?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
md: export various start/stop interfaces
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blob012482a900fc1b6be518e80f8e5c47be45667517
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 const int rw = bio_data_dir(bio);
219 mddev_t *mddev = q->queuedata;
220 int rv;
221 int cpu;
222
223 if (mddev == NULL || mddev->pers == NULL) {
224 bio_io_error(bio);
225 return 0;
226 }
227 rcu_read_lock();
228 if (mddev->suspended || mddev->barrier) {
229 DEFINE_WAIT(__wait);
230 for (;;) {
231 prepare_to_wait(&mddev->sb_wait, &__wait,
232 TASK_UNINTERRUPTIBLE);
233 if (!mddev->suspended && !mddev->barrier)
234 break;
235 rcu_read_unlock();
236 schedule();
237 rcu_read_lock();
238 }
239 finish_wait(&mddev->sb_wait, &__wait);
240 }
241 atomic_inc(&mddev->active_io);
242 rcu_read_unlock();
243
244 rv = mddev->pers->make_request(mddev, bio);
245
246 cpu = part_stat_lock();
247 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
248 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
249 bio_sectors(bio));
250 part_stat_unlock();
251
252 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
253 wake_up(&mddev->sb_wait);
254
255 return rv;
256 }
257
258 /* mddev_suspend makes sure no new requests are submitted
259 * to the device, and that any requests that have been submitted
260 * are completely handled.
261 * Once ->stop is called and completes, the module will be completely
262 * unused.
263 */
264 void mddev_suspend(mddev_t *mddev)
265 {
266 BUG_ON(mddev->suspended);
267 mddev->suspended = 1;
268 synchronize_rcu();
269 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
270 mddev->pers->quiesce(mddev, 1);
271 }
272 EXPORT_SYMBOL_GPL(mddev_suspend);
273
274 void mddev_resume(mddev_t *mddev)
275 {
276 mddev->suspended = 0;
277 wake_up(&mddev->sb_wait);
278 mddev->pers->quiesce(mddev, 0);
279 }
280 EXPORT_SYMBOL_GPL(mddev_resume);
281
282 int mddev_congested(mddev_t *mddev, int bits)
283 {
284 if (mddev->barrier)
285 return 1;
286 return mddev->suspended;
287 }
288 EXPORT_SYMBOL(mddev_congested);
289
290 /*
291 * Generic barrier handling for md
292 */
293
294 #define POST_REQUEST_BARRIER ((void*)1)
295
296 static void md_end_barrier(struct bio *bio, int err)
297 {
298 mdk_rdev_t *rdev = bio->bi_private;
299 mddev_t *mddev = rdev->mddev;
300 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
301 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
302
303 rdev_dec_pending(rdev, mddev);
304
305 if (atomic_dec_and_test(&mddev->flush_pending)) {
306 if (mddev->barrier == POST_REQUEST_BARRIER) {
307 /* This was a post-request barrier */
308 mddev->barrier = NULL;
309 wake_up(&mddev->sb_wait);
310 } else
311 /* The pre-request barrier has finished */
312 schedule_work(&mddev->barrier_work);
313 }
314 bio_put(bio);
315 }
316
317 static void submit_barriers(mddev_t *mddev)
318 {
319 mdk_rdev_t *rdev;
320
321 rcu_read_lock();
322 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
323 if (rdev->raid_disk >= 0 &&
324 !test_bit(Faulty, &rdev->flags)) {
325 /* Take two references, one is dropped
326 * when request finishes, one after
327 * we reclaim rcu_read_lock
328 */
329 struct bio *bi;
330 atomic_inc(&rdev->nr_pending);
331 atomic_inc(&rdev->nr_pending);
332 rcu_read_unlock();
333 bi = bio_alloc(GFP_KERNEL, 0);
334 bi->bi_end_io = md_end_barrier;
335 bi->bi_private = rdev;
336 bi->bi_bdev = rdev->bdev;
337 atomic_inc(&mddev->flush_pending);
338 submit_bio(WRITE_BARRIER, bi);
339 rcu_read_lock();
340 rdev_dec_pending(rdev, mddev);
341 }
342 rcu_read_unlock();
343 }
344
345 static void md_submit_barrier(struct work_struct *ws)
346 {
347 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
348 struct bio *bio = mddev->barrier;
349
350 atomic_set(&mddev->flush_pending, 1);
351
352 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
353 bio_endio(bio, -EOPNOTSUPP);
354 else if (bio->bi_size == 0)
355 /* an empty barrier - all done */
356 bio_endio(bio, 0);
357 else {
358 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
359 if (mddev->pers->make_request(mddev, bio))
360 generic_make_request(bio);
361 mddev->barrier = POST_REQUEST_BARRIER;
362 submit_barriers(mddev);
363 }
364 if (atomic_dec_and_test(&mddev->flush_pending)) {
365 mddev->barrier = NULL;
366 wake_up(&mddev->sb_wait);
367 }
368 }
369
370 void md_barrier_request(mddev_t *mddev, struct bio *bio)
371 {
372 spin_lock_irq(&mddev->write_lock);
373 wait_event_lock_irq(mddev->sb_wait,
374 !mddev->barrier,
375 mddev->write_lock, /*nothing*/);
376 mddev->barrier = bio;
377 spin_unlock_irq(&mddev->write_lock);
378
379 atomic_set(&mddev->flush_pending, 1);
380 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
381
382 submit_barriers(mddev);
383
384 if (atomic_dec_and_test(&mddev->flush_pending))
385 schedule_work(&mddev->barrier_work);
386 }
387 EXPORT_SYMBOL(md_barrier_request);
388
389 static inline mddev_t *mddev_get(mddev_t *mddev)
390 {
391 atomic_inc(&mddev->active);
392 return mddev;
393 }
394
395 static void mddev_delayed_delete(struct work_struct *ws);
396
397 static void mddev_put(mddev_t *mddev)
398 {
399 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
400 return;
401 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
402 mddev->ctime == 0 && !mddev->hold_active) {
403 /* Array is not configured at all, and not held active,
404 * so destroy it */
405 list_del(&mddev->all_mddevs);
406 if (mddev->gendisk) {
407 /* we did a probe so need to clean up.
408 * Call schedule_work inside the spinlock
409 * so that flush_scheduled_work() after
410 * mddev_find will succeed in waiting for the
411 * work to be done.
412 */
413 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
414 schedule_work(&mddev->del_work);
415 } else
416 kfree(mddev);
417 }
418 spin_unlock(&all_mddevs_lock);
419 }
420
421 void mddev_init(mddev_t *mddev)
422 {
423 mutex_init(&mddev->open_mutex);
424 mutex_init(&mddev->reconfig_mutex);
425 mutex_init(&mddev->bitmap_info.mutex);
426 INIT_LIST_HEAD(&mddev->disks);
427 INIT_LIST_HEAD(&mddev->all_mddevs);
428 init_timer(&mddev->safemode_timer);
429 atomic_set(&mddev->active, 1);
430 atomic_set(&mddev->openers, 0);
431 atomic_set(&mddev->active_io, 0);
432 spin_lock_init(&mddev->write_lock);
433 atomic_set(&mddev->flush_pending, 0);
434 init_waitqueue_head(&mddev->sb_wait);
435 init_waitqueue_head(&mddev->recovery_wait);
436 mddev->reshape_position = MaxSector;
437 mddev->resync_min = 0;
438 mddev->resync_max = MaxSector;
439 mddev->level = LEVEL_NONE;
440 }
441 EXPORT_SYMBOL_GPL(mddev_init);
442
443 static mddev_t * mddev_find(dev_t unit)
444 {
445 mddev_t *mddev, *new = NULL;
446
447 retry:
448 spin_lock(&all_mddevs_lock);
449
450 if (unit) {
451 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
452 if (mddev->unit == unit) {
453 mddev_get(mddev);
454 spin_unlock(&all_mddevs_lock);
455 kfree(new);
456 return mddev;
457 }
458
459 if (new) {
460 list_add(&new->all_mddevs, &all_mddevs);
461 spin_unlock(&all_mddevs_lock);
462 new->hold_active = UNTIL_IOCTL;
463 return new;
464 }
465 } else if (new) {
466 /* find an unused unit number */
467 static int next_minor = 512;
468 int start = next_minor;
469 int is_free = 0;
470 int dev = 0;
471 while (!is_free) {
472 dev = MKDEV(MD_MAJOR, next_minor);
473 next_minor++;
474 if (next_minor > MINORMASK)
475 next_minor = 0;
476 if (next_minor == start) {
477 /* Oh dear, all in use. */
478 spin_unlock(&all_mddevs_lock);
479 kfree(new);
480 return NULL;
481 }
482
483 is_free = 1;
484 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
485 if (mddev->unit == dev) {
486 is_free = 0;
487 break;
488 }
489 }
490 new->unit = dev;
491 new->md_minor = MINOR(dev);
492 new->hold_active = UNTIL_STOP;
493 list_add(&new->all_mddevs, &all_mddevs);
494 spin_unlock(&all_mddevs_lock);
495 return new;
496 }
497 spin_unlock(&all_mddevs_lock);
498
499 new = kzalloc(sizeof(*new), GFP_KERNEL);
500 if (!new)
501 return NULL;
502
503 new->unit = unit;
504 if (MAJOR(unit) == MD_MAJOR)
505 new->md_minor = MINOR(unit);
506 else
507 new->md_minor = MINOR(unit) >> MdpMinorShift;
508
509 mddev_init(new);
510
511 goto retry;
512 }
513
514 static inline int mddev_lock(mddev_t * mddev)
515 {
516 return mutex_lock_interruptible(&mddev->reconfig_mutex);
517 }
518
519 static inline int mddev_is_locked(mddev_t *mddev)
520 {
521 return mutex_is_locked(&mddev->reconfig_mutex);
522 }
523
524 static inline int mddev_trylock(mddev_t * mddev)
525 {
526 return mutex_trylock(&mddev->reconfig_mutex);
527 }
528
529 static struct attribute_group md_redundancy_group;
530
531 static void mddev_unlock(mddev_t * mddev)
532 {
533 if (mddev->to_remove) {
534 /* These cannot be removed under reconfig_mutex as
535 * an access to the files will try to take reconfig_mutex
536 * while holding the file unremovable, which leads to
537 * a deadlock.
538 * So hold open_mutex instead - we are allowed to take
539 * it while holding reconfig_mutex, and md_run can
540 * use it to wait for the remove to complete.
541 */
542 struct attribute_group *to_remove = mddev->to_remove;
543 mddev->to_remove = NULL;
544 mutex_lock(&mddev->open_mutex);
545 mutex_unlock(&mddev->reconfig_mutex);
546
547 if (mddev->kobj.sd) {
548 if (to_remove != &md_redundancy_group)
549 sysfs_remove_group(&mddev->kobj, to_remove);
550 if (mddev->pers == NULL ||
551 mddev->pers->sync_request == NULL) {
552 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
553 if (mddev->sysfs_action)
554 sysfs_put(mddev->sysfs_action);
555 mddev->sysfs_action = NULL;
556 }
557 }
558 mutex_unlock(&mddev->open_mutex);
559 } else
560 mutex_unlock(&mddev->reconfig_mutex);
561
562 md_wakeup_thread(mddev->thread);
563 }
564
565 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
566 {
567 mdk_rdev_t *rdev;
568
569 list_for_each_entry(rdev, &mddev->disks, same_set)
570 if (rdev->desc_nr == nr)
571 return rdev;
572
573 return NULL;
574 }
575
576 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
577 {
578 mdk_rdev_t *rdev;
579
580 list_for_each_entry(rdev, &mddev->disks, same_set)
581 if (rdev->bdev->bd_dev == dev)
582 return rdev;
583
584 return NULL;
585 }
586
587 static struct mdk_personality *find_pers(int level, char *clevel)
588 {
589 struct mdk_personality *pers;
590 list_for_each_entry(pers, &pers_list, list) {
591 if (level != LEVEL_NONE && pers->level == level)
592 return pers;
593 if (strcmp(pers->name, clevel)==0)
594 return pers;
595 }
596 return NULL;
597 }
598
599 /* return the offset of the super block in 512byte sectors */
600 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
601 {
602 sector_t num_sectors = bdev->bd_inode->i_size / 512;
603 return MD_NEW_SIZE_SECTORS(num_sectors);
604 }
605
606 static int alloc_disk_sb(mdk_rdev_t * rdev)
607 {
608 if (rdev->sb_page)
609 MD_BUG();
610
611 rdev->sb_page = alloc_page(GFP_KERNEL);
612 if (!rdev->sb_page) {
613 printk(KERN_ALERT "md: out of memory.\n");
614 return -ENOMEM;
615 }
616
617 return 0;
618 }
619
620 static void free_disk_sb(mdk_rdev_t * rdev)
621 {
622 if (rdev->sb_page) {
623 put_page(rdev->sb_page);
624 rdev->sb_loaded = 0;
625 rdev->sb_page = NULL;
626 rdev->sb_start = 0;
627 rdev->sectors = 0;
628 }
629 }
630
631
632 static void super_written(struct bio *bio, int error)
633 {
634 mdk_rdev_t *rdev = bio->bi_private;
635 mddev_t *mddev = rdev->mddev;
636
637 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
638 printk("md: super_written gets error=%d, uptodate=%d\n",
639 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
640 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
641 md_error(mddev, rdev);
642 }
643
644 if (atomic_dec_and_test(&mddev->pending_writes))
645 wake_up(&mddev->sb_wait);
646 bio_put(bio);
647 }
648
649 static void super_written_barrier(struct bio *bio, int error)
650 {
651 struct bio *bio2 = bio->bi_private;
652 mdk_rdev_t *rdev = bio2->bi_private;
653 mddev_t *mddev = rdev->mddev;
654
655 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
656 error == -EOPNOTSUPP) {
657 unsigned long flags;
658 /* barriers don't appear to be supported :-( */
659 set_bit(BarriersNotsupp, &rdev->flags);
660 mddev->barriers_work = 0;
661 spin_lock_irqsave(&mddev->write_lock, flags);
662 bio2->bi_next = mddev->biolist;
663 mddev->biolist = bio2;
664 spin_unlock_irqrestore(&mddev->write_lock, flags);
665 wake_up(&mddev->sb_wait);
666 bio_put(bio);
667 } else {
668 bio_put(bio2);
669 bio->bi_private = rdev;
670 super_written(bio, error);
671 }
672 }
673
674 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
675 sector_t sector, int size, struct page *page)
676 {
677 /* write first size bytes of page to sector of rdev
678 * Increment mddev->pending_writes before returning
679 * and decrement it on completion, waking up sb_wait
680 * if zero is reached.
681 * If an error occurred, call md_error
682 *
683 * As we might need to resubmit the request if BIO_RW_BARRIER
684 * causes ENOTSUPP, we allocate a spare bio...
685 */
686 struct bio *bio = bio_alloc(GFP_NOIO, 1);
687 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
688
689 bio->bi_bdev = rdev->bdev;
690 bio->bi_sector = sector;
691 bio_add_page(bio, page, size, 0);
692 bio->bi_private = rdev;
693 bio->bi_end_io = super_written;
694 bio->bi_rw = rw;
695
696 atomic_inc(&mddev->pending_writes);
697 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
698 struct bio *rbio;
699 rw |= (1<<BIO_RW_BARRIER);
700 rbio = bio_clone(bio, GFP_NOIO);
701 rbio->bi_private = bio;
702 rbio->bi_end_io = super_written_barrier;
703 submit_bio(rw, rbio);
704 } else
705 submit_bio(rw, bio);
706 }
707
708 void md_super_wait(mddev_t *mddev)
709 {
710 /* wait for all superblock writes that were scheduled to complete.
711 * if any had to be retried (due to BARRIER problems), retry them
712 */
713 DEFINE_WAIT(wq);
714 for(;;) {
715 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
716 if (atomic_read(&mddev->pending_writes)==0)
717 break;
718 while (mddev->biolist) {
719 struct bio *bio;
720 spin_lock_irq(&mddev->write_lock);
721 bio = mddev->biolist;
722 mddev->biolist = bio->bi_next ;
723 bio->bi_next = NULL;
724 spin_unlock_irq(&mddev->write_lock);
725 submit_bio(bio->bi_rw, bio);
726 }
727 schedule();
728 }
729 finish_wait(&mddev->sb_wait, &wq);
730 }
731
732 static void bi_complete(struct bio *bio, int error)
733 {
734 complete((struct completion*)bio->bi_private);
735 }
736
737 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
738 struct page *page, int rw)
739 {
740 struct bio *bio = bio_alloc(GFP_NOIO, 1);
741 struct completion event;
742 int ret;
743
744 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
745
746 bio->bi_bdev = bdev;
747 bio->bi_sector = sector;
748 bio_add_page(bio, page, size, 0);
749 init_completion(&event);
750 bio->bi_private = &event;
751 bio->bi_end_io = bi_complete;
752 submit_bio(rw, bio);
753 wait_for_completion(&event);
754
755 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
756 bio_put(bio);
757 return ret;
758 }
759 EXPORT_SYMBOL_GPL(sync_page_io);
760
761 static int read_disk_sb(mdk_rdev_t * rdev, int size)
762 {
763 char b[BDEVNAME_SIZE];
764 if (!rdev->sb_page) {
765 MD_BUG();
766 return -EINVAL;
767 }
768 if (rdev->sb_loaded)
769 return 0;
770
771
772 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
773 goto fail;
774 rdev->sb_loaded = 1;
775 return 0;
776
777 fail:
778 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
779 bdevname(rdev->bdev,b));
780 return -EINVAL;
781 }
782
783 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
784 {
785 return sb1->set_uuid0 == sb2->set_uuid0 &&
786 sb1->set_uuid1 == sb2->set_uuid1 &&
787 sb1->set_uuid2 == sb2->set_uuid2 &&
788 sb1->set_uuid3 == sb2->set_uuid3;
789 }
790
791 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
792 {
793 int ret;
794 mdp_super_t *tmp1, *tmp2;
795
796 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
797 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
798
799 if (!tmp1 || !tmp2) {
800 ret = 0;
801 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
802 goto abort;
803 }
804
805 *tmp1 = *sb1;
806 *tmp2 = *sb2;
807
808 /*
809 * nr_disks is not constant
810 */
811 tmp1->nr_disks = 0;
812 tmp2->nr_disks = 0;
813
814 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
815 abort:
816 kfree(tmp1);
817 kfree(tmp2);
818 return ret;
819 }
820
821
822 static u32 md_csum_fold(u32 csum)
823 {
824 csum = (csum & 0xffff) + (csum >> 16);
825 return (csum & 0xffff) + (csum >> 16);
826 }
827
828 static unsigned int calc_sb_csum(mdp_super_t * sb)
829 {
830 u64 newcsum = 0;
831 u32 *sb32 = (u32*)sb;
832 int i;
833 unsigned int disk_csum, csum;
834
835 disk_csum = sb->sb_csum;
836 sb->sb_csum = 0;
837
838 for (i = 0; i < MD_SB_BYTES/4 ; i++)
839 newcsum += sb32[i];
840 csum = (newcsum & 0xffffffff) + (newcsum>>32);
841
842
843 #ifdef CONFIG_ALPHA
844 /* This used to use csum_partial, which was wrong for several
845 * reasons including that different results are returned on
846 * different architectures. It isn't critical that we get exactly
847 * the same return value as before (we always csum_fold before
848 * testing, and that removes any differences). However as we
849 * know that csum_partial always returned a 16bit value on
850 * alphas, do a fold to maximise conformity to previous behaviour.
851 */
852 sb->sb_csum = md_csum_fold(disk_csum);
853 #else
854 sb->sb_csum = disk_csum;
855 #endif
856 return csum;
857 }
858
859
860 /*
861 * Handle superblock details.
862 * We want to be able to handle multiple superblock formats
863 * so we have a common interface to them all, and an array of
864 * different handlers.
865 * We rely on user-space to write the initial superblock, and support
866 * reading and updating of superblocks.
867 * Interface methods are:
868 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
869 * loads and validates a superblock on dev.
870 * if refdev != NULL, compare superblocks on both devices
871 * Return:
872 * 0 - dev has a superblock that is compatible with refdev
873 * 1 - dev has a superblock that is compatible and newer than refdev
874 * so dev should be used as the refdev in future
875 * -EINVAL superblock incompatible or invalid
876 * -othererror e.g. -EIO
877 *
878 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
879 * Verify that dev is acceptable into mddev.
880 * The first time, mddev->raid_disks will be 0, and data from
881 * dev should be merged in. Subsequent calls check that dev
882 * is new enough. Return 0 or -EINVAL
883 *
884 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
885 * Update the superblock for rdev with data in mddev
886 * This does not write to disc.
887 *
888 */
889
890 struct super_type {
891 char *name;
892 struct module *owner;
893 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
894 int minor_version);
895 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
896 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
897 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
898 sector_t num_sectors);
899 };
900
901 /*
902 * Check that the given mddev has no bitmap.
903 *
904 * This function is called from the run method of all personalities that do not
905 * support bitmaps. It prints an error message and returns non-zero if mddev
906 * has a bitmap. Otherwise, it returns 0.
907 *
908 */
909 int md_check_no_bitmap(mddev_t *mddev)
910 {
911 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
912 return 0;
913 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
914 mdname(mddev), mddev->pers->name);
915 return 1;
916 }
917 EXPORT_SYMBOL(md_check_no_bitmap);
918
919 /*
920 * load_super for 0.90.0
921 */
922 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
923 {
924 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
925 mdp_super_t *sb;
926 int ret;
927
928 /*
929 * Calculate the position of the superblock (512byte sectors),
930 * it's at the end of the disk.
931 *
932 * It also happens to be a multiple of 4Kb.
933 */
934 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
935
936 ret = read_disk_sb(rdev, MD_SB_BYTES);
937 if (ret) return ret;
938
939 ret = -EINVAL;
940
941 bdevname(rdev->bdev, b);
942 sb = (mdp_super_t*)page_address(rdev->sb_page);
943
944 if (sb->md_magic != MD_SB_MAGIC) {
945 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
946 b);
947 goto abort;
948 }
949
950 if (sb->major_version != 0 ||
951 sb->minor_version < 90 ||
952 sb->minor_version > 91) {
953 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
954 sb->major_version, sb->minor_version,
955 b);
956 goto abort;
957 }
958
959 if (sb->raid_disks <= 0)
960 goto abort;
961
962 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
963 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
964 b);
965 goto abort;
966 }
967
968 rdev->preferred_minor = sb->md_minor;
969 rdev->data_offset = 0;
970 rdev->sb_size = MD_SB_BYTES;
971
972 if (sb->level == LEVEL_MULTIPATH)
973 rdev->desc_nr = -1;
974 else
975 rdev->desc_nr = sb->this_disk.number;
976
977 if (!refdev) {
978 ret = 1;
979 } else {
980 __u64 ev1, ev2;
981 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
982 if (!uuid_equal(refsb, sb)) {
983 printk(KERN_WARNING "md: %s has different UUID to %s\n",
984 b, bdevname(refdev->bdev,b2));
985 goto abort;
986 }
987 if (!sb_equal(refsb, sb)) {
988 printk(KERN_WARNING "md: %s has same UUID"
989 " but different superblock to %s\n",
990 b, bdevname(refdev->bdev, b2));
991 goto abort;
992 }
993 ev1 = md_event(sb);
994 ev2 = md_event(refsb);
995 if (ev1 > ev2)
996 ret = 1;
997 else
998 ret = 0;
999 }
1000 rdev->sectors = rdev->sb_start;
1001
1002 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1003 /* "this cannot possibly happen" ... */
1004 ret = -EINVAL;
1005
1006 abort:
1007 return ret;
1008 }
1009
1010 /*
1011 * validate_super for 0.90.0
1012 */
1013 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1014 {
1015 mdp_disk_t *desc;
1016 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1017 __u64 ev1 = md_event(sb);
1018
1019 rdev->raid_disk = -1;
1020 clear_bit(Faulty, &rdev->flags);
1021 clear_bit(In_sync, &rdev->flags);
1022 clear_bit(WriteMostly, &rdev->flags);
1023 clear_bit(BarriersNotsupp, &rdev->flags);
1024
1025 if (mddev->raid_disks == 0) {
1026 mddev->major_version = 0;
1027 mddev->minor_version = sb->minor_version;
1028 mddev->patch_version = sb->patch_version;
1029 mddev->external = 0;
1030 mddev->chunk_sectors = sb->chunk_size >> 9;
1031 mddev->ctime = sb->ctime;
1032 mddev->utime = sb->utime;
1033 mddev->level = sb->level;
1034 mddev->clevel[0] = 0;
1035 mddev->layout = sb->layout;
1036 mddev->raid_disks = sb->raid_disks;
1037 mddev->dev_sectors = sb->size * 2;
1038 mddev->events = ev1;
1039 mddev->bitmap_info.offset = 0;
1040 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1041
1042 if (mddev->minor_version >= 91) {
1043 mddev->reshape_position = sb->reshape_position;
1044 mddev->delta_disks = sb->delta_disks;
1045 mddev->new_level = sb->new_level;
1046 mddev->new_layout = sb->new_layout;
1047 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1048 } else {
1049 mddev->reshape_position = MaxSector;
1050 mddev->delta_disks = 0;
1051 mddev->new_level = mddev->level;
1052 mddev->new_layout = mddev->layout;
1053 mddev->new_chunk_sectors = mddev->chunk_sectors;
1054 }
1055
1056 if (sb->state & (1<<MD_SB_CLEAN))
1057 mddev->recovery_cp = MaxSector;
1058 else {
1059 if (sb->events_hi == sb->cp_events_hi &&
1060 sb->events_lo == sb->cp_events_lo) {
1061 mddev->recovery_cp = sb->recovery_cp;
1062 } else
1063 mddev->recovery_cp = 0;
1064 }
1065
1066 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1067 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1068 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1069 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1070
1071 mddev->max_disks = MD_SB_DISKS;
1072
1073 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1074 mddev->bitmap_info.file == NULL)
1075 mddev->bitmap_info.offset =
1076 mddev->bitmap_info.default_offset;
1077
1078 } else if (mddev->pers == NULL) {
1079 /* Insist on good event counter while assembling, except
1080 * for spares (which don't need an event count) */
1081 ++ev1;
1082 if (sb->disks[rdev->desc_nr].state & (
1083 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1084 if (ev1 < mddev->events)
1085 return -EINVAL;
1086 } else if (mddev->bitmap) {
1087 /* if adding to array with a bitmap, then we can accept an
1088 * older device ... but not too old.
1089 */
1090 if (ev1 < mddev->bitmap->events_cleared)
1091 return 0;
1092 } else {
1093 if (ev1 < mddev->events)
1094 /* just a hot-add of a new device, leave raid_disk at -1 */
1095 return 0;
1096 }
1097
1098 if (mddev->level != LEVEL_MULTIPATH) {
1099 desc = sb->disks + rdev->desc_nr;
1100
1101 if (desc->state & (1<<MD_DISK_FAULTY))
1102 set_bit(Faulty, &rdev->flags);
1103 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1104 desc->raid_disk < mddev->raid_disks */) {
1105 set_bit(In_sync, &rdev->flags);
1106 rdev->raid_disk = desc->raid_disk;
1107 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1108 /* active but not in sync implies recovery up to
1109 * reshape position. We don't know exactly where
1110 * that is, so set to zero for now */
1111 if (mddev->minor_version >= 91) {
1112 rdev->recovery_offset = 0;
1113 rdev->raid_disk = desc->raid_disk;
1114 }
1115 }
1116 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1117 set_bit(WriteMostly, &rdev->flags);
1118 } else /* MULTIPATH are always insync */
1119 set_bit(In_sync, &rdev->flags);
1120 return 0;
1121 }
1122
1123 /*
1124 * sync_super for 0.90.0
1125 */
1126 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1127 {
1128 mdp_super_t *sb;
1129 mdk_rdev_t *rdev2;
1130 int next_spare = mddev->raid_disks;
1131
1132
1133 /* make rdev->sb match mddev data..
1134 *
1135 * 1/ zero out disks
1136 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1137 * 3/ any empty disks < next_spare become removed
1138 *
1139 * disks[0] gets initialised to REMOVED because
1140 * we cannot be sure from other fields if it has
1141 * been initialised or not.
1142 */
1143 int i;
1144 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1145
1146 rdev->sb_size = MD_SB_BYTES;
1147
1148 sb = (mdp_super_t*)page_address(rdev->sb_page);
1149
1150 memset(sb, 0, sizeof(*sb));
1151
1152 sb->md_magic = MD_SB_MAGIC;
1153 sb->major_version = mddev->major_version;
1154 sb->patch_version = mddev->patch_version;
1155 sb->gvalid_words = 0; /* ignored */
1156 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1157 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1158 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1159 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1160
1161 sb->ctime = mddev->ctime;
1162 sb->level = mddev->level;
1163 sb->size = mddev->dev_sectors / 2;
1164 sb->raid_disks = mddev->raid_disks;
1165 sb->md_minor = mddev->md_minor;
1166 sb->not_persistent = 0;
1167 sb->utime = mddev->utime;
1168 sb->state = 0;
1169 sb->events_hi = (mddev->events>>32);
1170 sb->events_lo = (u32)mddev->events;
1171
1172 if (mddev->reshape_position == MaxSector)
1173 sb->minor_version = 90;
1174 else {
1175 sb->minor_version = 91;
1176 sb->reshape_position = mddev->reshape_position;
1177 sb->new_level = mddev->new_level;
1178 sb->delta_disks = mddev->delta_disks;
1179 sb->new_layout = mddev->new_layout;
1180 sb->new_chunk = mddev->new_chunk_sectors << 9;
1181 }
1182 mddev->minor_version = sb->minor_version;
1183 if (mddev->in_sync)
1184 {
1185 sb->recovery_cp = mddev->recovery_cp;
1186 sb->cp_events_hi = (mddev->events>>32);
1187 sb->cp_events_lo = (u32)mddev->events;
1188 if (mddev->recovery_cp == MaxSector)
1189 sb->state = (1<< MD_SB_CLEAN);
1190 } else
1191 sb->recovery_cp = 0;
1192
1193 sb->layout = mddev->layout;
1194 sb->chunk_size = mddev->chunk_sectors << 9;
1195
1196 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1197 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1198
1199 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1200 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1201 mdp_disk_t *d;
1202 int desc_nr;
1203 int is_active = test_bit(In_sync, &rdev2->flags);
1204
1205 if (rdev2->raid_disk >= 0 &&
1206 sb->minor_version >= 91)
1207 /* we have nowhere to store the recovery_offset,
1208 * but if it is not below the reshape_position,
1209 * we can piggy-back on that.
1210 */
1211 is_active = 1;
1212 if (rdev2->raid_disk < 0 ||
1213 test_bit(Faulty, &rdev2->flags))
1214 is_active = 0;
1215 if (is_active)
1216 desc_nr = rdev2->raid_disk;
1217 else
1218 desc_nr = next_spare++;
1219 rdev2->desc_nr = desc_nr;
1220 d = &sb->disks[rdev2->desc_nr];
1221 nr_disks++;
1222 d->number = rdev2->desc_nr;
1223 d->major = MAJOR(rdev2->bdev->bd_dev);
1224 d->minor = MINOR(rdev2->bdev->bd_dev);
1225 if (is_active)
1226 d->raid_disk = rdev2->raid_disk;
1227 else
1228 d->raid_disk = rdev2->desc_nr; /* compatibility */
1229 if (test_bit(Faulty, &rdev2->flags))
1230 d->state = (1<<MD_DISK_FAULTY);
1231 else if (is_active) {
1232 d->state = (1<<MD_DISK_ACTIVE);
1233 if (test_bit(In_sync, &rdev2->flags))
1234 d->state |= (1<<MD_DISK_SYNC);
1235 active++;
1236 working++;
1237 } else {
1238 d->state = 0;
1239 spare++;
1240 working++;
1241 }
1242 if (test_bit(WriteMostly, &rdev2->flags))
1243 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1244 }
1245 /* now set the "removed" and "faulty" bits on any missing devices */
1246 for (i=0 ; i < mddev->raid_disks ; i++) {
1247 mdp_disk_t *d = &sb->disks[i];
1248 if (d->state == 0 && d->number == 0) {
1249 d->number = i;
1250 d->raid_disk = i;
1251 d->state = (1<<MD_DISK_REMOVED);
1252 d->state |= (1<<MD_DISK_FAULTY);
1253 failed++;
1254 }
1255 }
1256 sb->nr_disks = nr_disks;
1257 sb->active_disks = active;
1258 sb->working_disks = working;
1259 sb->failed_disks = failed;
1260 sb->spare_disks = spare;
1261
1262 sb->this_disk = sb->disks[rdev->desc_nr];
1263 sb->sb_csum = calc_sb_csum(sb);
1264 }
1265
1266 /*
1267 * rdev_size_change for 0.90.0
1268 */
1269 static unsigned long long
1270 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1271 {
1272 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1273 return 0; /* component must fit device */
1274 if (rdev->mddev->bitmap_info.offset)
1275 return 0; /* can't move bitmap */
1276 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1277 if (!num_sectors || num_sectors > rdev->sb_start)
1278 num_sectors = rdev->sb_start;
1279 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1280 rdev->sb_page);
1281 md_super_wait(rdev->mddev);
1282 return num_sectors / 2; /* kB for sysfs */
1283 }
1284
1285
1286 /*
1287 * version 1 superblock
1288 */
1289
1290 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1291 {
1292 __le32 disk_csum;
1293 u32 csum;
1294 unsigned long long newcsum;
1295 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1296 __le32 *isuper = (__le32*)sb;
1297 int i;
1298
1299 disk_csum = sb->sb_csum;
1300 sb->sb_csum = 0;
1301 newcsum = 0;
1302 for (i=0; size>=4; size -= 4 )
1303 newcsum += le32_to_cpu(*isuper++);
1304
1305 if (size == 2)
1306 newcsum += le16_to_cpu(*(__le16*) isuper);
1307
1308 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1309 sb->sb_csum = disk_csum;
1310 return cpu_to_le32(csum);
1311 }
1312
1313 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1314 {
1315 struct mdp_superblock_1 *sb;
1316 int ret;
1317 sector_t sb_start;
1318 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1319 int bmask;
1320
1321 /*
1322 * Calculate the position of the superblock in 512byte sectors.
1323 * It is always aligned to a 4K boundary and
1324 * depeding on minor_version, it can be:
1325 * 0: At least 8K, but less than 12K, from end of device
1326 * 1: At start of device
1327 * 2: 4K from start of device.
1328 */
1329 switch(minor_version) {
1330 case 0:
1331 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1332 sb_start -= 8*2;
1333 sb_start &= ~(sector_t)(4*2-1);
1334 break;
1335 case 1:
1336 sb_start = 0;
1337 break;
1338 case 2:
1339 sb_start = 8;
1340 break;
1341 default:
1342 return -EINVAL;
1343 }
1344 rdev->sb_start = sb_start;
1345
1346 /* superblock is rarely larger than 1K, but it can be larger,
1347 * and it is safe to read 4k, so we do that
1348 */
1349 ret = read_disk_sb(rdev, 4096);
1350 if (ret) return ret;
1351
1352
1353 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1354
1355 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1356 sb->major_version != cpu_to_le32(1) ||
1357 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1358 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1359 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1360 return -EINVAL;
1361
1362 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1363 printk("md: invalid superblock checksum on %s\n",
1364 bdevname(rdev->bdev,b));
1365 return -EINVAL;
1366 }
1367 if (le64_to_cpu(sb->data_size) < 10) {
1368 printk("md: data_size too small on %s\n",
1369 bdevname(rdev->bdev,b));
1370 return -EINVAL;
1371 }
1372
1373 rdev->preferred_minor = 0xffff;
1374 rdev->data_offset = le64_to_cpu(sb->data_offset);
1375 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1376
1377 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1378 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1379 if (rdev->sb_size & bmask)
1380 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1381
1382 if (minor_version
1383 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1384 return -EINVAL;
1385
1386 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1387 rdev->desc_nr = -1;
1388 else
1389 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1390
1391 if (!refdev) {
1392 ret = 1;
1393 } else {
1394 __u64 ev1, ev2;
1395 struct mdp_superblock_1 *refsb =
1396 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1397
1398 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1399 sb->level != refsb->level ||
1400 sb->layout != refsb->layout ||
1401 sb->chunksize != refsb->chunksize) {
1402 printk(KERN_WARNING "md: %s has strangely different"
1403 " superblock to %s\n",
1404 bdevname(rdev->bdev,b),
1405 bdevname(refdev->bdev,b2));
1406 return -EINVAL;
1407 }
1408 ev1 = le64_to_cpu(sb->events);
1409 ev2 = le64_to_cpu(refsb->events);
1410
1411 if (ev1 > ev2)
1412 ret = 1;
1413 else
1414 ret = 0;
1415 }
1416 if (minor_version)
1417 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1418 le64_to_cpu(sb->data_offset);
1419 else
1420 rdev->sectors = rdev->sb_start;
1421 if (rdev->sectors < le64_to_cpu(sb->data_size))
1422 return -EINVAL;
1423 rdev->sectors = le64_to_cpu(sb->data_size);
1424 if (le64_to_cpu(sb->size) > rdev->sectors)
1425 return -EINVAL;
1426 return ret;
1427 }
1428
1429 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1430 {
1431 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1432 __u64 ev1 = le64_to_cpu(sb->events);
1433
1434 rdev->raid_disk = -1;
1435 clear_bit(Faulty, &rdev->flags);
1436 clear_bit(In_sync, &rdev->flags);
1437 clear_bit(WriteMostly, &rdev->flags);
1438 clear_bit(BarriersNotsupp, &rdev->flags);
1439
1440 if (mddev->raid_disks == 0) {
1441 mddev->major_version = 1;
1442 mddev->patch_version = 0;
1443 mddev->external = 0;
1444 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1445 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1446 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1447 mddev->level = le32_to_cpu(sb->level);
1448 mddev->clevel[0] = 0;
1449 mddev->layout = le32_to_cpu(sb->layout);
1450 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1451 mddev->dev_sectors = le64_to_cpu(sb->size);
1452 mddev->events = ev1;
1453 mddev->bitmap_info.offset = 0;
1454 mddev->bitmap_info.default_offset = 1024 >> 9;
1455
1456 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1457 memcpy(mddev->uuid, sb->set_uuid, 16);
1458
1459 mddev->max_disks = (4096-256)/2;
1460
1461 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1462 mddev->bitmap_info.file == NULL )
1463 mddev->bitmap_info.offset =
1464 (__s32)le32_to_cpu(sb->bitmap_offset);
1465
1466 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1467 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1468 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1469 mddev->new_level = le32_to_cpu(sb->new_level);
1470 mddev->new_layout = le32_to_cpu(sb->new_layout);
1471 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1472 } else {
1473 mddev->reshape_position = MaxSector;
1474 mddev->delta_disks = 0;
1475 mddev->new_level = mddev->level;
1476 mddev->new_layout = mddev->layout;
1477 mddev->new_chunk_sectors = mddev->chunk_sectors;
1478 }
1479
1480 } else if (mddev->pers == NULL) {
1481 /* Insist of good event counter while assembling, except for
1482 * spares (which don't need an event count) */
1483 ++ev1;
1484 if (rdev->desc_nr >= 0 &&
1485 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1486 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1487 if (ev1 < mddev->events)
1488 return -EINVAL;
1489 } else if (mddev->bitmap) {
1490 /* If adding to array with a bitmap, then we can accept an
1491 * older device, but not too old.
1492 */
1493 if (ev1 < mddev->bitmap->events_cleared)
1494 return 0;
1495 } else {
1496 if (ev1 < mddev->events)
1497 /* just a hot-add of a new device, leave raid_disk at -1 */
1498 return 0;
1499 }
1500 if (mddev->level != LEVEL_MULTIPATH) {
1501 int role;
1502 if (rdev->desc_nr < 0 ||
1503 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1504 role = 0xffff;
1505 rdev->desc_nr = -1;
1506 } else
1507 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1508 switch(role) {
1509 case 0xffff: /* spare */
1510 break;
1511 case 0xfffe: /* faulty */
1512 set_bit(Faulty, &rdev->flags);
1513 break;
1514 default:
1515 if ((le32_to_cpu(sb->feature_map) &
1516 MD_FEATURE_RECOVERY_OFFSET))
1517 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1518 else
1519 set_bit(In_sync, &rdev->flags);
1520 rdev->raid_disk = role;
1521 break;
1522 }
1523 if (sb->devflags & WriteMostly1)
1524 set_bit(WriteMostly, &rdev->flags);
1525 } else /* MULTIPATH are always insync */
1526 set_bit(In_sync, &rdev->flags);
1527
1528 return 0;
1529 }
1530
1531 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1532 {
1533 struct mdp_superblock_1 *sb;
1534 mdk_rdev_t *rdev2;
1535 int max_dev, i;
1536 /* make rdev->sb match mddev and rdev data. */
1537
1538 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1539
1540 sb->feature_map = 0;
1541 sb->pad0 = 0;
1542 sb->recovery_offset = cpu_to_le64(0);
1543 memset(sb->pad1, 0, sizeof(sb->pad1));
1544 memset(sb->pad2, 0, sizeof(sb->pad2));
1545 memset(sb->pad3, 0, sizeof(sb->pad3));
1546
1547 sb->utime = cpu_to_le64((__u64)mddev->utime);
1548 sb->events = cpu_to_le64(mddev->events);
1549 if (mddev->in_sync)
1550 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1551 else
1552 sb->resync_offset = cpu_to_le64(0);
1553
1554 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1555
1556 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1557 sb->size = cpu_to_le64(mddev->dev_sectors);
1558 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1559 sb->level = cpu_to_le32(mddev->level);
1560 sb->layout = cpu_to_le32(mddev->layout);
1561
1562 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1563 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1564 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1565 }
1566
1567 if (rdev->raid_disk >= 0 &&
1568 !test_bit(In_sync, &rdev->flags)) {
1569 sb->feature_map |=
1570 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1571 sb->recovery_offset =
1572 cpu_to_le64(rdev->recovery_offset);
1573 }
1574
1575 if (mddev->reshape_position != MaxSector) {
1576 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1577 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1578 sb->new_layout = cpu_to_le32(mddev->new_layout);
1579 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1580 sb->new_level = cpu_to_le32(mddev->new_level);
1581 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1582 }
1583
1584 max_dev = 0;
1585 list_for_each_entry(rdev2, &mddev->disks, same_set)
1586 if (rdev2->desc_nr+1 > max_dev)
1587 max_dev = rdev2->desc_nr+1;
1588
1589 if (max_dev > le32_to_cpu(sb->max_dev)) {
1590 int bmask;
1591 sb->max_dev = cpu_to_le32(max_dev);
1592 rdev->sb_size = max_dev * 2 + 256;
1593 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1594 if (rdev->sb_size & bmask)
1595 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1596 }
1597 for (i=0; i<max_dev;i++)
1598 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1599
1600 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1601 i = rdev2->desc_nr;
1602 if (test_bit(Faulty, &rdev2->flags))
1603 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1604 else if (test_bit(In_sync, &rdev2->flags))
1605 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1606 else if (rdev2->raid_disk >= 0)
1607 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1608 else
1609 sb->dev_roles[i] = cpu_to_le16(0xffff);
1610 }
1611
1612 sb->sb_csum = calc_sb_1_csum(sb);
1613 }
1614
1615 static unsigned long long
1616 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1617 {
1618 struct mdp_superblock_1 *sb;
1619 sector_t max_sectors;
1620 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1621 return 0; /* component must fit device */
1622 if (rdev->sb_start < rdev->data_offset) {
1623 /* minor versions 1 and 2; superblock before data */
1624 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1625 max_sectors -= rdev->data_offset;
1626 if (!num_sectors || num_sectors > max_sectors)
1627 num_sectors = max_sectors;
1628 } else if (rdev->mddev->bitmap_info.offset) {
1629 /* minor version 0 with bitmap we can't move */
1630 return 0;
1631 } else {
1632 /* minor version 0; superblock after data */
1633 sector_t sb_start;
1634 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1635 sb_start &= ~(sector_t)(4*2 - 1);
1636 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1637 if (!num_sectors || num_sectors > max_sectors)
1638 num_sectors = max_sectors;
1639 rdev->sb_start = sb_start;
1640 }
1641 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1642 sb->data_size = cpu_to_le64(num_sectors);
1643 sb->super_offset = rdev->sb_start;
1644 sb->sb_csum = calc_sb_1_csum(sb);
1645 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1646 rdev->sb_page);
1647 md_super_wait(rdev->mddev);
1648 return num_sectors / 2; /* kB for sysfs */
1649 }
1650
1651 static struct super_type super_types[] = {
1652 [0] = {
1653 .name = "0.90.0",
1654 .owner = THIS_MODULE,
1655 .load_super = super_90_load,
1656 .validate_super = super_90_validate,
1657 .sync_super = super_90_sync,
1658 .rdev_size_change = super_90_rdev_size_change,
1659 },
1660 [1] = {
1661 .name = "md-1",
1662 .owner = THIS_MODULE,
1663 .load_super = super_1_load,
1664 .validate_super = super_1_validate,
1665 .sync_super = super_1_sync,
1666 .rdev_size_change = super_1_rdev_size_change,
1667 },
1668 };
1669
1670 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1671 {
1672 mdk_rdev_t *rdev, *rdev2;
1673
1674 rcu_read_lock();
1675 rdev_for_each_rcu(rdev, mddev1)
1676 rdev_for_each_rcu(rdev2, mddev2)
1677 if (rdev->bdev->bd_contains ==
1678 rdev2->bdev->bd_contains) {
1679 rcu_read_unlock();
1680 return 1;
1681 }
1682 rcu_read_unlock();
1683 return 0;
1684 }
1685
1686 static LIST_HEAD(pending_raid_disks);
1687
1688 /*
1689 * Try to register data integrity profile for an mddev
1690 *
1691 * This is called when an array is started and after a disk has been kicked
1692 * from the array. It only succeeds if all working and active component devices
1693 * are integrity capable with matching profiles.
1694 */
1695 int md_integrity_register(mddev_t *mddev)
1696 {
1697 mdk_rdev_t *rdev, *reference = NULL;
1698
1699 if (list_empty(&mddev->disks))
1700 return 0; /* nothing to do */
1701 if (blk_get_integrity(mddev->gendisk))
1702 return 0; /* already registered */
1703 list_for_each_entry(rdev, &mddev->disks, same_set) {
1704 /* skip spares and non-functional disks */
1705 if (test_bit(Faulty, &rdev->flags))
1706 continue;
1707 if (rdev->raid_disk < 0)
1708 continue;
1709 /*
1710 * If at least one rdev is not integrity capable, we can not
1711 * enable data integrity for the md device.
1712 */
1713 if (!bdev_get_integrity(rdev->bdev))
1714 return -EINVAL;
1715 if (!reference) {
1716 /* Use the first rdev as the reference */
1717 reference = rdev;
1718 continue;
1719 }
1720 /* does this rdev's profile match the reference profile? */
1721 if (blk_integrity_compare(reference->bdev->bd_disk,
1722 rdev->bdev->bd_disk) < 0)
1723 return -EINVAL;
1724 }
1725 /*
1726 * All component devices are integrity capable and have matching
1727 * profiles, register the common profile for the md device.
1728 */
1729 if (blk_integrity_register(mddev->gendisk,
1730 bdev_get_integrity(reference->bdev)) != 0) {
1731 printk(KERN_ERR "md: failed to register integrity for %s\n",
1732 mdname(mddev));
1733 return -EINVAL;
1734 }
1735 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1736 mdname(mddev));
1737 return 0;
1738 }
1739 EXPORT_SYMBOL(md_integrity_register);
1740
1741 /* Disable data integrity if non-capable/non-matching disk is being added */
1742 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1743 {
1744 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1745 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1746
1747 if (!bi_mddev) /* nothing to do */
1748 return;
1749 if (rdev->raid_disk < 0) /* skip spares */
1750 return;
1751 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1752 rdev->bdev->bd_disk) >= 0)
1753 return;
1754 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1755 blk_integrity_unregister(mddev->gendisk);
1756 }
1757 EXPORT_SYMBOL(md_integrity_add_rdev);
1758
1759 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1760 {
1761 char b[BDEVNAME_SIZE];
1762 struct kobject *ko;
1763 char *s;
1764 int err;
1765
1766 if (rdev->mddev) {
1767 MD_BUG();
1768 return -EINVAL;
1769 }
1770
1771 /* prevent duplicates */
1772 if (find_rdev(mddev, rdev->bdev->bd_dev))
1773 return -EEXIST;
1774
1775 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1776 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1777 rdev->sectors < mddev->dev_sectors)) {
1778 if (mddev->pers) {
1779 /* Cannot change size, so fail
1780 * If mddev->level <= 0, then we don't care
1781 * about aligning sizes (e.g. linear)
1782 */
1783 if (mddev->level > 0)
1784 return -ENOSPC;
1785 } else
1786 mddev->dev_sectors = rdev->sectors;
1787 }
1788
1789 /* Verify rdev->desc_nr is unique.
1790 * If it is -1, assign a free number, else
1791 * check number is not in use
1792 */
1793 if (rdev->desc_nr < 0) {
1794 int choice = 0;
1795 if (mddev->pers) choice = mddev->raid_disks;
1796 while (find_rdev_nr(mddev, choice))
1797 choice++;
1798 rdev->desc_nr = choice;
1799 } else {
1800 if (find_rdev_nr(mddev, rdev->desc_nr))
1801 return -EBUSY;
1802 }
1803 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1804 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1805 mdname(mddev), mddev->max_disks);
1806 return -EBUSY;
1807 }
1808 bdevname(rdev->bdev,b);
1809 while ( (s=strchr(b, '/')) != NULL)
1810 *s = '!';
1811
1812 rdev->mddev = mddev;
1813 printk(KERN_INFO "md: bind<%s>\n", b);
1814
1815 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1816 goto fail;
1817
1818 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1819 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1820 /* failure here is OK */;
1821 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1822
1823 list_add_rcu(&rdev->same_set, &mddev->disks);
1824 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1825
1826 /* May as well allow recovery to be retried once */
1827 mddev->recovery_disabled = 0;
1828
1829 return 0;
1830
1831 fail:
1832 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1833 b, mdname(mddev));
1834 return err;
1835 }
1836
1837 static void md_delayed_delete(struct work_struct *ws)
1838 {
1839 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1840 kobject_del(&rdev->kobj);
1841 kobject_put(&rdev->kobj);
1842 }
1843
1844 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1845 {
1846 char b[BDEVNAME_SIZE];
1847 if (!rdev->mddev) {
1848 MD_BUG();
1849 return;
1850 }
1851 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1852 list_del_rcu(&rdev->same_set);
1853 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1854 rdev->mddev = NULL;
1855 sysfs_remove_link(&rdev->kobj, "block");
1856 sysfs_put(rdev->sysfs_state);
1857 rdev->sysfs_state = NULL;
1858 /* We need to delay this, otherwise we can deadlock when
1859 * writing to 'remove' to "dev/state". We also need
1860 * to delay it due to rcu usage.
1861 */
1862 synchronize_rcu();
1863 INIT_WORK(&rdev->del_work, md_delayed_delete);
1864 kobject_get(&rdev->kobj);
1865 schedule_work(&rdev->del_work);
1866 }
1867
1868 /*
1869 * prevent the device from being mounted, repartitioned or
1870 * otherwise reused by a RAID array (or any other kernel
1871 * subsystem), by bd_claiming the device.
1872 */
1873 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1874 {
1875 int err = 0;
1876 struct block_device *bdev;
1877 char b[BDEVNAME_SIZE];
1878
1879 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1880 if (IS_ERR(bdev)) {
1881 printk(KERN_ERR "md: could not open %s.\n",
1882 __bdevname(dev, b));
1883 return PTR_ERR(bdev);
1884 }
1885 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1886 if (err) {
1887 printk(KERN_ERR "md: could not bd_claim %s.\n",
1888 bdevname(bdev, b));
1889 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1890 return err;
1891 }
1892 if (!shared)
1893 set_bit(AllReserved, &rdev->flags);
1894 rdev->bdev = bdev;
1895 return err;
1896 }
1897
1898 static void unlock_rdev(mdk_rdev_t *rdev)
1899 {
1900 struct block_device *bdev = rdev->bdev;
1901 rdev->bdev = NULL;
1902 if (!bdev)
1903 MD_BUG();
1904 bd_release(bdev);
1905 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1906 }
1907
1908 void md_autodetect_dev(dev_t dev);
1909
1910 static void export_rdev(mdk_rdev_t * rdev)
1911 {
1912 char b[BDEVNAME_SIZE];
1913 printk(KERN_INFO "md: export_rdev(%s)\n",
1914 bdevname(rdev->bdev,b));
1915 if (rdev->mddev)
1916 MD_BUG();
1917 free_disk_sb(rdev);
1918 #ifndef MODULE
1919 if (test_bit(AutoDetected, &rdev->flags))
1920 md_autodetect_dev(rdev->bdev->bd_dev);
1921 #endif
1922 unlock_rdev(rdev);
1923 kobject_put(&rdev->kobj);
1924 }
1925
1926 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1927 {
1928 unbind_rdev_from_array(rdev);
1929 export_rdev(rdev);
1930 }
1931
1932 static void export_array(mddev_t *mddev)
1933 {
1934 mdk_rdev_t *rdev, *tmp;
1935
1936 rdev_for_each(rdev, tmp, mddev) {
1937 if (!rdev->mddev) {
1938 MD_BUG();
1939 continue;
1940 }
1941 kick_rdev_from_array(rdev);
1942 }
1943 if (!list_empty(&mddev->disks))
1944 MD_BUG();
1945 mddev->raid_disks = 0;
1946 mddev->major_version = 0;
1947 }
1948
1949 static void print_desc(mdp_disk_t *desc)
1950 {
1951 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1952 desc->major,desc->minor,desc->raid_disk,desc->state);
1953 }
1954
1955 static void print_sb_90(mdp_super_t *sb)
1956 {
1957 int i;
1958
1959 printk(KERN_INFO
1960 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1961 sb->major_version, sb->minor_version, sb->patch_version,
1962 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1963 sb->ctime);
1964 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1965 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1966 sb->md_minor, sb->layout, sb->chunk_size);
1967 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1968 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1969 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1970 sb->failed_disks, sb->spare_disks,
1971 sb->sb_csum, (unsigned long)sb->events_lo);
1972
1973 printk(KERN_INFO);
1974 for (i = 0; i < MD_SB_DISKS; i++) {
1975 mdp_disk_t *desc;
1976
1977 desc = sb->disks + i;
1978 if (desc->number || desc->major || desc->minor ||
1979 desc->raid_disk || (desc->state && (desc->state != 4))) {
1980 printk(" D %2d: ", i);
1981 print_desc(desc);
1982 }
1983 }
1984 printk(KERN_INFO "md: THIS: ");
1985 print_desc(&sb->this_disk);
1986 }
1987
1988 static void print_sb_1(struct mdp_superblock_1 *sb)
1989 {
1990 __u8 *uuid;
1991
1992 uuid = sb->set_uuid;
1993 printk(KERN_INFO
1994 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1995 "md: Name: \"%s\" CT:%llu\n",
1996 le32_to_cpu(sb->major_version),
1997 le32_to_cpu(sb->feature_map),
1998 uuid,
1999 sb->set_name,
2000 (unsigned long long)le64_to_cpu(sb->ctime)
2001 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2002
2003 uuid = sb->device_uuid;
2004 printk(KERN_INFO
2005 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2006 " RO:%llu\n"
2007 "md: Dev:%08x UUID: %pU\n"
2008 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2009 "md: (MaxDev:%u) \n",
2010 le32_to_cpu(sb->level),
2011 (unsigned long long)le64_to_cpu(sb->size),
2012 le32_to_cpu(sb->raid_disks),
2013 le32_to_cpu(sb->layout),
2014 le32_to_cpu(sb->chunksize),
2015 (unsigned long long)le64_to_cpu(sb->data_offset),
2016 (unsigned long long)le64_to_cpu(sb->data_size),
2017 (unsigned long long)le64_to_cpu(sb->super_offset),
2018 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2019 le32_to_cpu(sb->dev_number),
2020 uuid,
2021 sb->devflags,
2022 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2023 (unsigned long long)le64_to_cpu(sb->events),
2024 (unsigned long long)le64_to_cpu(sb->resync_offset),
2025 le32_to_cpu(sb->sb_csum),
2026 le32_to_cpu(sb->max_dev)
2027 );
2028 }
2029
2030 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2031 {
2032 char b[BDEVNAME_SIZE];
2033 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2034 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2035 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2036 rdev->desc_nr);
2037 if (rdev->sb_loaded) {
2038 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2039 switch (major_version) {
2040 case 0:
2041 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2042 break;
2043 case 1:
2044 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2045 break;
2046 }
2047 } else
2048 printk(KERN_INFO "md: no rdev superblock!\n");
2049 }
2050
2051 static void md_print_devices(void)
2052 {
2053 struct list_head *tmp;
2054 mdk_rdev_t *rdev;
2055 mddev_t *mddev;
2056 char b[BDEVNAME_SIZE];
2057
2058 printk("\n");
2059 printk("md: **********************************\n");
2060 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2061 printk("md: **********************************\n");
2062 for_each_mddev(mddev, tmp) {
2063
2064 if (mddev->bitmap)
2065 bitmap_print_sb(mddev->bitmap);
2066 else
2067 printk("%s: ", mdname(mddev));
2068 list_for_each_entry(rdev, &mddev->disks, same_set)
2069 printk("<%s>", bdevname(rdev->bdev,b));
2070 printk("\n");
2071
2072 list_for_each_entry(rdev, &mddev->disks, same_set)
2073 print_rdev(rdev, mddev->major_version);
2074 }
2075 printk("md: **********************************\n");
2076 printk("\n");
2077 }
2078
2079
2080 static void sync_sbs(mddev_t * mddev, int nospares)
2081 {
2082 /* Update each superblock (in-memory image), but
2083 * if we are allowed to, skip spares which already
2084 * have the right event counter, or have one earlier
2085 * (which would mean they aren't being marked as dirty
2086 * with the rest of the array)
2087 */
2088 mdk_rdev_t *rdev;
2089
2090 /* First make sure individual recovery_offsets are correct */
2091 list_for_each_entry(rdev, &mddev->disks, same_set) {
2092 if (rdev->raid_disk >= 0 &&
2093 mddev->delta_disks >= 0 &&
2094 !test_bit(In_sync, &rdev->flags) &&
2095 mddev->curr_resync_completed > rdev->recovery_offset)
2096 rdev->recovery_offset = mddev->curr_resync_completed;
2097
2098 }
2099 list_for_each_entry(rdev, &mddev->disks, same_set) {
2100 if (rdev->sb_events == mddev->events ||
2101 (nospares &&
2102 rdev->raid_disk < 0 &&
2103 rdev->sb_events+1 == mddev->events)) {
2104 /* Don't update this superblock */
2105 rdev->sb_loaded = 2;
2106 } else {
2107 super_types[mddev->major_version].
2108 sync_super(mddev, rdev);
2109 rdev->sb_loaded = 1;
2110 }
2111 }
2112 }
2113
2114 static void md_update_sb(mddev_t * mddev, int force_change)
2115 {
2116 mdk_rdev_t *rdev;
2117 int sync_req;
2118 int nospares = 0;
2119
2120 mddev->utime = get_seconds();
2121 if (mddev->external)
2122 return;
2123 repeat:
2124 spin_lock_irq(&mddev->write_lock);
2125
2126 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2127 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2128 force_change = 1;
2129 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2130 /* just a clean<-> dirty transition, possibly leave spares alone,
2131 * though if events isn't the right even/odd, we will have to do
2132 * spares after all
2133 */
2134 nospares = 1;
2135 if (force_change)
2136 nospares = 0;
2137 if (mddev->degraded)
2138 /* If the array is degraded, then skipping spares is both
2139 * dangerous and fairly pointless.
2140 * Dangerous because a device that was removed from the array
2141 * might have a event_count that still looks up-to-date,
2142 * so it can be re-added without a resync.
2143 * Pointless because if there are any spares to skip,
2144 * then a recovery will happen and soon that array won't
2145 * be degraded any more and the spare can go back to sleep then.
2146 */
2147 nospares = 0;
2148
2149 sync_req = mddev->in_sync;
2150
2151 /* If this is just a dirty<->clean transition, and the array is clean
2152 * and 'events' is odd, we can roll back to the previous clean state */
2153 if (nospares
2154 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2155 && mddev->can_decrease_events
2156 && mddev->events != 1) {
2157 mddev->events--;
2158 mddev->can_decrease_events = 0;
2159 } else {
2160 /* otherwise we have to go forward and ... */
2161 mddev->events ++;
2162 mddev->can_decrease_events = nospares;
2163 }
2164
2165 if (!mddev->events) {
2166 /*
2167 * oops, this 64-bit counter should never wrap.
2168 * Either we are in around ~1 trillion A.C., assuming
2169 * 1 reboot per second, or we have a bug:
2170 */
2171 MD_BUG();
2172 mddev->events --;
2173 }
2174
2175 /*
2176 * do not write anything to disk if using
2177 * nonpersistent superblocks
2178 */
2179 if (!mddev->persistent) {
2180 if (!mddev->external)
2181 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2182
2183 spin_unlock_irq(&mddev->write_lock);
2184 wake_up(&mddev->sb_wait);
2185 return;
2186 }
2187 sync_sbs(mddev, nospares);
2188 spin_unlock_irq(&mddev->write_lock);
2189
2190 dprintk(KERN_INFO
2191 "md: updating %s RAID superblock on device (in sync %d)\n",
2192 mdname(mddev),mddev->in_sync);
2193
2194 bitmap_update_sb(mddev->bitmap);
2195 list_for_each_entry(rdev, &mddev->disks, same_set) {
2196 char b[BDEVNAME_SIZE];
2197 dprintk(KERN_INFO "md: ");
2198 if (rdev->sb_loaded != 1)
2199 continue; /* no noise on spare devices */
2200 if (test_bit(Faulty, &rdev->flags))
2201 dprintk("(skipping faulty ");
2202
2203 dprintk("%s ", bdevname(rdev->bdev,b));
2204 if (!test_bit(Faulty, &rdev->flags)) {
2205 md_super_write(mddev,rdev,
2206 rdev->sb_start, rdev->sb_size,
2207 rdev->sb_page);
2208 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2209 bdevname(rdev->bdev,b),
2210 (unsigned long long)rdev->sb_start);
2211 rdev->sb_events = mddev->events;
2212
2213 } else
2214 dprintk(")\n");
2215 if (mddev->level == LEVEL_MULTIPATH)
2216 /* only need to write one superblock... */
2217 break;
2218 }
2219 md_super_wait(mddev);
2220 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2221
2222 spin_lock_irq(&mddev->write_lock);
2223 if (mddev->in_sync != sync_req ||
2224 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2225 /* have to write it out again */
2226 spin_unlock_irq(&mddev->write_lock);
2227 goto repeat;
2228 }
2229 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2230 spin_unlock_irq(&mddev->write_lock);
2231 wake_up(&mddev->sb_wait);
2232 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2233 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2234
2235 }
2236
2237 /* words written to sysfs files may, or may not, be \n terminated.
2238 * We want to accept with case. For this we use cmd_match.
2239 */
2240 static int cmd_match(const char *cmd, const char *str)
2241 {
2242 /* See if cmd, written into a sysfs file, matches
2243 * str. They must either be the same, or cmd can
2244 * have a trailing newline
2245 */
2246 while (*cmd && *str && *cmd == *str) {
2247 cmd++;
2248 str++;
2249 }
2250 if (*cmd == '\n')
2251 cmd++;
2252 if (*str || *cmd)
2253 return 0;
2254 return 1;
2255 }
2256
2257 struct rdev_sysfs_entry {
2258 struct attribute attr;
2259 ssize_t (*show)(mdk_rdev_t *, char *);
2260 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2261 };
2262
2263 static ssize_t
2264 state_show(mdk_rdev_t *rdev, char *page)
2265 {
2266 char *sep = "";
2267 size_t len = 0;
2268
2269 if (test_bit(Faulty, &rdev->flags)) {
2270 len+= sprintf(page+len, "%sfaulty",sep);
2271 sep = ",";
2272 }
2273 if (test_bit(In_sync, &rdev->flags)) {
2274 len += sprintf(page+len, "%sin_sync",sep);
2275 sep = ",";
2276 }
2277 if (test_bit(WriteMostly, &rdev->flags)) {
2278 len += sprintf(page+len, "%swrite_mostly",sep);
2279 sep = ",";
2280 }
2281 if (test_bit(Blocked, &rdev->flags)) {
2282 len += sprintf(page+len, "%sblocked", sep);
2283 sep = ",";
2284 }
2285 if (!test_bit(Faulty, &rdev->flags) &&
2286 !test_bit(In_sync, &rdev->flags)) {
2287 len += sprintf(page+len, "%sspare", sep);
2288 sep = ",";
2289 }
2290 return len+sprintf(page+len, "\n");
2291 }
2292
2293 static ssize_t
2294 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2295 {
2296 /* can write
2297 * faulty - simulates and error
2298 * remove - disconnects the device
2299 * writemostly - sets write_mostly
2300 * -writemostly - clears write_mostly
2301 * blocked - sets the Blocked flag
2302 * -blocked - clears the Blocked flag
2303 * insync - sets Insync providing device isn't active
2304 */
2305 int err = -EINVAL;
2306 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2307 md_error(rdev->mddev, rdev);
2308 err = 0;
2309 } else if (cmd_match(buf, "remove")) {
2310 if (rdev->raid_disk >= 0)
2311 err = -EBUSY;
2312 else {
2313 mddev_t *mddev = rdev->mddev;
2314 kick_rdev_from_array(rdev);
2315 if (mddev->pers)
2316 md_update_sb(mddev, 1);
2317 md_new_event(mddev);
2318 err = 0;
2319 }
2320 } else if (cmd_match(buf, "writemostly")) {
2321 set_bit(WriteMostly, &rdev->flags);
2322 err = 0;
2323 } else if (cmd_match(buf, "-writemostly")) {
2324 clear_bit(WriteMostly, &rdev->flags);
2325 err = 0;
2326 } else if (cmd_match(buf, "blocked")) {
2327 set_bit(Blocked, &rdev->flags);
2328 err = 0;
2329 } else if (cmd_match(buf, "-blocked")) {
2330 clear_bit(Blocked, &rdev->flags);
2331 wake_up(&rdev->blocked_wait);
2332 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2333 md_wakeup_thread(rdev->mddev->thread);
2334
2335 err = 0;
2336 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2337 set_bit(In_sync, &rdev->flags);
2338 err = 0;
2339 }
2340 if (!err)
2341 sysfs_notify_dirent_safe(rdev->sysfs_state);
2342 return err ? err : len;
2343 }
2344 static struct rdev_sysfs_entry rdev_state =
2345 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2346
2347 static ssize_t
2348 errors_show(mdk_rdev_t *rdev, char *page)
2349 {
2350 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2351 }
2352
2353 static ssize_t
2354 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2355 {
2356 char *e;
2357 unsigned long n = simple_strtoul(buf, &e, 10);
2358 if (*buf && (*e == 0 || *e == '\n')) {
2359 atomic_set(&rdev->corrected_errors, n);
2360 return len;
2361 }
2362 return -EINVAL;
2363 }
2364 static struct rdev_sysfs_entry rdev_errors =
2365 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2366
2367 static ssize_t
2368 slot_show(mdk_rdev_t *rdev, char *page)
2369 {
2370 if (rdev->raid_disk < 0)
2371 return sprintf(page, "none\n");
2372 else
2373 return sprintf(page, "%d\n", rdev->raid_disk);
2374 }
2375
2376 static ssize_t
2377 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2378 {
2379 char *e;
2380 int err;
2381 char nm[20];
2382 int slot = simple_strtoul(buf, &e, 10);
2383 if (strncmp(buf, "none", 4)==0)
2384 slot = -1;
2385 else if (e==buf || (*e && *e!= '\n'))
2386 return -EINVAL;
2387 if (rdev->mddev->pers && slot == -1) {
2388 /* Setting 'slot' on an active array requires also
2389 * updating the 'rd%d' link, and communicating
2390 * with the personality with ->hot_*_disk.
2391 * For now we only support removing
2392 * failed/spare devices. This normally happens automatically,
2393 * but not when the metadata is externally managed.
2394 */
2395 if (rdev->raid_disk == -1)
2396 return -EEXIST;
2397 /* personality does all needed checks */
2398 if (rdev->mddev->pers->hot_add_disk == NULL)
2399 return -EINVAL;
2400 err = rdev->mddev->pers->
2401 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2402 if (err)
2403 return err;
2404 sprintf(nm, "rd%d", rdev->raid_disk);
2405 sysfs_remove_link(&rdev->mddev->kobj, nm);
2406 rdev->raid_disk = -1;
2407 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2408 md_wakeup_thread(rdev->mddev->thread);
2409 } else if (rdev->mddev->pers) {
2410 mdk_rdev_t *rdev2;
2411 /* Activating a spare .. or possibly reactivating
2412 * if we ever get bitmaps working here.
2413 */
2414
2415 if (rdev->raid_disk != -1)
2416 return -EBUSY;
2417
2418 if (rdev->mddev->pers->hot_add_disk == NULL)
2419 return -EINVAL;
2420
2421 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2422 if (rdev2->raid_disk == slot)
2423 return -EEXIST;
2424
2425 rdev->raid_disk = slot;
2426 if (test_bit(In_sync, &rdev->flags))
2427 rdev->saved_raid_disk = slot;
2428 else
2429 rdev->saved_raid_disk = -1;
2430 err = rdev->mddev->pers->
2431 hot_add_disk(rdev->mddev, rdev);
2432 if (err) {
2433 rdev->raid_disk = -1;
2434 return err;
2435 } else
2436 sysfs_notify_dirent_safe(rdev->sysfs_state);
2437 sprintf(nm, "rd%d", rdev->raid_disk);
2438 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2439 /* failure here is OK */;
2440 /* don't wakeup anyone, leave that to userspace. */
2441 } else {
2442 if (slot >= rdev->mddev->raid_disks)
2443 return -ENOSPC;
2444 rdev->raid_disk = slot;
2445 /* assume it is working */
2446 clear_bit(Faulty, &rdev->flags);
2447 clear_bit(WriteMostly, &rdev->flags);
2448 set_bit(In_sync, &rdev->flags);
2449 sysfs_notify_dirent_safe(rdev->sysfs_state);
2450 }
2451 return len;
2452 }
2453
2454
2455 static struct rdev_sysfs_entry rdev_slot =
2456 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2457
2458 static ssize_t
2459 offset_show(mdk_rdev_t *rdev, char *page)
2460 {
2461 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2462 }
2463
2464 static ssize_t
2465 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2466 {
2467 char *e;
2468 unsigned long long offset = simple_strtoull(buf, &e, 10);
2469 if (e==buf || (*e && *e != '\n'))
2470 return -EINVAL;
2471 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2472 return -EBUSY;
2473 if (rdev->sectors && rdev->mddev->external)
2474 /* Must set offset before size, so overlap checks
2475 * can be sane */
2476 return -EBUSY;
2477 rdev->data_offset = offset;
2478 return len;
2479 }
2480
2481 static struct rdev_sysfs_entry rdev_offset =
2482 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2483
2484 static ssize_t
2485 rdev_size_show(mdk_rdev_t *rdev, char *page)
2486 {
2487 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2488 }
2489
2490 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2491 {
2492 /* check if two start/length pairs overlap */
2493 if (s1+l1 <= s2)
2494 return 0;
2495 if (s2+l2 <= s1)
2496 return 0;
2497 return 1;
2498 }
2499
2500 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2501 {
2502 unsigned long long blocks;
2503 sector_t new;
2504
2505 if (strict_strtoull(buf, 10, &blocks) < 0)
2506 return -EINVAL;
2507
2508 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2509 return -EINVAL; /* sector conversion overflow */
2510
2511 new = blocks * 2;
2512 if (new != blocks * 2)
2513 return -EINVAL; /* unsigned long long to sector_t overflow */
2514
2515 *sectors = new;
2516 return 0;
2517 }
2518
2519 static ssize_t
2520 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2521 {
2522 mddev_t *my_mddev = rdev->mddev;
2523 sector_t oldsectors = rdev->sectors;
2524 sector_t sectors;
2525
2526 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2527 return -EINVAL;
2528 if (my_mddev->pers && rdev->raid_disk >= 0) {
2529 if (my_mddev->persistent) {
2530 sectors = super_types[my_mddev->major_version].
2531 rdev_size_change(rdev, sectors);
2532 if (!sectors)
2533 return -EBUSY;
2534 } else if (!sectors)
2535 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2536 rdev->data_offset;
2537 }
2538 if (sectors < my_mddev->dev_sectors)
2539 return -EINVAL; /* component must fit device */
2540
2541 rdev->sectors = sectors;
2542 if (sectors > oldsectors && my_mddev->external) {
2543 /* need to check that all other rdevs with the same ->bdev
2544 * do not overlap. We need to unlock the mddev to avoid
2545 * a deadlock. We have already changed rdev->sectors, and if
2546 * we have to change it back, we will have the lock again.
2547 */
2548 mddev_t *mddev;
2549 int overlap = 0;
2550 struct list_head *tmp;
2551
2552 mddev_unlock(my_mddev);
2553 for_each_mddev(mddev, tmp) {
2554 mdk_rdev_t *rdev2;
2555
2556 mddev_lock(mddev);
2557 list_for_each_entry(rdev2, &mddev->disks, same_set)
2558 if (test_bit(AllReserved, &rdev2->flags) ||
2559 (rdev->bdev == rdev2->bdev &&
2560 rdev != rdev2 &&
2561 overlaps(rdev->data_offset, rdev->sectors,
2562 rdev2->data_offset,
2563 rdev2->sectors))) {
2564 overlap = 1;
2565 break;
2566 }
2567 mddev_unlock(mddev);
2568 if (overlap) {
2569 mddev_put(mddev);
2570 break;
2571 }
2572 }
2573 mddev_lock(my_mddev);
2574 if (overlap) {
2575 /* Someone else could have slipped in a size
2576 * change here, but doing so is just silly.
2577 * We put oldsectors back because we *know* it is
2578 * safe, and trust userspace not to race with
2579 * itself
2580 */
2581 rdev->sectors = oldsectors;
2582 return -EBUSY;
2583 }
2584 }
2585 return len;
2586 }
2587
2588 static struct rdev_sysfs_entry rdev_size =
2589 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2590
2591
2592 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2593 {
2594 unsigned long long recovery_start = rdev->recovery_offset;
2595
2596 if (test_bit(In_sync, &rdev->flags) ||
2597 recovery_start == MaxSector)
2598 return sprintf(page, "none\n");
2599
2600 return sprintf(page, "%llu\n", recovery_start);
2601 }
2602
2603 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2604 {
2605 unsigned long long recovery_start;
2606
2607 if (cmd_match(buf, "none"))
2608 recovery_start = MaxSector;
2609 else if (strict_strtoull(buf, 10, &recovery_start))
2610 return -EINVAL;
2611
2612 if (rdev->mddev->pers &&
2613 rdev->raid_disk >= 0)
2614 return -EBUSY;
2615
2616 rdev->recovery_offset = recovery_start;
2617 if (recovery_start == MaxSector)
2618 set_bit(In_sync, &rdev->flags);
2619 else
2620 clear_bit(In_sync, &rdev->flags);
2621 return len;
2622 }
2623
2624 static struct rdev_sysfs_entry rdev_recovery_start =
2625 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2626
2627 static struct attribute *rdev_default_attrs[] = {
2628 &rdev_state.attr,
2629 &rdev_errors.attr,
2630 &rdev_slot.attr,
2631 &rdev_offset.attr,
2632 &rdev_size.attr,
2633 &rdev_recovery_start.attr,
2634 NULL,
2635 };
2636 static ssize_t
2637 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2638 {
2639 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2640 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2641 mddev_t *mddev = rdev->mddev;
2642 ssize_t rv;
2643
2644 if (!entry->show)
2645 return -EIO;
2646
2647 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2648 if (!rv) {
2649 if (rdev->mddev == NULL)
2650 rv = -EBUSY;
2651 else
2652 rv = entry->show(rdev, page);
2653 mddev_unlock(mddev);
2654 }
2655 return rv;
2656 }
2657
2658 static ssize_t
2659 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2660 const char *page, size_t length)
2661 {
2662 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2663 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2664 ssize_t rv;
2665 mddev_t *mddev = rdev->mddev;
2666
2667 if (!entry->store)
2668 return -EIO;
2669 if (!capable(CAP_SYS_ADMIN))
2670 return -EACCES;
2671 rv = mddev ? mddev_lock(mddev): -EBUSY;
2672 if (!rv) {
2673 if (rdev->mddev == NULL)
2674 rv = -EBUSY;
2675 else
2676 rv = entry->store(rdev, page, length);
2677 mddev_unlock(mddev);
2678 }
2679 return rv;
2680 }
2681
2682 static void rdev_free(struct kobject *ko)
2683 {
2684 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2685 kfree(rdev);
2686 }
2687 static const struct sysfs_ops rdev_sysfs_ops = {
2688 .show = rdev_attr_show,
2689 .store = rdev_attr_store,
2690 };
2691 static struct kobj_type rdev_ktype = {
2692 .release = rdev_free,
2693 .sysfs_ops = &rdev_sysfs_ops,
2694 .default_attrs = rdev_default_attrs,
2695 };
2696
2697 void md_rdev_init(mdk_rdev_t *rdev)
2698 {
2699 rdev->desc_nr = -1;
2700 rdev->saved_raid_disk = -1;
2701 rdev->raid_disk = -1;
2702 rdev->flags = 0;
2703 rdev->data_offset = 0;
2704 rdev->sb_events = 0;
2705 rdev->last_read_error.tv_sec = 0;
2706 rdev->last_read_error.tv_nsec = 0;
2707 atomic_set(&rdev->nr_pending, 0);
2708 atomic_set(&rdev->read_errors, 0);
2709 atomic_set(&rdev->corrected_errors, 0);
2710
2711 INIT_LIST_HEAD(&rdev->same_set);
2712 init_waitqueue_head(&rdev->blocked_wait);
2713 }
2714 EXPORT_SYMBOL_GPL(md_rdev_init);
2715 /*
2716 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2717 *
2718 * mark the device faulty if:
2719 *
2720 * - the device is nonexistent (zero size)
2721 * - the device has no valid superblock
2722 *
2723 * a faulty rdev _never_ has rdev->sb set.
2724 */
2725 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2726 {
2727 char b[BDEVNAME_SIZE];
2728 int err;
2729 mdk_rdev_t *rdev;
2730 sector_t size;
2731
2732 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2733 if (!rdev) {
2734 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2735 return ERR_PTR(-ENOMEM);
2736 }
2737
2738 md_rdev_init(rdev);
2739 if ((err = alloc_disk_sb(rdev)))
2740 goto abort_free;
2741
2742 err = lock_rdev(rdev, newdev, super_format == -2);
2743 if (err)
2744 goto abort_free;
2745
2746 kobject_init(&rdev->kobj, &rdev_ktype);
2747
2748 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2749 if (!size) {
2750 printk(KERN_WARNING
2751 "md: %s has zero or unknown size, marking faulty!\n",
2752 bdevname(rdev->bdev,b));
2753 err = -EINVAL;
2754 goto abort_free;
2755 }
2756
2757 if (super_format >= 0) {
2758 err = super_types[super_format].
2759 load_super(rdev, NULL, super_minor);
2760 if (err == -EINVAL) {
2761 printk(KERN_WARNING
2762 "md: %s does not have a valid v%d.%d "
2763 "superblock, not importing!\n",
2764 bdevname(rdev->bdev,b),
2765 super_format, super_minor);
2766 goto abort_free;
2767 }
2768 if (err < 0) {
2769 printk(KERN_WARNING
2770 "md: could not read %s's sb, not importing!\n",
2771 bdevname(rdev->bdev,b));
2772 goto abort_free;
2773 }
2774 }
2775
2776 return rdev;
2777
2778 abort_free:
2779 if (rdev->sb_page) {
2780 if (rdev->bdev)
2781 unlock_rdev(rdev);
2782 free_disk_sb(rdev);
2783 }
2784 kfree(rdev);
2785 return ERR_PTR(err);
2786 }
2787
2788 /*
2789 * Check a full RAID array for plausibility
2790 */
2791
2792
2793 static void analyze_sbs(mddev_t * mddev)
2794 {
2795 int i;
2796 mdk_rdev_t *rdev, *freshest, *tmp;
2797 char b[BDEVNAME_SIZE];
2798
2799 freshest = NULL;
2800 rdev_for_each(rdev, tmp, mddev)
2801 switch (super_types[mddev->major_version].
2802 load_super(rdev, freshest, mddev->minor_version)) {
2803 case 1:
2804 freshest = rdev;
2805 break;
2806 case 0:
2807 break;
2808 default:
2809 printk( KERN_ERR \
2810 "md: fatal superblock inconsistency in %s"
2811 " -- removing from array\n",
2812 bdevname(rdev->bdev,b));
2813 kick_rdev_from_array(rdev);
2814 }
2815
2816
2817 super_types[mddev->major_version].
2818 validate_super(mddev, freshest);
2819
2820 i = 0;
2821 rdev_for_each(rdev, tmp, mddev) {
2822 if (mddev->max_disks &&
2823 (rdev->desc_nr >= mddev->max_disks ||
2824 i > mddev->max_disks)) {
2825 printk(KERN_WARNING
2826 "md: %s: %s: only %d devices permitted\n",
2827 mdname(mddev), bdevname(rdev->bdev, b),
2828 mddev->max_disks);
2829 kick_rdev_from_array(rdev);
2830 continue;
2831 }
2832 if (rdev != freshest)
2833 if (super_types[mddev->major_version].
2834 validate_super(mddev, rdev)) {
2835 printk(KERN_WARNING "md: kicking non-fresh %s"
2836 " from array!\n",
2837 bdevname(rdev->bdev,b));
2838 kick_rdev_from_array(rdev);
2839 continue;
2840 }
2841 if (mddev->level == LEVEL_MULTIPATH) {
2842 rdev->desc_nr = i++;
2843 rdev->raid_disk = rdev->desc_nr;
2844 set_bit(In_sync, &rdev->flags);
2845 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2846 rdev->raid_disk = -1;
2847 clear_bit(In_sync, &rdev->flags);
2848 }
2849 }
2850 }
2851
2852 /* Read a fixed-point number.
2853 * Numbers in sysfs attributes should be in "standard" units where
2854 * possible, so time should be in seconds.
2855 * However we internally use a a much smaller unit such as
2856 * milliseconds or jiffies.
2857 * This function takes a decimal number with a possible fractional
2858 * component, and produces an integer which is the result of
2859 * multiplying that number by 10^'scale'.
2860 * all without any floating-point arithmetic.
2861 */
2862 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2863 {
2864 unsigned long result = 0;
2865 long decimals = -1;
2866 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2867 if (*cp == '.')
2868 decimals = 0;
2869 else if (decimals < scale) {
2870 unsigned int value;
2871 value = *cp - '0';
2872 result = result * 10 + value;
2873 if (decimals >= 0)
2874 decimals++;
2875 }
2876 cp++;
2877 }
2878 if (*cp == '\n')
2879 cp++;
2880 if (*cp)
2881 return -EINVAL;
2882 if (decimals < 0)
2883 decimals = 0;
2884 while (decimals < scale) {
2885 result *= 10;
2886 decimals ++;
2887 }
2888 *res = result;
2889 return 0;
2890 }
2891
2892
2893 static void md_safemode_timeout(unsigned long data);
2894
2895 static ssize_t
2896 safe_delay_show(mddev_t *mddev, char *page)
2897 {
2898 int msec = (mddev->safemode_delay*1000)/HZ;
2899 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2900 }
2901 static ssize_t
2902 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2903 {
2904 unsigned long msec;
2905
2906 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2907 return -EINVAL;
2908 if (msec == 0)
2909 mddev->safemode_delay = 0;
2910 else {
2911 unsigned long old_delay = mddev->safemode_delay;
2912 mddev->safemode_delay = (msec*HZ)/1000;
2913 if (mddev->safemode_delay == 0)
2914 mddev->safemode_delay = 1;
2915 if (mddev->safemode_delay < old_delay)
2916 md_safemode_timeout((unsigned long)mddev);
2917 }
2918 return len;
2919 }
2920 static struct md_sysfs_entry md_safe_delay =
2921 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2922
2923 static ssize_t
2924 level_show(mddev_t *mddev, char *page)
2925 {
2926 struct mdk_personality *p = mddev->pers;
2927 if (p)
2928 return sprintf(page, "%s\n", p->name);
2929 else if (mddev->clevel[0])
2930 return sprintf(page, "%s\n", mddev->clevel);
2931 else if (mddev->level != LEVEL_NONE)
2932 return sprintf(page, "%d\n", mddev->level);
2933 else
2934 return 0;
2935 }
2936
2937 static ssize_t
2938 level_store(mddev_t *mddev, const char *buf, size_t len)
2939 {
2940 char clevel[16];
2941 ssize_t rv = len;
2942 struct mdk_personality *pers;
2943 long level;
2944 void *priv;
2945 mdk_rdev_t *rdev;
2946
2947 if (mddev->pers == NULL) {
2948 if (len == 0)
2949 return 0;
2950 if (len >= sizeof(mddev->clevel))
2951 return -ENOSPC;
2952 strncpy(mddev->clevel, buf, len);
2953 if (mddev->clevel[len-1] == '\n')
2954 len--;
2955 mddev->clevel[len] = 0;
2956 mddev->level = LEVEL_NONE;
2957 return rv;
2958 }
2959
2960 /* request to change the personality. Need to ensure:
2961 * - array is not engaged in resync/recovery/reshape
2962 * - old personality can be suspended
2963 * - new personality will access other array.
2964 */
2965
2966 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2967 return -EBUSY;
2968
2969 if (!mddev->pers->quiesce) {
2970 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2971 mdname(mddev), mddev->pers->name);
2972 return -EINVAL;
2973 }
2974
2975 /* Now find the new personality */
2976 if (len == 0 || len >= sizeof(clevel))
2977 return -EINVAL;
2978 strncpy(clevel, buf, len);
2979 if (clevel[len-1] == '\n')
2980 len--;
2981 clevel[len] = 0;
2982 if (strict_strtol(clevel, 10, &level))
2983 level = LEVEL_NONE;
2984
2985 if (request_module("md-%s", clevel) != 0)
2986 request_module("md-level-%s", clevel);
2987 spin_lock(&pers_lock);
2988 pers = find_pers(level, clevel);
2989 if (!pers || !try_module_get(pers->owner)) {
2990 spin_unlock(&pers_lock);
2991 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
2992 return -EINVAL;
2993 }
2994 spin_unlock(&pers_lock);
2995
2996 if (pers == mddev->pers) {
2997 /* Nothing to do! */
2998 module_put(pers->owner);
2999 return rv;
3000 }
3001 if (!pers->takeover) {
3002 module_put(pers->owner);
3003 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3004 mdname(mddev), clevel);
3005 return -EINVAL;
3006 }
3007
3008 list_for_each_entry(rdev, &mddev->disks, same_set)
3009 rdev->new_raid_disk = rdev->raid_disk;
3010
3011 /* ->takeover must set new_* and/or delta_disks
3012 * if it succeeds, and may set them when it fails.
3013 */
3014 priv = pers->takeover(mddev);
3015 if (IS_ERR(priv)) {
3016 mddev->new_level = mddev->level;
3017 mddev->new_layout = mddev->layout;
3018 mddev->new_chunk_sectors = mddev->chunk_sectors;
3019 mddev->raid_disks -= mddev->delta_disks;
3020 mddev->delta_disks = 0;
3021 module_put(pers->owner);
3022 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3023 mdname(mddev), clevel);
3024 return PTR_ERR(priv);
3025 }
3026
3027 /* Looks like we have a winner */
3028 mddev_suspend(mddev);
3029 mddev->pers->stop(mddev);
3030
3031 if (mddev->pers->sync_request == NULL &&
3032 pers->sync_request != NULL) {
3033 /* need to add the md_redundancy_group */
3034 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3035 printk(KERN_WARNING
3036 "md: cannot register extra attributes for %s\n",
3037 mdname(mddev));
3038 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3039 }
3040 if (mddev->pers->sync_request != NULL &&
3041 pers->sync_request == NULL) {
3042 /* need to remove the md_redundancy_group */
3043 if (mddev->to_remove == NULL)
3044 mddev->to_remove = &md_redundancy_group;
3045 }
3046
3047 if (mddev->pers->sync_request == NULL &&
3048 mddev->external) {
3049 /* We are converting from a no-redundancy array
3050 * to a redundancy array and metadata is managed
3051 * externally so we need to be sure that writes
3052 * won't block due to a need to transition
3053 * clean->dirty
3054 * until external management is started.
3055 */
3056 mddev->in_sync = 0;
3057 mddev->safemode_delay = 0;
3058 mddev->safemode = 0;
3059 }
3060
3061 list_for_each_entry(rdev, &mddev->disks, same_set) {
3062 char nm[20];
3063 if (rdev->raid_disk < 0)
3064 continue;
3065 if (rdev->new_raid_disk > mddev->raid_disks)
3066 rdev->new_raid_disk = -1;
3067 if (rdev->new_raid_disk == rdev->raid_disk)
3068 continue;
3069 sprintf(nm, "rd%d", rdev->raid_disk);
3070 sysfs_remove_link(&mddev->kobj, nm);
3071 }
3072 list_for_each_entry(rdev, &mddev->disks, same_set) {
3073 if (rdev->raid_disk < 0)
3074 continue;
3075 if (rdev->new_raid_disk == rdev->raid_disk)
3076 continue;
3077 rdev->raid_disk = rdev->new_raid_disk;
3078 if (rdev->raid_disk < 0)
3079 clear_bit(In_sync, &rdev->flags);
3080 else {
3081 char nm[20];
3082 sprintf(nm, "rd%d", rdev->raid_disk);
3083 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3084 printk("md: cannot register %s for %s after level change\n",
3085 nm, mdname(mddev));
3086 }
3087 }
3088
3089 module_put(mddev->pers->owner);
3090 mddev->pers = pers;
3091 mddev->private = priv;
3092 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3093 mddev->level = mddev->new_level;
3094 mddev->layout = mddev->new_layout;
3095 mddev->chunk_sectors = mddev->new_chunk_sectors;
3096 mddev->delta_disks = 0;
3097 if (mddev->pers->sync_request == NULL) {
3098 /* this is now an array without redundancy, so
3099 * it must always be in_sync
3100 */
3101 mddev->in_sync = 1;
3102 del_timer_sync(&mddev->safemode_timer);
3103 }
3104 pers->run(mddev);
3105 mddev_resume(mddev);
3106 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3107 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3108 md_wakeup_thread(mddev->thread);
3109 sysfs_notify(&mddev->kobj, NULL, "level");
3110 md_new_event(mddev);
3111 return rv;
3112 }
3113
3114 static struct md_sysfs_entry md_level =
3115 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3116
3117
3118 static ssize_t
3119 layout_show(mddev_t *mddev, char *page)
3120 {
3121 /* just a number, not meaningful for all levels */
3122 if (mddev->reshape_position != MaxSector &&
3123 mddev->layout != mddev->new_layout)
3124 return sprintf(page, "%d (%d)\n",
3125 mddev->new_layout, mddev->layout);
3126 return sprintf(page, "%d\n", mddev->layout);
3127 }
3128
3129 static ssize_t
3130 layout_store(mddev_t *mddev, const char *buf, size_t len)
3131 {
3132 char *e;
3133 unsigned long n = simple_strtoul(buf, &e, 10);
3134
3135 if (!*buf || (*e && *e != '\n'))
3136 return -EINVAL;
3137
3138 if (mddev->pers) {
3139 int err;
3140 if (mddev->pers->check_reshape == NULL)
3141 return -EBUSY;
3142 mddev->new_layout = n;
3143 err = mddev->pers->check_reshape(mddev);
3144 if (err) {
3145 mddev->new_layout = mddev->layout;
3146 return err;
3147 }
3148 } else {
3149 mddev->new_layout = n;
3150 if (mddev->reshape_position == MaxSector)
3151 mddev->layout = n;
3152 }
3153 return len;
3154 }
3155 static struct md_sysfs_entry md_layout =
3156 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3157
3158
3159 static ssize_t
3160 raid_disks_show(mddev_t *mddev, char *page)
3161 {
3162 if (mddev->raid_disks == 0)
3163 return 0;
3164 if (mddev->reshape_position != MaxSector &&
3165 mddev->delta_disks != 0)
3166 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3167 mddev->raid_disks - mddev->delta_disks);
3168 return sprintf(page, "%d\n", mddev->raid_disks);
3169 }
3170
3171 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3172
3173 static ssize_t
3174 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3175 {
3176 char *e;
3177 int rv = 0;
3178 unsigned long n = simple_strtoul(buf, &e, 10);
3179
3180 if (!*buf || (*e && *e != '\n'))
3181 return -EINVAL;
3182
3183 if (mddev->pers)
3184 rv = update_raid_disks(mddev, n);
3185 else if (mddev->reshape_position != MaxSector) {
3186 int olddisks = mddev->raid_disks - mddev->delta_disks;
3187 mddev->delta_disks = n - olddisks;
3188 mddev->raid_disks = n;
3189 } else
3190 mddev->raid_disks = n;
3191 return rv ? rv : len;
3192 }
3193 static struct md_sysfs_entry md_raid_disks =
3194 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3195
3196 static ssize_t
3197 chunk_size_show(mddev_t *mddev, char *page)
3198 {
3199 if (mddev->reshape_position != MaxSector &&
3200 mddev->chunk_sectors != mddev->new_chunk_sectors)
3201 return sprintf(page, "%d (%d)\n",
3202 mddev->new_chunk_sectors << 9,
3203 mddev->chunk_sectors << 9);
3204 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3205 }
3206
3207 static ssize_t
3208 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3209 {
3210 char *e;
3211 unsigned long n = simple_strtoul(buf, &e, 10);
3212
3213 if (!*buf || (*e && *e != '\n'))
3214 return -EINVAL;
3215
3216 if (mddev->pers) {
3217 int err;
3218 if (mddev->pers->check_reshape == NULL)
3219 return -EBUSY;
3220 mddev->new_chunk_sectors = n >> 9;
3221 err = mddev->pers->check_reshape(mddev);
3222 if (err) {
3223 mddev->new_chunk_sectors = mddev->chunk_sectors;
3224 return err;
3225 }
3226 } else {
3227 mddev->new_chunk_sectors = n >> 9;
3228 if (mddev->reshape_position == MaxSector)
3229 mddev->chunk_sectors = n >> 9;
3230 }
3231 return len;
3232 }
3233 static struct md_sysfs_entry md_chunk_size =
3234 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3235
3236 static ssize_t
3237 resync_start_show(mddev_t *mddev, char *page)
3238 {
3239 if (mddev->recovery_cp == MaxSector)
3240 return sprintf(page, "none\n");
3241 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3242 }
3243
3244 static ssize_t
3245 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3246 {
3247 char *e;
3248 unsigned long long n = simple_strtoull(buf, &e, 10);
3249
3250 if (mddev->pers)
3251 return -EBUSY;
3252 if (cmd_match(buf, "none"))
3253 n = MaxSector;
3254 else if (!*buf || (*e && *e != '\n'))
3255 return -EINVAL;
3256
3257 mddev->recovery_cp = n;
3258 return len;
3259 }
3260 static struct md_sysfs_entry md_resync_start =
3261 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3262
3263 /*
3264 * The array state can be:
3265 *
3266 * clear
3267 * No devices, no size, no level
3268 * Equivalent to STOP_ARRAY ioctl
3269 * inactive
3270 * May have some settings, but array is not active
3271 * all IO results in error
3272 * When written, doesn't tear down array, but just stops it
3273 * suspended (not supported yet)
3274 * All IO requests will block. The array can be reconfigured.
3275 * Writing this, if accepted, will block until array is quiescent
3276 * readonly
3277 * no resync can happen. no superblocks get written.
3278 * write requests fail
3279 * read-auto
3280 * like readonly, but behaves like 'clean' on a write request.
3281 *
3282 * clean - no pending writes, but otherwise active.
3283 * When written to inactive array, starts without resync
3284 * If a write request arrives then
3285 * if metadata is known, mark 'dirty' and switch to 'active'.
3286 * if not known, block and switch to write-pending
3287 * If written to an active array that has pending writes, then fails.
3288 * active
3289 * fully active: IO and resync can be happening.
3290 * When written to inactive array, starts with resync
3291 *
3292 * write-pending
3293 * clean, but writes are blocked waiting for 'active' to be written.
3294 *
3295 * active-idle
3296 * like active, but no writes have been seen for a while (100msec).
3297 *
3298 */
3299 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3300 write_pending, active_idle, bad_word};
3301 static char *array_states[] = {
3302 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3303 "write-pending", "active-idle", NULL };
3304
3305 static int match_word(const char *word, char **list)
3306 {
3307 int n;
3308 for (n=0; list[n]; n++)
3309 if (cmd_match(word, list[n]))
3310 break;
3311 return n;
3312 }
3313
3314 static ssize_t
3315 array_state_show(mddev_t *mddev, char *page)
3316 {
3317 enum array_state st = inactive;
3318
3319 if (mddev->pers)
3320 switch(mddev->ro) {
3321 case 1:
3322 st = readonly;
3323 break;
3324 case 2:
3325 st = read_auto;
3326 break;
3327 case 0:
3328 if (mddev->in_sync)
3329 st = clean;
3330 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3331 st = write_pending;
3332 else if (mddev->safemode)
3333 st = active_idle;
3334 else
3335 st = active;
3336 }
3337 else {
3338 if (list_empty(&mddev->disks) &&
3339 mddev->raid_disks == 0 &&
3340 mddev->dev_sectors == 0)
3341 st = clear;
3342 else
3343 st = inactive;
3344 }
3345 return sprintf(page, "%s\n", array_states[st]);
3346 }
3347
3348 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3349 static int md_set_readonly(mddev_t * mddev, int is_open);
3350 static int do_md_run(mddev_t * mddev);
3351 static int restart_array(mddev_t *mddev);
3352
3353 static ssize_t
3354 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3355 {
3356 int err = -EINVAL;
3357 enum array_state st = match_word(buf, array_states);
3358 switch(st) {
3359 case bad_word:
3360 break;
3361 case clear:
3362 /* stopping an active array */
3363 if (atomic_read(&mddev->openers) > 0)
3364 return -EBUSY;
3365 err = do_md_stop(mddev, 0, 0);
3366 break;
3367 case inactive:
3368 /* stopping an active array */
3369 if (mddev->pers) {
3370 if (atomic_read(&mddev->openers) > 0)
3371 return -EBUSY;
3372 err = do_md_stop(mddev, 2, 0);
3373 } else
3374 err = 0; /* already inactive */
3375 break;
3376 case suspended:
3377 break; /* not supported yet */
3378 case readonly:
3379 if (mddev->pers)
3380 err = md_set_readonly(mddev, 0);
3381 else {
3382 mddev->ro = 1;
3383 set_disk_ro(mddev->gendisk, 1);
3384 err = do_md_run(mddev);
3385 }
3386 break;
3387 case read_auto:
3388 if (mddev->pers) {
3389 if (mddev->ro == 0)
3390 err = md_set_readonly(mddev, 0);
3391 else if (mddev->ro == 1)
3392 err = restart_array(mddev);
3393 if (err == 0) {
3394 mddev->ro = 2;
3395 set_disk_ro(mddev->gendisk, 0);
3396 }
3397 } else {
3398 mddev->ro = 2;
3399 err = do_md_run(mddev);
3400 }
3401 break;
3402 case clean:
3403 if (mddev->pers) {
3404 restart_array(mddev);
3405 spin_lock_irq(&mddev->write_lock);
3406 if (atomic_read(&mddev->writes_pending) == 0) {
3407 if (mddev->in_sync == 0) {
3408 mddev->in_sync = 1;
3409 if (mddev->safemode == 1)
3410 mddev->safemode = 0;
3411 if (mddev->persistent)
3412 set_bit(MD_CHANGE_CLEAN,
3413 &mddev->flags);
3414 }
3415 err = 0;
3416 } else
3417 err = -EBUSY;
3418 spin_unlock_irq(&mddev->write_lock);
3419 } else
3420 err = -EINVAL;
3421 break;
3422 case active:
3423 if (mddev->pers) {
3424 restart_array(mddev);
3425 if (mddev->external)
3426 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3427 wake_up(&mddev->sb_wait);
3428 err = 0;
3429 } else {
3430 mddev->ro = 0;
3431 set_disk_ro(mddev->gendisk, 0);
3432 err = do_md_run(mddev);
3433 }
3434 break;
3435 case write_pending:
3436 case active_idle:
3437 /* these cannot be set */
3438 break;
3439 }
3440 if (err)
3441 return err;
3442 else {
3443 sysfs_notify_dirent_safe(mddev->sysfs_state);
3444 return len;
3445 }
3446 }
3447 static struct md_sysfs_entry md_array_state =
3448 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3449
3450 static ssize_t
3451 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3452 return sprintf(page, "%d\n",
3453 atomic_read(&mddev->max_corr_read_errors));
3454 }
3455
3456 static ssize_t
3457 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3458 {
3459 char *e;
3460 unsigned long n = simple_strtoul(buf, &e, 10);
3461
3462 if (*buf && (*e == 0 || *e == '\n')) {
3463 atomic_set(&mddev->max_corr_read_errors, n);
3464 return len;
3465 }
3466 return -EINVAL;
3467 }
3468
3469 static struct md_sysfs_entry max_corr_read_errors =
3470 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3471 max_corrected_read_errors_store);
3472
3473 static ssize_t
3474 null_show(mddev_t *mddev, char *page)
3475 {
3476 return -EINVAL;
3477 }
3478
3479 static ssize_t
3480 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3481 {
3482 /* buf must be %d:%d\n? giving major and minor numbers */
3483 /* The new device is added to the array.
3484 * If the array has a persistent superblock, we read the
3485 * superblock to initialise info and check validity.
3486 * Otherwise, only checking done is that in bind_rdev_to_array,
3487 * which mainly checks size.
3488 */
3489 char *e;
3490 int major = simple_strtoul(buf, &e, 10);
3491 int minor;
3492 dev_t dev;
3493 mdk_rdev_t *rdev;
3494 int err;
3495
3496 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3497 return -EINVAL;
3498 minor = simple_strtoul(e+1, &e, 10);
3499 if (*e && *e != '\n')
3500 return -EINVAL;
3501 dev = MKDEV(major, minor);
3502 if (major != MAJOR(dev) ||
3503 minor != MINOR(dev))
3504 return -EOVERFLOW;
3505
3506
3507 if (mddev->persistent) {
3508 rdev = md_import_device(dev, mddev->major_version,
3509 mddev->minor_version);
3510 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3511 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3512 mdk_rdev_t, same_set);
3513 err = super_types[mddev->major_version]
3514 .load_super(rdev, rdev0, mddev->minor_version);
3515 if (err < 0)
3516 goto out;
3517 }
3518 } else if (mddev->external)
3519 rdev = md_import_device(dev, -2, -1);
3520 else
3521 rdev = md_import_device(dev, -1, -1);
3522
3523 if (IS_ERR(rdev))
3524 return PTR_ERR(rdev);
3525 err = bind_rdev_to_array(rdev, mddev);
3526 out:
3527 if (err)
3528 export_rdev(rdev);
3529 return err ? err : len;
3530 }
3531
3532 static struct md_sysfs_entry md_new_device =
3533 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3534
3535 static ssize_t
3536 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3537 {
3538 char *end;
3539 unsigned long chunk, end_chunk;
3540
3541 if (!mddev->bitmap)
3542 goto out;
3543 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3544 while (*buf) {
3545 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3546 if (buf == end) break;
3547 if (*end == '-') { /* range */
3548 buf = end + 1;
3549 end_chunk = simple_strtoul(buf, &end, 0);
3550 if (buf == end) break;
3551 }
3552 if (*end && !isspace(*end)) break;
3553 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3554 buf = skip_spaces(end);
3555 }
3556 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3557 out:
3558 return len;
3559 }
3560
3561 static struct md_sysfs_entry md_bitmap =
3562 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3563
3564 static ssize_t
3565 size_show(mddev_t *mddev, char *page)
3566 {
3567 return sprintf(page, "%llu\n",
3568 (unsigned long long)mddev->dev_sectors / 2);
3569 }
3570
3571 static int update_size(mddev_t *mddev, sector_t num_sectors);
3572
3573 static ssize_t
3574 size_store(mddev_t *mddev, const char *buf, size_t len)
3575 {
3576 /* If array is inactive, we can reduce the component size, but
3577 * not increase it (except from 0).
3578 * If array is active, we can try an on-line resize
3579 */
3580 sector_t sectors;
3581 int err = strict_blocks_to_sectors(buf, &sectors);
3582
3583 if (err < 0)
3584 return err;
3585 if (mddev->pers) {
3586 err = update_size(mddev, sectors);
3587 md_update_sb(mddev, 1);
3588 } else {
3589 if (mddev->dev_sectors == 0 ||
3590 mddev->dev_sectors > sectors)
3591 mddev->dev_sectors = sectors;
3592 else
3593 err = -ENOSPC;
3594 }
3595 return err ? err : len;
3596 }
3597
3598 static struct md_sysfs_entry md_size =
3599 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3600
3601
3602 /* Metdata version.
3603 * This is one of
3604 * 'none' for arrays with no metadata (good luck...)
3605 * 'external' for arrays with externally managed metadata,
3606 * or N.M for internally known formats
3607 */
3608 static ssize_t
3609 metadata_show(mddev_t *mddev, char *page)
3610 {
3611 if (mddev->persistent)
3612 return sprintf(page, "%d.%d\n",
3613 mddev->major_version, mddev->minor_version);
3614 else if (mddev->external)
3615 return sprintf(page, "external:%s\n", mddev->metadata_type);
3616 else
3617 return sprintf(page, "none\n");
3618 }
3619
3620 static ssize_t
3621 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3622 {
3623 int major, minor;
3624 char *e;
3625 /* Changing the details of 'external' metadata is
3626 * always permitted. Otherwise there must be
3627 * no devices attached to the array.
3628 */
3629 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3630 ;
3631 else if (!list_empty(&mddev->disks))
3632 return -EBUSY;
3633
3634 if (cmd_match(buf, "none")) {
3635 mddev->persistent = 0;
3636 mddev->external = 0;
3637 mddev->major_version = 0;
3638 mddev->minor_version = 90;
3639 return len;
3640 }
3641 if (strncmp(buf, "external:", 9) == 0) {
3642 size_t namelen = len-9;
3643 if (namelen >= sizeof(mddev->metadata_type))
3644 namelen = sizeof(mddev->metadata_type)-1;
3645 strncpy(mddev->metadata_type, buf+9, namelen);
3646 mddev->metadata_type[namelen] = 0;
3647 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3648 mddev->metadata_type[--namelen] = 0;
3649 mddev->persistent = 0;
3650 mddev->external = 1;
3651 mddev->major_version = 0;
3652 mddev->minor_version = 90;
3653 return len;
3654 }
3655 major = simple_strtoul(buf, &e, 10);
3656 if (e==buf || *e != '.')
3657 return -EINVAL;
3658 buf = e+1;
3659 minor = simple_strtoul(buf, &e, 10);
3660 if (e==buf || (*e && *e != '\n') )
3661 return -EINVAL;
3662 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3663 return -ENOENT;
3664 mddev->major_version = major;
3665 mddev->minor_version = minor;
3666 mddev->persistent = 1;
3667 mddev->external = 0;
3668 return len;
3669 }
3670
3671 static struct md_sysfs_entry md_metadata =
3672 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3673
3674 static ssize_t
3675 action_show(mddev_t *mddev, char *page)
3676 {
3677 char *type = "idle";
3678 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3679 type = "frozen";
3680 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3681 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3682 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3683 type = "reshape";
3684 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3685 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3686 type = "resync";
3687 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3688 type = "check";
3689 else
3690 type = "repair";
3691 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3692 type = "recover";
3693 }
3694 return sprintf(page, "%s\n", type);
3695 }
3696
3697 static ssize_t
3698 action_store(mddev_t *mddev, const char *page, size_t len)
3699 {
3700 if (!mddev->pers || !mddev->pers->sync_request)
3701 return -EINVAL;
3702
3703 if (cmd_match(page, "frozen"))
3704 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3705 else
3706 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3707
3708 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3709 if (mddev->sync_thread) {
3710 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3711 md_unregister_thread(mddev->sync_thread);
3712 mddev->sync_thread = NULL;
3713 mddev->recovery = 0;
3714 }
3715 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3716 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3717 return -EBUSY;
3718 else if (cmd_match(page, "resync"))
3719 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3720 else if (cmd_match(page, "recover")) {
3721 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3722 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3723 } else if (cmd_match(page, "reshape")) {
3724 int err;
3725 if (mddev->pers->start_reshape == NULL)
3726 return -EINVAL;
3727 err = mddev->pers->start_reshape(mddev);
3728 if (err)
3729 return err;
3730 sysfs_notify(&mddev->kobj, NULL, "degraded");
3731 } else {
3732 if (cmd_match(page, "check"))
3733 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3734 else if (!cmd_match(page, "repair"))
3735 return -EINVAL;
3736 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3737 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3738 }
3739 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3740 md_wakeup_thread(mddev->thread);
3741 sysfs_notify_dirent_safe(mddev->sysfs_action);
3742 return len;
3743 }
3744
3745 static ssize_t
3746 mismatch_cnt_show(mddev_t *mddev, char *page)
3747 {
3748 return sprintf(page, "%llu\n",
3749 (unsigned long long) mddev->resync_mismatches);
3750 }
3751
3752 static struct md_sysfs_entry md_scan_mode =
3753 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3754
3755
3756 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3757
3758 static ssize_t
3759 sync_min_show(mddev_t *mddev, char *page)
3760 {
3761 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3762 mddev->sync_speed_min ? "local": "system");
3763 }
3764
3765 static ssize_t
3766 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3767 {
3768 int min;
3769 char *e;
3770 if (strncmp(buf, "system", 6)==0) {
3771 mddev->sync_speed_min = 0;
3772 return len;
3773 }
3774 min = simple_strtoul(buf, &e, 10);
3775 if (buf == e || (*e && *e != '\n') || min <= 0)
3776 return -EINVAL;
3777 mddev->sync_speed_min = min;
3778 return len;
3779 }
3780
3781 static struct md_sysfs_entry md_sync_min =
3782 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3783
3784 static ssize_t
3785 sync_max_show(mddev_t *mddev, char *page)
3786 {
3787 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3788 mddev->sync_speed_max ? "local": "system");
3789 }
3790
3791 static ssize_t
3792 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3793 {
3794 int max;
3795 char *e;
3796 if (strncmp(buf, "system", 6)==0) {
3797 mddev->sync_speed_max = 0;
3798 return len;
3799 }
3800 max = simple_strtoul(buf, &e, 10);
3801 if (buf == e || (*e && *e != '\n') || max <= 0)
3802 return -EINVAL;
3803 mddev->sync_speed_max = max;
3804 return len;
3805 }
3806
3807 static struct md_sysfs_entry md_sync_max =
3808 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3809
3810 static ssize_t
3811 degraded_show(mddev_t *mddev, char *page)
3812 {
3813 return sprintf(page, "%d\n", mddev->degraded);
3814 }
3815 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3816
3817 static ssize_t
3818 sync_force_parallel_show(mddev_t *mddev, char *page)
3819 {
3820 return sprintf(page, "%d\n", mddev->parallel_resync);
3821 }
3822
3823 static ssize_t
3824 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3825 {
3826 long n;
3827
3828 if (strict_strtol(buf, 10, &n))
3829 return -EINVAL;
3830
3831 if (n != 0 && n != 1)
3832 return -EINVAL;
3833
3834 mddev->parallel_resync = n;
3835
3836 if (mddev->sync_thread)
3837 wake_up(&resync_wait);
3838
3839 return len;
3840 }
3841
3842 /* force parallel resync, even with shared block devices */
3843 static struct md_sysfs_entry md_sync_force_parallel =
3844 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3845 sync_force_parallel_show, sync_force_parallel_store);
3846
3847 static ssize_t
3848 sync_speed_show(mddev_t *mddev, char *page)
3849 {
3850 unsigned long resync, dt, db;
3851 if (mddev->curr_resync == 0)
3852 return sprintf(page, "none\n");
3853 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3854 dt = (jiffies - mddev->resync_mark) / HZ;
3855 if (!dt) dt++;
3856 db = resync - mddev->resync_mark_cnt;
3857 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3858 }
3859
3860 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3861
3862 static ssize_t
3863 sync_completed_show(mddev_t *mddev, char *page)
3864 {
3865 unsigned long max_sectors, resync;
3866
3867 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3868 return sprintf(page, "none\n");
3869
3870 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3871 max_sectors = mddev->resync_max_sectors;
3872 else
3873 max_sectors = mddev->dev_sectors;
3874
3875 resync = mddev->curr_resync_completed;
3876 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3877 }
3878
3879 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3880
3881 static ssize_t
3882 min_sync_show(mddev_t *mddev, char *page)
3883 {
3884 return sprintf(page, "%llu\n",
3885 (unsigned long long)mddev->resync_min);
3886 }
3887 static ssize_t
3888 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3889 {
3890 unsigned long long min;
3891 if (strict_strtoull(buf, 10, &min))
3892 return -EINVAL;
3893 if (min > mddev->resync_max)
3894 return -EINVAL;
3895 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3896 return -EBUSY;
3897
3898 /* Must be a multiple of chunk_size */
3899 if (mddev->chunk_sectors) {
3900 sector_t temp = min;
3901 if (sector_div(temp, mddev->chunk_sectors))
3902 return -EINVAL;
3903 }
3904 mddev->resync_min = min;
3905
3906 return len;
3907 }
3908
3909 static struct md_sysfs_entry md_min_sync =
3910 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3911
3912 static ssize_t
3913 max_sync_show(mddev_t *mddev, char *page)
3914 {
3915 if (mddev->resync_max == MaxSector)
3916 return sprintf(page, "max\n");
3917 else
3918 return sprintf(page, "%llu\n",
3919 (unsigned long long)mddev->resync_max);
3920 }
3921 static ssize_t
3922 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3923 {
3924 if (strncmp(buf, "max", 3) == 0)
3925 mddev->resync_max = MaxSector;
3926 else {
3927 unsigned long long max;
3928 if (strict_strtoull(buf, 10, &max))
3929 return -EINVAL;
3930 if (max < mddev->resync_min)
3931 return -EINVAL;
3932 if (max < mddev->resync_max &&
3933 mddev->ro == 0 &&
3934 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3935 return -EBUSY;
3936
3937 /* Must be a multiple of chunk_size */
3938 if (mddev->chunk_sectors) {
3939 sector_t temp = max;
3940 if (sector_div(temp, mddev->chunk_sectors))
3941 return -EINVAL;
3942 }
3943 mddev->resync_max = max;
3944 }
3945 wake_up(&mddev->recovery_wait);
3946 return len;
3947 }
3948
3949 static struct md_sysfs_entry md_max_sync =
3950 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3951
3952 static ssize_t
3953 suspend_lo_show(mddev_t *mddev, char *page)
3954 {
3955 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3956 }
3957
3958 static ssize_t
3959 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3960 {
3961 char *e;
3962 unsigned long long new = simple_strtoull(buf, &e, 10);
3963
3964 if (mddev->pers == NULL ||
3965 mddev->pers->quiesce == NULL)
3966 return -EINVAL;
3967 if (buf == e || (*e && *e != '\n'))
3968 return -EINVAL;
3969 if (new >= mddev->suspend_hi ||
3970 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3971 mddev->suspend_lo = new;
3972 mddev->pers->quiesce(mddev, 2);
3973 return len;
3974 } else
3975 return -EINVAL;
3976 }
3977 static struct md_sysfs_entry md_suspend_lo =
3978 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3979
3980
3981 static ssize_t
3982 suspend_hi_show(mddev_t *mddev, char *page)
3983 {
3984 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3985 }
3986
3987 static ssize_t
3988 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3989 {
3990 char *e;
3991 unsigned long long new = simple_strtoull(buf, &e, 10);
3992
3993 if (mddev->pers == NULL ||
3994 mddev->pers->quiesce == NULL)
3995 return -EINVAL;
3996 if (buf == e || (*e && *e != '\n'))
3997 return -EINVAL;
3998 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3999 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4000 mddev->suspend_hi = new;
4001 mddev->pers->quiesce(mddev, 1);
4002 mddev->pers->quiesce(mddev, 0);
4003 return len;
4004 } else
4005 return -EINVAL;
4006 }
4007 static struct md_sysfs_entry md_suspend_hi =
4008 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4009
4010 static ssize_t
4011 reshape_position_show(mddev_t *mddev, char *page)
4012 {
4013 if (mddev->reshape_position != MaxSector)
4014 return sprintf(page, "%llu\n",
4015 (unsigned long long)mddev->reshape_position);
4016 strcpy(page, "none\n");
4017 return 5;
4018 }
4019
4020 static ssize_t
4021 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4022 {
4023 char *e;
4024 unsigned long long new = simple_strtoull(buf, &e, 10);
4025 if (mddev->pers)
4026 return -EBUSY;
4027 if (buf == e || (*e && *e != '\n'))
4028 return -EINVAL;
4029 mddev->reshape_position = new;
4030 mddev->delta_disks = 0;
4031 mddev->new_level = mddev->level;
4032 mddev->new_layout = mddev->layout;
4033 mddev->new_chunk_sectors = mddev->chunk_sectors;
4034 return len;
4035 }
4036
4037 static struct md_sysfs_entry md_reshape_position =
4038 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4039 reshape_position_store);
4040
4041 static ssize_t
4042 array_size_show(mddev_t *mddev, char *page)
4043 {
4044 if (mddev->external_size)
4045 return sprintf(page, "%llu\n",
4046 (unsigned long long)mddev->array_sectors/2);
4047 else
4048 return sprintf(page, "default\n");
4049 }
4050
4051 static ssize_t
4052 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4053 {
4054 sector_t sectors;
4055
4056 if (strncmp(buf, "default", 7) == 0) {
4057 if (mddev->pers)
4058 sectors = mddev->pers->size(mddev, 0, 0);
4059 else
4060 sectors = mddev->array_sectors;
4061
4062 mddev->external_size = 0;
4063 } else {
4064 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4065 return -EINVAL;
4066 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4067 return -E2BIG;
4068
4069 mddev->external_size = 1;
4070 }
4071
4072 mddev->array_sectors = sectors;
4073 set_capacity(mddev->gendisk, mddev->array_sectors);
4074 if (mddev->pers)
4075 revalidate_disk(mddev->gendisk);
4076
4077 return len;
4078 }
4079
4080 static struct md_sysfs_entry md_array_size =
4081 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4082 array_size_store);
4083
4084 static struct attribute *md_default_attrs[] = {
4085 &md_level.attr,
4086 &md_layout.attr,
4087 &md_raid_disks.attr,
4088 &md_chunk_size.attr,
4089 &md_size.attr,
4090 &md_resync_start.attr,
4091 &md_metadata.attr,
4092 &md_new_device.attr,
4093 &md_safe_delay.attr,
4094 &md_array_state.attr,
4095 &md_reshape_position.attr,
4096 &md_array_size.attr,
4097 &max_corr_read_errors.attr,
4098 NULL,
4099 };
4100
4101 static struct attribute *md_redundancy_attrs[] = {
4102 &md_scan_mode.attr,
4103 &md_mismatches.attr,
4104 &md_sync_min.attr,
4105 &md_sync_max.attr,
4106 &md_sync_speed.attr,
4107 &md_sync_force_parallel.attr,
4108 &md_sync_completed.attr,
4109 &md_min_sync.attr,
4110 &md_max_sync.attr,
4111 &md_suspend_lo.attr,
4112 &md_suspend_hi.attr,
4113 &md_bitmap.attr,
4114 &md_degraded.attr,
4115 NULL,
4116 };
4117 static struct attribute_group md_redundancy_group = {
4118 .name = NULL,
4119 .attrs = md_redundancy_attrs,
4120 };
4121
4122
4123 static ssize_t
4124 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4125 {
4126 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4127 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4128 ssize_t rv;
4129
4130 if (!entry->show)
4131 return -EIO;
4132 rv = mddev_lock(mddev);
4133 if (!rv) {
4134 rv = entry->show(mddev, page);
4135 mddev_unlock(mddev);
4136 }
4137 return rv;
4138 }
4139
4140 static ssize_t
4141 md_attr_store(struct kobject *kobj, struct attribute *attr,
4142 const char *page, size_t length)
4143 {
4144 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4145 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4146 ssize_t rv;
4147
4148 if (!entry->store)
4149 return -EIO;
4150 if (!capable(CAP_SYS_ADMIN))
4151 return -EACCES;
4152 rv = mddev_lock(mddev);
4153 if (mddev->hold_active == UNTIL_IOCTL)
4154 mddev->hold_active = 0;
4155 if (!rv) {
4156 rv = entry->store(mddev, page, length);
4157 mddev_unlock(mddev);
4158 }
4159 return rv;
4160 }
4161
4162 static void md_free(struct kobject *ko)
4163 {
4164 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4165
4166 if (mddev->sysfs_state)
4167 sysfs_put(mddev->sysfs_state);
4168
4169 if (mddev->gendisk) {
4170 del_gendisk(mddev->gendisk);
4171 put_disk(mddev->gendisk);
4172 }
4173 if (mddev->queue)
4174 blk_cleanup_queue(mddev->queue);
4175
4176 kfree(mddev);
4177 }
4178
4179 static const struct sysfs_ops md_sysfs_ops = {
4180 .show = md_attr_show,
4181 .store = md_attr_store,
4182 };
4183 static struct kobj_type md_ktype = {
4184 .release = md_free,
4185 .sysfs_ops = &md_sysfs_ops,
4186 .default_attrs = md_default_attrs,
4187 };
4188
4189 int mdp_major = 0;
4190
4191 static void mddev_delayed_delete(struct work_struct *ws)
4192 {
4193 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4194
4195 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4196 kobject_del(&mddev->kobj);
4197 kobject_put(&mddev->kobj);
4198 }
4199
4200 static int md_alloc(dev_t dev, char *name)
4201 {
4202 static DEFINE_MUTEX(disks_mutex);
4203 mddev_t *mddev = mddev_find(dev);
4204 struct gendisk *disk;
4205 int partitioned;
4206 int shift;
4207 int unit;
4208 int error;
4209
4210 if (!mddev)
4211 return -ENODEV;
4212
4213 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4214 shift = partitioned ? MdpMinorShift : 0;
4215 unit = MINOR(mddev->unit) >> shift;
4216
4217 /* wait for any previous instance if this device
4218 * to be completed removed (mddev_delayed_delete).
4219 */
4220 flush_scheduled_work();
4221
4222 mutex_lock(&disks_mutex);
4223 error = -EEXIST;
4224 if (mddev->gendisk)
4225 goto abort;
4226
4227 if (name) {
4228 /* Need to ensure that 'name' is not a duplicate.
4229 */
4230 mddev_t *mddev2;
4231 spin_lock(&all_mddevs_lock);
4232
4233 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4234 if (mddev2->gendisk &&
4235 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4236 spin_unlock(&all_mddevs_lock);
4237 goto abort;
4238 }
4239 spin_unlock(&all_mddevs_lock);
4240 }
4241
4242 error = -ENOMEM;
4243 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4244 if (!mddev->queue)
4245 goto abort;
4246 mddev->queue->queuedata = mddev;
4247
4248 /* Can be unlocked because the queue is new: no concurrency */
4249 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4250
4251 blk_queue_make_request(mddev->queue, md_make_request);
4252
4253 disk = alloc_disk(1 << shift);
4254 if (!disk) {
4255 blk_cleanup_queue(mddev->queue);
4256 mddev->queue = NULL;
4257 goto abort;
4258 }
4259 disk->major = MAJOR(mddev->unit);
4260 disk->first_minor = unit << shift;
4261 if (name)
4262 strcpy(disk->disk_name, name);
4263 else if (partitioned)
4264 sprintf(disk->disk_name, "md_d%d", unit);
4265 else
4266 sprintf(disk->disk_name, "md%d", unit);
4267 disk->fops = &md_fops;
4268 disk->private_data = mddev;
4269 disk->queue = mddev->queue;
4270 /* Allow extended partitions. This makes the
4271 * 'mdp' device redundant, but we can't really
4272 * remove it now.
4273 */
4274 disk->flags |= GENHD_FL_EXT_DEVT;
4275 add_disk(disk);
4276 mddev->gendisk = disk;
4277 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4278 &disk_to_dev(disk)->kobj, "%s", "md");
4279 if (error) {
4280 /* This isn't possible, but as kobject_init_and_add is marked
4281 * __must_check, we must do something with the result
4282 */
4283 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4284 disk->disk_name);
4285 error = 0;
4286 }
4287 if (mddev->kobj.sd &&
4288 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4289 printk(KERN_DEBUG "pointless warning\n");
4290 abort:
4291 mutex_unlock(&disks_mutex);
4292 if (!error && mddev->kobj.sd) {
4293 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4294 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4295 }
4296 mddev_put(mddev);
4297 return error;
4298 }
4299
4300 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4301 {
4302 md_alloc(dev, NULL);
4303 return NULL;
4304 }
4305
4306 static int add_named_array(const char *val, struct kernel_param *kp)
4307 {
4308 /* val must be "md_*" where * is not all digits.
4309 * We allocate an array with a large free minor number, and
4310 * set the name to val. val must not already be an active name.
4311 */
4312 int len = strlen(val);
4313 char buf[DISK_NAME_LEN];
4314
4315 while (len && val[len-1] == '\n')
4316 len--;
4317 if (len >= DISK_NAME_LEN)
4318 return -E2BIG;
4319 strlcpy(buf, val, len+1);
4320 if (strncmp(buf, "md_", 3) != 0)
4321 return -EINVAL;
4322 return md_alloc(0, buf);
4323 }
4324
4325 static void md_safemode_timeout(unsigned long data)
4326 {
4327 mddev_t *mddev = (mddev_t *) data;
4328
4329 if (!atomic_read(&mddev->writes_pending)) {
4330 mddev->safemode = 1;
4331 if (mddev->external)
4332 sysfs_notify_dirent_safe(mddev->sysfs_state);
4333 }
4334 md_wakeup_thread(mddev->thread);
4335 }
4336
4337 static int start_dirty_degraded;
4338
4339 int md_run(mddev_t *mddev)
4340 {
4341 int err;
4342 mdk_rdev_t *rdev;
4343 struct mdk_personality *pers;
4344
4345 if (list_empty(&mddev->disks))
4346 /* cannot run an array with no devices.. */
4347 return -EINVAL;
4348
4349 if (mddev->pers)
4350 return -EBUSY;
4351
4352 /* These two calls synchronise us with the
4353 * sysfs_remove_group calls in mddev_unlock,
4354 * so they must have completed.
4355 */
4356 mutex_lock(&mddev->open_mutex);
4357 mutex_unlock(&mddev->open_mutex);
4358
4359 /*
4360 * Analyze all RAID superblock(s)
4361 */
4362 if (!mddev->raid_disks) {
4363 if (!mddev->persistent)
4364 return -EINVAL;
4365 analyze_sbs(mddev);
4366 }
4367
4368 if (mddev->level != LEVEL_NONE)
4369 request_module("md-level-%d", mddev->level);
4370 else if (mddev->clevel[0])
4371 request_module("md-%s", mddev->clevel);
4372
4373 /*
4374 * Drop all container device buffers, from now on
4375 * the only valid external interface is through the md
4376 * device.
4377 */
4378 list_for_each_entry(rdev, &mddev->disks, same_set) {
4379 if (test_bit(Faulty, &rdev->flags))
4380 continue;
4381 sync_blockdev(rdev->bdev);
4382 invalidate_bdev(rdev->bdev);
4383
4384 /* perform some consistency tests on the device.
4385 * We don't want the data to overlap the metadata,
4386 * Internal Bitmap issues have been handled elsewhere.
4387 */
4388 if (rdev->data_offset < rdev->sb_start) {
4389 if (mddev->dev_sectors &&
4390 rdev->data_offset + mddev->dev_sectors
4391 > rdev->sb_start) {
4392 printk("md: %s: data overlaps metadata\n",
4393 mdname(mddev));
4394 return -EINVAL;
4395 }
4396 } else {
4397 if (rdev->sb_start + rdev->sb_size/512
4398 > rdev->data_offset) {
4399 printk("md: %s: metadata overlaps data\n",
4400 mdname(mddev));
4401 return -EINVAL;
4402 }
4403 }
4404 sysfs_notify_dirent_safe(rdev->sysfs_state);
4405 }
4406
4407 spin_lock(&pers_lock);
4408 pers = find_pers(mddev->level, mddev->clevel);
4409 if (!pers || !try_module_get(pers->owner)) {
4410 spin_unlock(&pers_lock);
4411 if (mddev->level != LEVEL_NONE)
4412 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4413 mddev->level);
4414 else
4415 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4416 mddev->clevel);
4417 return -EINVAL;
4418 }
4419 mddev->pers = pers;
4420 spin_unlock(&pers_lock);
4421 if (mddev->level != pers->level) {
4422 mddev->level = pers->level;
4423 mddev->new_level = pers->level;
4424 }
4425 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4426
4427 if (mddev->reshape_position != MaxSector &&
4428 pers->start_reshape == NULL) {
4429 /* This personality cannot handle reshaping... */
4430 mddev->pers = NULL;
4431 module_put(pers->owner);
4432 return -EINVAL;
4433 }
4434
4435 if (pers->sync_request) {
4436 /* Warn if this is a potentially silly
4437 * configuration.
4438 */
4439 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4440 mdk_rdev_t *rdev2;
4441 int warned = 0;
4442
4443 list_for_each_entry(rdev, &mddev->disks, same_set)
4444 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4445 if (rdev < rdev2 &&
4446 rdev->bdev->bd_contains ==
4447 rdev2->bdev->bd_contains) {
4448 printk(KERN_WARNING
4449 "%s: WARNING: %s appears to be"
4450 " on the same physical disk as"
4451 " %s.\n",
4452 mdname(mddev),
4453 bdevname(rdev->bdev,b),
4454 bdevname(rdev2->bdev,b2));
4455 warned = 1;
4456 }
4457 }
4458
4459 if (warned)
4460 printk(KERN_WARNING
4461 "True protection against single-disk"
4462 " failure might be compromised.\n");
4463 }
4464
4465 mddev->recovery = 0;
4466 /* may be over-ridden by personality */
4467 mddev->resync_max_sectors = mddev->dev_sectors;
4468
4469 mddev->barriers_work = 1;
4470 mddev->ok_start_degraded = start_dirty_degraded;
4471
4472 if (start_readonly && mddev->ro == 0)
4473 mddev->ro = 2; /* read-only, but switch on first write */
4474
4475 err = mddev->pers->run(mddev);
4476 if (err)
4477 printk(KERN_ERR "md: pers->run() failed ...\n");
4478 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4479 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4480 " but 'external_size' not in effect?\n", __func__);
4481 printk(KERN_ERR
4482 "md: invalid array_size %llu > default size %llu\n",
4483 (unsigned long long)mddev->array_sectors / 2,
4484 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4485 err = -EINVAL;
4486 mddev->pers->stop(mddev);
4487 }
4488 if (err == 0 && mddev->pers->sync_request) {
4489 err = bitmap_create(mddev);
4490 if (err) {
4491 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4492 mdname(mddev), err);
4493 mddev->pers->stop(mddev);
4494 }
4495 }
4496 if (err) {
4497 module_put(mddev->pers->owner);
4498 mddev->pers = NULL;
4499 bitmap_destroy(mddev);
4500 return err;
4501 }
4502 if (mddev->pers->sync_request) {
4503 if (mddev->kobj.sd &&
4504 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4505 printk(KERN_WARNING
4506 "md: cannot register extra attributes for %s\n",
4507 mdname(mddev));
4508 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4509 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4510 mddev->ro = 0;
4511
4512 atomic_set(&mddev->writes_pending,0);
4513 atomic_set(&mddev->max_corr_read_errors,
4514 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4515 mddev->safemode = 0;
4516 mddev->safemode_timer.function = md_safemode_timeout;
4517 mddev->safemode_timer.data = (unsigned long) mddev;
4518 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4519 mddev->in_sync = 1;
4520
4521 list_for_each_entry(rdev, &mddev->disks, same_set)
4522 if (rdev->raid_disk >= 0) {
4523 char nm[20];
4524 sprintf(nm, "rd%d", rdev->raid_disk);
4525 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4526 /* failure here is OK */;
4527 }
4528
4529 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4530
4531 if (mddev->flags)
4532 md_update_sb(mddev, 0);
4533
4534 md_wakeup_thread(mddev->thread);
4535 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4536
4537 md_new_event(mddev);
4538 sysfs_notify_dirent_safe(mddev->sysfs_state);
4539 sysfs_notify_dirent_safe(mddev->sysfs_action);
4540 sysfs_notify(&mddev->kobj, NULL, "degraded");
4541 return 0;
4542 }
4543 EXPORT_SYMBOL_GPL(md_run);
4544
4545 static int do_md_run(mddev_t *mddev)
4546 {
4547 int err;
4548
4549 err = md_run(mddev);
4550 if (err)
4551 goto out;
4552
4553 set_capacity(mddev->gendisk, mddev->array_sectors);
4554 revalidate_disk(mddev->gendisk);
4555 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4556 out:
4557 return err;
4558 }
4559
4560 static int restart_array(mddev_t *mddev)
4561 {
4562 struct gendisk *disk = mddev->gendisk;
4563
4564 /* Complain if it has no devices */
4565 if (list_empty(&mddev->disks))
4566 return -ENXIO;
4567 if (!mddev->pers)
4568 return -EINVAL;
4569 if (!mddev->ro)
4570 return -EBUSY;
4571 mddev->safemode = 0;
4572 mddev->ro = 0;
4573 set_disk_ro(disk, 0);
4574 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4575 mdname(mddev));
4576 /* Kick recovery or resync if necessary */
4577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4578 md_wakeup_thread(mddev->thread);
4579 md_wakeup_thread(mddev->sync_thread);
4580 sysfs_notify_dirent_safe(mddev->sysfs_state);
4581 return 0;
4582 }
4583
4584 /* similar to deny_write_access, but accounts for our holding a reference
4585 * to the file ourselves */
4586 static int deny_bitmap_write_access(struct file * file)
4587 {
4588 struct inode *inode = file->f_mapping->host;
4589
4590 spin_lock(&inode->i_lock);
4591 if (atomic_read(&inode->i_writecount) > 1) {
4592 spin_unlock(&inode->i_lock);
4593 return -ETXTBSY;
4594 }
4595 atomic_set(&inode->i_writecount, -1);
4596 spin_unlock(&inode->i_lock);
4597
4598 return 0;
4599 }
4600
4601 void restore_bitmap_write_access(struct file *file)
4602 {
4603 struct inode *inode = file->f_mapping->host;
4604
4605 spin_lock(&inode->i_lock);
4606 atomic_set(&inode->i_writecount, 1);
4607 spin_unlock(&inode->i_lock);
4608 }
4609
4610 static void md_clean(mddev_t *mddev)
4611 {
4612 mddev->array_sectors = 0;
4613 mddev->external_size = 0;
4614 mddev->dev_sectors = 0;
4615 mddev->raid_disks = 0;
4616 mddev->recovery_cp = 0;
4617 mddev->resync_min = 0;
4618 mddev->resync_max = MaxSector;
4619 mddev->reshape_position = MaxSector;
4620 mddev->external = 0;
4621 mddev->persistent = 0;
4622 mddev->level = LEVEL_NONE;
4623 mddev->clevel[0] = 0;
4624 mddev->flags = 0;
4625 mddev->ro = 0;
4626 mddev->metadata_type[0] = 0;
4627 mddev->chunk_sectors = 0;
4628 mddev->ctime = mddev->utime = 0;
4629 mddev->layout = 0;
4630 mddev->max_disks = 0;
4631 mddev->events = 0;
4632 mddev->can_decrease_events = 0;
4633 mddev->delta_disks = 0;
4634 mddev->new_level = LEVEL_NONE;
4635 mddev->new_layout = 0;
4636 mddev->new_chunk_sectors = 0;
4637 mddev->curr_resync = 0;
4638 mddev->resync_mismatches = 0;
4639 mddev->suspend_lo = mddev->suspend_hi = 0;
4640 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4641 mddev->recovery = 0;
4642 mddev->in_sync = 0;
4643 mddev->degraded = 0;
4644 mddev->barriers_work = 0;
4645 mddev->safemode = 0;
4646 mddev->bitmap_info.offset = 0;
4647 mddev->bitmap_info.default_offset = 0;
4648 mddev->bitmap_info.chunksize = 0;
4649 mddev->bitmap_info.daemon_sleep = 0;
4650 mddev->bitmap_info.max_write_behind = 0;
4651 }
4652
4653 void md_stop_writes(mddev_t *mddev)
4654 {
4655 if (mddev->sync_thread) {
4656 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4657 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4658 md_unregister_thread(mddev->sync_thread);
4659 mddev->sync_thread = NULL;
4660 }
4661
4662 del_timer_sync(&mddev->safemode_timer);
4663
4664 bitmap_flush(mddev);
4665 md_super_wait(mddev);
4666
4667 if (!mddev->in_sync || mddev->flags) {
4668 /* mark array as shutdown cleanly */
4669 mddev->in_sync = 1;
4670 md_update_sb(mddev, 1);
4671 }
4672 }
4673 EXPORT_SYMBOL_GPL(md_stop_writes);
4674
4675 void md_stop(mddev_t *mddev)
4676 {
4677 mddev->pers->stop(mddev);
4678 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4679 mddev->to_remove = &md_redundancy_group;
4680 module_put(mddev->pers->owner);
4681 mddev->pers = NULL;
4682 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4683 }
4684 EXPORT_SYMBOL_GPL(md_stop);
4685
4686 static int md_set_readonly(mddev_t *mddev, int is_open)
4687 {
4688 int err = 0;
4689 mutex_lock(&mddev->open_mutex);
4690 if (atomic_read(&mddev->openers) > is_open) {
4691 printk("md: %s still in use.\n",mdname(mddev));
4692 err = -EBUSY;
4693 goto out;
4694 }
4695 if (mddev->pers) {
4696 md_stop_writes(mddev);
4697
4698 err = -ENXIO;
4699 if (mddev->ro==1)
4700 goto out;
4701 mddev->ro = 1;
4702 set_disk_ro(mddev->gendisk, 1);
4703 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4704 sysfs_notify_dirent_safe(mddev->sysfs_state);
4705 err = 0;
4706 }
4707 out:
4708 mutex_unlock(&mddev->open_mutex);
4709 return err;
4710 }
4711
4712 /* mode:
4713 * 0 - completely stop and dis-assemble array
4714 * 2 - stop but do not disassemble array
4715 */
4716 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4717 {
4718 int err = 0;
4719 struct gendisk *disk = mddev->gendisk;
4720 mdk_rdev_t *rdev;
4721
4722 mutex_lock(&mddev->open_mutex);
4723 if (atomic_read(&mddev->openers) > is_open) {
4724 printk("md: %s still in use.\n",mdname(mddev));
4725 err = -EBUSY;
4726 } else if (mddev->pers) {
4727
4728 if (mddev->ro)
4729 set_disk_ro(disk, 0);
4730
4731 md_stop_writes(mddev);
4732 md_stop(mddev);
4733 mddev->queue->merge_bvec_fn = NULL;
4734 mddev->queue->unplug_fn = NULL;
4735 mddev->queue->backing_dev_info.congested_fn = NULL;
4736
4737 /* tell userspace to handle 'inactive' */
4738 sysfs_notify_dirent_safe(mddev->sysfs_state);
4739
4740 list_for_each_entry(rdev, &mddev->disks, same_set)
4741 if (rdev->raid_disk >= 0) {
4742 char nm[20];
4743 sprintf(nm, "rd%d", rdev->raid_disk);
4744 sysfs_remove_link(&mddev->kobj, nm);
4745 }
4746
4747 set_capacity(disk, 0);
4748 revalidate_disk(disk);
4749
4750 if (mddev->ro)
4751 mddev->ro = 0;
4752
4753 err = 0;
4754 }
4755 mutex_unlock(&mddev->open_mutex);
4756 if (err)
4757 return err;
4758 /*
4759 * Free resources if final stop
4760 */
4761 if (mode == 0) {
4762
4763 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4764
4765 bitmap_destroy(mddev);
4766 if (mddev->bitmap_info.file) {
4767 restore_bitmap_write_access(mddev->bitmap_info.file);
4768 fput(mddev->bitmap_info.file);
4769 mddev->bitmap_info.file = NULL;
4770 }
4771 mddev->bitmap_info.offset = 0;
4772
4773 export_array(mddev);
4774
4775 md_clean(mddev);
4776 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4777 if (mddev->hold_active == UNTIL_STOP)
4778 mddev->hold_active = 0;
4779
4780 }
4781 err = 0;
4782 blk_integrity_unregister(disk);
4783 md_new_event(mddev);
4784 sysfs_notify_dirent_safe(mddev->sysfs_state);
4785 return err;
4786 }
4787
4788 #ifndef MODULE
4789 static void autorun_array(mddev_t *mddev)
4790 {
4791 mdk_rdev_t *rdev;
4792 int err;
4793
4794 if (list_empty(&mddev->disks))
4795 return;
4796
4797 printk(KERN_INFO "md: running: ");
4798
4799 list_for_each_entry(rdev, &mddev->disks, same_set) {
4800 char b[BDEVNAME_SIZE];
4801 printk("<%s>", bdevname(rdev->bdev,b));
4802 }
4803 printk("\n");
4804
4805 err = do_md_run(mddev);
4806 if (err) {
4807 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4808 do_md_stop(mddev, 0, 0);
4809 }
4810 }
4811
4812 /*
4813 * lets try to run arrays based on all disks that have arrived
4814 * until now. (those are in pending_raid_disks)
4815 *
4816 * the method: pick the first pending disk, collect all disks with
4817 * the same UUID, remove all from the pending list and put them into
4818 * the 'same_array' list. Then order this list based on superblock
4819 * update time (freshest comes first), kick out 'old' disks and
4820 * compare superblocks. If everything's fine then run it.
4821 *
4822 * If "unit" is allocated, then bump its reference count
4823 */
4824 static void autorun_devices(int part)
4825 {
4826 mdk_rdev_t *rdev0, *rdev, *tmp;
4827 mddev_t *mddev;
4828 char b[BDEVNAME_SIZE];
4829
4830 printk(KERN_INFO "md: autorun ...\n");
4831 while (!list_empty(&pending_raid_disks)) {
4832 int unit;
4833 dev_t dev;
4834 LIST_HEAD(candidates);
4835 rdev0 = list_entry(pending_raid_disks.next,
4836 mdk_rdev_t, same_set);
4837
4838 printk(KERN_INFO "md: considering %s ...\n",
4839 bdevname(rdev0->bdev,b));
4840 INIT_LIST_HEAD(&candidates);
4841 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4842 if (super_90_load(rdev, rdev0, 0) >= 0) {
4843 printk(KERN_INFO "md: adding %s ...\n",
4844 bdevname(rdev->bdev,b));
4845 list_move(&rdev->same_set, &candidates);
4846 }
4847 /*
4848 * now we have a set of devices, with all of them having
4849 * mostly sane superblocks. It's time to allocate the
4850 * mddev.
4851 */
4852 if (part) {
4853 dev = MKDEV(mdp_major,
4854 rdev0->preferred_minor << MdpMinorShift);
4855 unit = MINOR(dev) >> MdpMinorShift;
4856 } else {
4857 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4858 unit = MINOR(dev);
4859 }
4860 if (rdev0->preferred_minor != unit) {
4861 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4862 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4863 break;
4864 }
4865
4866 md_probe(dev, NULL, NULL);
4867 mddev = mddev_find(dev);
4868 if (!mddev || !mddev->gendisk) {
4869 if (mddev)
4870 mddev_put(mddev);
4871 printk(KERN_ERR
4872 "md: cannot allocate memory for md drive.\n");
4873 break;
4874 }
4875 if (mddev_lock(mddev))
4876 printk(KERN_WARNING "md: %s locked, cannot run\n",
4877 mdname(mddev));
4878 else if (mddev->raid_disks || mddev->major_version
4879 || !list_empty(&mddev->disks)) {
4880 printk(KERN_WARNING
4881 "md: %s already running, cannot run %s\n",
4882 mdname(mddev), bdevname(rdev0->bdev,b));
4883 mddev_unlock(mddev);
4884 } else {
4885 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4886 mddev->persistent = 1;
4887 rdev_for_each_list(rdev, tmp, &candidates) {
4888 list_del_init(&rdev->same_set);
4889 if (bind_rdev_to_array(rdev, mddev))
4890 export_rdev(rdev);
4891 }
4892 autorun_array(mddev);
4893 mddev_unlock(mddev);
4894 }
4895 /* on success, candidates will be empty, on error
4896 * it won't...
4897 */
4898 rdev_for_each_list(rdev, tmp, &candidates) {
4899 list_del_init(&rdev->same_set);
4900 export_rdev(rdev);
4901 }
4902 mddev_put(mddev);
4903 }
4904 printk(KERN_INFO "md: ... autorun DONE.\n");
4905 }
4906 #endif /* !MODULE */
4907
4908 static int get_version(void __user * arg)
4909 {
4910 mdu_version_t ver;
4911
4912 ver.major = MD_MAJOR_VERSION;
4913 ver.minor = MD_MINOR_VERSION;
4914 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4915
4916 if (copy_to_user(arg, &ver, sizeof(ver)))
4917 return -EFAULT;
4918
4919 return 0;
4920 }
4921
4922 static int get_array_info(mddev_t * mddev, void __user * arg)
4923 {
4924 mdu_array_info_t info;
4925 int nr,working,insync,failed,spare;
4926 mdk_rdev_t *rdev;
4927
4928 nr=working=insync=failed=spare=0;
4929 list_for_each_entry(rdev, &mddev->disks, same_set) {
4930 nr++;
4931 if (test_bit(Faulty, &rdev->flags))
4932 failed++;
4933 else {
4934 working++;
4935 if (test_bit(In_sync, &rdev->flags))
4936 insync++;
4937 else
4938 spare++;
4939 }
4940 }
4941
4942 info.major_version = mddev->major_version;
4943 info.minor_version = mddev->minor_version;
4944 info.patch_version = MD_PATCHLEVEL_VERSION;
4945 info.ctime = mddev->ctime;
4946 info.level = mddev->level;
4947 info.size = mddev->dev_sectors / 2;
4948 if (info.size != mddev->dev_sectors / 2) /* overflow */
4949 info.size = -1;
4950 info.nr_disks = nr;
4951 info.raid_disks = mddev->raid_disks;
4952 info.md_minor = mddev->md_minor;
4953 info.not_persistent= !mddev->persistent;
4954
4955 info.utime = mddev->utime;
4956 info.state = 0;
4957 if (mddev->in_sync)
4958 info.state = (1<<MD_SB_CLEAN);
4959 if (mddev->bitmap && mddev->bitmap_info.offset)
4960 info.state = (1<<MD_SB_BITMAP_PRESENT);
4961 info.active_disks = insync;
4962 info.working_disks = working;
4963 info.failed_disks = failed;
4964 info.spare_disks = spare;
4965
4966 info.layout = mddev->layout;
4967 info.chunk_size = mddev->chunk_sectors << 9;
4968
4969 if (copy_to_user(arg, &info, sizeof(info)))
4970 return -EFAULT;
4971
4972 return 0;
4973 }
4974
4975 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4976 {
4977 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4978 char *ptr, *buf = NULL;
4979 int err = -ENOMEM;
4980
4981 if (md_allow_write(mddev))
4982 file = kmalloc(sizeof(*file), GFP_NOIO);
4983 else
4984 file = kmalloc(sizeof(*file), GFP_KERNEL);
4985
4986 if (!file)
4987 goto out;
4988
4989 /* bitmap disabled, zero the first byte and copy out */
4990 if (!mddev->bitmap || !mddev->bitmap->file) {
4991 file->pathname[0] = '\0';
4992 goto copy_out;
4993 }
4994
4995 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4996 if (!buf)
4997 goto out;
4998
4999 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5000 if (IS_ERR(ptr))
5001 goto out;
5002
5003 strcpy(file->pathname, ptr);
5004
5005 copy_out:
5006 err = 0;
5007 if (copy_to_user(arg, file, sizeof(*file)))
5008 err = -EFAULT;
5009 out:
5010 kfree(buf);
5011 kfree(file);
5012 return err;
5013 }
5014
5015 static int get_disk_info(mddev_t * mddev, void __user * arg)
5016 {
5017 mdu_disk_info_t info;
5018 mdk_rdev_t *rdev;
5019
5020 if (copy_from_user(&info, arg, sizeof(info)))
5021 return -EFAULT;
5022
5023 rdev = find_rdev_nr(mddev, info.number);
5024 if (rdev) {
5025 info.major = MAJOR(rdev->bdev->bd_dev);
5026 info.minor = MINOR(rdev->bdev->bd_dev);
5027 info.raid_disk = rdev->raid_disk;
5028 info.state = 0;
5029 if (test_bit(Faulty, &rdev->flags))
5030 info.state |= (1<<MD_DISK_FAULTY);
5031 else if (test_bit(In_sync, &rdev->flags)) {
5032 info.state |= (1<<MD_DISK_ACTIVE);
5033 info.state |= (1<<MD_DISK_SYNC);
5034 }
5035 if (test_bit(WriteMostly, &rdev->flags))
5036 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5037 } else {
5038 info.major = info.minor = 0;
5039 info.raid_disk = -1;
5040 info.state = (1<<MD_DISK_REMOVED);
5041 }
5042
5043 if (copy_to_user(arg, &info, sizeof(info)))
5044 return -EFAULT;
5045
5046 return 0;
5047 }
5048
5049 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5050 {
5051 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5052 mdk_rdev_t *rdev;
5053 dev_t dev = MKDEV(info->major,info->minor);
5054
5055 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5056 return -EOVERFLOW;
5057
5058 if (!mddev->raid_disks) {
5059 int err;
5060 /* expecting a device which has a superblock */
5061 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5062 if (IS_ERR(rdev)) {
5063 printk(KERN_WARNING
5064 "md: md_import_device returned %ld\n",
5065 PTR_ERR(rdev));
5066 return PTR_ERR(rdev);
5067 }
5068 if (!list_empty(&mddev->disks)) {
5069 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5070 mdk_rdev_t, same_set);
5071 err = super_types[mddev->major_version]
5072 .load_super(rdev, rdev0, mddev->minor_version);
5073 if (err < 0) {
5074 printk(KERN_WARNING
5075 "md: %s has different UUID to %s\n",
5076 bdevname(rdev->bdev,b),
5077 bdevname(rdev0->bdev,b2));
5078 export_rdev(rdev);
5079 return -EINVAL;
5080 }
5081 }
5082 err = bind_rdev_to_array(rdev, mddev);
5083 if (err)
5084 export_rdev(rdev);
5085 return err;
5086 }
5087
5088 /*
5089 * add_new_disk can be used once the array is assembled
5090 * to add "hot spares". They must already have a superblock
5091 * written
5092 */
5093 if (mddev->pers) {
5094 int err;
5095 if (!mddev->pers->hot_add_disk) {
5096 printk(KERN_WARNING
5097 "%s: personality does not support diskops!\n",
5098 mdname(mddev));
5099 return -EINVAL;
5100 }
5101 if (mddev->persistent)
5102 rdev = md_import_device(dev, mddev->major_version,
5103 mddev->minor_version);
5104 else
5105 rdev = md_import_device(dev, -1, -1);
5106 if (IS_ERR(rdev)) {
5107 printk(KERN_WARNING
5108 "md: md_import_device returned %ld\n",
5109 PTR_ERR(rdev));
5110 return PTR_ERR(rdev);
5111 }
5112 /* set save_raid_disk if appropriate */
5113 if (!mddev->persistent) {
5114 if (info->state & (1<<MD_DISK_SYNC) &&
5115 info->raid_disk < mddev->raid_disks)
5116 rdev->raid_disk = info->raid_disk;
5117 else
5118 rdev->raid_disk = -1;
5119 } else
5120 super_types[mddev->major_version].
5121 validate_super(mddev, rdev);
5122 rdev->saved_raid_disk = rdev->raid_disk;
5123
5124 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5125 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5126 set_bit(WriteMostly, &rdev->flags);
5127 else
5128 clear_bit(WriteMostly, &rdev->flags);
5129
5130 rdev->raid_disk = -1;
5131 err = bind_rdev_to_array(rdev, mddev);
5132 if (!err && !mddev->pers->hot_remove_disk) {
5133 /* If there is hot_add_disk but no hot_remove_disk
5134 * then added disks for geometry changes,
5135 * and should be added immediately.
5136 */
5137 super_types[mddev->major_version].
5138 validate_super(mddev, rdev);
5139 err = mddev->pers->hot_add_disk(mddev, rdev);
5140 if (err)
5141 unbind_rdev_from_array(rdev);
5142 }
5143 if (err)
5144 export_rdev(rdev);
5145 else
5146 sysfs_notify_dirent_safe(rdev->sysfs_state);
5147
5148 md_update_sb(mddev, 1);
5149 if (mddev->degraded)
5150 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5152 md_wakeup_thread(mddev->thread);
5153 return err;
5154 }
5155
5156 /* otherwise, add_new_disk is only allowed
5157 * for major_version==0 superblocks
5158 */
5159 if (mddev->major_version != 0) {
5160 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5161 mdname(mddev));
5162 return -EINVAL;
5163 }
5164
5165 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5166 int err;
5167 rdev = md_import_device(dev, -1, 0);
5168 if (IS_ERR(rdev)) {
5169 printk(KERN_WARNING
5170 "md: error, md_import_device() returned %ld\n",
5171 PTR_ERR(rdev));
5172 return PTR_ERR(rdev);
5173 }
5174 rdev->desc_nr = info->number;
5175 if (info->raid_disk < mddev->raid_disks)
5176 rdev->raid_disk = info->raid_disk;
5177 else
5178 rdev->raid_disk = -1;
5179
5180 if (rdev->raid_disk < mddev->raid_disks)
5181 if (info->state & (1<<MD_DISK_SYNC))
5182 set_bit(In_sync, &rdev->flags);
5183
5184 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5185 set_bit(WriteMostly, &rdev->flags);
5186
5187 if (!mddev->persistent) {
5188 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5189 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5190 } else
5191 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5192 rdev->sectors = rdev->sb_start;
5193
5194 err = bind_rdev_to_array(rdev, mddev);
5195 if (err) {
5196 export_rdev(rdev);
5197 return err;
5198 }
5199 }
5200
5201 return 0;
5202 }
5203
5204 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5205 {
5206 char b[BDEVNAME_SIZE];
5207 mdk_rdev_t *rdev;
5208
5209 rdev = find_rdev(mddev, dev);
5210 if (!rdev)
5211 return -ENXIO;
5212
5213 if (rdev->raid_disk >= 0)
5214 goto busy;
5215
5216 kick_rdev_from_array(rdev);
5217 md_update_sb(mddev, 1);
5218 md_new_event(mddev);
5219
5220 return 0;
5221 busy:
5222 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5223 bdevname(rdev->bdev,b), mdname(mddev));
5224 return -EBUSY;
5225 }
5226
5227 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5228 {
5229 char b[BDEVNAME_SIZE];
5230 int err;
5231 mdk_rdev_t *rdev;
5232
5233 if (!mddev->pers)
5234 return -ENODEV;
5235
5236 if (mddev->major_version != 0) {
5237 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5238 " version-0 superblocks.\n",
5239 mdname(mddev));
5240 return -EINVAL;
5241 }
5242 if (!mddev->pers->hot_add_disk) {
5243 printk(KERN_WARNING
5244 "%s: personality does not support diskops!\n",
5245 mdname(mddev));
5246 return -EINVAL;
5247 }
5248
5249 rdev = md_import_device(dev, -1, 0);
5250 if (IS_ERR(rdev)) {
5251 printk(KERN_WARNING
5252 "md: error, md_import_device() returned %ld\n",
5253 PTR_ERR(rdev));
5254 return -EINVAL;
5255 }
5256
5257 if (mddev->persistent)
5258 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5259 else
5260 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5261
5262 rdev->sectors = rdev->sb_start;
5263
5264 if (test_bit(Faulty, &rdev->flags)) {
5265 printk(KERN_WARNING
5266 "md: can not hot-add faulty %s disk to %s!\n",
5267 bdevname(rdev->bdev,b), mdname(mddev));
5268 err = -EINVAL;
5269 goto abort_export;
5270 }
5271 clear_bit(In_sync, &rdev->flags);
5272 rdev->desc_nr = -1;
5273 rdev->saved_raid_disk = -1;
5274 err = bind_rdev_to_array(rdev, mddev);
5275 if (err)
5276 goto abort_export;
5277
5278 /*
5279 * The rest should better be atomic, we can have disk failures
5280 * noticed in interrupt contexts ...
5281 */
5282
5283 rdev->raid_disk = -1;
5284
5285 md_update_sb(mddev, 1);
5286
5287 /*
5288 * Kick recovery, maybe this spare has to be added to the
5289 * array immediately.
5290 */
5291 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5292 md_wakeup_thread(mddev->thread);
5293 md_new_event(mddev);
5294 return 0;
5295
5296 abort_export:
5297 export_rdev(rdev);
5298 return err;
5299 }
5300
5301 static int set_bitmap_file(mddev_t *mddev, int fd)
5302 {
5303 int err;
5304
5305 if (mddev->pers) {
5306 if (!mddev->pers->quiesce)
5307 return -EBUSY;
5308 if (mddev->recovery || mddev->sync_thread)
5309 return -EBUSY;
5310 /* we should be able to change the bitmap.. */
5311 }
5312
5313
5314 if (fd >= 0) {
5315 if (mddev->bitmap)
5316 return -EEXIST; /* cannot add when bitmap is present */
5317 mddev->bitmap_info.file = fget(fd);
5318
5319 if (mddev->bitmap_info.file == NULL) {
5320 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5321 mdname(mddev));
5322 return -EBADF;
5323 }
5324
5325 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5326 if (err) {
5327 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5328 mdname(mddev));
5329 fput(mddev->bitmap_info.file);
5330 mddev->bitmap_info.file = NULL;
5331 return err;
5332 }
5333 mddev->bitmap_info.offset = 0; /* file overrides offset */
5334 } else if (mddev->bitmap == NULL)
5335 return -ENOENT; /* cannot remove what isn't there */
5336 err = 0;
5337 if (mddev->pers) {
5338 mddev->pers->quiesce(mddev, 1);
5339 if (fd >= 0)
5340 err = bitmap_create(mddev);
5341 if (fd < 0 || err) {
5342 bitmap_destroy(mddev);
5343 fd = -1; /* make sure to put the file */
5344 }
5345 mddev->pers->quiesce(mddev, 0);
5346 }
5347 if (fd < 0) {
5348 if (mddev->bitmap_info.file) {
5349 restore_bitmap_write_access(mddev->bitmap_info.file);
5350 fput(mddev->bitmap_info.file);
5351 }
5352 mddev->bitmap_info.file = NULL;
5353 }
5354
5355 return err;
5356 }
5357
5358 /*
5359 * set_array_info is used two different ways
5360 * The original usage is when creating a new array.
5361 * In this usage, raid_disks is > 0 and it together with
5362 * level, size, not_persistent,layout,chunksize determine the
5363 * shape of the array.
5364 * This will always create an array with a type-0.90.0 superblock.
5365 * The newer usage is when assembling an array.
5366 * In this case raid_disks will be 0, and the major_version field is
5367 * use to determine which style super-blocks are to be found on the devices.
5368 * The minor and patch _version numbers are also kept incase the
5369 * super_block handler wishes to interpret them.
5370 */
5371 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5372 {
5373
5374 if (info->raid_disks == 0) {
5375 /* just setting version number for superblock loading */
5376 if (info->major_version < 0 ||
5377 info->major_version >= ARRAY_SIZE(super_types) ||
5378 super_types[info->major_version].name == NULL) {
5379 /* maybe try to auto-load a module? */
5380 printk(KERN_INFO
5381 "md: superblock version %d not known\n",
5382 info->major_version);
5383 return -EINVAL;
5384 }
5385 mddev->major_version = info->major_version;
5386 mddev->minor_version = info->minor_version;
5387 mddev->patch_version = info->patch_version;
5388 mddev->persistent = !info->not_persistent;
5389 /* ensure mddev_put doesn't delete this now that there
5390 * is some minimal configuration.
5391 */
5392 mddev->ctime = get_seconds();
5393 return 0;
5394 }
5395 mddev->major_version = MD_MAJOR_VERSION;
5396 mddev->minor_version = MD_MINOR_VERSION;
5397 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5398 mddev->ctime = get_seconds();
5399
5400 mddev->level = info->level;
5401 mddev->clevel[0] = 0;
5402 mddev->dev_sectors = 2 * (sector_t)info->size;
5403 mddev->raid_disks = info->raid_disks;
5404 /* don't set md_minor, it is determined by which /dev/md* was
5405 * openned
5406 */
5407 if (info->state & (1<<MD_SB_CLEAN))
5408 mddev->recovery_cp = MaxSector;
5409 else
5410 mddev->recovery_cp = 0;
5411 mddev->persistent = ! info->not_persistent;
5412 mddev->external = 0;
5413
5414 mddev->layout = info->layout;
5415 mddev->chunk_sectors = info->chunk_size >> 9;
5416
5417 mddev->max_disks = MD_SB_DISKS;
5418
5419 if (mddev->persistent)
5420 mddev->flags = 0;
5421 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5422
5423 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5424 mddev->bitmap_info.offset = 0;
5425
5426 mddev->reshape_position = MaxSector;
5427
5428 /*
5429 * Generate a 128 bit UUID
5430 */
5431 get_random_bytes(mddev->uuid, 16);
5432
5433 mddev->new_level = mddev->level;
5434 mddev->new_chunk_sectors = mddev->chunk_sectors;
5435 mddev->new_layout = mddev->layout;
5436 mddev->delta_disks = 0;
5437
5438 return 0;
5439 }
5440
5441 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5442 {
5443 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5444
5445 if (mddev->external_size)
5446 return;
5447
5448 mddev->array_sectors = array_sectors;
5449 }
5450 EXPORT_SYMBOL(md_set_array_sectors);
5451
5452 static int update_size(mddev_t *mddev, sector_t num_sectors)
5453 {
5454 mdk_rdev_t *rdev;
5455 int rv;
5456 int fit = (num_sectors == 0);
5457
5458 if (mddev->pers->resize == NULL)
5459 return -EINVAL;
5460 /* The "num_sectors" is the number of sectors of each device that
5461 * is used. This can only make sense for arrays with redundancy.
5462 * linear and raid0 always use whatever space is available. We can only
5463 * consider changing this number if no resync or reconstruction is
5464 * happening, and if the new size is acceptable. It must fit before the
5465 * sb_start or, if that is <data_offset, it must fit before the size
5466 * of each device. If num_sectors is zero, we find the largest size
5467 * that fits.
5468
5469 */
5470 if (mddev->sync_thread)
5471 return -EBUSY;
5472 if (mddev->bitmap)
5473 /* Sorry, cannot grow a bitmap yet, just remove it,
5474 * grow, and re-add.
5475 */
5476 return -EBUSY;
5477 list_for_each_entry(rdev, &mddev->disks, same_set) {
5478 sector_t avail = rdev->sectors;
5479
5480 if (fit && (num_sectors == 0 || num_sectors > avail))
5481 num_sectors = avail;
5482 if (avail < num_sectors)
5483 return -ENOSPC;
5484 }
5485 rv = mddev->pers->resize(mddev, num_sectors);
5486 if (!rv)
5487 revalidate_disk(mddev->gendisk);
5488 return rv;
5489 }
5490
5491 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5492 {
5493 int rv;
5494 /* change the number of raid disks */
5495 if (mddev->pers->check_reshape == NULL)
5496 return -EINVAL;
5497 if (raid_disks <= 0 ||
5498 (mddev->max_disks && raid_disks >= mddev->max_disks))
5499 return -EINVAL;
5500 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5501 return -EBUSY;
5502 mddev->delta_disks = raid_disks - mddev->raid_disks;
5503
5504 rv = mddev->pers->check_reshape(mddev);
5505 return rv;
5506 }
5507
5508
5509 /*
5510 * update_array_info is used to change the configuration of an
5511 * on-line array.
5512 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5513 * fields in the info are checked against the array.
5514 * Any differences that cannot be handled will cause an error.
5515 * Normally, only one change can be managed at a time.
5516 */
5517 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5518 {
5519 int rv = 0;
5520 int cnt = 0;
5521 int state = 0;
5522
5523 /* calculate expected state,ignoring low bits */
5524 if (mddev->bitmap && mddev->bitmap_info.offset)
5525 state |= (1 << MD_SB_BITMAP_PRESENT);
5526
5527 if (mddev->major_version != info->major_version ||
5528 mddev->minor_version != info->minor_version ||
5529 /* mddev->patch_version != info->patch_version || */
5530 mddev->ctime != info->ctime ||
5531 mddev->level != info->level ||
5532 /* mddev->layout != info->layout || */
5533 !mddev->persistent != info->not_persistent||
5534 mddev->chunk_sectors != info->chunk_size >> 9 ||
5535 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5536 ((state^info->state) & 0xfffffe00)
5537 )
5538 return -EINVAL;
5539 /* Check there is only one change */
5540 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5541 cnt++;
5542 if (mddev->raid_disks != info->raid_disks)
5543 cnt++;
5544 if (mddev->layout != info->layout)
5545 cnt++;
5546 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5547 cnt++;
5548 if (cnt == 0)
5549 return 0;
5550 if (cnt > 1)
5551 return -EINVAL;
5552
5553 if (mddev->layout != info->layout) {
5554 /* Change layout
5555 * we don't need to do anything at the md level, the
5556 * personality will take care of it all.
5557 */
5558 if (mddev->pers->check_reshape == NULL)
5559 return -EINVAL;
5560 else {
5561 mddev->new_layout = info->layout;
5562 rv = mddev->pers->check_reshape(mddev);
5563 if (rv)
5564 mddev->new_layout = mddev->layout;
5565 return rv;
5566 }
5567 }
5568 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5569 rv = update_size(mddev, (sector_t)info->size * 2);
5570
5571 if (mddev->raid_disks != info->raid_disks)
5572 rv = update_raid_disks(mddev, info->raid_disks);
5573
5574 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5575 if (mddev->pers->quiesce == NULL)
5576 return -EINVAL;
5577 if (mddev->recovery || mddev->sync_thread)
5578 return -EBUSY;
5579 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5580 /* add the bitmap */
5581 if (mddev->bitmap)
5582 return -EEXIST;
5583 if (mddev->bitmap_info.default_offset == 0)
5584 return -EINVAL;
5585 mddev->bitmap_info.offset =
5586 mddev->bitmap_info.default_offset;
5587 mddev->pers->quiesce(mddev, 1);
5588 rv = bitmap_create(mddev);
5589 if (rv)
5590 bitmap_destroy(mddev);
5591 mddev->pers->quiesce(mddev, 0);
5592 } else {
5593 /* remove the bitmap */
5594 if (!mddev->bitmap)
5595 return -ENOENT;
5596 if (mddev->bitmap->file)
5597 return -EINVAL;
5598 mddev->pers->quiesce(mddev, 1);
5599 bitmap_destroy(mddev);
5600 mddev->pers->quiesce(mddev, 0);
5601 mddev->bitmap_info.offset = 0;
5602 }
5603 }
5604 md_update_sb(mddev, 1);
5605 return rv;
5606 }
5607
5608 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5609 {
5610 mdk_rdev_t *rdev;
5611
5612 if (mddev->pers == NULL)
5613 return -ENODEV;
5614
5615 rdev = find_rdev(mddev, dev);
5616 if (!rdev)
5617 return -ENODEV;
5618
5619 md_error(mddev, rdev);
5620 return 0;
5621 }
5622
5623 /*
5624 * We have a problem here : there is no easy way to give a CHS
5625 * virtual geometry. We currently pretend that we have a 2 heads
5626 * 4 sectors (with a BIG number of cylinders...). This drives
5627 * dosfs just mad... ;-)
5628 */
5629 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5630 {
5631 mddev_t *mddev = bdev->bd_disk->private_data;
5632
5633 geo->heads = 2;
5634 geo->sectors = 4;
5635 geo->cylinders = mddev->array_sectors / 8;
5636 return 0;
5637 }
5638
5639 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5640 unsigned int cmd, unsigned long arg)
5641 {
5642 int err = 0;
5643 void __user *argp = (void __user *)arg;
5644 mddev_t *mddev = NULL;
5645 int ro;
5646
5647 if (!capable(CAP_SYS_ADMIN))
5648 return -EACCES;
5649
5650 /*
5651 * Commands dealing with the RAID driver but not any
5652 * particular array:
5653 */
5654 switch (cmd)
5655 {
5656 case RAID_VERSION:
5657 err = get_version(argp);
5658 goto done;
5659
5660 case PRINT_RAID_DEBUG:
5661 err = 0;
5662 md_print_devices();
5663 goto done;
5664
5665 #ifndef MODULE
5666 case RAID_AUTORUN:
5667 err = 0;
5668 autostart_arrays(arg);
5669 goto done;
5670 #endif
5671 default:;
5672 }
5673
5674 /*
5675 * Commands creating/starting a new array:
5676 */
5677
5678 mddev = bdev->bd_disk->private_data;
5679
5680 if (!mddev) {
5681 BUG();
5682 goto abort;
5683 }
5684
5685 err = mddev_lock(mddev);
5686 if (err) {
5687 printk(KERN_INFO
5688 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5689 err, cmd);
5690 goto abort;
5691 }
5692
5693 switch (cmd)
5694 {
5695 case SET_ARRAY_INFO:
5696 {
5697 mdu_array_info_t info;
5698 if (!arg)
5699 memset(&info, 0, sizeof(info));
5700 else if (copy_from_user(&info, argp, sizeof(info))) {
5701 err = -EFAULT;
5702 goto abort_unlock;
5703 }
5704 if (mddev->pers) {
5705 err = update_array_info(mddev, &info);
5706 if (err) {
5707 printk(KERN_WARNING "md: couldn't update"
5708 " array info. %d\n", err);
5709 goto abort_unlock;
5710 }
5711 goto done_unlock;
5712 }
5713 if (!list_empty(&mddev->disks)) {
5714 printk(KERN_WARNING
5715 "md: array %s already has disks!\n",
5716 mdname(mddev));
5717 err = -EBUSY;
5718 goto abort_unlock;
5719 }
5720 if (mddev->raid_disks) {
5721 printk(KERN_WARNING
5722 "md: array %s already initialised!\n",
5723 mdname(mddev));
5724 err = -EBUSY;
5725 goto abort_unlock;
5726 }
5727 err = set_array_info(mddev, &info);
5728 if (err) {
5729 printk(KERN_WARNING "md: couldn't set"
5730 " array info. %d\n", err);
5731 goto abort_unlock;
5732 }
5733 }
5734 goto done_unlock;
5735
5736 default:;
5737 }
5738
5739 /*
5740 * Commands querying/configuring an existing array:
5741 */
5742 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5743 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5744 if ((!mddev->raid_disks && !mddev->external)
5745 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5746 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5747 && cmd != GET_BITMAP_FILE) {
5748 err = -ENODEV;
5749 goto abort_unlock;
5750 }
5751
5752 /*
5753 * Commands even a read-only array can execute:
5754 */
5755 switch (cmd)
5756 {
5757 case GET_ARRAY_INFO:
5758 err = get_array_info(mddev, argp);
5759 goto done_unlock;
5760
5761 case GET_BITMAP_FILE:
5762 err = get_bitmap_file(mddev, argp);
5763 goto done_unlock;
5764
5765 case GET_DISK_INFO:
5766 err = get_disk_info(mddev, argp);
5767 goto done_unlock;
5768
5769 case RESTART_ARRAY_RW:
5770 err = restart_array(mddev);
5771 goto done_unlock;
5772
5773 case STOP_ARRAY:
5774 err = do_md_stop(mddev, 0, 1);
5775 goto done_unlock;
5776
5777 case STOP_ARRAY_RO:
5778 err = md_set_readonly(mddev, 1);
5779 goto done_unlock;
5780
5781 case BLKROSET:
5782 if (get_user(ro, (int __user *)(arg))) {
5783 err = -EFAULT;
5784 goto done_unlock;
5785 }
5786 err = -EINVAL;
5787
5788 /* if the bdev is going readonly the value of mddev->ro
5789 * does not matter, no writes are coming
5790 */
5791 if (ro)
5792 goto done_unlock;
5793
5794 /* are we are already prepared for writes? */
5795 if (mddev->ro != 1)
5796 goto done_unlock;
5797
5798 /* transitioning to readauto need only happen for
5799 * arrays that call md_write_start
5800 */
5801 if (mddev->pers) {
5802 err = restart_array(mddev);
5803 if (err == 0) {
5804 mddev->ro = 2;
5805 set_disk_ro(mddev->gendisk, 0);
5806 }
5807 }
5808 goto done_unlock;
5809 }
5810
5811 /*
5812 * The remaining ioctls are changing the state of the
5813 * superblock, so we do not allow them on read-only arrays.
5814 * However non-MD ioctls (e.g. get-size) will still come through
5815 * here and hit the 'default' below, so only disallow
5816 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5817 */
5818 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5819 if (mddev->ro == 2) {
5820 mddev->ro = 0;
5821 sysfs_notify_dirent_safe(mddev->sysfs_state);
5822 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5823 md_wakeup_thread(mddev->thread);
5824 } else {
5825 err = -EROFS;
5826 goto abort_unlock;
5827 }
5828 }
5829
5830 switch (cmd)
5831 {
5832 case ADD_NEW_DISK:
5833 {
5834 mdu_disk_info_t info;
5835 if (copy_from_user(&info, argp, sizeof(info)))
5836 err = -EFAULT;
5837 else
5838 err = add_new_disk(mddev, &info);
5839 goto done_unlock;
5840 }
5841
5842 case HOT_REMOVE_DISK:
5843 err = hot_remove_disk(mddev, new_decode_dev(arg));
5844 goto done_unlock;
5845
5846 case HOT_ADD_DISK:
5847 err = hot_add_disk(mddev, new_decode_dev(arg));
5848 goto done_unlock;
5849
5850 case SET_DISK_FAULTY:
5851 err = set_disk_faulty(mddev, new_decode_dev(arg));
5852 goto done_unlock;
5853
5854 case RUN_ARRAY:
5855 err = do_md_run(mddev);
5856 goto done_unlock;
5857
5858 case SET_BITMAP_FILE:
5859 err = set_bitmap_file(mddev, (int)arg);
5860 goto done_unlock;
5861
5862 default:
5863 err = -EINVAL;
5864 goto abort_unlock;
5865 }
5866
5867 done_unlock:
5868 abort_unlock:
5869 if (mddev->hold_active == UNTIL_IOCTL &&
5870 err != -EINVAL)
5871 mddev->hold_active = 0;
5872 mddev_unlock(mddev);
5873
5874 return err;
5875 done:
5876 if (err)
5877 MD_BUG();
5878 abort:
5879 return err;
5880 }
5881 #ifdef CONFIG_COMPAT
5882 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5883 unsigned int cmd, unsigned long arg)
5884 {
5885 switch (cmd) {
5886 case HOT_REMOVE_DISK:
5887 case HOT_ADD_DISK:
5888 case SET_DISK_FAULTY:
5889 case SET_BITMAP_FILE:
5890 /* These take in integer arg, do not convert */
5891 break;
5892 default:
5893 arg = (unsigned long)compat_ptr(arg);
5894 break;
5895 }
5896
5897 return md_ioctl(bdev, mode, cmd, arg);
5898 }
5899 #endif /* CONFIG_COMPAT */
5900
5901 static int md_open(struct block_device *bdev, fmode_t mode)
5902 {
5903 /*
5904 * Succeed if we can lock the mddev, which confirms that
5905 * it isn't being stopped right now.
5906 */
5907 mddev_t *mddev = mddev_find(bdev->bd_dev);
5908 int err;
5909
5910 if (mddev->gendisk != bdev->bd_disk) {
5911 /* we are racing with mddev_put which is discarding this
5912 * bd_disk.
5913 */
5914 mddev_put(mddev);
5915 /* Wait until bdev->bd_disk is definitely gone */
5916 flush_scheduled_work();
5917 /* Then retry the open from the top */
5918 return -ERESTARTSYS;
5919 }
5920 BUG_ON(mddev != bdev->bd_disk->private_data);
5921
5922 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5923 goto out;
5924
5925 err = 0;
5926 atomic_inc(&mddev->openers);
5927 mutex_unlock(&mddev->open_mutex);
5928
5929 check_disk_size_change(mddev->gendisk, bdev);
5930 out:
5931 return err;
5932 }
5933
5934 static int md_release(struct gendisk *disk, fmode_t mode)
5935 {
5936 mddev_t *mddev = disk->private_data;
5937
5938 BUG_ON(!mddev);
5939 atomic_dec(&mddev->openers);
5940 mddev_put(mddev);
5941
5942 return 0;
5943 }
5944 static const struct block_device_operations md_fops =
5945 {
5946 .owner = THIS_MODULE,
5947 .open = md_open,
5948 .release = md_release,
5949 .ioctl = md_ioctl,
5950 #ifdef CONFIG_COMPAT
5951 .compat_ioctl = md_compat_ioctl,
5952 #endif
5953 .getgeo = md_getgeo,
5954 };
5955
5956 static int md_thread(void * arg)
5957 {
5958 mdk_thread_t *thread = arg;
5959
5960 /*
5961 * md_thread is a 'system-thread', it's priority should be very
5962 * high. We avoid resource deadlocks individually in each
5963 * raid personality. (RAID5 does preallocation) We also use RR and
5964 * the very same RT priority as kswapd, thus we will never get
5965 * into a priority inversion deadlock.
5966 *
5967 * we definitely have to have equal or higher priority than
5968 * bdflush, otherwise bdflush will deadlock if there are too
5969 * many dirty RAID5 blocks.
5970 */
5971
5972 allow_signal(SIGKILL);
5973 while (!kthread_should_stop()) {
5974
5975 /* We need to wait INTERRUPTIBLE so that
5976 * we don't add to the load-average.
5977 * That means we need to be sure no signals are
5978 * pending
5979 */
5980 if (signal_pending(current))
5981 flush_signals(current);
5982
5983 wait_event_interruptible_timeout
5984 (thread->wqueue,
5985 test_bit(THREAD_WAKEUP, &thread->flags)
5986 || kthread_should_stop(),
5987 thread->timeout);
5988
5989 clear_bit(THREAD_WAKEUP, &thread->flags);
5990
5991 thread->run(thread->mddev);
5992 }
5993
5994 return 0;
5995 }
5996
5997 void md_wakeup_thread(mdk_thread_t *thread)
5998 {
5999 if (thread) {
6000 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6001 set_bit(THREAD_WAKEUP, &thread->flags);
6002 wake_up(&thread->wqueue);
6003 }
6004 }
6005
6006 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6007 const char *name)
6008 {
6009 mdk_thread_t *thread;
6010
6011 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6012 if (!thread)
6013 return NULL;
6014
6015 init_waitqueue_head(&thread->wqueue);
6016
6017 thread->run = run;
6018 thread->mddev = mddev;
6019 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6020 thread->tsk = kthread_run(md_thread, thread,
6021 "%s_%s",
6022 mdname(thread->mddev),
6023 name ?: mddev->pers->name);
6024 if (IS_ERR(thread->tsk)) {
6025 kfree(thread);
6026 return NULL;
6027 }
6028 return thread;
6029 }
6030
6031 void md_unregister_thread(mdk_thread_t *thread)
6032 {
6033 if (!thread)
6034 return;
6035 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6036
6037 kthread_stop(thread->tsk);
6038 kfree(thread);
6039 }
6040
6041 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6042 {
6043 if (!mddev) {
6044 MD_BUG();
6045 return;
6046 }
6047
6048 if (!rdev || test_bit(Faulty, &rdev->flags))
6049 return;
6050
6051 if (mddev->external)
6052 set_bit(Blocked, &rdev->flags);
6053 /*
6054 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6055 mdname(mddev),
6056 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6057 __builtin_return_address(0),__builtin_return_address(1),
6058 __builtin_return_address(2),__builtin_return_address(3));
6059 */
6060 if (!mddev->pers)
6061 return;
6062 if (!mddev->pers->error_handler)
6063 return;
6064 mddev->pers->error_handler(mddev,rdev);
6065 if (mddev->degraded)
6066 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6067 sysfs_notify_dirent_safe(rdev->sysfs_state);
6068 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6069 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6070 md_wakeup_thread(mddev->thread);
6071 md_new_event_inintr(mddev);
6072 }
6073
6074 /* seq_file implementation /proc/mdstat */
6075
6076 static void status_unused(struct seq_file *seq)
6077 {
6078 int i = 0;
6079 mdk_rdev_t *rdev;
6080
6081 seq_printf(seq, "unused devices: ");
6082
6083 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6084 char b[BDEVNAME_SIZE];
6085 i++;
6086 seq_printf(seq, "%s ",
6087 bdevname(rdev->bdev,b));
6088 }
6089 if (!i)
6090 seq_printf(seq, "<none>");
6091
6092 seq_printf(seq, "\n");
6093 }
6094
6095
6096 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6097 {
6098 sector_t max_sectors, resync, res;
6099 unsigned long dt, db;
6100 sector_t rt;
6101 int scale;
6102 unsigned int per_milli;
6103
6104 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6105
6106 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6107 max_sectors = mddev->resync_max_sectors;
6108 else
6109 max_sectors = mddev->dev_sectors;
6110
6111 /*
6112 * Should not happen.
6113 */
6114 if (!max_sectors) {
6115 MD_BUG();
6116 return;
6117 }
6118 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6119 * in a sector_t, and (max_sectors>>scale) will fit in a
6120 * u32, as those are the requirements for sector_div.
6121 * Thus 'scale' must be at least 10
6122 */
6123 scale = 10;
6124 if (sizeof(sector_t) > sizeof(unsigned long)) {
6125 while ( max_sectors/2 > (1ULL<<(scale+32)))
6126 scale++;
6127 }
6128 res = (resync>>scale)*1000;
6129 sector_div(res, (u32)((max_sectors>>scale)+1));
6130
6131 per_milli = res;
6132 {
6133 int i, x = per_milli/50, y = 20-x;
6134 seq_printf(seq, "[");
6135 for (i = 0; i < x; i++)
6136 seq_printf(seq, "=");
6137 seq_printf(seq, ">");
6138 for (i = 0; i < y; i++)
6139 seq_printf(seq, ".");
6140 seq_printf(seq, "] ");
6141 }
6142 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6143 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6144 "reshape" :
6145 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6146 "check" :
6147 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6148 "resync" : "recovery"))),
6149 per_milli/10, per_milli % 10,
6150 (unsigned long long) resync/2,
6151 (unsigned long long) max_sectors/2);
6152
6153 /*
6154 * dt: time from mark until now
6155 * db: blocks written from mark until now
6156 * rt: remaining time
6157 *
6158 * rt is a sector_t, so could be 32bit or 64bit.
6159 * So we divide before multiply in case it is 32bit and close
6160 * to the limit.
6161 * We scale the divisor (db) by 32 to avoid loosing precision
6162 * near the end of resync when the number of remaining sectors
6163 * is close to 'db'.
6164 * We then divide rt by 32 after multiplying by db to compensate.
6165 * The '+1' avoids division by zero if db is very small.
6166 */
6167 dt = ((jiffies - mddev->resync_mark) / HZ);
6168 if (!dt) dt++;
6169 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6170 - mddev->resync_mark_cnt;
6171
6172 rt = max_sectors - resync; /* number of remaining sectors */
6173 sector_div(rt, db/32+1);
6174 rt *= dt;
6175 rt >>= 5;
6176
6177 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6178 ((unsigned long)rt % 60)/6);
6179
6180 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6181 }
6182
6183 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6184 {
6185 struct list_head *tmp;
6186 loff_t l = *pos;
6187 mddev_t *mddev;
6188
6189 if (l >= 0x10000)
6190 return NULL;
6191 if (!l--)
6192 /* header */
6193 return (void*)1;
6194
6195 spin_lock(&all_mddevs_lock);
6196 list_for_each(tmp,&all_mddevs)
6197 if (!l--) {
6198 mddev = list_entry(tmp, mddev_t, all_mddevs);
6199 mddev_get(mddev);
6200 spin_unlock(&all_mddevs_lock);
6201 return mddev;
6202 }
6203 spin_unlock(&all_mddevs_lock);
6204 if (!l--)
6205 return (void*)2;/* tail */
6206 return NULL;
6207 }
6208
6209 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6210 {
6211 struct list_head *tmp;
6212 mddev_t *next_mddev, *mddev = v;
6213
6214 ++*pos;
6215 if (v == (void*)2)
6216 return NULL;
6217
6218 spin_lock(&all_mddevs_lock);
6219 if (v == (void*)1)
6220 tmp = all_mddevs.next;
6221 else
6222 tmp = mddev->all_mddevs.next;
6223 if (tmp != &all_mddevs)
6224 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6225 else {
6226 next_mddev = (void*)2;
6227 *pos = 0x10000;
6228 }
6229 spin_unlock(&all_mddevs_lock);
6230
6231 if (v != (void*)1)
6232 mddev_put(mddev);
6233 return next_mddev;
6234
6235 }
6236
6237 static void md_seq_stop(struct seq_file *seq, void *v)
6238 {
6239 mddev_t *mddev = v;
6240
6241 if (mddev && v != (void*)1 && v != (void*)2)
6242 mddev_put(mddev);
6243 }
6244
6245 struct mdstat_info {
6246 int event;
6247 };
6248
6249 static int md_seq_show(struct seq_file *seq, void *v)
6250 {
6251 mddev_t *mddev = v;
6252 sector_t sectors;
6253 mdk_rdev_t *rdev;
6254 struct mdstat_info *mi = seq->private;
6255 struct bitmap *bitmap;
6256
6257 if (v == (void*)1) {
6258 struct mdk_personality *pers;
6259 seq_printf(seq, "Personalities : ");
6260 spin_lock(&pers_lock);
6261 list_for_each_entry(pers, &pers_list, list)
6262 seq_printf(seq, "[%s] ", pers->name);
6263
6264 spin_unlock(&pers_lock);
6265 seq_printf(seq, "\n");
6266 mi->event = atomic_read(&md_event_count);
6267 return 0;
6268 }
6269 if (v == (void*)2) {
6270 status_unused(seq);
6271 return 0;
6272 }
6273
6274 if (mddev_lock(mddev) < 0)
6275 return -EINTR;
6276
6277 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6278 seq_printf(seq, "%s : %sactive", mdname(mddev),
6279 mddev->pers ? "" : "in");
6280 if (mddev->pers) {
6281 if (mddev->ro==1)
6282 seq_printf(seq, " (read-only)");
6283 if (mddev->ro==2)
6284 seq_printf(seq, " (auto-read-only)");
6285 seq_printf(seq, " %s", mddev->pers->name);
6286 }
6287
6288 sectors = 0;
6289 list_for_each_entry(rdev, &mddev->disks, same_set) {
6290 char b[BDEVNAME_SIZE];
6291 seq_printf(seq, " %s[%d]",
6292 bdevname(rdev->bdev,b), rdev->desc_nr);
6293 if (test_bit(WriteMostly, &rdev->flags))
6294 seq_printf(seq, "(W)");
6295 if (test_bit(Faulty, &rdev->flags)) {
6296 seq_printf(seq, "(F)");
6297 continue;
6298 } else if (rdev->raid_disk < 0)
6299 seq_printf(seq, "(S)"); /* spare */
6300 sectors += rdev->sectors;
6301 }
6302
6303 if (!list_empty(&mddev->disks)) {
6304 if (mddev->pers)
6305 seq_printf(seq, "\n %llu blocks",
6306 (unsigned long long)
6307 mddev->array_sectors / 2);
6308 else
6309 seq_printf(seq, "\n %llu blocks",
6310 (unsigned long long)sectors / 2);
6311 }
6312 if (mddev->persistent) {
6313 if (mddev->major_version != 0 ||
6314 mddev->minor_version != 90) {
6315 seq_printf(seq," super %d.%d",
6316 mddev->major_version,
6317 mddev->minor_version);
6318 }
6319 } else if (mddev->external)
6320 seq_printf(seq, " super external:%s",
6321 mddev->metadata_type);
6322 else
6323 seq_printf(seq, " super non-persistent");
6324
6325 if (mddev->pers) {
6326 mddev->pers->status(seq, mddev);
6327 seq_printf(seq, "\n ");
6328 if (mddev->pers->sync_request) {
6329 if (mddev->curr_resync > 2) {
6330 status_resync(seq, mddev);
6331 seq_printf(seq, "\n ");
6332 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6333 seq_printf(seq, "\tresync=DELAYED\n ");
6334 else if (mddev->recovery_cp < MaxSector)
6335 seq_printf(seq, "\tresync=PENDING\n ");
6336 }
6337 } else
6338 seq_printf(seq, "\n ");
6339
6340 if ((bitmap = mddev->bitmap)) {
6341 unsigned long chunk_kb;
6342 unsigned long flags;
6343 spin_lock_irqsave(&bitmap->lock, flags);
6344 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6345 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6346 "%lu%s chunk",
6347 bitmap->pages - bitmap->missing_pages,
6348 bitmap->pages,
6349 (bitmap->pages - bitmap->missing_pages)
6350 << (PAGE_SHIFT - 10),
6351 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6352 chunk_kb ? "KB" : "B");
6353 if (bitmap->file) {
6354 seq_printf(seq, ", file: ");
6355 seq_path(seq, &bitmap->file->f_path, " \t\n");
6356 }
6357
6358 seq_printf(seq, "\n");
6359 spin_unlock_irqrestore(&bitmap->lock, flags);
6360 }
6361
6362 seq_printf(seq, "\n");
6363 }
6364 mddev_unlock(mddev);
6365
6366 return 0;
6367 }
6368
6369 static const struct seq_operations md_seq_ops = {
6370 .start = md_seq_start,
6371 .next = md_seq_next,
6372 .stop = md_seq_stop,
6373 .show = md_seq_show,
6374 };
6375
6376 static int md_seq_open(struct inode *inode, struct file *file)
6377 {
6378 int error;
6379 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6380 if (mi == NULL)
6381 return -ENOMEM;
6382
6383 error = seq_open(file, &md_seq_ops);
6384 if (error)
6385 kfree(mi);
6386 else {
6387 struct seq_file *p = file->private_data;
6388 p->private = mi;
6389 mi->event = atomic_read(&md_event_count);
6390 }
6391 return error;
6392 }
6393
6394 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6395 {
6396 struct seq_file *m = filp->private_data;
6397 struct mdstat_info *mi = m->private;
6398 int mask;
6399
6400 poll_wait(filp, &md_event_waiters, wait);
6401
6402 /* always allow read */
6403 mask = POLLIN | POLLRDNORM;
6404
6405 if (mi->event != atomic_read(&md_event_count))
6406 mask |= POLLERR | POLLPRI;
6407 return mask;
6408 }
6409
6410 static const struct file_operations md_seq_fops = {
6411 .owner = THIS_MODULE,
6412 .open = md_seq_open,
6413 .read = seq_read,
6414 .llseek = seq_lseek,
6415 .release = seq_release_private,
6416 .poll = mdstat_poll,
6417 };
6418
6419 int register_md_personality(struct mdk_personality *p)
6420 {
6421 spin_lock(&pers_lock);
6422 list_add_tail(&p->list, &pers_list);
6423 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6424 spin_unlock(&pers_lock);
6425 return 0;
6426 }
6427
6428 int unregister_md_personality(struct mdk_personality *p)
6429 {
6430 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6431 spin_lock(&pers_lock);
6432 list_del_init(&p->list);
6433 spin_unlock(&pers_lock);
6434 return 0;
6435 }
6436
6437 static int is_mddev_idle(mddev_t *mddev, int init)
6438 {
6439 mdk_rdev_t * rdev;
6440 int idle;
6441 int curr_events;
6442
6443 idle = 1;
6444 rcu_read_lock();
6445 rdev_for_each_rcu(rdev, mddev) {
6446 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6447 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6448 (int)part_stat_read(&disk->part0, sectors[1]) -
6449 atomic_read(&disk->sync_io);
6450 /* sync IO will cause sync_io to increase before the disk_stats
6451 * as sync_io is counted when a request starts, and
6452 * disk_stats is counted when it completes.
6453 * So resync activity will cause curr_events to be smaller than
6454 * when there was no such activity.
6455 * non-sync IO will cause disk_stat to increase without
6456 * increasing sync_io so curr_events will (eventually)
6457 * be larger than it was before. Once it becomes
6458 * substantially larger, the test below will cause
6459 * the array to appear non-idle, and resync will slow
6460 * down.
6461 * If there is a lot of outstanding resync activity when
6462 * we set last_event to curr_events, then all that activity
6463 * completing might cause the array to appear non-idle
6464 * and resync will be slowed down even though there might
6465 * not have been non-resync activity. This will only
6466 * happen once though. 'last_events' will soon reflect
6467 * the state where there is little or no outstanding
6468 * resync requests, and further resync activity will
6469 * always make curr_events less than last_events.
6470 *
6471 */
6472 if (init || curr_events - rdev->last_events > 64) {
6473 rdev->last_events = curr_events;
6474 idle = 0;
6475 }
6476 }
6477 rcu_read_unlock();
6478 return idle;
6479 }
6480
6481 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6482 {
6483 /* another "blocks" (512byte) blocks have been synced */
6484 atomic_sub(blocks, &mddev->recovery_active);
6485 wake_up(&mddev->recovery_wait);
6486 if (!ok) {
6487 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6488 md_wakeup_thread(mddev->thread);
6489 // stop recovery, signal do_sync ....
6490 }
6491 }
6492
6493
6494 /* md_write_start(mddev, bi)
6495 * If we need to update some array metadata (e.g. 'active' flag
6496 * in superblock) before writing, schedule a superblock update
6497 * and wait for it to complete.
6498 */
6499 void md_write_start(mddev_t *mddev, struct bio *bi)
6500 {
6501 int did_change = 0;
6502 if (bio_data_dir(bi) != WRITE)
6503 return;
6504
6505 BUG_ON(mddev->ro == 1);
6506 if (mddev->ro == 2) {
6507 /* need to switch to read/write */
6508 mddev->ro = 0;
6509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6510 md_wakeup_thread(mddev->thread);
6511 md_wakeup_thread(mddev->sync_thread);
6512 did_change = 1;
6513 }
6514 atomic_inc(&mddev->writes_pending);
6515 if (mddev->safemode == 1)
6516 mddev->safemode = 0;
6517 if (mddev->in_sync) {
6518 spin_lock_irq(&mddev->write_lock);
6519 if (mddev->in_sync) {
6520 mddev->in_sync = 0;
6521 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6522 md_wakeup_thread(mddev->thread);
6523 did_change = 1;
6524 }
6525 spin_unlock_irq(&mddev->write_lock);
6526 }
6527 if (did_change)
6528 sysfs_notify_dirent_safe(mddev->sysfs_state);
6529 wait_event(mddev->sb_wait,
6530 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6531 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6532 }
6533
6534 void md_write_end(mddev_t *mddev)
6535 {
6536 if (atomic_dec_and_test(&mddev->writes_pending)) {
6537 if (mddev->safemode == 2)
6538 md_wakeup_thread(mddev->thread);
6539 else if (mddev->safemode_delay)
6540 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6541 }
6542 }
6543
6544 /* md_allow_write(mddev)
6545 * Calling this ensures that the array is marked 'active' so that writes
6546 * may proceed without blocking. It is important to call this before
6547 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6548 * Must be called with mddev_lock held.
6549 *
6550 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6551 * is dropped, so return -EAGAIN after notifying userspace.
6552 */
6553 int md_allow_write(mddev_t *mddev)
6554 {
6555 if (!mddev->pers)
6556 return 0;
6557 if (mddev->ro)
6558 return 0;
6559 if (!mddev->pers->sync_request)
6560 return 0;
6561
6562 spin_lock_irq(&mddev->write_lock);
6563 if (mddev->in_sync) {
6564 mddev->in_sync = 0;
6565 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6566 if (mddev->safemode_delay &&
6567 mddev->safemode == 0)
6568 mddev->safemode = 1;
6569 spin_unlock_irq(&mddev->write_lock);
6570 md_update_sb(mddev, 0);
6571 sysfs_notify_dirent_safe(mddev->sysfs_state);
6572 } else
6573 spin_unlock_irq(&mddev->write_lock);
6574
6575 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6576 return -EAGAIN;
6577 else
6578 return 0;
6579 }
6580 EXPORT_SYMBOL_GPL(md_allow_write);
6581
6582 #define SYNC_MARKS 10
6583 #define SYNC_MARK_STEP (3*HZ)
6584 void md_do_sync(mddev_t *mddev)
6585 {
6586 mddev_t *mddev2;
6587 unsigned int currspeed = 0,
6588 window;
6589 sector_t max_sectors,j, io_sectors;
6590 unsigned long mark[SYNC_MARKS];
6591 sector_t mark_cnt[SYNC_MARKS];
6592 int last_mark,m;
6593 struct list_head *tmp;
6594 sector_t last_check;
6595 int skipped = 0;
6596 mdk_rdev_t *rdev;
6597 char *desc;
6598
6599 /* just incase thread restarts... */
6600 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6601 return;
6602 if (mddev->ro) /* never try to sync a read-only array */
6603 return;
6604
6605 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6606 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6607 desc = "data-check";
6608 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6609 desc = "requested-resync";
6610 else
6611 desc = "resync";
6612 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6613 desc = "reshape";
6614 else
6615 desc = "recovery";
6616
6617 /* we overload curr_resync somewhat here.
6618 * 0 == not engaged in resync at all
6619 * 2 == checking that there is no conflict with another sync
6620 * 1 == like 2, but have yielded to allow conflicting resync to
6621 * commense
6622 * other == active in resync - this many blocks
6623 *
6624 * Before starting a resync we must have set curr_resync to
6625 * 2, and then checked that every "conflicting" array has curr_resync
6626 * less than ours. When we find one that is the same or higher
6627 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6628 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6629 * This will mean we have to start checking from the beginning again.
6630 *
6631 */
6632
6633 do {
6634 mddev->curr_resync = 2;
6635
6636 try_again:
6637 if (kthread_should_stop())
6638 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6639
6640 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6641 goto skip;
6642 for_each_mddev(mddev2, tmp) {
6643 if (mddev2 == mddev)
6644 continue;
6645 if (!mddev->parallel_resync
6646 && mddev2->curr_resync
6647 && match_mddev_units(mddev, mddev2)) {
6648 DEFINE_WAIT(wq);
6649 if (mddev < mddev2 && mddev->curr_resync == 2) {
6650 /* arbitrarily yield */
6651 mddev->curr_resync = 1;
6652 wake_up(&resync_wait);
6653 }
6654 if (mddev > mddev2 && mddev->curr_resync == 1)
6655 /* no need to wait here, we can wait the next
6656 * time 'round when curr_resync == 2
6657 */
6658 continue;
6659 /* We need to wait 'interruptible' so as not to
6660 * contribute to the load average, and not to
6661 * be caught by 'softlockup'
6662 */
6663 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6664 if (!kthread_should_stop() &&
6665 mddev2->curr_resync >= mddev->curr_resync) {
6666 printk(KERN_INFO "md: delaying %s of %s"
6667 " until %s has finished (they"
6668 " share one or more physical units)\n",
6669 desc, mdname(mddev), mdname(mddev2));
6670 mddev_put(mddev2);
6671 if (signal_pending(current))
6672 flush_signals(current);
6673 schedule();
6674 finish_wait(&resync_wait, &wq);
6675 goto try_again;
6676 }
6677 finish_wait(&resync_wait, &wq);
6678 }
6679 }
6680 } while (mddev->curr_resync < 2);
6681
6682 j = 0;
6683 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6684 /* resync follows the size requested by the personality,
6685 * which defaults to physical size, but can be virtual size
6686 */
6687 max_sectors = mddev->resync_max_sectors;
6688 mddev->resync_mismatches = 0;
6689 /* we don't use the checkpoint if there's a bitmap */
6690 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6691 j = mddev->resync_min;
6692 else if (!mddev->bitmap)
6693 j = mddev->recovery_cp;
6694
6695 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6696 max_sectors = mddev->dev_sectors;
6697 else {
6698 /* recovery follows the physical size of devices */
6699 max_sectors = mddev->dev_sectors;
6700 j = MaxSector;
6701 rcu_read_lock();
6702 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6703 if (rdev->raid_disk >= 0 &&
6704 !test_bit(Faulty, &rdev->flags) &&
6705 !test_bit(In_sync, &rdev->flags) &&
6706 rdev->recovery_offset < j)
6707 j = rdev->recovery_offset;
6708 rcu_read_unlock();
6709 }
6710
6711 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6712 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6713 " %d KB/sec/disk.\n", speed_min(mddev));
6714 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6715 "(but not more than %d KB/sec) for %s.\n",
6716 speed_max(mddev), desc);
6717
6718 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6719
6720 io_sectors = 0;
6721 for (m = 0; m < SYNC_MARKS; m++) {
6722 mark[m] = jiffies;
6723 mark_cnt[m] = io_sectors;
6724 }
6725 last_mark = 0;
6726 mddev->resync_mark = mark[last_mark];
6727 mddev->resync_mark_cnt = mark_cnt[last_mark];
6728
6729 /*
6730 * Tune reconstruction:
6731 */
6732 window = 32*(PAGE_SIZE/512);
6733 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6734 window/2,(unsigned long long) max_sectors/2);
6735
6736 atomic_set(&mddev->recovery_active, 0);
6737 last_check = 0;
6738
6739 if (j>2) {
6740 printk(KERN_INFO
6741 "md: resuming %s of %s from checkpoint.\n",
6742 desc, mdname(mddev));
6743 mddev->curr_resync = j;
6744 }
6745 mddev->curr_resync_completed = mddev->curr_resync;
6746
6747 while (j < max_sectors) {
6748 sector_t sectors;
6749
6750 skipped = 0;
6751
6752 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6753 ((mddev->curr_resync > mddev->curr_resync_completed &&
6754 (mddev->curr_resync - mddev->curr_resync_completed)
6755 > (max_sectors >> 4)) ||
6756 (j - mddev->curr_resync_completed)*2
6757 >= mddev->resync_max - mddev->curr_resync_completed
6758 )) {
6759 /* time to update curr_resync_completed */
6760 blk_unplug(mddev->queue);
6761 wait_event(mddev->recovery_wait,
6762 atomic_read(&mddev->recovery_active) == 0);
6763 mddev->curr_resync_completed =
6764 mddev->curr_resync;
6765 if (mddev->persistent)
6766 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6767 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6768 }
6769
6770 while (j >= mddev->resync_max && !kthread_should_stop()) {
6771 /* As this condition is controlled by user-space,
6772 * we can block indefinitely, so use '_interruptible'
6773 * to avoid triggering warnings.
6774 */
6775 flush_signals(current); /* just in case */
6776 wait_event_interruptible(mddev->recovery_wait,
6777 mddev->resync_max > j
6778 || kthread_should_stop());
6779 }
6780
6781 if (kthread_should_stop())
6782 goto interrupted;
6783
6784 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6785 currspeed < speed_min(mddev));
6786 if (sectors == 0) {
6787 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6788 goto out;
6789 }
6790
6791 if (!skipped) { /* actual IO requested */
6792 io_sectors += sectors;
6793 atomic_add(sectors, &mddev->recovery_active);
6794 }
6795
6796 j += sectors;
6797 if (j>1) mddev->curr_resync = j;
6798 mddev->curr_mark_cnt = io_sectors;
6799 if (last_check == 0)
6800 /* this is the earliers that rebuilt will be
6801 * visible in /proc/mdstat
6802 */
6803 md_new_event(mddev);
6804
6805 if (last_check + window > io_sectors || j == max_sectors)
6806 continue;
6807
6808 last_check = io_sectors;
6809
6810 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6811 break;
6812
6813 repeat:
6814 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6815 /* step marks */
6816 int next = (last_mark+1) % SYNC_MARKS;
6817
6818 mddev->resync_mark = mark[next];
6819 mddev->resync_mark_cnt = mark_cnt[next];
6820 mark[next] = jiffies;
6821 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6822 last_mark = next;
6823 }
6824
6825
6826 if (kthread_should_stop())
6827 goto interrupted;
6828
6829
6830 /*
6831 * this loop exits only if either when we are slower than
6832 * the 'hard' speed limit, or the system was IO-idle for
6833 * a jiffy.
6834 * the system might be non-idle CPU-wise, but we only care
6835 * about not overloading the IO subsystem. (things like an
6836 * e2fsck being done on the RAID array should execute fast)
6837 */
6838 blk_unplug(mddev->queue);
6839 cond_resched();
6840
6841 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6842 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6843
6844 if (currspeed > speed_min(mddev)) {
6845 if ((currspeed > speed_max(mddev)) ||
6846 !is_mddev_idle(mddev, 0)) {
6847 msleep(500);
6848 goto repeat;
6849 }
6850 }
6851 }
6852 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6853 /*
6854 * this also signals 'finished resyncing' to md_stop
6855 */
6856 out:
6857 blk_unplug(mddev->queue);
6858
6859 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6860
6861 /* tell personality that we are finished */
6862 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6863
6864 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6865 mddev->curr_resync > 2) {
6866 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6867 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6868 if (mddev->curr_resync >= mddev->recovery_cp) {
6869 printk(KERN_INFO
6870 "md: checkpointing %s of %s.\n",
6871 desc, mdname(mddev));
6872 mddev->recovery_cp = mddev->curr_resync;
6873 }
6874 } else
6875 mddev->recovery_cp = MaxSector;
6876 } else {
6877 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6878 mddev->curr_resync = MaxSector;
6879 rcu_read_lock();
6880 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6881 if (rdev->raid_disk >= 0 &&
6882 mddev->delta_disks >= 0 &&
6883 !test_bit(Faulty, &rdev->flags) &&
6884 !test_bit(In_sync, &rdev->flags) &&
6885 rdev->recovery_offset < mddev->curr_resync)
6886 rdev->recovery_offset = mddev->curr_resync;
6887 rcu_read_unlock();
6888 }
6889 }
6890 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6891
6892 skip:
6893 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6894 /* We completed so min/max setting can be forgotten if used. */
6895 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6896 mddev->resync_min = 0;
6897 mddev->resync_max = MaxSector;
6898 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6899 mddev->resync_min = mddev->curr_resync_completed;
6900 mddev->curr_resync = 0;
6901 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6902 mddev->curr_resync_completed = 0;
6903 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6904 wake_up(&resync_wait);
6905 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6906 md_wakeup_thread(mddev->thread);
6907 return;
6908
6909 interrupted:
6910 /*
6911 * got a signal, exit.
6912 */
6913 printk(KERN_INFO
6914 "md: md_do_sync() got signal ... exiting\n");
6915 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6916 goto out;
6917
6918 }
6919 EXPORT_SYMBOL_GPL(md_do_sync);
6920
6921
6922 static int remove_and_add_spares(mddev_t *mddev)
6923 {
6924 mdk_rdev_t *rdev;
6925 int spares = 0;
6926
6927 mddev->curr_resync_completed = 0;
6928
6929 list_for_each_entry(rdev, &mddev->disks, same_set)
6930 if (rdev->raid_disk >= 0 &&
6931 !test_bit(Blocked, &rdev->flags) &&
6932 (test_bit(Faulty, &rdev->flags) ||
6933 ! test_bit(In_sync, &rdev->flags)) &&
6934 atomic_read(&rdev->nr_pending)==0) {
6935 if (mddev->pers->hot_remove_disk(
6936 mddev, rdev->raid_disk)==0) {
6937 char nm[20];
6938 sprintf(nm,"rd%d", rdev->raid_disk);
6939 sysfs_remove_link(&mddev->kobj, nm);
6940 rdev->raid_disk = -1;
6941 }
6942 }
6943
6944 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6945 list_for_each_entry(rdev, &mddev->disks, same_set) {
6946 if (rdev->raid_disk >= 0 &&
6947 !test_bit(In_sync, &rdev->flags) &&
6948 !test_bit(Blocked, &rdev->flags))
6949 spares++;
6950 if (rdev->raid_disk < 0
6951 && !test_bit(Faulty, &rdev->flags)) {
6952 rdev->recovery_offset = 0;
6953 if (mddev->pers->
6954 hot_add_disk(mddev, rdev) == 0) {
6955 char nm[20];
6956 sprintf(nm, "rd%d", rdev->raid_disk);
6957 if (sysfs_create_link(&mddev->kobj,
6958 &rdev->kobj, nm))
6959 /* failure here is OK */;
6960 spares++;
6961 md_new_event(mddev);
6962 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6963 } else
6964 break;
6965 }
6966 }
6967 }
6968 return spares;
6969 }
6970 /*
6971 * This routine is regularly called by all per-raid-array threads to
6972 * deal with generic issues like resync and super-block update.
6973 * Raid personalities that don't have a thread (linear/raid0) do not
6974 * need this as they never do any recovery or update the superblock.
6975 *
6976 * It does not do any resync itself, but rather "forks" off other threads
6977 * to do that as needed.
6978 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6979 * "->recovery" and create a thread at ->sync_thread.
6980 * When the thread finishes it sets MD_RECOVERY_DONE
6981 * and wakeups up this thread which will reap the thread and finish up.
6982 * This thread also removes any faulty devices (with nr_pending == 0).
6983 *
6984 * The overall approach is:
6985 * 1/ if the superblock needs updating, update it.
6986 * 2/ If a recovery thread is running, don't do anything else.
6987 * 3/ If recovery has finished, clean up, possibly marking spares active.
6988 * 4/ If there are any faulty devices, remove them.
6989 * 5/ If array is degraded, try to add spares devices
6990 * 6/ If array has spares or is not in-sync, start a resync thread.
6991 */
6992 void md_check_recovery(mddev_t *mddev)
6993 {
6994 mdk_rdev_t *rdev;
6995
6996
6997 if (mddev->bitmap)
6998 bitmap_daemon_work(mddev);
6999
7000 if (mddev->ro)
7001 return;
7002
7003 if (signal_pending(current)) {
7004 if (mddev->pers->sync_request && !mddev->external) {
7005 printk(KERN_INFO "md: %s in immediate safe mode\n",
7006 mdname(mddev));
7007 mddev->safemode = 2;
7008 }
7009 flush_signals(current);
7010 }
7011
7012 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7013 return;
7014 if ( ! (
7015 (mddev->flags && !mddev->external) ||
7016 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7017 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7018 (mddev->external == 0 && mddev->safemode == 1) ||
7019 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7020 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7021 ))
7022 return;
7023
7024 if (mddev_trylock(mddev)) {
7025 int spares = 0;
7026
7027 if (mddev->ro) {
7028 /* Only thing we do on a ro array is remove
7029 * failed devices.
7030 */
7031 remove_and_add_spares(mddev);
7032 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7033 goto unlock;
7034 }
7035
7036 if (!mddev->external) {
7037 int did_change = 0;
7038 spin_lock_irq(&mddev->write_lock);
7039 if (mddev->safemode &&
7040 !atomic_read(&mddev->writes_pending) &&
7041 !mddev->in_sync &&
7042 mddev->recovery_cp == MaxSector) {
7043 mddev->in_sync = 1;
7044 did_change = 1;
7045 if (mddev->persistent)
7046 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7047 }
7048 if (mddev->safemode == 1)
7049 mddev->safemode = 0;
7050 spin_unlock_irq(&mddev->write_lock);
7051 if (did_change)
7052 sysfs_notify_dirent_safe(mddev->sysfs_state);
7053 }
7054
7055 if (mddev->flags)
7056 md_update_sb(mddev, 0);
7057
7058 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7059 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7060 /* resync/recovery still happening */
7061 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7062 goto unlock;
7063 }
7064 if (mddev->sync_thread) {
7065 /* resync has finished, collect result */
7066 md_unregister_thread(mddev->sync_thread);
7067 mddev->sync_thread = NULL;
7068 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7069 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7070 /* success...*/
7071 /* activate any spares */
7072 if (mddev->pers->spare_active(mddev))
7073 sysfs_notify(&mddev->kobj, NULL,
7074 "degraded");
7075 }
7076 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7077 mddev->pers->finish_reshape)
7078 mddev->pers->finish_reshape(mddev);
7079 md_update_sb(mddev, 1);
7080
7081 /* if array is no-longer degraded, then any saved_raid_disk
7082 * information must be scrapped
7083 */
7084 if (!mddev->degraded)
7085 list_for_each_entry(rdev, &mddev->disks, same_set)
7086 rdev->saved_raid_disk = -1;
7087
7088 mddev->recovery = 0;
7089 /* flag recovery needed just to double check */
7090 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7091 sysfs_notify_dirent_safe(mddev->sysfs_action);
7092 md_new_event(mddev);
7093 goto unlock;
7094 }
7095 /* Set RUNNING before clearing NEEDED to avoid
7096 * any transients in the value of "sync_action".
7097 */
7098 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7099 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7100 /* Clear some bits that don't mean anything, but
7101 * might be left set
7102 */
7103 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7104 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7105
7106 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7107 goto unlock;
7108 /* no recovery is running.
7109 * remove any failed drives, then
7110 * add spares if possible.
7111 * Spare are also removed and re-added, to allow
7112 * the personality to fail the re-add.
7113 */
7114
7115 if (mddev->reshape_position != MaxSector) {
7116 if (mddev->pers->check_reshape == NULL ||
7117 mddev->pers->check_reshape(mddev) != 0)
7118 /* Cannot proceed */
7119 goto unlock;
7120 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7121 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7122 } else if ((spares = remove_and_add_spares(mddev))) {
7123 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7124 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7125 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7126 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7127 } else if (mddev->recovery_cp < MaxSector) {
7128 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7129 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7130 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7131 /* nothing to be done ... */
7132 goto unlock;
7133
7134 if (mddev->pers->sync_request) {
7135 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7136 /* We are adding a device or devices to an array
7137 * which has the bitmap stored on all devices.
7138 * So make sure all bitmap pages get written
7139 */
7140 bitmap_write_all(mddev->bitmap);
7141 }
7142 mddev->sync_thread = md_register_thread(md_do_sync,
7143 mddev,
7144 "resync");
7145 if (!mddev->sync_thread) {
7146 printk(KERN_ERR "%s: could not start resync"
7147 " thread...\n",
7148 mdname(mddev));
7149 /* leave the spares where they are, it shouldn't hurt */
7150 mddev->recovery = 0;
7151 } else
7152 md_wakeup_thread(mddev->sync_thread);
7153 sysfs_notify_dirent_safe(mddev->sysfs_action);
7154 md_new_event(mddev);
7155 }
7156 unlock:
7157 if (!mddev->sync_thread) {
7158 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7159 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7160 &mddev->recovery))
7161 if (mddev->sysfs_action)
7162 sysfs_notify_dirent_safe(mddev->sysfs_action);
7163 }
7164 mddev_unlock(mddev);
7165 }
7166 }
7167
7168 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7169 {
7170 sysfs_notify_dirent_safe(rdev->sysfs_state);
7171 wait_event_timeout(rdev->blocked_wait,
7172 !test_bit(Blocked, &rdev->flags),
7173 msecs_to_jiffies(5000));
7174 rdev_dec_pending(rdev, mddev);
7175 }
7176 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7177
7178 static int md_notify_reboot(struct notifier_block *this,
7179 unsigned long code, void *x)
7180 {
7181 struct list_head *tmp;
7182 mddev_t *mddev;
7183
7184 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7185
7186 printk(KERN_INFO "md: stopping all md devices.\n");
7187
7188 for_each_mddev(mddev, tmp)
7189 if (mddev_trylock(mddev)) {
7190 /* Force a switch to readonly even array
7191 * appears to still be in use. Hence
7192 * the '100'.
7193 */
7194 md_set_readonly(mddev, 100);
7195 mddev_unlock(mddev);
7196 }
7197 /*
7198 * certain more exotic SCSI devices are known to be
7199 * volatile wrt too early system reboots. While the
7200 * right place to handle this issue is the given
7201 * driver, we do want to have a safe RAID driver ...
7202 */
7203 mdelay(1000*1);
7204 }
7205 return NOTIFY_DONE;
7206 }
7207
7208 static struct notifier_block md_notifier = {
7209 .notifier_call = md_notify_reboot,
7210 .next = NULL,
7211 .priority = INT_MAX, /* before any real devices */
7212 };
7213
7214 static void md_geninit(void)
7215 {
7216 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7217
7218 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7219 }
7220
7221 static int __init md_init(void)
7222 {
7223 if (register_blkdev(MD_MAJOR, "md"))
7224 return -1;
7225 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7226 unregister_blkdev(MD_MAJOR, "md");
7227 return -1;
7228 }
7229 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7230 md_probe, NULL, NULL);
7231 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7232 md_probe, NULL, NULL);
7233
7234 register_reboot_notifier(&md_notifier);
7235 raid_table_header = register_sysctl_table(raid_root_table);
7236
7237 md_geninit();
7238 return 0;
7239 }
7240
7241
7242 #ifndef MODULE
7243
7244 /*
7245 * Searches all registered partitions for autorun RAID arrays
7246 * at boot time.
7247 */
7248
7249 static LIST_HEAD(all_detected_devices);
7250 struct detected_devices_node {
7251 struct list_head list;
7252 dev_t dev;
7253 };
7254
7255 void md_autodetect_dev(dev_t dev)
7256 {
7257 struct detected_devices_node *node_detected_dev;
7258
7259 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7260 if (node_detected_dev) {
7261 node_detected_dev->dev = dev;
7262 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7263 } else {
7264 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7265 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7266 }
7267 }
7268
7269
7270 static void autostart_arrays(int part)
7271 {
7272 mdk_rdev_t *rdev;
7273 struct detected_devices_node *node_detected_dev;
7274 dev_t dev;
7275 int i_scanned, i_passed;
7276
7277 i_scanned = 0;
7278 i_passed = 0;
7279
7280 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7281
7282 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7283 i_scanned++;
7284 node_detected_dev = list_entry(all_detected_devices.next,
7285 struct detected_devices_node, list);
7286 list_del(&node_detected_dev->list);
7287 dev = node_detected_dev->dev;
7288 kfree(node_detected_dev);
7289 rdev = md_import_device(dev,0, 90);
7290 if (IS_ERR(rdev))
7291 continue;
7292
7293 if (test_bit(Faulty, &rdev->flags)) {
7294 MD_BUG();
7295 continue;
7296 }
7297 set_bit(AutoDetected, &rdev->flags);
7298 list_add(&rdev->same_set, &pending_raid_disks);
7299 i_passed++;
7300 }
7301
7302 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7303 i_scanned, i_passed);
7304
7305 autorun_devices(part);
7306 }
7307
7308 #endif /* !MODULE */
7309
7310 static __exit void md_exit(void)
7311 {
7312 mddev_t *mddev;
7313 struct list_head *tmp;
7314
7315 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7316 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7317
7318 unregister_blkdev(MD_MAJOR,"md");
7319 unregister_blkdev(mdp_major, "mdp");
7320 unregister_reboot_notifier(&md_notifier);
7321 unregister_sysctl_table(raid_table_header);
7322 remove_proc_entry("mdstat", NULL);
7323 for_each_mddev(mddev, tmp) {
7324 export_array(mddev);
7325 mddev->hold_active = 0;
7326 }
7327 }
7328
7329 subsys_initcall(md_init);
7330 module_exit(md_exit)
7331
7332 static int get_ro(char *buffer, struct kernel_param *kp)
7333 {
7334 return sprintf(buffer, "%d", start_readonly);
7335 }
7336 static int set_ro(const char *val, struct kernel_param *kp)
7337 {
7338 char *e;
7339 int num = simple_strtoul(val, &e, 10);
7340 if (*val && (*e == '\0' || *e == '\n')) {
7341 start_readonly = num;
7342 return 0;
7343 }
7344 return -EINVAL;
7345 }
7346
7347 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7348 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7349
7350 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7351
7352 EXPORT_SYMBOL(register_md_personality);
7353 EXPORT_SYMBOL(unregister_md_personality);
7354 EXPORT_SYMBOL(md_error);
7355 EXPORT_SYMBOL(md_done_sync);
7356 EXPORT_SYMBOL(md_write_start);
7357 EXPORT_SYMBOL(md_write_end);
7358 EXPORT_SYMBOL(md_register_thread);
7359 EXPORT_SYMBOL(md_unregister_thread);
7360 EXPORT_SYMBOL(md_wakeup_thread);
7361 EXPORT_SYMBOL(md_check_recovery);
7362 MODULE_LICENSE("GPL");
7363 MODULE_DESCRIPTION("MD RAID framework");
7364 MODULE_ALIAS("md");
7365 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree