?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
omap: rx51_defconfig: Remove CONFIG_SYSFS_DEPRECATED*=y options
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / md / md.c
blobfdc1890b6ac58310bd988f518ca4751df1a95cd4
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 "md.h"
53 #include "bitmap.h"
54
55 #define DEBUG 0
56 #define dprintk(x...) ((void)(DEBUG && printk(x)))
57
58
59 #ifndef MODULE
60 static void autostart_arrays(int part);
61 #endif
62
63 static LIST_HEAD(pers_list);
64 static DEFINE_SPINLOCK(pers_lock);
65
66 static void md_print_devices(void);
67
68 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
69
70 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71
72 /*
73 * Default number of read corrections we'll attempt on an rdev
74 * before ejecting it from the array. We divide the read error
75 * count by 2 for every hour elapsed between read errors.
76 */
77 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
78 /*
79 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80 * is 1000 KB/sec, so the extra system load does not show up that much.
81 * Increase it if you want to have more _guaranteed_ speed. Note that
82 * the RAID driver will use the maximum available bandwidth if the IO
83 * subsystem is idle. There is also an 'absolute maximum' reconstruction
84 * speed limit - in case reconstruction slows down your system despite
85 * idle IO detection.
86 *
87 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88 * or /sys/block/mdX/md/sync_speed_{min,max}
89 */
90
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
94 {
95 return mddev->sync_speed_min ?
96 mddev->sync_speed_min : sysctl_speed_limit_min;
97 }
98
99 static inline int speed_max(mddev_t *mddev)
100 {
101 return mddev->sync_speed_max ?
102 mddev->sync_speed_max : sysctl_speed_limit_max;
103 }
104
105 static struct ctl_table_header *raid_table_header;
106
107 static ctl_table raid_table[] = {
108 {
109 .procname = "speed_limit_min",
110 .data = &sysctl_speed_limit_min,
111 .maxlen = sizeof(int),
112 .mode = S_IRUGO|S_IWUSR,
113 .proc_handler = proc_dointvec,
114 },
115 {
116 .procname = "speed_limit_max",
117 .data = &sysctl_speed_limit_max,
118 .maxlen = sizeof(int),
119 .mode = S_IRUGO|S_IWUSR,
120 .proc_handler = proc_dointvec,
121 },
122 { }
123 };
124
125 static ctl_table raid_dir_table[] = {
126 {
127 .procname = "raid",
128 .maxlen = 0,
129 .mode = S_IRUGO|S_IXUGO,
130 .child = raid_table,
131 },
132 { }
133 };
134
135 static ctl_table raid_root_table[] = {
136 {
137 .procname = "dev",
138 .maxlen = 0,
139 .mode = 0555,
140 .child = raid_dir_table,
141 },
142 { }
143 };
144
145 static const struct block_device_operations md_fops;
146
147 static int start_readonly;
148
149 /*
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
153 * count increases.
154 *
155 * Events are:
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
158 */
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
160 static atomic_t md_event_count;
161 void md_new_event(mddev_t *mddev)
162 {
163 atomic_inc(&md_event_count);
164 wake_up(&md_event_waiters);
165 }
166 EXPORT_SYMBOL_GPL(md_new_event);
167
168 /* Alternate version that can be called from interrupts
169 * when calling sysfs_notify isn't needed.
170 */
171 static void md_new_event_inintr(mddev_t *mddev)
172 {
173 atomic_inc(&md_event_count);
174 wake_up(&md_event_waiters);
175 }
176
177 /*
178 * Enables to iterate over all existing md arrays
179 * all_mddevs_lock protects this list.
180 */
181 static LIST_HEAD(all_mddevs);
182 static DEFINE_SPINLOCK(all_mddevs_lock);
183
184
185 /*
186 * iterates through all used mddevs in the system.
187 * We take care to grab the all_mddevs_lock whenever navigating
188 * the list, and to always hold a refcount when unlocked.
189 * Any code which breaks out of this loop while own
190 * a reference to the current mddev and must mddev_put it.
191 */
192 #define for_each_mddev(mddev,tmp) \
193 \
194 for (({ spin_lock(&all_mddevs_lock); \
195 tmp = all_mddevs.next; \
196 mddev = NULL;}); \
197 ({ if (tmp != &all_mddevs) \
198 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
199 spin_unlock(&all_mddevs_lock); \
200 if (mddev) mddev_put(mddev); \
201 mddev = list_entry(tmp, mddev_t, all_mddevs); \
202 tmp != &all_mddevs;}); \
203 ({ spin_lock(&all_mddevs_lock); \
204 tmp = tmp->next;}) \
205 )
206
207
208 /* Rather than calling directly into the personality make_request function,
209 * IO requests come here first so that we can check if the device is
210 * being suspended pending a reconfiguration.
211 * We hold a refcount over the call to ->make_request. By the time that
212 * call has finished, the bio has been linked into some internal structure
213 * and so is visible to ->quiesce(), so we don't need the refcount any more.
214 */
215 static int md_make_request(struct request_queue *q, struct bio *bio)
216 {
217 mddev_t *mddev = q->queuedata;
218 int rv;
219 if (mddev == NULL || mddev->pers == NULL) {
220 bio_io_error(bio);
221 return 0;
222 }
223 rcu_read_lock();
224 if (mddev->suspended || mddev->barrier) {
225 DEFINE_WAIT(__wait);
226 for (;;) {
227 prepare_to_wait(&mddev->sb_wait, &__wait,
228 TASK_UNINTERRUPTIBLE);
229 if (!mddev->suspended && !mddev->barrier)
230 break;
231 rcu_read_unlock();
232 schedule();
233 rcu_read_lock();
234 }
235 finish_wait(&mddev->sb_wait, &__wait);
236 }
237 atomic_inc(&mddev->active_io);
238 rcu_read_unlock();
239 rv = mddev->pers->make_request(q, bio);
240 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
241 wake_up(&mddev->sb_wait);
242
243 return rv;
244 }
245
246 static void mddev_suspend(mddev_t *mddev)
247 {
248 BUG_ON(mddev->suspended);
249 mddev->suspended = 1;
250 synchronize_rcu();
251 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
252 mddev->pers->quiesce(mddev, 1);
253 md_unregister_thread(mddev->thread);
254 mddev->thread = NULL;
255 /* we now know that no code is executing in the personality module,
256 * except possibly the tail end of a ->bi_end_io function, but that
257 * is certain to complete before the module has a chance to get
258 * unloaded
259 */
260 }
261
262 static void mddev_resume(mddev_t *mddev)
263 {
264 mddev->suspended = 0;
265 wake_up(&mddev->sb_wait);
266 mddev->pers->quiesce(mddev, 0);
267 }
268
269 int mddev_congested(mddev_t *mddev, int bits)
270 {
271 if (mddev->barrier)
272 return 1;
273 return mddev->suspended;
274 }
275 EXPORT_SYMBOL(mddev_congested);
276
277 /*
278 * Generic barrier handling for md
279 */
280
281 #define POST_REQUEST_BARRIER ((void*)1)
282
283 static void md_end_barrier(struct bio *bio, int err)
284 {
285 mdk_rdev_t *rdev = bio->bi_private;
286 mddev_t *mddev = rdev->mddev;
287 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
288 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
289
290 rdev_dec_pending(rdev, mddev);
291
292 if (atomic_dec_and_test(&mddev->flush_pending)) {
293 if (mddev->barrier == POST_REQUEST_BARRIER) {
294 /* This was a post-request barrier */
295 mddev->barrier = NULL;
296 wake_up(&mddev->sb_wait);
297 } else
298 /* The pre-request barrier has finished */
299 schedule_work(&mddev->barrier_work);
300 }
301 bio_put(bio);
302 }
303
304 static void submit_barriers(mddev_t *mddev)
305 {
306 mdk_rdev_t *rdev;
307
308 rcu_read_lock();
309 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
310 if (rdev->raid_disk >= 0 &&
311 !test_bit(Faulty, &rdev->flags)) {
312 /* Take two references, one is dropped
313 * when request finishes, one after
314 * we reclaim rcu_read_lock
315 */
316 struct bio *bi;
317 atomic_inc(&rdev->nr_pending);
318 atomic_inc(&rdev->nr_pending);
319 rcu_read_unlock();
320 bi = bio_alloc(GFP_KERNEL, 0);
321 bi->bi_end_io = md_end_barrier;
322 bi->bi_private = rdev;
323 bi->bi_bdev = rdev->bdev;
324 atomic_inc(&mddev->flush_pending);
325 submit_bio(WRITE_BARRIER, bi);
326 rcu_read_lock();
327 rdev_dec_pending(rdev, mddev);
328 }
329 rcu_read_unlock();
330 }
331
332 static void md_submit_barrier(struct work_struct *ws)
333 {
334 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
335 struct bio *bio = mddev->barrier;
336
337 atomic_set(&mddev->flush_pending, 1);
338
339 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
340 bio_endio(bio, -EOPNOTSUPP);
341 else if (bio->bi_size == 0)
342 /* an empty barrier - all done */
343 bio_endio(bio, 0);
344 else {
345 bio->bi_rw &= ~(1<<BIO_RW_BARRIER);
346 if (mddev->pers->make_request(mddev->queue, bio))
347 generic_make_request(bio);
348 mddev->barrier = POST_REQUEST_BARRIER;
349 submit_barriers(mddev);
350 }
351 if (atomic_dec_and_test(&mddev->flush_pending)) {
352 mddev->barrier = NULL;
353 wake_up(&mddev->sb_wait);
354 }
355 }
356
357 void md_barrier_request(mddev_t *mddev, struct bio *bio)
358 {
359 spin_lock_irq(&mddev->write_lock);
360 wait_event_lock_irq(mddev->sb_wait,
361 !mddev->barrier,
362 mddev->write_lock, /*nothing*/);
363 mddev->barrier = bio;
364 spin_unlock_irq(&mddev->write_lock);
365
366 atomic_set(&mddev->flush_pending, 1);
367 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
368
369 submit_barriers(mddev);
370
371 if (atomic_dec_and_test(&mddev->flush_pending))
372 schedule_work(&mddev->barrier_work);
373 }
374 EXPORT_SYMBOL(md_barrier_request);
375
376 static inline mddev_t *mddev_get(mddev_t *mddev)
377 {
378 atomic_inc(&mddev->active);
379 return mddev;
380 }
381
382 static void mddev_delayed_delete(struct work_struct *ws);
383
384 static void mddev_put(mddev_t *mddev)
385 {
386 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
387 return;
388 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
389 mddev->ctime == 0 && !mddev->hold_active) {
390 /* Array is not configured at all, and not held active,
391 * so destroy it */
392 list_del(&mddev->all_mddevs);
393 if (mddev->gendisk) {
394 /* we did a probe so need to clean up.
395 * Call schedule_work inside the spinlock
396 * so that flush_scheduled_work() after
397 * mddev_find will succeed in waiting for the
398 * work to be done.
399 */
400 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
401 schedule_work(&mddev->del_work);
402 } else
403 kfree(mddev);
404 }
405 spin_unlock(&all_mddevs_lock);
406 }
407
408 static mddev_t * mddev_find(dev_t unit)
409 {
410 mddev_t *mddev, *new = NULL;
411
412 retry:
413 spin_lock(&all_mddevs_lock);
414
415 if (unit) {
416 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
417 if (mddev->unit == unit) {
418 mddev_get(mddev);
419 spin_unlock(&all_mddevs_lock);
420 kfree(new);
421 return mddev;
422 }
423
424 if (new) {
425 list_add(&new->all_mddevs, &all_mddevs);
426 spin_unlock(&all_mddevs_lock);
427 new->hold_active = UNTIL_IOCTL;
428 return new;
429 }
430 } else if (new) {
431 /* find an unused unit number */
432 static int next_minor = 512;
433 int start = next_minor;
434 int is_free = 0;
435 int dev = 0;
436 while (!is_free) {
437 dev = MKDEV(MD_MAJOR, next_minor);
438 next_minor++;
439 if (next_minor > MINORMASK)
440 next_minor = 0;
441 if (next_minor == start) {
442 /* Oh dear, all in use. */
443 spin_unlock(&all_mddevs_lock);
444 kfree(new);
445 return NULL;
446 }
447
448 is_free = 1;
449 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
450 if (mddev->unit == dev) {
451 is_free = 0;
452 break;
453 }
454 }
455 new->unit = dev;
456 new->md_minor = MINOR(dev);
457 new->hold_active = UNTIL_STOP;
458 list_add(&new->all_mddevs, &all_mddevs);
459 spin_unlock(&all_mddevs_lock);
460 return new;
461 }
462 spin_unlock(&all_mddevs_lock);
463
464 new = kzalloc(sizeof(*new), GFP_KERNEL);
465 if (!new)
466 return NULL;
467
468 new->unit = unit;
469 if (MAJOR(unit) == MD_MAJOR)
470 new->md_minor = MINOR(unit);
471 else
472 new->md_minor = MINOR(unit) >> MdpMinorShift;
473
474 mutex_init(&new->open_mutex);
475 mutex_init(&new->reconfig_mutex);
476 mutex_init(&new->bitmap_info.mutex);
477 INIT_LIST_HEAD(&new->disks);
478 INIT_LIST_HEAD(&new->all_mddevs);
479 init_timer(&new->safemode_timer);
480 atomic_set(&new->active, 1);
481 atomic_set(&new->openers, 0);
482 atomic_set(&new->active_io, 0);
483 spin_lock_init(&new->write_lock);
484 atomic_set(&new->flush_pending, 0);
485 init_waitqueue_head(&new->sb_wait);
486 init_waitqueue_head(&new->recovery_wait);
487 new->reshape_position = MaxSector;
488 new->resync_min = 0;
489 new->resync_max = MaxSector;
490 new->level = LEVEL_NONE;
491
492 goto retry;
493 }
494
495 static inline int mddev_lock(mddev_t * mddev)
496 {
497 return mutex_lock_interruptible(&mddev->reconfig_mutex);
498 }
499
500 static inline int mddev_is_locked(mddev_t *mddev)
501 {
502 return mutex_is_locked(&mddev->reconfig_mutex);
503 }
504
505 static inline int mddev_trylock(mddev_t * mddev)
506 {
507 return mutex_trylock(&mddev->reconfig_mutex);
508 }
509
510 static inline void mddev_unlock(mddev_t * mddev)
511 {
512 mutex_unlock(&mddev->reconfig_mutex);
513
514 md_wakeup_thread(mddev->thread);
515 }
516
517 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
518 {
519 mdk_rdev_t *rdev;
520
521 list_for_each_entry(rdev, &mddev->disks, same_set)
522 if (rdev->desc_nr == nr)
523 return rdev;
524
525 return NULL;
526 }
527
528 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
529 {
530 mdk_rdev_t *rdev;
531
532 list_for_each_entry(rdev, &mddev->disks, same_set)
533 if (rdev->bdev->bd_dev == dev)
534 return rdev;
535
536 return NULL;
537 }
538
539 static struct mdk_personality *find_pers(int level, char *clevel)
540 {
541 struct mdk_personality *pers;
542 list_for_each_entry(pers, &pers_list, list) {
543 if (level != LEVEL_NONE && pers->level == level)
544 return pers;
545 if (strcmp(pers->name, clevel)==0)
546 return pers;
547 }
548 return NULL;
549 }
550
551 /* return the offset of the super block in 512byte sectors */
552 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
553 {
554 sector_t num_sectors = bdev->bd_inode->i_size / 512;
555 return MD_NEW_SIZE_SECTORS(num_sectors);
556 }
557
558 static int alloc_disk_sb(mdk_rdev_t * rdev)
559 {
560 if (rdev->sb_page)
561 MD_BUG();
562
563 rdev->sb_page = alloc_page(GFP_KERNEL);
564 if (!rdev->sb_page) {
565 printk(KERN_ALERT "md: out of memory.\n");
566 return -ENOMEM;
567 }
568
569 return 0;
570 }
571
572 static void free_disk_sb(mdk_rdev_t * rdev)
573 {
574 if (rdev->sb_page) {
575 put_page(rdev->sb_page);
576 rdev->sb_loaded = 0;
577 rdev->sb_page = NULL;
578 rdev->sb_start = 0;
579 rdev->sectors = 0;
580 }
581 }
582
583
584 static void super_written(struct bio *bio, int error)
585 {
586 mdk_rdev_t *rdev = bio->bi_private;
587 mddev_t *mddev = rdev->mddev;
588
589 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
590 printk("md: super_written gets error=%d, uptodate=%d\n",
591 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
592 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
593 md_error(mddev, rdev);
594 }
595
596 if (atomic_dec_and_test(&mddev->pending_writes))
597 wake_up(&mddev->sb_wait);
598 bio_put(bio);
599 }
600
601 static void super_written_barrier(struct bio *bio, int error)
602 {
603 struct bio *bio2 = bio->bi_private;
604 mdk_rdev_t *rdev = bio2->bi_private;
605 mddev_t *mddev = rdev->mddev;
606
607 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
608 error == -EOPNOTSUPP) {
609 unsigned long flags;
610 /* barriers don't appear to be supported :-( */
611 set_bit(BarriersNotsupp, &rdev->flags);
612 mddev->barriers_work = 0;
613 spin_lock_irqsave(&mddev->write_lock, flags);
614 bio2->bi_next = mddev->biolist;
615 mddev->biolist = bio2;
616 spin_unlock_irqrestore(&mddev->write_lock, flags);
617 wake_up(&mddev->sb_wait);
618 bio_put(bio);
619 } else {
620 bio_put(bio2);
621 bio->bi_private = rdev;
622 super_written(bio, error);
623 }
624 }
625
626 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
627 sector_t sector, int size, struct page *page)
628 {
629 /* write first size bytes of page to sector of rdev
630 * Increment mddev->pending_writes before returning
631 * and decrement it on completion, waking up sb_wait
632 * if zero is reached.
633 * If an error occurred, call md_error
634 *
635 * As we might need to resubmit the request if BIO_RW_BARRIER
636 * causes ENOTSUPP, we allocate a spare bio...
637 */
638 struct bio *bio = bio_alloc(GFP_NOIO, 1);
639 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
640
641 bio->bi_bdev = rdev->bdev;
642 bio->bi_sector = sector;
643 bio_add_page(bio, page, size, 0);
644 bio->bi_private = rdev;
645 bio->bi_end_io = super_written;
646 bio->bi_rw = rw;
647
648 atomic_inc(&mddev->pending_writes);
649 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
650 struct bio *rbio;
651 rw |= (1<<BIO_RW_BARRIER);
652 rbio = bio_clone(bio, GFP_NOIO);
653 rbio->bi_private = bio;
654 rbio->bi_end_io = super_written_barrier;
655 submit_bio(rw, rbio);
656 } else
657 submit_bio(rw, bio);
658 }
659
660 void md_super_wait(mddev_t *mddev)
661 {
662 /* wait for all superblock writes that were scheduled to complete.
663 * if any had to be retried (due to BARRIER problems), retry them
664 */
665 DEFINE_WAIT(wq);
666 for(;;) {
667 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
668 if (atomic_read(&mddev->pending_writes)==0)
669 break;
670 while (mddev->biolist) {
671 struct bio *bio;
672 spin_lock_irq(&mddev->write_lock);
673 bio = mddev->biolist;
674 mddev->biolist = bio->bi_next ;
675 bio->bi_next = NULL;
676 spin_unlock_irq(&mddev->write_lock);
677 submit_bio(bio->bi_rw, bio);
678 }
679 schedule();
680 }
681 finish_wait(&mddev->sb_wait, &wq);
682 }
683
684 static void bi_complete(struct bio *bio, int error)
685 {
686 complete((struct completion*)bio->bi_private);
687 }
688
689 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
690 struct page *page, int rw)
691 {
692 struct bio *bio = bio_alloc(GFP_NOIO, 1);
693 struct completion event;
694 int ret;
695
696 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
697
698 bio->bi_bdev = bdev;
699 bio->bi_sector = sector;
700 bio_add_page(bio, page, size, 0);
701 init_completion(&event);
702 bio->bi_private = &event;
703 bio->bi_end_io = bi_complete;
704 submit_bio(rw, bio);
705 wait_for_completion(&event);
706
707 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
708 bio_put(bio);
709 return ret;
710 }
711 EXPORT_SYMBOL_GPL(sync_page_io);
712
713 static int read_disk_sb(mdk_rdev_t * rdev, int size)
714 {
715 char b[BDEVNAME_SIZE];
716 if (!rdev->sb_page) {
717 MD_BUG();
718 return -EINVAL;
719 }
720 if (rdev->sb_loaded)
721 return 0;
722
723
724 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
725 goto fail;
726 rdev->sb_loaded = 1;
727 return 0;
728
729 fail:
730 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
731 bdevname(rdev->bdev,b));
732 return -EINVAL;
733 }
734
735 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
736 {
737 return sb1->set_uuid0 == sb2->set_uuid0 &&
738 sb1->set_uuid1 == sb2->set_uuid1 &&
739 sb1->set_uuid2 == sb2->set_uuid2 &&
740 sb1->set_uuid3 == sb2->set_uuid3;
741 }
742
743 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
744 {
745 int ret;
746 mdp_super_t *tmp1, *tmp2;
747
748 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
749 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
750
751 if (!tmp1 || !tmp2) {
752 ret = 0;
753 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
754 goto abort;
755 }
756
757 *tmp1 = *sb1;
758 *tmp2 = *sb2;
759
760 /*
761 * nr_disks is not constant
762 */
763 tmp1->nr_disks = 0;
764 tmp2->nr_disks = 0;
765
766 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
767 abort:
768 kfree(tmp1);
769 kfree(tmp2);
770 return ret;
771 }
772
773
774 static u32 md_csum_fold(u32 csum)
775 {
776 csum = (csum & 0xffff) + (csum >> 16);
777 return (csum & 0xffff) + (csum >> 16);
778 }
779
780 static unsigned int calc_sb_csum(mdp_super_t * sb)
781 {
782 u64 newcsum = 0;
783 u32 *sb32 = (u32*)sb;
784 int i;
785 unsigned int disk_csum, csum;
786
787 disk_csum = sb->sb_csum;
788 sb->sb_csum = 0;
789
790 for (i = 0; i < MD_SB_BYTES/4 ; i++)
791 newcsum += sb32[i];
792 csum = (newcsum & 0xffffffff) + (newcsum>>32);
793
794
795 #ifdef CONFIG_ALPHA
796 /* This used to use csum_partial, which was wrong for several
797 * reasons including that different results are returned on
798 * different architectures. It isn't critical that we get exactly
799 * the same return value as before (we always csum_fold before
800 * testing, and that removes any differences). However as we
801 * know that csum_partial always returned a 16bit value on
802 * alphas, do a fold to maximise conformity to previous behaviour.
803 */
804 sb->sb_csum = md_csum_fold(disk_csum);
805 #else
806 sb->sb_csum = disk_csum;
807 #endif
808 return csum;
809 }
810
811
812 /*
813 * Handle superblock details.
814 * We want to be able to handle multiple superblock formats
815 * so we have a common interface to them all, and an array of
816 * different handlers.
817 * We rely on user-space to write the initial superblock, and support
818 * reading and updating of superblocks.
819 * Interface methods are:
820 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
821 * loads and validates a superblock on dev.
822 * if refdev != NULL, compare superblocks on both devices
823 * Return:
824 * 0 - dev has a superblock that is compatible with refdev
825 * 1 - dev has a superblock that is compatible and newer than refdev
826 * so dev should be used as the refdev in future
827 * -EINVAL superblock incompatible or invalid
828 * -othererror e.g. -EIO
829 *
830 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
831 * Verify that dev is acceptable into mddev.
832 * The first time, mddev->raid_disks will be 0, and data from
833 * dev should be merged in. Subsequent calls check that dev
834 * is new enough. Return 0 or -EINVAL
835 *
836 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
837 * Update the superblock for rdev with data in mddev
838 * This does not write to disc.
839 *
840 */
841
842 struct super_type {
843 char *name;
844 struct module *owner;
845 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
846 int minor_version);
847 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
848 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
849 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
850 sector_t num_sectors);
851 };
852
853 /*
854 * Check that the given mddev has no bitmap.
855 *
856 * This function is called from the run method of all personalities that do not
857 * support bitmaps. It prints an error message and returns non-zero if mddev
858 * has a bitmap. Otherwise, it returns 0.
859 *
860 */
861 int md_check_no_bitmap(mddev_t *mddev)
862 {
863 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
864 return 0;
865 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
866 mdname(mddev), mddev->pers->name);
867 return 1;
868 }
869 EXPORT_SYMBOL(md_check_no_bitmap);
870
871 /*
872 * load_super for 0.90.0
873 */
874 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
875 {
876 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
877 mdp_super_t *sb;
878 int ret;
879
880 /*
881 * Calculate the position of the superblock (512byte sectors),
882 * it's at the end of the disk.
883 *
884 * It also happens to be a multiple of 4Kb.
885 */
886 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
887
888 ret = read_disk_sb(rdev, MD_SB_BYTES);
889 if (ret) return ret;
890
891 ret = -EINVAL;
892
893 bdevname(rdev->bdev, b);
894 sb = (mdp_super_t*)page_address(rdev->sb_page);
895
896 if (sb->md_magic != MD_SB_MAGIC) {
897 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
898 b);
899 goto abort;
900 }
901
902 if (sb->major_version != 0 ||
903 sb->minor_version < 90 ||
904 sb->minor_version > 91) {
905 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
906 sb->major_version, sb->minor_version,
907 b);
908 goto abort;
909 }
910
911 if (sb->raid_disks <= 0)
912 goto abort;
913
914 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
915 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
916 b);
917 goto abort;
918 }
919
920 rdev->preferred_minor = sb->md_minor;
921 rdev->data_offset = 0;
922 rdev->sb_size = MD_SB_BYTES;
923
924 if (sb->level == LEVEL_MULTIPATH)
925 rdev->desc_nr = -1;
926 else
927 rdev->desc_nr = sb->this_disk.number;
928
929 if (!refdev) {
930 ret = 1;
931 } else {
932 __u64 ev1, ev2;
933 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
934 if (!uuid_equal(refsb, sb)) {
935 printk(KERN_WARNING "md: %s has different UUID to %s\n",
936 b, bdevname(refdev->bdev,b2));
937 goto abort;
938 }
939 if (!sb_equal(refsb, sb)) {
940 printk(KERN_WARNING "md: %s has same UUID"
941 " but different superblock to %s\n",
942 b, bdevname(refdev->bdev, b2));
943 goto abort;
944 }
945 ev1 = md_event(sb);
946 ev2 = md_event(refsb);
947 if (ev1 > ev2)
948 ret = 1;
949 else
950 ret = 0;
951 }
952 rdev->sectors = rdev->sb_start;
953
954 if (rdev->sectors < sb->size * 2 && sb->level > 1)
955 /* "this cannot possibly happen" ... */
956 ret = -EINVAL;
957
958 abort:
959 return ret;
960 }
961
962 /*
963 * validate_super for 0.90.0
964 */
965 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
966 {
967 mdp_disk_t *desc;
968 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
969 __u64 ev1 = md_event(sb);
970
971 rdev->raid_disk = -1;
972 clear_bit(Faulty, &rdev->flags);
973 clear_bit(In_sync, &rdev->flags);
974 clear_bit(WriteMostly, &rdev->flags);
975 clear_bit(BarriersNotsupp, &rdev->flags);
976
977 if (mddev->raid_disks == 0) {
978 mddev->major_version = 0;
979 mddev->minor_version = sb->minor_version;
980 mddev->patch_version = sb->patch_version;
981 mddev->external = 0;
982 mddev->chunk_sectors = sb->chunk_size >> 9;
983 mddev->ctime = sb->ctime;
984 mddev->utime = sb->utime;
985 mddev->level = sb->level;
986 mddev->clevel[0] = 0;
987 mddev->layout = sb->layout;
988 mddev->raid_disks = sb->raid_disks;
989 mddev->dev_sectors = sb->size * 2;
990 mddev->events = ev1;
991 mddev->bitmap_info.offset = 0;
992 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
993
994 if (mddev->minor_version >= 91) {
995 mddev->reshape_position = sb->reshape_position;
996 mddev->delta_disks = sb->delta_disks;
997 mddev->new_level = sb->new_level;
998 mddev->new_layout = sb->new_layout;
999 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1000 } else {
1001 mddev->reshape_position = MaxSector;
1002 mddev->delta_disks = 0;
1003 mddev->new_level = mddev->level;
1004 mddev->new_layout = mddev->layout;
1005 mddev->new_chunk_sectors = mddev->chunk_sectors;
1006 }
1007
1008 if (sb->state & (1<<MD_SB_CLEAN))
1009 mddev->recovery_cp = MaxSector;
1010 else {
1011 if (sb->events_hi == sb->cp_events_hi &&
1012 sb->events_lo == sb->cp_events_lo) {
1013 mddev->recovery_cp = sb->recovery_cp;
1014 } else
1015 mddev->recovery_cp = 0;
1016 }
1017
1018 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1019 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1020 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1021 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1022
1023 mddev->max_disks = MD_SB_DISKS;
1024
1025 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1026 mddev->bitmap_info.file == NULL)
1027 mddev->bitmap_info.offset =
1028 mddev->bitmap_info.default_offset;
1029
1030 } else if (mddev->pers == NULL) {
1031 /* Insist on good event counter while assembling */
1032 ++ev1;
1033 if (ev1 < mddev->events)
1034 return -EINVAL;
1035 } else if (mddev->bitmap) {
1036 /* if adding to array with a bitmap, then we can accept an
1037 * older device ... but not too old.
1038 */
1039 if (ev1 < mddev->bitmap->events_cleared)
1040 return 0;
1041 } else {
1042 if (ev1 < mddev->events)
1043 /* just a hot-add of a new device, leave raid_disk at -1 */
1044 return 0;
1045 }
1046
1047 if (mddev->level != LEVEL_MULTIPATH) {
1048 desc = sb->disks + rdev->desc_nr;
1049
1050 if (desc->state & (1<<MD_DISK_FAULTY))
1051 set_bit(Faulty, &rdev->flags);
1052 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1053 desc->raid_disk < mddev->raid_disks */) {
1054 set_bit(In_sync, &rdev->flags);
1055 rdev->raid_disk = desc->raid_disk;
1056 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1057 /* active but not in sync implies recovery up to
1058 * reshape position. We don't know exactly where
1059 * that is, so set to zero for now */
1060 if (mddev->minor_version >= 91) {
1061 rdev->recovery_offset = 0;
1062 rdev->raid_disk = desc->raid_disk;
1063 }
1064 }
1065 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1066 set_bit(WriteMostly, &rdev->flags);
1067 } else /* MULTIPATH are always insync */
1068 set_bit(In_sync, &rdev->flags);
1069 return 0;
1070 }
1071
1072 /*
1073 * sync_super for 0.90.0
1074 */
1075 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1076 {
1077 mdp_super_t *sb;
1078 mdk_rdev_t *rdev2;
1079 int next_spare = mddev->raid_disks;
1080
1081
1082 /* make rdev->sb match mddev data..
1083 *
1084 * 1/ zero out disks
1085 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1086 * 3/ any empty disks < next_spare become removed
1087 *
1088 * disks[0] gets initialised to REMOVED because
1089 * we cannot be sure from other fields if it has
1090 * been initialised or not.
1091 */
1092 int i;
1093 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1094
1095 rdev->sb_size = MD_SB_BYTES;
1096
1097 sb = (mdp_super_t*)page_address(rdev->sb_page);
1098
1099 memset(sb, 0, sizeof(*sb));
1100
1101 sb->md_magic = MD_SB_MAGIC;
1102 sb->major_version = mddev->major_version;
1103 sb->patch_version = mddev->patch_version;
1104 sb->gvalid_words = 0; /* ignored */
1105 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1106 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1107 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1108 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1109
1110 sb->ctime = mddev->ctime;
1111 sb->level = mddev->level;
1112 sb->size = mddev->dev_sectors / 2;
1113 sb->raid_disks = mddev->raid_disks;
1114 sb->md_minor = mddev->md_minor;
1115 sb->not_persistent = 0;
1116 sb->utime = mddev->utime;
1117 sb->state = 0;
1118 sb->events_hi = (mddev->events>>32);
1119 sb->events_lo = (u32)mddev->events;
1120
1121 if (mddev->reshape_position == MaxSector)
1122 sb->minor_version = 90;
1123 else {
1124 sb->minor_version = 91;
1125 sb->reshape_position = mddev->reshape_position;
1126 sb->new_level = mddev->new_level;
1127 sb->delta_disks = mddev->delta_disks;
1128 sb->new_layout = mddev->new_layout;
1129 sb->new_chunk = mddev->new_chunk_sectors << 9;
1130 }
1131 mddev->minor_version = sb->minor_version;
1132 if (mddev->in_sync)
1133 {
1134 sb->recovery_cp = mddev->recovery_cp;
1135 sb->cp_events_hi = (mddev->events>>32);
1136 sb->cp_events_lo = (u32)mddev->events;
1137 if (mddev->recovery_cp == MaxSector)
1138 sb->state = (1<< MD_SB_CLEAN);
1139 } else
1140 sb->recovery_cp = 0;
1141
1142 sb->layout = mddev->layout;
1143 sb->chunk_size = mddev->chunk_sectors << 9;
1144
1145 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1146 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1147
1148 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1149 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1150 mdp_disk_t *d;
1151 int desc_nr;
1152 int is_active = test_bit(In_sync, &rdev2->flags);
1153
1154 if (rdev2->raid_disk >= 0 &&
1155 sb->minor_version >= 91)
1156 /* we have nowhere to store the recovery_offset,
1157 * but if it is not below the reshape_position,
1158 * we can piggy-back on that.
1159 */
1160 is_active = 1;
1161 if (rdev2->raid_disk < 0 ||
1162 test_bit(Faulty, &rdev2->flags))
1163 is_active = 0;
1164 if (is_active)
1165 desc_nr = rdev2->raid_disk;
1166 else
1167 desc_nr = next_spare++;
1168 rdev2->desc_nr = desc_nr;
1169 d = &sb->disks[rdev2->desc_nr];
1170 nr_disks++;
1171 d->number = rdev2->desc_nr;
1172 d->major = MAJOR(rdev2->bdev->bd_dev);
1173 d->minor = MINOR(rdev2->bdev->bd_dev);
1174 if (is_active)
1175 d->raid_disk = rdev2->raid_disk;
1176 else
1177 d->raid_disk = rdev2->desc_nr; /* compatibility */
1178 if (test_bit(Faulty, &rdev2->flags))
1179 d->state = (1<<MD_DISK_FAULTY);
1180 else if (is_active) {
1181 d->state = (1<<MD_DISK_ACTIVE);
1182 if (test_bit(In_sync, &rdev2->flags))
1183 d->state |= (1<<MD_DISK_SYNC);
1184 active++;
1185 working++;
1186 } else {
1187 d->state = 0;
1188 spare++;
1189 working++;
1190 }
1191 if (test_bit(WriteMostly, &rdev2->flags))
1192 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1193 }
1194 /* now set the "removed" and "faulty" bits on any missing devices */
1195 for (i=0 ; i < mddev->raid_disks ; i++) {
1196 mdp_disk_t *d = &sb->disks[i];
1197 if (d->state == 0 && d->number == 0) {
1198 d->number = i;
1199 d->raid_disk = i;
1200 d->state = (1<<MD_DISK_REMOVED);
1201 d->state |= (1<<MD_DISK_FAULTY);
1202 failed++;
1203 }
1204 }
1205 sb->nr_disks = nr_disks;
1206 sb->active_disks = active;
1207 sb->working_disks = working;
1208 sb->failed_disks = failed;
1209 sb->spare_disks = spare;
1210
1211 sb->this_disk = sb->disks[rdev->desc_nr];
1212 sb->sb_csum = calc_sb_csum(sb);
1213 }
1214
1215 /*
1216 * rdev_size_change for 0.90.0
1217 */
1218 static unsigned long long
1219 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1220 {
1221 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1222 return 0; /* component must fit device */
1223 if (rdev->mddev->bitmap_info.offset)
1224 return 0; /* can't move bitmap */
1225 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1226 if (!num_sectors || num_sectors > rdev->sb_start)
1227 num_sectors = rdev->sb_start;
1228 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1229 rdev->sb_page);
1230 md_super_wait(rdev->mddev);
1231 return num_sectors / 2; /* kB for sysfs */
1232 }
1233
1234
1235 /*
1236 * version 1 superblock
1237 */
1238
1239 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1240 {
1241 __le32 disk_csum;
1242 u32 csum;
1243 unsigned long long newcsum;
1244 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1245 __le32 *isuper = (__le32*)sb;
1246 int i;
1247
1248 disk_csum = sb->sb_csum;
1249 sb->sb_csum = 0;
1250 newcsum = 0;
1251 for (i=0; size>=4; size -= 4 )
1252 newcsum += le32_to_cpu(*isuper++);
1253
1254 if (size == 2)
1255 newcsum += le16_to_cpu(*(__le16*) isuper);
1256
1257 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1258 sb->sb_csum = disk_csum;
1259 return cpu_to_le32(csum);
1260 }
1261
1262 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1263 {
1264 struct mdp_superblock_1 *sb;
1265 int ret;
1266 sector_t sb_start;
1267 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1268 int bmask;
1269
1270 /*
1271 * Calculate the position of the superblock in 512byte sectors.
1272 * It is always aligned to a 4K boundary and
1273 * depeding on minor_version, it can be:
1274 * 0: At least 8K, but less than 12K, from end of device
1275 * 1: At start of device
1276 * 2: 4K from start of device.
1277 */
1278 switch(minor_version) {
1279 case 0:
1280 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1281 sb_start -= 8*2;
1282 sb_start &= ~(sector_t)(4*2-1);
1283 break;
1284 case 1:
1285 sb_start = 0;
1286 break;
1287 case 2:
1288 sb_start = 8;
1289 break;
1290 default:
1291 return -EINVAL;
1292 }
1293 rdev->sb_start = sb_start;
1294
1295 /* superblock is rarely larger than 1K, but it can be larger,
1296 * and it is safe to read 4k, so we do that
1297 */
1298 ret = read_disk_sb(rdev, 4096);
1299 if (ret) return ret;
1300
1301
1302 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1303
1304 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1305 sb->major_version != cpu_to_le32(1) ||
1306 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1307 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1308 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1309 return -EINVAL;
1310
1311 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1312 printk("md: invalid superblock checksum on %s\n",
1313 bdevname(rdev->bdev,b));
1314 return -EINVAL;
1315 }
1316 if (le64_to_cpu(sb->data_size) < 10) {
1317 printk("md: data_size too small on %s\n",
1318 bdevname(rdev->bdev,b));
1319 return -EINVAL;
1320 }
1321
1322 rdev->preferred_minor = 0xffff;
1323 rdev->data_offset = le64_to_cpu(sb->data_offset);
1324 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1325
1326 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1327 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1328 if (rdev->sb_size & bmask)
1329 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1330
1331 if (minor_version
1332 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1333 return -EINVAL;
1334
1335 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1336 rdev->desc_nr = -1;
1337 else
1338 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1339
1340 if (!refdev) {
1341 ret = 1;
1342 } else {
1343 __u64 ev1, ev2;
1344 struct mdp_superblock_1 *refsb =
1345 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1346
1347 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1348 sb->level != refsb->level ||
1349 sb->layout != refsb->layout ||
1350 sb->chunksize != refsb->chunksize) {
1351 printk(KERN_WARNING "md: %s has strangely different"
1352 " superblock to %s\n",
1353 bdevname(rdev->bdev,b),
1354 bdevname(refdev->bdev,b2));
1355 return -EINVAL;
1356 }
1357 ev1 = le64_to_cpu(sb->events);
1358 ev2 = le64_to_cpu(refsb->events);
1359
1360 if (ev1 > ev2)
1361 ret = 1;
1362 else
1363 ret = 0;
1364 }
1365 if (minor_version)
1366 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1367 le64_to_cpu(sb->data_offset);
1368 else
1369 rdev->sectors = rdev->sb_start;
1370 if (rdev->sectors < le64_to_cpu(sb->data_size))
1371 return -EINVAL;
1372 rdev->sectors = le64_to_cpu(sb->data_size);
1373 if (le64_to_cpu(sb->size) > rdev->sectors)
1374 return -EINVAL;
1375 return ret;
1376 }
1377
1378 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1379 {
1380 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1381 __u64 ev1 = le64_to_cpu(sb->events);
1382
1383 rdev->raid_disk = -1;
1384 clear_bit(Faulty, &rdev->flags);
1385 clear_bit(In_sync, &rdev->flags);
1386 clear_bit(WriteMostly, &rdev->flags);
1387 clear_bit(BarriersNotsupp, &rdev->flags);
1388
1389 if (mddev->raid_disks == 0) {
1390 mddev->major_version = 1;
1391 mddev->patch_version = 0;
1392 mddev->external = 0;
1393 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1394 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1395 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1396 mddev->level = le32_to_cpu(sb->level);
1397 mddev->clevel[0] = 0;
1398 mddev->layout = le32_to_cpu(sb->layout);
1399 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1400 mddev->dev_sectors = le64_to_cpu(sb->size);
1401 mddev->events = ev1;
1402 mddev->bitmap_info.offset = 0;
1403 mddev->bitmap_info.default_offset = 1024 >> 9;
1404
1405 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1406 memcpy(mddev->uuid, sb->set_uuid, 16);
1407
1408 mddev->max_disks = (4096-256)/2;
1409
1410 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1411 mddev->bitmap_info.file == NULL )
1412 mddev->bitmap_info.offset =
1413 (__s32)le32_to_cpu(sb->bitmap_offset);
1414
1415 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1416 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1417 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1418 mddev->new_level = le32_to_cpu(sb->new_level);
1419 mddev->new_layout = le32_to_cpu(sb->new_layout);
1420 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1421 } else {
1422 mddev->reshape_position = MaxSector;
1423 mddev->delta_disks = 0;
1424 mddev->new_level = mddev->level;
1425 mddev->new_layout = mddev->layout;
1426 mddev->new_chunk_sectors = mddev->chunk_sectors;
1427 }
1428
1429 } else if (mddev->pers == NULL) {
1430 /* Insist of good event counter while assembling */
1431 ++ev1;
1432 if (ev1 < mddev->events)
1433 return -EINVAL;
1434 } else if (mddev->bitmap) {
1435 /* If adding to array with a bitmap, then we can accept an
1436 * older device, but not too old.
1437 */
1438 if (ev1 < mddev->bitmap->events_cleared)
1439 return 0;
1440 } else {
1441 if (ev1 < mddev->events)
1442 /* just a hot-add of a new device, leave raid_disk at -1 */
1443 return 0;
1444 }
1445 if (mddev->level != LEVEL_MULTIPATH) {
1446 int role;
1447 if (rdev->desc_nr < 0 ||
1448 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1449 role = 0xffff;
1450 rdev->desc_nr = -1;
1451 } else
1452 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1453 switch(role) {
1454 case 0xffff: /* spare */
1455 break;
1456 case 0xfffe: /* faulty */
1457 set_bit(Faulty, &rdev->flags);
1458 break;
1459 default:
1460 if ((le32_to_cpu(sb->feature_map) &
1461 MD_FEATURE_RECOVERY_OFFSET))
1462 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1463 else
1464 set_bit(In_sync, &rdev->flags);
1465 rdev->raid_disk = role;
1466 break;
1467 }
1468 if (sb->devflags & WriteMostly1)
1469 set_bit(WriteMostly, &rdev->flags);
1470 } else /* MULTIPATH are always insync */
1471 set_bit(In_sync, &rdev->flags);
1472
1473 return 0;
1474 }
1475
1476 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1477 {
1478 struct mdp_superblock_1 *sb;
1479 mdk_rdev_t *rdev2;
1480 int max_dev, i;
1481 /* make rdev->sb match mddev and rdev data. */
1482
1483 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1484
1485 sb->feature_map = 0;
1486 sb->pad0 = 0;
1487 sb->recovery_offset = cpu_to_le64(0);
1488 memset(sb->pad1, 0, sizeof(sb->pad1));
1489 memset(sb->pad2, 0, sizeof(sb->pad2));
1490 memset(sb->pad3, 0, sizeof(sb->pad3));
1491
1492 sb->utime = cpu_to_le64((__u64)mddev->utime);
1493 sb->events = cpu_to_le64(mddev->events);
1494 if (mddev->in_sync)
1495 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1496 else
1497 sb->resync_offset = cpu_to_le64(0);
1498
1499 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1500
1501 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1502 sb->size = cpu_to_le64(mddev->dev_sectors);
1503 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1504 sb->level = cpu_to_le32(mddev->level);
1505 sb->layout = cpu_to_le32(mddev->layout);
1506
1507 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1508 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1509 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1510 }
1511
1512 if (rdev->raid_disk >= 0 &&
1513 !test_bit(In_sync, &rdev->flags)) {
1514 sb->feature_map |=
1515 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1516 sb->recovery_offset =
1517 cpu_to_le64(rdev->recovery_offset);
1518 }
1519
1520 if (mddev->reshape_position != MaxSector) {
1521 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1522 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1523 sb->new_layout = cpu_to_le32(mddev->new_layout);
1524 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1525 sb->new_level = cpu_to_le32(mddev->new_level);
1526 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1527 }
1528
1529 max_dev = 0;
1530 list_for_each_entry(rdev2, &mddev->disks, same_set)
1531 if (rdev2->desc_nr+1 > max_dev)
1532 max_dev = rdev2->desc_nr+1;
1533
1534 if (max_dev > le32_to_cpu(sb->max_dev)) {
1535 int bmask;
1536 sb->max_dev = cpu_to_le32(max_dev);
1537 rdev->sb_size = max_dev * 2 + 256;
1538 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1539 if (rdev->sb_size & bmask)
1540 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1541 }
1542 for (i=0; i<max_dev;i++)
1543 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1544
1545 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1546 i = rdev2->desc_nr;
1547 if (test_bit(Faulty, &rdev2->flags))
1548 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1549 else if (test_bit(In_sync, &rdev2->flags))
1550 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1551 else if (rdev2->raid_disk >= 0)
1552 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1553 else
1554 sb->dev_roles[i] = cpu_to_le16(0xffff);
1555 }
1556
1557 sb->sb_csum = calc_sb_1_csum(sb);
1558 }
1559
1560 static unsigned long long
1561 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1562 {
1563 struct mdp_superblock_1 *sb;
1564 sector_t max_sectors;
1565 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1566 return 0; /* component must fit device */
1567 if (rdev->sb_start < rdev->data_offset) {
1568 /* minor versions 1 and 2; superblock before data */
1569 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1570 max_sectors -= rdev->data_offset;
1571 if (!num_sectors || num_sectors > max_sectors)
1572 num_sectors = max_sectors;
1573 } else if (rdev->mddev->bitmap_info.offset) {
1574 /* minor version 0 with bitmap we can't move */
1575 return 0;
1576 } else {
1577 /* minor version 0; superblock after data */
1578 sector_t sb_start;
1579 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1580 sb_start &= ~(sector_t)(4*2 - 1);
1581 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1582 if (!num_sectors || num_sectors > max_sectors)
1583 num_sectors = max_sectors;
1584 rdev->sb_start = sb_start;
1585 }
1586 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1587 sb->data_size = cpu_to_le64(num_sectors);
1588 sb->super_offset = rdev->sb_start;
1589 sb->sb_csum = calc_sb_1_csum(sb);
1590 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1591 rdev->sb_page);
1592 md_super_wait(rdev->mddev);
1593 return num_sectors / 2; /* kB for sysfs */
1594 }
1595
1596 static struct super_type super_types[] = {
1597 [0] = {
1598 .name = "0.90.0",
1599 .owner = THIS_MODULE,
1600 .load_super = super_90_load,
1601 .validate_super = super_90_validate,
1602 .sync_super = super_90_sync,
1603 .rdev_size_change = super_90_rdev_size_change,
1604 },
1605 [1] = {
1606 .name = "md-1",
1607 .owner = THIS_MODULE,
1608 .load_super = super_1_load,
1609 .validate_super = super_1_validate,
1610 .sync_super = super_1_sync,
1611 .rdev_size_change = super_1_rdev_size_change,
1612 },
1613 };
1614
1615 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1616 {
1617 mdk_rdev_t *rdev, *rdev2;
1618
1619 rcu_read_lock();
1620 rdev_for_each_rcu(rdev, mddev1)
1621 rdev_for_each_rcu(rdev2, mddev2)
1622 if (rdev->bdev->bd_contains ==
1623 rdev2->bdev->bd_contains) {
1624 rcu_read_unlock();
1625 return 1;
1626 }
1627 rcu_read_unlock();
1628 return 0;
1629 }
1630
1631 static LIST_HEAD(pending_raid_disks);
1632
1633 /*
1634 * Try to register data integrity profile for an mddev
1635 *
1636 * This is called when an array is started and after a disk has been kicked
1637 * from the array. It only succeeds if all working and active component devices
1638 * are integrity capable with matching profiles.
1639 */
1640 int md_integrity_register(mddev_t *mddev)
1641 {
1642 mdk_rdev_t *rdev, *reference = NULL;
1643
1644 if (list_empty(&mddev->disks))
1645 return 0; /* nothing to do */
1646 if (blk_get_integrity(mddev->gendisk))
1647 return 0; /* already registered */
1648 list_for_each_entry(rdev, &mddev->disks, same_set) {
1649 /* skip spares and non-functional disks */
1650 if (test_bit(Faulty, &rdev->flags))
1651 continue;
1652 if (rdev->raid_disk < 0)
1653 continue;
1654 /*
1655 * If at least one rdev is not integrity capable, we can not
1656 * enable data integrity for the md device.
1657 */
1658 if (!bdev_get_integrity(rdev->bdev))
1659 return -EINVAL;
1660 if (!reference) {
1661 /* Use the first rdev as the reference */
1662 reference = rdev;
1663 continue;
1664 }
1665 /* does this rdev's profile match the reference profile? */
1666 if (blk_integrity_compare(reference->bdev->bd_disk,
1667 rdev->bdev->bd_disk) < 0)
1668 return -EINVAL;
1669 }
1670 /*
1671 * All component devices are integrity capable and have matching
1672 * profiles, register the common profile for the md device.
1673 */
1674 if (blk_integrity_register(mddev->gendisk,
1675 bdev_get_integrity(reference->bdev)) != 0) {
1676 printk(KERN_ERR "md: failed to register integrity for %s\n",
1677 mdname(mddev));
1678 return -EINVAL;
1679 }
1680 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1681 mdname(mddev));
1682 return 0;
1683 }
1684 EXPORT_SYMBOL(md_integrity_register);
1685
1686 /* Disable data integrity if non-capable/non-matching disk is being added */
1687 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1688 {
1689 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1690 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1691
1692 if (!bi_mddev) /* nothing to do */
1693 return;
1694 if (rdev->raid_disk < 0) /* skip spares */
1695 return;
1696 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1697 rdev->bdev->bd_disk) >= 0)
1698 return;
1699 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1700 blk_integrity_unregister(mddev->gendisk);
1701 }
1702 EXPORT_SYMBOL(md_integrity_add_rdev);
1703
1704 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1705 {
1706 char b[BDEVNAME_SIZE];
1707 struct kobject *ko;
1708 char *s;
1709 int err;
1710
1711 if (rdev->mddev) {
1712 MD_BUG();
1713 return -EINVAL;
1714 }
1715
1716 /* prevent duplicates */
1717 if (find_rdev(mddev, rdev->bdev->bd_dev))
1718 return -EEXIST;
1719
1720 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1721 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1722 rdev->sectors < mddev->dev_sectors)) {
1723 if (mddev->pers) {
1724 /* Cannot change size, so fail
1725 * If mddev->level <= 0, then we don't care
1726 * about aligning sizes (e.g. linear)
1727 */
1728 if (mddev->level > 0)
1729 return -ENOSPC;
1730 } else
1731 mddev->dev_sectors = rdev->sectors;
1732 }
1733
1734 /* Verify rdev->desc_nr is unique.
1735 * If it is -1, assign a free number, else
1736 * check number is not in use
1737 */
1738 if (rdev->desc_nr < 0) {
1739 int choice = 0;
1740 if (mddev->pers) choice = mddev->raid_disks;
1741 while (find_rdev_nr(mddev, choice))
1742 choice++;
1743 rdev->desc_nr = choice;
1744 } else {
1745 if (find_rdev_nr(mddev, rdev->desc_nr))
1746 return -EBUSY;
1747 }
1748 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1749 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1750 mdname(mddev), mddev->max_disks);
1751 return -EBUSY;
1752 }
1753 bdevname(rdev->bdev,b);
1754 while ( (s=strchr(b, '/')) != NULL)
1755 *s = '!';
1756
1757 rdev->mddev = mddev;
1758 printk(KERN_INFO "md: bind<%s>\n", b);
1759
1760 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1761 goto fail;
1762
1763 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1764 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1765 kobject_del(&rdev->kobj);
1766 goto fail;
1767 }
1768 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1769
1770 list_add_rcu(&rdev->same_set, &mddev->disks);
1771 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1772
1773 /* May as well allow recovery to be retried once */
1774 mddev->recovery_disabled = 0;
1775
1776 return 0;
1777
1778 fail:
1779 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1780 b, mdname(mddev));
1781 return err;
1782 }
1783
1784 static void md_delayed_delete(struct work_struct *ws)
1785 {
1786 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1787 kobject_del(&rdev->kobj);
1788 kobject_put(&rdev->kobj);
1789 }
1790
1791 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1792 {
1793 char b[BDEVNAME_SIZE];
1794 if (!rdev->mddev) {
1795 MD_BUG();
1796 return;
1797 }
1798 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1799 list_del_rcu(&rdev->same_set);
1800 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1801 rdev->mddev = NULL;
1802 sysfs_remove_link(&rdev->kobj, "block");
1803 sysfs_put(rdev->sysfs_state);
1804 rdev->sysfs_state = NULL;
1805 /* We need to delay this, otherwise we can deadlock when
1806 * writing to 'remove' to "dev/state". We also need
1807 * to delay it due to rcu usage.
1808 */
1809 synchronize_rcu();
1810 INIT_WORK(&rdev->del_work, md_delayed_delete);
1811 kobject_get(&rdev->kobj);
1812 schedule_work(&rdev->del_work);
1813 }
1814
1815 /*
1816 * prevent the device from being mounted, repartitioned or
1817 * otherwise reused by a RAID array (or any other kernel
1818 * subsystem), by bd_claiming the device.
1819 */
1820 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1821 {
1822 int err = 0;
1823 struct block_device *bdev;
1824 char b[BDEVNAME_SIZE];
1825
1826 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1827 if (IS_ERR(bdev)) {
1828 printk(KERN_ERR "md: could not open %s.\n",
1829 __bdevname(dev, b));
1830 return PTR_ERR(bdev);
1831 }
1832 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1833 if (err) {
1834 printk(KERN_ERR "md: could not bd_claim %s.\n",
1835 bdevname(bdev, b));
1836 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1837 return err;
1838 }
1839 if (!shared)
1840 set_bit(AllReserved, &rdev->flags);
1841 rdev->bdev = bdev;
1842 return err;
1843 }
1844
1845 static void unlock_rdev(mdk_rdev_t *rdev)
1846 {
1847 struct block_device *bdev = rdev->bdev;
1848 rdev->bdev = NULL;
1849 if (!bdev)
1850 MD_BUG();
1851 bd_release(bdev);
1852 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1853 }
1854
1855 void md_autodetect_dev(dev_t dev);
1856
1857 static void export_rdev(mdk_rdev_t * rdev)
1858 {
1859 char b[BDEVNAME_SIZE];
1860 printk(KERN_INFO "md: export_rdev(%s)\n",
1861 bdevname(rdev->bdev,b));
1862 if (rdev->mddev)
1863 MD_BUG();
1864 free_disk_sb(rdev);
1865 #ifndef MODULE
1866 if (test_bit(AutoDetected, &rdev->flags))
1867 md_autodetect_dev(rdev->bdev->bd_dev);
1868 #endif
1869 unlock_rdev(rdev);
1870 kobject_put(&rdev->kobj);
1871 }
1872
1873 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1874 {
1875 unbind_rdev_from_array(rdev);
1876 export_rdev(rdev);
1877 }
1878
1879 static void export_array(mddev_t *mddev)
1880 {
1881 mdk_rdev_t *rdev, *tmp;
1882
1883 rdev_for_each(rdev, tmp, mddev) {
1884 if (!rdev->mddev) {
1885 MD_BUG();
1886 continue;
1887 }
1888 kick_rdev_from_array(rdev);
1889 }
1890 if (!list_empty(&mddev->disks))
1891 MD_BUG();
1892 mddev->raid_disks = 0;
1893 mddev->major_version = 0;
1894 }
1895
1896 static void print_desc(mdp_disk_t *desc)
1897 {
1898 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1899 desc->major,desc->minor,desc->raid_disk,desc->state);
1900 }
1901
1902 static void print_sb_90(mdp_super_t *sb)
1903 {
1904 int i;
1905
1906 printk(KERN_INFO
1907 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1908 sb->major_version, sb->minor_version, sb->patch_version,
1909 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1910 sb->ctime);
1911 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1912 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1913 sb->md_minor, sb->layout, sb->chunk_size);
1914 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1915 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1916 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1917 sb->failed_disks, sb->spare_disks,
1918 sb->sb_csum, (unsigned long)sb->events_lo);
1919
1920 printk(KERN_INFO);
1921 for (i = 0; i < MD_SB_DISKS; i++) {
1922 mdp_disk_t *desc;
1923
1924 desc = sb->disks + i;
1925 if (desc->number || desc->major || desc->minor ||
1926 desc->raid_disk || (desc->state && (desc->state != 4))) {
1927 printk(" D %2d: ", i);
1928 print_desc(desc);
1929 }
1930 }
1931 printk(KERN_INFO "md: THIS: ");
1932 print_desc(&sb->this_disk);
1933 }
1934
1935 static void print_sb_1(struct mdp_superblock_1 *sb)
1936 {
1937 __u8 *uuid;
1938
1939 uuid = sb->set_uuid;
1940 printk(KERN_INFO
1941 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
1942 "md: Name: \"%s\" CT:%llu\n",
1943 le32_to_cpu(sb->major_version),
1944 le32_to_cpu(sb->feature_map),
1945 uuid,
1946 sb->set_name,
1947 (unsigned long long)le64_to_cpu(sb->ctime)
1948 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1949
1950 uuid = sb->device_uuid;
1951 printk(KERN_INFO
1952 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1953 " RO:%llu\n"
1954 "md: Dev:%08x UUID: %pU\n"
1955 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1956 "md: (MaxDev:%u) \n",
1957 le32_to_cpu(sb->level),
1958 (unsigned long long)le64_to_cpu(sb->size),
1959 le32_to_cpu(sb->raid_disks),
1960 le32_to_cpu(sb->layout),
1961 le32_to_cpu(sb->chunksize),
1962 (unsigned long long)le64_to_cpu(sb->data_offset),
1963 (unsigned long long)le64_to_cpu(sb->data_size),
1964 (unsigned long long)le64_to_cpu(sb->super_offset),
1965 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1966 le32_to_cpu(sb->dev_number),
1967 uuid,
1968 sb->devflags,
1969 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1970 (unsigned long long)le64_to_cpu(sb->events),
1971 (unsigned long long)le64_to_cpu(sb->resync_offset),
1972 le32_to_cpu(sb->sb_csum),
1973 le32_to_cpu(sb->max_dev)
1974 );
1975 }
1976
1977 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1978 {
1979 char b[BDEVNAME_SIZE];
1980 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1981 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1982 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1983 rdev->desc_nr);
1984 if (rdev->sb_loaded) {
1985 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1986 switch (major_version) {
1987 case 0:
1988 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1989 break;
1990 case 1:
1991 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1992 break;
1993 }
1994 } else
1995 printk(KERN_INFO "md: no rdev superblock!\n");
1996 }
1997
1998 static void md_print_devices(void)
1999 {
2000 struct list_head *tmp;
2001 mdk_rdev_t *rdev;
2002 mddev_t *mddev;
2003 char b[BDEVNAME_SIZE];
2004
2005 printk("\n");
2006 printk("md: **********************************\n");
2007 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2008 printk("md: **********************************\n");
2009 for_each_mddev(mddev, tmp) {
2010
2011 if (mddev->bitmap)
2012 bitmap_print_sb(mddev->bitmap);
2013 else
2014 printk("%s: ", mdname(mddev));
2015 list_for_each_entry(rdev, &mddev->disks, same_set)
2016 printk("<%s>", bdevname(rdev->bdev,b));
2017 printk("\n");
2018
2019 list_for_each_entry(rdev, &mddev->disks, same_set)
2020 print_rdev(rdev, mddev->major_version);
2021 }
2022 printk("md: **********************************\n");
2023 printk("\n");
2024 }
2025
2026
2027 static void sync_sbs(mddev_t * mddev, int nospares)
2028 {
2029 /* Update each superblock (in-memory image), but
2030 * if we are allowed to, skip spares which already
2031 * have the right event counter, or have one earlier
2032 * (which would mean they aren't being marked as dirty
2033 * with the rest of the array)
2034 */
2035 mdk_rdev_t *rdev;
2036
2037 /* First make sure individual recovery_offsets are correct */
2038 list_for_each_entry(rdev, &mddev->disks, same_set) {
2039 if (rdev->raid_disk >= 0 &&
2040 !test_bit(In_sync, &rdev->flags) &&
2041 mddev->curr_resync_completed > rdev->recovery_offset)
2042 rdev->recovery_offset = mddev->curr_resync_completed;
2043
2044 }
2045 list_for_each_entry(rdev, &mddev->disks, same_set) {
2046 if (rdev->sb_events == mddev->events ||
2047 (nospares &&
2048 rdev->raid_disk < 0 &&
2049 (rdev->sb_events&1)==0 &&
2050 rdev->sb_events+1 == mddev->events)) {
2051 /* Don't update this superblock */
2052 rdev->sb_loaded = 2;
2053 } else {
2054 super_types[mddev->major_version].
2055 sync_super(mddev, rdev);
2056 rdev->sb_loaded = 1;
2057 }
2058 }
2059 }
2060
2061 static void md_update_sb(mddev_t * mddev, int force_change)
2062 {
2063 mdk_rdev_t *rdev;
2064 int sync_req;
2065 int nospares = 0;
2066
2067 mddev->utime = get_seconds();
2068 if (mddev->external)
2069 return;
2070 repeat:
2071 spin_lock_irq(&mddev->write_lock);
2072
2073 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2074 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2075 force_change = 1;
2076 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2077 /* just a clean<-> dirty transition, possibly leave spares alone,
2078 * though if events isn't the right even/odd, we will have to do
2079 * spares after all
2080 */
2081 nospares = 1;
2082 if (force_change)
2083 nospares = 0;
2084 if (mddev->degraded)
2085 /* If the array is degraded, then skipping spares is both
2086 * dangerous and fairly pointless.
2087 * Dangerous because a device that was removed from the array
2088 * might have a event_count that still looks up-to-date,
2089 * so it can be re-added without a resync.
2090 * Pointless because if there are any spares to skip,
2091 * then a recovery will happen and soon that array won't
2092 * be degraded any more and the spare can go back to sleep then.
2093 */
2094 nospares = 0;
2095
2096 sync_req = mddev->in_sync;
2097
2098 /* If this is just a dirty<->clean transition, and the array is clean
2099 * and 'events' is odd, we can roll back to the previous clean state */
2100 if (nospares
2101 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2102 && (mddev->events & 1)
2103 && mddev->events != 1)
2104 mddev->events--;
2105 else {
2106 /* otherwise we have to go forward and ... */
2107 mddev->events ++;
2108 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
2109 /* .. if the array isn't clean, an 'even' event must also go
2110 * to spares. */
2111 if ((mddev->events&1)==0)
2112 nospares = 0;
2113 } else {
2114 /* otherwise an 'odd' event must go to spares */
2115 if ((mddev->events&1))
2116 nospares = 0;
2117 }
2118 }
2119
2120 if (!mddev->events) {
2121 /*
2122 * oops, this 64-bit counter should never wrap.
2123 * Either we are in around ~1 trillion A.C., assuming
2124 * 1 reboot per second, or we have a bug:
2125 */
2126 MD_BUG();
2127 mddev->events --;
2128 }
2129
2130 /*
2131 * do not write anything to disk if using
2132 * nonpersistent superblocks
2133 */
2134 if (!mddev->persistent) {
2135 if (!mddev->external)
2136 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2137
2138 spin_unlock_irq(&mddev->write_lock);
2139 wake_up(&mddev->sb_wait);
2140 return;
2141 }
2142 sync_sbs(mddev, nospares);
2143 spin_unlock_irq(&mddev->write_lock);
2144
2145 dprintk(KERN_INFO
2146 "md: updating %s RAID superblock on device (in sync %d)\n",
2147 mdname(mddev),mddev->in_sync);
2148
2149 bitmap_update_sb(mddev->bitmap);
2150 list_for_each_entry(rdev, &mddev->disks, same_set) {
2151 char b[BDEVNAME_SIZE];
2152 dprintk(KERN_INFO "md: ");
2153 if (rdev->sb_loaded != 1)
2154 continue; /* no noise on spare devices */
2155 if (test_bit(Faulty, &rdev->flags))
2156 dprintk("(skipping faulty ");
2157
2158 dprintk("%s ", bdevname(rdev->bdev,b));
2159 if (!test_bit(Faulty, &rdev->flags)) {
2160 md_super_write(mddev,rdev,
2161 rdev->sb_start, rdev->sb_size,
2162 rdev->sb_page);
2163 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2164 bdevname(rdev->bdev,b),
2165 (unsigned long long)rdev->sb_start);
2166 rdev->sb_events = mddev->events;
2167
2168 } else
2169 dprintk(")\n");
2170 if (mddev->level == LEVEL_MULTIPATH)
2171 /* only need to write one superblock... */
2172 break;
2173 }
2174 md_super_wait(mddev);
2175 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2176
2177 spin_lock_irq(&mddev->write_lock);
2178 if (mddev->in_sync != sync_req ||
2179 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2180 /* have to write it out again */
2181 spin_unlock_irq(&mddev->write_lock);
2182 goto repeat;
2183 }
2184 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2185 spin_unlock_irq(&mddev->write_lock);
2186 wake_up(&mddev->sb_wait);
2187 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2188 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2189
2190 }
2191
2192 /* words written to sysfs files may, or may not, be \n terminated.
2193 * We want to accept with case. For this we use cmd_match.
2194 */
2195 static int cmd_match(const char *cmd, const char *str)
2196 {
2197 /* See if cmd, written into a sysfs file, matches
2198 * str. They must either be the same, or cmd can
2199 * have a trailing newline
2200 */
2201 while (*cmd && *str && *cmd == *str) {
2202 cmd++;
2203 str++;
2204 }
2205 if (*cmd == '\n')
2206 cmd++;
2207 if (*str || *cmd)
2208 return 0;
2209 return 1;
2210 }
2211
2212 struct rdev_sysfs_entry {
2213 struct attribute attr;
2214 ssize_t (*show)(mdk_rdev_t *, char *);
2215 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2216 };
2217
2218 static ssize_t
2219 state_show(mdk_rdev_t *rdev, char *page)
2220 {
2221 char *sep = "";
2222 size_t len = 0;
2223
2224 if (test_bit(Faulty, &rdev->flags)) {
2225 len+= sprintf(page+len, "%sfaulty",sep);
2226 sep = ",";
2227 }
2228 if (test_bit(In_sync, &rdev->flags)) {
2229 len += sprintf(page+len, "%sin_sync",sep);
2230 sep = ",";
2231 }
2232 if (test_bit(WriteMostly, &rdev->flags)) {
2233 len += sprintf(page+len, "%swrite_mostly",sep);
2234 sep = ",";
2235 }
2236 if (test_bit(Blocked, &rdev->flags)) {
2237 len += sprintf(page+len, "%sblocked", sep);
2238 sep = ",";
2239 }
2240 if (!test_bit(Faulty, &rdev->flags) &&
2241 !test_bit(In_sync, &rdev->flags)) {
2242 len += sprintf(page+len, "%sspare", sep);
2243 sep = ",";
2244 }
2245 return len+sprintf(page+len, "\n");
2246 }
2247
2248 static ssize_t
2249 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2250 {
2251 /* can write
2252 * faulty - simulates and error
2253 * remove - disconnects the device
2254 * writemostly - sets write_mostly
2255 * -writemostly - clears write_mostly
2256 * blocked - sets the Blocked flag
2257 * -blocked - clears the Blocked flag
2258 * insync - sets Insync providing device isn't active
2259 */
2260 int err = -EINVAL;
2261 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2262 md_error(rdev->mddev, rdev);
2263 err = 0;
2264 } else if (cmd_match(buf, "remove")) {
2265 if (rdev->raid_disk >= 0)
2266 err = -EBUSY;
2267 else {
2268 mddev_t *mddev = rdev->mddev;
2269 kick_rdev_from_array(rdev);
2270 if (mddev->pers)
2271 md_update_sb(mddev, 1);
2272 md_new_event(mddev);
2273 err = 0;
2274 }
2275 } else if (cmd_match(buf, "writemostly")) {
2276 set_bit(WriteMostly, &rdev->flags);
2277 err = 0;
2278 } else if (cmd_match(buf, "-writemostly")) {
2279 clear_bit(WriteMostly, &rdev->flags);
2280 err = 0;
2281 } else if (cmd_match(buf, "blocked")) {
2282 set_bit(Blocked, &rdev->flags);
2283 err = 0;
2284 } else if (cmd_match(buf, "-blocked")) {
2285 clear_bit(Blocked, &rdev->flags);
2286 wake_up(&rdev->blocked_wait);
2287 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2288 md_wakeup_thread(rdev->mddev->thread);
2289
2290 err = 0;
2291 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2292 set_bit(In_sync, &rdev->flags);
2293 err = 0;
2294 }
2295 if (!err && rdev->sysfs_state)
2296 sysfs_notify_dirent(rdev->sysfs_state);
2297 return err ? err : len;
2298 }
2299 static struct rdev_sysfs_entry rdev_state =
2300 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2301
2302 static ssize_t
2303 errors_show(mdk_rdev_t *rdev, char *page)
2304 {
2305 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2306 }
2307
2308 static ssize_t
2309 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2310 {
2311 char *e;
2312 unsigned long n = simple_strtoul(buf, &e, 10);
2313 if (*buf && (*e == 0 || *e == '\n')) {
2314 atomic_set(&rdev->corrected_errors, n);
2315 return len;
2316 }
2317 return -EINVAL;
2318 }
2319 static struct rdev_sysfs_entry rdev_errors =
2320 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2321
2322 static ssize_t
2323 slot_show(mdk_rdev_t *rdev, char *page)
2324 {
2325 if (rdev->raid_disk < 0)
2326 return sprintf(page, "none\n");
2327 else
2328 return sprintf(page, "%d\n", rdev->raid_disk);
2329 }
2330
2331 static ssize_t
2332 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2333 {
2334 char *e;
2335 int err;
2336 char nm[20];
2337 int slot = simple_strtoul(buf, &e, 10);
2338 if (strncmp(buf, "none", 4)==0)
2339 slot = -1;
2340 else if (e==buf || (*e && *e!= '\n'))
2341 return -EINVAL;
2342 if (rdev->mddev->pers && slot == -1) {
2343 /* Setting 'slot' on an active array requires also
2344 * updating the 'rd%d' link, and communicating
2345 * with the personality with ->hot_*_disk.
2346 * For now we only support removing
2347 * failed/spare devices. This normally happens automatically,
2348 * but not when the metadata is externally managed.
2349 */
2350 if (rdev->raid_disk == -1)
2351 return -EEXIST;
2352 /* personality does all needed checks */
2353 if (rdev->mddev->pers->hot_add_disk == NULL)
2354 return -EINVAL;
2355 err = rdev->mddev->pers->
2356 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2357 if (err)
2358 return err;
2359 sprintf(nm, "rd%d", rdev->raid_disk);
2360 sysfs_remove_link(&rdev->mddev->kobj, nm);
2361 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2362 md_wakeup_thread(rdev->mddev->thread);
2363 } else if (rdev->mddev->pers) {
2364 mdk_rdev_t *rdev2;
2365 /* Activating a spare .. or possibly reactivating
2366 * if we ever get bitmaps working here.
2367 */
2368
2369 if (rdev->raid_disk != -1)
2370 return -EBUSY;
2371
2372 if (rdev->mddev->pers->hot_add_disk == NULL)
2373 return -EINVAL;
2374
2375 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2376 if (rdev2->raid_disk == slot)
2377 return -EEXIST;
2378
2379 rdev->raid_disk = slot;
2380 if (test_bit(In_sync, &rdev->flags))
2381 rdev->saved_raid_disk = slot;
2382 else
2383 rdev->saved_raid_disk = -1;
2384 err = rdev->mddev->pers->
2385 hot_add_disk(rdev->mddev, rdev);
2386 if (err) {
2387 rdev->raid_disk = -1;
2388 return err;
2389 } else
2390 sysfs_notify_dirent(rdev->sysfs_state);
2391 sprintf(nm, "rd%d", rdev->raid_disk);
2392 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2393 printk(KERN_WARNING
2394 "md: cannot register "
2395 "%s for %s\n",
2396 nm, mdname(rdev->mddev));
2397
2398 /* don't wakeup anyone, leave that to userspace. */
2399 } else {
2400 if (slot >= rdev->mddev->raid_disks)
2401 return -ENOSPC;
2402 rdev->raid_disk = slot;
2403 /* assume it is working */
2404 clear_bit(Faulty, &rdev->flags);
2405 clear_bit(WriteMostly, &rdev->flags);
2406 set_bit(In_sync, &rdev->flags);
2407 sysfs_notify_dirent(rdev->sysfs_state);
2408 }
2409 return len;
2410 }
2411
2412
2413 static struct rdev_sysfs_entry rdev_slot =
2414 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2415
2416 static ssize_t
2417 offset_show(mdk_rdev_t *rdev, char *page)
2418 {
2419 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2420 }
2421
2422 static ssize_t
2423 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2424 {
2425 char *e;
2426 unsigned long long offset = simple_strtoull(buf, &e, 10);
2427 if (e==buf || (*e && *e != '\n'))
2428 return -EINVAL;
2429 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2430 return -EBUSY;
2431 if (rdev->sectors && rdev->mddev->external)
2432 /* Must set offset before size, so overlap checks
2433 * can be sane */
2434 return -EBUSY;
2435 rdev->data_offset = offset;
2436 return len;
2437 }
2438
2439 static struct rdev_sysfs_entry rdev_offset =
2440 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2441
2442 static ssize_t
2443 rdev_size_show(mdk_rdev_t *rdev, char *page)
2444 {
2445 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2446 }
2447
2448 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2449 {
2450 /* check if two start/length pairs overlap */
2451 if (s1+l1 <= s2)
2452 return 0;
2453 if (s2+l2 <= s1)
2454 return 0;
2455 return 1;
2456 }
2457
2458 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2459 {
2460 unsigned long long blocks;
2461 sector_t new;
2462
2463 if (strict_strtoull(buf, 10, &blocks) < 0)
2464 return -EINVAL;
2465
2466 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2467 return -EINVAL; /* sector conversion overflow */
2468
2469 new = blocks * 2;
2470 if (new != blocks * 2)
2471 return -EINVAL; /* unsigned long long to sector_t overflow */
2472
2473 *sectors = new;
2474 return 0;
2475 }
2476
2477 static ssize_t
2478 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2479 {
2480 mddev_t *my_mddev = rdev->mddev;
2481 sector_t oldsectors = rdev->sectors;
2482 sector_t sectors;
2483
2484 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2485 return -EINVAL;
2486 if (my_mddev->pers && rdev->raid_disk >= 0) {
2487 if (my_mddev->persistent) {
2488 sectors = super_types[my_mddev->major_version].
2489 rdev_size_change(rdev, sectors);
2490 if (!sectors)
2491 return -EBUSY;
2492 } else if (!sectors)
2493 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2494 rdev->data_offset;
2495 }
2496 if (sectors < my_mddev->dev_sectors)
2497 return -EINVAL; /* component must fit device */
2498
2499 rdev->sectors = sectors;
2500 if (sectors > oldsectors && my_mddev->external) {
2501 /* need to check that all other rdevs with the same ->bdev
2502 * do not overlap. We need to unlock the mddev to avoid
2503 * a deadlock. We have already changed rdev->sectors, and if
2504 * we have to change it back, we will have the lock again.
2505 */
2506 mddev_t *mddev;
2507 int overlap = 0;
2508 struct list_head *tmp;
2509
2510 mddev_unlock(my_mddev);
2511 for_each_mddev(mddev, tmp) {
2512 mdk_rdev_t *rdev2;
2513
2514 mddev_lock(mddev);
2515 list_for_each_entry(rdev2, &mddev->disks, same_set)
2516 if (test_bit(AllReserved, &rdev2->flags) ||
2517 (rdev->bdev == rdev2->bdev &&
2518 rdev != rdev2 &&
2519 overlaps(rdev->data_offset, rdev->sectors,
2520 rdev2->data_offset,
2521 rdev2->sectors))) {
2522 overlap = 1;
2523 break;
2524 }
2525 mddev_unlock(mddev);
2526 if (overlap) {
2527 mddev_put(mddev);
2528 break;
2529 }
2530 }
2531 mddev_lock(my_mddev);
2532 if (overlap) {
2533 /* Someone else could have slipped in a size
2534 * change here, but doing so is just silly.
2535 * We put oldsectors back because we *know* it is
2536 * safe, and trust userspace not to race with
2537 * itself
2538 */
2539 rdev->sectors = oldsectors;
2540 return -EBUSY;
2541 }
2542 }
2543 return len;
2544 }
2545
2546 static struct rdev_sysfs_entry rdev_size =
2547 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2548
2549
2550 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2551 {
2552 unsigned long long recovery_start = rdev->recovery_offset;
2553
2554 if (test_bit(In_sync, &rdev->flags) ||
2555 recovery_start == MaxSector)
2556 return sprintf(page, "none\n");
2557
2558 return sprintf(page, "%llu\n", recovery_start);
2559 }
2560
2561 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2562 {
2563 unsigned long long recovery_start;
2564
2565 if (cmd_match(buf, "none"))
2566 recovery_start = MaxSector;
2567 else if (strict_strtoull(buf, 10, &recovery_start))
2568 return -EINVAL;
2569
2570 if (rdev->mddev->pers &&
2571 rdev->raid_disk >= 0)
2572 return -EBUSY;
2573
2574 rdev->recovery_offset = recovery_start;
2575 if (recovery_start == MaxSector)
2576 set_bit(In_sync, &rdev->flags);
2577 else
2578 clear_bit(In_sync, &rdev->flags);
2579 return len;
2580 }
2581
2582 static struct rdev_sysfs_entry rdev_recovery_start =
2583 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2584
2585 static struct attribute *rdev_default_attrs[] = {
2586 &rdev_state.attr,
2587 &rdev_errors.attr,
2588 &rdev_slot.attr,
2589 &rdev_offset.attr,
2590 &rdev_size.attr,
2591 &rdev_recovery_start.attr,
2592 NULL,
2593 };
2594 static ssize_t
2595 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2596 {
2597 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2598 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2599 mddev_t *mddev = rdev->mddev;
2600 ssize_t rv;
2601
2602 if (!entry->show)
2603 return -EIO;
2604
2605 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2606 if (!rv) {
2607 if (rdev->mddev == NULL)
2608 rv = -EBUSY;
2609 else
2610 rv = entry->show(rdev, page);
2611 mddev_unlock(mddev);
2612 }
2613 return rv;
2614 }
2615
2616 static ssize_t
2617 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2618 const char *page, size_t length)
2619 {
2620 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2621 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2622 ssize_t rv;
2623 mddev_t *mddev = rdev->mddev;
2624
2625 if (!entry->store)
2626 return -EIO;
2627 if (!capable(CAP_SYS_ADMIN))
2628 return -EACCES;
2629 rv = mddev ? mddev_lock(mddev): -EBUSY;
2630 if (!rv) {
2631 if (rdev->mddev == NULL)
2632 rv = -EBUSY;
2633 else
2634 rv = entry->store(rdev, page, length);
2635 mddev_unlock(mddev);
2636 }
2637 return rv;
2638 }
2639
2640 static void rdev_free(struct kobject *ko)
2641 {
2642 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2643 kfree(rdev);
2644 }
2645 static const struct sysfs_ops rdev_sysfs_ops = {
2646 .show = rdev_attr_show,
2647 .store = rdev_attr_store,
2648 };
2649 static struct kobj_type rdev_ktype = {
2650 .release = rdev_free,
2651 .sysfs_ops = &rdev_sysfs_ops,
2652 .default_attrs = rdev_default_attrs,
2653 };
2654
2655 /*
2656 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2657 *
2658 * mark the device faulty if:
2659 *
2660 * - the device is nonexistent (zero size)
2661 * - the device has no valid superblock
2662 *
2663 * a faulty rdev _never_ has rdev->sb set.
2664 */
2665 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2666 {
2667 char b[BDEVNAME_SIZE];
2668 int err;
2669 mdk_rdev_t *rdev;
2670 sector_t size;
2671
2672 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2673 if (!rdev) {
2674 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2675 return ERR_PTR(-ENOMEM);
2676 }
2677
2678 if ((err = alloc_disk_sb(rdev)))
2679 goto abort_free;
2680
2681 err = lock_rdev(rdev, newdev, super_format == -2);
2682 if (err)
2683 goto abort_free;
2684
2685 kobject_init(&rdev->kobj, &rdev_ktype);
2686
2687 rdev->desc_nr = -1;
2688 rdev->saved_raid_disk = -1;
2689 rdev->raid_disk = -1;
2690 rdev->flags = 0;
2691 rdev->data_offset = 0;
2692 rdev->sb_events = 0;
2693 rdev->last_read_error.tv_sec = 0;
2694 rdev->last_read_error.tv_nsec = 0;
2695 atomic_set(&rdev->nr_pending, 0);
2696 atomic_set(&rdev->read_errors, 0);
2697 atomic_set(&rdev->corrected_errors, 0);
2698
2699 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2700 if (!size) {
2701 printk(KERN_WARNING
2702 "md: %s has zero or unknown size, marking faulty!\n",
2703 bdevname(rdev->bdev,b));
2704 err = -EINVAL;
2705 goto abort_free;
2706 }
2707
2708 if (super_format >= 0) {
2709 err = super_types[super_format].
2710 load_super(rdev, NULL, super_minor);
2711 if (err == -EINVAL) {
2712 printk(KERN_WARNING
2713 "md: %s does not have a valid v%d.%d "
2714 "superblock, not importing!\n",
2715 bdevname(rdev->bdev,b),
2716 super_format, super_minor);
2717 goto abort_free;
2718 }
2719 if (err < 0) {
2720 printk(KERN_WARNING
2721 "md: could not read %s's sb, not importing!\n",
2722 bdevname(rdev->bdev,b));
2723 goto abort_free;
2724 }
2725 }
2726
2727 INIT_LIST_HEAD(&rdev->same_set);
2728 init_waitqueue_head(&rdev->blocked_wait);
2729
2730 return rdev;
2731
2732 abort_free:
2733 if (rdev->sb_page) {
2734 if (rdev->bdev)
2735 unlock_rdev(rdev);
2736 free_disk_sb(rdev);
2737 }
2738 kfree(rdev);
2739 return ERR_PTR(err);
2740 }
2741
2742 /*
2743 * Check a full RAID array for plausibility
2744 */
2745
2746
2747 static void analyze_sbs(mddev_t * mddev)
2748 {
2749 int i;
2750 mdk_rdev_t *rdev, *freshest, *tmp;
2751 char b[BDEVNAME_SIZE];
2752
2753 freshest = NULL;
2754 rdev_for_each(rdev, tmp, mddev)
2755 switch (super_types[mddev->major_version].
2756 load_super(rdev, freshest, mddev->minor_version)) {
2757 case 1:
2758 freshest = rdev;
2759 break;
2760 case 0:
2761 break;
2762 default:
2763 printk( KERN_ERR \
2764 "md: fatal superblock inconsistency in %s"
2765 " -- removing from array\n",
2766 bdevname(rdev->bdev,b));
2767 kick_rdev_from_array(rdev);
2768 }
2769
2770
2771 super_types[mddev->major_version].
2772 validate_super(mddev, freshest);
2773
2774 i = 0;
2775 rdev_for_each(rdev, tmp, mddev) {
2776 if (rdev->desc_nr >= mddev->max_disks ||
2777 i > mddev->max_disks) {
2778 printk(KERN_WARNING
2779 "md: %s: %s: only %d devices permitted\n",
2780 mdname(mddev), bdevname(rdev->bdev, b),
2781 mddev->max_disks);
2782 kick_rdev_from_array(rdev);
2783 continue;
2784 }
2785 if (rdev != freshest)
2786 if (super_types[mddev->major_version].
2787 validate_super(mddev, rdev)) {
2788 printk(KERN_WARNING "md: kicking non-fresh %s"
2789 " from array!\n",
2790 bdevname(rdev->bdev,b));
2791 kick_rdev_from_array(rdev);
2792 continue;
2793 }
2794 if (mddev->level == LEVEL_MULTIPATH) {
2795 rdev->desc_nr = i++;
2796 rdev->raid_disk = rdev->desc_nr;
2797 set_bit(In_sync, &rdev->flags);
2798 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2799 rdev->raid_disk = -1;
2800 clear_bit(In_sync, &rdev->flags);
2801 }
2802 }
2803 }
2804
2805 /* Read a fixed-point number.
2806 * Numbers in sysfs attributes should be in "standard" units where
2807 * possible, so time should be in seconds.
2808 * However we internally use a a much smaller unit such as
2809 * milliseconds or jiffies.
2810 * This function takes a decimal number with a possible fractional
2811 * component, and produces an integer which is the result of
2812 * multiplying that number by 10^'scale'.
2813 * all without any floating-point arithmetic.
2814 */
2815 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2816 {
2817 unsigned long result = 0;
2818 long decimals = -1;
2819 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2820 if (*cp == '.')
2821 decimals = 0;
2822 else if (decimals < scale) {
2823 unsigned int value;
2824 value = *cp - '0';
2825 result = result * 10 + value;
2826 if (decimals >= 0)
2827 decimals++;
2828 }
2829 cp++;
2830 }
2831 if (*cp == '\n')
2832 cp++;
2833 if (*cp)
2834 return -EINVAL;
2835 if (decimals < 0)
2836 decimals = 0;
2837 while (decimals < scale) {
2838 result *= 10;
2839 decimals ++;
2840 }
2841 *res = result;
2842 return 0;
2843 }
2844
2845
2846 static void md_safemode_timeout(unsigned long data);
2847
2848 static ssize_t
2849 safe_delay_show(mddev_t *mddev, char *page)
2850 {
2851 int msec = (mddev->safemode_delay*1000)/HZ;
2852 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2853 }
2854 static ssize_t
2855 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2856 {
2857 unsigned long msec;
2858
2859 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2860 return -EINVAL;
2861 if (msec == 0)
2862 mddev->safemode_delay = 0;
2863 else {
2864 unsigned long old_delay = mddev->safemode_delay;
2865 mddev->safemode_delay = (msec*HZ)/1000;
2866 if (mddev->safemode_delay == 0)
2867 mddev->safemode_delay = 1;
2868 if (mddev->safemode_delay < old_delay)
2869 md_safemode_timeout((unsigned long)mddev);
2870 }
2871 return len;
2872 }
2873 static struct md_sysfs_entry md_safe_delay =
2874 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2875
2876 static ssize_t
2877 level_show(mddev_t *mddev, char *page)
2878 {
2879 struct mdk_personality *p = mddev->pers;
2880 if (p)
2881 return sprintf(page, "%s\n", p->name);
2882 else if (mddev->clevel[0])
2883 return sprintf(page, "%s\n", mddev->clevel);
2884 else if (mddev->level != LEVEL_NONE)
2885 return sprintf(page, "%d\n", mddev->level);
2886 else
2887 return 0;
2888 }
2889
2890 static ssize_t
2891 level_store(mddev_t *mddev, const char *buf, size_t len)
2892 {
2893 char level[16];
2894 ssize_t rv = len;
2895 struct mdk_personality *pers;
2896 void *priv;
2897 mdk_rdev_t *rdev;
2898
2899 if (mddev->pers == NULL) {
2900 if (len == 0)
2901 return 0;
2902 if (len >= sizeof(mddev->clevel))
2903 return -ENOSPC;
2904 strncpy(mddev->clevel, buf, len);
2905 if (mddev->clevel[len-1] == '\n')
2906 len--;
2907 mddev->clevel[len] = 0;
2908 mddev->level = LEVEL_NONE;
2909 return rv;
2910 }
2911
2912 /* request to change the personality. Need to ensure:
2913 * - array is not engaged in resync/recovery/reshape
2914 * - old personality can be suspended
2915 * - new personality will access other array.
2916 */
2917
2918 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2919 return -EBUSY;
2920
2921 if (!mddev->pers->quiesce) {
2922 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2923 mdname(mddev), mddev->pers->name);
2924 return -EINVAL;
2925 }
2926
2927 /* Now find the new personality */
2928 if (len == 0 || len >= sizeof(level))
2929 return -EINVAL;
2930 strncpy(level, buf, len);
2931 if (level[len-1] == '\n')
2932 len--;
2933 level[len] = 0;
2934
2935 request_module("md-%s", level);
2936 spin_lock(&pers_lock);
2937 pers = find_pers(LEVEL_NONE, level);
2938 if (!pers || !try_module_get(pers->owner)) {
2939 spin_unlock(&pers_lock);
2940 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2941 return -EINVAL;
2942 }
2943 spin_unlock(&pers_lock);
2944
2945 if (pers == mddev->pers) {
2946 /* Nothing to do! */
2947 module_put(pers->owner);
2948 return rv;
2949 }
2950 if (!pers->takeover) {
2951 module_put(pers->owner);
2952 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2953 mdname(mddev), level);
2954 return -EINVAL;
2955 }
2956
2957 /* ->takeover must set new_* and/or delta_disks
2958 * if it succeeds, and may set them when it fails.
2959 */
2960 priv = pers->takeover(mddev);
2961 if (IS_ERR(priv)) {
2962 mddev->new_level = mddev->level;
2963 mddev->new_layout = mddev->layout;
2964 mddev->new_chunk_sectors = mddev->chunk_sectors;
2965 mddev->raid_disks -= mddev->delta_disks;
2966 mddev->delta_disks = 0;
2967 module_put(pers->owner);
2968 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2969 mdname(mddev), level);
2970 return PTR_ERR(priv);
2971 }
2972
2973 /* Looks like we have a winner */
2974 mddev_suspend(mddev);
2975 mddev->pers->stop(mddev);
2976 module_put(mddev->pers->owner);
2977 /* Invalidate devices that are now superfluous */
2978 list_for_each_entry(rdev, &mddev->disks, same_set)
2979 if (rdev->raid_disk >= mddev->raid_disks) {
2980 rdev->raid_disk = -1;
2981 clear_bit(In_sync, &rdev->flags);
2982 }
2983 mddev->pers = pers;
2984 mddev->private = priv;
2985 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2986 mddev->level = mddev->new_level;
2987 mddev->layout = mddev->new_layout;
2988 mddev->chunk_sectors = mddev->new_chunk_sectors;
2989 mddev->delta_disks = 0;
2990 pers->run(mddev);
2991 mddev_resume(mddev);
2992 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2993 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2994 md_wakeup_thread(mddev->thread);
2995 return rv;
2996 }
2997
2998 static struct md_sysfs_entry md_level =
2999 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3000
3001
3002 static ssize_t
3003 layout_show(mddev_t *mddev, char *page)
3004 {
3005 /* just a number, not meaningful for all levels */
3006 if (mddev->reshape_position != MaxSector &&
3007 mddev->layout != mddev->new_layout)
3008 return sprintf(page, "%d (%d)\n",
3009 mddev->new_layout, mddev->layout);
3010 return sprintf(page, "%d\n", mddev->layout);
3011 }
3012
3013 static ssize_t
3014 layout_store(mddev_t *mddev, const char *buf, size_t len)
3015 {
3016 char *e;
3017 unsigned long n = simple_strtoul(buf, &e, 10);
3018
3019 if (!*buf || (*e && *e != '\n'))
3020 return -EINVAL;
3021
3022 if (mddev->pers) {
3023 int err;
3024 if (mddev->pers->check_reshape == NULL)
3025 return -EBUSY;
3026 mddev->new_layout = n;
3027 err = mddev->pers->check_reshape(mddev);
3028 if (err) {
3029 mddev->new_layout = mddev->layout;
3030 return err;
3031 }
3032 } else {
3033 mddev->new_layout = n;
3034 if (mddev->reshape_position == MaxSector)
3035 mddev->layout = n;
3036 }
3037 return len;
3038 }
3039 static struct md_sysfs_entry md_layout =
3040 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3041
3042
3043 static ssize_t
3044 raid_disks_show(mddev_t *mddev, char *page)
3045 {
3046 if (mddev->raid_disks == 0)
3047 return 0;
3048 if (mddev->reshape_position != MaxSector &&
3049 mddev->delta_disks != 0)
3050 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3051 mddev->raid_disks - mddev->delta_disks);
3052 return sprintf(page, "%d\n", mddev->raid_disks);
3053 }
3054
3055 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3056
3057 static ssize_t
3058 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3059 {
3060 char *e;
3061 int rv = 0;
3062 unsigned long n = simple_strtoul(buf, &e, 10);
3063
3064 if (!*buf || (*e && *e != '\n'))
3065 return -EINVAL;
3066
3067 if (mddev->pers)
3068 rv = update_raid_disks(mddev, n);
3069 else if (mddev->reshape_position != MaxSector) {
3070 int olddisks = mddev->raid_disks - mddev->delta_disks;
3071 mddev->delta_disks = n - olddisks;
3072 mddev->raid_disks = n;
3073 } else
3074 mddev->raid_disks = n;
3075 return rv ? rv : len;
3076 }
3077 static struct md_sysfs_entry md_raid_disks =
3078 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3079
3080 static ssize_t
3081 chunk_size_show(mddev_t *mddev, char *page)
3082 {
3083 if (mddev->reshape_position != MaxSector &&
3084 mddev->chunk_sectors != mddev->new_chunk_sectors)
3085 return sprintf(page, "%d (%d)\n",
3086 mddev->new_chunk_sectors << 9,
3087 mddev->chunk_sectors << 9);
3088 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3089 }
3090
3091 static ssize_t
3092 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3093 {
3094 char *e;
3095 unsigned long n = simple_strtoul(buf, &e, 10);
3096
3097 if (!*buf || (*e && *e != '\n'))
3098 return -EINVAL;
3099
3100 if (mddev->pers) {
3101 int err;
3102 if (mddev->pers->check_reshape == NULL)
3103 return -EBUSY;
3104 mddev->new_chunk_sectors = n >> 9;
3105 err = mddev->pers->check_reshape(mddev);
3106 if (err) {
3107 mddev->new_chunk_sectors = mddev->chunk_sectors;
3108 return err;
3109 }
3110 } else {
3111 mddev->new_chunk_sectors = n >> 9;
3112 if (mddev->reshape_position == MaxSector)
3113 mddev->chunk_sectors = n >> 9;
3114 }
3115 return len;
3116 }
3117 static struct md_sysfs_entry md_chunk_size =
3118 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3119
3120 static ssize_t
3121 resync_start_show(mddev_t *mddev, char *page)
3122 {
3123 if (mddev->recovery_cp == MaxSector)
3124 return sprintf(page, "none\n");
3125 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3126 }
3127
3128 static ssize_t
3129 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3130 {
3131 char *e;
3132 unsigned long long n = simple_strtoull(buf, &e, 10);
3133
3134 if (mddev->pers)
3135 return -EBUSY;
3136 if (cmd_match(buf, "none"))
3137 n = MaxSector;
3138 else if (!*buf || (*e && *e != '\n'))
3139 return -EINVAL;
3140
3141 mddev->recovery_cp = n;
3142 return len;
3143 }
3144 static struct md_sysfs_entry md_resync_start =
3145 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3146
3147 /*
3148 * The array state can be:
3149 *
3150 * clear
3151 * No devices, no size, no level
3152 * Equivalent to STOP_ARRAY ioctl
3153 * inactive
3154 * May have some settings, but array is not active
3155 * all IO results in error
3156 * When written, doesn't tear down array, but just stops it
3157 * suspended (not supported yet)
3158 * All IO requests will block. The array can be reconfigured.
3159 * Writing this, if accepted, will block until array is quiescent
3160 * readonly
3161 * no resync can happen. no superblocks get written.
3162 * write requests fail
3163 * read-auto
3164 * like readonly, but behaves like 'clean' on a write request.
3165 *
3166 * clean - no pending writes, but otherwise active.
3167 * When written to inactive array, starts without resync
3168 * If a write request arrives then
3169 * if metadata is known, mark 'dirty' and switch to 'active'.
3170 * if not known, block and switch to write-pending
3171 * If written to an active array that has pending writes, then fails.
3172 * active
3173 * fully active: IO and resync can be happening.
3174 * When written to inactive array, starts with resync
3175 *
3176 * write-pending
3177 * clean, but writes are blocked waiting for 'active' to be written.
3178 *
3179 * active-idle
3180 * like active, but no writes have been seen for a while (100msec).
3181 *
3182 */
3183 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3184 write_pending, active_idle, bad_word};
3185 static char *array_states[] = {
3186 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3187 "write-pending", "active-idle", NULL };
3188
3189 static int match_word(const char *word, char **list)
3190 {
3191 int n;
3192 for (n=0; list[n]; n++)
3193 if (cmd_match(word, list[n]))
3194 break;
3195 return n;
3196 }
3197
3198 static ssize_t
3199 array_state_show(mddev_t *mddev, char *page)
3200 {
3201 enum array_state st = inactive;
3202
3203 if (mddev->pers)
3204 switch(mddev->ro) {
3205 case 1:
3206 st = readonly;
3207 break;
3208 case 2:
3209 st = read_auto;
3210 break;
3211 case 0:
3212 if (mddev->in_sync)
3213 st = clean;
3214 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3215 st = write_pending;
3216 else if (mddev->safemode)
3217 st = active_idle;
3218 else
3219 st = active;
3220 }
3221 else {
3222 if (list_empty(&mddev->disks) &&
3223 mddev->raid_disks == 0 &&
3224 mddev->dev_sectors == 0)
3225 st = clear;
3226 else
3227 st = inactive;
3228 }
3229 return sprintf(page, "%s\n", array_states[st]);
3230 }
3231
3232 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3233 static int do_md_run(mddev_t * mddev);
3234 static int restart_array(mddev_t *mddev);
3235
3236 static ssize_t
3237 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3238 {
3239 int err = -EINVAL;
3240 enum array_state st = match_word(buf, array_states);
3241 switch(st) {
3242 case bad_word:
3243 break;
3244 case clear:
3245 /* stopping an active array */
3246 if (atomic_read(&mddev->openers) > 0)
3247 return -EBUSY;
3248 err = do_md_stop(mddev, 0, 0);
3249 break;
3250 case inactive:
3251 /* stopping an active array */
3252 if (mddev->pers) {
3253 if (atomic_read(&mddev->openers) > 0)
3254 return -EBUSY;
3255 err = do_md_stop(mddev, 2, 0);
3256 } else
3257 err = 0; /* already inactive */
3258 break;
3259 case suspended:
3260 break; /* not supported yet */
3261 case readonly:
3262 if (mddev->pers)
3263 err = do_md_stop(mddev, 1, 0);
3264 else {
3265 mddev->ro = 1;
3266 set_disk_ro(mddev->gendisk, 1);
3267 err = do_md_run(mddev);
3268 }
3269 break;
3270 case read_auto:
3271 if (mddev->pers) {
3272 if (mddev->ro == 0)
3273 err = do_md_stop(mddev, 1, 0);
3274 else if (mddev->ro == 1)
3275 err = restart_array(mddev);
3276 if (err == 0) {
3277 mddev->ro = 2;
3278 set_disk_ro(mddev->gendisk, 0);
3279 }
3280 } else {
3281 mddev->ro = 2;
3282 err = do_md_run(mddev);
3283 }
3284 break;
3285 case clean:
3286 if (mddev->pers) {
3287 restart_array(mddev);
3288 spin_lock_irq(&mddev->write_lock);
3289 if (atomic_read(&mddev->writes_pending) == 0) {
3290 if (mddev->in_sync == 0) {
3291 mddev->in_sync = 1;
3292 if (mddev->safemode == 1)
3293 mddev->safemode = 0;
3294 if (mddev->persistent)
3295 set_bit(MD_CHANGE_CLEAN,
3296 &mddev->flags);
3297 }
3298 err = 0;
3299 } else
3300 err = -EBUSY;
3301 spin_unlock_irq(&mddev->write_lock);
3302 } else
3303 err = -EINVAL;
3304 break;
3305 case active:
3306 if (mddev->pers) {
3307 restart_array(mddev);
3308 if (mddev->external)
3309 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3310 wake_up(&mddev->sb_wait);
3311 err = 0;
3312 } else {
3313 mddev->ro = 0;
3314 set_disk_ro(mddev->gendisk, 0);
3315 err = do_md_run(mddev);
3316 }
3317 break;
3318 case write_pending:
3319 case active_idle:
3320 /* these cannot be set */
3321 break;
3322 }
3323 if (err)
3324 return err;
3325 else {
3326 sysfs_notify_dirent(mddev->sysfs_state);
3327 return len;
3328 }
3329 }
3330 static struct md_sysfs_entry md_array_state =
3331 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3332
3333 static ssize_t
3334 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3335 return sprintf(page, "%d\n",
3336 atomic_read(&mddev->max_corr_read_errors));
3337 }
3338
3339 static ssize_t
3340 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3341 {
3342 char *e;
3343 unsigned long n = simple_strtoul(buf, &e, 10);
3344
3345 if (*buf && (*e == 0 || *e == '\n')) {
3346 atomic_set(&mddev->max_corr_read_errors, n);
3347 return len;
3348 }
3349 return -EINVAL;
3350 }
3351
3352 static struct md_sysfs_entry max_corr_read_errors =
3353 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3354 max_corrected_read_errors_store);
3355
3356 static ssize_t
3357 null_show(mddev_t *mddev, char *page)
3358 {
3359 return -EINVAL;
3360 }
3361
3362 static ssize_t
3363 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3364 {
3365 /* buf must be %d:%d\n? giving major and minor numbers */
3366 /* The new device is added to the array.
3367 * If the array has a persistent superblock, we read the
3368 * superblock to initialise info and check validity.
3369 * Otherwise, only checking done is that in bind_rdev_to_array,
3370 * which mainly checks size.
3371 */
3372 char *e;
3373 int major = simple_strtoul(buf, &e, 10);
3374 int minor;
3375 dev_t dev;
3376 mdk_rdev_t *rdev;
3377 int err;
3378
3379 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3380 return -EINVAL;
3381 minor = simple_strtoul(e+1, &e, 10);
3382 if (*e && *e != '\n')
3383 return -EINVAL;
3384 dev = MKDEV(major, minor);
3385 if (major != MAJOR(dev) ||
3386 minor != MINOR(dev))
3387 return -EOVERFLOW;
3388
3389
3390 if (mddev->persistent) {
3391 rdev = md_import_device(dev, mddev->major_version,
3392 mddev->minor_version);
3393 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3394 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3395 mdk_rdev_t, same_set);
3396 err = super_types[mddev->major_version]
3397 .load_super(rdev, rdev0, mddev->minor_version);
3398 if (err < 0)
3399 goto out;
3400 }
3401 } else if (mddev->external)
3402 rdev = md_import_device(dev, -2, -1);
3403 else
3404 rdev = md_import_device(dev, -1, -1);
3405
3406 if (IS_ERR(rdev))
3407 return PTR_ERR(rdev);
3408 err = bind_rdev_to_array(rdev, mddev);
3409 out:
3410 if (err)
3411 export_rdev(rdev);
3412 return err ? err : len;
3413 }
3414
3415 static struct md_sysfs_entry md_new_device =
3416 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3417
3418 static ssize_t
3419 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3420 {
3421 char *end;
3422 unsigned long chunk, end_chunk;
3423
3424 if (!mddev->bitmap)
3425 goto out;
3426 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3427 while (*buf) {
3428 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3429 if (buf == end) break;
3430 if (*end == '-') { /* range */
3431 buf = end + 1;
3432 end_chunk = simple_strtoul(buf, &end, 0);
3433 if (buf == end) break;
3434 }
3435 if (*end && !isspace(*end)) break;
3436 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3437 buf = skip_spaces(end);
3438 }
3439 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3440 out:
3441 return len;
3442 }
3443
3444 static struct md_sysfs_entry md_bitmap =
3445 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3446
3447 static ssize_t
3448 size_show(mddev_t *mddev, char *page)
3449 {
3450 return sprintf(page, "%llu\n",
3451 (unsigned long long)mddev->dev_sectors / 2);
3452 }
3453
3454 static int update_size(mddev_t *mddev, sector_t num_sectors);
3455
3456 static ssize_t
3457 size_store(mddev_t *mddev, const char *buf, size_t len)
3458 {
3459 /* If array is inactive, we can reduce the component size, but
3460 * not increase it (except from 0).
3461 * If array is active, we can try an on-line resize
3462 */
3463 sector_t sectors;
3464 int err = strict_blocks_to_sectors(buf, &sectors);
3465
3466 if (err < 0)
3467 return err;
3468 if (mddev->pers) {
3469 err = update_size(mddev, sectors);
3470 md_update_sb(mddev, 1);
3471 } else {
3472 if (mddev->dev_sectors == 0 ||
3473 mddev->dev_sectors > sectors)
3474 mddev->dev_sectors = sectors;
3475 else
3476 err = -ENOSPC;
3477 }
3478 return err ? err : len;
3479 }
3480
3481 static struct md_sysfs_entry md_size =
3482 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3483
3484
3485 /* Metdata version.
3486 * This is one of
3487 * 'none' for arrays with no metadata (good luck...)
3488 * 'external' for arrays with externally managed metadata,
3489 * or N.M for internally known formats
3490 */
3491 static ssize_t
3492 metadata_show(mddev_t *mddev, char *page)
3493 {
3494 if (mddev->persistent)
3495 return sprintf(page, "%d.%d\n",
3496 mddev->major_version, mddev->minor_version);
3497 else if (mddev->external)
3498 return sprintf(page, "external:%s\n", mddev->metadata_type);
3499 else
3500 return sprintf(page, "none\n");
3501 }
3502
3503 static ssize_t
3504 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3505 {
3506 int major, minor;
3507 char *e;
3508 /* Changing the details of 'external' metadata is
3509 * always permitted. Otherwise there must be
3510 * no devices attached to the array.
3511 */
3512 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3513 ;
3514 else if (!list_empty(&mddev->disks))
3515 return -EBUSY;
3516
3517 if (cmd_match(buf, "none")) {
3518 mddev->persistent = 0;
3519 mddev->external = 0;
3520 mddev->major_version = 0;
3521 mddev->minor_version = 90;
3522 return len;
3523 }
3524 if (strncmp(buf, "external:", 9) == 0) {
3525 size_t namelen = len-9;
3526 if (namelen >= sizeof(mddev->metadata_type))
3527 namelen = sizeof(mddev->metadata_type)-1;
3528 strncpy(mddev->metadata_type, buf+9, namelen);
3529 mddev->metadata_type[namelen] = 0;
3530 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3531 mddev->metadata_type[--namelen] = 0;
3532 mddev->persistent = 0;
3533 mddev->external = 1;
3534 mddev->major_version = 0;
3535 mddev->minor_version = 90;
3536 return len;
3537 }
3538 major = simple_strtoul(buf, &e, 10);
3539 if (e==buf || *e != '.')
3540 return -EINVAL;
3541 buf = e+1;
3542 minor = simple_strtoul(buf, &e, 10);
3543 if (e==buf || (*e && *e != '\n') )
3544 return -EINVAL;
3545 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3546 return -ENOENT;
3547 mddev->major_version = major;
3548 mddev->minor_version = minor;
3549 mddev->persistent = 1;
3550 mddev->external = 0;
3551 return len;
3552 }
3553
3554 static struct md_sysfs_entry md_metadata =
3555 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3556
3557 static ssize_t
3558 action_show(mddev_t *mddev, char *page)
3559 {
3560 char *type = "idle";
3561 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3562 type = "frozen";
3563 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3564 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3565 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3566 type = "reshape";
3567 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3568 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3569 type = "resync";
3570 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3571 type = "check";
3572 else
3573 type = "repair";
3574 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3575 type = "recover";
3576 }
3577 return sprintf(page, "%s\n", type);
3578 }
3579
3580 static ssize_t
3581 action_store(mddev_t *mddev, const char *page, size_t len)
3582 {
3583 if (!mddev->pers || !mddev->pers->sync_request)
3584 return -EINVAL;
3585
3586 if (cmd_match(page, "frozen"))
3587 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3588 else
3589 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3590
3591 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3592 if (mddev->sync_thread) {
3593 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3594 md_unregister_thread(mddev->sync_thread);
3595 mddev->sync_thread = NULL;
3596 mddev->recovery = 0;
3597 }
3598 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3599 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3600 return -EBUSY;
3601 else if (cmd_match(page, "resync"))
3602 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3603 else if (cmd_match(page, "recover")) {
3604 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3605 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3606 } else if (cmd_match(page, "reshape")) {
3607 int err;
3608 if (mddev->pers->start_reshape == NULL)
3609 return -EINVAL;
3610 err = mddev->pers->start_reshape(mddev);
3611 if (err)
3612 return err;
3613 sysfs_notify(&mddev->kobj, NULL, "degraded");
3614 } else {
3615 if (cmd_match(page, "check"))
3616 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3617 else if (!cmd_match(page, "repair"))
3618 return -EINVAL;
3619 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3620 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3621 }
3622 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3623 md_wakeup_thread(mddev->thread);
3624 sysfs_notify_dirent(mddev->sysfs_action);
3625 return len;
3626 }
3627
3628 static ssize_t
3629 mismatch_cnt_show(mddev_t *mddev, char *page)
3630 {
3631 return sprintf(page, "%llu\n",
3632 (unsigned long long) mddev->resync_mismatches);
3633 }
3634
3635 static struct md_sysfs_entry md_scan_mode =
3636 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3637
3638
3639 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3640
3641 static ssize_t
3642 sync_min_show(mddev_t *mddev, char *page)
3643 {
3644 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3645 mddev->sync_speed_min ? "local": "system");
3646 }
3647
3648 static ssize_t
3649 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3650 {
3651 int min;
3652 char *e;
3653 if (strncmp(buf, "system", 6)==0) {
3654 mddev->sync_speed_min = 0;
3655 return len;
3656 }
3657 min = simple_strtoul(buf, &e, 10);
3658 if (buf == e || (*e && *e != '\n') || min <= 0)
3659 return -EINVAL;
3660 mddev->sync_speed_min = min;
3661 return len;
3662 }
3663
3664 static struct md_sysfs_entry md_sync_min =
3665 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3666
3667 static ssize_t
3668 sync_max_show(mddev_t *mddev, char *page)
3669 {
3670 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3671 mddev->sync_speed_max ? "local": "system");
3672 }
3673
3674 static ssize_t
3675 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3676 {
3677 int max;
3678 char *e;
3679 if (strncmp(buf, "system", 6)==0) {
3680 mddev->sync_speed_max = 0;
3681 return len;
3682 }
3683 max = simple_strtoul(buf, &e, 10);
3684 if (buf == e || (*e && *e != '\n') || max <= 0)
3685 return -EINVAL;
3686 mddev->sync_speed_max = max;
3687 return len;
3688 }
3689
3690 static struct md_sysfs_entry md_sync_max =
3691 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3692
3693 static ssize_t
3694 degraded_show(mddev_t *mddev, char *page)
3695 {
3696 return sprintf(page, "%d\n", mddev->degraded);
3697 }
3698 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3699
3700 static ssize_t
3701 sync_force_parallel_show(mddev_t *mddev, char *page)
3702 {
3703 return sprintf(page, "%d\n", mddev->parallel_resync);
3704 }
3705
3706 static ssize_t
3707 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3708 {
3709 long n;
3710
3711 if (strict_strtol(buf, 10, &n))
3712 return -EINVAL;
3713
3714 if (n != 0 && n != 1)
3715 return -EINVAL;
3716
3717 mddev->parallel_resync = n;
3718
3719 if (mddev->sync_thread)
3720 wake_up(&resync_wait);
3721
3722 return len;
3723 }
3724
3725 /* force parallel resync, even with shared block devices */
3726 static struct md_sysfs_entry md_sync_force_parallel =
3727 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3728 sync_force_parallel_show, sync_force_parallel_store);
3729
3730 static ssize_t
3731 sync_speed_show(mddev_t *mddev, char *page)
3732 {
3733 unsigned long resync, dt, db;
3734 if (mddev->curr_resync == 0)
3735 return sprintf(page, "none\n");
3736 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3737 dt = (jiffies - mddev->resync_mark) / HZ;
3738 if (!dt) dt++;
3739 db = resync - mddev->resync_mark_cnt;
3740 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3741 }
3742
3743 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3744
3745 static ssize_t
3746 sync_completed_show(mddev_t *mddev, char *page)
3747 {
3748 unsigned long max_sectors, resync;
3749
3750 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3751 return sprintf(page, "none\n");
3752
3753 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3754 max_sectors = mddev->resync_max_sectors;
3755 else
3756 max_sectors = mddev->dev_sectors;
3757
3758 resync = mddev->curr_resync_completed;
3759 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3760 }
3761
3762 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3763
3764 static ssize_t
3765 min_sync_show(mddev_t *mddev, char *page)
3766 {
3767 return sprintf(page, "%llu\n",
3768 (unsigned long long)mddev->resync_min);
3769 }
3770 static ssize_t
3771 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3772 {
3773 unsigned long long min;
3774 if (strict_strtoull(buf, 10, &min))
3775 return -EINVAL;
3776 if (min > mddev->resync_max)
3777 return -EINVAL;
3778 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3779 return -EBUSY;
3780
3781 /* Must be a multiple of chunk_size */
3782 if (mddev->chunk_sectors) {
3783 sector_t temp = min;
3784 if (sector_div(temp, mddev->chunk_sectors))
3785 return -EINVAL;
3786 }
3787 mddev->resync_min = min;
3788
3789 return len;
3790 }
3791
3792 static struct md_sysfs_entry md_min_sync =
3793 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3794
3795 static ssize_t
3796 max_sync_show(mddev_t *mddev, char *page)
3797 {
3798 if (mddev->resync_max == MaxSector)
3799 return sprintf(page, "max\n");
3800 else
3801 return sprintf(page, "%llu\n",
3802 (unsigned long long)mddev->resync_max);
3803 }
3804 static ssize_t
3805 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3806 {
3807 if (strncmp(buf, "max", 3) == 0)
3808 mddev->resync_max = MaxSector;
3809 else {
3810 unsigned long long max;
3811 if (strict_strtoull(buf, 10, &max))
3812 return -EINVAL;
3813 if (max < mddev->resync_min)
3814 return -EINVAL;
3815 if (max < mddev->resync_max &&
3816 mddev->ro == 0 &&
3817 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3818 return -EBUSY;
3819
3820 /* Must be a multiple of chunk_size */
3821 if (mddev->chunk_sectors) {
3822 sector_t temp = max;
3823 if (sector_div(temp, mddev->chunk_sectors))
3824 return -EINVAL;
3825 }
3826 mddev->resync_max = max;
3827 }
3828 wake_up(&mddev->recovery_wait);
3829 return len;
3830 }
3831
3832 static struct md_sysfs_entry md_max_sync =
3833 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3834
3835 static ssize_t
3836 suspend_lo_show(mddev_t *mddev, char *page)
3837 {
3838 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3839 }
3840
3841 static ssize_t
3842 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3843 {
3844 char *e;
3845 unsigned long long new = simple_strtoull(buf, &e, 10);
3846
3847 if (mddev->pers == NULL ||
3848 mddev->pers->quiesce == NULL)
3849 return -EINVAL;
3850 if (buf == e || (*e && *e != '\n'))
3851 return -EINVAL;
3852 if (new >= mddev->suspend_hi ||
3853 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3854 mddev->suspend_lo = new;
3855 mddev->pers->quiesce(mddev, 2);
3856 return len;
3857 } else
3858 return -EINVAL;
3859 }
3860 static struct md_sysfs_entry md_suspend_lo =
3861 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3862
3863
3864 static ssize_t
3865 suspend_hi_show(mddev_t *mddev, char *page)
3866 {
3867 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3868 }
3869
3870 static ssize_t
3871 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3872 {
3873 char *e;
3874 unsigned long long new = simple_strtoull(buf, &e, 10);
3875
3876 if (mddev->pers == NULL ||
3877 mddev->pers->quiesce == NULL)
3878 return -EINVAL;
3879 if (buf == e || (*e && *e != '\n'))
3880 return -EINVAL;
3881 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3882 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3883 mddev->suspend_hi = new;
3884 mddev->pers->quiesce(mddev, 1);
3885 mddev->pers->quiesce(mddev, 0);
3886 return len;
3887 } else
3888 return -EINVAL;
3889 }
3890 static struct md_sysfs_entry md_suspend_hi =
3891 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3892
3893 static ssize_t
3894 reshape_position_show(mddev_t *mddev, char *page)
3895 {
3896 if (mddev->reshape_position != MaxSector)
3897 return sprintf(page, "%llu\n",
3898 (unsigned long long)mddev->reshape_position);
3899 strcpy(page, "none\n");
3900 return 5;
3901 }
3902
3903 static ssize_t
3904 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3905 {
3906 char *e;
3907 unsigned long long new = simple_strtoull(buf, &e, 10);
3908 if (mddev->pers)
3909 return -EBUSY;
3910 if (buf == e || (*e && *e != '\n'))
3911 return -EINVAL;
3912 mddev->reshape_position = new;
3913 mddev->delta_disks = 0;
3914 mddev->new_level = mddev->level;
3915 mddev->new_layout = mddev->layout;
3916 mddev->new_chunk_sectors = mddev->chunk_sectors;
3917 return len;
3918 }
3919
3920 static struct md_sysfs_entry md_reshape_position =
3921 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3922 reshape_position_store);
3923
3924 static ssize_t
3925 array_size_show(mddev_t *mddev, char *page)
3926 {
3927 if (mddev->external_size)
3928 return sprintf(page, "%llu\n",
3929 (unsigned long long)mddev->array_sectors/2);
3930 else
3931 return sprintf(page, "default\n");
3932 }
3933
3934 static ssize_t
3935 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3936 {
3937 sector_t sectors;
3938
3939 if (strncmp(buf, "default", 7) == 0) {
3940 if (mddev->pers)
3941 sectors = mddev->pers->size(mddev, 0, 0);
3942 else
3943 sectors = mddev->array_sectors;
3944
3945 mddev->external_size = 0;
3946 } else {
3947 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3948 return -EINVAL;
3949 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3950 return -E2BIG;
3951
3952 mddev->external_size = 1;
3953 }
3954
3955 mddev->array_sectors = sectors;
3956 set_capacity(mddev->gendisk, mddev->array_sectors);
3957 if (mddev->pers)
3958 revalidate_disk(mddev->gendisk);
3959
3960 return len;
3961 }
3962
3963 static struct md_sysfs_entry md_array_size =
3964 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3965 array_size_store);
3966
3967 static struct attribute *md_default_attrs[] = {
3968 &md_level.attr,
3969 &md_layout.attr,
3970 &md_raid_disks.attr,
3971 &md_chunk_size.attr,
3972 &md_size.attr,
3973 &md_resync_start.attr,
3974 &md_metadata.attr,
3975 &md_new_device.attr,
3976 &md_safe_delay.attr,
3977 &md_array_state.attr,
3978 &md_reshape_position.attr,
3979 &md_array_size.attr,
3980 &max_corr_read_errors.attr,
3981 NULL,
3982 };
3983
3984 static struct attribute *md_redundancy_attrs[] = {
3985 &md_scan_mode.attr,
3986 &md_mismatches.attr,
3987 &md_sync_min.attr,
3988 &md_sync_max.attr,
3989 &md_sync_speed.attr,
3990 &md_sync_force_parallel.attr,
3991 &md_sync_completed.attr,
3992 &md_min_sync.attr,
3993 &md_max_sync.attr,
3994 &md_suspend_lo.attr,
3995 &md_suspend_hi.attr,
3996 &md_bitmap.attr,
3997 &md_degraded.attr,
3998 NULL,
3999 };
4000 static struct attribute_group md_redundancy_group = {
4001 .name = NULL,
4002 .attrs = md_redundancy_attrs,
4003 };
4004
4005
4006 static ssize_t
4007 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4008 {
4009 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4010 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4011 ssize_t rv;
4012
4013 if (!entry->show)
4014 return -EIO;
4015 rv = mddev_lock(mddev);
4016 if (!rv) {
4017 rv = entry->show(mddev, page);
4018 mddev_unlock(mddev);
4019 }
4020 return rv;
4021 }
4022
4023 static ssize_t
4024 md_attr_store(struct kobject *kobj, struct attribute *attr,
4025 const char *page, size_t length)
4026 {
4027 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4028 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4029 ssize_t rv;
4030
4031 if (!entry->store)
4032 return -EIO;
4033 if (!capable(CAP_SYS_ADMIN))
4034 return -EACCES;
4035 rv = mddev_lock(mddev);
4036 if (mddev->hold_active == UNTIL_IOCTL)
4037 mddev->hold_active = 0;
4038 if (!rv) {
4039 rv = entry->store(mddev, page, length);
4040 mddev_unlock(mddev);
4041 }
4042 return rv;
4043 }
4044
4045 static void md_free(struct kobject *ko)
4046 {
4047 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4048
4049 if (mddev->sysfs_state)
4050 sysfs_put(mddev->sysfs_state);
4051
4052 if (mddev->gendisk) {
4053 del_gendisk(mddev->gendisk);
4054 put_disk(mddev->gendisk);
4055 }
4056 if (mddev->queue)
4057 blk_cleanup_queue(mddev->queue);
4058
4059 kfree(mddev);
4060 }
4061
4062 static const struct sysfs_ops md_sysfs_ops = {
4063 .show = md_attr_show,
4064 .store = md_attr_store,
4065 };
4066 static struct kobj_type md_ktype = {
4067 .release = md_free,
4068 .sysfs_ops = &md_sysfs_ops,
4069 .default_attrs = md_default_attrs,
4070 };
4071
4072 int mdp_major = 0;
4073
4074 static void mddev_delayed_delete(struct work_struct *ws)
4075 {
4076 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4077
4078 if (mddev->private) {
4079 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4080 if (mddev->private != (void*)1)
4081 sysfs_remove_group(&mddev->kobj, mddev->private);
4082 if (mddev->sysfs_action)
4083 sysfs_put(mddev->sysfs_action);
4084 mddev->sysfs_action = NULL;
4085 mddev->private = NULL;
4086 }
4087 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4088 kobject_del(&mddev->kobj);
4089 kobject_put(&mddev->kobj);
4090 }
4091
4092 static int md_alloc(dev_t dev, char *name)
4093 {
4094 static DEFINE_MUTEX(disks_mutex);
4095 mddev_t *mddev = mddev_find(dev);
4096 struct gendisk *disk;
4097 int partitioned;
4098 int shift;
4099 int unit;
4100 int error;
4101
4102 if (!mddev)
4103 return -ENODEV;
4104
4105 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4106 shift = partitioned ? MdpMinorShift : 0;
4107 unit = MINOR(mddev->unit) >> shift;
4108
4109 /* wait for any previous instance if this device
4110 * to be completed removed (mddev_delayed_delete).
4111 */
4112 flush_scheduled_work();
4113
4114 mutex_lock(&disks_mutex);
4115 error = -EEXIST;
4116 if (mddev->gendisk)
4117 goto abort;
4118
4119 if (name) {
4120 /* Need to ensure that 'name' is not a duplicate.
4121 */
4122 mddev_t *mddev2;
4123 spin_lock(&all_mddevs_lock);
4124
4125 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4126 if (mddev2->gendisk &&
4127 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4128 spin_unlock(&all_mddevs_lock);
4129 goto abort;
4130 }
4131 spin_unlock(&all_mddevs_lock);
4132 }
4133
4134 error = -ENOMEM;
4135 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4136 if (!mddev->queue)
4137 goto abort;
4138 mddev->queue->queuedata = mddev;
4139
4140 /* Can be unlocked because the queue is new: no concurrency */
4141 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4142
4143 blk_queue_make_request(mddev->queue, md_make_request);
4144
4145 disk = alloc_disk(1 << shift);
4146 if (!disk) {
4147 blk_cleanup_queue(mddev->queue);
4148 mddev->queue = NULL;
4149 goto abort;
4150 }
4151 disk->major = MAJOR(mddev->unit);
4152 disk->first_minor = unit << shift;
4153 if (name)
4154 strcpy(disk->disk_name, name);
4155 else if (partitioned)
4156 sprintf(disk->disk_name, "md_d%d", unit);
4157 else
4158 sprintf(disk->disk_name, "md%d", unit);
4159 disk->fops = &md_fops;
4160 disk->private_data = mddev;
4161 disk->queue = mddev->queue;
4162 /* Allow extended partitions. This makes the
4163 * 'mdp' device redundant, but we can't really
4164 * remove it now.
4165 */
4166 disk->flags |= GENHD_FL_EXT_DEVT;
4167 add_disk(disk);
4168 mddev->gendisk = disk;
4169 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4170 &disk_to_dev(disk)->kobj, "%s", "md");
4171 if (error) {
4172 /* This isn't possible, but as kobject_init_and_add is marked
4173 * __must_check, we must do something with the result
4174 */
4175 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4176 disk->disk_name);
4177 error = 0;
4178 }
4179 if (sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4180 printk(KERN_DEBUG "pointless warning\n");
4181 abort:
4182 mutex_unlock(&disks_mutex);
4183 if (!error) {
4184 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4185 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
4186 }
4187 mddev_put(mddev);
4188 return error;
4189 }
4190
4191 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4192 {
4193 md_alloc(dev, NULL);
4194 return NULL;
4195 }
4196
4197 static int add_named_array(const char *val, struct kernel_param *kp)
4198 {
4199 /* val must be "md_*" where * is not all digits.
4200 * We allocate an array with a large free minor number, and
4201 * set the name to val. val must not already be an active name.
4202 */
4203 int len = strlen(val);
4204 char buf[DISK_NAME_LEN];
4205
4206 while (len && val[len-1] == '\n')
4207 len--;
4208 if (len >= DISK_NAME_LEN)
4209 return -E2BIG;
4210 strlcpy(buf, val, len+1);
4211 if (strncmp(buf, "md_", 3) != 0)
4212 return -EINVAL;
4213 return md_alloc(0, buf);
4214 }
4215
4216 static void md_safemode_timeout(unsigned long data)
4217 {
4218 mddev_t *mddev = (mddev_t *) data;
4219
4220 if (!atomic_read(&mddev->writes_pending)) {
4221 mddev->safemode = 1;
4222 if (mddev->external)
4223 sysfs_notify_dirent(mddev->sysfs_state);
4224 }
4225 md_wakeup_thread(mddev->thread);
4226 }
4227
4228 static int start_dirty_degraded;
4229
4230 static int do_md_run(mddev_t * mddev)
4231 {
4232 int err;
4233 mdk_rdev_t *rdev;
4234 struct gendisk *disk;
4235 struct mdk_personality *pers;
4236
4237 if (list_empty(&mddev->disks))
4238 /* cannot run an array with no devices.. */
4239 return -EINVAL;
4240
4241 if (mddev->pers)
4242 return -EBUSY;
4243
4244 /*
4245 * Analyze all RAID superblock(s)
4246 */
4247 if (!mddev->raid_disks) {
4248 if (!mddev->persistent)
4249 return -EINVAL;
4250 analyze_sbs(mddev);
4251 }
4252
4253 if (mddev->level != LEVEL_NONE)
4254 request_module("md-level-%d", mddev->level);
4255 else if (mddev->clevel[0])
4256 request_module("md-%s", mddev->clevel);
4257
4258 /*
4259 * Drop all container device buffers, from now on
4260 * the only valid external interface is through the md
4261 * device.
4262 */
4263 list_for_each_entry(rdev, &mddev->disks, same_set) {
4264 if (test_bit(Faulty, &rdev->flags))
4265 continue;
4266 sync_blockdev(rdev->bdev);
4267 invalidate_bdev(rdev->bdev);
4268
4269 /* perform some consistency tests on the device.
4270 * We don't want the data to overlap the metadata,
4271 * Internal Bitmap issues have been handled elsewhere.
4272 */
4273 if (rdev->data_offset < rdev->sb_start) {
4274 if (mddev->dev_sectors &&
4275 rdev->data_offset + mddev->dev_sectors
4276 > rdev->sb_start) {
4277 printk("md: %s: data overlaps metadata\n",
4278 mdname(mddev));
4279 return -EINVAL;
4280 }
4281 } else {
4282 if (rdev->sb_start + rdev->sb_size/512
4283 > rdev->data_offset) {
4284 printk("md: %s: metadata overlaps data\n",
4285 mdname(mddev));
4286 return -EINVAL;
4287 }
4288 }
4289 sysfs_notify_dirent(rdev->sysfs_state);
4290 }
4291
4292 disk = mddev->gendisk;
4293
4294 spin_lock(&pers_lock);
4295 pers = find_pers(mddev->level, mddev->clevel);
4296 if (!pers || !try_module_get(pers->owner)) {
4297 spin_unlock(&pers_lock);
4298 if (mddev->level != LEVEL_NONE)
4299 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4300 mddev->level);
4301 else
4302 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4303 mddev->clevel);
4304 return -EINVAL;
4305 }
4306 mddev->pers = pers;
4307 spin_unlock(&pers_lock);
4308 if (mddev->level != pers->level) {
4309 mddev->level = pers->level;
4310 mddev->new_level = pers->level;
4311 }
4312 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4313
4314 if (mddev->reshape_position != MaxSector &&
4315 pers->start_reshape == NULL) {
4316 /* This personality cannot handle reshaping... */
4317 mddev->pers = NULL;
4318 module_put(pers->owner);
4319 return -EINVAL;
4320 }
4321
4322 if (pers->sync_request) {
4323 /* Warn if this is a potentially silly
4324 * configuration.
4325 */
4326 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4327 mdk_rdev_t *rdev2;
4328 int warned = 0;
4329
4330 list_for_each_entry(rdev, &mddev->disks, same_set)
4331 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4332 if (rdev < rdev2 &&
4333 rdev->bdev->bd_contains ==
4334 rdev2->bdev->bd_contains) {
4335 printk(KERN_WARNING
4336 "%s: WARNING: %s appears to be"
4337 " on the same physical disk as"
4338 " %s.\n",
4339 mdname(mddev),
4340 bdevname(rdev->bdev,b),
4341 bdevname(rdev2->bdev,b2));
4342 warned = 1;
4343 }
4344 }
4345
4346 if (warned)
4347 printk(KERN_WARNING
4348 "True protection against single-disk"
4349 " failure might be compromised.\n");
4350 }
4351
4352 mddev->recovery = 0;
4353 /* may be over-ridden by personality */
4354 mddev->resync_max_sectors = mddev->dev_sectors;
4355
4356 mddev->barriers_work = 1;
4357 mddev->ok_start_degraded = start_dirty_degraded;
4358
4359 if (start_readonly && mddev->ro == 0)
4360 mddev->ro = 2; /* read-only, but switch on first write */
4361
4362 err = mddev->pers->run(mddev);
4363 if (err)
4364 printk(KERN_ERR "md: pers->run() failed ...\n");
4365 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4366 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4367 " but 'external_size' not in effect?\n", __func__);
4368 printk(KERN_ERR
4369 "md: invalid array_size %llu > default size %llu\n",
4370 (unsigned long long)mddev->array_sectors / 2,
4371 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4372 err = -EINVAL;
4373 mddev->pers->stop(mddev);
4374 }
4375 if (err == 0 && mddev->pers->sync_request) {
4376 err = bitmap_create(mddev);
4377 if (err) {
4378 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4379 mdname(mddev), err);
4380 mddev->pers->stop(mddev);
4381 }
4382 }
4383 if (err) {
4384 module_put(mddev->pers->owner);
4385 mddev->pers = NULL;
4386 bitmap_destroy(mddev);
4387 return err;
4388 }
4389 if (mddev->pers->sync_request) {
4390 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4391 printk(KERN_WARNING
4392 "md: cannot register extra attributes for %s\n",
4393 mdname(mddev));
4394 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4395 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4396 mddev->ro = 0;
4397
4398 atomic_set(&mddev->writes_pending,0);
4399 atomic_set(&mddev->max_corr_read_errors,
4400 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4401 mddev->safemode = 0;
4402 mddev->safemode_timer.function = md_safemode_timeout;
4403 mddev->safemode_timer.data = (unsigned long) mddev;
4404 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4405 mddev->in_sync = 1;
4406
4407 list_for_each_entry(rdev, &mddev->disks, same_set)
4408 if (rdev->raid_disk >= 0) {
4409 char nm[20];
4410 sprintf(nm, "rd%d", rdev->raid_disk);
4411 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4412 printk("md: cannot register %s for %s\n",
4413 nm, mdname(mddev));
4414 }
4415
4416 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4417
4418 if (mddev->flags)
4419 md_update_sb(mddev, 0);
4420
4421 set_capacity(disk, mddev->array_sectors);
4422
4423 md_wakeup_thread(mddev->thread);
4424 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4425
4426 revalidate_disk(mddev->gendisk);
4427 mddev->changed = 1;
4428 md_new_event(mddev);
4429 sysfs_notify_dirent(mddev->sysfs_state);
4430 if (mddev->sysfs_action)
4431 sysfs_notify_dirent(mddev->sysfs_action);
4432 sysfs_notify(&mddev->kobj, NULL, "degraded");
4433 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4434 return 0;
4435 }
4436
4437 static int restart_array(mddev_t *mddev)
4438 {
4439 struct gendisk *disk = mddev->gendisk;
4440
4441 /* Complain if it has no devices */
4442 if (list_empty(&mddev->disks))
4443 return -ENXIO;
4444 if (!mddev->pers)
4445 return -EINVAL;
4446 if (!mddev->ro)
4447 return -EBUSY;
4448 mddev->safemode = 0;
4449 mddev->ro = 0;
4450 set_disk_ro(disk, 0);
4451 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4452 mdname(mddev));
4453 /* Kick recovery or resync if necessary */
4454 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4455 md_wakeup_thread(mddev->thread);
4456 md_wakeup_thread(mddev->sync_thread);
4457 sysfs_notify_dirent(mddev->sysfs_state);
4458 return 0;
4459 }
4460
4461 /* similar to deny_write_access, but accounts for our holding a reference
4462 * to the file ourselves */
4463 static int deny_bitmap_write_access(struct file * file)
4464 {
4465 struct inode *inode = file->f_mapping->host;
4466
4467 spin_lock(&inode->i_lock);
4468 if (atomic_read(&inode->i_writecount) > 1) {
4469 spin_unlock(&inode->i_lock);
4470 return -ETXTBSY;
4471 }
4472 atomic_set(&inode->i_writecount, -1);
4473 spin_unlock(&inode->i_lock);
4474
4475 return 0;
4476 }
4477
4478 void restore_bitmap_write_access(struct file *file)
4479 {
4480 struct inode *inode = file->f_mapping->host;
4481
4482 spin_lock(&inode->i_lock);
4483 atomic_set(&inode->i_writecount, 1);
4484 spin_unlock(&inode->i_lock);
4485 }
4486
4487 /* mode:
4488 * 0 - completely stop and dis-assemble array
4489 * 1 - switch to readonly
4490 * 2 - stop but do not disassemble array
4491 */
4492 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4493 {
4494 int err = 0;
4495 struct gendisk *disk = mddev->gendisk;
4496 mdk_rdev_t *rdev;
4497
4498 mutex_lock(&mddev->open_mutex);
4499 if (atomic_read(&mddev->openers) > is_open) {
4500 printk("md: %s still in use.\n",mdname(mddev));
4501 err = -EBUSY;
4502 } else if (mddev->pers) {
4503
4504 if (mddev->sync_thread) {
4505 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4506 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4507 md_unregister_thread(mddev->sync_thread);
4508 mddev->sync_thread = NULL;
4509 }
4510
4511 del_timer_sync(&mddev->safemode_timer);
4512
4513 switch(mode) {
4514 case 1: /* readonly */
4515 err = -ENXIO;
4516 if (mddev->ro==1)
4517 goto out;
4518 mddev->ro = 1;
4519 break;
4520 case 0: /* disassemble */
4521 case 2: /* stop */
4522 bitmap_flush(mddev);
4523 md_super_wait(mddev);
4524 if (mddev->ro)
4525 set_disk_ro(disk, 0);
4526
4527 mddev->pers->stop(mddev);
4528 mddev->queue->merge_bvec_fn = NULL;
4529 mddev->queue->unplug_fn = NULL;
4530 mddev->queue->backing_dev_info.congested_fn = NULL;
4531 module_put(mddev->pers->owner);
4532 if (mddev->pers->sync_request && mddev->private == NULL)
4533 mddev->private = (void*)1;
4534 mddev->pers = NULL;
4535 /* tell userspace to handle 'inactive' */
4536 sysfs_notify_dirent(mddev->sysfs_state);
4537
4538 list_for_each_entry(rdev, &mddev->disks, same_set)
4539 if (rdev->raid_disk >= 0) {
4540 char nm[20];
4541 sprintf(nm, "rd%d", rdev->raid_disk);
4542 sysfs_remove_link(&mddev->kobj, nm);
4543 }
4544
4545 set_capacity(disk, 0);
4546 mddev->changed = 1;
4547
4548 if (mddev->ro)
4549 mddev->ro = 0;
4550 }
4551 if (!mddev->in_sync || mddev->flags) {
4552 /* mark array as shutdown cleanly */
4553 mddev->in_sync = 1;
4554 md_update_sb(mddev, 1);
4555 }
4556 if (mode == 1)
4557 set_disk_ro(disk, 1);
4558 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4559 err = 0;
4560 }
4561 out:
4562 mutex_unlock(&mddev->open_mutex);
4563 if (err)
4564 return err;
4565 /*
4566 * Free resources if final stop
4567 */
4568 if (mode == 0) {
4569
4570 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4571
4572 bitmap_destroy(mddev);
4573 if (mddev->bitmap_info.file) {
4574 restore_bitmap_write_access(mddev->bitmap_info.file);
4575 fput(mddev->bitmap_info.file);
4576 mddev->bitmap_info.file = NULL;
4577 }
4578 mddev->bitmap_info.offset = 0;
4579
4580 export_array(mddev);
4581
4582 mddev->array_sectors = 0;
4583 mddev->external_size = 0;
4584 mddev->dev_sectors = 0;
4585 mddev->raid_disks = 0;
4586 mddev->recovery_cp = 0;
4587 mddev->resync_min = 0;
4588 mddev->resync_max = MaxSector;
4589 mddev->reshape_position = MaxSector;
4590 mddev->external = 0;
4591 mddev->persistent = 0;
4592 mddev->level = LEVEL_NONE;
4593 mddev->clevel[0] = 0;
4594 mddev->flags = 0;
4595 mddev->ro = 0;
4596 mddev->metadata_type[0] = 0;
4597 mddev->chunk_sectors = 0;
4598 mddev->ctime = mddev->utime = 0;
4599 mddev->layout = 0;
4600 mddev->max_disks = 0;
4601 mddev->events = 0;
4602 mddev->delta_disks = 0;
4603 mddev->new_level = LEVEL_NONE;
4604 mddev->new_layout = 0;
4605 mddev->new_chunk_sectors = 0;
4606 mddev->curr_resync = 0;
4607 mddev->resync_mismatches = 0;
4608 mddev->suspend_lo = mddev->suspend_hi = 0;
4609 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4610 mddev->recovery = 0;
4611 mddev->in_sync = 0;
4612 mddev->changed = 0;
4613 mddev->degraded = 0;
4614 mddev->barriers_work = 0;
4615 mddev->safemode = 0;
4616 mddev->bitmap_info.offset = 0;
4617 mddev->bitmap_info.default_offset = 0;
4618 mddev->bitmap_info.chunksize = 0;
4619 mddev->bitmap_info.daemon_sleep = 0;
4620 mddev->bitmap_info.max_write_behind = 0;
4621 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4622 if (mddev->hold_active == UNTIL_STOP)
4623 mddev->hold_active = 0;
4624
4625 } else if (mddev->pers)
4626 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4627 mdname(mddev));
4628 err = 0;
4629 blk_integrity_unregister(disk);
4630 md_new_event(mddev);
4631 sysfs_notify_dirent(mddev->sysfs_state);
4632 return err;
4633 }
4634
4635 #ifndef MODULE
4636 static void autorun_array(mddev_t *mddev)
4637 {
4638 mdk_rdev_t *rdev;
4639 int err;
4640
4641 if (list_empty(&mddev->disks))
4642 return;
4643
4644 printk(KERN_INFO "md: running: ");
4645
4646 list_for_each_entry(rdev, &mddev->disks, same_set) {
4647 char b[BDEVNAME_SIZE];
4648 printk("<%s>", bdevname(rdev->bdev,b));
4649 }
4650 printk("\n");
4651
4652 err = do_md_run(mddev);
4653 if (err) {
4654 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4655 do_md_stop(mddev, 0, 0);
4656 }
4657 }
4658
4659 /*
4660 * lets try to run arrays based on all disks that have arrived
4661 * until now. (those are in pending_raid_disks)
4662 *
4663 * the method: pick the first pending disk, collect all disks with
4664 * the same UUID, remove all from the pending list and put them into
4665 * the 'same_array' list. Then order this list based on superblock
4666 * update time (freshest comes first), kick out 'old' disks and
4667 * compare superblocks. If everything's fine then run it.
4668 *
4669 * If "unit" is allocated, then bump its reference count
4670 */
4671 static void autorun_devices(int part)
4672 {
4673 mdk_rdev_t *rdev0, *rdev, *tmp;
4674 mddev_t *mddev;
4675 char b[BDEVNAME_SIZE];
4676
4677 printk(KERN_INFO "md: autorun ...\n");
4678 while (!list_empty(&pending_raid_disks)) {
4679 int unit;
4680 dev_t dev;
4681 LIST_HEAD(candidates);
4682 rdev0 = list_entry(pending_raid_disks.next,
4683 mdk_rdev_t, same_set);
4684
4685 printk(KERN_INFO "md: considering %s ...\n",
4686 bdevname(rdev0->bdev,b));
4687 INIT_LIST_HEAD(&candidates);
4688 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4689 if (super_90_load(rdev, rdev0, 0) >= 0) {
4690 printk(KERN_INFO "md: adding %s ...\n",
4691 bdevname(rdev->bdev,b));
4692 list_move(&rdev->same_set, &candidates);
4693 }
4694 /*
4695 * now we have a set of devices, with all of them having
4696 * mostly sane superblocks. It's time to allocate the
4697 * mddev.
4698 */
4699 if (part) {
4700 dev = MKDEV(mdp_major,
4701 rdev0->preferred_minor << MdpMinorShift);
4702 unit = MINOR(dev) >> MdpMinorShift;
4703 } else {
4704 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4705 unit = MINOR(dev);
4706 }
4707 if (rdev0->preferred_minor != unit) {
4708 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4709 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4710 break;
4711 }
4712
4713 md_probe(dev, NULL, NULL);
4714 mddev = mddev_find(dev);
4715 if (!mddev || !mddev->gendisk) {
4716 if (mddev)
4717 mddev_put(mddev);
4718 printk(KERN_ERR
4719 "md: cannot allocate memory for md drive.\n");
4720 break;
4721 }
4722 if (mddev_lock(mddev))
4723 printk(KERN_WARNING "md: %s locked, cannot run\n",
4724 mdname(mddev));
4725 else if (mddev->raid_disks || mddev->major_version
4726 || !list_empty(&mddev->disks)) {
4727 printk(KERN_WARNING
4728 "md: %s already running, cannot run %s\n",
4729 mdname(mddev), bdevname(rdev0->bdev,b));
4730 mddev_unlock(mddev);
4731 } else {
4732 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4733 mddev->persistent = 1;
4734 rdev_for_each_list(rdev, tmp, &candidates) {
4735 list_del_init(&rdev->same_set);
4736 if (bind_rdev_to_array(rdev, mddev))
4737 export_rdev(rdev);
4738 }
4739 autorun_array(mddev);
4740 mddev_unlock(mddev);
4741 }
4742 /* on success, candidates will be empty, on error
4743 * it won't...
4744 */
4745 rdev_for_each_list(rdev, tmp, &candidates) {
4746 list_del_init(&rdev->same_set);
4747 export_rdev(rdev);
4748 }
4749 mddev_put(mddev);
4750 }
4751 printk(KERN_INFO "md: ... autorun DONE.\n");
4752 }
4753 #endif /* !MODULE */
4754
4755 static int get_version(void __user * arg)
4756 {
4757 mdu_version_t ver;
4758
4759 ver.major = MD_MAJOR_VERSION;
4760 ver.minor = MD_MINOR_VERSION;
4761 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4762
4763 if (copy_to_user(arg, &ver, sizeof(ver)))
4764 return -EFAULT;
4765
4766 return 0;
4767 }
4768
4769 static int get_array_info(mddev_t * mddev, void __user * arg)
4770 {
4771 mdu_array_info_t info;
4772 int nr,working,insync,failed,spare;
4773 mdk_rdev_t *rdev;
4774
4775 nr=working=insync=failed=spare=0;
4776 list_for_each_entry(rdev, &mddev->disks, same_set) {
4777 nr++;
4778 if (test_bit(Faulty, &rdev->flags))
4779 failed++;
4780 else {
4781 working++;
4782 if (test_bit(In_sync, &rdev->flags))
4783 insync++;
4784 else
4785 spare++;
4786 }
4787 }
4788
4789 info.major_version = mddev->major_version;
4790 info.minor_version = mddev->minor_version;
4791 info.patch_version = MD_PATCHLEVEL_VERSION;
4792 info.ctime = mddev->ctime;
4793 info.level = mddev->level;
4794 info.size = mddev->dev_sectors / 2;
4795 if (info.size != mddev->dev_sectors / 2) /* overflow */
4796 info.size = -1;
4797 info.nr_disks = nr;
4798 info.raid_disks = mddev->raid_disks;
4799 info.md_minor = mddev->md_minor;
4800 info.not_persistent= !mddev->persistent;
4801
4802 info.utime = mddev->utime;
4803 info.state = 0;
4804 if (mddev->in_sync)
4805 info.state = (1<<MD_SB_CLEAN);
4806 if (mddev->bitmap && mddev->bitmap_info.offset)
4807 info.state = (1<<MD_SB_BITMAP_PRESENT);
4808 info.active_disks = insync;
4809 info.working_disks = working;
4810 info.failed_disks = failed;
4811 info.spare_disks = spare;
4812
4813 info.layout = mddev->layout;
4814 info.chunk_size = mddev->chunk_sectors << 9;
4815
4816 if (copy_to_user(arg, &info, sizeof(info)))
4817 return -EFAULT;
4818
4819 return 0;
4820 }
4821
4822 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4823 {
4824 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4825 char *ptr, *buf = NULL;
4826 int err = -ENOMEM;
4827
4828 if (md_allow_write(mddev))
4829 file = kmalloc(sizeof(*file), GFP_NOIO);
4830 else
4831 file = kmalloc(sizeof(*file), GFP_KERNEL);
4832
4833 if (!file)
4834 goto out;
4835
4836 /* bitmap disabled, zero the first byte and copy out */
4837 if (!mddev->bitmap || !mddev->bitmap->file) {
4838 file->pathname[0] = '\0';
4839 goto copy_out;
4840 }
4841
4842 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4843 if (!buf)
4844 goto out;
4845
4846 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4847 if (IS_ERR(ptr))
4848 goto out;
4849
4850 strcpy(file->pathname, ptr);
4851
4852 copy_out:
4853 err = 0;
4854 if (copy_to_user(arg, file, sizeof(*file)))
4855 err = -EFAULT;
4856 out:
4857 kfree(buf);
4858 kfree(file);
4859 return err;
4860 }
4861
4862 static int get_disk_info(mddev_t * mddev, void __user * arg)
4863 {
4864 mdu_disk_info_t info;
4865 mdk_rdev_t *rdev;
4866
4867 if (copy_from_user(&info, arg, sizeof(info)))
4868 return -EFAULT;
4869
4870 rdev = find_rdev_nr(mddev, info.number);
4871 if (rdev) {
4872 info.major = MAJOR(rdev->bdev->bd_dev);
4873 info.minor = MINOR(rdev->bdev->bd_dev);
4874 info.raid_disk = rdev->raid_disk;
4875 info.state = 0;
4876 if (test_bit(Faulty, &rdev->flags))
4877 info.state |= (1<<MD_DISK_FAULTY);
4878 else if (test_bit(In_sync, &rdev->flags)) {
4879 info.state |= (1<<MD_DISK_ACTIVE);
4880 info.state |= (1<<MD_DISK_SYNC);
4881 }
4882 if (test_bit(WriteMostly, &rdev->flags))
4883 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4884 } else {
4885 info.major = info.minor = 0;
4886 info.raid_disk = -1;
4887 info.state = (1<<MD_DISK_REMOVED);
4888 }
4889
4890 if (copy_to_user(arg, &info, sizeof(info)))
4891 return -EFAULT;
4892
4893 return 0;
4894 }
4895
4896 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4897 {
4898 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4899 mdk_rdev_t *rdev;
4900 dev_t dev = MKDEV(info->major,info->minor);
4901
4902 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4903 return -EOVERFLOW;
4904
4905 if (!mddev->raid_disks) {
4906 int err;
4907 /* expecting a device which has a superblock */
4908 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4909 if (IS_ERR(rdev)) {
4910 printk(KERN_WARNING
4911 "md: md_import_device returned %ld\n",
4912 PTR_ERR(rdev));
4913 return PTR_ERR(rdev);
4914 }
4915 if (!list_empty(&mddev->disks)) {
4916 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4917 mdk_rdev_t, same_set);
4918 err = super_types[mddev->major_version]
4919 .load_super(rdev, rdev0, mddev->minor_version);
4920 if (err < 0) {
4921 printk(KERN_WARNING
4922 "md: %s has different UUID to %s\n",
4923 bdevname(rdev->bdev,b),
4924 bdevname(rdev0->bdev,b2));
4925 export_rdev(rdev);
4926 return -EINVAL;
4927 }
4928 }
4929 err = bind_rdev_to_array(rdev, mddev);
4930 if (err)
4931 export_rdev(rdev);
4932 return err;
4933 }
4934
4935 /*
4936 * add_new_disk can be used once the array is assembled
4937 * to add "hot spares". They must already have a superblock
4938 * written
4939 */
4940 if (mddev->pers) {
4941 int err;
4942 if (!mddev->pers->hot_add_disk) {
4943 printk(KERN_WARNING
4944 "%s: personality does not support diskops!\n",
4945 mdname(mddev));
4946 return -EINVAL;
4947 }
4948 if (mddev->persistent)
4949 rdev = md_import_device(dev, mddev->major_version,
4950 mddev->minor_version);
4951 else
4952 rdev = md_import_device(dev, -1, -1);
4953 if (IS_ERR(rdev)) {
4954 printk(KERN_WARNING
4955 "md: md_import_device returned %ld\n",
4956 PTR_ERR(rdev));
4957 return PTR_ERR(rdev);
4958 }
4959 /* set save_raid_disk if appropriate */
4960 if (!mddev->persistent) {
4961 if (info->state & (1<<MD_DISK_SYNC) &&
4962 info->raid_disk < mddev->raid_disks)
4963 rdev->raid_disk = info->raid_disk;
4964 else
4965 rdev->raid_disk = -1;
4966 } else
4967 super_types[mddev->major_version].
4968 validate_super(mddev, rdev);
4969 rdev->saved_raid_disk = rdev->raid_disk;
4970
4971 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4972 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4973 set_bit(WriteMostly, &rdev->flags);
4974 else
4975 clear_bit(WriteMostly, &rdev->flags);
4976
4977 rdev->raid_disk = -1;
4978 err = bind_rdev_to_array(rdev, mddev);
4979 if (!err && !mddev->pers->hot_remove_disk) {
4980 /* If there is hot_add_disk but no hot_remove_disk
4981 * then added disks for geometry changes,
4982 * and should be added immediately.
4983 */
4984 super_types[mddev->major_version].
4985 validate_super(mddev, rdev);
4986 err = mddev->pers->hot_add_disk(mddev, rdev);
4987 if (err)
4988 unbind_rdev_from_array(rdev);
4989 }
4990 if (err)
4991 export_rdev(rdev);
4992 else
4993 sysfs_notify_dirent(rdev->sysfs_state);
4994
4995 md_update_sb(mddev, 1);
4996 if (mddev->degraded)
4997 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4998 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4999 md_wakeup_thread(mddev->thread);
5000 return err;
5001 }
5002
5003 /* otherwise, add_new_disk is only allowed
5004 * for major_version==0 superblocks
5005 */
5006 if (mddev->major_version != 0) {
5007 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5008 mdname(mddev));
5009 return -EINVAL;
5010 }
5011
5012 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5013 int err;
5014 rdev = md_import_device(dev, -1, 0);
5015 if (IS_ERR(rdev)) {
5016 printk(KERN_WARNING
5017 "md: error, md_import_device() returned %ld\n",
5018 PTR_ERR(rdev));
5019 return PTR_ERR(rdev);
5020 }
5021 rdev->desc_nr = info->number;
5022 if (info->raid_disk < mddev->raid_disks)
5023 rdev->raid_disk = info->raid_disk;
5024 else
5025 rdev->raid_disk = -1;
5026
5027 if (rdev->raid_disk < mddev->raid_disks)
5028 if (info->state & (1<<MD_DISK_SYNC))
5029 set_bit(In_sync, &rdev->flags);
5030
5031 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5032 set_bit(WriteMostly, &rdev->flags);
5033
5034 if (!mddev->persistent) {
5035 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5036 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5037 } else
5038 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5039 rdev->sectors = rdev->sb_start;
5040
5041 err = bind_rdev_to_array(rdev, mddev);
5042 if (err) {
5043 export_rdev(rdev);
5044 return err;
5045 }
5046 }
5047
5048 return 0;
5049 }
5050
5051 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5052 {
5053 char b[BDEVNAME_SIZE];
5054 mdk_rdev_t *rdev;
5055
5056 rdev = find_rdev(mddev, dev);
5057 if (!rdev)
5058 return -ENXIO;
5059
5060 if (rdev->raid_disk >= 0)
5061 goto busy;
5062
5063 kick_rdev_from_array(rdev);
5064 md_update_sb(mddev, 1);
5065 md_new_event(mddev);
5066
5067 return 0;
5068 busy:
5069 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5070 bdevname(rdev->bdev,b), mdname(mddev));
5071 return -EBUSY;
5072 }
5073
5074 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5075 {
5076 char b[BDEVNAME_SIZE];
5077 int err;
5078 mdk_rdev_t *rdev;
5079
5080 if (!mddev->pers)
5081 return -ENODEV;
5082
5083 if (mddev->major_version != 0) {
5084 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5085 " version-0 superblocks.\n",
5086 mdname(mddev));
5087 return -EINVAL;
5088 }
5089 if (!mddev->pers->hot_add_disk) {
5090 printk(KERN_WARNING
5091 "%s: personality does not support diskops!\n",
5092 mdname(mddev));
5093 return -EINVAL;
5094 }
5095
5096 rdev = md_import_device(dev, -1, 0);
5097 if (IS_ERR(rdev)) {
5098 printk(KERN_WARNING
5099 "md: error, md_import_device() returned %ld\n",
5100 PTR_ERR(rdev));
5101 return -EINVAL;
5102 }
5103
5104 if (mddev->persistent)
5105 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5106 else
5107 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5108
5109 rdev->sectors = rdev->sb_start;
5110
5111 if (test_bit(Faulty, &rdev->flags)) {
5112 printk(KERN_WARNING
5113 "md: can not hot-add faulty %s disk to %s!\n",
5114 bdevname(rdev->bdev,b), mdname(mddev));
5115 err = -EINVAL;
5116 goto abort_export;
5117 }
5118 clear_bit(In_sync, &rdev->flags);
5119 rdev->desc_nr = -1;
5120 rdev->saved_raid_disk = -1;
5121 err = bind_rdev_to_array(rdev, mddev);
5122 if (err)
5123 goto abort_export;
5124
5125 /*
5126 * The rest should better be atomic, we can have disk failures
5127 * noticed in interrupt contexts ...
5128 */
5129
5130 rdev->raid_disk = -1;
5131
5132 md_update_sb(mddev, 1);
5133
5134 /*
5135 * Kick recovery, maybe this spare has to be added to the
5136 * array immediately.
5137 */
5138 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5139 md_wakeup_thread(mddev->thread);
5140 md_new_event(mddev);
5141 return 0;
5142
5143 abort_export:
5144 export_rdev(rdev);
5145 return err;
5146 }
5147
5148 static int set_bitmap_file(mddev_t *mddev, int fd)
5149 {
5150 int err;
5151
5152 if (mddev->pers) {
5153 if (!mddev->pers->quiesce)
5154 return -EBUSY;
5155 if (mddev->recovery || mddev->sync_thread)
5156 return -EBUSY;
5157 /* we should be able to change the bitmap.. */
5158 }
5159
5160
5161 if (fd >= 0) {
5162 if (mddev->bitmap)
5163 return -EEXIST; /* cannot add when bitmap is present */
5164 mddev->bitmap_info.file = fget(fd);
5165
5166 if (mddev->bitmap_info.file == NULL) {
5167 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5168 mdname(mddev));
5169 return -EBADF;
5170 }
5171
5172 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5173 if (err) {
5174 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5175 mdname(mddev));
5176 fput(mddev->bitmap_info.file);
5177 mddev->bitmap_info.file = NULL;
5178 return err;
5179 }
5180 mddev->bitmap_info.offset = 0; /* file overrides offset */
5181 } else if (mddev->bitmap == NULL)
5182 return -ENOENT; /* cannot remove what isn't there */
5183 err = 0;
5184 if (mddev->pers) {
5185 mddev->pers->quiesce(mddev, 1);
5186 if (fd >= 0)
5187 err = bitmap_create(mddev);
5188 if (fd < 0 || err) {
5189 bitmap_destroy(mddev);
5190 fd = -1; /* make sure to put the file */
5191 }
5192 mddev->pers->quiesce(mddev, 0);
5193 }
5194 if (fd < 0) {
5195 if (mddev->bitmap_info.file) {
5196 restore_bitmap_write_access(mddev->bitmap_info.file);
5197 fput(mddev->bitmap_info.file);
5198 }
5199 mddev->bitmap_info.file = NULL;
5200 }
5201
5202 return err;
5203 }
5204
5205 /*
5206 * set_array_info is used two different ways
5207 * The original usage is when creating a new array.
5208 * In this usage, raid_disks is > 0 and it together with
5209 * level, size, not_persistent,layout,chunksize determine the
5210 * shape of the array.
5211 * This will always create an array with a type-0.90.0 superblock.
5212 * The newer usage is when assembling an array.
5213 * In this case raid_disks will be 0, and the major_version field is
5214 * use to determine which style super-blocks are to be found on the devices.
5215 * The minor and patch _version numbers are also kept incase the
5216 * super_block handler wishes to interpret them.
5217 */
5218 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5219 {
5220
5221 if (info->raid_disks == 0) {
5222 /* just setting version number for superblock loading */
5223 if (info->major_version < 0 ||
5224 info->major_version >= ARRAY_SIZE(super_types) ||
5225 super_types[info->major_version].name == NULL) {
5226 /* maybe try to auto-load a module? */
5227 printk(KERN_INFO
5228 "md: superblock version %d not known\n",
5229 info->major_version);
5230 return -EINVAL;
5231 }
5232 mddev->major_version = info->major_version;
5233 mddev->minor_version = info->minor_version;
5234 mddev->patch_version = info->patch_version;
5235 mddev->persistent = !info->not_persistent;
5236 /* ensure mddev_put doesn't delete this now that there
5237 * is some minimal configuration.
5238 */
5239 mddev->ctime = get_seconds();
5240 return 0;
5241 }
5242 mddev->major_version = MD_MAJOR_VERSION;
5243 mddev->minor_version = MD_MINOR_VERSION;
5244 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5245 mddev->ctime = get_seconds();
5246
5247 mddev->level = info->level;
5248 mddev->clevel[0] = 0;
5249 mddev->dev_sectors = 2 * (sector_t)info->size;
5250 mddev->raid_disks = info->raid_disks;
5251 /* don't set md_minor, it is determined by which /dev/md* was
5252 * openned
5253 */
5254 if (info->state & (1<<MD_SB_CLEAN))
5255 mddev->recovery_cp = MaxSector;
5256 else
5257 mddev->recovery_cp = 0;
5258 mddev->persistent = ! info->not_persistent;
5259 mddev->external = 0;
5260
5261 mddev->layout = info->layout;
5262 mddev->chunk_sectors = info->chunk_size >> 9;
5263
5264 mddev->max_disks = MD_SB_DISKS;
5265
5266 if (mddev->persistent)
5267 mddev->flags = 0;
5268 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5269
5270 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5271 mddev->bitmap_info.offset = 0;
5272
5273 mddev->reshape_position = MaxSector;
5274
5275 /*
5276 * Generate a 128 bit UUID
5277 */
5278 get_random_bytes(mddev->uuid, 16);
5279
5280 mddev->new_level = mddev->level;
5281 mddev->new_chunk_sectors = mddev->chunk_sectors;
5282 mddev->new_layout = mddev->layout;
5283 mddev->delta_disks = 0;
5284
5285 return 0;
5286 }
5287
5288 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5289 {
5290 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5291
5292 if (mddev->external_size)
5293 return;
5294
5295 mddev->array_sectors = array_sectors;
5296 }
5297 EXPORT_SYMBOL(md_set_array_sectors);
5298
5299 static int update_size(mddev_t *mddev, sector_t num_sectors)
5300 {
5301 mdk_rdev_t *rdev;
5302 int rv;
5303 int fit = (num_sectors == 0);
5304
5305 if (mddev->pers->resize == NULL)
5306 return -EINVAL;
5307 /* The "num_sectors" is the number of sectors of each device that
5308 * is used. This can only make sense for arrays with redundancy.
5309 * linear and raid0 always use whatever space is available. We can only
5310 * consider changing this number if no resync or reconstruction is
5311 * happening, and if the new size is acceptable. It must fit before the
5312 * sb_start or, if that is <data_offset, it must fit before the size
5313 * of each device. If num_sectors is zero, we find the largest size
5314 * that fits.
5315
5316 */
5317 if (mddev->sync_thread)
5318 return -EBUSY;
5319 if (mddev->bitmap)
5320 /* Sorry, cannot grow a bitmap yet, just remove it,
5321 * grow, and re-add.
5322 */
5323 return -EBUSY;
5324 list_for_each_entry(rdev, &mddev->disks, same_set) {
5325 sector_t avail = rdev->sectors;
5326
5327 if (fit && (num_sectors == 0 || num_sectors > avail))
5328 num_sectors = avail;
5329 if (avail < num_sectors)
5330 return -ENOSPC;
5331 }
5332 rv = mddev->pers->resize(mddev, num_sectors);
5333 if (!rv)
5334 revalidate_disk(mddev->gendisk);
5335 return rv;
5336 }
5337
5338 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5339 {
5340 int rv;
5341 /* change the number of raid disks */
5342 if (mddev->pers->check_reshape == NULL)
5343 return -EINVAL;
5344 if (raid_disks <= 0 ||
5345 raid_disks >= mddev->max_disks)
5346 return -EINVAL;
5347 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5348 return -EBUSY;
5349 mddev->delta_disks = raid_disks - mddev->raid_disks;
5350
5351 rv = mddev->pers->check_reshape(mddev);
5352 return rv;
5353 }
5354
5355
5356 /*
5357 * update_array_info is used to change the configuration of an
5358 * on-line array.
5359 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5360 * fields in the info are checked against the array.
5361 * Any differences that cannot be handled will cause an error.
5362 * Normally, only one change can be managed at a time.
5363 */
5364 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5365 {
5366 int rv = 0;
5367 int cnt = 0;
5368 int state = 0;
5369
5370 /* calculate expected state,ignoring low bits */
5371 if (mddev->bitmap && mddev->bitmap_info.offset)
5372 state |= (1 << MD_SB_BITMAP_PRESENT);
5373
5374 if (mddev->major_version != info->major_version ||
5375 mddev->minor_version != info->minor_version ||
5376 /* mddev->patch_version != info->patch_version || */
5377 mddev->ctime != info->ctime ||
5378 mddev->level != info->level ||
5379 /* mddev->layout != info->layout || */
5380 !mddev->persistent != info->not_persistent||
5381 mddev->chunk_sectors != info->chunk_size >> 9 ||
5382 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5383 ((state^info->state) & 0xfffffe00)
5384 )
5385 return -EINVAL;
5386 /* Check there is only one change */
5387 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5388 cnt++;
5389 if (mddev->raid_disks != info->raid_disks)
5390 cnt++;
5391 if (mddev->layout != info->layout)
5392 cnt++;
5393 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5394 cnt++;
5395 if (cnt == 0)
5396 return 0;
5397 if (cnt > 1)
5398 return -EINVAL;
5399
5400 if (mddev->layout != info->layout) {
5401 /* Change layout
5402 * we don't need to do anything at the md level, the
5403 * personality will take care of it all.
5404 */
5405 if (mddev->pers->check_reshape == NULL)
5406 return -EINVAL;
5407 else {
5408 mddev->new_layout = info->layout;
5409 rv = mddev->pers->check_reshape(mddev);
5410 if (rv)
5411 mddev->new_layout = mddev->layout;
5412 return rv;
5413 }
5414 }
5415 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5416 rv = update_size(mddev, (sector_t)info->size * 2);
5417
5418 if (mddev->raid_disks != info->raid_disks)
5419 rv = update_raid_disks(mddev, info->raid_disks);
5420
5421 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5422 if (mddev->pers->quiesce == NULL)
5423 return -EINVAL;
5424 if (mddev->recovery || mddev->sync_thread)
5425 return -EBUSY;
5426 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5427 /* add the bitmap */
5428 if (mddev->bitmap)
5429 return -EEXIST;
5430 if (mddev->bitmap_info.default_offset == 0)
5431 return -EINVAL;
5432 mddev->bitmap_info.offset =
5433 mddev->bitmap_info.default_offset;
5434 mddev->pers->quiesce(mddev, 1);
5435 rv = bitmap_create(mddev);
5436 if (rv)
5437 bitmap_destroy(mddev);
5438 mddev->pers->quiesce(mddev, 0);
5439 } else {
5440 /* remove the bitmap */
5441 if (!mddev->bitmap)
5442 return -ENOENT;
5443 if (mddev->bitmap->file)
5444 return -EINVAL;
5445 mddev->pers->quiesce(mddev, 1);
5446 bitmap_destroy(mddev);
5447 mddev->pers->quiesce(mddev, 0);
5448 mddev->bitmap_info.offset = 0;
5449 }
5450 }
5451 md_update_sb(mddev, 1);
5452 return rv;
5453 }
5454
5455 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5456 {
5457 mdk_rdev_t *rdev;
5458
5459 if (mddev->pers == NULL)
5460 return -ENODEV;
5461
5462 rdev = find_rdev(mddev, dev);
5463 if (!rdev)
5464 return -ENODEV;
5465
5466 md_error(mddev, rdev);
5467 return 0;
5468 }
5469
5470 /*
5471 * We have a problem here : there is no easy way to give a CHS
5472 * virtual geometry. We currently pretend that we have a 2 heads
5473 * 4 sectors (with a BIG number of cylinders...). This drives
5474 * dosfs just mad... ;-)
5475 */
5476 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5477 {
5478 mddev_t *mddev = bdev->bd_disk->private_data;
5479
5480 geo->heads = 2;
5481 geo->sectors = 4;
5482 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5483 return 0;
5484 }
5485
5486 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5487 unsigned int cmd, unsigned long arg)
5488 {
5489 int err = 0;
5490 void __user *argp = (void __user *)arg;
5491 mddev_t *mddev = NULL;
5492
5493 if (!capable(CAP_SYS_ADMIN))
5494 return -EACCES;
5495
5496 /*
5497 * Commands dealing with the RAID driver but not any
5498 * particular array:
5499 */
5500 switch (cmd)
5501 {
5502 case RAID_VERSION:
5503 err = get_version(argp);
5504 goto done;
5505
5506 case PRINT_RAID_DEBUG:
5507 err = 0;
5508 md_print_devices();
5509 goto done;
5510
5511 #ifndef MODULE
5512 case RAID_AUTORUN:
5513 err = 0;
5514 autostart_arrays(arg);
5515 goto done;
5516 #endif
5517 default:;
5518 }
5519
5520 /*
5521 * Commands creating/starting a new array:
5522 */
5523
5524 mddev = bdev->bd_disk->private_data;
5525
5526 if (!mddev) {
5527 BUG();
5528 goto abort;
5529 }
5530
5531 err = mddev_lock(mddev);
5532 if (err) {
5533 printk(KERN_INFO
5534 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5535 err, cmd);
5536 goto abort;
5537 }
5538
5539 switch (cmd)
5540 {
5541 case SET_ARRAY_INFO:
5542 {
5543 mdu_array_info_t info;
5544 if (!arg)
5545 memset(&info, 0, sizeof(info));
5546 else if (copy_from_user(&info, argp, sizeof(info))) {
5547 err = -EFAULT;
5548 goto abort_unlock;
5549 }
5550 if (mddev->pers) {
5551 err = update_array_info(mddev, &info);
5552 if (err) {
5553 printk(KERN_WARNING "md: couldn't update"
5554 " array info. %d\n", err);
5555 goto abort_unlock;
5556 }
5557 goto done_unlock;
5558 }
5559 if (!list_empty(&mddev->disks)) {
5560 printk(KERN_WARNING
5561 "md: array %s already has disks!\n",
5562 mdname(mddev));
5563 err = -EBUSY;
5564 goto abort_unlock;
5565 }
5566 if (mddev->raid_disks) {
5567 printk(KERN_WARNING
5568 "md: array %s already initialised!\n",
5569 mdname(mddev));
5570 err = -EBUSY;
5571 goto abort_unlock;
5572 }
5573 err = set_array_info(mddev, &info);
5574 if (err) {
5575 printk(KERN_WARNING "md: couldn't set"
5576 " array info. %d\n", err);
5577 goto abort_unlock;
5578 }
5579 }
5580 goto done_unlock;
5581
5582 default:;
5583 }
5584
5585 /*
5586 * Commands querying/configuring an existing array:
5587 */
5588 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5589 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5590 if ((!mddev->raid_disks && !mddev->external)
5591 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5592 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5593 && cmd != GET_BITMAP_FILE) {
5594 err = -ENODEV;
5595 goto abort_unlock;
5596 }
5597
5598 /*
5599 * Commands even a read-only array can execute:
5600 */
5601 switch (cmd)
5602 {
5603 case GET_ARRAY_INFO:
5604 err = get_array_info(mddev, argp);
5605 goto done_unlock;
5606
5607 case GET_BITMAP_FILE:
5608 err = get_bitmap_file(mddev, argp);
5609 goto done_unlock;
5610
5611 case GET_DISK_INFO:
5612 err = get_disk_info(mddev, argp);
5613 goto done_unlock;
5614
5615 case RESTART_ARRAY_RW:
5616 err = restart_array(mddev);
5617 goto done_unlock;
5618
5619 case STOP_ARRAY:
5620 err = do_md_stop(mddev, 0, 1);
5621 goto done_unlock;
5622
5623 case STOP_ARRAY_RO:
5624 err = do_md_stop(mddev, 1, 1);
5625 goto done_unlock;
5626
5627 }
5628
5629 /*
5630 * The remaining ioctls are changing the state of the
5631 * superblock, so we do not allow them on read-only arrays.
5632 * However non-MD ioctls (e.g. get-size) will still come through
5633 * here and hit the 'default' below, so only disallow
5634 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5635 */
5636 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5637 if (mddev->ro == 2) {
5638 mddev->ro = 0;
5639 sysfs_notify_dirent(mddev->sysfs_state);
5640 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5641 md_wakeup_thread(mddev->thread);
5642 } else {
5643 err = -EROFS;
5644 goto abort_unlock;
5645 }
5646 }
5647
5648 switch (cmd)
5649 {
5650 case ADD_NEW_DISK:
5651 {
5652 mdu_disk_info_t info;
5653 if (copy_from_user(&info, argp, sizeof(info)))
5654 err = -EFAULT;
5655 else
5656 err = add_new_disk(mddev, &info);
5657 goto done_unlock;
5658 }
5659
5660 case HOT_REMOVE_DISK:
5661 err = hot_remove_disk(mddev, new_decode_dev(arg));
5662 goto done_unlock;
5663
5664 case HOT_ADD_DISK:
5665 err = hot_add_disk(mddev, new_decode_dev(arg));
5666 goto done_unlock;
5667
5668 case SET_DISK_FAULTY:
5669 err = set_disk_faulty(mddev, new_decode_dev(arg));
5670 goto done_unlock;
5671
5672 case RUN_ARRAY:
5673 err = do_md_run(mddev);
5674 goto done_unlock;
5675
5676 case SET_BITMAP_FILE:
5677 err = set_bitmap_file(mddev, (int)arg);
5678 goto done_unlock;
5679
5680 default:
5681 err = -EINVAL;
5682 goto abort_unlock;
5683 }
5684
5685 done_unlock:
5686 abort_unlock:
5687 if (mddev->hold_active == UNTIL_IOCTL &&
5688 err != -EINVAL)
5689 mddev->hold_active = 0;
5690 mddev_unlock(mddev);
5691
5692 return err;
5693 done:
5694 if (err)
5695 MD_BUG();
5696 abort:
5697 return err;
5698 }
5699 #ifdef CONFIG_COMPAT
5700 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5701 unsigned int cmd, unsigned long arg)
5702 {
5703 switch (cmd) {
5704 case HOT_REMOVE_DISK:
5705 case HOT_ADD_DISK:
5706 case SET_DISK_FAULTY:
5707 case SET_BITMAP_FILE:
5708 /* These take in integer arg, do not convert */
5709 break;
5710 default:
5711 arg = (unsigned long)compat_ptr(arg);
5712 break;
5713 }
5714
5715 return md_ioctl(bdev, mode, cmd, arg);
5716 }
5717 #endif /* CONFIG_COMPAT */
5718
5719 static int md_open(struct block_device *bdev, fmode_t mode)
5720 {
5721 /*
5722 * Succeed if we can lock the mddev, which confirms that
5723 * it isn't being stopped right now.
5724 */
5725 mddev_t *mddev = mddev_find(bdev->bd_dev);
5726 int err;
5727
5728 if (mddev->gendisk != bdev->bd_disk) {
5729 /* we are racing with mddev_put which is discarding this
5730 * bd_disk.
5731 */
5732 mddev_put(mddev);
5733 /* Wait until bdev->bd_disk is definitely gone */
5734 flush_scheduled_work();
5735 /* Then retry the open from the top */
5736 return -ERESTARTSYS;
5737 }
5738 BUG_ON(mddev != bdev->bd_disk->private_data);
5739
5740 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5741 goto out;
5742
5743 err = 0;
5744 atomic_inc(&mddev->openers);
5745 mutex_unlock(&mddev->open_mutex);
5746
5747 check_disk_change(bdev);
5748 out:
5749 return err;
5750 }
5751
5752 static int md_release(struct gendisk *disk, fmode_t mode)
5753 {
5754 mddev_t *mddev = disk->private_data;
5755
5756 BUG_ON(!mddev);
5757 atomic_dec(&mddev->openers);
5758 mddev_put(mddev);
5759
5760 return 0;
5761 }
5762
5763 static int md_media_changed(struct gendisk *disk)
5764 {
5765 mddev_t *mddev = disk->private_data;
5766
5767 return mddev->changed;
5768 }
5769
5770 static int md_revalidate(struct gendisk *disk)
5771 {
5772 mddev_t *mddev = disk->private_data;
5773
5774 mddev->changed = 0;
5775 return 0;
5776 }
5777 static const struct block_device_operations md_fops =
5778 {
5779 .owner = THIS_MODULE,
5780 .open = md_open,
5781 .release = md_release,
5782 .ioctl = md_ioctl,
5783 #ifdef CONFIG_COMPAT
5784 .compat_ioctl = md_compat_ioctl,
5785 #endif
5786 .getgeo = md_getgeo,
5787 .media_changed = md_media_changed,
5788 .revalidate_disk= md_revalidate,
5789 };
5790
5791 static int md_thread(void * arg)
5792 {
5793 mdk_thread_t *thread = arg;
5794
5795 /*
5796 * md_thread is a 'system-thread', it's priority should be very
5797 * high. We avoid resource deadlocks individually in each
5798 * raid personality. (RAID5 does preallocation) We also use RR and
5799 * the very same RT priority as kswapd, thus we will never get
5800 * into a priority inversion deadlock.
5801 *
5802 * we definitely have to have equal or higher priority than
5803 * bdflush, otherwise bdflush will deadlock if there are too
5804 * many dirty RAID5 blocks.
5805 */
5806
5807 allow_signal(SIGKILL);
5808 while (!kthread_should_stop()) {
5809
5810 /* We need to wait INTERRUPTIBLE so that
5811 * we don't add to the load-average.
5812 * That means we need to be sure no signals are
5813 * pending
5814 */
5815 if (signal_pending(current))
5816 flush_signals(current);
5817
5818 wait_event_interruptible_timeout
5819 (thread->wqueue,
5820 test_bit(THREAD_WAKEUP, &thread->flags)
5821 || kthread_should_stop(),
5822 thread->timeout);
5823
5824 clear_bit(THREAD_WAKEUP, &thread->flags);
5825
5826 thread->run(thread->mddev);
5827 }
5828
5829 return 0;
5830 }
5831
5832 void md_wakeup_thread(mdk_thread_t *thread)
5833 {
5834 if (thread) {
5835 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5836 set_bit(THREAD_WAKEUP, &thread->flags);
5837 wake_up(&thread->wqueue);
5838 }
5839 }
5840
5841 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5842 const char *name)
5843 {
5844 mdk_thread_t *thread;
5845
5846 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5847 if (!thread)
5848 return NULL;
5849
5850 init_waitqueue_head(&thread->wqueue);
5851
5852 thread->run = run;
5853 thread->mddev = mddev;
5854 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5855 thread->tsk = kthread_run(md_thread, thread,
5856 "%s_%s",
5857 mdname(thread->mddev),
5858 name ?: mddev->pers->name);
5859 if (IS_ERR(thread->tsk)) {
5860 kfree(thread);
5861 return NULL;
5862 }
5863 return thread;
5864 }
5865
5866 void md_unregister_thread(mdk_thread_t *thread)
5867 {
5868 if (!thread)
5869 return;
5870 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5871
5872 kthread_stop(thread->tsk);
5873 kfree(thread);
5874 }
5875
5876 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5877 {
5878 if (!mddev) {
5879 MD_BUG();
5880 return;
5881 }
5882
5883 if (!rdev || test_bit(Faulty, &rdev->flags))
5884 return;
5885
5886 if (mddev->external)
5887 set_bit(Blocked, &rdev->flags);
5888 /*
5889 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5890 mdname(mddev),
5891 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5892 __builtin_return_address(0),__builtin_return_address(1),
5893 __builtin_return_address(2),__builtin_return_address(3));
5894 */
5895 if (!mddev->pers)
5896 return;
5897 if (!mddev->pers->error_handler)
5898 return;
5899 mddev->pers->error_handler(mddev,rdev);
5900 if (mddev->degraded)
5901 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5902 set_bit(StateChanged, &rdev->flags);
5903 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5904 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5905 md_wakeup_thread(mddev->thread);
5906 md_new_event_inintr(mddev);
5907 }
5908
5909 /* seq_file implementation /proc/mdstat */
5910
5911 static void status_unused(struct seq_file *seq)
5912 {
5913 int i = 0;
5914 mdk_rdev_t *rdev;
5915
5916 seq_printf(seq, "unused devices: ");
5917
5918 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5919 char b[BDEVNAME_SIZE];
5920 i++;
5921 seq_printf(seq, "%s ",
5922 bdevname(rdev->bdev,b));
5923 }
5924 if (!i)
5925 seq_printf(seq, "<none>");
5926
5927 seq_printf(seq, "\n");
5928 }
5929
5930
5931 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5932 {
5933 sector_t max_sectors, resync, res;
5934 unsigned long dt, db;
5935 sector_t rt;
5936 int scale;
5937 unsigned int per_milli;
5938
5939 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5940
5941 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5942 max_sectors = mddev->resync_max_sectors;
5943 else
5944 max_sectors = mddev->dev_sectors;
5945
5946 /*
5947 * Should not happen.
5948 */
5949 if (!max_sectors) {
5950 MD_BUG();
5951 return;
5952 }
5953 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5954 * in a sector_t, and (max_sectors>>scale) will fit in a
5955 * u32, as those are the requirements for sector_div.
5956 * Thus 'scale' must be at least 10
5957 */
5958 scale = 10;
5959 if (sizeof(sector_t) > sizeof(unsigned long)) {
5960 while ( max_sectors/2 > (1ULL<<(scale+32)))
5961 scale++;
5962 }
5963 res = (resync>>scale)*1000;
5964 sector_div(res, (u32)((max_sectors>>scale)+1));
5965
5966 per_milli = res;
5967 {
5968 int i, x = per_milli/50, y = 20-x;
5969 seq_printf(seq, "[");
5970 for (i = 0; i < x; i++)
5971 seq_printf(seq, "=");
5972 seq_printf(seq, ">");
5973 for (i = 0; i < y; i++)
5974 seq_printf(seq, ".");
5975 seq_printf(seq, "] ");
5976 }
5977 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5978 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5979 "reshape" :
5980 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5981 "check" :
5982 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5983 "resync" : "recovery"))),
5984 per_milli/10, per_milli % 10,
5985 (unsigned long long) resync/2,
5986 (unsigned long long) max_sectors/2);
5987
5988 /*
5989 * dt: time from mark until now
5990 * db: blocks written from mark until now
5991 * rt: remaining time
5992 *
5993 * rt is a sector_t, so could be 32bit or 64bit.
5994 * So we divide before multiply in case it is 32bit and close
5995 * to the limit.
5996 * We scale the divisor (db) by 32 to avoid loosing precision
5997 * near the end of resync when the number of remaining sectors
5998 * is close to 'db'.
5999 * We then divide rt by 32 after multiplying by db to compensate.
6000 * The '+1' avoids division by zero if db is very small.
6001 */
6002 dt = ((jiffies - mddev->resync_mark) / HZ);
6003 if (!dt) dt++;
6004 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6005 - mddev->resync_mark_cnt;
6006
6007 rt = max_sectors - resync; /* number of remaining sectors */
6008 sector_div(rt, db/32+1);
6009 rt *= dt;
6010 rt >>= 5;
6011
6012 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6013 ((unsigned long)rt % 60)/6);
6014
6015 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6016 }
6017
6018 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6019 {
6020 struct list_head *tmp;
6021 loff_t l = *pos;
6022 mddev_t *mddev;
6023
6024 if (l >= 0x10000)
6025 return NULL;
6026 if (!l--)
6027 /* header */
6028 return (void*)1;
6029
6030 spin_lock(&all_mddevs_lock);
6031 list_for_each(tmp,&all_mddevs)
6032 if (!l--) {
6033 mddev = list_entry(tmp, mddev_t, all_mddevs);
6034 mddev_get(mddev);
6035 spin_unlock(&all_mddevs_lock);
6036 return mddev;
6037 }
6038 spin_unlock(&all_mddevs_lock);
6039 if (!l--)
6040 return (void*)2;/* tail */
6041 return NULL;
6042 }
6043
6044 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6045 {
6046 struct list_head *tmp;
6047 mddev_t *next_mddev, *mddev = v;
6048
6049 ++*pos;
6050 if (v == (void*)2)
6051 return NULL;
6052
6053 spin_lock(&all_mddevs_lock);
6054 if (v == (void*)1)
6055 tmp = all_mddevs.next;
6056 else
6057 tmp = mddev->all_mddevs.next;
6058 if (tmp != &all_mddevs)
6059 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6060 else {
6061 next_mddev = (void*)2;
6062 *pos = 0x10000;
6063 }
6064 spin_unlock(&all_mddevs_lock);
6065
6066 if (v != (void*)1)
6067 mddev_put(mddev);
6068 return next_mddev;
6069
6070 }
6071
6072 static void md_seq_stop(struct seq_file *seq, void *v)
6073 {
6074 mddev_t *mddev = v;
6075
6076 if (mddev && v != (void*)1 && v != (void*)2)
6077 mddev_put(mddev);
6078 }
6079
6080 struct mdstat_info {
6081 int event;
6082 };
6083
6084 static int md_seq_show(struct seq_file *seq, void *v)
6085 {
6086 mddev_t *mddev = v;
6087 sector_t sectors;
6088 mdk_rdev_t *rdev;
6089 struct mdstat_info *mi = seq->private;
6090 struct bitmap *bitmap;
6091
6092 if (v == (void*)1) {
6093 struct mdk_personality *pers;
6094 seq_printf(seq, "Personalities : ");
6095 spin_lock(&pers_lock);
6096 list_for_each_entry(pers, &pers_list, list)
6097 seq_printf(seq, "[%s] ", pers->name);
6098
6099 spin_unlock(&pers_lock);
6100 seq_printf(seq, "\n");
6101 mi->event = atomic_read(&md_event_count);
6102 return 0;
6103 }
6104 if (v == (void*)2) {
6105 status_unused(seq);
6106 return 0;
6107 }
6108
6109 if (mddev_lock(mddev) < 0)
6110 return -EINTR;
6111
6112 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6113 seq_printf(seq, "%s : %sactive", mdname(mddev),
6114 mddev->pers ? "" : "in");
6115 if (mddev->pers) {
6116 if (mddev->ro==1)
6117 seq_printf(seq, " (read-only)");
6118 if (mddev->ro==2)
6119 seq_printf(seq, " (auto-read-only)");
6120 seq_printf(seq, " %s", mddev->pers->name);
6121 }
6122
6123 sectors = 0;
6124 list_for_each_entry(rdev, &mddev->disks, same_set) {
6125 char b[BDEVNAME_SIZE];
6126 seq_printf(seq, " %s[%d]",
6127 bdevname(rdev->bdev,b), rdev->desc_nr);
6128 if (test_bit(WriteMostly, &rdev->flags))
6129 seq_printf(seq, "(W)");
6130 if (test_bit(Faulty, &rdev->flags)) {
6131 seq_printf(seq, "(F)");
6132 continue;
6133 } else if (rdev->raid_disk < 0)
6134 seq_printf(seq, "(S)"); /* spare */
6135 sectors += rdev->sectors;
6136 }
6137
6138 if (!list_empty(&mddev->disks)) {
6139 if (mddev->pers)
6140 seq_printf(seq, "\n %llu blocks",
6141 (unsigned long long)
6142 mddev->array_sectors / 2);
6143 else
6144 seq_printf(seq, "\n %llu blocks",
6145 (unsigned long long)sectors / 2);
6146 }
6147 if (mddev->persistent) {
6148 if (mddev->major_version != 0 ||
6149 mddev->minor_version != 90) {
6150 seq_printf(seq," super %d.%d",
6151 mddev->major_version,
6152 mddev->minor_version);
6153 }
6154 } else if (mddev->external)
6155 seq_printf(seq, " super external:%s",
6156 mddev->metadata_type);
6157 else
6158 seq_printf(seq, " super non-persistent");
6159
6160 if (mddev->pers) {
6161 mddev->pers->status(seq, mddev);
6162 seq_printf(seq, "\n ");
6163 if (mddev->pers->sync_request) {
6164 if (mddev->curr_resync > 2) {
6165 status_resync(seq, mddev);
6166 seq_printf(seq, "\n ");
6167 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6168 seq_printf(seq, "\tresync=DELAYED\n ");
6169 else if (mddev->recovery_cp < MaxSector)
6170 seq_printf(seq, "\tresync=PENDING\n ");
6171 }
6172 } else
6173 seq_printf(seq, "\n ");
6174
6175 if ((bitmap = mddev->bitmap)) {
6176 unsigned long chunk_kb;
6177 unsigned long flags;
6178 spin_lock_irqsave(&bitmap->lock, flags);
6179 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6180 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6181 "%lu%s chunk",
6182 bitmap->pages - bitmap->missing_pages,
6183 bitmap->pages,
6184 (bitmap->pages - bitmap->missing_pages)
6185 << (PAGE_SHIFT - 10),
6186 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6187 chunk_kb ? "KB" : "B");
6188 if (bitmap->file) {
6189 seq_printf(seq, ", file: ");
6190 seq_path(seq, &bitmap->file->f_path, " \t\n");
6191 }
6192
6193 seq_printf(seq, "\n");
6194 spin_unlock_irqrestore(&bitmap->lock, flags);
6195 }
6196
6197 seq_printf(seq, "\n");
6198 }
6199 mddev_unlock(mddev);
6200
6201 return 0;
6202 }
6203
6204 static const struct seq_operations md_seq_ops = {
6205 .start = md_seq_start,
6206 .next = md_seq_next,
6207 .stop = md_seq_stop,
6208 .show = md_seq_show,
6209 };
6210
6211 static int md_seq_open(struct inode *inode, struct file *file)
6212 {
6213 int error;
6214 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6215 if (mi == NULL)
6216 return -ENOMEM;
6217
6218 error = seq_open(file, &md_seq_ops);
6219 if (error)
6220 kfree(mi);
6221 else {
6222 struct seq_file *p = file->private_data;
6223 p->private = mi;
6224 mi->event = atomic_read(&md_event_count);
6225 }
6226 return error;
6227 }
6228
6229 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6230 {
6231 struct seq_file *m = filp->private_data;
6232 struct mdstat_info *mi = m->private;
6233 int mask;
6234
6235 poll_wait(filp, &md_event_waiters, wait);
6236
6237 /* always allow read */
6238 mask = POLLIN | POLLRDNORM;
6239
6240 if (mi->event != atomic_read(&md_event_count))
6241 mask |= POLLERR | POLLPRI;
6242 return mask;
6243 }
6244
6245 static const struct file_operations md_seq_fops = {
6246 .owner = THIS_MODULE,
6247 .open = md_seq_open,
6248 .read = seq_read,
6249 .llseek = seq_lseek,
6250 .release = seq_release_private,
6251 .poll = mdstat_poll,
6252 };
6253
6254 int register_md_personality(struct mdk_personality *p)
6255 {
6256 spin_lock(&pers_lock);
6257 list_add_tail(&p->list, &pers_list);
6258 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6259 spin_unlock(&pers_lock);
6260 return 0;
6261 }
6262
6263 int unregister_md_personality(struct mdk_personality *p)
6264 {
6265 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6266 spin_lock(&pers_lock);
6267 list_del_init(&p->list);
6268 spin_unlock(&pers_lock);
6269 return 0;
6270 }
6271
6272 static int is_mddev_idle(mddev_t *mddev, int init)
6273 {
6274 mdk_rdev_t * rdev;
6275 int idle;
6276 int curr_events;
6277
6278 idle = 1;
6279 rcu_read_lock();
6280 rdev_for_each_rcu(rdev, mddev) {
6281 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6282 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6283 (int)part_stat_read(&disk->part0, sectors[1]) -
6284 atomic_read(&disk->sync_io);
6285 /* sync IO will cause sync_io to increase before the disk_stats
6286 * as sync_io is counted when a request starts, and
6287 * disk_stats is counted when it completes.
6288 * So resync activity will cause curr_events to be smaller than
6289 * when there was no such activity.
6290 * non-sync IO will cause disk_stat to increase without
6291 * increasing sync_io so curr_events will (eventually)
6292 * be larger than it was before. Once it becomes
6293 * substantially larger, the test below will cause
6294 * the array to appear non-idle, and resync will slow
6295 * down.
6296 * If there is a lot of outstanding resync activity when
6297 * we set last_event to curr_events, then all that activity
6298 * completing might cause the array to appear non-idle
6299 * and resync will be slowed down even though there might
6300 * not have been non-resync activity. This will only
6301 * happen once though. 'last_events' will soon reflect
6302 * the state where there is little or no outstanding
6303 * resync requests, and further resync activity will
6304 * always make curr_events less than last_events.
6305 *
6306 */
6307 if (init || curr_events - rdev->last_events > 64) {
6308 rdev->last_events = curr_events;
6309 idle = 0;
6310 }
6311 }
6312 rcu_read_unlock();
6313 return idle;
6314 }
6315
6316 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6317 {
6318 /* another "blocks" (512byte) blocks have been synced */
6319 atomic_sub(blocks, &mddev->recovery_active);
6320 wake_up(&mddev->recovery_wait);
6321 if (!ok) {
6322 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6323 md_wakeup_thread(mddev->thread);
6324 // stop recovery, signal do_sync ....
6325 }
6326 }
6327
6328
6329 /* md_write_start(mddev, bi)
6330 * If we need to update some array metadata (e.g. 'active' flag
6331 * in superblock) before writing, schedule a superblock update
6332 * and wait for it to complete.
6333 */
6334 void md_write_start(mddev_t *mddev, struct bio *bi)
6335 {
6336 int did_change = 0;
6337 if (bio_data_dir(bi) != WRITE)
6338 return;
6339
6340 BUG_ON(mddev->ro == 1);
6341 if (mddev->ro == 2) {
6342 /* need to switch to read/write */
6343 mddev->ro = 0;
6344 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6345 md_wakeup_thread(mddev->thread);
6346 md_wakeup_thread(mddev->sync_thread);
6347 did_change = 1;
6348 }
6349 atomic_inc(&mddev->writes_pending);
6350 if (mddev->safemode == 1)
6351 mddev->safemode = 0;
6352 if (mddev->in_sync) {
6353 spin_lock_irq(&mddev->write_lock);
6354 if (mddev->in_sync) {
6355 mddev->in_sync = 0;
6356 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6357 md_wakeup_thread(mddev->thread);
6358 did_change = 1;
6359 }
6360 spin_unlock_irq(&mddev->write_lock);
6361 }
6362 if (did_change)
6363 sysfs_notify_dirent(mddev->sysfs_state);
6364 wait_event(mddev->sb_wait,
6365 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6366 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6367 }
6368
6369 void md_write_end(mddev_t *mddev)
6370 {
6371 if (atomic_dec_and_test(&mddev->writes_pending)) {
6372 if (mddev->safemode == 2)
6373 md_wakeup_thread(mddev->thread);
6374 else if (mddev->safemode_delay)
6375 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6376 }
6377 }
6378
6379 /* md_allow_write(mddev)
6380 * Calling this ensures that the array is marked 'active' so that writes
6381 * may proceed without blocking. It is important to call this before
6382 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6383 * Must be called with mddev_lock held.
6384 *
6385 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6386 * is dropped, so return -EAGAIN after notifying userspace.
6387 */
6388 int md_allow_write(mddev_t *mddev)
6389 {
6390 if (!mddev->pers)
6391 return 0;
6392 if (mddev->ro)
6393 return 0;
6394 if (!mddev->pers->sync_request)
6395 return 0;
6396
6397 spin_lock_irq(&mddev->write_lock);
6398 if (mddev->in_sync) {
6399 mddev->in_sync = 0;
6400 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6401 if (mddev->safemode_delay &&
6402 mddev->safemode == 0)
6403 mddev->safemode = 1;
6404 spin_unlock_irq(&mddev->write_lock);
6405 md_update_sb(mddev, 0);
6406 sysfs_notify_dirent(mddev->sysfs_state);
6407 } else
6408 spin_unlock_irq(&mddev->write_lock);
6409
6410 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6411 return -EAGAIN;
6412 else
6413 return 0;
6414 }
6415 EXPORT_SYMBOL_GPL(md_allow_write);
6416
6417 #define SYNC_MARKS 10
6418 #define SYNC_MARK_STEP (3*HZ)
6419 void md_do_sync(mddev_t *mddev)
6420 {
6421 mddev_t *mddev2;
6422 unsigned int currspeed = 0,
6423 window;
6424 sector_t max_sectors,j, io_sectors;
6425 unsigned long mark[SYNC_MARKS];
6426 sector_t mark_cnt[SYNC_MARKS];
6427 int last_mark,m;
6428 struct list_head *tmp;
6429 sector_t last_check;
6430 int skipped = 0;
6431 mdk_rdev_t *rdev;
6432 char *desc;
6433
6434 /* just incase thread restarts... */
6435 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6436 return;
6437 if (mddev->ro) /* never try to sync a read-only array */
6438 return;
6439
6440 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6441 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6442 desc = "data-check";
6443 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6444 desc = "requested-resync";
6445 else
6446 desc = "resync";
6447 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6448 desc = "reshape";
6449 else
6450 desc = "recovery";
6451
6452 /* we overload curr_resync somewhat here.
6453 * 0 == not engaged in resync at all
6454 * 2 == checking that there is no conflict with another sync
6455 * 1 == like 2, but have yielded to allow conflicting resync to
6456 * commense
6457 * other == active in resync - this many blocks
6458 *
6459 * Before starting a resync we must have set curr_resync to
6460 * 2, and then checked that every "conflicting" array has curr_resync
6461 * less than ours. When we find one that is the same or higher
6462 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6463 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6464 * This will mean we have to start checking from the beginning again.
6465 *
6466 */
6467
6468 do {
6469 mddev->curr_resync = 2;
6470
6471 try_again:
6472 if (kthread_should_stop())
6473 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6474
6475 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6476 goto skip;
6477 for_each_mddev(mddev2, tmp) {
6478 if (mddev2 == mddev)
6479 continue;
6480 if (!mddev->parallel_resync
6481 && mddev2->curr_resync
6482 && match_mddev_units(mddev, mddev2)) {
6483 DEFINE_WAIT(wq);
6484 if (mddev < mddev2 && mddev->curr_resync == 2) {
6485 /* arbitrarily yield */
6486 mddev->curr_resync = 1;
6487 wake_up(&resync_wait);
6488 }
6489 if (mddev > mddev2 && mddev->curr_resync == 1)
6490 /* no need to wait here, we can wait the next
6491 * time 'round when curr_resync == 2
6492 */
6493 continue;
6494 /* We need to wait 'interruptible' so as not to
6495 * contribute to the load average, and not to
6496 * be caught by 'softlockup'
6497 */
6498 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6499 if (!kthread_should_stop() &&
6500 mddev2->curr_resync >= mddev->curr_resync) {
6501 printk(KERN_INFO "md: delaying %s of %s"
6502 " until %s has finished (they"
6503 " share one or more physical units)\n",
6504 desc, mdname(mddev), mdname(mddev2));
6505 mddev_put(mddev2);
6506 if (signal_pending(current))
6507 flush_signals(current);
6508 schedule();
6509 finish_wait(&resync_wait, &wq);
6510 goto try_again;
6511 }
6512 finish_wait(&resync_wait, &wq);
6513 }
6514 }
6515 } while (mddev->curr_resync < 2);
6516
6517 j = 0;
6518 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6519 /* resync follows the size requested by the personality,
6520 * which defaults to physical size, but can be virtual size
6521 */
6522 max_sectors = mddev->resync_max_sectors;
6523 mddev->resync_mismatches = 0;
6524 /* we don't use the checkpoint if there's a bitmap */
6525 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6526 j = mddev->resync_min;
6527 else if (!mddev->bitmap)
6528 j = mddev->recovery_cp;
6529
6530 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6531 max_sectors = mddev->dev_sectors;
6532 else {
6533 /* recovery follows the physical size of devices */
6534 max_sectors = mddev->dev_sectors;
6535 j = MaxSector;
6536 rcu_read_lock();
6537 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6538 if (rdev->raid_disk >= 0 &&
6539 !test_bit(Faulty, &rdev->flags) &&
6540 !test_bit(In_sync, &rdev->flags) &&
6541 rdev->recovery_offset < j)
6542 j = rdev->recovery_offset;
6543 rcu_read_unlock();
6544 }
6545
6546 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6547 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6548 " %d KB/sec/disk.\n", speed_min(mddev));
6549 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6550 "(but not more than %d KB/sec) for %s.\n",
6551 speed_max(mddev), desc);
6552
6553 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6554
6555 io_sectors = 0;
6556 for (m = 0; m < SYNC_MARKS; m++) {
6557 mark[m] = jiffies;
6558 mark_cnt[m] = io_sectors;
6559 }
6560 last_mark = 0;
6561 mddev->resync_mark = mark[last_mark];
6562 mddev->resync_mark_cnt = mark_cnt[last_mark];
6563
6564 /*
6565 * Tune reconstruction:
6566 */
6567 window = 32*(PAGE_SIZE/512);
6568 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6569 window/2,(unsigned long long) max_sectors/2);
6570
6571 atomic_set(&mddev->recovery_active, 0);
6572 last_check = 0;
6573
6574 if (j>2) {
6575 printk(KERN_INFO
6576 "md: resuming %s of %s from checkpoint.\n",
6577 desc, mdname(mddev));
6578 mddev->curr_resync = j;
6579 }
6580 mddev->curr_resync_completed = mddev->curr_resync;
6581
6582 while (j < max_sectors) {
6583 sector_t sectors;
6584
6585 skipped = 0;
6586
6587 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6588 ((mddev->curr_resync > mddev->curr_resync_completed &&
6589 (mddev->curr_resync - mddev->curr_resync_completed)
6590 > (max_sectors >> 4)) ||
6591 (j - mddev->curr_resync_completed)*2
6592 >= mddev->resync_max - mddev->curr_resync_completed
6593 )) {
6594 /* time to update curr_resync_completed */
6595 blk_unplug(mddev->queue);
6596 wait_event(mddev->recovery_wait,
6597 atomic_read(&mddev->recovery_active) == 0);
6598 mddev->curr_resync_completed =
6599 mddev->curr_resync;
6600 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6601 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6602 }
6603
6604 while (j >= mddev->resync_max && !kthread_should_stop()) {
6605 /* As this condition is controlled by user-space,
6606 * we can block indefinitely, so use '_interruptible'
6607 * to avoid triggering warnings.
6608 */
6609 flush_signals(current); /* just in case */
6610 wait_event_interruptible(mddev->recovery_wait,
6611 mddev->resync_max > j
6612 || kthread_should_stop());
6613 }
6614
6615 if (kthread_should_stop())
6616 goto interrupted;
6617
6618 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6619 currspeed < speed_min(mddev));
6620 if (sectors == 0) {
6621 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6622 goto out;
6623 }
6624
6625 if (!skipped) { /* actual IO requested */
6626 io_sectors += sectors;
6627 atomic_add(sectors, &mddev->recovery_active);
6628 }
6629
6630 j += sectors;
6631 if (j>1) mddev->curr_resync = j;
6632 mddev->curr_mark_cnt = io_sectors;
6633 if (last_check == 0)
6634 /* this is the earliers that rebuilt will be
6635 * visible in /proc/mdstat
6636 */
6637 md_new_event(mddev);
6638
6639 if (last_check + window > io_sectors || j == max_sectors)
6640 continue;
6641
6642 last_check = io_sectors;
6643
6644 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6645 break;
6646
6647 repeat:
6648 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6649 /* step marks */
6650 int next = (last_mark+1) % SYNC_MARKS;
6651
6652 mddev->resync_mark = mark[next];
6653 mddev->resync_mark_cnt = mark_cnt[next];
6654 mark[next] = jiffies;
6655 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6656 last_mark = next;
6657 }
6658
6659
6660 if (kthread_should_stop())
6661 goto interrupted;
6662
6663
6664 /*
6665 * this loop exits only if either when we are slower than
6666 * the 'hard' speed limit, or the system was IO-idle for
6667 * a jiffy.
6668 * the system might be non-idle CPU-wise, but we only care
6669 * about not overloading the IO subsystem. (things like an
6670 * e2fsck being done on the RAID array should execute fast)
6671 */
6672 blk_unplug(mddev->queue);
6673 cond_resched();
6674
6675 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6676 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6677
6678 if (currspeed > speed_min(mddev)) {
6679 if ((currspeed > speed_max(mddev)) ||
6680 !is_mddev_idle(mddev, 0)) {
6681 msleep(500);
6682 goto repeat;
6683 }
6684 }
6685 }
6686 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6687 /*
6688 * this also signals 'finished resyncing' to md_stop
6689 */
6690 out:
6691 blk_unplug(mddev->queue);
6692
6693 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6694
6695 /* tell personality that we are finished */
6696 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6697
6698 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6699 mddev->curr_resync > 2) {
6700 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6701 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6702 if (mddev->curr_resync >= mddev->recovery_cp) {
6703 printk(KERN_INFO
6704 "md: checkpointing %s of %s.\n",
6705 desc, mdname(mddev));
6706 mddev->recovery_cp = mddev->curr_resync;
6707 }
6708 } else
6709 mddev->recovery_cp = MaxSector;
6710 } else {
6711 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6712 mddev->curr_resync = MaxSector;
6713 rcu_read_lock();
6714 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6715 if (rdev->raid_disk >= 0 &&
6716 !test_bit(Faulty, &rdev->flags) &&
6717 !test_bit(In_sync, &rdev->flags) &&
6718 rdev->recovery_offset < mddev->curr_resync)
6719 rdev->recovery_offset = mddev->curr_resync;
6720 rcu_read_unlock();
6721 }
6722 }
6723 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6724
6725 skip:
6726 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6727 /* We completed so min/max setting can be forgotten if used. */
6728 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6729 mddev->resync_min = 0;
6730 mddev->resync_max = MaxSector;
6731 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6732 mddev->resync_min = mddev->curr_resync_completed;
6733 mddev->curr_resync = 0;
6734 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6735 mddev->curr_resync_completed = 0;
6736 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6737 wake_up(&resync_wait);
6738 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6739 md_wakeup_thread(mddev->thread);
6740 return;
6741
6742 interrupted:
6743 /*
6744 * got a signal, exit.
6745 */
6746 printk(KERN_INFO
6747 "md: md_do_sync() got signal ... exiting\n");
6748 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6749 goto out;
6750
6751 }
6752 EXPORT_SYMBOL_GPL(md_do_sync);
6753
6754
6755 static int remove_and_add_spares(mddev_t *mddev)
6756 {
6757 mdk_rdev_t *rdev;
6758 int spares = 0;
6759
6760 mddev->curr_resync_completed = 0;
6761
6762 list_for_each_entry(rdev, &mddev->disks, same_set)
6763 if (rdev->raid_disk >= 0 &&
6764 !test_bit(Blocked, &rdev->flags) &&
6765 (test_bit(Faulty, &rdev->flags) ||
6766 ! test_bit(In_sync, &rdev->flags)) &&
6767 atomic_read(&rdev->nr_pending)==0) {
6768 if (mddev->pers->hot_remove_disk(
6769 mddev, rdev->raid_disk)==0) {
6770 char nm[20];
6771 sprintf(nm,"rd%d", rdev->raid_disk);
6772 sysfs_remove_link(&mddev->kobj, nm);
6773 rdev->raid_disk = -1;
6774 }
6775 }
6776
6777 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6778 list_for_each_entry(rdev, &mddev->disks, same_set) {
6779 if (rdev->raid_disk >= 0 &&
6780 !test_bit(In_sync, &rdev->flags) &&
6781 !test_bit(Blocked, &rdev->flags))
6782 spares++;
6783 if (rdev->raid_disk < 0
6784 && !test_bit(Faulty, &rdev->flags)) {
6785 rdev->recovery_offset = 0;
6786 if (mddev->pers->
6787 hot_add_disk(mddev, rdev) == 0) {
6788 char nm[20];
6789 sprintf(nm, "rd%d", rdev->raid_disk);
6790 if (sysfs_create_link(&mddev->kobj,
6791 &rdev->kobj, nm))
6792 printk(KERN_WARNING
6793 "md: cannot register "
6794 "%s for %s\n",
6795 nm, mdname(mddev));
6796 spares++;
6797 md_new_event(mddev);
6798 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6799 } else
6800 break;
6801 }
6802 }
6803 }
6804 return spares;
6805 }
6806 /*
6807 * This routine is regularly called by all per-raid-array threads to
6808 * deal with generic issues like resync and super-block update.
6809 * Raid personalities that don't have a thread (linear/raid0) do not
6810 * need this as they never do any recovery or update the superblock.
6811 *
6812 * It does not do any resync itself, but rather "forks" off other threads
6813 * to do that as needed.
6814 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6815 * "->recovery" and create a thread at ->sync_thread.
6816 * When the thread finishes it sets MD_RECOVERY_DONE
6817 * and wakeups up this thread which will reap the thread and finish up.
6818 * This thread also removes any faulty devices (with nr_pending == 0).
6819 *
6820 * The overall approach is:
6821 * 1/ if the superblock needs updating, update it.
6822 * 2/ If a recovery thread is running, don't do anything else.
6823 * 3/ If recovery has finished, clean up, possibly marking spares active.
6824 * 4/ If there are any faulty devices, remove them.
6825 * 5/ If array is degraded, try to add spares devices
6826 * 6/ If array has spares or is not in-sync, start a resync thread.
6827 */
6828 void md_check_recovery(mddev_t *mddev)
6829 {
6830 mdk_rdev_t *rdev;
6831
6832
6833 if (mddev->bitmap)
6834 bitmap_daemon_work(mddev);
6835
6836 if (mddev->ro)
6837 return;
6838
6839 if (signal_pending(current)) {
6840 if (mddev->pers->sync_request && !mddev->external) {
6841 printk(KERN_INFO "md: %s in immediate safe mode\n",
6842 mdname(mddev));
6843 mddev->safemode = 2;
6844 }
6845 flush_signals(current);
6846 }
6847
6848 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6849 return;
6850 if ( ! (
6851 (mddev->flags && !mddev->external) ||
6852 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6853 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6854 (mddev->external == 0 && mddev->safemode == 1) ||
6855 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6856 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6857 ))
6858 return;
6859
6860 if (mddev_trylock(mddev)) {
6861 int spares = 0;
6862
6863 if (mddev->ro) {
6864 /* Only thing we do on a ro array is remove
6865 * failed devices.
6866 */
6867 remove_and_add_spares(mddev);
6868 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6869 goto unlock;
6870 }
6871
6872 if (!mddev->external) {
6873 int did_change = 0;
6874 spin_lock_irq(&mddev->write_lock);
6875 if (mddev->safemode &&
6876 !atomic_read(&mddev->writes_pending) &&
6877 !mddev->in_sync &&
6878 mddev->recovery_cp == MaxSector) {
6879 mddev->in_sync = 1;
6880 did_change = 1;
6881 if (mddev->persistent)
6882 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6883 }
6884 if (mddev->safemode == 1)
6885 mddev->safemode = 0;
6886 spin_unlock_irq(&mddev->write_lock);
6887 if (did_change)
6888 sysfs_notify_dirent(mddev->sysfs_state);
6889 }
6890
6891 if (mddev->flags)
6892 md_update_sb(mddev, 0);
6893
6894 list_for_each_entry(rdev, &mddev->disks, same_set)
6895 if (test_and_clear_bit(StateChanged, &rdev->flags))
6896 sysfs_notify_dirent(rdev->sysfs_state);
6897
6898
6899 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6900 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6901 /* resync/recovery still happening */
6902 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6903 goto unlock;
6904 }
6905 if (mddev->sync_thread) {
6906 /* resync has finished, collect result */
6907 md_unregister_thread(mddev->sync_thread);
6908 mddev->sync_thread = NULL;
6909 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6910 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6911 /* success...*/
6912 /* activate any spares */
6913 if (mddev->pers->spare_active(mddev))
6914 sysfs_notify(&mddev->kobj, NULL,
6915 "degraded");
6916 }
6917 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6918 mddev->pers->finish_reshape)
6919 mddev->pers->finish_reshape(mddev);
6920 md_update_sb(mddev, 1);
6921
6922 /* if array is no-longer degraded, then any saved_raid_disk
6923 * information must be scrapped
6924 */
6925 if (!mddev->degraded)
6926 list_for_each_entry(rdev, &mddev->disks, same_set)
6927 rdev->saved_raid_disk = -1;
6928
6929 mddev->recovery = 0;
6930 /* flag recovery needed just to double check */
6931 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6932 sysfs_notify_dirent(mddev->sysfs_action);
6933 md_new_event(mddev);
6934 goto unlock;
6935 }
6936 /* Set RUNNING before clearing NEEDED to avoid
6937 * any transients in the value of "sync_action".
6938 */
6939 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6940 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6941 /* Clear some bits that don't mean anything, but
6942 * might be left set
6943 */
6944 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6945 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6946
6947 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6948 goto unlock;
6949 /* no recovery is running.
6950 * remove any failed drives, then
6951 * add spares if possible.
6952 * Spare are also removed and re-added, to allow
6953 * the personality to fail the re-add.
6954 */
6955
6956 if (mddev->reshape_position != MaxSector) {
6957 if (mddev->pers->check_reshape == NULL ||
6958 mddev->pers->check_reshape(mddev) != 0)
6959 /* Cannot proceed */
6960 goto unlock;
6961 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6962 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6963 } else if ((spares = remove_and_add_spares(mddev))) {
6964 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6965 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6966 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6967 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6968 } else if (mddev->recovery_cp < MaxSector) {
6969 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6970 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6971 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6972 /* nothing to be done ... */
6973 goto unlock;
6974
6975 if (mddev->pers->sync_request) {
6976 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6977 /* We are adding a device or devices to an array
6978 * which has the bitmap stored on all devices.
6979 * So make sure all bitmap pages get written
6980 */
6981 bitmap_write_all(mddev->bitmap);
6982 }
6983 mddev->sync_thread = md_register_thread(md_do_sync,
6984 mddev,
6985 "resync");
6986 if (!mddev->sync_thread) {
6987 printk(KERN_ERR "%s: could not start resync"
6988 " thread...\n",
6989 mdname(mddev));
6990 /* leave the spares where they are, it shouldn't hurt */
6991 mddev->recovery = 0;
6992 } else
6993 md_wakeup_thread(mddev->sync_thread);
6994 sysfs_notify_dirent(mddev->sysfs_action);
6995 md_new_event(mddev);
6996 }
6997 unlock:
6998 if (!mddev->sync_thread) {
6999 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7000 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7001 &mddev->recovery))
7002 if (mddev->sysfs_action)
7003 sysfs_notify_dirent(mddev->sysfs_action);
7004 }
7005 mddev_unlock(mddev);
7006 }
7007 }
7008
7009 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7010 {
7011 sysfs_notify_dirent(rdev->sysfs_state);
7012 wait_event_timeout(rdev->blocked_wait,
7013 !test_bit(Blocked, &rdev->flags),
7014 msecs_to_jiffies(5000));
7015 rdev_dec_pending(rdev, mddev);
7016 }
7017 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7018
7019 static int md_notify_reboot(struct notifier_block *this,
7020 unsigned long code, void *x)
7021 {
7022 struct list_head *tmp;
7023 mddev_t *mddev;
7024
7025 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7026
7027 printk(KERN_INFO "md: stopping all md devices.\n");
7028
7029 for_each_mddev(mddev, tmp)
7030 if (mddev_trylock(mddev)) {
7031 /* Force a switch to readonly even array
7032 * appears to still be in use. Hence
7033 * the '100'.
7034 */
7035 do_md_stop(mddev, 1, 100);
7036 mddev_unlock(mddev);
7037 }
7038 /*
7039 * certain more exotic SCSI devices are known to be
7040 * volatile wrt too early system reboots. While the
7041 * right place to handle this issue is the given
7042 * driver, we do want to have a safe RAID driver ...
7043 */
7044 mdelay(1000*1);
7045 }
7046 return NOTIFY_DONE;
7047 }
7048
7049 static struct notifier_block md_notifier = {
7050 .notifier_call = md_notify_reboot,
7051 .next = NULL,
7052 .priority = INT_MAX, /* before any real devices */
7053 };
7054
7055 static void md_geninit(void)
7056 {
7057 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7058
7059 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7060 }
7061
7062 static int __init md_init(void)
7063 {
7064 if (register_blkdev(MD_MAJOR, "md"))
7065 return -1;
7066 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7067 unregister_blkdev(MD_MAJOR, "md");
7068 return -1;
7069 }
7070 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7071 md_probe, NULL, NULL);
7072 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7073 md_probe, NULL, NULL);
7074
7075 register_reboot_notifier(&md_notifier);
7076 raid_table_header = register_sysctl_table(raid_root_table);
7077
7078 md_geninit();
7079 return 0;
7080 }
7081
7082
7083 #ifndef MODULE
7084
7085 /*
7086 * Searches all registered partitions for autorun RAID arrays
7087 * at boot time.
7088 */
7089
7090 static LIST_HEAD(all_detected_devices);
7091 struct detected_devices_node {
7092 struct list_head list;
7093 dev_t dev;
7094 };
7095
7096 void md_autodetect_dev(dev_t dev)
7097 {
7098 struct detected_devices_node *node_detected_dev;
7099
7100 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7101 if (node_detected_dev) {
7102 node_detected_dev->dev = dev;
7103 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7104 } else {
7105 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7106 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7107 }
7108 }
7109
7110
7111 static void autostart_arrays(int part)
7112 {
7113 mdk_rdev_t *rdev;
7114 struct detected_devices_node *node_detected_dev;
7115 dev_t dev;
7116 int i_scanned, i_passed;
7117
7118 i_scanned = 0;
7119 i_passed = 0;
7120
7121 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7122
7123 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7124 i_scanned++;
7125 node_detected_dev = list_entry(all_detected_devices.next,
7126 struct detected_devices_node, list);
7127 list_del(&node_detected_dev->list);
7128 dev = node_detected_dev->dev;
7129 kfree(node_detected_dev);
7130 rdev = md_import_device(dev,0, 90);
7131 if (IS_ERR(rdev))
7132 continue;
7133
7134 if (test_bit(Faulty, &rdev->flags)) {
7135 MD_BUG();
7136 continue;
7137 }
7138 set_bit(AutoDetected, &rdev->flags);
7139 list_add(&rdev->same_set, &pending_raid_disks);
7140 i_passed++;
7141 }
7142
7143 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7144 i_scanned, i_passed);
7145
7146 autorun_devices(part);
7147 }
7148
7149 #endif /* !MODULE */
7150
7151 static __exit void md_exit(void)
7152 {
7153 mddev_t *mddev;
7154 struct list_head *tmp;
7155
7156 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7157 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7158
7159 unregister_blkdev(MD_MAJOR,"md");
7160 unregister_blkdev(mdp_major, "mdp");
7161 unregister_reboot_notifier(&md_notifier);
7162 unregister_sysctl_table(raid_table_header);
7163 remove_proc_entry("mdstat", NULL);
7164 for_each_mddev(mddev, tmp) {
7165 export_array(mddev);
7166 mddev->hold_active = 0;
7167 }
7168 }
7169
7170 subsys_initcall(md_init);
7171 module_exit(md_exit)
7172
7173 static int get_ro(char *buffer, struct kernel_param *kp)
7174 {
7175 return sprintf(buffer, "%d", start_readonly);
7176 }
7177 static int set_ro(const char *val, struct kernel_param *kp)
7178 {
7179 char *e;
7180 int num = simple_strtoul(val, &e, 10);
7181 if (*val && (*e == '\0' || *e == '\n')) {
7182 start_readonly = num;
7183 return 0;
7184 }
7185 return -EINVAL;
7186 }
7187
7188 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7189 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7190
7191 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7192
7193 EXPORT_SYMBOL(register_md_personality);
7194 EXPORT_SYMBOL(unregister_md_personality);
7195 EXPORT_SYMBOL(md_error);
7196 EXPORT_SYMBOL(md_done_sync);
7197 EXPORT_SYMBOL(md_write_start);
7198 EXPORT_SYMBOL(md_write_end);
7199 EXPORT_SYMBOL(md_register_thread);
7200 EXPORT_SYMBOL(md_unregister_thread);
7201 EXPORT_SYMBOL(md_wakeup_thread);
7202 EXPORT_SYMBOL(md_check_recovery);
7203 MODULE_LICENSE("GPL");
7204 MODULE_DESCRIPTION("MD RAID framework");
7205 MODULE_ALIAS("md");
7206 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree