallocation) We also use RR and
5806 * the very same RT priority as kswapd, thus we will never get
5807 * into a priority inversion deadlock.
5808 *
5809 * we definitely have to have equal or higher priority than
5810 * bdflush, otherwise bdflush will deadlock if there are too
5811 * many dirty RAID5 blocks.
5812 */
5813
5814 allow_signal(SIGKILL);
5815 while (!kthread_should_stop()) {
5816
5817 /* We need to wait INTERRUPTIBLE so that
5818 * we don't add to the load-average.
5819 * That means we need to be sure no signals are
5820 * pending
5821 */
5822 if (signal_pending(current))
5823 flush_signals(current);
5824
5825 wait_event_interruptible_timeout
5826 (thread->wqueue,
5827 test_bit(THREAD_WAKEUP, &thread->flags)
5828 || kthread_should_stop(),
5829 thread->timeout);
5830
5831 clear_bit(THREAD_WAKEUP, &thread->flags);
5832
5833 thread->run(thread->mddev);
5834 }
5835
5836 return 0;
5837 }
5838
5839 void md_wakeup_thread(mdk_thread_t *thread)
5840 {
5841 if (thread) {
5842 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5843 set_bit(THREAD_WAKEUP, &thread->flags);
5844 wake_up(&thread->wqueue);
5845 }
5846 }
5847
5848 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5849 const char *name)
5850 {
5851 mdk_thread_t *thread;
5852
5853 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5854 if (!thread)
5855 return NULL;
5856
5857 init_waitqueue_head(&thread->wqueue);
5858
5859 thread->run = run;
5860 thread->mddev = mddev;
5861 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5862 thread->tsk = kthread_run(md_thread, thread,
5863 "%s_%s",
5864 mdname(thread->mddev),
5865 name ?: mddev->pers->name);
5866 if (IS_ERR(thread->tsk)) {
5867 kfree(thread);
5868 return NULL;
5869 }
5870 return thread;
5871 }
5872
5873 void md_unregister_thread(mdk_thread_t *thread)
5874 {
5875 if (!thread)
5876 return;
5877 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5878
5879 kthread_stop(thread->tsk);
5880 kfree(thread);
5881 }
5882
5883 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5884 {
5885 if (!mddev) {
5886 MD_BUG();
5887 return;
5888 }
5889
5890 if (!rdev || test_bit(Faulty, &rdev->flags))
5891 return;
5892
5893 if (mddev->external)
5894 set_bit(Blocked, &rdev->flags);
5895 /*
5896 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5897 mdname(mddev),
5898 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5899 __builtin_return_address(0),__builtin_return_address(1),
5900 __builtin_return_address(2),__builtin_return_address(3));
5901 */
5902 if (!mddev->pers)
5903 return;
5904 if (!mddev->pers->error_handler)
5905 return;
5906 mddev->pers->error_handler(mddev,rdev);
5907 if (mddev->degraded)
5908 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5909 set_bit(StateChanged, &rdev->flags);
5910 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5911 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5912 md_wakeup_thread(mddev->thread);
5913 md_new_event_inintr(mddev);
5914 }
5915
5916 /* seq_file implementation /proc/mdstat */
5917
5918 static void status_unused(struct seq_file *seq)
5919 {
5920 int i = 0;
5921 mdk_rdev_t *rdev;
5922
5923 seq_printf(seq, "unused devices: ");
5924
5925 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5926 char b[BDEVNAME_SIZE];
5927 i++;
5928 seq_printf(seq, "%s ",
5929 bdevname(rdev->bdev,b));
5930 }
5931 if (!i)
5932 seq_printf(seq, "<none>");
5933
5934 seq_printf(seq, "\n");
5935 }
5936
5937
5938 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5939 {
5940 sector_t max_sectors, resync, res;
5941 unsigned long dt, db;
5942 sector_t rt;
5943 int scale;
5944 unsigned int per_milli;
5945
5946 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5947
5948 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5949 max_sectors = mddev->resync_max_sectors;
5950 else
5951 max_sectors = mddev->dev_sectors;
5952
5953 /*
5954 * Should not happen.
5955 */
5956 if (!max_sectors) {
5957 MD_BUG();
5958 return;
5959 }
5960 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5961 * in a sector_t, and (max_sectors>>scale) will fit in a
5962 * u32, as those are the requirements for sector_div.
5963 * Thus 'scale' must be at least 10
5964 */
5965 scale = 10;
5966 if (sizeof(sector_t) > sizeof(unsigned long)) {
5967 while ( max_sectors/2 > (1ULL<<(scale+32)))
5968 scale++;
5969 }
5970 res = (resync>>scale)*1000;
5971 sector_div(res, (u32)((max_sectors>>scale)+1));
5972
5973 per_milli = res;
5974 {
5975 int i, x = per_milli/50, y = 20-x;
5976 seq_printf(seq, "[");
5977 for (i = 0; i < x; i++)
5978 seq_printf(seq, "=");
5979 seq_printf(seq, ">");
5980 for (i = 0; i < y; i++)
5981 seq_printf(seq, ".");
5982 seq_printf(seq, "] ");
5983 }
5984 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5985 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5986 "reshape" :
5987 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5988 "check" :
5989 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5990 "resync" : "recovery"))),
5991 per_milli/10, per_milli % 10,
5992 (unsigned long long) resync/2,
5993 (unsigned long long) max_sectors/2);
5994
5995 /*
5996 * dt: time from mark until now
5997 * db: blocks written from mark until now
5998 * rt: remaining time
5999 *
6000 * rt is a sector_t, so could be 32bit or 64bit.
6001 * So we divide before multiply in case it is 32bit and close
6002 * to the limit.
6003 * We scale the divisor (db) by 32 to avoid loosing precision
6004 * near the end of resync when the number of remaining sectors
6005 * is close to 'db'.
6006 * We then divide rt by 32 after multiplying by db to compensate.
6007 * The '+1' avoids division by zero if db is very small.
6008 */
6009 dt = ((jiffies - mddev->resync_mark) / HZ);
6010 if (!dt) dt++;
6011 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6012 - mddev->resync_mark_cnt;
6013
6014 rt = max_sectors - resync; /* number of remaining sectors */
6015 sector_div(rt, db/32+1);
6016 rt *= dt;
6017 rt >>= 5;
6018
6019 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6020 ((unsigned long)rt % 60)/6);
6021
6022 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6023 }
6024
6025 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6026 {
6027 struct list_head *tmp;
6028 loff_t l = *pos;
6029 mddev_t *mddev;
6030
6031 if (l >= 0x10000)
6032 return NULL;
6033 if (!l--)
6034 /* header */
6035 return (void*)1;
6036
6037 spin_lock(&all_mddevs_lock);
6038 list_for_each(tmp,&all_mddevs)
6039 if (!l--) {
6040 mddev = list_entry(tmp, mddev_t, all_mddevs);
6041 mddev_get(mddev);
6042 spin_unlock(&all_mddevs_lock);
6043 return mddev;
6044 }
6045 spin_unlock(&all_mddevs_lock);
6046 if (!l--)
6047 return (void*)2;/* tail */
6048 return NULL;
6049 }
6050
6051 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6052 {
6053 struct list_head *tmp;
6054 mddev_t *next_mddev, *mddev = v;
6055
6056 ++*pos;
6057 if (v == (void*)2)
6058 return NULL;
6059
6060 spin_lock(&all_mddevs_lock);
6061 if (v == (void*)1)
6062 tmp = all_mddevs.next;
6063 else
6064 tmp = mddev->all_mddevs.next;
6065 if (tmp != &all_mddevs)
6066 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6067 else {
6068 next_mddev = (void*)2;
6069 *pos = 0x10000;
6070 }
6071 spin_unlock(&all_mddevs_lock);
6072
6073 if (v != (void*)1)
6074 mddev_put(mddev);
6075 return next_mddev;
6076
6077 }
6078
6079 static void md_seq_stop(struct seq_file *seq, void *v)
6080 {
6081 mddev_t *mddev = v;
6082
6083 if (mddev && v != (void*)1 && v != (void*)2)
6084 mddev_put(mddev);
6085 }
6086
6087 struct mdstat_info {
6088 int event;
6089 };
6090
6091 static int md_seq_show(struct seq_file *seq, void *v)
6092 {
6093 mddev_t *mddev = v;
6094 sector_t sectors;
6095 mdk_rdev_t *rdev;
6096 struct mdstat_info *mi = seq->private;
6097 struct bitmap *bitmap;
6098
6099 if (v == (void*)1) {
6100 struct mdk_personality *pers;
6101 seq_printf(seq, "Personalities : ");
6102 spin_lock(&pers_lock);
6103 list_for_each_entry(pers, &pers_list, list)
6104 seq_printf(seq, "[%s] ", pers->name);
6105
6106 spin_unlock(&pers_lock);
6107 seq_printf(seq, "\n");
6108 mi->event = atomic_read(&md_event_count);
6109 return 0;
6110 }
6111 if (v == (void*)2) {
6112 status_unused(seq);
6113 return 0;
6114 }
6115
6116 if (mddev_lock(mddev) < 0)
6117 return -EINTR;
6118
6119 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6120 seq_printf(seq, "%s : %sactive", mdname(mddev),
6121 mddev->pers ? "" : "in");
6122 if (mddev->pers) {
6123 if (mddev->ro==1)
6124 seq_printf(seq, " (read-only)");
6125 if (mddev->ro==2)
6126 seq_printf(seq, " (auto-read-only)");
6127 seq_printf(seq, " %s", mddev->pers->name);
6128 }
6129
6130 sectors = 0;
6131 list_for_each_entry(rdev, &mddev->disks, same_set) {
6132 char b[BDEVNAME_SIZE];
6133 seq_printf(seq, " %s[%d]",
6134 bdevname(rdev->bdev,b), rdev->desc_nr);
6135 if (test_bit(WriteMostly, &rdev->flags))
6136 seq_printf(seq, "(W)");
6137 if (test_bit(Faulty, &rdev->flags)) {
6138 seq_printf(seq, "(F)");
6139 continue;
6140 } else if (rdev->raid_disk < 0)
6141 seq_printf(seq, "(S)"); /* spare */
6142 sectors += rdev->sectors;
6143 }
6144
6145 if (!list_empty(&mddev->disks)) {
6146 if (mddev->pers)
6147 seq_printf(seq, "\n %llu blocks",
6148 (unsigned long long)
6149 mddev->array_sectors / 2);
6150 else
6151 seq_printf(seq, "\n %llu blocks",
6152 (unsigned long long)sectors / 2);
6153 }
6154 if (mddev->persistent) {
6155 if (mddev->major_version != 0 ||
6156 mddev->minor_version != 90) {
6157 seq_printf(seq," super %d.%d",
6158 mddev->major_version,
6159 mddev->minor_version);
6160 }
6161 } else if (mddev->external)
6162 seq_printf(seq, " super external:%s",
6163 mddev->metadata_type);
6164 else
6165 seq_printf(seq, " super non-persistent");
6166
6167 if (mddev->pers) {
6168 mddev->pers->status(seq, mddev);
6169 seq_printf(seq, "\n ");
6170 if (mddev->pers->sync_request) {
6171 if (mddev->curr_resync > 2) {
6172 status_resync(seq, mddev);
6173 seq_printf(seq, "\n ");
6174 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6175 seq_printf(seq, "\tresync=DELAYED\n ");
6176 else if (mddev->recovery_cp < MaxSector)
6177 seq_printf(seq, "\tresync=PENDING\n ");
6178 }
6179 } else
6180 seq_printf(seq, "\n ");
6181
6182 if ((bitmap = mddev->bitmap)) {
6183 unsigned long chunk_kb;
6184 unsigned long flags;
6185 spin_lock_irqsave(&bitmap->lock, flags);
6186 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6187 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6188 "%lu%s chunk",
6189 bitmap->pages - bitmap->missing_pages,
6190 bitmap->pages,
6191 (bitmap->pages - bitmap->missing_pages)
6192 << (PAGE_SHIFT - 10),
6193 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6194 chunk_kb ? "KB" : "B");
6195 if (bitmap->file) {
6196 seq_printf(seq, ", file: ");
6197 seq_path(seq, &bitmap->file->f_path, " \t\n");
6198 }
6199
6200 seq_printf(seq, "\n");
6201 spin_unlock_irqrestore(&bitmap->lock, flags);
6202 }
6203
6204 seq_printf(seq, "\n");
6205 }
6206 mddev_unlock(mddev);
6207
6208 return 0;
6209 }
6210
6211 static const struct seq_operations md_seq_ops = {
6212 .start = md_seq_start,
6213 .next = md_seq_next,
6214 .stop = md_seq_stop,
6215 .show = md_seq_show,
6216 };
6217
6218 static int md_seq_open(struct inode *inode, struct file *file)
6219 {
6220 int error;
6221 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6222 if (mi == NULL)
6223 return -ENOMEM;
6224
6225 error = seq_open(file, &md_seq_ops);
6226 if (error)
6227 kfree(mi);
6228 else {
6229 struct seq_file *p = file->private_data;
6230 p->private = mi;
6231 mi->event = atomic_read(&md_event_count);
6232 }
6233 return error;
6234 }
6235
6236 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6237 {
6238 struct seq_file *m = filp->private_data;
6239 struct mdstat_info *mi = m->private;
6240 int mask;
6241
6242 poll_wait(filp, &md_event_waiters, wait);
6243
6244 /* always allow read */
6245 mask = POLLIN | POLLRDNORM;
6246
6247 if (mi->event != atomic_read(&md_event_count))
6248 mask |= POLLERR | POLLPRI;
6249 return mask;
6250 }
6251
6252 static const struct file_operations md_seq_fops = {
6253 .owner = THIS_MODULE,
6254 .open = md_seq_open,
6255 .read = seq_read,
6256 .llseek = seq_lseek,
6257 .release = seq_release_private,
6258 .poll = mdstat_poll,
6259 };
6260
6261 int register_md_personality(struct mdk_personality *p)
6262 {
6263 spin_lock(&pers_lock);
6264 list_add_tail(&p->list, &pers_list);
6265 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6266 spin_unlock(&pers_lock);
6267 return 0;
6268 }
6269
6270 int unregister_md_personality(struct mdk_personality *p)
6271 {
6272 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6273 spin_lock(&pers_lock);
6274 list_del_init(&p->list);
6275 spin_unlock(&pers_lock);
6276 return 0;
6277 }
6278
6279 static int is_mddev_idle(mddev_t *mddev, int init)
6280 {
6281 mdk_rdev_t * rdev;
6282 int idle;
6283 int curr_events;
6284
6285 idle = 1;
6286 rcu_read_lock();
6287 rdev_for_each_rcu(rdev, mddev) {
6288 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6289 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6290 (int)part_stat_read(&disk->part0, sectors[1]) -
6291 atomic_read(&disk->sync_io);
6292 /* sync IO will cause sync_io to increase before the disk_stats
6293 * as sync_io is counted when a request starts, and
6294 * disk_stats is counted when it completes.
6295 * So resync activity will cause curr_events to be smaller than
6296 * when there was no such activity.
6297 * non-sync IO will cause disk_stat to increase without
6298 * increasing sync_io so curr_events will (eventually)
6299 * be larger than it was before. Once it becomes
6300 * substantially larger, the test below will cause
6301 * the array to appear non-idle, and resync will slow
6302 * down.
6303 * If there is a lot of outstanding resync activity when
6304 * we set last_event to curr_events, then all that activity
6305 * completing might cause the array to appear non-idle
6306 * and resync will be slowed down even though there might
6307 * not have been non-resync activity. This will only
6308 * happen once though. 'last_events' will soon reflect
6309 * the state where there is little or no outstanding
6310 * resync requests, and further resync activity will
6311 * always make curr_events less than last_events.
6312 *
6313 */
6314 if (init || curr_events - rdev->last_events > 64) {
6315 rdev->last_events = curr_events;
6316 idle = 0;
6317 }
6318 }
6319 rcu_read_unlock();
6320 return idle;
6321 }
6322
6323 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6324 {
6325 /* another "blocks" (512byte) blocks have been synced */
6326 atomic_sub(blocks, &mddev->recovery_active);
6327 wake_up(&mddev->recovery_wait);
6328 if (!ok) {
6329 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6330 md_wakeup_thread(mddev->thread);
6331 // stop recovery, signal do_sync ....
6332 }
6333 }
6334
6335
6336 /* md_write_start(mddev, bi)
6337 * If we need to update some array metadata (e.g. 'active' flag
6338 * in superblock) before writing, schedule a superblock update
6339 * and wait for it to complete.
6340 */
6341 void md_write_start(mddev_t *mddev, struct bio *bi)
6342 {
6343 int did_change = 0;
6344 if (bio_data_dir(bi) != WRITE)
6345 return;
6346
6347 BUG_ON(mddev->ro == 1);
6348 if (mddev->ro == 2) {
6349 /* need to switch to read/write */
6350 mddev->ro = 0;
6351 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6352 md_wakeup_thread(mddev->thread);
6353 md_wakeup_thread(mddev->sync_thread);
6354 did_change = 1;
6355 }
6356 atomic_inc(&mddev->writes_pending);
6357 if (mddev->safemode == 1)
6358 mddev->safemode = 0;
6359 if (mddev->in_sync) {
6360 spin_lock_irq(&mddev->write_lock);
6361 if (mddev->in_sync) {
6362 mddev->in_sync = 0;
6363 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6364 md_wakeup_thread(mddev->thread);
6365 did_change = 1;
6366 }
6367 spin_unlock_irq(&mddev->write_lock);
6368 }
6369 if (did_change)
6370 sysfs_notify_dirent(mddev->sysfs_state);
6371 wait_event(mddev->sb_wait,
6372 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6373 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6374 }
6375
6376 void md_write_end(mddev_t *mddev)
6377 {
6378 if (atomic_dec_and_test(&mddev->writes_pending)) {
6379 if (mddev->safemode == 2)
6380 md_wakeup_thread(mddev->thread);
6381 else if (mddev->safemode_delay)
6382 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6383 }
6384 }
6385
6386 /* md_allow_write(mddev)
6387 * Calling this ensures that the array is marked 'active' so that writes
6388 * may proceed without blocking. It is important to call this before
6389 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6390 * Must be called with mddev_lock held.
6391 *
6392 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6393 * is dropped, so return -EAGAIN after notifying userspace.
6394 */
6395 int md_allow_write(mddev_t *mddev)
6396 {
6397 if (!mddev->pers)
6398 return 0;
6399 if (mddev->ro)
6400 return 0;
6401 if (!mddev->pers->sync_request)
6402 return 0;
6403
6404 spin_lock_irq(&mddev->write_lock);
6405 if (mddev->in_sync) {
6406 mddev->in_sync = 0;
6407 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6408 if (mddev->safemode_delay &&
6409 mddev->safemode == 0)
6410 mddev->safemode = 1;
6411 spin_unlock_irq(&mddev->write_lock);
6412 md_update_sb(mddev, 0);
6413 sysfs_notify_dirent(mddev->sysfs_state);
6414 } else
6415 spin_unlock_irq(&mddev->write_lock);
6416
6417 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6418 return -EAGAIN;
6419 else
6420 return 0;
6421 }
6422 EXPORT_SYMBOL_GPL(md_allow_write);
6423
6424 #define SYNC_MARKS 10
6425 #define SYNC_MARK_STEP (3*HZ)
6426 void md_do_sync(mddev_t *mddev)
6427 {
6428 mddev_t *mddev2;
6429 unsigned int currspeed = 0,
6430 window;
6431 sector_t max_sectors,j, io_sectors;
6432 unsigned long mark[SYNC_MARKS];
6433 sector_t mark_cnt[SYNC_MARKS];
6434 int last_mark,m;
6435 struct list_head *tmp;
6436 sector_t last_check;
6437 int skipped = 0;
6438 mdk_rdev_t *rdev;
6439 char *desc;
6440
6441 /* just incase thread restarts... */
6442 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6443 return;
6444 if (mddev->ro) /* never try to sync a read-only array */
6445 return;
6446
6447 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6448 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6449 desc = "data-check";
6450 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6451 desc = "requested-resync";
6452 else
6453 desc = "resync";
6454 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6455 desc = "reshape";
6456 else
6457 desc = "recovery";
6458
6459 /* we overload curr_resync somewhat here.
6460 * 0 == not engaged in resync at all
6461 * 2 == checking that there is no conflict with another sync
6462 * 1 == like 2, but have yielded to allow conflicting resync to
6463 * commense
6464 * other == active in resync - this many blocks
6465 *
6466 * Before starting a resync we must have set curr_resync to
6467 * 2, and then checked that every "conflicting" array has curr_resync
6468 * less than ours. When we find one that is the same or higher
6469 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6470 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6471 * This will mean we have to start checking from the beginning again.
6472 *
6473 */
6474
6475 do {
6476 mddev->curr_resync = 2;
6477
6478 try_again:
6479 if (kthread_should_stop())
6480 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6481
6482 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6483 goto skip;
6484 for_each_mddev(mddev2, tmp) {
6485 if (mddev2 == mddev)
6486 continue;
6487 if (!mddev->parallel_resync
6488 && mddev2->curr_resync
6489 && match_mddev_units(mddev, mddev2)) {
6490 DEFINE_WAIT(wq);
6491 if (mddev < mddev2 && mddev->curr_resync == 2) {
6492 /* arbitrarily yield */
6493 mddev->curr_resync = 1;
6494 wake_up(&resync_wait);
6495 }
6496 if (mddev > mddev2 && mddev->curr_resync == 1)
6497 /* no need to wait here, we can wait the next
6498 * time 'round when curr_resync == 2
6499 */
6500 continue;
6501 /* We need to wait 'interruptible' so as not to
6502 * contribute to the load average, and not to
6503 * be caught by 'softlockup'
6504 */
6505 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6506 if (!kthread_should_stop() &&
6507 mddev2->curr_resync >= mddev->curr_resync) {
6508 printk(KERN_INFO "md: delaying %s of %s"
6509 " until %s has finished (they"
6510 " share one or more physical units)\n",
6511 desc, mdname(mddev), mdname(mddev2));
6512 mddev_put(mddev2);
6513 if (signal_pending(current))
6514 flush_signals(current);
6515 schedule();
6516 finish_wait(&resync_wait, &wq);
6517 goto try_again;
6518 }
6519 finish_wait(&resync_wait, &wq);
6520 }
6521 }
6522 } while (mddev->curr_resync < 2);
6523
6524 j = 0;
6525 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6526 /* resync follows the size requested by the personality,
6527 * which defaults to physical size, but can be virtual size
6528 */
6529 max_sectors = mddev->resync_max_sectors;
6530 mddev->resync_mismatches = 0;
6531 /* we don't use the checkpoint if there's a bitmap */
6532 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6533 j = mddev->resync_min;
6534 else if (!mddev->bitmap)
6535 j = mddev->recovery_cp;
6536
6537 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6538 max_sectors = mddev->dev_sectors;
6539 else {
6540 /* recovery follows the physical size of devices */
6541 max_sectors = mddev->dev_sectors;
6542 j = MaxSector;
6543 rcu_read_lock();
6544 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6545 if (rdev->raid_disk >= 0 &&
6546 !test_bit(Faulty, &rdev->flags) &&
6547 !test_bit(In_sync, &rdev->flags) &&
6548 rdev->recovery_offset < j)
6549 j = rdev->recovery_offset;
6550 rcu_read_unlock();
6551 }
6552
6553 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6554 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6555 " %d KB/sec/disk.\n", speed_min(mddev));
6556 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6557 "(but not more than %d KB/sec) for %s.\n",
6558 speed_max(mddev), desc);
6559
6560 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6561
6562 io_sectors = 0;
6563 for (m = 0; m < SYNC_MARKS; m++) {
6564 mark[m] = jiffies;
6565 mark_cnt[m] = io_sectors;
6566 }
6567 last_mark = 0;
6568 mddev->resync_mark = mark[last_mark];
6569 mddev->resync_mark_cnt = mark_cnt[last_mark];
6570
6571 /*
6572 * Tune reconstruction:
6573 */
6574 window = 32*(PAGE_SIZE/512);
6575 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6576 window/2,(unsigned long long) max_sectors/2);
6577
6578 atomic_set(&mddev->recovery_active, 0);
6579 last_check = 0;
6580
6581 if (j>2) {
6582 printk(KERN_INFO
6583 "md: resuming %s of %s from checkpoint.\n",
6584 desc, mdname(mddev));
6585 mddev->curr_resync = j;
6586 }
6587 mddev->curr_resync_completed = mddev->curr_resync;
6588
6589 while (j < max_sectors) {
6590 sector_t sectors;
6591
6592 skipped = 0;
6593
6594 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6595 ((mddev->curr_resync > mddev->curr_resync_completed &&
6596 (mddev->curr_resync - mddev->curr_resync_completed)
6597 > (max_sectors >> 4)) ||
6598 (j - mddev->curr_resync_completed)*2
6599 >= mddev->resync_max - mddev->curr_resync_completed
6600 )) {
6601 /* time to update curr_resync_completed */
6602 blk_unplug(mddev->queue);
6603 wait_event(mddev->recovery_wait,
6604 atomic_read(&mddev->recovery_active) == 0);
6605 mddev->curr_resync_completed =
6606 mddev->curr_resync;
6607 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6608 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6609 }
6610
6611 while (j >= mddev->resync_max && !kthread_should_stop()) {
6612 /* As this condition is controlled by user-space,
6613 * we can block indefinitely, so use '_interruptible'
6614 * to avoid triggering warnings.
6615 */
6616 flush_signals(current); /* just in case */
6617 wait_event_interruptible(mddev->recovery_wait,
6618 mddev->resync_max > j
6619 || kthread_should_stop());
6620 }
6621
6622 if (kthread_should_stop())
6623 goto interrupted;
6624
6625 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6626 currspeed < speed_min(mddev));
6627 if (sectors == 0) {
6628 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6629 goto out;
6630 }
6631
6632 if (!skipped) { /* actual IO requested */
6633 io_sectors += sectors;
6634 atomic_add(sectors, &mddev->recovery_active);
6635 }
6636
6637 j += sectors;
6638 if (j>1) mddev->curr_resync = j;
6639 mddev->curr_mark_cnt = io_sectors;
6640 if (last_check == 0)
6641 /* this is the earliers that rebuilt will be
6642 * visible in /proc/mdstat
6643 */
6644 md_new_event(mddev);
6645
6646 if (last_check + window > io_sectors || j == max_sectors)
6647 continue;
6648
6649 last_check = io_sectors;
6650
6651 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6652 break;
6653
6654 repeat:
6655 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6656 /* step marks */
6657 int next = (last_mark+1) % SYNC_MARKS;
6658
6659 mddev->resync_mark = mark[next];
6660 mddev->resync_mark_cnt = mark_cnt[next];
6661 mark[next] = jiffies;
6662 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6663 last_mark = next;
6664 }
6665
6666
6667 if (kthread_should_stop())
6668 goto interrupted;
6669
6670
6671 /*
6672 * this loop exits only if either when we are slower than
6673 * the 'hard' speed limit, or the system was IO-idle for
6674 * a jiffy.
6675 * the system might be non-idle CPU-wise, but we only care
6676 * about not overloading the IO subsystem. (things like an
6677 * e2fsck being done on the RAID array should execute fast)
6678 */
6679 blk_unplug(mddev->queue);
6680 cond_resched();
6681
6682 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6683 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6684
6685 if (currspeed > speed_min(mddev)) {
6686 if ((currspeed > speed_max(mddev)) ||
6687 !is_mddev_idle(mddev, 0)) {
6688 msleep(500);
6689 goto repeat;
6690 }
6691 }
6692 }
6693 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6694 /*
6695 * this also signals 'finished resyncing' to md_stop
6696 */
6697 out:
6698 blk_unplug(mddev->queue);
6699
6700 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6701
6702 /* tell personality that we are finished */
6703 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6704
6705 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6706 mddev->curr_resync > 2) {
6707 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6708 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6709 if (mddev->curr_resync >= mddev->recovery_cp) {
6710 printk(KERN_INFO
6711 "md: checkpointing %s of %s.\n",
6712 desc, mdname(mddev));
6713 mddev->recovery_cp = mddev->curr_resync;
6714 }
6715 } else
6716 mddev->recovery_cp = MaxSector;
6717 } else {
6718 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6719 mddev->curr_resync = MaxSector;
6720 rcu_read_lock();
6721 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6722 if (rdev->raid_disk >= 0 &&
6723 !test_bit(Faulty, &rdev->flags) &&
6724 !test_bit(In_sync, &rdev->flags) &&
6725 rdev->recovery_offset < mddev->curr_resync)
6726 rdev->recovery_offset = mddev->curr_resync;
6727 rcu_read_unlock();
6728 }
6729 }
6730 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6731
6732 skip:
6733 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6734 /* We completed so min/max setting can be forgotten if used. */
6735 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6736 mddev->resync_min = 0;
6737 mddev->resync_max = MaxSector;
6738 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6739 mddev->resync_min = mddev->curr_resync_completed;
6740 mddev->curr_resync = 0;
6741 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6742 mddev->curr_resync_completed = 0;
6743 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6744 wake_up(&resync_wait);
6745 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6746 md_wakeup_thread(mddev->thread);
6747 return;
6748
6749 interrupted:
6750 /*
6751 * got a signal, exit.
6752 */
6753 printk(KERN_INFO
6754 "md: md_do_sync() got signal ... exiting\n");
6755 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6756 goto out;
6757
6758 }
6759 EXPORT_SYMBOL_GPL(md_do_sync);
6760
6761
6762 static int remove_and_add_spares(mddev_t *mddev)
6763 {
6764 mdk_rdev_t *rdev;
6765 int spares = 0;
6766
6767 mddev->curr_resync_completed = 0;
6768
6769 list_for_each_entry(rdev, &mddev->disks, same_set)
6770 if (rdev->raid_disk >= 0 &&
6771 !test_bit(Blocked, &rdev->flags) &&
6772 (test_bit(Faulty, &rdev->flags) ||
6773 ! test_bit(In_sync, &rdev->flags)) &&
6774 atomic_read(&rdev->nr_pending)==0) {
6775 if (mddev->pers->hot_remove_disk(
6776 mddev, rdev->raid_disk)==0) {
6777 char nm[20];
6778 sprintf(nm,"rd%d", rdev->raid_disk);
6779 sysfs_remove_link(&mddev->kobj, nm);
6780 rdev->raid_disk = -1;
6781 }
6782 }
6783
6784 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6785 list_for_each_entry(rdev, &mddev->disks, same_set) {
6786 if (rdev->raid_disk >= 0 &&
6787 !test_bit(In_sync, &rdev->flags) &&
6788 !test_bit(Blocked, &rdev->flags))
6789 spares++;
6790 if (rdev->raid_disk < 0
6791 && !test_bit(Faulty, &rdev->flags)) {
6792 rdev->recovery_offset = 0;
6793 if (mddev->pers->
6794 hot_add_disk(mddev, rdev) == 0) {
6795 char nm[20];
6796 sprintf(nm, "rd%d", rdev->raid_disk);
6797 if (sysfs_create_link(&mddev->kobj,
6798 &rdev->kobj, nm))
6799 printk(KERN_WARNING
6800 "md: cannot register "
6801 "%s for %s\n",
6802 nm, mdname(mddev));
6803 spares++;
6804 md_new_event(mddev);
6805 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6806 } else
6807 break;
6808 }
6809 }
6810 }
6811 return spares;
6812 }
6813 /*
6814 * This routine is regularly called by all per-raid-array threads to
6815 * deal with generic issues like resync and super-block update.
6816 * Raid personalities that don't have a thread (linear/raid0) do not
6817 * need this as they never do any recovery or update the superblock.
6818 *
6819 * It does not do any resync itself, but rather "forks" off other threads
6820 * to do that as needed.
6821 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6822 * "->recovery" and create a thread at ->sync_thread.
6823 * When the thread finishes it sets MD_RECOVERY_DONE
6824 * and wakeups up this thread which will reap the thread and finish up.
6825 * This thread also removes any faulty devices (with nr_pending == 0).
6826 *
6827 * The overall approach is:
6828 * 1/ if the superblock needs updating, update it.
6829 * 2/ If a recovery thread is running, don't do anything else.
6830 * 3/ If recovery has finished, clean up, possibly marking spares active.
6831 * 4/ If there are any faulty devices, remove them.
6832 * 5/ If array is degraded, try to add spares devices
6833 * 6/ If array has spares or is not in-sync, start a resync thread.
6834 */
6835 void md_check_recovery(mddev_t *mddev)
6836 {
6837 mdk_rdev_t *rdev;
6838
6839
6840 if (mddev->bitmap)
6841 bitmap_daemon_work(mddev);
6842
6843 if (mddev->ro)
6844 return;
6845
6846 if (signal_pending(current)) {
6847 if (mddev->pers->sync_request && !mddev->external) {
6848 printk(KERN_INFO "md: %s in immediate safe mode\n",
6849 mdname(mddev));
6850 mddev->safemode = 2;
6851 }
6852 flush_signals(current);
6853 }
6854
6855 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6856 return;
6857 if ( ! (
6858 (mddev->flags && !mddev->external) ||
6859 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6860 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6861 (mddev->external == 0 && mddev->safemode == 1) ||
6862 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6863 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6864 ))
6865 return;
6866
6867 if (mddev_trylock(mddev)) {
6868 int spares = 0;
6869
6870 if (mddev->ro) {
6871 /* Only thing we do on a ro array is remove
6872 * failed devices.
6873 */
6874 remove_and_add_spares(mddev);
6875 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6876 goto unlock;
6877 }
6878
6879 if (!mddev->external) {
6880 int did_change = 0;
6881 spin_lock_irq(&mddev->write_lock);
6882 if (mddev->safemode &&
6883 !atomic_read(&mddev->writes_pending) &&
6884 !mddev->in_sync &&
6885 mddev->recovery_cp == MaxSector) {
6886 mddev->in_sync = 1;
6887 did_change = 1;
6888 if (mddev->persistent)
6889 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6890 }
6891 if (mddev->safemode == 1)
6892 mddev->safemode = 0;
6893 spin_unlock_irq(&mddev->write_lock);
6894 if (did_change)
6895 sysfs_notify_dirent(mddev->sysfs_state);
6896 }
6897
6898 if (mddev->flags)
6899 md_update_sb(mddev, 0);
6900
6901 list_for_each_entry(rdev, &mddev->disks, same_set)
6902 if (test_and_clear_bit(StateChanged, &rdev->flags))
6903 sysfs_notify_dirent(rdev->sysfs_state);
6904
6905
6906 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6907 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6908 /* resync/recovery still happening */
6909 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6910 goto unlock;
6911 }
6912 if (mddev->sync_thread) {
6913 /* resync has finished, collect result */
6914 md_unregister_thread(mddev->sync_thread);
6915 mddev->sync_thread = NULL;
6916 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6917 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6918 /* success...*/
6919 /* activate any spares */
6920 if (mddev->pers->spare_active(mddev))
6921 sysfs_notify(&mddev->kobj, NULL,
6922 "degraded");
6923 }
6924 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6925 mddev->pers->finish_reshape)
6926 mddev->pers->finish_reshape(mddev);
6927 md_update_sb(mddev, 1);
6928
6929 /* if array is no-longer degraded, then any saved_raid_disk
6930 * information must be scrapped
6931 */
6932 if (!mddev->degraded)
6933 list_for_each_entry(rdev, &mddev->disks, same_set)
6934 rdev->saved_raid_disk = -1;
6935
6936 mddev->recovery = 0;
6937 /* flag recovery needed just to double check */
6938 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6939 sysfs_notify_dirent(mddev->sysfs_action);
6940 md_new_event(mddev);
6941 goto unlock;
6942 }
6943 /* Set RUNNING before clearing NEEDED to avoid
6944 * any transients in the value of "sync_action".
6945 */
6946 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6947 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6948 /* Clear some bits that don't mean anything, but
6949 * might be left set
6950 */
6951 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6952 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6953
6954 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6955 goto unlock;
6956 /* no recovery is running.
6957 * remove any failed drives, then
6958 * add spares if possible.
6959 * Spare are also removed and re-added, to allow
6960 * the personality to fail the re-add.
6961 */
6962
6963 if (mddev->reshape_position != MaxSector) {
6964 if (mddev->pers->check_reshape == NULL ||
6965 mddev->pers->check_reshape(mddev) != 0)
6966 /* Cannot proceed */
6967 goto unlock;
6968 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6969 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6970 } else if ((spares = remove_and_add_spares(mddev))) {
6971 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6972 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6973 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6974 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6975 } else if (mddev->recovery_cp < MaxSector) {
6976 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6977 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6978 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6979 /* nothing to be done ... */
6980 goto unlock;
6981
6982 if (mddev->pers->sync_request) {
6983 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6984 /* We are adding a device or devices to an array
6985 * which has the bitmap stored on all devices.
6986 * So make sure all bitmap pages get written
6987 */
6988 bitmap_write_all(mddev->bitmap);
6989 }
6990 mddev->sync_thread = md_register_thread(md_do_sync,
6991 mddev,
6992 "resync");
6993 if (!mddev->sync_thread) {
6994 printk(KERN_ERR "%s: could not start resync"
6995 " thread...\n",
6996 mdname(mddev));
6997 /* leave the spares where they are, it shouldn't hurt */
6998 mddev->recovery = 0;
6999 } else
7000 md_wakeup_thread(mddev->sync_thread);
7001 sysfs_notify_dirent(mddev->sysfs_action);
7002 md_new_event(mddev);
7003 }
7004 unlock:
7005 if (!mddev->sync_thread) {
7006 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7007 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7008 &mddev->recovery))
7009 if (mddev->sysfs_action)
7010 sysfs_notify_dirent(mddev->sysfs_action);
7011 }
7012 mddev_unlock(mddev);
7013 }
7014 }
7015
7016 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7017 {
7018 sysfs_notify_dirent(rdev->sysfs_state);
7019 wait_event_timeout(rdev->blocked_wait,
7020 !test_bit(Blocked, &rdev->flags),
7021 msecs_to_jiffies(5000));
7022 rdev_dec_pending(rdev, mddev);
7023 }
7024 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7025
7026 static int md_notify_reboot(struct notifier_block *this,
7027 unsigned long code, void *x)
7028 {
7029 struct list_head *tmp;
7030 mddev_t *mddev;
7031
7032 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7033
7034 printk(KERN_INFO "md: stopping all md devices.\n");
7035
7036 for_each_mddev(mddev, tmp)
7037 if (mddev_trylock(mddev)) {
7038 /* Force a switch to readonly even array
7039 * appears to still be in use. Hence
7040 * the '100'.
7041 */
7042 do_md_stop(mddev, 1, 100);
7043 mddev_unlock(mddev);
7044 }
7045 /*
7046 * certain more exotic SCSI devices are known to be
7047 * volatile wrt too early system reboots. While the
7048 * right place to handle this issue is the given
7049 * driver, we do want to have a safe RAID driver ...
7050 */
7051 mdelay(1000*1);
7052 }
7053 return NOTIFY_DONE;
7054 }
7055
7056 static struct notifier_block md_notifier = {
7057 .notifier_call = md_notify_reboot,
7058 .next = NULL,
7059 .priority = INT_MAX, /* before any real devices */
7060 };
7061
7062 static void md_geninit(void)
7063 {
7064 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7065
7066 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7067 }
7068
7069 static int __init md_init(void)
7070 {
7071 if (register_blkdev(MD_MAJOR, "md"))
7072 return -1;
7073 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7074 unregister_blkdev(MD_MAJOR, "md");
7075 return -1;
7076 }
7077 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7078 md_probe, NULL, NULL);
7079 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7080 md_probe, NULL, NULL);
7081
7082 register_reboot_notifier(&md_notifier);
7083 raid_table_header = register_sysctl_table(raid_root_table);
7084
7085 md_geninit();
7086 return 0;
7087 }
7088
7089
7090 #ifndef MODULE
7091
7092 /*
7093 * Searches all registered partitions for autorun RAID arrays
7094 * at boot time.
7095 */
7096
7097 static LIST_HEAD(all_detected_devices);
7098 struct detected_devices_node {
7099 struct list_head list;
7100 dev_t dev;
7101 };
7102
7103 void md_autodetect_dev(dev_t dev)
7104 {
7105 struct detected_devices_node *node_detected_dev;
7106
7107 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7108 if (node_detected_dev) {
7109 node_detected_dev->dev = dev;
7110 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7111 } else {
7112 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7113 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7114 }
7115 }
7116
7117
7118 static void autostart_arrays(int part)
7119 {
7120 mdk_rdev_t *rdev;
7121 struct detected_devices_node *node_detected_dev;
7122 dev_t dev;
7123 int i_scanned, i_passed;
7124
7125 i_scanned = 0;
7126 i_passed = 0;
7127
7128 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7129
7130 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7131 i_scanned++;
7132 node_detected_dev = list_entry(all_detected_devices.next,
7133 struct detected_devices_node, list);
7134 list_del(&node_detected_dev->list);
7135 dev = node_detected_dev->dev;
7136 kfree(node_detected_dev);
7137 rdev = md_import_device(dev,0, 90);
7138 if (IS_ERR(rdev))
7139 continue;
7140
7141 if (test_bit(Faulty, &rdev->flags)) {
7142 MD_BUG();
7143 continue;
7144 }
7145 set_bit(AutoDetected, &rdev->flags);
7146 list_add(&rdev->same_set, &pending_raid_disks);
7147 i_passed++;
7148 }
7149
7150 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7151 i_scanned, i_passed);
7152
7153 autorun_devices(part);
7154 }
7155
7156 #endif /* !MODULE */
7157
7158 static __exit void md_exit(void)
7159 {
7160 mddev_t *mddev;
7161 struct list_head *tmp;
7162
7163 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7164 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7165
7166 unregister_blkdev(MD_MAJOR,"md");
7167 unregister_blkdev(mdp_major, "mdp");
7168 unregister_reboot_notifier(&md_notifier);
7169 unregister_sysctl_table(raid_table_header);
7170 remove_proc_entry("mdstat", NULL);
7171 for_each_mddev(mddev, tmp) {
7172 export_array(mddev);
7173 mddev->hold_active = 0;
7174 }
7175 }
7176
7177 subsys_initcall(md_init);
7178 module_exit(md_exit)
7179
7180 static int get_ro(char *buffer, struct kernel_param *kp)
7181 {
7182 return sprintf(buffer, "%d", start_readonly);
7183 }
7184 static int set_ro(const char *val, struct kernel_param *kp)
7185 {
7186 char *e;
7187 int num = simple_strtoul(val, &e, 10);
7188 if (*val && (*e == '\0' || *e == '\n')) {
7189 start_readonly = num;
7190 return 0;
7191 }
7192 return -EINVAL;
7193 }
7194
7195 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7196 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7197
7198 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7199
7200 EXPORT_SYMBOL(register_md_personality);
7201 EXPORT_SYMBOL(unregister_md_personality);
7202 EXPORT_SYMBOL(md_error);
7203 EXPORT_SYMBOL(md_done_sync);
7204 EXPORT_SYMBOL(md_write_start);
7205 EXPORT_SYMBOL(md_write_end);
7206 EXPORT_SYMBOL(md_register_thread);
7207 EXPORT_SYMBOL(md_unregister_thread);
7208 EXPORT_SYMBOL(md_wakeup_thread);
7209 EXPORT_SYMBOL(md_check_recovery);
7210 MODULE_LICENSE("GPL");
7211 MODULE_DESCRIPTION("MD RAID framework");
7212 MODULE_ALIAS("md");
7213 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel-based virtual machine