?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
GUI: Fix Tomato RAF theme for all builds. Compilation typo.
[tomato.git] / release / src-rt-6.x.4708 / linux / linux-2.6.36 / drivers / md / md.c
blob012859708a1b2a10315691810c61f7dd78d7481b
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/smp_lock.h>
40 #include <linux/buffer_head.h> /* for invalidate_bdev */
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
54 #include "md.h"
55 #include "bitmap.h"
56
57 #define DEBUG 0
58 #define dprintk(x...) ((void)(DEBUG && printk(x)))
59
60
61 #ifndef MODULE
62 static void autostart_arrays(int part);
63 #endif
64
65 static LIST_HEAD(pers_list);
66 static DEFINE_SPINLOCK(pers_lock);
67
68 static void md_print_devices(void);
69
70 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
71
72 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
73
74 /*
75 * Default number of read corrections we'll attempt on an rdev
76 * before ejecting it from the array. We divide the read error
77 * count by 2 for every hour elapsed between read errors.
78 */
79 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
80 /*
81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
82 * is 1000 KB/sec, so the extra system load does not show up that much.
83 * Increase it if you want to have more _guaranteed_ speed. Note that
84 * the RAID driver will use the maximum available bandwidth if the IO
85 * subsystem is idle. There is also an 'absolute maximum' reconstruction
86 * speed limit - in case reconstruction slows down your system despite
87 * idle IO detection.
88 *
89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
90 * or /sys/block/mdX/md/sync_speed_{min,max}
91 */
92
93 static int sysctl_speed_limit_min = 1000;
94 static int sysctl_speed_limit_max = 200000;
95 static inline int speed_min(mddev_t *mddev)
96 {
97 return mddev->sync_speed_min ?
98 mddev->sync_speed_min : sysctl_speed_limit_min;
99 }
100
101 static inline int speed_max(mddev_t *mddev)
102 {
103 return mddev->sync_speed_max ?
104 mddev->sync_speed_max : sysctl_speed_limit_max;
105 }
106
107 static struct ctl_table_header *raid_table_header;
108
109 static ctl_table raid_table[] = {
110 {
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = proc_dointvec,
116 },
117 {
118 .procname = "speed_limit_max",
119 .data = &sysctl_speed_limit_max,
120 .maxlen = sizeof(int),
121 .mode = S_IRUGO|S_IWUSR,
122 .proc_handler = proc_dointvec,
123 },
124 { }
125 };
126
127 static ctl_table raid_dir_table[] = {
128 {
129 .procname = "raid",
130 .maxlen = 0,
131 .mode = S_IRUGO|S_IXUGO,
132 .child = raid_table,
133 },
134 { }
135 };
136
137 static ctl_table raid_root_table[] = {
138 {
139 .procname = "dev",
140 .maxlen = 0,
141 .mode = 0555,
142 .child = raid_dir_table,
143 },
144 { }
145 };
146
147 static const struct block_device_operations md_fops;
148
149 static int start_readonly;
150
151 /*
152 * We have a system wide 'event count' that is incremented
153 * on any 'interesting' event, and readers of /proc/mdstat
154 * can use 'poll' or 'select' to find out when the event
155 * count increases.
156 *
157 * Events are:
158 * start array, stop array, error, add device, remove device,
159 * start build, activate spare
160 */
161 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
162 static atomic_t md_event_count;
163 void md_new_event(mddev_t *mddev)
164 {
165 atomic_inc(&md_event_count);
166 wake_up(&md_event_waiters);
167 }
168 EXPORT_SYMBOL_GPL(md_new_event);
169
170 /* Alternate version that can be called from interrupts
171 * when calling sysfs_notify isn't needed.
172 */
173 static void md_new_event_inintr(mddev_t *mddev)
174 {
175 atomic_inc(&md_event_count);
176 wake_up(&md_event_waiters);
177 }
178
179 /*
180 * Enables to iterate over all existing md arrays
181 * all_mddevs_lock protects this list.
182 */
183 static LIST_HEAD(all_mddevs);
184 static DEFINE_SPINLOCK(all_mddevs_lock);
185
186
187 /*
188 * iterates through all used mddevs in the system.
189 * We take care to grab the all_mddevs_lock whenever navigating
190 * the list, and to always hold a refcount when unlocked.
191 * Any code which breaks out of this loop while own
192 * a reference to the current mddev and must mddev_put it.
193 */
194 #define for_each_mddev(mddev,tmp) \
195 \
196 for (({ spin_lock(&all_mddevs_lock); \
197 tmp = all_mddevs.next; \
198 mddev = NULL;}); \
199 ({ if (tmp != &all_mddevs) \
200 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
201 spin_unlock(&all_mddevs_lock); \
202 if (mddev) mddev_put(mddev); \
203 mddev = list_entry(tmp, mddev_t, all_mddevs); \
204 tmp != &all_mddevs;}); \
205 ({ spin_lock(&all_mddevs_lock); \
206 tmp = tmp->next;}) \
207 )
208
209
210 /* Rather than calling directly into the personality make_request function,
211 * IO requests come here first so that we can check if the device is
212 * being suspended pending a reconfiguration.
213 * We hold a refcount over the call to ->make_request. By the time that
214 * call has finished, the bio has been linked into some internal structure
215 * and so is visible to ->quiesce(), so we don't need the refcount any more.
216 */
217 static int md_make_request(struct request_queue *q, struct bio *bio)
218 {
219 const int rw = bio_data_dir(bio);
220 mddev_t *mddev = q->queuedata;
221 int rv;
222 int cpu;
223 unsigned int sectors;
224
225 if (mddev == NULL || mddev->pers == NULL
226 || !mddev->ready) {
227 bio_io_error(bio);
228 return 0;
229 }
230 smp_rmb(); /* Ensure implications of 'active' are visible */
231 rcu_read_lock();
232 if (mddev->suspended || mddev->barrier) {
233 DEFINE_WAIT(__wait);
234 for (;;) {
235 prepare_to_wait(&mddev->sb_wait, &__wait,
236 TASK_UNINTERRUPTIBLE);
237 if (!mddev->suspended && !mddev->barrier)
238 break;
239 rcu_read_unlock();
240 schedule();
241 rcu_read_lock();
242 }
243 finish_wait(&mddev->sb_wait, &__wait);
244 }
245 atomic_inc(&mddev->active_io);
246 rcu_read_unlock();
247
248 /*
249 * save the sectors now since our bio can
250 * go away inside make_request
251 */
252 sectors = bio_sectors(bio);
253 rv = mddev->pers->make_request(mddev, bio);
254
255 cpu = part_stat_lock();
256 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
257 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
258 part_stat_unlock();
259
260 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
261 wake_up(&mddev->sb_wait);
262
263 return rv;
264 }
265
266 /* mddev_suspend makes sure no new requests are submitted
267 * to the device, and that any requests that have been submitted
268 * are completely handled.
269 * Once ->stop is called and completes, the module will be completely
270 * unused.
271 */
272 void mddev_suspend(mddev_t *mddev)
273 {
274 BUG_ON(mddev->suspended);
275 mddev->suspended = 1;
276 synchronize_rcu();
277 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
278 mddev->pers->quiesce(mddev, 1);
279 }
280 EXPORT_SYMBOL_GPL(mddev_suspend);
281
282 void mddev_resume(mddev_t *mddev)
283 {
284 mddev->suspended = 0;
285 wake_up(&mddev->sb_wait);
286 mddev->pers->quiesce(mddev, 0);
287 }
288 EXPORT_SYMBOL_GPL(mddev_resume);
289
290 int mddev_congested(mddev_t *mddev, int bits)
291 {
292 if (mddev->barrier)
293 return 1;
294 return mddev->suspended;
295 }
296 EXPORT_SYMBOL(mddev_congested);
297
298 /*
299 * Generic barrier handling for md
300 */
301
302 #define POST_REQUEST_BARRIER ((void*)1)
303
304 static void md_end_barrier(struct bio *bio, int err)
305 {
306 mdk_rdev_t *rdev = bio->bi_private;
307 mddev_t *mddev = rdev->mddev;
308 if (err == -EOPNOTSUPP && mddev->barrier != POST_REQUEST_BARRIER)
309 set_bit(BIO_EOPNOTSUPP, &mddev->barrier->bi_flags);
310
311 rdev_dec_pending(rdev, mddev);
312
313 if (atomic_dec_and_test(&mddev->flush_pending)) {
314 if (mddev->barrier == POST_REQUEST_BARRIER) {
315 /* This was a post-request barrier */
316 mddev->barrier = NULL;
317 wake_up(&mddev->sb_wait);
318 } else
319 /* The pre-request barrier has finished */
320 schedule_work(&mddev->barrier_work);
321 }
322 bio_put(bio);
323 }
324
325 static void submit_barriers(mddev_t *mddev)
326 {
327 mdk_rdev_t *rdev;
328
329 rcu_read_lock();
330 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
331 if (rdev->raid_disk >= 0 &&
332 !test_bit(Faulty, &rdev->flags)) {
333 /* Take two references, one is dropped
334 * when request finishes, one after
335 * we reclaim rcu_read_lock
336 */
337 struct bio *bi;
338 atomic_inc(&rdev->nr_pending);
339 atomic_inc(&rdev->nr_pending);
340 rcu_read_unlock();
341 bi = bio_alloc(GFP_KERNEL, 0);
342 bi->bi_end_io = md_end_barrier;
343 bi->bi_private = rdev;
344 bi->bi_bdev = rdev->bdev;
345 atomic_inc(&mddev->flush_pending);
346 submit_bio(WRITE_BARRIER, bi);
347 rcu_read_lock();
348 rdev_dec_pending(rdev, mddev);
349 }
350 rcu_read_unlock();
351 }
352
353 static void md_submit_barrier(struct work_struct *ws)
354 {
355 mddev_t *mddev = container_of(ws, mddev_t, barrier_work);
356 struct bio *bio = mddev->barrier;
357
358 atomic_set(&mddev->flush_pending, 1);
359
360 if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
361 bio_endio(bio, -EOPNOTSUPP);
362 else if (bio->bi_size == 0)
363 /* an empty barrier - all done */
364 bio_endio(bio, 0);
365 else {
366 bio->bi_rw &= ~REQ_HARDBARRIER;
367 if (mddev->pers->make_request(mddev, bio))
368 generic_make_request(bio);
369 mddev->barrier = POST_REQUEST_BARRIER;
370 submit_barriers(mddev);
371 }
372 if (atomic_dec_and_test(&mddev->flush_pending)) {
373 mddev->barrier = NULL;
374 wake_up(&mddev->sb_wait);
375 }
376 }
377
378 void md_barrier_request(mddev_t *mddev, struct bio *bio)
379 {
380 spin_lock_irq(&mddev->write_lock);
381 wait_event_lock_irq(mddev->sb_wait,
382 !mddev->barrier,
383 mddev->write_lock, /*nothing*/);
384 mddev->barrier = bio;
385 spin_unlock_irq(&mddev->write_lock);
386
387 atomic_set(&mddev->flush_pending, 1);
388 INIT_WORK(&mddev->barrier_work, md_submit_barrier);
389
390 submit_barriers(mddev);
391
392 if (atomic_dec_and_test(&mddev->flush_pending))
393 schedule_work(&mddev->barrier_work);
394 }
395 EXPORT_SYMBOL(md_barrier_request);
396
397 /* Support for plugging.
398 * This mirrors the plugging support in request_queue, but does not
399 * require having a whole queue
400 */
401 static void plugger_work(struct work_struct *work)
402 {
403 struct plug_handle *plug =
404 container_of(work, struct plug_handle, unplug_work);
405 plug->unplug_fn(plug);
406 }
407 static void plugger_timeout(unsigned long data)
408 {
409 struct plug_handle *plug = (void *)data;
410 kblockd_schedule_work(NULL, &plug->unplug_work);
411 }
412 void plugger_init(struct plug_handle *plug,
413 void (*unplug_fn)(struct plug_handle *))
414 {
415 plug->unplug_flag = 0;
416 plug->unplug_fn = unplug_fn;
417 init_timer(&plug->unplug_timer);
418 plug->unplug_timer.function = plugger_timeout;
419 plug->unplug_timer.data = (unsigned long)plug;
420 INIT_WORK(&plug->unplug_work, plugger_work);
421 }
422 EXPORT_SYMBOL_GPL(plugger_init);
423
424 void plugger_set_plug(struct plug_handle *plug)
425 {
426 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
427 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
428 }
429 EXPORT_SYMBOL_GPL(plugger_set_plug);
430
431 int plugger_remove_plug(struct plug_handle *plug)
432 {
433 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
434 del_timer(&plug->unplug_timer);
435 return 1;
436 } else
437 return 0;
438 }
439 EXPORT_SYMBOL_GPL(plugger_remove_plug);
440
441
442 static inline mddev_t *mddev_get(mddev_t *mddev)
443 {
444 atomic_inc(&mddev->active);
445 return mddev;
446 }
447
448 static void mddev_delayed_delete(struct work_struct *ws);
449
450 static void mddev_put(mddev_t *mddev)
451 {
452 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
453 return;
454 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
455 mddev->ctime == 0 && !mddev->hold_active) {
456 /* Array is not configured at all, and not held active,
457 * so destroy it */
458 list_del(&mddev->all_mddevs);
459 if (mddev->gendisk) {
460 /* we did a probe so need to clean up.
461 * Call schedule_work inside the spinlock
462 * so that flush_scheduled_work() after
463 * mddev_find will succeed in waiting for the
464 * work to be done.
465 */
466 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
467 schedule_work(&mddev->del_work);
468 } else
469 kfree(mddev);
470 }
471 spin_unlock(&all_mddevs_lock);
472 }
473
474 void mddev_init(mddev_t *mddev)
475 {
476 mutex_init(&mddev->open_mutex);
477 mutex_init(&mddev->reconfig_mutex);
478 mutex_init(&mddev->bitmap_info.mutex);
479 INIT_LIST_HEAD(&mddev->disks);
480 INIT_LIST_HEAD(&mddev->all_mddevs);
481 init_timer(&mddev->safemode_timer);
482 atomic_set(&mddev->active, 1);
483 atomic_set(&mddev->openers, 0);
484 atomic_set(&mddev->active_io, 0);
485 spin_lock_init(&mddev->write_lock);
486 atomic_set(&mddev->flush_pending, 0);
487 init_waitqueue_head(&mddev->sb_wait);
488 init_waitqueue_head(&mddev->recovery_wait);
489 mddev->reshape_position = MaxSector;
490 mddev->resync_min = 0;
491 mddev->resync_max = MaxSector;
492 mddev->level = LEVEL_NONE;
493 }
494 EXPORT_SYMBOL_GPL(mddev_init);
495
496 static mddev_t * mddev_find(dev_t unit)
497 {
498 mddev_t *mddev, *new = NULL;
499
500 retry:
501 spin_lock(&all_mddevs_lock);
502
503 if (unit) {
504 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
505 if (mddev->unit == unit) {
506 mddev_get(mddev);
507 spin_unlock(&all_mddevs_lock);
508 kfree(new);
509 return mddev;
510 }
511
512 if (new) {
513 list_add(&new->all_mddevs, &all_mddevs);
514 spin_unlock(&all_mddevs_lock);
515 new->hold_active = UNTIL_IOCTL;
516 return new;
517 }
518 } else if (new) {
519 /* find an unused unit number */
520 static int next_minor = 512;
521 int start = next_minor;
522 int is_free = 0;
523 int dev = 0;
524 while (!is_free) {
525 dev = MKDEV(MD_MAJOR, next_minor);
526 next_minor++;
527 if (next_minor > MINORMASK)
528 next_minor = 0;
529 if (next_minor == start) {
530 /* Oh dear, all in use. */
531 spin_unlock(&all_mddevs_lock);
532 kfree(new);
533 return NULL;
534 }
535
536 is_free = 1;
537 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
538 if (mddev->unit == dev) {
539 is_free = 0;
540 break;
541 }
542 }
543 new->unit = dev;
544 new->md_minor = MINOR(dev);
545 new->hold_active = UNTIL_STOP;
546 list_add(&new->all_mddevs, &all_mddevs);
547 spin_unlock(&all_mddevs_lock);
548 return new;
549 }
550 spin_unlock(&all_mddevs_lock);
551
552 new = kzalloc(sizeof(*new), GFP_KERNEL);
553 if (!new)
554 return NULL;
555
556 new->unit = unit;
557 if (MAJOR(unit) == MD_MAJOR)
558 new->md_minor = MINOR(unit);
559 else
560 new->md_minor = MINOR(unit) >> MdpMinorShift;
561
562 mddev_init(new);
563
564 goto retry;
565 }
566
567 static inline int mddev_lock(mddev_t * mddev)
568 {
569 return mutex_lock_interruptible(&mddev->reconfig_mutex);
570 }
571
572 static inline int mddev_is_locked(mddev_t *mddev)
573 {
574 return mutex_is_locked(&mddev->reconfig_mutex);
575 }
576
577 static inline int mddev_trylock(mddev_t * mddev)
578 {
579 return mutex_trylock(&mddev->reconfig_mutex);
580 }
581
582 static struct attribute_group md_redundancy_group;
583
584 static void mddev_unlock(mddev_t * mddev)
585 {
586 if (mddev->to_remove) {
587 /* These cannot be removed under reconfig_mutex as
588 * an access to the files will try to take reconfig_mutex
589 * while holding the file unremovable, which leads to
590 * a deadlock.
591 * So hold set sysfs_active while the remove in happeing,
592 * and anything else which might set ->to_remove or my
593 * otherwise change the sysfs namespace will fail with
594 * -EBUSY if sysfs_active is still set.
595 * We set sysfs_active under reconfig_mutex and elsewhere
596 * test it under the same mutex to ensure its correct value
597 * is seen.
598 */
599 struct attribute_group *to_remove = mddev->to_remove;
600 mddev->to_remove = NULL;
601 mddev->sysfs_active = 1;
602 mutex_unlock(&mddev->reconfig_mutex);
603
604 if (mddev->kobj.sd) {
605 if (to_remove != &md_redundancy_group)
606 sysfs_remove_group(&mddev->kobj, to_remove);
607 if (mddev->pers == NULL ||
608 mddev->pers->sync_request == NULL) {
609 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
610 if (mddev->sysfs_action)
611 sysfs_put(mddev->sysfs_action);
612 mddev->sysfs_action = NULL;
613 }
614 }
615 mddev->sysfs_active = 0;
616 } else
617 mutex_unlock(&mddev->reconfig_mutex);
618
619 md_wakeup_thread(mddev->thread);
620 }
621
622 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
623 {
624 mdk_rdev_t *rdev;
625
626 list_for_each_entry(rdev, &mddev->disks, same_set)
627 if (rdev->desc_nr == nr)
628 return rdev;
629
630 return NULL;
631 }
632
633 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
634 {
635 mdk_rdev_t *rdev;
636
637 list_for_each_entry(rdev, &mddev->disks, same_set)
638 if (rdev->bdev->bd_dev == dev)
639 return rdev;
640
641 return NULL;
642 }
643
644 static struct mdk_personality *find_pers(int level, char *clevel)
645 {
646 struct mdk_personality *pers;
647 list_for_each_entry(pers, &pers_list, list) {
648 if (level != LEVEL_NONE && pers->level == level)
649 return pers;
650 if (strcmp(pers->name, clevel)==0)
651 return pers;
652 }
653 return NULL;
654 }
655
656 /* return the offset of the super block in 512byte sectors */
657 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
658 {
659 sector_t num_sectors = bdev->bd_inode->i_size / 512;
660 return MD_NEW_SIZE_SECTORS(num_sectors);
661 }
662
663 static int alloc_disk_sb(mdk_rdev_t * rdev)
664 {
665 if (rdev->sb_page)
666 MD_BUG();
667
668 rdev->sb_page = alloc_page(GFP_KERNEL);
669 if (!rdev->sb_page) {
670 printk(KERN_ALERT "md: out of memory.\n");
671 return -ENOMEM;
672 }
673
674 return 0;
675 }
676
677 static void free_disk_sb(mdk_rdev_t * rdev)
678 {
679 if (rdev->sb_page) {
680 put_page(rdev->sb_page);
681 rdev->sb_loaded = 0;
682 rdev->sb_page = NULL;
683 rdev->sb_start = 0;
684 rdev->sectors = 0;
685 }
686 }
687
688
689 static void super_written(struct bio *bio, int error)
690 {
691 mdk_rdev_t *rdev = bio->bi_private;
692 mddev_t *mddev = rdev->mddev;
693
694 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
695 printk("md: super_written gets error=%d, uptodate=%d\n",
696 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
697 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
698 md_error(mddev, rdev);
699 }
700
701 if (atomic_dec_and_test(&mddev->pending_writes))
702 wake_up(&mddev->sb_wait);
703 bio_put(bio);
704 }
705
706 static void super_written_barrier(struct bio *bio, int error)
707 {
708 struct bio *bio2 = bio->bi_private;
709 mdk_rdev_t *rdev = bio2->bi_private;
710 mddev_t *mddev = rdev->mddev;
711
712 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
713 error == -EOPNOTSUPP) {
714 unsigned long flags;
715 /* barriers don't appear to be supported :-( */
716 set_bit(BarriersNotsupp, &rdev->flags);
717 mddev->barriers_work = 0;
718 spin_lock_irqsave(&mddev->write_lock, flags);
719 bio2->bi_next = mddev->biolist;
720 mddev->biolist = bio2;
721 spin_unlock_irqrestore(&mddev->write_lock, flags);
722 wake_up(&mddev->sb_wait);
723 bio_put(bio);
724 } else {
725 bio_put(bio2);
726 bio->bi_private = rdev;
727 super_written(bio, error);
728 }
729 }
730
731 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
732 sector_t sector, int size, struct page *page)
733 {
734 /* write first size bytes of page to sector of rdev
735 * Increment mddev->pending_writes before returning
736 * and decrement it on completion, waking up sb_wait
737 * if zero is reached.
738 * If an error occurred, call md_error
739 *
740 * As we might need to resubmit the request if REQ_HARDBARRIER
741 * causes ENOTSUPP, we allocate a spare bio...
742 */
743 struct bio *bio = bio_alloc(GFP_NOIO, 1);
744 int rw = REQ_WRITE | REQ_SYNC | REQ_UNPLUG;
745
746 bio->bi_bdev = rdev->bdev;
747 bio->bi_sector = sector;
748 bio_add_page(bio, page, size, 0);
749 bio->bi_private = rdev;
750 bio->bi_end_io = super_written;
751 bio->bi_rw = rw;
752
753 atomic_inc(&mddev->pending_writes);
754 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
755 struct bio *rbio;
756 rw |= REQ_HARDBARRIER;
757 rbio = bio_clone(bio, GFP_NOIO);
758 rbio->bi_private = bio;
759 rbio->bi_end_io = super_written_barrier;
760 submit_bio(rw, rbio);
761 } else
762 submit_bio(rw, bio);
763 }
764
765 void md_super_wait(mddev_t *mddev)
766 {
767 /* wait for all superblock writes that were scheduled to complete.
768 * if any had to be retried (due to BARRIER problems), retry them
769 */
770 DEFINE_WAIT(wq);
771 for(;;) {
772 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
773 if (atomic_read(&mddev->pending_writes)==0)
774 break;
775 while (mddev->biolist) {
776 struct bio *bio;
777 spin_lock_irq(&mddev->write_lock);
778 bio = mddev->biolist;
779 mddev->biolist = bio->bi_next ;
780 bio->bi_next = NULL;
781 spin_unlock_irq(&mddev->write_lock);
782 submit_bio(bio->bi_rw, bio);
783 }
784 schedule();
785 }
786 finish_wait(&mddev->sb_wait, &wq);
787 }
788
789 static void bi_complete(struct bio *bio, int error)
790 {
791 complete((struct completion*)bio->bi_private);
792 }
793
794 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
795 struct page *page, int rw)
796 {
797 struct bio *bio = bio_alloc(GFP_NOIO, 1);
798 struct completion event;
799 int ret;
800
801 rw |= REQ_SYNC | REQ_UNPLUG;
802
803 bio->bi_bdev = bdev;
804 bio->bi_sector = sector;
805 bio_add_page(bio, page, size, 0);
806 init_completion(&event);
807 bio->bi_private = &event;
808 bio->bi_end_io = bi_complete;
809 submit_bio(rw, bio);
810 wait_for_completion(&event);
811
812 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
813 bio_put(bio);
814 return ret;
815 }
816 EXPORT_SYMBOL_GPL(sync_page_io);
817
818 static int read_disk_sb(mdk_rdev_t * rdev, int size)
819 {
820 char b[BDEVNAME_SIZE];
821 if (!rdev->sb_page) {
822 MD_BUG();
823 return -EINVAL;
824 }
825 if (rdev->sb_loaded)
826 return 0;
827
828
829 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
830 goto fail;
831 rdev->sb_loaded = 1;
832 return 0;
833
834 fail:
835 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
836 bdevname(rdev->bdev,b));
837 return -EINVAL;
838 }
839
840 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
841 {
842 return sb1->set_uuid0 == sb2->set_uuid0 &&
843 sb1->set_uuid1 == sb2->set_uuid1 &&
844 sb1->set_uuid2 == sb2->set_uuid2 &&
845 sb1->set_uuid3 == sb2->set_uuid3;
846 }
847
848 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
849 {
850 int ret;
851 mdp_super_t *tmp1, *tmp2;
852
853 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
854 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
855
856 if (!tmp1 || !tmp2) {
857 ret = 0;
858 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
859 goto abort;
860 }
861
862 *tmp1 = *sb1;
863 *tmp2 = *sb2;
864
865 /*
866 * nr_disks is not constant
867 */
868 tmp1->nr_disks = 0;
869 tmp2->nr_disks = 0;
870
871 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
872 abort:
873 kfree(tmp1);
874 kfree(tmp2);
875 return ret;
876 }
877
878
879 static u32 md_csum_fold(u32 csum)
880 {
881 csum = (csum & 0xffff) + (csum >> 16);
882 return (csum & 0xffff) + (csum >> 16);
883 }
884
885 static unsigned int calc_sb_csum(mdp_super_t * sb)
886 {
887 u64 newcsum = 0;
888 u32 *sb32 = (u32*)sb;
889 int i;
890 unsigned int disk_csum, csum;
891
892 disk_csum = sb->sb_csum;
893 sb->sb_csum = 0;
894
895 for (i = 0; i < MD_SB_BYTES/4 ; i++)
896 newcsum += sb32[i];
897 csum = (newcsum & 0xffffffff) + (newcsum>>32);
898
899
900 #ifdef CONFIG_ALPHA
901 /* This used to use csum_partial, which was wrong for several
902 * reasons including that different results are returned on
903 * different architectures. It isn't critical that we get exactly
904 * the same return value as before (we always csum_fold before
905 * testing, and that removes any differences). However as we
906 * know that csum_partial always returned a 16bit value on
907 * alphas, do a fold to maximise conformity to previous behaviour.
908 */
909 sb->sb_csum = md_csum_fold(disk_csum);
910 #else
911 sb->sb_csum = disk_csum;
912 #endif
913 return csum;
914 }
915
916
917 /*
918 * Handle superblock details.
919 * We want to be able to handle multiple superblock formats
920 * so we have a common interface to them all, and an array of
921 * different handlers.
922 * We rely on user-space to write the initial superblock, and support
923 * reading and updating of superblocks.
924 * Interface methods are:
925 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
926 * loads and validates a superblock on dev.
927 * if refdev != NULL, compare superblocks on both devices
928 * Return:
929 * 0 - dev has a superblock that is compatible with refdev
930 * 1 - dev has a superblock that is compatible and newer than refdev
931 * so dev should be used as the refdev in future
932 * -EINVAL superblock incompatible or invalid
933 * -othererror e.g. -EIO
934 *
935 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
936 * Verify that dev is acceptable into mddev.
937 * The first time, mddev->raid_disks will be 0, and data from
938 * dev should be merged in. Subsequent calls check that dev
939 * is new enough. Return 0 or -EINVAL
940 *
941 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
942 * Update the superblock for rdev with data in mddev
943 * This does not write to disc.
944 *
945 */
946
947 struct super_type {
948 char *name;
949 struct module *owner;
950 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
951 int minor_version);
952 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
953 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
954 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
955 sector_t num_sectors);
956 };
957
958 /*
959 * Check that the given mddev has no bitmap.
960 *
961 * This function is called from the run method of all personalities that do not
962 * support bitmaps. It prints an error message and returns non-zero if mddev
963 * has a bitmap. Otherwise, it returns 0.
964 *
965 */
966 int md_check_no_bitmap(mddev_t *mddev)
967 {
968 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
969 return 0;
970 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
971 mdname(mddev), mddev->pers->name);
972 return 1;
973 }
974 EXPORT_SYMBOL(md_check_no_bitmap);
975
976 /*
977 * load_super for 0.90.0
978 */
979 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
980 {
981 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
982 mdp_super_t *sb;
983 int ret;
984
985 /*
986 * Calculate the position of the superblock (512byte sectors),
987 * it's at the end of the disk.
988 *
989 * It also happens to be a multiple of 4Kb.
990 */
991 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
992
993 ret = read_disk_sb(rdev, MD_SB_BYTES);
994 if (ret) return ret;
995
996 ret = -EINVAL;
997
998 bdevname(rdev->bdev, b);
999 sb = (mdp_super_t*)page_address(rdev->sb_page);
1000
1001 if (sb->md_magic != MD_SB_MAGIC) {
1002 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1003 b);
1004 goto abort;
1005 }
1006
1007 if (sb->major_version != 0 ||
1008 sb->minor_version < 90 ||
1009 sb->minor_version > 91) {
1010 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1011 sb->major_version, sb->minor_version,
1012 b);
1013 goto abort;
1014 }
1015
1016 if (sb->raid_disks <= 0)
1017 goto abort;
1018
1019 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1020 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1021 b);
1022 goto abort;
1023 }
1024
1025 rdev->preferred_minor = sb->md_minor;
1026 rdev->data_offset = 0;
1027 rdev->sb_size = MD_SB_BYTES;
1028
1029 if (sb->level == LEVEL_MULTIPATH)
1030 rdev->desc_nr = -1;
1031 else
1032 rdev->desc_nr = sb->this_disk.number;
1033
1034 if (!refdev) {
1035 ret = 1;
1036 } else {
1037 __u64 ev1, ev2;
1038 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1039 if (!uuid_equal(refsb, sb)) {
1040 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1041 b, bdevname(refdev->bdev,b2));
1042 goto abort;
1043 }
1044 if (!sb_equal(refsb, sb)) {
1045 printk(KERN_WARNING "md: %s has same UUID"
1046 " but different superblock to %s\n",
1047 b, bdevname(refdev->bdev, b2));
1048 goto abort;
1049 }
1050 ev1 = md_event(sb);
1051 ev2 = md_event(refsb);
1052 if (ev1 > ev2)
1053 ret = 1;
1054 else
1055 ret = 0;
1056 }
1057 rdev->sectors = rdev->sb_start;
1058
1059 if (rdev->sectors < sb->size * 2 && sb->level > 1)
1060 /* "this cannot possibly happen" ... */
1061 ret = -EINVAL;
1062
1063 abort:
1064 return ret;
1065 }
1066
1067 /*
1068 * validate_super for 0.90.0
1069 */
1070 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1071 {
1072 mdp_disk_t *desc;
1073 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
1074 __u64 ev1 = md_event(sb);
1075
1076 rdev->raid_disk = -1;
1077 clear_bit(Faulty, &rdev->flags);
1078 clear_bit(In_sync, &rdev->flags);
1079 clear_bit(WriteMostly, &rdev->flags);
1080 clear_bit(BarriersNotsupp, &rdev->flags);
1081
1082 if (mddev->raid_disks == 0) {
1083 mddev->major_version = 0;
1084 mddev->minor_version = sb->minor_version;
1085 mddev->patch_version = sb->patch_version;
1086 mddev->external = 0;
1087 mddev->chunk_sectors = sb->chunk_size >> 9;
1088 mddev->ctime = sb->ctime;
1089 mddev->utime = sb->utime;
1090 mddev->level = sb->level;
1091 mddev->clevel[0] = 0;
1092 mddev->layout = sb->layout;
1093 mddev->raid_disks = sb->raid_disks;
1094 mddev->dev_sectors = sb->size * 2;
1095 mddev->events = ev1;
1096 mddev->bitmap_info.offset = 0;
1097 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1098
1099 if (mddev->minor_version >= 91) {
1100 mddev->reshape_position = sb->reshape_position;
1101 mddev->delta_disks = sb->delta_disks;
1102 mddev->new_level = sb->new_level;
1103 mddev->new_layout = sb->new_layout;
1104 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1105 } else {
1106 mddev->reshape_position = MaxSector;
1107 mddev->delta_disks = 0;
1108 mddev->new_level = mddev->level;
1109 mddev->new_layout = mddev->layout;
1110 mddev->new_chunk_sectors = mddev->chunk_sectors;
1111 }
1112
1113 if (sb->state & (1<<MD_SB_CLEAN))
1114 mddev->recovery_cp = MaxSector;
1115 else {
1116 if (sb->events_hi == sb->cp_events_hi &&
1117 sb->events_lo == sb->cp_events_lo) {
1118 mddev->recovery_cp = sb->recovery_cp;
1119 } else
1120 mddev->recovery_cp = 0;
1121 }
1122
1123 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1124 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1125 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1126 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1127
1128 mddev->max_disks = MD_SB_DISKS;
1129
1130 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1131 mddev->bitmap_info.file == NULL)
1132 mddev->bitmap_info.offset =
1133 mddev->bitmap_info.default_offset;
1134
1135 } else if (mddev->pers == NULL) {
1136 /* Insist on good event counter while assembling, except
1137 * for spares (which don't need an event count) */
1138 ++ev1;
1139 if (sb->disks[rdev->desc_nr].state & (
1140 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1141 if (ev1 < mddev->events)
1142 return -EINVAL;
1143 } else if (mddev->bitmap) {
1144 /* if adding to array with a bitmap, then we can accept an
1145 * older device ... but not too old.
1146 */
1147 if (ev1 < mddev->bitmap->events_cleared)
1148 return 0;
1149 } else {
1150 if (ev1 < mddev->events)
1151 /* just a hot-add of a new device, leave raid_disk at -1 */
1152 return 0;
1153 }
1154
1155 if (mddev->level != LEVEL_MULTIPATH) {
1156 desc = sb->disks + rdev->desc_nr;
1157
1158 if (desc->state & (1<<MD_DISK_FAULTY))
1159 set_bit(Faulty, &rdev->flags);
1160 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1161 desc->raid_disk < mddev->raid_disks */) {
1162 set_bit(In_sync, &rdev->flags);
1163 rdev->raid_disk = desc->raid_disk;
1164 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1165 /* active but not in sync implies recovery up to
1166 * reshape position. We don't know exactly where
1167 * that is, so set to zero for now */
1168 if (mddev->minor_version >= 91) {
1169 rdev->recovery_offset = 0;
1170 rdev->raid_disk = desc->raid_disk;
1171 }
1172 }
1173 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1174 set_bit(WriteMostly, &rdev->flags);
1175 } else /* MULTIPATH are always insync */
1176 set_bit(In_sync, &rdev->flags);
1177 return 0;
1178 }
1179
1180 /*
1181 * sync_super for 0.90.0
1182 */
1183 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1184 {
1185 mdp_super_t *sb;
1186 mdk_rdev_t *rdev2;
1187 int next_spare = mddev->raid_disks;
1188
1189
1190 /* make rdev->sb match mddev data..
1191 *
1192 * 1/ zero out disks
1193 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1194 * 3/ any empty disks < next_spare become removed
1195 *
1196 * disks[0] gets initialised to REMOVED because
1197 * we cannot be sure from other fields if it has
1198 * been initialised or not.
1199 */
1200 int i;
1201 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1202
1203 rdev->sb_size = MD_SB_BYTES;
1204
1205 sb = (mdp_super_t*)page_address(rdev->sb_page);
1206
1207 memset(sb, 0, sizeof(*sb));
1208
1209 sb->md_magic = MD_SB_MAGIC;
1210 sb->major_version = mddev->major_version;
1211 sb->patch_version = mddev->patch_version;
1212 sb->gvalid_words = 0; /* ignored */
1213 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1214 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1215 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1216 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1217
1218 sb->ctime = mddev->ctime;
1219 sb->level = mddev->level;
1220 sb->size = mddev->dev_sectors / 2;
1221 sb->raid_disks = mddev->raid_disks;
1222 sb->md_minor = mddev->md_minor;
1223 sb->not_persistent = 0;
1224 sb->utime = mddev->utime;
1225 sb->state = 0;
1226 sb->events_hi = (mddev->events>>32);
1227 sb->events_lo = (u32)mddev->events;
1228
1229 if (mddev->reshape_position == MaxSector)
1230 sb->minor_version = 90;
1231 else {
1232 sb->minor_version = 91;
1233 sb->reshape_position = mddev->reshape_position;
1234 sb->new_level = mddev->new_level;
1235 sb->delta_disks = mddev->delta_disks;
1236 sb->new_layout = mddev->new_layout;
1237 sb->new_chunk = mddev->new_chunk_sectors << 9;
1238 }
1239 mddev->minor_version = sb->minor_version;
1240 if (mddev->in_sync)
1241 {
1242 sb->recovery_cp = mddev->recovery_cp;
1243 sb->cp_events_hi = (mddev->events>>32);
1244 sb->cp_events_lo = (u32)mddev->events;
1245 if (mddev->recovery_cp == MaxSector)
1246 sb->state = (1<< MD_SB_CLEAN);
1247 } else
1248 sb->recovery_cp = 0;
1249
1250 sb->layout = mddev->layout;
1251 sb->chunk_size = mddev->chunk_sectors << 9;
1252
1253 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1254 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1255
1256 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1257 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1258 mdp_disk_t *d;
1259 int desc_nr;
1260 int is_active = test_bit(In_sync, &rdev2->flags);
1261
1262 if (rdev2->raid_disk >= 0 &&
1263 sb->minor_version >= 91)
1264 /* we have nowhere to store the recovery_offset,
1265 * but if it is not below the reshape_position,
1266 * we can piggy-back on that.
1267 */
1268 is_active = 1;
1269 if (rdev2->raid_disk < 0 ||
1270 test_bit(Faulty, &rdev2->flags))
1271 is_active = 0;
1272 if (is_active)
1273 desc_nr = rdev2->raid_disk;
1274 else
1275 desc_nr = next_spare++;
1276 rdev2->desc_nr = desc_nr;
1277 d = &sb->disks[rdev2->desc_nr];
1278 nr_disks++;
1279 d->number = rdev2->desc_nr;
1280 d->major = MAJOR(rdev2->bdev->bd_dev);
1281 d->minor = MINOR(rdev2->bdev->bd_dev);
1282 if (is_active)
1283 d->raid_disk = rdev2->raid_disk;
1284 else
1285 d->raid_disk = rdev2->desc_nr; /* compatibility */
1286 if (test_bit(Faulty, &rdev2->flags))
1287 d->state = (1<<MD_DISK_FAULTY);
1288 else if (is_active) {
1289 d->state = (1<<MD_DISK_ACTIVE);
1290 if (test_bit(In_sync, &rdev2->flags))
1291 d->state |= (1<<MD_DISK_SYNC);
1292 active++;
1293 working++;
1294 } else {
1295 d->state = 0;
1296 spare++;
1297 working++;
1298 }
1299 if (test_bit(WriteMostly, &rdev2->flags))
1300 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1301 }
1302 /* now set the "removed" and "faulty" bits on any missing devices */
1303 for (i=0 ; i < mddev->raid_disks ; i++) {
1304 mdp_disk_t *d = &sb->disks[i];
1305 if (d->state == 0 && d->number == 0) {
1306 d->number = i;
1307 d->raid_disk = i;
1308 d->state = (1<<MD_DISK_REMOVED);
1309 d->state |= (1<<MD_DISK_FAULTY);
1310 failed++;
1311 }
1312 }
1313 sb->nr_disks = nr_disks;
1314 sb->active_disks = active;
1315 sb->working_disks = working;
1316 sb->failed_disks = failed;
1317 sb->spare_disks = spare;
1318
1319 sb->this_disk = sb->disks[rdev->desc_nr];
1320 sb->sb_csum = calc_sb_csum(sb);
1321 }
1322
1323 /*
1324 * rdev_size_change for 0.90.0
1325 */
1326 static unsigned long long
1327 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1328 {
1329 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1330 return 0; /* component must fit device */
1331 if (rdev->mddev->bitmap_info.offset)
1332 return 0; /* can't move bitmap */
1333 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1334 if (!num_sectors || num_sectors > rdev->sb_start)
1335 num_sectors = rdev->sb_start;
1336 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1337 rdev->sb_page);
1338 md_super_wait(rdev->mddev);
1339 return num_sectors;
1340 }
1341
1342
1343 /*
1344 * version 1 superblock
1345 */
1346
1347 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1348 {
1349 __le32 disk_csum;
1350 u32 csum;
1351 unsigned long long newcsum;
1352 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1353 __le32 *isuper = (__le32*)sb;
1354 int i;
1355
1356 disk_csum = sb->sb_csum;
1357 sb->sb_csum = 0;
1358 newcsum = 0;
1359 for (i=0; size>=4; size -= 4 )
1360 newcsum += le32_to_cpu(*isuper++);
1361
1362 if (size == 2)
1363 newcsum += le16_to_cpu(*(__le16*) isuper);
1364
1365 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1366 sb->sb_csum = disk_csum;
1367 return cpu_to_le32(csum);
1368 }
1369
1370 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1371 {
1372 struct mdp_superblock_1 *sb;
1373 int ret;
1374 sector_t sb_start;
1375 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1376 int bmask;
1377
1378 /*
1379 * Calculate the position of the superblock in 512byte sectors.
1380 * It is always aligned to a 4K boundary and
1381 * depeding on minor_version, it can be:
1382 * 0: At least 8K, but less than 12K, from end of device
1383 * 1: At start of device
1384 * 2: 4K from start of device.
1385 */
1386 switch(minor_version) {
1387 case 0:
1388 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1389 sb_start -= 8*2;
1390 sb_start &= ~(sector_t)(4*2-1);
1391 break;
1392 case 1:
1393 sb_start = 0;
1394 break;
1395 case 2:
1396 sb_start = 8;
1397 break;
1398 default:
1399 return -EINVAL;
1400 }
1401 rdev->sb_start = sb_start;
1402
1403 /* superblock is rarely larger than 1K, but it can be larger,
1404 * and it is safe to read 4k, so we do that
1405 */
1406 ret = read_disk_sb(rdev, 4096);
1407 if (ret) return ret;
1408
1409
1410 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1411
1412 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1413 sb->major_version != cpu_to_le32(1) ||
1414 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1415 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1416 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417 return -EINVAL;
1418
1419 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1420 printk("md: invalid superblock checksum on %s\n",
1421 bdevname(rdev->bdev,b));
1422 return -EINVAL;
1423 }
1424 if (le64_to_cpu(sb->data_size) < 10) {
1425 printk("md: data_size too small on %s\n",
1426 bdevname(rdev->bdev,b));
1427 return -EINVAL;
1428 }
1429
1430 rdev->preferred_minor = 0xffff;
1431 rdev->data_offset = le64_to_cpu(sb->data_offset);
1432 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1433
1434 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1435 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1436 if (rdev->sb_size & bmask)
1437 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1438
1439 if (minor_version
1440 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1441 return -EINVAL;
1442
1443 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1444 rdev->desc_nr = -1;
1445 else
1446 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1447
1448 if (!refdev) {
1449 ret = 1;
1450 } else {
1451 __u64 ev1, ev2;
1452 struct mdp_superblock_1 *refsb =
1453 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1454
1455 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1456 sb->level != refsb->level ||
1457 sb->layout != refsb->layout ||
1458 sb->chunksize != refsb->chunksize) {
1459 printk(KERN_WARNING "md: %s has strangely different"
1460 " superblock to %s\n",
1461 bdevname(rdev->bdev,b),
1462 bdevname(refdev->bdev,b2));
1463 return -EINVAL;
1464 }
1465 ev1 = le64_to_cpu(sb->events);
1466 ev2 = le64_to_cpu(refsb->events);
1467
1468 if (ev1 > ev2)
1469 ret = 1;
1470 else
1471 ret = 0;
1472 }
1473 if (minor_version)
1474 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1475 le64_to_cpu(sb->data_offset);
1476 else
1477 rdev->sectors = rdev->sb_start;
1478 if (rdev->sectors < le64_to_cpu(sb->data_size))
1479 return -EINVAL;
1480 rdev->sectors = le64_to_cpu(sb->data_size);
1481 if (le64_to_cpu(sb->size) > rdev->sectors)
1482 return -EINVAL;
1483 return ret;
1484 }
1485
1486 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1487 {
1488 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1489 __u64 ev1 = le64_to_cpu(sb->events);
1490
1491 rdev->raid_disk = -1;
1492 clear_bit(Faulty, &rdev->flags);
1493 clear_bit(In_sync, &rdev->flags);
1494 clear_bit(WriteMostly, &rdev->flags);
1495 clear_bit(BarriersNotsupp, &rdev->flags);
1496
1497 if (mddev->raid_disks == 0) {
1498 mddev->major_version = 1;
1499 mddev->patch_version = 0;
1500 mddev->external = 0;
1501 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1502 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1503 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1504 mddev->level = le32_to_cpu(sb->level);
1505 mddev->clevel[0] = 0;
1506 mddev->layout = le32_to_cpu(sb->layout);
1507 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1508 mddev->dev_sectors = le64_to_cpu(sb->size);
1509 mddev->events = ev1;
1510 mddev->bitmap_info.offset = 0;
1511 mddev->bitmap_info.default_offset = 1024 >> 9;
1512
1513 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1514 memcpy(mddev->uuid, sb->set_uuid, 16);
1515
1516 mddev->max_disks = (4096-256)/2;
1517
1518 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1519 mddev->bitmap_info.file == NULL )
1520 mddev->bitmap_info.offset =
1521 (__s32)le32_to_cpu(sb->bitmap_offset);
1522
1523 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1524 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1525 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1526 mddev->new_level = le32_to_cpu(sb->new_level);
1527 mddev->new_layout = le32_to_cpu(sb->new_layout);
1528 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1529 } else {
1530 mddev->reshape_position = MaxSector;
1531 mddev->delta_disks = 0;
1532 mddev->new_level = mddev->level;
1533 mddev->new_layout = mddev->layout;
1534 mddev->new_chunk_sectors = mddev->chunk_sectors;
1535 }
1536
1537 } else if (mddev->pers == NULL) {
1538 /* Insist of good event counter while assembling, except for
1539 * spares (which don't need an event count) */
1540 ++ev1;
1541 if (rdev->desc_nr >= 0 &&
1542 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1543 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1544 if (ev1 < mddev->events)
1545 return -EINVAL;
1546 } else if (mddev->bitmap) {
1547 /* If adding to array with a bitmap, then we can accept an
1548 * older device, but not too old.
1549 */
1550 if (ev1 < mddev->bitmap->events_cleared)
1551 return 0;
1552 } else {
1553 if (ev1 < mddev->events)
1554 /* just a hot-add of a new device, leave raid_disk at -1 */
1555 return 0;
1556 }
1557 if (mddev->level != LEVEL_MULTIPATH) {
1558 int role;
1559 if (rdev->desc_nr < 0 ||
1560 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1561 role = 0xffff;
1562 rdev->desc_nr = -1;
1563 } else
1564 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1565 switch(role) {
1566 case 0xffff: /* spare */
1567 break;
1568 case 0xfffe: /* faulty */
1569 set_bit(Faulty, &rdev->flags);
1570 break;
1571 default:
1572 if ((le32_to_cpu(sb->feature_map) &
1573 MD_FEATURE_RECOVERY_OFFSET))
1574 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1575 else
1576 set_bit(In_sync, &rdev->flags);
1577 rdev->raid_disk = role;
1578 break;
1579 }
1580 if (sb->devflags & WriteMostly1)
1581 set_bit(WriteMostly, &rdev->flags);
1582 } else /* MULTIPATH are always insync */
1583 set_bit(In_sync, &rdev->flags);
1584
1585 return 0;
1586 }
1587
1588 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1589 {
1590 struct mdp_superblock_1 *sb;
1591 mdk_rdev_t *rdev2;
1592 int max_dev, i;
1593 /* make rdev->sb match mddev and rdev data. */
1594
1595 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1596
1597 sb->feature_map = 0;
1598 sb->pad0 = 0;
1599 sb->recovery_offset = cpu_to_le64(0);
1600 memset(sb->pad1, 0, sizeof(sb->pad1));
1601 memset(sb->pad2, 0, sizeof(sb->pad2));
1602 memset(sb->pad3, 0, sizeof(sb->pad3));
1603
1604 sb->utime = cpu_to_le64((__u64)mddev->utime);
1605 sb->events = cpu_to_le64(mddev->events);
1606 if (mddev->in_sync)
1607 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1608 else
1609 sb->resync_offset = cpu_to_le64(0);
1610
1611 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1612
1613 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1614 sb->size = cpu_to_le64(mddev->dev_sectors);
1615 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1616 sb->level = cpu_to_le32(mddev->level);
1617 sb->layout = cpu_to_le32(mddev->layout);
1618
1619 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1620 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1621 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1622 }
1623
1624 if (rdev->raid_disk >= 0 &&
1625 !test_bit(In_sync, &rdev->flags)) {
1626 sb->feature_map |=
1627 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1628 sb->recovery_offset =
1629 cpu_to_le64(rdev->recovery_offset);
1630 }
1631
1632 if (mddev->reshape_position != MaxSector) {
1633 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1634 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1635 sb->new_layout = cpu_to_le32(mddev->new_layout);
1636 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1637 sb->new_level = cpu_to_le32(mddev->new_level);
1638 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1639 }
1640
1641 max_dev = 0;
1642 list_for_each_entry(rdev2, &mddev->disks, same_set)
1643 if (rdev2->desc_nr+1 > max_dev)
1644 max_dev = rdev2->desc_nr+1;
1645
1646 if (max_dev > le32_to_cpu(sb->max_dev)) {
1647 int bmask;
1648 sb->max_dev = cpu_to_le32(max_dev);
1649 rdev->sb_size = max_dev * 2 + 256;
1650 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1651 if (rdev->sb_size & bmask)
1652 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1653 } else
1654 max_dev = le32_to_cpu(sb->max_dev);
1655
1656 for (i=0; i<max_dev;i++)
1657 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1658
1659 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1660 i = rdev2->desc_nr;
1661 if (test_bit(Faulty, &rdev2->flags))
1662 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1663 else if (test_bit(In_sync, &rdev2->flags))
1664 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1665 else if (rdev2->raid_disk >= 0)
1666 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1667 else
1668 sb->dev_roles[i] = cpu_to_le16(0xffff);
1669 }
1670
1671 sb->sb_csum = calc_sb_1_csum(sb);
1672 }
1673
1674 static unsigned long long
1675 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1676 {
1677 struct mdp_superblock_1 *sb;
1678 sector_t max_sectors;
1679 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1680 return 0; /* component must fit device */
1681 if (rdev->sb_start < rdev->data_offset) {
1682 /* minor versions 1 and 2; superblock before data */
1683 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1684 max_sectors -= rdev->data_offset;
1685 if (!num_sectors || num_sectors > max_sectors)
1686 num_sectors = max_sectors;
1687 } else if (rdev->mddev->bitmap_info.offset) {
1688 /* minor version 0 with bitmap we can't move */
1689 return 0;
1690 } else {
1691 /* minor version 0; superblock after data */
1692 sector_t sb_start;
1693 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1694 sb_start &= ~(sector_t)(4*2 - 1);
1695 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1696 if (!num_sectors || num_sectors > max_sectors)
1697 num_sectors = max_sectors;
1698 rdev->sb_start = sb_start;
1699 }
1700 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1701 sb->data_size = cpu_to_le64(num_sectors);
1702 sb->super_offset = rdev->sb_start;
1703 sb->sb_csum = calc_sb_1_csum(sb);
1704 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1705 rdev->sb_page);
1706 md_super_wait(rdev->mddev);
1707 return num_sectors;
1708 }
1709
1710 static struct super_type super_types[] = {
1711 [0] = {
1712 .name = "0.90.0",
1713 .owner = THIS_MODULE,
1714 .load_super = super_90_load,
1715 .validate_super = super_90_validate,
1716 .sync_super = super_90_sync,
1717 .rdev_size_change = super_90_rdev_size_change,
1718 },
1719 [1] = {
1720 .name = "md-1",
1721 .owner = THIS_MODULE,
1722 .load_super = super_1_load,
1723 .validate_super = super_1_validate,
1724 .sync_super = super_1_sync,
1725 .rdev_size_change = super_1_rdev_size_change,
1726 },
1727 };
1728
1729 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1730 {
1731 mdk_rdev_t *rdev, *rdev2;
1732
1733 rcu_read_lock();
1734 rdev_for_each_rcu(rdev, mddev1)
1735 rdev_for_each_rcu(rdev2, mddev2)
1736 if (rdev->bdev->bd_contains ==
1737 rdev2->bdev->bd_contains) {
1738 rcu_read_unlock();
1739 return 1;
1740 }
1741 rcu_read_unlock();
1742 return 0;
1743 }
1744
1745 static LIST_HEAD(pending_raid_disks);
1746
1747 /*
1748 * Try to register data integrity profile for an mddev
1749 *
1750 * This is called when an array is started and after a disk has been kicked
1751 * from the array. It only succeeds if all working and active component devices
1752 * are integrity capable with matching profiles.
1753 */
1754 int md_integrity_register(mddev_t *mddev)
1755 {
1756 mdk_rdev_t *rdev, *reference = NULL;
1757
1758 if (list_empty(&mddev->disks))
1759 return 0; /* nothing to do */
1760 if (blk_get_integrity(mddev->gendisk))
1761 return 0; /* already registered */
1762 list_for_each_entry(rdev, &mddev->disks, same_set) {
1763 /* skip spares and non-functional disks */
1764 if (test_bit(Faulty, &rdev->flags))
1765 continue;
1766 if (rdev->raid_disk < 0)
1767 continue;
1768 /*
1769 * If at least one rdev is not integrity capable, we can not
1770 * enable data integrity for the md device.
1771 */
1772 if (!bdev_get_integrity(rdev->bdev))
1773 return -EINVAL;
1774 if (!reference) {
1775 /* Use the first rdev as the reference */
1776 reference = rdev;
1777 continue;
1778 }
1779 /* does this rdev's profile match the reference profile? */
1780 if (blk_integrity_compare(reference->bdev->bd_disk,
1781 rdev->bdev->bd_disk) < 0)
1782 return -EINVAL;
1783 }
1784 /*
1785 * All component devices are integrity capable and have matching
1786 * profiles, register the common profile for the md device.
1787 */
1788 if (blk_integrity_register(mddev->gendisk,
1789 bdev_get_integrity(reference->bdev)) != 0) {
1790 printk(KERN_ERR "md: failed to register integrity for %s\n",
1791 mdname(mddev));
1792 return -EINVAL;
1793 }
1794 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1795 mdname(mddev));
1796 return 0;
1797 }
1798 EXPORT_SYMBOL(md_integrity_register);
1799
1800 /* Disable data integrity if non-capable/non-matching disk is being added */
1801 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1802 {
1803 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1804 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1805
1806 if (!bi_mddev) /* nothing to do */
1807 return;
1808 if (rdev->raid_disk < 0) /* skip spares */
1809 return;
1810 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1811 rdev->bdev->bd_disk) >= 0)
1812 return;
1813 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1814 blk_integrity_unregister(mddev->gendisk);
1815 }
1816 EXPORT_SYMBOL(md_integrity_add_rdev);
1817
1818 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1819 {
1820 char b[BDEVNAME_SIZE];
1821 struct kobject *ko;
1822 char *s;
1823 int err;
1824
1825 if (rdev->mddev) {
1826 MD_BUG();
1827 return -EINVAL;
1828 }
1829
1830 /* prevent duplicates */
1831 if (find_rdev(mddev, rdev->bdev->bd_dev))
1832 return -EEXIST;
1833
1834 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1835 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1836 rdev->sectors < mddev->dev_sectors)) {
1837 if (mddev->pers) {
1838 /* Cannot change size, so fail
1839 * If mddev->level <= 0, then we don't care
1840 * about aligning sizes (e.g. linear)
1841 */
1842 if (mddev->level > 0)
1843 return -ENOSPC;
1844 } else
1845 mddev->dev_sectors = rdev->sectors;
1846 }
1847
1848 /* Verify rdev->desc_nr is unique.
1849 * If it is -1, assign a free number, else
1850 * check number is not in use
1851 */
1852 if (rdev->desc_nr < 0) {
1853 int choice = 0;
1854 if (mddev->pers) choice = mddev->raid_disks;
1855 while (find_rdev_nr(mddev, choice))
1856 choice++;
1857 rdev->desc_nr = choice;
1858 } else {
1859 if (find_rdev_nr(mddev, rdev->desc_nr))
1860 return -EBUSY;
1861 }
1862 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1863 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1864 mdname(mddev), mddev->max_disks);
1865 return -EBUSY;
1866 }
1867 bdevname(rdev->bdev,b);
1868 while ( (s=strchr(b, '/')) != NULL)
1869 *s = '!';
1870
1871 rdev->mddev = mddev;
1872 printk(KERN_INFO "md: bind<%s>\n", b);
1873
1874 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1875 goto fail;
1876
1877 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1878 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1879 /* failure here is OK */;
1880 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
1881
1882 list_add_rcu(&rdev->same_set, &mddev->disks);
1883 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1884
1885 /* May as well allow recovery to be retried once */
1886 mddev->recovery_disabled = 0;
1887
1888 return 0;
1889
1890 fail:
1891 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1892 b, mdname(mddev));
1893 return err;
1894 }
1895
1896 static void md_delayed_delete(struct work_struct *ws)
1897 {
1898 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1899 kobject_del(&rdev->kobj);
1900 kobject_put(&rdev->kobj);
1901 }
1902
1903 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1904 {
1905 char b[BDEVNAME_SIZE];
1906 if (!rdev->mddev) {
1907 MD_BUG();
1908 return;
1909 }
1910 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1911 list_del_rcu(&rdev->same_set);
1912 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1913 rdev->mddev = NULL;
1914 sysfs_remove_link(&rdev->kobj, "block");
1915 sysfs_put(rdev->sysfs_state);
1916 rdev->sysfs_state = NULL;
1917 /* We need to delay this, otherwise we can deadlock when
1918 * writing to 'remove' to "dev/state". We also need
1919 * to delay it due to rcu usage.
1920 */
1921 synchronize_rcu();
1922 INIT_WORK(&rdev->del_work, md_delayed_delete);
1923 kobject_get(&rdev->kobj);
1924 schedule_work(&rdev->del_work);
1925 }
1926
1927 /*
1928 * prevent the device from being mounted, repartitioned or
1929 * otherwise reused by a RAID array (or any other kernel
1930 * subsystem), by bd_claiming the device.
1931 */
1932 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1933 {
1934 int err = 0;
1935 struct block_device *bdev;
1936 char b[BDEVNAME_SIZE];
1937
1938 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1939 if (IS_ERR(bdev)) {
1940 printk(KERN_ERR "md: could not open %s.\n",
1941 __bdevname(dev, b));
1942 return PTR_ERR(bdev);
1943 }
1944 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1945 if (err) {
1946 printk(KERN_ERR "md: could not bd_claim %s.\n",
1947 bdevname(bdev, b));
1948 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1949 return err;
1950 }
1951 if (!shared)
1952 set_bit(AllReserved, &rdev->flags);
1953 rdev->bdev = bdev;
1954 return err;
1955 }
1956
1957 static void unlock_rdev(mdk_rdev_t *rdev)
1958 {
1959 struct block_device *bdev = rdev->bdev;
1960 rdev->bdev = NULL;
1961 if (!bdev)
1962 MD_BUG();
1963 bd_release(bdev);
1964 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1965 }
1966
1967 void md_autodetect_dev(dev_t dev);
1968
1969 static void export_rdev(mdk_rdev_t * rdev)
1970 {
1971 char b[BDEVNAME_SIZE];
1972 printk(KERN_INFO "md: export_rdev(%s)\n",
1973 bdevname(rdev->bdev,b));
1974 if (rdev->mddev)
1975 MD_BUG();
1976 free_disk_sb(rdev);
1977 #ifndef MODULE
1978 if (test_bit(AutoDetected, &rdev->flags))
1979 md_autodetect_dev(rdev->bdev->bd_dev);
1980 #endif
1981 unlock_rdev(rdev);
1982 kobject_put(&rdev->kobj);
1983 }
1984
1985 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1986 {
1987 unbind_rdev_from_array(rdev);
1988 export_rdev(rdev);
1989 }
1990
1991 static void export_array(mddev_t *mddev)
1992 {
1993 mdk_rdev_t *rdev, *tmp;
1994
1995 rdev_for_each(rdev, tmp, mddev) {
1996 if (!rdev->mddev) {
1997 MD_BUG();
1998 continue;
1999 }
2000 kick_rdev_from_array(rdev);
2001 }
2002 if (!list_empty(&mddev->disks))
2003 MD_BUG();
2004 mddev->raid_disks = 0;
2005 mddev->major_version = 0;
2006 }
2007
2008 static void print_desc(mdp_disk_t *desc)
2009 {
2010 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2011 desc->major,desc->minor,desc->raid_disk,desc->state);
2012 }
2013
2014 static void print_sb_90(mdp_super_t *sb)
2015 {
2016 int i;
2017
2018 printk(KERN_INFO
2019 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2020 sb->major_version, sb->minor_version, sb->patch_version,
2021 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2022 sb->ctime);
2023 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2024 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2025 sb->md_minor, sb->layout, sb->chunk_size);
2026 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2027 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2028 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2029 sb->failed_disks, sb->spare_disks,
2030 sb->sb_csum, (unsigned long)sb->events_lo);
2031
2032 printk(KERN_INFO);
2033 for (i = 0; i < MD_SB_DISKS; i++) {
2034 mdp_disk_t *desc;
2035
2036 desc = sb->disks + i;
2037 if (desc->number || desc->major || desc->minor ||
2038 desc->raid_disk || (desc->state && (desc->state != 4))) {
2039 printk(" D %2d: ", i);
2040 print_desc(desc);
2041 }
2042 }
2043 printk(KERN_INFO "md: THIS: ");
2044 print_desc(&sb->this_disk);
2045 }
2046
2047 static void print_sb_1(struct mdp_superblock_1 *sb)
2048 {
2049 __u8 *uuid;
2050
2051 uuid = sb->set_uuid;
2052 printk(KERN_INFO
2053 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2054 "md: Name: \"%s\" CT:%llu\n",
2055 le32_to_cpu(sb->major_version),
2056 le32_to_cpu(sb->feature_map),
2057 uuid,
2058 sb->set_name,
2059 (unsigned long long)le64_to_cpu(sb->ctime)
2060 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2061
2062 uuid = sb->device_uuid;
2063 printk(KERN_INFO
2064 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2065 " RO:%llu\n"
2066 "md: Dev:%08x UUID: %pU\n"
2067 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2068 "md: (MaxDev:%u) \n",
2069 le32_to_cpu(sb->level),
2070 (unsigned long long)le64_to_cpu(sb->size),
2071 le32_to_cpu(sb->raid_disks),
2072 le32_to_cpu(sb->layout),
2073 le32_to_cpu(sb->chunksize),
2074 (unsigned long long)le64_to_cpu(sb->data_offset),
2075 (unsigned long long)le64_to_cpu(sb->data_size),
2076 (unsigned long long)le64_to_cpu(sb->super_offset),
2077 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2078 le32_to_cpu(sb->dev_number),
2079 uuid,
2080 sb->devflags,
2081 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2082 (unsigned long long)le64_to_cpu(sb->events),
2083 (unsigned long long)le64_to_cpu(sb->resync_offset),
2084 le32_to_cpu(sb->sb_csum),
2085 le32_to_cpu(sb->max_dev)
2086 );
2087 }
2088
2089 static void print_rdev(mdk_rdev_t *rdev, int major_version)
2090 {
2091 char b[BDEVNAME_SIZE];
2092 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2093 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2094 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2095 rdev->desc_nr);
2096 if (rdev->sb_loaded) {
2097 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2098 switch (major_version) {
2099 case 0:
2100 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2101 break;
2102 case 1:
2103 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2104 break;
2105 }
2106 } else
2107 printk(KERN_INFO "md: no rdev superblock!\n");
2108 }
2109
2110 static void md_print_devices(void)
2111 {
2112 struct list_head *tmp;
2113 mdk_rdev_t *rdev;
2114 mddev_t *mddev;
2115 char b[BDEVNAME_SIZE];
2116
2117 printk("\n");
2118 printk("md: **********************************\n");
2119 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2120 printk("md: **********************************\n");
2121 for_each_mddev(mddev, tmp) {
2122
2123 if (mddev->bitmap)
2124 bitmap_print_sb(mddev->bitmap);
2125 else
2126 printk("%s: ", mdname(mddev));
2127 list_for_each_entry(rdev, &mddev->disks, same_set)
2128 printk("<%s>", bdevname(rdev->bdev,b));
2129 printk("\n");
2130
2131 list_for_each_entry(rdev, &mddev->disks, same_set)
2132 print_rdev(rdev, mddev->major_version);
2133 }
2134 printk("md: **********************************\n");
2135 printk("\n");
2136 }
2137
2138
2139 static void sync_sbs(mddev_t * mddev, int nospares)
2140 {
2141 /* Update each superblock (in-memory image), but
2142 * if we are allowed to, skip spares which already
2143 * have the right event counter, or have one earlier
2144 * (which would mean they aren't being marked as dirty
2145 * with the rest of the array)
2146 */
2147 mdk_rdev_t *rdev;
2148 list_for_each_entry(rdev, &mddev->disks, same_set) {
2149 if (rdev->sb_events == mddev->events ||
2150 (nospares &&
2151 rdev->raid_disk < 0 &&
2152 rdev->sb_events+1 == mddev->events)) {
2153 /* Don't update this superblock */
2154 rdev->sb_loaded = 2;
2155 } else {
2156 super_types[mddev->major_version].
2157 sync_super(mddev, rdev);
2158 rdev->sb_loaded = 1;
2159 }
2160 }
2161 }
2162
2163 static void md_update_sb(mddev_t * mddev, int force_change)
2164 {
2165 mdk_rdev_t *rdev;
2166 int sync_req;
2167 int nospares = 0;
2168
2169 repeat:
2170 /* First make sure individual recovery_offsets are correct */
2171 list_for_each_entry(rdev, &mddev->disks, same_set) {
2172 if (rdev->raid_disk >= 0 &&
2173 mddev->delta_disks >= 0 &&
2174 !test_bit(In_sync, &rdev->flags) &&
2175 mddev->curr_resync_completed > rdev->recovery_offset)
2176 rdev->recovery_offset = mddev->curr_resync_completed;
2177
2178 }
2179 if (!mddev->persistent) {
2180 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2181 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2182 if (!mddev->external)
2183 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2184 wake_up(&mddev->sb_wait);
2185 return;
2186 }
2187
2188 spin_lock_irq(&mddev->write_lock);
2189
2190 mddev->utime = get_seconds();
2191
2192 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2193 force_change = 1;
2194 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2195 /* just a clean<-> dirty transition, possibly leave spares alone,
2196 * though if events isn't the right even/odd, we will have to do
2197 * spares after all
2198 */
2199 nospares = 1;
2200 if (force_change)
2201 nospares = 0;
2202 if (mddev->degraded)
2203 /* If the array is degraded, then skipping spares is both
2204 * dangerous and fairly pointless.
2205 * Dangerous because a device that was removed from the array
2206 * might have a event_count that still looks up-to-date,
2207 * so it can be re-added without a resync.
2208 * Pointless because if there are any spares to skip,
2209 * then a recovery will happen and soon that array won't
2210 * be degraded any more and the spare can go back to sleep then.
2211 */
2212 nospares = 0;
2213
2214 sync_req = mddev->in_sync;
2215
2216 /* If this is just a dirty<->clean transition, and the array is clean
2217 * and 'events' is odd, we can roll back to the previous clean state */
2218 if (nospares
2219 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2220 && mddev->can_decrease_events
2221 && mddev->events != 1) {
2222 mddev->events--;
2223 mddev->can_decrease_events = 0;
2224 } else {
2225 /* otherwise we have to go forward and ... */
2226 mddev->events ++;
2227 mddev->can_decrease_events = nospares;
2228 }
2229
2230 if (!mddev->events) {
2231 /*
2232 * oops, this 64-bit counter should never wrap.
2233 * Either we are in around ~1 trillion A.C., assuming
2234 * 1 reboot per second, or we have a bug:
2235 */
2236 MD_BUG();
2237 mddev->events --;
2238 }
2239 sync_sbs(mddev, nospares);
2240 spin_unlock_irq(&mddev->write_lock);
2241
2242 dprintk(KERN_INFO
2243 "md: updating %s RAID superblock on device (in sync %d)\n",
2244 mdname(mddev),mddev->in_sync);
2245
2246 bitmap_update_sb(mddev->bitmap);
2247 list_for_each_entry(rdev, &mddev->disks, same_set) {
2248 char b[BDEVNAME_SIZE];
2249 dprintk(KERN_INFO "md: ");
2250 if (rdev->sb_loaded != 1)
2251 continue; /* no noise on spare devices */
2252 if (test_bit(Faulty, &rdev->flags))
2253 dprintk("(skipping faulty ");
2254
2255 dprintk("%s ", bdevname(rdev->bdev,b));
2256 if (!test_bit(Faulty, &rdev->flags)) {
2257 md_super_write(mddev,rdev,
2258 rdev->sb_start, rdev->sb_size,
2259 rdev->sb_page);
2260 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2261 bdevname(rdev->bdev,b),
2262 (unsigned long long)rdev->sb_start);
2263 rdev->sb_events = mddev->events;
2264
2265 } else
2266 dprintk(")\n");
2267 if (mddev->level == LEVEL_MULTIPATH)
2268 /* only need to write one superblock... */
2269 break;
2270 }
2271 md_super_wait(mddev);
2272 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2273
2274 spin_lock_irq(&mddev->write_lock);
2275 if (mddev->in_sync != sync_req ||
2276 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2277 /* have to write it out again */
2278 spin_unlock_irq(&mddev->write_lock);
2279 goto repeat;
2280 }
2281 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2282 spin_unlock_irq(&mddev->write_lock);
2283 wake_up(&mddev->sb_wait);
2284 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2285 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2286
2287 }
2288
2289 /* words written to sysfs files may, or may not, be \n terminated.
2290 * We want to accept with case. For this we use cmd_match.
2291 */
2292 static int cmd_match(const char *cmd, const char *str)
2293 {
2294 /* See if cmd, written into a sysfs file, matches
2295 * str. They must either be the same, or cmd can
2296 * have a trailing newline
2297 */
2298 while (*cmd && *str && *cmd == *str) {
2299 cmd++;
2300 str++;
2301 }
2302 if (*cmd == '\n')
2303 cmd++;
2304 if (*str || *cmd)
2305 return 0;
2306 return 1;
2307 }
2308
2309 struct rdev_sysfs_entry {
2310 struct attribute attr;
2311 ssize_t (*show)(mdk_rdev_t *, char *);
2312 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2313 };
2314
2315 static ssize_t
2316 state_show(mdk_rdev_t *rdev, char *page)
2317 {
2318 char *sep = "";
2319 size_t len = 0;
2320
2321 if (test_bit(Faulty, &rdev->flags)) {
2322 len+= sprintf(page+len, "%sfaulty",sep);
2323 sep = ",";
2324 }
2325 if (test_bit(In_sync, &rdev->flags)) {
2326 len += sprintf(page+len, "%sin_sync",sep);
2327 sep = ",";
2328 }
2329 if (test_bit(WriteMostly, &rdev->flags)) {
2330 len += sprintf(page+len, "%swrite_mostly",sep);
2331 sep = ",";
2332 }
2333 if (test_bit(Blocked, &rdev->flags)) {
2334 len += sprintf(page+len, "%sblocked", sep);
2335 sep = ",";
2336 }
2337 if (!test_bit(Faulty, &rdev->flags) &&
2338 !test_bit(In_sync, &rdev->flags)) {
2339 len += sprintf(page+len, "%sspare", sep);
2340 sep = ",";
2341 }
2342 return len+sprintf(page+len, "\n");
2343 }
2344
2345 static ssize_t
2346 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2347 {
2348 /* can write
2349 * faulty - simulates and error
2350 * remove - disconnects the device
2351 * writemostly - sets write_mostly
2352 * -writemostly - clears write_mostly
2353 * blocked - sets the Blocked flag
2354 * -blocked - clears the Blocked flag
2355 * insync - sets Insync providing device isn't active
2356 */
2357 int err = -EINVAL;
2358 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2359 md_error(rdev->mddev, rdev);
2360 err = 0;
2361 } else if (cmd_match(buf, "remove")) {
2362 if (rdev->raid_disk >= 0)
2363 err = -EBUSY;
2364 else {
2365 mddev_t *mddev = rdev->mddev;
2366 kick_rdev_from_array(rdev);
2367 if (mddev->pers)
2368 md_update_sb(mddev, 1);
2369 md_new_event(mddev);
2370 err = 0;
2371 }
2372 } else if (cmd_match(buf, "writemostly")) {
2373 set_bit(WriteMostly, &rdev->flags);
2374 err = 0;
2375 } else if (cmd_match(buf, "-writemostly")) {
2376 clear_bit(WriteMostly, &rdev->flags);
2377 err = 0;
2378 } else if (cmd_match(buf, "blocked")) {
2379 set_bit(Blocked, &rdev->flags);
2380 err = 0;
2381 } else if (cmd_match(buf, "-blocked")) {
2382 clear_bit(Blocked, &rdev->flags);
2383 wake_up(&rdev->blocked_wait);
2384 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2385 md_wakeup_thread(rdev->mddev->thread);
2386
2387 err = 0;
2388 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2389 set_bit(In_sync, &rdev->flags);
2390 err = 0;
2391 }
2392 if (!err)
2393 sysfs_notify_dirent_safe(rdev->sysfs_state);
2394 return err ? err : len;
2395 }
2396 static struct rdev_sysfs_entry rdev_state =
2397 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2398
2399 static ssize_t
2400 errors_show(mdk_rdev_t *rdev, char *page)
2401 {
2402 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2403 }
2404
2405 static ssize_t
2406 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2407 {
2408 char *e;
2409 unsigned long n = simple_strtoul(buf, &e, 10);
2410 if (*buf && (*e == 0 || *e == '\n')) {
2411 atomic_set(&rdev->corrected_errors, n);
2412 return len;
2413 }
2414 return -EINVAL;
2415 }
2416 static struct rdev_sysfs_entry rdev_errors =
2417 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2418
2419 static ssize_t
2420 slot_show(mdk_rdev_t *rdev, char *page)
2421 {
2422 if (rdev->raid_disk < 0)
2423 return sprintf(page, "none\n");
2424 else
2425 return sprintf(page, "%d\n", rdev->raid_disk);
2426 }
2427
2428 static ssize_t
2429 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2430 {
2431 char *e;
2432 int err;
2433 char nm[20];
2434 int slot = simple_strtoul(buf, &e, 10);
2435 if (strncmp(buf, "none", 4)==0)
2436 slot = -1;
2437 else if (e==buf || (*e && *e!= '\n'))
2438 return -EINVAL;
2439 if (rdev->mddev->pers && slot == -1) {
2440 /* Setting 'slot' on an active array requires also
2441 * updating the 'rd%d' link, and communicating
2442 * with the personality with ->hot_*_disk.
2443 * For now we only support removing
2444 * failed/spare devices. This normally happens automatically,
2445 * but not when the metadata is externally managed.
2446 */
2447 if (rdev->raid_disk == -1)
2448 return -EEXIST;
2449 /* personality does all needed checks */
2450 if (rdev->mddev->pers->hot_add_disk == NULL)
2451 return -EINVAL;
2452 err = rdev->mddev->pers->
2453 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2454 if (err)
2455 return err;
2456 sprintf(nm, "rd%d", rdev->raid_disk);
2457 sysfs_remove_link(&rdev->mddev->kobj, nm);
2458 rdev->raid_disk = -1;
2459 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2460 md_wakeup_thread(rdev->mddev->thread);
2461 } else if (rdev->mddev->pers) {
2462 mdk_rdev_t *rdev2;
2463 /* Activating a spare .. or possibly reactivating
2464 * if we ever get bitmaps working here.
2465 */
2466
2467 if (rdev->raid_disk != -1)
2468 return -EBUSY;
2469
2470 if (rdev->mddev->pers->hot_add_disk == NULL)
2471 return -EINVAL;
2472
2473 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2474 if (rdev2->raid_disk == slot)
2475 return -EEXIST;
2476
2477 rdev->raid_disk = slot;
2478 if (test_bit(In_sync, &rdev->flags))
2479 rdev->saved_raid_disk = slot;
2480 else
2481 rdev->saved_raid_disk = -1;
2482 err = rdev->mddev->pers->
2483 hot_add_disk(rdev->mddev, rdev);
2484 if (err) {
2485 rdev->raid_disk = -1;
2486 return err;
2487 } else
2488 sysfs_notify_dirent_safe(rdev->sysfs_state);
2489 sprintf(nm, "rd%d", rdev->raid_disk);
2490 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2491 /* failure here is OK */;
2492 /* don't wakeup anyone, leave that to userspace. */
2493 } else {
2494 if (slot >= rdev->mddev->raid_disks)
2495 return -ENOSPC;
2496 rdev->raid_disk = slot;
2497 /* assume it is working */
2498 clear_bit(Faulty, &rdev->flags);
2499 clear_bit(WriteMostly, &rdev->flags);
2500 set_bit(In_sync, &rdev->flags);
2501 sysfs_notify_dirent_safe(rdev->sysfs_state);
2502 }
2503 return len;
2504 }
2505
2506
2507 static struct rdev_sysfs_entry rdev_slot =
2508 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2509
2510 static ssize_t
2511 offset_show(mdk_rdev_t *rdev, char *page)
2512 {
2513 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2514 }
2515
2516 static ssize_t
2517 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2518 {
2519 char *e;
2520 unsigned long long offset = simple_strtoull(buf, &e, 10);
2521 if (e==buf || (*e && *e != '\n'))
2522 return -EINVAL;
2523 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2524 return -EBUSY;
2525 if (rdev->sectors && rdev->mddev->external)
2526 /* Must set offset before size, so overlap checks
2527 * can be sane */
2528 return -EBUSY;
2529 rdev->data_offset = offset;
2530 return len;
2531 }
2532
2533 static struct rdev_sysfs_entry rdev_offset =
2534 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2535
2536 static ssize_t
2537 rdev_size_show(mdk_rdev_t *rdev, char *page)
2538 {
2539 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2540 }
2541
2542 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2543 {
2544 /* check if two start/length pairs overlap */
2545 if (s1+l1 <= s2)
2546 return 0;
2547 if (s2+l2 <= s1)
2548 return 0;
2549 return 1;
2550 }
2551
2552 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2553 {
2554 unsigned long long blocks;
2555 sector_t new;
2556
2557 if (strict_strtoull(buf, 10, &blocks) < 0)
2558 return -EINVAL;
2559
2560 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2561 return -EINVAL; /* sector conversion overflow */
2562
2563 new = blocks * 2;
2564 if (new != blocks * 2)
2565 return -EINVAL; /* unsigned long long to sector_t overflow */
2566
2567 *sectors = new;
2568 return 0;
2569 }
2570
2571 static ssize_t
2572 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2573 {
2574 mddev_t *my_mddev = rdev->mddev;
2575 sector_t oldsectors = rdev->sectors;
2576 sector_t sectors;
2577
2578 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2579 return -EINVAL;
2580 if (my_mddev->pers && rdev->raid_disk >= 0) {
2581 if (my_mddev->persistent) {
2582 sectors = super_types[my_mddev->major_version].
2583 rdev_size_change(rdev, sectors);
2584 if (!sectors)
2585 return -EBUSY;
2586 } else if (!sectors)
2587 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2588 rdev->data_offset;
2589 }
2590 if (sectors < my_mddev->dev_sectors)
2591 return -EINVAL; /* component must fit device */
2592
2593 rdev->sectors = sectors;
2594 if (sectors > oldsectors && my_mddev->external) {
2595 /* need to check that all other rdevs with the same ->bdev
2596 * do not overlap. We need to unlock the mddev to avoid
2597 * a deadlock. We have already changed rdev->sectors, and if
2598 * we have to change it back, we will have the lock again.
2599 */
2600 mddev_t *mddev;
2601 int overlap = 0;
2602 struct list_head *tmp;
2603
2604 mddev_unlock(my_mddev);
2605 for_each_mddev(mddev, tmp) {
2606 mdk_rdev_t *rdev2;
2607
2608 mddev_lock(mddev);
2609 list_for_each_entry(rdev2, &mddev->disks, same_set)
2610 if (test_bit(AllReserved, &rdev2->flags) ||
2611 (rdev->bdev == rdev2->bdev &&
2612 rdev != rdev2 &&
2613 overlaps(rdev->data_offset, rdev->sectors,
2614 rdev2->data_offset,
2615 rdev2->sectors))) {
2616 overlap = 1;
2617 break;
2618 }
2619 mddev_unlock(mddev);
2620 if (overlap) {
2621 mddev_put(mddev);
2622 break;
2623 }
2624 }
2625 mddev_lock(my_mddev);
2626 if (overlap) {
2627 /* Someone else could have slipped in a size
2628 * change here, but doing so is just silly.
2629 * We put oldsectors back because we *know* it is
2630 * safe, and trust userspace not to race with
2631 * itself
2632 */
2633 rdev->sectors = oldsectors;
2634 return -EBUSY;
2635 }
2636 }
2637 return len;
2638 }
2639
2640 static struct rdev_sysfs_entry rdev_size =
2641 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2642
2643
2644 static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2645 {
2646 unsigned long long recovery_start = rdev->recovery_offset;
2647
2648 if (test_bit(In_sync, &rdev->flags) ||
2649 recovery_start == MaxSector)
2650 return sprintf(page, "none\n");
2651
2652 return sprintf(page, "%llu\n", recovery_start);
2653 }
2654
2655 static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2656 {
2657 unsigned long long recovery_start;
2658
2659 if (cmd_match(buf, "none"))
2660 recovery_start = MaxSector;
2661 else if (strict_strtoull(buf, 10, &recovery_start))
2662 return -EINVAL;
2663
2664 if (rdev->mddev->pers &&
2665 rdev->raid_disk >= 0)
2666 return -EBUSY;
2667
2668 rdev->recovery_offset = recovery_start;
2669 if (recovery_start == MaxSector)
2670 set_bit(In_sync, &rdev->flags);
2671 else
2672 clear_bit(In_sync, &rdev->flags);
2673 return len;
2674 }
2675
2676 static struct rdev_sysfs_entry rdev_recovery_start =
2677 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2678
2679 static struct attribute *rdev_default_attrs[] = {
2680 &rdev_state.attr,
2681 &rdev_errors.attr,
2682 &rdev_slot.attr,
2683 &rdev_offset.attr,
2684 &rdev_size.attr,
2685 &rdev_recovery_start.attr,
2686 NULL,
2687 };
2688 static ssize_t
2689 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2690 {
2691 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2692 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2693 mddev_t *mddev = rdev->mddev;
2694 ssize_t rv;
2695
2696 if (!entry->show)
2697 return -EIO;
2698
2699 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2700 if (!rv) {
2701 if (rdev->mddev == NULL)
2702 rv = -EBUSY;
2703 else
2704 rv = entry->show(rdev, page);
2705 mddev_unlock(mddev);
2706 }
2707 return rv;
2708 }
2709
2710 static ssize_t
2711 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2712 const char *page, size_t length)
2713 {
2714 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2715 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2716 ssize_t rv;
2717 mddev_t *mddev = rdev->mddev;
2718
2719 if (!entry->store)
2720 return -EIO;
2721 if (!capable(CAP_SYS_ADMIN))
2722 return -EACCES;
2723 rv = mddev ? mddev_lock(mddev): -EBUSY;
2724 if (!rv) {
2725 if (rdev->mddev == NULL)
2726 rv = -EBUSY;
2727 else
2728 rv = entry->store(rdev, page, length);
2729 mddev_unlock(mddev);
2730 }
2731 return rv;
2732 }
2733
2734 static void rdev_free(struct kobject *ko)
2735 {
2736 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2737 kfree(rdev);
2738 }
2739 static const struct sysfs_ops rdev_sysfs_ops = {
2740 .show = rdev_attr_show,
2741 .store = rdev_attr_store,
2742 };
2743 static struct kobj_type rdev_ktype = {
2744 .release = rdev_free,
2745 .sysfs_ops = &rdev_sysfs_ops,
2746 .default_attrs = rdev_default_attrs,
2747 };
2748
2749 void md_rdev_init(mdk_rdev_t *rdev)
2750 {
2751 rdev->desc_nr = -1;
2752 rdev->saved_raid_disk = -1;
2753 rdev->raid_disk = -1;
2754 rdev->flags = 0;
2755 rdev->data_offset = 0;
2756 rdev->sb_events = 0;
2757 rdev->last_read_error.tv_sec = 0;
2758 rdev->last_read_error.tv_nsec = 0;
2759 atomic_set(&rdev->nr_pending, 0);
2760 atomic_set(&rdev->read_errors, 0);
2761 atomic_set(&rdev->corrected_errors, 0);
2762
2763 INIT_LIST_HEAD(&rdev->same_set);
2764 init_waitqueue_head(&rdev->blocked_wait);
2765 }
2766 EXPORT_SYMBOL_GPL(md_rdev_init);
2767 /*
2768 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2769 *
2770 * mark the device faulty if:
2771 *
2772 * - the device is nonexistent (zero size)
2773 * - the device has no valid superblock
2774 *
2775 * a faulty rdev _never_ has rdev->sb set.
2776 */
2777 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2778 {
2779 char b[BDEVNAME_SIZE];
2780 int err;
2781 mdk_rdev_t *rdev;
2782 sector_t size;
2783
2784 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2785 if (!rdev) {
2786 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2787 return ERR_PTR(-ENOMEM);
2788 }
2789
2790 md_rdev_init(rdev);
2791 if ((err = alloc_disk_sb(rdev)))
2792 goto abort_free;
2793
2794 err = lock_rdev(rdev, newdev, super_format == -2);
2795 if (err)
2796 goto abort_free;
2797
2798 kobject_init(&rdev->kobj, &rdev_ktype);
2799
2800 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2801 if (!size) {
2802 printk(KERN_WARNING
2803 "md: %s has zero or unknown size, marking faulty!\n",
2804 bdevname(rdev->bdev,b));
2805 err = -EINVAL;
2806 goto abort_free;
2807 }
2808
2809 if (super_format >= 0) {
2810 err = super_types[super_format].
2811 load_super(rdev, NULL, super_minor);
2812 if (err == -EINVAL) {
2813 printk(KERN_WARNING
2814 "md: %s does not have a valid v%d.%d "
2815 "superblock, not importing!\n",
2816 bdevname(rdev->bdev,b),
2817 super_format, super_minor);
2818 goto abort_free;
2819 }
2820 if (err < 0) {
2821 printk(KERN_WARNING
2822 "md: could not read %s's sb, not importing!\n",
2823 bdevname(rdev->bdev,b));
2824 goto abort_free;
2825 }
2826 }
2827
2828 return rdev;
2829
2830 abort_free:
2831 if (rdev->sb_page) {
2832 if (rdev->bdev)
2833 unlock_rdev(rdev);
2834 free_disk_sb(rdev);
2835 }
2836 kfree(rdev);
2837 return ERR_PTR(err);
2838 }
2839
2840 /*
2841 * Check a full RAID array for plausibility
2842 */
2843
2844
2845 static void analyze_sbs(mddev_t * mddev)
2846 {
2847 int i;
2848 mdk_rdev_t *rdev, *freshest, *tmp;
2849 char b[BDEVNAME_SIZE];
2850
2851 freshest = NULL;
2852 rdev_for_each(rdev, tmp, mddev)
2853 switch (super_types[mddev->major_version].
2854 load_super(rdev, freshest, mddev->minor_version)) {
2855 case 1:
2856 freshest = rdev;
2857 break;
2858 case 0:
2859 break;
2860 default:
2861 printk( KERN_ERR \
2862 "md: fatal superblock inconsistency in %s"
2863 " -- removing from array\n",
2864 bdevname(rdev->bdev,b));
2865 kick_rdev_from_array(rdev);
2866 }
2867
2868
2869 super_types[mddev->major_version].
2870 validate_super(mddev, freshest);
2871
2872 i = 0;
2873 rdev_for_each(rdev, tmp, mddev) {
2874 if (mddev->max_disks &&
2875 (rdev->desc_nr >= mddev->max_disks ||
2876 i > mddev->max_disks)) {
2877 printk(KERN_WARNING
2878 "md: %s: %s: only %d devices permitted\n",
2879 mdname(mddev), bdevname(rdev->bdev, b),
2880 mddev->max_disks);
2881 kick_rdev_from_array(rdev);
2882 continue;
2883 }
2884 if (rdev != freshest)
2885 if (super_types[mddev->major_version].
2886 validate_super(mddev, rdev)) {
2887 printk(KERN_WARNING "md: kicking non-fresh %s"
2888 " from array!\n",
2889 bdevname(rdev->bdev,b));
2890 kick_rdev_from_array(rdev);
2891 continue;
2892 }
2893 if (mddev->level == LEVEL_MULTIPATH) {
2894 rdev->desc_nr = i++;
2895 rdev->raid_disk = rdev->desc_nr;
2896 set_bit(In_sync, &rdev->flags);
2897 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2898 rdev->raid_disk = -1;
2899 clear_bit(In_sync, &rdev->flags);
2900 }
2901 }
2902 }
2903
2904 /* Read a fixed-point number.
2905 * Numbers in sysfs attributes should be in "standard" units where
2906 * possible, so time should be in seconds.
2907 * However we internally use a a much smaller unit such as
2908 * milliseconds or jiffies.
2909 * This function takes a decimal number with a possible fractional
2910 * component, and produces an integer which is the result of
2911 * multiplying that number by 10^'scale'.
2912 * all without any floating-point arithmetic.
2913 */
2914 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2915 {
2916 unsigned long result = 0;
2917 long decimals = -1;
2918 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2919 if (*cp == '.')
2920 decimals = 0;
2921 else if (decimals < scale) {
2922 unsigned int value;
2923 value = *cp - '0';
2924 result = result * 10 + value;
2925 if (decimals >= 0)
2926 decimals++;
2927 }
2928 cp++;
2929 }
2930 if (*cp == '\n')
2931 cp++;
2932 if (*cp)
2933 return -EINVAL;
2934 if (decimals < 0)
2935 decimals = 0;
2936 while (decimals < scale) {
2937 result *= 10;
2938 decimals ++;
2939 }
2940 *res = result;
2941 return 0;
2942 }
2943
2944
2945 static void md_safemode_timeout(unsigned long data);
2946
2947 static ssize_t
2948 safe_delay_show(mddev_t *mddev, char *page)
2949 {
2950 int msec = (mddev->safemode_delay*1000)/HZ;
2951 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2952 }
2953 static ssize_t
2954 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2955 {
2956 unsigned long msec;
2957
2958 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
2959 return -EINVAL;
2960 if (msec == 0)
2961 mddev->safemode_delay = 0;
2962 else {
2963 unsigned long old_delay = mddev->safemode_delay;
2964 mddev->safemode_delay = (msec*HZ)/1000;
2965 if (mddev->safemode_delay == 0)
2966 mddev->safemode_delay = 1;
2967 if (mddev->safemode_delay < old_delay)
2968 md_safemode_timeout((unsigned long)mddev);
2969 }
2970 return len;
2971 }
2972 static struct md_sysfs_entry md_safe_delay =
2973 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2974
2975 static ssize_t
2976 level_show(mddev_t *mddev, char *page)
2977 {
2978 struct mdk_personality *p = mddev->pers;
2979 if (p)
2980 return sprintf(page, "%s\n", p->name);
2981 else if (mddev->clevel[0])
2982 return sprintf(page, "%s\n", mddev->clevel);
2983 else if (mddev->level != LEVEL_NONE)
2984 return sprintf(page, "%d\n", mddev->level);
2985 else
2986 return 0;
2987 }
2988
2989 static ssize_t
2990 level_store(mddev_t *mddev, const char *buf, size_t len)
2991 {
2992 char clevel[16];
2993 ssize_t rv = len;
2994 struct mdk_personality *pers;
2995 long level;
2996 void *priv;
2997 mdk_rdev_t *rdev;
2998
2999 if (mddev->pers == NULL) {
3000 if (len == 0)
3001 return 0;
3002 if (len >= sizeof(mddev->clevel))
3003 return -ENOSPC;
3004 strncpy(mddev->clevel, buf, len);
3005 if (mddev->clevel[len-1] == '\n')
3006 len--;
3007 mddev->clevel[len] = 0;
3008 mddev->level = LEVEL_NONE;
3009 return rv;
3010 }
3011
3012 /* request to change the personality. Need to ensure:
3013 * - array is not engaged in resync/recovery/reshape
3014 * - old personality can be suspended
3015 * - new personality will access other array.
3016 */
3017
3018 if (mddev->sync_thread ||
3019 mddev->reshape_position != MaxSector ||
3020 mddev->sysfs_active)
3021 return -EBUSY;
3022
3023 if (!mddev->pers->quiesce) {
3024 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3025 mdname(mddev), mddev->pers->name);
3026 return -EINVAL;
3027 }
3028
3029 /* Now find the new personality */
3030 if (len == 0 || len >= sizeof(clevel))
3031 return -EINVAL;
3032 strncpy(clevel, buf, len);
3033 if (clevel[len-1] == '\n')
3034 len--;
3035 clevel[len] = 0;
3036 if (strict_strtol(clevel, 10, &level))
3037 level = LEVEL_NONE;
3038
3039 if (request_module("md-%s", clevel) != 0)
3040 request_module("md-level-%s", clevel);
3041 spin_lock(&pers_lock);
3042 pers = find_pers(level, clevel);
3043 if (!pers || !try_module_get(pers->owner)) {
3044 spin_unlock(&pers_lock);
3045 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3046 return -EINVAL;
3047 }
3048 spin_unlock(&pers_lock);
3049
3050 if (pers == mddev->pers) {
3051 /* Nothing to do! */
3052 module_put(pers->owner);
3053 return rv;
3054 }
3055 if (!pers->takeover) {
3056 module_put(pers->owner);
3057 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3058 mdname(mddev), clevel);
3059 return -EINVAL;
3060 }
3061
3062 list_for_each_entry(rdev, &mddev->disks, same_set)
3063 rdev->new_raid_disk = rdev->raid_disk;
3064
3065 /* ->takeover must set new_* and/or delta_disks
3066 * if it succeeds, and may set them when it fails.
3067 */
3068 priv = pers->takeover(mddev);
3069 if (IS_ERR(priv)) {
3070 mddev->new_level = mddev->level;
3071 mddev->new_layout = mddev->layout;
3072 mddev->new_chunk_sectors = mddev->chunk_sectors;
3073 mddev->raid_disks -= mddev->delta_disks;
3074 mddev->delta_disks = 0;
3075 module_put(pers->owner);
3076 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3077 mdname(mddev), clevel);
3078 return PTR_ERR(priv);
3079 }
3080
3081 /* Looks like we have a winner */
3082 mddev_suspend(mddev);
3083 mddev->pers->stop(mddev);
3084
3085 if (mddev->pers->sync_request == NULL &&
3086 pers->sync_request != NULL) {
3087 /* need to add the md_redundancy_group */
3088 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3089 printk(KERN_WARNING
3090 "md: cannot register extra attributes for %s\n",
3091 mdname(mddev));
3092 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3093 }
3094 if (mddev->pers->sync_request != NULL &&
3095 pers->sync_request == NULL) {
3096 /* need to remove the md_redundancy_group */
3097 if (mddev->to_remove == NULL)
3098 mddev->to_remove = &md_redundancy_group;
3099 }
3100
3101 if (mddev->pers->sync_request == NULL &&
3102 mddev->external) {
3103 /* We are converting from a no-redundancy array
3104 * to a redundancy array and metadata is managed
3105 * externally so we need to be sure that writes
3106 * won't block due to a need to transition
3107 * clean->dirty
3108 * until external management is started.
3109 */
3110 mddev->in_sync = 0;
3111 mddev->safemode_delay = 0;
3112 mddev->safemode = 0;
3113 }
3114
3115 list_for_each_entry(rdev, &mddev->disks, same_set) {
3116 char nm[20];
3117 if (rdev->raid_disk < 0)
3118 continue;
3119 if (rdev->new_raid_disk >= mddev->raid_disks)
3120 rdev->new_raid_disk = -1;
3121 if (rdev->new_raid_disk == rdev->raid_disk)
3122 continue;
3123 sprintf(nm, "rd%d", rdev->raid_disk);
3124 sysfs_remove_link(&mddev->kobj, nm);
3125 }
3126 list_for_each_entry(rdev, &mddev->disks, same_set) {
3127 if (rdev->raid_disk < 0)
3128 continue;
3129 if (rdev->new_raid_disk == rdev->raid_disk)
3130 continue;
3131 rdev->raid_disk = rdev->new_raid_disk;
3132 if (rdev->raid_disk < 0)
3133 clear_bit(In_sync, &rdev->flags);
3134 else {
3135 char nm[20];
3136 sprintf(nm, "rd%d", rdev->raid_disk);
3137 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3138 printk("md: cannot register %s for %s after level change\n",
3139 nm, mdname(mddev));
3140 }
3141 }
3142
3143 module_put(mddev->pers->owner);
3144 mddev->pers = pers;
3145 mddev->private = priv;
3146 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3147 mddev->level = mddev->new_level;
3148 mddev->layout = mddev->new_layout;
3149 mddev->chunk_sectors = mddev->new_chunk_sectors;
3150 mddev->delta_disks = 0;
3151 if (mddev->pers->sync_request == NULL) {
3152 /* this is now an array without redundancy, so
3153 * it must always be in_sync
3154 */
3155 mddev->in_sync = 1;
3156 del_timer_sync(&mddev->safemode_timer);
3157 }
3158 pers->run(mddev);
3159 mddev_resume(mddev);
3160 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3161 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3162 md_wakeup_thread(mddev->thread);
3163 sysfs_notify(&mddev->kobj, NULL, "level");
3164 md_new_event(mddev);
3165 return rv;
3166 }
3167
3168 static struct md_sysfs_entry md_level =
3169 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3170
3171
3172 static ssize_t
3173 layout_show(mddev_t *mddev, char *page)
3174 {
3175 /* just a number, not meaningful for all levels */
3176 if (mddev->reshape_position != MaxSector &&
3177 mddev->layout != mddev->new_layout)
3178 return sprintf(page, "%d (%d)\n",
3179 mddev->new_layout, mddev->layout);
3180 return sprintf(page, "%d\n", mddev->layout);
3181 }
3182
3183 static ssize_t
3184 layout_store(mddev_t *mddev, const char *buf, size_t len)
3185 {
3186 char *e;
3187 unsigned long n = simple_strtoul(buf, &e, 10);
3188
3189 if (!*buf || (*e && *e != '\n'))
3190 return -EINVAL;
3191
3192 if (mddev->pers) {
3193 int err;
3194 if (mddev->pers->check_reshape == NULL)
3195 return -EBUSY;
3196 mddev->new_layout = n;
3197 err = mddev->pers->check_reshape(mddev);
3198 if (err) {
3199 mddev->new_layout = mddev->layout;
3200 return err;
3201 }
3202 } else {
3203 mddev->new_layout = n;
3204 if (mddev->reshape_position == MaxSector)
3205 mddev->layout = n;
3206 }
3207 return len;
3208 }
3209 static struct md_sysfs_entry md_layout =
3210 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3211
3212
3213 static ssize_t
3214 raid_disks_show(mddev_t *mddev, char *page)
3215 {
3216 if (mddev->raid_disks == 0)
3217 return 0;
3218 if (mddev->reshape_position != MaxSector &&
3219 mddev->delta_disks != 0)
3220 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3221 mddev->raid_disks - mddev->delta_disks);
3222 return sprintf(page, "%d\n", mddev->raid_disks);
3223 }
3224
3225 static int update_raid_disks(mddev_t *mddev, int raid_disks);
3226
3227 static ssize_t
3228 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3229 {
3230 char *e;
3231 int rv = 0;
3232 unsigned long n = simple_strtoul(buf, &e, 10);
3233
3234 if (!*buf || (*e && *e != '\n'))
3235 return -EINVAL;
3236
3237 if (mddev->pers)
3238 rv = update_raid_disks(mddev, n);
3239 else if (mddev->reshape_position != MaxSector) {
3240 int olddisks = mddev->raid_disks - mddev->delta_disks;
3241 mddev->delta_disks = n - olddisks;
3242 mddev->raid_disks = n;
3243 } else
3244 mddev->raid_disks = n;
3245 return rv ? rv : len;
3246 }
3247 static struct md_sysfs_entry md_raid_disks =
3248 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3249
3250 static ssize_t
3251 chunk_size_show(mddev_t *mddev, char *page)
3252 {
3253 if (mddev->reshape_position != MaxSector &&
3254 mddev->chunk_sectors != mddev->new_chunk_sectors)
3255 return sprintf(page, "%d (%d)\n",
3256 mddev->new_chunk_sectors << 9,
3257 mddev->chunk_sectors << 9);
3258 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3259 }
3260
3261 static ssize_t
3262 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3263 {
3264 char *e;
3265 unsigned long n = simple_strtoul(buf, &e, 10);
3266
3267 if (!*buf || (*e && *e != '\n'))
3268 return -EINVAL;
3269
3270 if (mddev->pers) {
3271 int err;
3272 if (mddev->pers->check_reshape == NULL)
3273 return -EBUSY;
3274 mddev->new_chunk_sectors = n >> 9;
3275 err = mddev->pers->check_reshape(mddev);
3276 if (err) {
3277 mddev->new_chunk_sectors = mddev->chunk_sectors;
3278 return err;
3279 }
3280 } else {
3281 mddev->new_chunk_sectors = n >> 9;
3282 if (mddev->reshape_position == MaxSector)
3283 mddev->chunk_sectors = n >> 9;
3284 }
3285 return len;
3286 }
3287 static struct md_sysfs_entry md_chunk_size =
3288 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3289
3290 static ssize_t
3291 resync_start_show(mddev_t *mddev, char *page)
3292 {
3293 if (mddev->recovery_cp == MaxSector)
3294 return sprintf(page, "none\n");
3295 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3296 }
3297
3298 static ssize_t
3299 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3300 {
3301 char *e;
3302 unsigned long long n = simple_strtoull(buf, &e, 10);
3303
3304 if (mddev->pers)
3305 return -EBUSY;
3306 if (cmd_match(buf, "none"))
3307 n = MaxSector;
3308 else if (!*buf || (*e && *e != '\n'))
3309 return -EINVAL;
3310
3311 mddev->recovery_cp = n;
3312 return len;
3313 }
3314 static struct md_sysfs_entry md_resync_start =
3315 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3316
3317 /*
3318 * The array state can be:
3319 *
3320 * clear
3321 * No devices, no size, no level
3322 * Equivalent to STOP_ARRAY ioctl
3323 * inactive
3324 * May have some settings, but array is not active
3325 * all IO results in error
3326 * When written, doesn't tear down array, but just stops it
3327 * suspended (not supported yet)
3328 * All IO requests will block. The array can be reconfigured.
3329 * Writing this, if accepted, will block until array is quiescent
3330 * readonly
3331 * no resync can happen. no superblocks get written.
3332 * write requests fail
3333 * read-auto
3334 * like readonly, but behaves like 'clean' on a write request.
3335 *
3336 * clean - no pending writes, but otherwise active.
3337 * When written to inactive array, starts without resync
3338 * If a write request arrives then
3339 * if metadata is known, mark 'dirty' and switch to 'active'.
3340 * if not known, block and switch to write-pending
3341 * If written to an active array that has pending writes, then fails.
3342 * active
3343 * fully active: IO and resync can be happening.
3344 * When written to inactive array, starts with resync
3345 *
3346 * write-pending
3347 * clean, but writes are blocked waiting for 'active' to be written.
3348 *
3349 * active-idle
3350 * like active, but no writes have been seen for a while (100msec).
3351 *
3352 */
3353 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3354 write_pending, active_idle, bad_word};
3355 static char *array_states[] = {
3356 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3357 "write-pending", "active-idle", NULL };
3358
3359 static int match_word(const char *word, char **list)
3360 {
3361 int n;
3362 for (n=0; list[n]; n++)
3363 if (cmd_match(word, list[n]))
3364 break;
3365 return n;
3366 }
3367
3368 static ssize_t
3369 array_state_show(mddev_t *mddev, char *page)
3370 {
3371 enum array_state st = inactive;
3372
3373 if (mddev->pers)
3374 switch(mddev->ro) {
3375 case 1:
3376 st = readonly;
3377 break;
3378 case 2:
3379 st = read_auto;
3380 break;
3381 case 0:
3382 if (mddev->in_sync)
3383 st = clean;
3384 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3385 st = write_pending;
3386 else if (mddev->safemode)
3387 st = active_idle;
3388 else
3389 st = active;
3390 }
3391 else {
3392 if (list_empty(&mddev->disks) &&
3393 mddev->raid_disks == 0 &&
3394 mddev->dev_sectors == 0)
3395 st = clear;
3396 else
3397 st = inactive;
3398 }
3399 return sprintf(page, "%s\n", array_states[st]);
3400 }
3401
3402 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3403 static int md_set_readonly(mddev_t * mddev, int is_open);
3404 static int do_md_run(mddev_t * mddev);
3405 static int restart_array(mddev_t *mddev);
3406
3407 static ssize_t
3408 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3409 {
3410 int err = -EINVAL;
3411 enum array_state st = match_word(buf, array_states);
3412 switch(st) {
3413 case bad_word:
3414 break;
3415 case clear:
3416 /* stopping an active array */
3417 if (atomic_read(&mddev->openers) > 0)
3418 return -EBUSY;
3419 err = do_md_stop(mddev, 0, 0);
3420 break;
3421 case inactive:
3422 /* stopping an active array */
3423 if (mddev->pers) {
3424 if (atomic_read(&mddev->openers) > 0)
3425 return -EBUSY;
3426 err = do_md_stop(mddev, 2, 0);
3427 } else
3428 err = 0; /* already inactive */
3429 break;
3430 case suspended:
3431 break; /* not supported yet */
3432 case readonly:
3433 if (mddev->pers)
3434 err = md_set_readonly(mddev, 0);
3435 else {
3436 mddev->ro = 1;
3437 set_disk_ro(mddev->gendisk, 1);
3438 err = do_md_run(mddev);
3439 }
3440 break;
3441 case read_auto:
3442 if (mddev->pers) {
3443 if (mddev->ro == 0)
3444 err = md_set_readonly(mddev, 0);
3445 else if (mddev->ro == 1)
3446 err = restart_array(mddev);
3447 if (err == 0) {
3448 mddev->ro = 2;
3449 set_disk_ro(mddev->gendisk, 0);
3450 }
3451 } else {
3452 mddev->ro = 2;
3453 err = do_md_run(mddev);
3454 }
3455 break;
3456 case clean:
3457 if (mddev->pers) {
3458 restart_array(mddev);
3459 spin_lock_irq(&mddev->write_lock);
3460 if (atomic_read(&mddev->writes_pending) == 0) {
3461 if (mddev->in_sync == 0) {
3462 mddev->in_sync = 1;
3463 if (mddev->safemode == 1)
3464 mddev->safemode = 0;
3465 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3466 }
3467 err = 0;
3468 } else
3469 err = -EBUSY;
3470 spin_unlock_irq(&mddev->write_lock);
3471 } else
3472 err = -EINVAL;
3473 break;
3474 case active:
3475 if (mddev->pers) {
3476 restart_array(mddev);
3477 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3478 wake_up(&mddev->sb_wait);
3479 err = 0;
3480 } else {
3481 mddev->ro = 0;
3482 set_disk_ro(mddev->gendisk, 0);
3483 err = do_md_run(mddev);
3484 }
3485 break;
3486 case write_pending:
3487 case active_idle:
3488 /* these cannot be set */
3489 break;
3490 }
3491 if (err)
3492 return err;
3493 else {
3494 sysfs_notify_dirent_safe(mddev->sysfs_state);
3495 return len;
3496 }
3497 }
3498 static struct md_sysfs_entry md_array_state =
3499 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3500
3501 static ssize_t
3502 max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3503 return sprintf(page, "%d\n",
3504 atomic_read(&mddev->max_corr_read_errors));
3505 }
3506
3507 static ssize_t
3508 max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3509 {
3510 char *e;
3511 unsigned long n = simple_strtoul(buf, &e, 10);
3512
3513 if (*buf && (*e == 0 || *e == '\n')) {
3514 atomic_set(&mddev->max_corr_read_errors, n);
3515 return len;
3516 }
3517 return -EINVAL;
3518 }
3519
3520 static struct md_sysfs_entry max_corr_read_errors =
3521 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3522 max_corrected_read_errors_store);
3523
3524 static ssize_t
3525 null_show(mddev_t *mddev, char *page)
3526 {
3527 return -EINVAL;
3528 }
3529
3530 static ssize_t
3531 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3532 {
3533 /* buf must be %d:%d\n? giving major and minor numbers */
3534 /* The new device is added to the array.
3535 * If the array has a persistent superblock, we read the
3536 * superblock to initialise info and check validity.
3537 * Otherwise, only checking done is that in bind_rdev_to_array,
3538 * which mainly checks size.
3539 */
3540 char *e;
3541 int major = simple_strtoul(buf, &e, 10);
3542 int minor;
3543 dev_t dev;
3544 mdk_rdev_t *rdev;
3545 int err;
3546
3547 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3548 return -EINVAL;
3549 minor = simple_strtoul(e+1, &e, 10);
3550 if (*e && *e != '\n')
3551 return -EINVAL;
3552 dev = MKDEV(major, minor);
3553 if (major != MAJOR(dev) ||
3554 minor != MINOR(dev))
3555 return -EOVERFLOW;
3556
3557
3558 if (mddev->persistent) {
3559 rdev = md_import_device(dev, mddev->major_version,
3560 mddev->minor_version);
3561 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3562 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3563 mdk_rdev_t, same_set);
3564 err = super_types[mddev->major_version]
3565 .load_super(rdev, rdev0, mddev->minor_version);
3566 if (err < 0)
3567 goto out;
3568 }
3569 } else if (mddev->external)
3570 rdev = md_import_device(dev, -2, -1);
3571 else
3572 rdev = md_import_device(dev, -1, -1);
3573
3574 if (IS_ERR(rdev))
3575 return PTR_ERR(rdev);
3576 err = bind_rdev_to_array(rdev, mddev);
3577 out:
3578 if (err)
3579 export_rdev(rdev);
3580 return err ? err : len;
3581 }
3582
3583 static struct md_sysfs_entry md_new_device =
3584 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3585
3586 static ssize_t
3587 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3588 {
3589 char *end;
3590 unsigned long chunk, end_chunk;
3591
3592 if (!mddev->bitmap)
3593 goto out;
3594 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3595 while (*buf) {
3596 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3597 if (buf == end) break;
3598 if (*end == '-') { /* range */
3599 buf = end + 1;
3600 end_chunk = simple_strtoul(buf, &end, 0);
3601 if (buf == end) break;
3602 }
3603 if (*end && !isspace(*end)) break;
3604 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3605 buf = skip_spaces(end);
3606 }
3607 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3608 out:
3609 return len;
3610 }
3611
3612 static struct md_sysfs_entry md_bitmap =
3613 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3614
3615 static ssize_t
3616 size_show(mddev_t *mddev, char *page)
3617 {
3618 return sprintf(page, "%llu\n",
3619 (unsigned long long)mddev->dev_sectors / 2);
3620 }
3621
3622 static int update_size(mddev_t *mddev, sector_t num_sectors);
3623
3624 static ssize_t
3625 size_store(mddev_t *mddev, const char *buf, size_t len)
3626 {
3627 /* If array is inactive, we can reduce the component size, but
3628 * not increase it (except from 0).
3629 * If array is active, we can try an on-line resize
3630 */
3631 sector_t sectors;
3632 int err = strict_blocks_to_sectors(buf, &sectors);
3633
3634 if (err < 0)
3635 return err;
3636 if (mddev->pers) {
3637 err = update_size(mddev, sectors);
3638 md_update_sb(mddev, 1);
3639 } else {
3640 if (mddev->dev_sectors == 0 ||
3641 mddev->dev_sectors > sectors)
3642 mddev->dev_sectors = sectors;
3643 else
3644 err = -ENOSPC;
3645 }
3646 return err ? err : len;
3647 }
3648
3649 static struct md_sysfs_entry md_size =
3650 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3651
3652
3653 /* Metdata version.
3654 * This is one of
3655 * 'none' for arrays with no metadata (good luck...)
3656 * 'external' for arrays with externally managed metadata,
3657 * or N.M for internally known formats
3658 */
3659 static ssize_t
3660 metadata_show(mddev_t *mddev, char *page)
3661 {
3662 if (mddev->persistent)
3663 return sprintf(page, "%d.%d\n",
3664 mddev->major_version, mddev->minor_version);
3665 else if (mddev->external)
3666 return sprintf(page, "external:%s\n", mddev->metadata_type);
3667 else
3668 return sprintf(page, "none\n");
3669 }
3670
3671 static ssize_t
3672 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3673 {
3674 int major, minor;
3675 char *e;
3676 /* Changing the details of 'external' metadata is
3677 * always permitted. Otherwise there must be
3678 * no devices attached to the array.
3679 */
3680 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3681 ;
3682 else if (!list_empty(&mddev->disks))
3683 return -EBUSY;
3684
3685 if (cmd_match(buf, "none")) {
3686 mddev->persistent = 0;
3687 mddev->external = 0;
3688 mddev->major_version = 0;
3689 mddev->minor_version = 90;
3690 return len;
3691 }
3692 if (strncmp(buf, "external:", 9) == 0) {
3693 size_t namelen = len-9;
3694 if (namelen >= sizeof(mddev->metadata_type))
3695 namelen = sizeof(mddev->metadata_type)-1;
3696 strncpy(mddev->metadata_type, buf+9, namelen);
3697 mddev->metadata_type[namelen] = 0;
3698 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3699 mddev->metadata_type[--namelen] = 0;
3700 mddev->persistent = 0;
3701 mddev->external = 1;
3702 mddev->major_version = 0;
3703 mddev->minor_version = 90;
3704 return len;
3705 }
3706 major = simple_strtoul(buf, &e, 10);
3707 if (e==buf || *e != '.')
3708 return -EINVAL;
3709 buf = e+1;
3710 minor = simple_strtoul(buf, &e, 10);
3711 if (e==buf || (*e && *e != '\n') )
3712 return -EINVAL;
3713 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3714 return -ENOENT;
3715 mddev->major_version = major;
3716 mddev->minor_version = minor;
3717 mddev->persistent = 1;
3718 mddev->external = 0;
3719 return len;
3720 }
3721
3722 static struct md_sysfs_entry md_metadata =
3723 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3724
3725 static ssize_t
3726 action_show(mddev_t *mddev, char *page)
3727 {
3728 char *type = "idle";
3729 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3730 type = "frozen";
3731 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3732 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3733 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3734 type = "reshape";
3735 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3736 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3737 type = "resync";
3738 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3739 type = "check";
3740 else
3741 type = "repair";
3742 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3743 type = "recover";
3744 }
3745 return sprintf(page, "%s\n", type);
3746 }
3747
3748 static ssize_t
3749 action_store(mddev_t *mddev, const char *page, size_t len)
3750 {
3751 if (!mddev->pers || !mddev->pers->sync_request)
3752 return -EINVAL;
3753
3754 if (cmd_match(page, "frozen"))
3755 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3756 else
3757 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3758
3759 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3760 if (mddev->sync_thread) {
3761 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3762 md_unregister_thread(mddev->sync_thread);
3763 mddev->sync_thread = NULL;
3764 mddev->recovery = 0;
3765 }
3766 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3767 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3768 return -EBUSY;
3769 else if (cmd_match(page, "resync"))
3770 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3771 else if (cmd_match(page, "recover")) {
3772 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3773 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3774 } else if (cmd_match(page, "reshape")) {
3775 int err;
3776 if (mddev->pers->start_reshape == NULL)
3777 return -EINVAL;
3778 err = mddev->pers->start_reshape(mddev);
3779 if (err)
3780 return err;
3781 sysfs_notify(&mddev->kobj, NULL, "degraded");
3782 } else {
3783 if (cmd_match(page, "check"))
3784 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3785 else if (!cmd_match(page, "repair"))
3786 return -EINVAL;
3787 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3788 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3789 }
3790 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3791 md_wakeup_thread(mddev->thread);
3792 sysfs_notify_dirent_safe(mddev->sysfs_action);
3793 return len;
3794 }
3795
3796 static ssize_t
3797 mismatch_cnt_show(mddev_t *mddev, char *page)
3798 {
3799 return sprintf(page, "%llu\n",
3800 (unsigned long long) mddev->resync_mismatches);
3801 }
3802
3803 static struct md_sysfs_entry md_scan_mode =
3804 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3805
3806
3807 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3808
3809 static ssize_t
3810 sync_min_show(mddev_t *mddev, char *page)
3811 {
3812 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3813 mddev->sync_speed_min ? "local": "system");
3814 }
3815
3816 static ssize_t
3817 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3818 {
3819 int min;
3820 char *e;
3821 if (strncmp(buf, "system", 6)==0) {
3822 mddev->sync_speed_min = 0;
3823 return len;
3824 }
3825 min = simple_strtoul(buf, &e, 10);
3826 if (buf == e || (*e && *e != '\n') || min <= 0)
3827 return -EINVAL;
3828 mddev->sync_speed_min = min;
3829 return len;
3830 }
3831
3832 static struct md_sysfs_entry md_sync_min =
3833 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3834
3835 static ssize_t
3836 sync_max_show(mddev_t *mddev, char *page)
3837 {
3838 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3839 mddev->sync_speed_max ? "local": "system");
3840 }
3841
3842 static ssize_t
3843 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3844 {
3845 int max;
3846 char *e;
3847 if (strncmp(buf, "system", 6)==0) {
3848 mddev->sync_speed_max = 0;
3849 return len;
3850 }
3851 max = simple_strtoul(buf, &e, 10);
3852 if (buf == e || (*e && *e != '\n') || max <= 0)
3853 return -EINVAL;
3854 mddev->sync_speed_max = max;
3855 return len;
3856 }
3857
3858 static struct md_sysfs_entry md_sync_max =
3859 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3860
3861 static ssize_t
3862 degraded_show(mddev_t *mddev, char *page)
3863 {
3864 return sprintf(page, "%d\n", mddev->degraded);
3865 }
3866 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3867
3868 static ssize_t
3869 sync_force_parallel_show(mddev_t *mddev, char *page)
3870 {
3871 return sprintf(page, "%d\n", mddev->parallel_resync);
3872 }
3873
3874 static ssize_t
3875 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3876 {
3877 long n;
3878
3879 if (strict_strtol(buf, 10, &n))
3880 return -EINVAL;
3881
3882 if (n != 0 && n != 1)
3883 return -EINVAL;
3884
3885 mddev->parallel_resync = n;
3886
3887 if (mddev->sync_thread)
3888 wake_up(&resync_wait);
3889
3890 return len;
3891 }
3892
3893 /* force parallel resync, even with shared block devices */
3894 static struct md_sysfs_entry md_sync_force_parallel =
3895 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3896 sync_force_parallel_show, sync_force_parallel_store);
3897
3898 static ssize_t
3899 sync_speed_show(mddev_t *mddev, char *page)
3900 {
3901 unsigned long resync, dt, db;
3902 if (mddev->curr_resync == 0)
3903 return sprintf(page, "none\n");
3904 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3905 dt = (jiffies - mddev->resync_mark) / HZ;
3906 if (!dt) dt++;
3907 db = resync - mddev->resync_mark_cnt;
3908 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3909 }
3910
3911 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3912
3913 static ssize_t
3914 sync_completed_show(mddev_t *mddev, char *page)
3915 {
3916 unsigned long max_sectors, resync;
3917
3918 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3919 return sprintf(page, "none\n");
3920
3921 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3922 max_sectors = mddev->resync_max_sectors;
3923 else
3924 max_sectors = mddev->dev_sectors;
3925
3926 resync = mddev->curr_resync_completed;
3927 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3928 }
3929
3930 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3931
3932 static ssize_t
3933 min_sync_show(mddev_t *mddev, char *page)
3934 {
3935 return sprintf(page, "%llu\n",
3936 (unsigned long long)mddev->resync_min);
3937 }
3938 static ssize_t
3939 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3940 {
3941 unsigned long long min;
3942 if (strict_strtoull(buf, 10, &min))
3943 return -EINVAL;
3944 if (min > mddev->resync_max)
3945 return -EINVAL;
3946 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3947 return -EBUSY;
3948
3949 /* Must be a multiple of chunk_size */
3950 if (mddev->chunk_sectors) {
3951 sector_t temp = min;
3952 if (sector_div(temp, mddev->chunk_sectors))
3953 return -EINVAL;
3954 }
3955 mddev->resync_min = min;
3956
3957 return len;
3958 }
3959
3960 static struct md_sysfs_entry md_min_sync =
3961 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3962
3963 static ssize_t
3964 max_sync_show(mddev_t *mddev, char *page)
3965 {
3966 if (mddev->resync_max == MaxSector)
3967 return sprintf(page, "max\n");
3968 else
3969 return sprintf(page, "%llu\n",
3970 (unsigned long long)mddev->resync_max);
3971 }
3972 static ssize_t
3973 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3974 {
3975 if (strncmp(buf, "max", 3) == 0)
3976 mddev->resync_max = MaxSector;
3977 else {
3978 unsigned long long max;
3979 if (strict_strtoull(buf, 10, &max))
3980 return -EINVAL;
3981 if (max < mddev->resync_min)
3982 return -EINVAL;
3983 if (max < mddev->resync_max &&
3984 mddev->ro == 0 &&
3985 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3986 return -EBUSY;
3987
3988 /* Must be a multiple of chunk_size */
3989 if (mddev->chunk_sectors) {
3990 sector_t temp = max;
3991 if (sector_div(temp, mddev->chunk_sectors))
3992 return -EINVAL;
3993 }
3994 mddev->resync_max = max;
3995 }
3996 wake_up(&mddev->recovery_wait);
3997 return len;
3998 }
3999
4000 static struct md_sysfs_entry md_max_sync =
4001 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4002
4003 static ssize_t
4004 suspend_lo_show(mddev_t *mddev, char *page)
4005 {
4006 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4007 }
4008
4009 static ssize_t
4010 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4011 {
4012 char *e;
4013 unsigned long long new = simple_strtoull(buf, &e, 10);
4014
4015 if (mddev->pers == NULL ||
4016 mddev->pers->quiesce == NULL)
4017 return -EINVAL;
4018 if (buf == e || (*e && *e != '\n'))
4019 return -EINVAL;
4020 if (new >= mddev->suspend_hi ||
4021 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4022 mddev->suspend_lo = new;
4023 mddev->pers->quiesce(mddev, 2);
4024 return len;
4025 } else
4026 return -EINVAL;
4027 }
4028 static struct md_sysfs_entry md_suspend_lo =
4029 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4030
4031
4032 static ssize_t
4033 suspend_hi_show(mddev_t *mddev, char *page)
4034 {
4035 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4036 }
4037
4038 static ssize_t
4039 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4040 {
4041 char *e;
4042 unsigned long long new = simple_strtoull(buf, &e, 10);
4043
4044 if (mddev->pers == NULL ||
4045 mddev->pers->quiesce == NULL)
4046 return -EINVAL;
4047 if (buf == e || (*e && *e != '\n'))
4048 return -EINVAL;
4049 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4050 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4051 mddev->suspend_hi = new;
4052 mddev->pers->quiesce(mddev, 1);
4053 mddev->pers->quiesce(mddev, 0);
4054 return len;
4055 } else
4056 return -EINVAL;
4057 }
4058 static struct md_sysfs_entry md_suspend_hi =
4059 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4060
4061 static ssize_t
4062 reshape_position_show(mddev_t *mddev, char *page)
4063 {
4064 if (mddev->reshape_position != MaxSector)
4065 return sprintf(page, "%llu\n",
4066 (unsigned long long)mddev->reshape_position);
4067 strcpy(page, "none\n");
4068 return 5;
4069 }
4070
4071 static ssize_t
4072 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4073 {
4074 char *e;
4075 unsigned long long new = simple_strtoull(buf, &e, 10);
4076 if (mddev->pers)
4077 return -EBUSY;
4078 if (buf == e || (*e && *e != '\n'))
4079 return -EINVAL;
4080 mddev->reshape_position = new;
4081 mddev->delta_disks = 0;
4082 mddev->new_level = mddev->level;
4083 mddev->new_layout = mddev->layout;
4084 mddev->new_chunk_sectors = mddev->chunk_sectors;
4085 return len;
4086 }
4087
4088 static struct md_sysfs_entry md_reshape_position =
4089 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4090 reshape_position_store);
4091
4092 static ssize_t
4093 array_size_show(mddev_t *mddev, char *page)
4094 {
4095 if (mddev->external_size)
4096 return sprintf(page, "%llu\n",
4097 (unsigned long long)mddev->array_sectors/2);
4098 else
4099 return sprintf(page, "default\n");
4100 }
4101
4102 static ssize_t
4103 array_size_store(mddev_t *mddev, const char *buf, size_t len)
4104 {
4105 sector_t sectors;
4106
4107 if (strncmp(buf, "default", 7) == 0) {
4108 if (mddev->pers)
4109 sectors = mddev->pers->size(mddev, 0, 0);
4110 else
4111 sectors = mddev->array_sectors;
4112
4113 mddev->external_size = 0;
4114 } else {
4115 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4116 return -EINVAL;
4117 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4118 return -E2BIG;
4119
4120 mddev->external_size = 1;
4121 }
4122
4123 mddev->array_sectors = sectors;
4124 set_capacity(mddev->gendisk, mddev->array_sectors);
4125 if (mddev->pers)
4126 revalidate_disk(mddev->gendisk);
4127
4128 return len;
4129 }
4130
4131 static struct md_sysfs_entry md_array_size =
4132 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4133 array_size_store);
4134
4135 static struct attribute *md_default_attrs[] = {
4136 &md_level.attr,
4137 &md_layout.attr,
4138 &md_raid_disks.attr,
4139 &md_chunk_size.attr,
4140 &md_size.attr,
4141 &md_resync_start.attr,
4142 &md_metadata.attr,
4143 &md_new_device.attr,
4144 &md_safe_delay.attr,
4145 &md_array_state.attr,
4146 &md_reshape_position.attr,
4147 &md_array_size.attr,
4148 &max_corr_read_errors.attr,
4149 NULL,
4150 };
4151
4152 static struct attribute *md_redundancy_attrs[] = {
4153 &md_scan_mode.attr,
4154 &md_mismatches.attr,
4155 &md_sync_min.attr,
4156 &md_sync_max.attr,
4157 &md_sync_speed.attr,
4158 &md_sync_force_parallel.attr,
4159 &md_sync_completed.attr,
4160 &md_min_sync.attr,
4161 &md_max_sync.attr,
4162 &md_suspend_lo.attr,
4163 &md_suspend_hi.attr,
4164 &md_bitmap.attr,
4165 &md_degraded.attr,
4166 NULL,
4167 };
4168 static struct attribute_group md_redundancy_group = {
4169 .name = NULL,
4170 .attrs = md_redundancy_attrs,
4171 };
4172
4173
4174 static ssize_t
4175 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4176 {
4177 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4178 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4179 ssize_t rv;
4180
4181 if (!entry->show)
4182 return -EIO;
4183 rv = mddev_lock(mddev);
4184 if (!rv) {
4185 rv = entry->show(mddev, page);
4186 mddev_unlock(mddev);
4187 }
4188 return rv;
4189 }
4190
4191 static ssize_t
4192 md_attr_store(struct kobject *kobj, struct attribute *attr,
4193 const char *page, size_t length)
4194 {
4195 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4196 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4197 ssize_t rv;
4198
4199 if (!entry->store)
4200 return -EIO;
4201 if (!capable(CAP_SYS_ADMIN))
4202 return -EACCES;
4203 rv = mddev_lock(mddev);
4204 if (mddev->hold_active == UNTIL_IOCTL)
4205 mddev->hold_active = 0;
4206 if (!rv) {
4207 rv = entry->store(mddev, page, length);
4208 mddev_unlock(mddev);
4209 }
4210 return rv;
4211 }
4212
4213 static void md_free(struct kobject *ko)
4214 {
4215 mddev_t *mddev = container_of(ko, mddev_t, kobj);
4216
4217 if (mddev->sysfs_state)
4218 sysfs_put(mddev->sysfs_state);
4219
4220 if (mddev->gendisk) {
4221 del_gendisk(mddev->gendisk);
4222 put_disk(mddev->gendisk);
4223 }
4224 if (mddev->queue)
4225 blk_cleanup_queue(mddev->queue);
4226
4227 kfree(mddev);
4228 }
4229
4230 static const struct sysfs_ops md_sysfs_ops = {
4231 .show = md_attr_show,
4232 .store = md_attr_store,
4233 };
4234 static struct kobj_type md_ktype = {
4235 .release = md_free,
4236 .sysfs_ops = &md_sysfs_ops,
4237 .default_attrs = md_default_attrs,
4238 };
4239
4240 int mdp_major = 0;
4241
4242 static void mddev_delayed_delete(struct work_struct *ws)
4243 {
4244 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4245
4246 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4247 kobject_del(&mddev->kobj);
4248 kobject_put(&mddev->kobj);
4249 }
4250
4251 static int md_alloc(dev_t dev, char *name)
4252 {
4253 static DEFINE_MUTEX(disks_mutex);
4254 mddev_t *mddev = mddev_find(dev);
4255 struct gendisk *disk;
4256 int partitioned;
4257 int shift;
4258 int unit;
4259 int error;
4260
4261 if (!mddev)
4262 return -ENODEV;
4263
4264 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4265 shift = partitioned ? MdpMinorShift : 0;
4266 unit = MINOR(mddev->unit) >> shift;
4267
4268 /* wait for any previous instance if this device
4269 * to be completed removed (mddev_delayed_delete).
4270 */
4271 flush_scheduled_work();
4272
4273 mutex_lock(&disks_mutex);
4274 error = -EEXIST;
4275 if (mddev->gendisk)
4276 goto abort;
4277
4278 if (name) {
4279 /* Need to ensure that 'name' is not a duplicate.
4280 */
4281 mddev_t *mddev2;
4282 spin_lock(&all_mddevs_lock);
4283
4284 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4285 if (mddev2->gendisk &&
4286 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4287 spin_unlock(&all_mddevs_lock);
4288 goto abort;
4289 }
4290 spin_unlock(&all_mddevs_lock);
4291 }
4292
4293 error = -ENOMEM;
4294 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4295 if (!mddev->queue)
4296 goto abort;
4297 mddev->queue->queuedata = mddev;
4298
4299 blk_queue_make_request(mddev->queue, md_make_request);
4300
4301 disk = alloc_disk(1 << shift);
4302 if (!disk) {
4303 blk_cleanup_queue(mddev->queue);
4304 mddev->queue = NULL;
4305 goto abort;
4306 }
4307 disk->major = MAJOR(mddev->unit);
4308 disk->first_minor = unit << shift;
4309 if (name)
4310 strcpy(disk->disk_name, name);
4311 else if (partitioned)
4312 sprintf(disk->disk_name, "md_d%d", unit);
4313 else
4314 sprintf(disk->disk_name, "md%d", unit);
4315 disk->fops = &md_fops;
4316 disk->private_data = mddev;
4317 disk->queue = mddev->queue;
4318 /* Allow extended partitions. This makes the
4319 * 'mdp' device redundant, but we can't really
4320 * remove it now.
4321 */
4322 disk->flags |= GENHD_FL_EXT_DEVT;
4323 add_disk(disk);
4324 mddev->gendisk = disk;
4325 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4326 &disk_to_dev(disk)->kobj, "%s", "md");
4327 if (error) {
4328 /* This isn't possible, but as kobject_init_and_add is marked
4329 * __must_check, we must do something with the result
4330 */
4331 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4332 disk->disk_name);
4333 error = 0;
4334 }
4335 if (mddev->kobj.sd &&
4336 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4337 printk(KERN_DEBUG "pointless warning\n");
4338 abort:
4339 mutex_unlock(&disks_mutex);
4340 if (!error && mddev->kobj.sd) {
4341 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4342 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4343 }
4344 mddev_put(mddev);
4345 return error;
4346 }
4347
4348 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4349 {
4350 md_alloc(dev, NULL);
4351 return NULL;
4352 }
4353
4354 static int add_named_array(const char *val, struct kernel_param *kp)
4355 {
4356 /* val must be "md_*" where * is not all digits.
4357 * We allocate an array with a large free minor number, and
4358 * set the name to val. val must not already be an active name.
4359 */
4360 int len = strlen(val);
4361 char buf[DISK_NAME_LEN];
4362
4363 while (len && val[len-1] == '\n')
4364 len--;
4365 if (len >= DISK_NAME_LEN)
4366 return -E2BIG;
4367 strlcpy(buf, val, len+1);
4368 if (strncmp(buf, "md_", 3) != 0)
4369 return -EINVAL;
4370 return md_alloc(0, buf);
4371 }
4372
4373 static void md_safemode_timeout(unsigned long data)
4374 {
4375 mddev_t *mddev = (mddev_t *) data;
4376
4377 if (!atomic_read(&mddev->writes_pending)) {
4378 mddev->safemode = 1;
4379 if (mddev->external)
4380 sysfs_notify_dirent_safe(mddev->sysfs_state);
4381 }
4382 md_wakeup_thread(mddev->thread);
4383 }
4384
4385 static int start_dirty_degraded;
4386
4387 int md_run(mddev_t *mddev)
4388 {
4389 int err;
4390 mdk_rdev_t *rdev;
4391 struct mdk_personality *pers;
4392
4393 if (list_empty(&mddev->disks))
4394 /* cannot run an array with no devices.. */
4395 return -EINVAL;
4396
4397 if (mddev->pers)
4398 return -EBUSY;
4399 /* Cannot run until previous stop completes properly */
4400 if (mddev->sysfs_active)
4401 return -EBUSY;
4402
4403 /*
4404 * Analyze all RAID superblock(s)
4405 */
4406 if (!mddev->raid_disks) {
4407 if (!mddev->persistent)
4408 return -EINVAL;
4409 analyze_sbs(mddev);
4410 }
4411
4412 if (mddev->level != LEVEL_NONE)
4413 request_module("md-level-%d", mddev->level);
4414 else if (mddev->clevel[0])
4415 request_module("md-%s", mddev->clevel);
4416
4417 /*
4418 * Drop all container device buffers, from now on
4419 * the only valid external interface is through the md
4420 * device.
4421 */
4422 list_for_each_entry(rdev, &mddev->disks, same_set) {
4423 if (test_bit(Faulty, &rdev->flags))
4424 continue;
4425 sync_blockdev(rdev->bdev);
4426 invalidate_bdev(rdev->bdev);
4427
4428 /* perform some consistency tests on the device.
4429 * We don't want the data to overlap the metadata,
4430 * Internal Bitmap issues have been handled elsewhere.
4431 */
4432 if (rdev->data_offset < rdev->sb_start) {
4433 if (mddev->dev_sectors &&
4434 rdev->data_offset + mddev->dev_sectors
4435 > rdev->sb_start) {
4436 printk("md: %s: data overlaps metadata\n",
4437 mdname(mddev));
4438 return -EINVAL;
4439 }
4440 } else {
4441 if (rdev->sb_start + rdev->sb_size/512
4442 > rdev->data_offset) {
4443 printk("md: %s: metadata overlaps data\n",
4444 mdname(mddev));
4445 return -EINVAL;
4446 }
4447 }
4448 sysfs_notify_dirent_safe(rdev->sysfs_state);
4449 }
4450
4451 spin_lock(&pers_lock);
4452 pers = find_pers(mddev->level, mddev->clevel);
4453 if (!pers || !try_module_get(pers->owner)) {
4454 spin_unlock(&pers_lock);
4455 if (mddev->level != LEVEL_NONE)
4456 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4457 mddev->level);
4458 else
4459 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4460 mddev->clevel);
4461 return -EINVAL;
4462 }
4463 mddev->pers = pers;
4464 spin_unlock(&pers_lock);
4465 if (mddev->level != pers->level) {
4466 mddev->level = pers->level;
4467 mddev->new_level = pers->level;
4468 }
4469 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4470
4471 if (mddev->reshape_position != MaxSector &&
4472 pers->start_reshape == NULL) {
4473 /* This personality cannot handle reshaping... */
4474 mddev->pers = NULL;
4475 module_put(pers->owner);
4476 return -EINVAL;
4477 }
4478
4479 if (pers->sync_request) {
4480 /* Warn if this is a potentially silly
4481 * configuration.
4482 */
4483 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4484 mdk_rdev_t *rdev2;
4485 int warned = 0;
4486
4487 list_for_each_entry(rdev, &mddev->disks, same_set)
4488 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4489 if (rdev < rdev2 &&
4490 rdev->bdev->bd_contains ==
4491 rdev2->bdev->bd_contains) {
4492 printk(KERN_WARNING
4493 "%s: WARNING: %s appears to be"
4494 " on the same physical disk as"
4495 " %s.\n",
4496 mdname(mddev),
4497 bdevname(rdev->bdev,b),
4498 bdevname(rdev2->bdev,b2));
4499 warned = 1;
4500 }
4501 }
4502
4503 if (warned)
4504 printk(KERN_WARNING
4505 "True protection against single-disk"
4506 " failure might be compromised.\n");
4507 }
4508
4509 mddev->recovery = 0;
4510 /* may be over-ridden by personality */
4511 mddev->resync_max_sectors = mddev->dev_sectors;
4512
4513 mddev->barriers_work = 1;
4514 mddev->ok_start_degraded = start_dirty_degraded;
4515
4516 if (start_readonly && mddev->ro == 0)
4517 mddev->ro = 2; /* read-only, but switch on first write */
4518
4519 err = mddev->pers->run(mddev);
4520 if (err)
4521 printk(KERN_ERR "md: pers->run() failed ...\n");
4522 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4523 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4524 " but 'external_size' not in effect?\n", __func__);
4525 printk(KERN_ERR
4526 "md: invalid array_size %llu > default size %llu\n",
4527 (unsigned long long)mddev->array_sectors / 2,
4528 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4529 err = -EINVAL;
4530 mddev->pers->stop(mddev);
4531 }
4532 if (err == 0 && mddev->pers->sync_request) {
4533 err = bitmap_create(mddev);
4534 if (err) {
4535 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4536 mdname(mddev), err);
4537 mddev->pers->stop(mddev);
4538 }
4539 }
4540 if (err) {
4541 module_put(mddev->pers->owner);
4542 mddev->pers = NULL;
4543 bitmap_destroy(mddev);
4544 return err;
4545 }
4546 if (mddev->pers->sync_request) {
4547 if (mddev->kobj.sd &&
4548 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4549 printk(KERN_WARNING
4550 "md: cannot register extra attributes for %s\n",
4551 mdname(mddev));
4552 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4553 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4554 mddev->ro = 0;
4555
4556 atomic_set(&mddev->writes_pending,0);
4557 atomic_set(&mddev->max_corr_read_errors,
4558 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4559 mddev->safemode = 0;
4560 mddev->safemode_timer.function = md_safemode_timeout;
4561 mddev->safemode_timer.data = (unsigned long) mddev;
4562 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4563 mddev->in_sync = 1;
4564 smp_wmb();
4565 mddev->ready = 1;
4566 list_for_each_entry(rdev, &mddev->disks, same_set)
4567 if (rdev->raid_disk >= 0) {
4568 char nm[20];
4569 sprintf(nm, "rd%d", rdev->raid_disk);
4570 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4571 /* failure here is OK */;
4572 }
4573
4574 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4575
4576 if (mddev->flags)
4577 md_update_sb(mddev, 0);
4578
4579 md_wakeup_thread(mddev->thread);
4580 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4581
4582 md_new_event(mddev);
4583 sysfs_notify_dirent_safe(mddev->sysfs_state);
4584 sysfs_notify_dirent_safe(mddev->sysfs_action);
4585 sysfs_notify(&mddev->kobj, NULL, "degraded");
4586 return 0;
4587 }
4588 EXPORT_SYMBOL_GPL(md_run);
4589
4590 static int do_md_run(mddev_t *mddev)
4591 {
4592 int err;
4593
4594 err = md_run(mddev);
4595 if (err)
4596 goto out;
4597 err = bitmap_load(mddev);
4598 if (err) {
4599 bitmap_destroy(mddev);
4600 goto out;
4601 }
4602 set_capacity(mddev->gendisk, mddev->array_sectors);
4603 revalidate_disk(mddev->gendisk);
4604 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4605 out:
4606 return err;
4607 }
4608
4609 static int restart_array(mddev_t *mddev)
4610 {
4611 struct gendisk *disk = mddev->gendisk;
4612
4613 /* Complain if it has no devices */
4614 if (list_empty(&mddev->disks))
4615 return -ENXIO;
4616 if (!mddev->pers)
4617 return -EINVAL;
4618 if (!mddev->ro)
4619 return -EBUSY;
4620 mddev->safemode = 0;
4621 mddev->ro = 0;
4622 set_disk_ro(disk, 0);
4623 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4624 mdname(mddev));
4625 /* Kick recovery or resync if necessary */
4626 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4627 md_wakeup_thread(mddev->thread);
4628 md_wakeup_thread(mddev->sync_thread);
4629 sysfs_notify_dirent_safe(mddev->sysfs_state);
4630 return 0;
4631 }
4632
4633 /* similar to deny_write_access, but accounts for our holding a reference
4634 * to the file ourselves */
4635 static int deny_bitmap_write_access(struct file * file)
4636 {
4637 struct inode *inode = file->f_mapping->host;
4638
4639 spin_lock(&inode->i_lock);
4640 if (atomic_read(&inode->i_writecount) > 1) {
4641 spin_unlock(&inode->i_lock);
4642 return -ETXTBSY;
4643 }
4644 atomic_set(&inode->i_writecount, -1);
4645 spin_unlock(&inode->i_lock);
4646
4647 return 0;
4648 }
4649
4650 void restore_bitmap_write_access(struct file *file)
4651 {
4652 struct inode *inode = file->f_mapping->host;
4653
4654 spin_lock(&inode->i_lock);
4655 atomic_set(&inode->i_writecount, 1);
4656 spin_unlock(&inode->i_lock);
4657 }
4658
4659 static void md_clean(mddev_t *mddev)
4660 {
4661 mddev->array_sectors = 0;
4662 mddev->external_size = 0;
4663 mddev->dev_sectors = 0;
4664 mddev->raid_disks = 0;
4665 mddev->recovery_cp = 0;
4666 mddev->resync_min = 0;
4667 mddev->resync_max = MaxSector;
4668 mddev->reshape_position = MaxSector;
4669 mddev->external = 0;
4670 mddev->persistent = 0;
4671 mddev->level = LEVEL_NONE;
4672 mddev->clevel[0] = 0;
4673 mddev->flags = 0;
4674 mddev->ro = 0;
4675 mddev->metadata_type[0] = 0;
4676 mddev->chunk_sectors = 0;
4677 mddev->ctime = mddev->utime = 0;
4678 mddev->layout = 0;
4679 mddev->max_disks = 0;
4680 mddev->events = 0;
4681 mddev->can_decrease_events = 0;
4682 mddev->delta_disks = 0;
4683 mddev->new_level = LEVEL_NONE;
4684 mddev->new_layout = 0;
4685 mddev->new_chunk_sectors = 0;
4686 mddev->curr_resync = 0;
4687 mddev->resync_mismatches = 0;
4688 mddev->suspend_lo = mddev->suspend_hi = 0;
4689 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4690 mddev->recovery = 0;
4691 mddev->in_sync = 0;
4692 mddev->degraded = 0;
4693 mddev->barriers_work = 0;
4694 mddev->safemode = 0;
4695 mddev->bitmap_info.offset = 0;
4696 mddev->bitmap_info.default_offset = 0;
4697 mddev->bitmap_info.chunksize = 0;
4698 mddev->bitmap_info.daemon_sleep = 0;
4699 mddev->bitmap_info.max_write_behind = 0;
4700 mddev->plug = NULL;
4701 }
4702
4703 void md_stop_writes(mddev_t *mddev)
4704 {
4705 if (mddev->sync_thread) {
4706 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4707 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4708 md_unregister_thread(mddev->sync_thread);
4709 mddev->sync_thread = NULL;
4710 }
4711
4712 del_timer_sync(&mddev->safemode_timer);
4713
4714 bitmap_flush(mddev);
4715 md_super_wait(mddev);
4716
4717 if (!mddev->in_sync || mddev->flags) {
4718 /* mark array as shutdown cleanly */
4719 mddev->in_sync = 1;
4720 md_update_sb(mddev, 1);
4721 }
4722 }
4723 EXPORT_SYMBOL_GPL(md_stop_writes);
4724
4725 void md_stop(mddev_t *mddev)
4726 {
4727 mddev->ready = 0;
4728 mddev->pers->stop(mddev);
4729 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4730 mddev->to_remove = &md_redundancy_group;
4731 module_put(mddev->pers->owner);
4732 mddev->pers = NULL;
4733 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4734 }
4735 EXPORT_SYMBOL_GPL(md_stop);
4736
4737 static int md_set_readonly(mddev_t *mddev, int is_open)
4738 {
4739 int err = 0;
4740 mutex_lock(&mddev->open_mutex);
4741 if (atomic_read(&mddev->openers) > is_open) {
4742 printk("md: %s still in use.\n",mdname(mddev));
4743 err = -EBUSY;
4744 goto out;
4745 }
4746 if (mddev->pers) {
4747 md_stop_writes(mddev);
4748
4749 err = -ENXIO;
4750 if (mddev->ro==1)
4751 goto out;
4752 mddev->ro = 1;
4753 set_disk_ro(mddev->gendisk, 1);
4754 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4755 sysfs_notify_dirent_safe(mddev->sysfs_state);
4756 err = 0;
4757 }
4758 out:
4759 mutex_unlock(&mddev->open_mutex);
4760 return err;
4761 }
4762
4763 /* mode:
4764 * 0 - completely stop and dis-assemble array
4765 * 2 - stop but do not disassemble array
4766 */
4767 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4768 {
4769 struct gendisk *disk = mddev->gendisk;
4770 mdk_rdev_t *rdev;
4771
4772 mutex_lock(&mddev->open_mutex);
4773 if (atomic_read(&mddev->openers) > is_open ||
4774 mddev->sysfs_active) {
4775 printk("md: %s still in use.\n",mdname(mddev));
4776 mutex_unlock(&mddev->open_mutex);
4777 return -EBUSY;
4778 }
4779
4780 if (mddev->pers) {
4781 if (mddev->ro)
4782 set_disk_ro(disk, 0);
4783
4784 md_stop_writes(mddev);
4785 md_stop(mddev);
4786 mddev->queue->merge_bvec_fn = NULL;
4787 mddev->queue->unplug_fn = NULL;
4788 mddev->queue->backing_dev_info.congested_fn = NULL;
4789
4790 /* tell userspace to handle 'inactive' */
4791 sysfs_notify_dirent_safe(mddev->sysfs_state);
4792
4793 list_for_each_entry(rdev, &mddev->disks, same_set)
4794 if (rdev->raid_disk >= 0) {
4795 char nm[20];
4796 sprintf(nm, "rd%d", rdev->raid_disk);
4797 sysfs_remove_link(&mddev->kobj, nm);
4798 }
4799
4800 set_capacity(disk, 0);
4801 mutex_unlock(&mddev->open_mutex);
4802 revalidate_disk(disk);
4803
4804 if (mddev->ro)
4805 mddev->ro = 0;
4806 } else
4807 mutex_unlock(&mddev->open_mutex);
4808 /*
4809 * Free resources if final stop
4810 */
4811 if (mode == 0) {
4812 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4813
4814 bitmap_destroy(mddev);
4815 if (mddev->bitmap_info.file) {
4816 restore_bitmap_write_access(mddev->bitmap_info.file);
4817 fput(mddev->bitmap_info.file);
4818 mddev->bitmap_info.file = NULL;
4819 }
4820 mddev->bitmap_info.offset = 0;
4821
4822 export_array(mddev);
4823
4824 md_clean(mddev);
4825 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4826 if (mddev->hold_active == UNTIL_STOP)
4827 mddev->hold_active = 0;
4828 }
4829 blk_integrity_unregister(disk);
4830 md_new_event(mddev);
4831 sysfs_notify_dirent_safe(mddev->sysfs_state);
4832 return 0;
4833 }
4834
4835 #ifndef MODULE
4836 static void autorun_array(mddev_t *mddev)
4837 {
4838 mdk_rdev_t *rdev;
4839 int err;
4840
4841 if (list_empty(&mddev->disks))
4842 return;
4843
4844 printk(KERN_INFO "md: running: ");
4845
4846 list_for_each_entry(rdev, &mddev->disks, same_set) {
4847 char b[BDEVNAME_SIZE];
4848 printk("<%s>", bdevname(rdev->bdev,b));
4849 }
4850 printk("\n");
4851
4852 err = do_md_run(mddev);
4853 if (err) {
4854 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4855 do_md_stop(mddev, 0, 0);
4856 }
4857 }
4858
4859 /*
4860 * lets try to run arrays based on all disks that have arrived
4861 * until now. (those are in pending_raid_disks)
4862 *
4863 * the method: pick the first pending disk, collect all disks with
4864 * the same UUID, remove all from the pending list and put them into
4865 * the 'same_array' list. Then order this list based on superblock
4866 * update time (freshest comes first), kick out 'old' disks and
4867 * compare superblocks. If everything's fine then run it.
4868 *
4869 * If "unit" is allocated, then bump its reference count
4870 */
4871 static void autorun_devices(int part)
4872 {
4873 mdk_rdev_t *rdev0, *rdev, *tmp;
4874 mddev_t *mddev;
4875 char b[BDEVNAME_SIZE];
4876
4877 printk(KERN_INFO "md: autorun ...\n");
4878 while (!list_empty(&pending_raid_disks)) {
4879 int unit;
4880 dev_t dev;
4881 LIST_HEAD(candidates);
4882 rdev0 = list_entry(pending_raid_disks.next,
4883 mdk_rdev_t, same_set);
4884
4885 printk(KERN_INFO "md: considering %s ...\n",
4886 bdevname(rdev0->bdev,b));
4887 INIT_LIST_HEAD(&candidates);
4888 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4889 if (super_90_load(rdev, rdev0, 0) >= 0) {
4890 printk(KERN_INFO "md: adding %s ...\n",
4891 bdevname(rdev->bdev,b));
4892 list_move(&rdev->same_set, &candidates);
4893 }
4894 /*
4895 * now we have a set of devices, with all of them having
4896 * mostly sane superblocks. It's time to allocate the
4897 * mddev.
4898 */
4899 if (part) {
4900 dev = MKDEV(mdp_major,
4901 rdev0->preferred_minor << MdpMinorShift);
4902 unit = MINOR(dev) >> MdpMinorShift;
4903 } else {
4904 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4905 unit = MINOR(dev);
4906 }
4907 if (rdev0->preferred_minor != unit) {
4908 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4909 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4910 break;
4911 }
4912
4913 md_probe(dev, NULL, NULL);
4914 mddev = mddev_find(dev);
4915 if (!mddev || !mddev->gendisk) {
4916 if (mddev)
4917 mddev_put(mddev);
4918 printk(KERN_ERR
4919 "md: cannot allocate memory for md drive.\n");
4920 break;
4921 }
4922 if (mddev_lock(mddev))
4923 printk(KERN_WARNING "md: %s locked, cannot run\n",
4924 mdname(mddev));
4925 else if (mddev->raid_disks || mddev->major_version
4926 || !list_empty(&mddev->disks)) {
4927 printk(KERN_WARNING
4928 "md: %s already running, cannot run %s\n",
4929 mdname(mddev), bdevname(rdev0->bdev,b));
4930 mddev_unlock(mddev);
4931 } else {
4932 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4933 mddev->persistent = 1;
4934 rdev_for_each_list(rdev, tmp, &candidates) {
4935 list_del_init(&rdev->same_set);
4936 if (bind_rdev_to_array(rdev, mddev))
4937 export_rdev(rdev);
4938 }
4939 autorun_array(mddev);
4940 mddev_unlock(mddev);
4941 }
4942 /* on success, candidates will be empty, on error
4943 * it won't...
4944 */
4945 rdev_for_each_list(rdev, tmp, &candidates) {
4946 list_del_init(&rdev->same_set);
4947 export_rdev(rdev);
4948 }
4949 mddev_put(mddev);
4950 }
4951 printk(KERN_INFO "md: ... autorun DONE.\n");
4952 }
4953 #endif /* !MODULE */
4954
4955 static int get_version(void __user * arg)
4956 {
4957 mdu_version_t ver;
4958
4959 ver.major = MD_MAJOR_VERSION;
4960 ver.minor = MD_MINOR_VERSION;
4961 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4962
4963 if (copy_to_user(arg, &ver, sizeof(ver)))
4964 return -EFAULT;
4965
4966 return 0;
4967 }
4968
4969 static int get_array_info(mddev_t * mddev, void __user * arg)
4970 {
4971 mdu_array_info_t info;
4972 int nr,working,insync,failed,spare;
4973 mdk_rdev_t *rdev;
4974
4975 nr=working=insync=failed=spare=0;
4976 list_for_each_entry(rdev, &mddev->disks, same_set) {
4977 nr++;
4978 if (test_bit(Faulty, &rdev->flags))
4979 failed++;
4980 else {
4981 working++;
4982 if (test_bit(In_sync, &rdev->flags))
4983 insync++;
4984 else
4985 spare++;
4986 }
4987 }
4988
4989 info.major_version = mddev->major_version;
4990 info.minor_version = mddev->minor_version;
4991 info.patch_version = MD_PATCHLEVEL_VERSION;
4992 info.ctime = mddev->ctime;
4993 info.level = mddev->level;
4994 info.size = mddev->dev_sectors / 2;
4995 if (info.size != mddev->dev_sectors / 2) /* overflow */
4996 info.size = -1;
4997 info.nr_disks = nr;
4998 info.raid_disks = mddev->raid_disks;
4999 info.md_minor = mddev->md_minor;
5000 info.not_persistent= !mddev->persistent;
5001
5002 info.utime = mddev->utime;
5003 info.state = 0;
5004 if (mddev->in_sync)
5005 info.state = (1<<MD_SB_CLEAN);
5006 if (mddev->bitmap && mddev->bitmap_info.offset)
5007 info.state = (1<<MD_SB_BITMAP_PRESENT);
5008 info.active_disks = insync;
5009 info.working_disks = working;
5010 info.failed_disks = failed;
5011 info.spare_disks = spare;
5012
5013 info.layout = mddev->layout;
5014 info.chunk_size = mddev->chunk_sectors << 9;
5015
5016 if (copy_to_user(arg, &info, sizeof(info)))
5017 return -EFAULT;
5018
5019 return 0;
5020 }
5021
5022 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5023 {
5024 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5025 char *ptr, *buf = NULL;
5026 int err = -ENOMEM;
5027
5028 if (md_allow_write(mddev))
5029 file = kmalloc(sizeof(*file), GFP_NOIO);
5030 else
5031 file = kmalloc(sizeof(*file), GFP_KERNEL);
5032
5033 if (!file)
5034 goto out;
5035
5036 /* bitmap disabled, zero the first byte and copy out */
5037 if (!mddev->bitmap || !mddev->bitmap->file) {
5038 file->pathname[0] = '\0';
5039 goto copy_out;
5040 }
5041
5042 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5043 if (!buf)
5044 goto out;
5045
5046 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5047 if (IS_ERR(ptr))
5048 goto out;
5049
5050 strcpy(file->pathname, ptr);
5051
5052 copy_out:
5053 err = 0;
5054 if (copy_to_user(arg, file, sizeof(*file)))
5055 err = -EFAULT;
5056 out:
5057 kfree(buf);
5058 kfree(file);
5059 return err;
5060 }
5061
5062 static int get_disk_info(mddev_t * mddev, void __user * arg)
5063 {
5064 mdu_disk_info_t info;
5065 mdk_rdev_t *rdev;
5066
5067 if (copy_from_user(&info, arg, sizeof(info)))
5068 return -EFAULT;
5069
5070 rdev = find_rdev_nr(mddev, info.number);
5071 if (rdev) {
5072 info.major = MAJOR(rdev->bdev->bd_dev);
5073 info.minor = MINOR(rdev->bdev->bd_dev);
5074 info.raid_disk = rdev->raid_disk;
5075 info.state = 0;
5076 if (test_bit(Faulty, &rdev->flags))
5077 info.state |= (1<<MD_DISK_FAULTY);
5078 else if (test_bit(In_sync, &rdev->flags)) {
5079 info.state |= (1<<MD_DISK_ACTIVE);
5080 info.state |= (1<<MD_DISK_SYNC);
5081 }
5082 if (test_bit(WriteMostly, &rdev->flags))
5083 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5084 } else {
5085 info.major = info.minor = 0;
5086 info.raid_disk = -1;
5087 info.state = (1<<MD_DISK_REMOVED);
5088 }
5089
5090 if (copy_to_user(arg, &info, sizeof(info)))
5091 return -EFAULT;
5092
5093 return 0;
5094 }
5095
5096 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5097 {
5098 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5099 mdk_rdev_t *rdev;
5100 dev_t dev = MKDEV(info->major,info->minor);
5101
5102 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5103 return -EOVERFLOW;
5104
5105 if (!mddev->raid_disks) {
5106 int err;
5107 /* expecting a device which has a superblock */
5108 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5109 if (IS_ERR(rdev)) {
5110 printk(KERN_WARNING
5111 "md: md_import_device returned %ld\n",
5112 PTR_ERR(rdev));
5113 return PTR_ERR(rdev);
5114 }
5115 if (!list_empty(&mddev->disks)) {
5116 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5117 mdk_rdev_t, same_set);
5118 err = super_types[mddev->major_version]
5119 .load_super(rdev, rdev0, mddev->minor_version);
5120 if (err < 0) {
5121 printk(KERN_WARNING
5122 "md: %s has different UUID to %s\n",
5123 bdevname(rdev->bdev,b),
5124 bdevname(rdev0->bdev,b2));
5125 export_rdev(rdev);
5126 return -EINVAL;
5127 }
5128 }
5129 err = bind_rdev_to_array(rdev, mddev);
5130 if (err)
5131 export_rdev(rdev);
5132 return err;
5133 }
5134
5135 /*
5136 * add_new_disk can be used once the array is assembled
5137 * to add "hot spares". They must already have a superblock
5138 * written
5139 */
5140 if (mddev->pers) {
5141 int err;
5142 if (!mddev->pers->hot_add_disk) {
5143 printk(KERN_WARNING
5144 "%s: personality does not support diskops!\n",
5145 mdname(mddev));
5146 return -EINVAL;
5147 }
5148 if (mddev->persistent)
5149 rdev = md_import_device(dev, mddev->major_version,
5150 mddev->minor_version);
5151 else
5152 rdev = md_import_device(dev, -1, -1);
5153 if (IS_ERR(rdev)) {
5154 printk(KERN_WARNING
5155 "md: md_import_device returned %ld\n",
5156 PTR_ERR(rdev));
5157 return PTR_ERR(rdev);
5158 }
5159 /* set saved_raid_disk if appropriate */
5160 if (!mddev->persistent) {
5161 if (info->state & (1<<MD_DISK_SYNC) &&
5162 info->raid_disk < mddev->raid_disks) {
5163 rdev->raid_disk = info->raid_disk;
5164 set_bit(In_sync, &rdev->flags);
5165 } else
5166 rdev->raid_disk = -1;
5167 } else
5168 super_types[mddev->major_version].
5169 validate_super(mddev, rdev);
5170 if (test_bit(In_sync, &rdev->flags))
5171 rdev->saved_raid_disk = rdev->raid_disk;
5172 else
5173 rdev->saved_raid_disk = -1;
5174
5175 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5176 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5177 set_bit(WriteMostly, &rdev->flags);
5178 else
5179 clear_bit(WriteMostly, &rdev->flags);
5180
5181 rdev->raid_disk = -1;
5182 err = bind_rdev_to_array(rdev, mddev);
5183 if (!err && !mddev->pers->hot_remove_disk) {
5184 /* If there is hot_add_disk but no hot_remove_disk
5185 * then added disks for geometry changes,
5186 * and should be added immediately.
5187 */
5188 super_types[mddev->major_version].
5189 validate_super(mddev, rdev);
5190 err = mddev->pers->hot_add_disk(mddev, rdev);
5191 if (err)
5192 unbind_rdev_from_array(rdev);
5193 }
5194 if (err)
5195 export_rdev(rdev);
5196 else
5197 sysfs_notify_dirent_safe(rdev->sysfs_state);
5198
5199 md_update_sb(mddev, 1);
5200 if (mddev->degraded)
5201 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5202 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5203 md_wakeup_thread(mddev->thread);
5204 return err;
5205 }
5206
5207 /* otherwise, add_new_disk is only allowed
5208 * for major_version==0 superblocks
5209 */
5210 if (mddev->major_version != 0) {
5211 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5212 mdname(mddev));
5213 return -EINVAL;
5214 }
5215
5216 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5217 int err;
5218 rdev = md_import_device(dev, -1, 0);
5219 if (IS_ERR(rdev)) {
5220 printk(KERN_WARNING
5221 "md: error, md_import_device() returned %ld\n",
5222 PTR_ERR(rdev));
5223 return PTR_ERR(rdev);
5224 }
5225 rdev->desc_nr = info->number;
5226 if (info->raid_disk < mddev->raid_disks)
5227 rdev->raid_disk = info->raid_disk;
5228 else
5229 rdev->raid_disk = -1;
5230
5231 if (rdev->raid_disk < mddev->raid_disks)
5232 if (info->state & (1<<MD_DISK_SYNC))
5233 set_bit(In_sync, &rdev->flags);
5234
5235 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5236 set_bit(WriteMostly, &rdev->flags);
5237
5238 if (!mddev->persistent) {
5239 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5240 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5241 } else
5242 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5243 rdev->sectors = rdev->sb_start;
5244
5245 err = bind_rdev_to_array(rdev, mddev);
5246 if (err) {
5247 export_rdev(rdev);
5248 return err;
5249 }
5250 }
5251
5252 return 0;
5253 }
5254
5255 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5256 {
5257 char b[BDEVNAME_SIZE];
5258 mdk_rdev_t *rdev;
5259
5260 rdev = find_rdev(mddev, dev);
5261 if (!rdev)
5262 return -ENXIO;
5263
5264 if (rdev->raid_disk >= 0)
5265 goto busy;
5266
5267 kick_rdev_from_array(rdev);
5268 md_update_sb(mddev, 1);
5269 md_new_event(mddev);
5270
5271 return 0;
5272 busy:
5273 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5274 bdevname(rdev->bdev,b), mdname(mddev));
5275 return -EBUSY;
5276 }
5277
5278 static int hot_add_disk(mddev_t * mddev, dev_t dev)
5279 {
5280 char b[BDEVNAME_SIZE];
5281 int err;
5282 mdk_rdev_t *rdev;
5283
5284 if (!mddev->pers)
5285 return -ENODEV;
5286
5287 if (mddev->major_version != 0) {
5288 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5289 " version-0 superblocks.\n",
5290 mdname(mddev));
5291 return -EINVAL;
5292 }
5293 if (!mddev->pers->hot_add_disk) {
5294 printk(KERN_WARNING
5295 "%s: personality does not support diskops!\n",
5296 mdname(mddev));
5297 return -EINVAL;
5298 }
5299
5300 rdev = md_import_device(dev, -1, 0);
5301 if (IS_ERR(rdev)) {
5302 printk(KERN_WARNING
5303 "md: error, md_import_device() returned %ld\n",
5304 PTR_ERR(rdev));
5305 return -EINVAL;
5306 }
5307
5308 if (mddev->persistent)
5309 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
5310 else
5311 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
5312
5313 rdev->sectors = rdev->sb_start;
5314
5315 if (test_bit(Faulty, &rdev->flags)) {
5316 printk(KERN_WARNING
5317 "md: can not hot-add faulty %s disk to %s!\n",
5318 bdevname(rdev->bdev,b), mdname(mddev));
5319 err = -EINVAL;
5320 goto abort_export;
5321 }
5322 clear_bit(In_sync, &rdev->flags);
5323 rdev->desc_nr = -1;
5324 rdev->saved_raid_disk = -1;
5325 err = bind_rdev_to_array(rdev, mddev);
5326 if (err)
5327 goto abort_export;
5328
5329 /*
5330 * The rest should better be atomic, we can have disk failures
5331 * noticed in interrupt contexts ...
5332 */
5333
5334 rdev->raid_disk = -1;
5335
5336 md_update_sb(mddev, 1);
5337
5338 /*
5339 * Kick recovery, maybe this spare has to be added to the
5340 * array immediately.
5341 */
5342 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5343 md_wakeup_thread(mddev->thread);
5344 md_new_event(mddev);
5345 return 0;
5346
5347 abort_export:
5348 export_rdev(rdev);
5349 return err;
5350 }
5351
5352 static int set_bitmap_file(mddev_t *mddev, int fd)
5353 {
5354 int err;
5355
5356 if (mddev->pers) {
5357 if (!mddev->pers->quiesce)
5358 return -EBUSY;
5359 if (mddev->recovery || mddev->sync_thread)
5360 return -EBUSY;
5361 /* we should be able to change the bitmap.. */
5362 }
5363
5364
5365 if (fd >= 0) {
5366 if (mddev->bitmap)
5367 return -EEXIST; /* cannot add when bitmap is present */
5368 mddev->bitmap_info.file = fget(fd);
5369
5370 if (mddev->bitmap_info.file == NULL) {
5371 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5372 mdname(mddev));
5373 return -EBADF;
5374 }
5375
5376 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5377 if (err) {
5378 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5379 mdname(mddev));
5380 fput(mddev->bitmap_info.file);
5381 mddev->bitmap_info.file = NULL;
5382 return err;
5383 }
5384 mddev->bitmap_info.offset = 0; /* file overrides offset */
5385 } else if (mddev->bitmap == NULL)
5386 return -ENOENT; /* cannot remove what isn't there */
5387 err = 0;
5388 if (mddev->pers) {
5389 mddev->pers->quiesce(mddev, 1);
5390 if (fd >= 0) {
5391 err = bitmap_create(mddev);
5392 if (!err)
5393 err = bitmap_load(mddev);
5394 }
5395 if (fd < 0 || err) {
5396 bitmap_destroy(mddev);
5397 fd = -1; /* make sure to put the file */
5398 }
5399 mddev->pers->quiesce(mddev, 0);
5400 }
5401 if (fd < 0) {
5402 if (mddev->bitmap_info.file) {
5403 restore_bitmap_write_access(mddev->bitmap_info.file);
5404 fput(mddev->bitmap_info.file);
5405 }
5406 mddev->bitmap_info.file = NULL;
5407 }
5408
5409 return err;
5410 }
5411
5412 /*
5413 * set_array_info is used two different ways
5414 * The original usage is when creating a new array.
5415 * In this usage, raid_disks is > 0 and it together with
5416 * level, size, not_persistent,layout,chunksize determine the
5417 * shape of the array.
5418 * This will always create an array with a type-0.90.0 superblock.
5419 * The newer usage is when assembling an array.
5420 * In this case raid_disks will be 0, and the major_version field is
5421 * use to determine which style super-blocks are to be found on the devices.
5422 * The minor and patch _version numbers are also kept incase the
5423 * super_block handler wishes to interpret them.
5424 */
5425 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5426 {
5427
5428 if (info->raid_disks == 0) {
5429 /* just setting version number for superblock loading */
5430 if (info->major_version < 0 ||
5431 info->major_version >= ARRAY_SIZE(super_types) ||
5432 super_types[info->major_version].name == NULL) {
5433 /* maybe try to auto-load a module? */
5434 printk(KERN_INFO
5435 "md: superblock version %d not known\n",
5436 info->major_version);
5437 return -EINVAL;
5438 }
5439 mddev->major_version = info->major_version;
5440 mddev->minor_version = info->minor_version;
5441 mddev->patch_version = info->patch_version;
5442 mddev->persistent = !info->not_persistent;
5443 /* ensure mddev_put doesn't delete this now that there
5444 * is some minimal configuration.
5445 */
5446 mddev->ctime = get_seconds();
5447 return 0;
5448 }
5449 mddev->major_version = MD_MAJOR_VERSION;
5450 mddev->minor_version = MD_MINOR_VERSION;
5451 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5452 mddev->ctime = get_seconds();
5453
5454 mddev->level = info->level;
5455 mddev->clevel[0] = 0;
5456 mddev->dev_sectors = 2 * (sector_t)info->size;
5457 mddev->raid_disks = info->raid_disks;
5458 /* don't set md_minor, it is determined by which /dev/md* was
5459 * openned
5460 */
5461 if (info->state & (1<<MD_SB_CLEAN))
5462 mddev->recovery_cp = MaxSector;
5463 else
5464 mddev->recovery_cp = 0;
5465 mddev->persistent = ! info->not_persistent;
5466 mddev->external = 0;
5467
5468 mddev->layout = info->layout;
5469 mddev->chunk_sectors = info->chunk_size >> 9;
5470
5471 mddev->max_disks = MD_SB_DISKS;
5472
5473 if (mddev->persistent)
5474 mddev->flags = 0;
5475 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5476
5477 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5478 mddev->bitmap_info.offset = 0;
5479
5480 mddev->reshape_position = MaxSector;
5481
5482 /*
5483 * Generate a 128 bit UUID
5484 */
5485 get_random_bytes(mddev->uuid, 16);
5486
5487 mddev->new_level = mddev->level;
5488 mddev->new_chunk_sectors = mddev->chunk_sectors;
5489 mddev->new_layout = mddev->layout;
5490 mddev->delta_disks = 0;
5491
5492 return 0;
5493 }
5494
5495 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5496 {
5497 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5498
5499 if (mddev->external_size)
5500 return;
5501
5502 mddev->array_sectors = array_sectors;
5503 }
5504 EXPORT_SYMBOL(md_set_array_sectors);
5505
5506 static int update_size(mddev_t *mddev, sector_t num_sectors)
5507 {
5508 mdk_rdev_t *rdev;
5509 int rv;
5510 int fit = (num_sectors == 0);
5511
5512 if (mddev->pers->resize == NULL)
5513 return -EINVAL;
5514 /* The "num_sectors" is the number of sectors of each device that
5515 * is used. This can only make sense for arrays with redundancy.
5516 * linear and raid0 always use whatever space is available. We can only
5517 * consider changing this number if no resync or reconstruction is
5518 * happening, and if the new size is acceptable. It must fit before the
5519 * sb_start or, if that is <data_offset, it must fit before the size
5520 * of each device. If num_sectors is zero, we find the largest size
5521 * that fits.
5522
5523 */
5524 if (mddev->sync_thread)
5525 return -EBUSY;
5526 if (mddev->bitmap)
5527 /* Sorry, cannot grow a bitmap yet, just remove it,
5528 * grow, and re-add.
5529 */
5530 return -EBUSY;
5531 list_for_each_entry(rdev, &mddev->disks, same_set) {
5532 sector_t avail = rdev->sectors;
5533
5534 if (fit && (num_sectors == 0 || num_sectors > avail))
5535 num_sectors = avail;
5536 if (avail < num_sectors)
5537 return -ENOSPC;
5538 }
5539 rv = mddev->pers->resize(mddev, num_sectors);
5540 if (!rv)
5541 revalidate_disk(mddev->gendisk);
5542 return rv;
5543 }
5544
5545 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5546 {
5547 int rv;
5548 /* change the number of raid disks */
5549 if (mddev->pers->check_reshape == NULL)
5550 return -EINVAL;
5551 if (raid_disks <= 0 ||
5552 (mddev->max_disks && raid_disks >= mddev->max_disks))
5553 return -EINVAL;
5554 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5555 return -EBUSY;
5556 mddev->delta_disks = raid_disks - mddev->raid_disks;
5557
5558 rv = mddev->pers->check_reshape(mddev);
5559 return rv;
5560 }
5561
5562
5563 /*
5564 * update_array_info is used to change the configuration of an
5565 * on-line array.
5566 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5567 * fields in the info are checked against the array.
5568 * Any differences that cannot be handled will cause an error.
5569 * Normally, only one change can be managed at a time.
5570 */
5571 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5572 {
5573 int rv = 0;
5574 int cnt = 0;
5575 int state = 0;
5576
5577 /* calculate expected state,ignoring low bits */
5578 if (mddev->bitmap && mddev->bitmap_info.offset)
5579 state |= (1 << MD_SB_BITMAP_PRESENT);
5580
5581 if (mddev->major_version != info->major_version ||
5582 mddev->minor_version != info->minor_version ||
5583 /* mddev->patch_version != info->patch_version || */
5584 mddev->ctime != info->ctime ||
5585 mddev->level != info->level ||
5586 /* mddev->layout != info->layout || */
5587 !mddev->persistent != info->not_persistent||
5588 mddev->chunk_sectors != info->chunk_size >> 9 ||
5589 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5590 ((state^info->state) & 0xfffffe00)
5591 )
5592 return -EINVAL;
5593 /* Check there is only one change */
5594 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5595 cnt++;
5596 if (mddev->raid_disks != info->raid_disks)
5597 cnt++;
5598 if (mddev->layout != info->layout)
5599 cnt++;
5600 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5601 cnt++;
5602 if (cnt == 0)
5603 return 0;
5604 if (cnt > 1)
5605 return -EINVAL;
5606
5607 if (mddev->layout != info->layout) {
5608 /* Change layout
5609 * we don't need to do anything at the md level, the
5610 * personality will take care of it all.
5611 */
5612 if (mddev->pers->check_reshape == NULL)
5613 return -EINVAL;
5614 else {
5615 mddev->new_layout = info->layout;
5616 rv = mddev->pers->check_reshape(mddev);
5617 if (rv)
5618 mddev->new_layout = mddev->layout;
5619 return rv;
5620 }
5621 }
5622 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5623 rv = update_size(mddev, (sector_t)info->size * 2);
5624
5625 if (mddev->raid_disks != info->raid_disks)
5626 rv = update_raid_disks(mddev, info->raid_disks);
5627
5628 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5629 if (mddev->pers->quiesce == NULL)
5630 return -EINVAL;
5631 if (mddev->recovery || mddev->sync_thread)
5632 return -EBUSY;
5633 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5634 /* add the bitmap */
5635 if (mddev->bitmap)
5636 return -EEXIST;
5637 if (mddev->bitmap_info.default_offset == 0)
5638 return -EINVAL;
5639 mddev->bitmap_info.offset =
5640 mddev->bitmap_info.default_offset;
5641 mddev->pers->quiesce(mddev, 1);
5642 rv = bitmap_create(mddev);
5643 if (!rv)
5644 rv = bitmap_load(mddev);
5645 if (rv)
5646 bitmap_destroy(mddev);
5647 mddev->pers->quiesce(mddev, 0);
5648 } else {
5649 /* remove the bitmap */
5650 if (!mddev->bitmap)
5651 return -ENOENT;
5652 if (mddev->bitmap->file)
5653 return -EINVAL;
5654 mddev->pers->quiesce(mddev, 1);
5655 bitmap_destroy(mddev);
5656 mddev->pers->quiesce(mddev, 0);
5657 mddev->bitmap_info.offset = 0;
5658 }
5659 }
5660 md_update_sb(mddev, 1);
5661 return rv;
5662 }
5663
5664 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5665 {
5666 mdk_rdev_t *rdev;
5667
5668 if (mddev->pers == NULL)
5669 return -ENODEV;
5670
5671 rdev = find_rdev(mddev, dev);
5672 if (!rdev)
5673 return -ENODEV;
5674
5675 md_error(mddev, rdev);
5676 return 0;
5677 }
5678
5679 /*
5680 * We have a problem here : there is no easy way to give a CHS
5681 * virtual geometry. We currently pretend that we have a 2 heads
5682 * 4 sectors (with a BIG number of cylinders...). This drives
5683 * dosfs just mad... ;-)
5684 */
5685 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5686 {
5687 mddev_t *mddev = bdev->bd_disk->private_data;
5688
5689 geo->heads = 2;
5690 geo->sectors = 4;
5691 geo->cylinders = mddev->array_sectors / 8;
5692 return 0;
5693 }
5694
5695 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5696 unsigned int cmd, unsigned long arg)
5697 {
5698 int err = 0;
5699 void __user *argp = (void __user *)arg;
5700 mddev_t *mddev = NULL;
5701 int ro;
5702
5703 if (!capable(CAP_SYS_ADMIN))
5704 return -EACCES;
5705
5706 /*
5707 * Commands dealing with the RAID driver but not any
5708 * particular array:
5709 */
5710 switch (cmd)
5711 {
5712 case RAID_VERSION:
5713 err = get_version(argp);
5714 goto done;
5715
5716 case PRINT_RAID_DEBUG:
5717 err = 0;
5718 md_print_devices();
5719 goto done;
5720
5721 #ifndef MODULE
5722 case RAID_AUTORUN:
5723 err = 0;
5724 autostart_arrays(arg);
5725 goto done;
5726 #endif
5727 default:;
5728 }
5729
5730 /*
5731 * Commands creating/starting a new array:
5732 */
5733
5734 mddev = bdev->bd_disk->private_data;
5735
5736 if (!mddev) {
5737 BUG();
5738 goto abort;
5739 }
5740
5741 err = mddev_lock(mddev);
5742 if (err) {
5743 printk(KERN_INFO
5744 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5745 err, cmd);
5746 goto abort;
5747 }
5748
5749 switch (cmd)
5750 {
5751 case SET_ARRAY_INFO:
5752 {
5753 mdu_array_info_t info;
5754 if (!arg)
5755 memset(&info, 0, sizeof(info));
5756 else if (copy_from_user(&info, argp, sizeof(info))) {
5757 err = -EFAULT;
5758 goto abort_unlock;
5759 }
5760 if (mddev->pers) {
5761 err = update_array_info(mddev, &info);
5762 if (err) {
5763 printk(KERN_WARNING "md: couldn't update"
5764 " array info. %d\n", err);
5765 goto abort_unlock;
5766 }
5767 goto done_unlock;
5768 }
5769 if (!list_empty(&mddev->disks)) {
5770 printk(KERN_WARNING
5771 "md: array %s already has disks!\n",
5772 mdname(mddev));
5773 err = -EBUSY;
5774 goto abort_unlock;
5775 }
5776 if (mddev->raid_disks) {
5777 printk(KERN_WARNING
5778 "md: array %s already initialised!\n",
5779 mdname(mddev));
5780 err = -EBUSY;
5781 goto abort_unlock;
5782 }
5783 err = set_array_info(mddev, &info);
5784 if (err) {
5785 printk(KERN_WARNING "md: couldn't set"
5786 " array info. %d\n", err);
5787 goto abort_unlock;
5788 }
5789 }
5790 goto done_unlock;
5791
5792 default:;
5793 }
5794
5795 /*
5796 * Commands querying/configuring an existing array:
5797 */
5798 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5799 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5800 if ((!mddev->raid_disks && !mddev->external)
5801 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5802 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5803 && cmd != GET_BITMAP_FILE) {
5804 err = -ENODEV;
5805 goto abort_unlock;
5806 }
5807
5808 /*
5809 * Commands even a read-only array can execute:
5810 */
5811 switch (cmd)
5812 {
5813 case GET_ARRAY_INFO:
5814 err = get_array_info(mddev, argp);
5815 goto done_unlock;
5816
5817 case GET_BITMAP_FILE:
5818 err = get_bitmap_file(mddev, argp);
5819 goto done_unlock;
5820
5821 case GET_DISK_INFO:
5822 err = get_disk_info(mddev, argp);
5823 goto done_unlock;
5824
5825 case RESTART_ARRAY_RW:
5826 err = restart_array(mddev);
5827 goto done_unlock;
5828
5829 case STOP_ARRAY:
5830 err = do_md_stop(mddev, 0, 1);
5831 goto done_unlock;
5832
5833 case STOP_ARRAY_RO:
5834 err = md_set_readonly(mddev, 1);
5835 goto done_unlock;
5836
5837 case BLKROSET:
5838 if (get_user(ro, (int __user *)(arg))) {
5839 err = -EFAULT;
5840 goto done_unlock;
5841 }
5842 err = -EINVAL;
5843
5844 /* if the bdev is going readonly the value of mddev->ro
5845 * does not matter, no writes are coming
5846 */
5847 if (ro)
5848 goto done_unlock;
5849
5850 /* are we are already prepared for writes? */
5851 if (mddev->ro != 1)
5852 goto done_unlock;
5853
5854 /* transitioning to readauto need only happen for
5855 * arrays that call md_write_start
5856 */
5857 if (mddev->pers) {
5858 err = restart_array(mddev);
5859 if (err == 0) {
5860 mddev->ro = 2;
5861 set_disk_ro(mddev->gendisk, 0);
5862 }
5863 }
5864 goto done_unlock;
5865 }
5866
5867 /*
5868 * The remaining ioctls are changing the state of the
5869 * superblock, so we do not allow them on read-only arrays.
5870 * However non-MD ioctls (e.g. get-size) will still come through
5871 * here and hit the 'default' below, so only disallow
5872 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5873 */
5874 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5875 if (mddev->ro == 2) {
5876 mddev->ro = 0;
5877 sysfs_notify_dirent_safe(mddev->sysfs_state);
5878 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5879 md_wakeup_thread(mddev->thread);
5880 } else {
5881 err = -EROFS;
5882 goto abort_unlock;
5883 }
5884 }
5885
5886 switch (cmd)
5887 {
5888 case ADD_NEW_DISK:
5889 {
5890 mdu_disk_info_t info;
5891 if (copy_from_user(&info, argp, sizeof(info)))
5892 err = -EFAULT;
5893 else
5894 err = add_new_disk(mddev, &info);
5895 goto done_unlock;
5896 }
5897
5898 case HOT_REMOVE_DISK:
5899 err = hot_remove_disk(mddev, new_decode_dev(arg));
5900 goto done_unlock;
5901
5902 case HOT_ADD_DISK:
5903 err = hot_add_disk(mddev, new_decode_dev(arg));
5904 goto done_unlock;
5905
5906 case SET_DISK_FAULTY:
5907 err = set_disk_faulty(mddev, new_decode_dev(arg));
5908 goto done_unlock;
5909
5910 case RUN_ARRAY:
5911 err = do_md_run(mddev);
5912 goto done_unlock;
5913
5914 case SET_BITMAP_FILE:
5915 err = set_bitmap_file(mddev, (int)arg);
5916 goto done_unlock;
5917
5918 default:
5919 err = -EINVAL;
5920 goto abort_unlock;
5921 }
5922
5923 done_unlock:
5924 abort_unlock:
5925 if (mddev->hold_active == UNTIL_IOCTL &&
5926 err != -EINVAL)
5927 mddev->hold_active = 0;
5928 mddev_unlock(mddev);
5929
5930 return err;
5931 done:
5932 if (err)
5933 MD_BUG();
5934 abort:
5935 return err;
5936 }
5937 #ifdef CONFIG_COMPAT
5938 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5939 unsigned int cmd, unsigned long arg)
5940 {
5941 switch (cmd) {
5942 case HOT_REMOVE_DISK:
5943 case HOT_ADD_DISK:
5944 case SET_DISK_FAULTY:
5945 case SET_BITMAP_FILE:
5946 /* These take in integer arg, do not convert */
5947 break;
5948 default:
5949 arg = (unsigned long)compat_ptr(arg);
5950 break;
5951 }
5952
5953 return md_ioctl(bdev, mode, cmd, arg);
5954 }
5955 #endif /* CONFIG_COMPAT */
5956
5957 static int md_open(struct block_device *bdev, fmode_t mode)
5958 {
5959 /*
5960 * Succeed if we can lock the mddev, which confirms that
5961 * it isn't being stopped right now.
5962 */
5963 mddev_t *mddev = mddev_find(bdev->bd_dev);
5964 int err;
5965
5966 lock_kernel();
5967 if (mddev->gendisk != bdev->bd_disk) {
5968 /* we are racing with mddev_put which is discarding this
5969 * bd_disk.
5970 */
5971 mddev_put(mddev);
5972 /* Wait until bdev->bd_disk is definitely gone */
5973 flush_scheduled_work();
5974 /* Then retry the open from the top */
5975 unlock_kernel();
5976 return -ERESTARTSYS;
5977 }
5978 BUG_ON(mddev != bdev->bd_disk->private_data);
5979
5980 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5981 goto out;
5982
5983 err = 0;
5984 atomic_inc(&mddev->openers);
5985 mutex_unlock(&mddev->open_mutex);
5986
5987 check_disk_size_change(mddev->gendisk, bdev);
5988 out:
5989 unlock_kernel();
5990 return err;
5991 }
5992
5993 static int md_release(struct gendisk *disk, fmode_t mode)
5994 {
5995 mddev_t *mddev = disk->private_data;
5996
5997 BUG_ON(!mddev);
5998 lock_kernel();
5999 atomic_dec(&mddev->openers);
6000 mddev_put(mddev);
6001 unlock_kernel();
6002
6003 return 0;
6004 }
6005 static const struct block_device_operations md_fops =
6006 {
6007 .owner = THIS_MODULE,
6008 .open = md_open,
6009 .release = md_release,
6010 .ioctl = md_ioctl,
6011 #ifdef CONFIG_COMPAT
6012 .compat_ioctl = md_compat_ioctl,
6013 #endif
6014 .getgeo = md_getgeo,
6015 };
6016
6017 static int md_thread(void * arg)
6018 {
6019 mdk_thread_t *thread = arg;
6020
6021 /*
6022 * md_thread is a 'system-thread', it's priority should be very
6023 * high. We avoid resource deadlocks individually in each
6024 * raid personality. (RAID5 does preallocation) We also use RR and
6025 * the very same RT priority as kswapd, thus we will never get
6026 * into a priority inversion deadlock.
6027 *
6028 * we definitely have to have equal or higher priority than
6029 * bdflush, otherwise bdflush will deadlock if there are too
6030 * many dirty RAID5 blocks.
6031 */
6032
6033 allow_signal(SIGKILL);
6034 while (!kthread_should_stop()) {
6035
6036 /* We need to wait INTERRUPTIBLE so that
6037 * we don't add to the load-average.
6038 * That means we need to be sure no signals are
6039 * pending
6040 */
6041 if (signal_pending(current))
6042 flush_signals(current);
6043
6044 wait_event_interruptible_timeout
6045 (thread->wqueue,
6046 test_bit(THREAD_WAKEUP, &thread->flags)
6047 || kthread_should_stop(),
6048 thread->timeout);
6049
6050 clear_bit(THREAD_WAKEUP, &thread->flags);
6051 if (!kthread_should_stop())
6052 thread->run(thread->mddev);
6053 }
6054
6055 return 0;
6056 }
6057
6058 void md_wakeup_thread(mdk_thread_t *thread)
6059 {
6060 if (thread) {
6061 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6062 set_bit(THREAD_WAKEUP, &thread->flags);
6063 wake_up(&thread->wqueue);
6064 }
6065 }
6066
6067 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6068 const char *name)
6069 {
6070 mdk_thread_t *thread;
6071
6072 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6073 if (!thread)
6074 return NULL;
6075
6076 init_waitqueue_head(&thread->wqueue);
6077
6078 thread->run = run;
6079 thread->mddev = mddev;
6080 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6081 thread->tsk = kthread_run(md_thread, thread,
6082 "%s_%s",
6083 mdname(thread->mddev),
6084 name ?: mddev->pers->name);
6085 if (IS_ERR(thread->tsk)) {
6086 kfree(thread);
6087 return NULL;
6088 }
6089 return thread;
6090 }
6091
6092 void md_unregister_thread(mdk_thread_t *thread)
6093 {
6094 if (!thread)
6095 return;
6096 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6097
6098 kthread_stop(thread->tsk);
6099 kfree(thread);
6100 }
6101
6102 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6103 {
6104 if (!mddev) {
6105 MD_BUG();
6106 return;
6107 }
6108
6109 if (!rdev || test_bit(Faulty, &rdev->flags))
6110 return;
6111
6112 if (mddev->external)
6113 set_bit(Blocked, &rdev->flags);
6114 /*
6115 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6116 mdname(mddev),
6117 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6118 __builtin_return_address(0),__builtin_return_address(1),
6119 __builtin_return_address(2),__builtin_return_address(3));
6120 */
6121 if (!mddev->pers)
6122 return;
6123 if (!mddev->pers->error_handler)
6124 return;
6125 mddev->pers->error_handler(mddev,rdev);
6126 if (mddev->degraded)
6127 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6128 sysfs_notify_dirent_safe(rdev->sysfs_state);
6129 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6130 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6131 md_wakeup_thread(mddev->thread);
6132 if (mddev->event_work.func)
6133 schedule_work(&mddev->event_work);
6134 md_new_event_inintr(mddev);
6135 }
6136
6137 /* seq_file implementation /proc/mdstat */
6138
6139 static void status_unused(struct seq_file *seq)
6140 {
6141 int i = 0;
6142 mdk_rdev_t *rdev;
6143
6144 seq_printf(seq, "unused devices: ");
6145
6146 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6147 char b[BDEVNAME_SIZE];
6148 i++;
6149 seq_printf(seq, "%s ",
6150 bdevname(rdev->bdev,b));
6151 }
6152 if (!i)
6153 seq_printf(seq, "<none>");
6154
6155 seq_printf(seq, "\n");
6156 }
6157
6158
6159 static void status_resync(struct seq_file *seq, mddev_t * mddev)
6160 {
6161 sector_t max_sectors, resync, res;
6162 unsigned long dt, db;
6163 sector_t rt;
6164 int scale;
6165 unsigned int per_milli;
6166
6167 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6168
6169 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6170 max_sectors = mddev->resync_max_sectors;
6171 else
6172 max_sectors = mddev->dev_sectors;
6173
6174 /*
6175 * Should not happen.
6176 */
6177 if (!max_sectors) {
6178 MD_BUG();
6179 return;
6180 }
6181 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6182 * in a sector_t, and (max_sectors>>scale) will fit in a
6183 * u32, as those are the requirements for sector_div.
6184 * Thus 'scale' must be at least 10
6185 */
6186 scale = 10;
6187 if (sizeof(sector_t) > sizeof(unsigned long)) {
6188 while ( max_sectors/2 > (1ULL<<(scale+32)))
6189 scale++;
6190 }
6191 res = (resync>>scale)*1000;
6192 sector_div(res, (u32)((max_sectors>>scale)+1));
6193
6194 per_milli = res;
6195 {
6196 int i, x = per_milli/50, y = 20-x;
6197 seq_printf(seq, "[");
6198 for (i = 0; i < x; i++)
6199 seq_printf(seq, "=");
6200 seq_printf(seq, ">");
6201 for (i = 0; i < y; i++)
6202 seq_printf(seq, ".");
6203 seq_printf(seq, "] ");
6204 }
6205 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6206 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6207 "reshape" :
6208 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6209 "check" :
6210 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6211 "resync" : "recovery"))),
6212 per_milli/10, per_milli % 10,
6213 (unsigned long long) resync/2,
6214 (unsigned long long) max_sectors/2);
6215
6216 /*
6217 * dt: time from mark until now
6218 * db: blocks written from mark until now
6219 * rt: remaining time
6220 *
6221 * rt is a sector_t, so could be 32bit or 64bit.
6222 * So we divide before multiply in case it is 32bit and close
6223 * to the limit.
6224 * We scale the divisor (db) by 32 to avoid loosing precision
6225 * near the end of resync when the number of remaining sectors
6226 * is close to 'db'.
6227 * We then divide rt by 32 after multiplying by db to compensate.
6228 * The '+1' avoids division by zero if db is very small.
6229 */
6230 dt = ((jiffies - mddev->resync_mark) / HZ);
6231 if (!dt) dt++;
6232 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6233 - mddev->resync_mark_cnt;
6234
6235 rt = max_sectors - resync; /* number of remaining sectors */
6236 sector_div(rt, db/32+1);
6237 rt *= dt;
6238 rt >>= 5;
6239
6240 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6241 ((unsigned long)rt % 60)/6);
6242
6243 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6244 }
6245
6246 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6247 {
6248 struct list_head *tmp;
6249 loff_t l = *pos;
6250 mddev_t *mddev;
6251
6252 if (l >= 0x10000)
6253 return NULL;
6254 if (!l--)
6255 /* header */
6256 return (void*)1;
6257
6258 spin_lock(&all_mddevs_lock);
6259 list_for_each(tmp,&all_mddevs)
6260 if (!l--) {
6261 mddev = list_entry(tmp, mddev_t, all_mddevs);
6262 mddev_get(mddev);
6263 spin_unlock(&all_mddevs_lock);
6264 return mddev;
6265 }
6266 spin_unlock(&all_mddevs_lock);
6267 if (!l--)
6268 return (void*)2;/* tail */
6269 return NULL;
6270 }
6271
6272 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6273 {
6274 struct list_head *tmp;
6275 mddev_t *next_mddev, *mddev = v;
6276
6277 ++*pos;
6278 if (v == (void*)2)
6279 return NULL;
6280
6281 spin_lock(&all_mddevs_lock);
6282 if (v == (void*)1)
6283 tmp = all_mddevs.next;
6284 else
6285 tmp = mddev->all_mddevs.next;
6286 if (tmp != &all_mddevs)
6287 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6288 else {
6289 next_mddev = (void*)2;
6290 *pos = 0x10000;
6291 }
6292 spin_unlock(&all_mddevs_lock);
6293
6294 if (v != (void*)1)
6295 mddev_put(mddev);
6296 return next_mddev;
6297
6298 }
6299
6300 static void md_seq_stop(struct seq_file *seq, void *v)
6301 {
6302 mddev_t *mddev = v;
6303
6304 if (mddev && v != (void*)1 && v != (void*)2)
6305 mddev_put(mddev);
6306 }
6307
6308 struct mdstat_info {
6309 int event;
6310 };
6311
6312 static int md_seq_show(struct seq_file *seq, void *v)
6313 {
6314 mddev_t *mddev = v;
6315 sector_t sectors;
6316 mdk_rdev_t *rdev;
6317 struct mdstat_info *mi = seq->private;
6318 struct bitmap *bitmap;
6319
6320 if (v == (void*)1) {
6321 struct mdk_personality *pers;
6322 seq_printf(seq, "Personalities : ");
6323 spin_lock(&pers_lock);
6324 list_for_each_entry(pers, &pers_list, list)
6325 seq_printf(seq, "[%s] ", pers->name);
6326
6327 spin_unlock(&pers_lock);
6328 seq_printf(seq, "\n");
6329 mi->event = atomic_read(&md_event_count);
6330 return 0;
6331 }
6332 if (v == (void*)2) {
6333 status_unused(seq);
6334 return 0;
6335 }
6336
6337 if (mddev_lock(mddev) < 0)
6338 return -EINTR;
6339
6340 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6341 seq_printf(seq, "%s : %sactive", mdname(mddev),
6342 mddev->pers ? "" : "in");
6343 if (mddev->pers) {
6344 if (mddev->ro==1)
6345 seq_printf(seq, " (read-only)");
6346 if (mddev->ro==2)
6347 seq_printf(seq, " (auto-read-only)");
6348 seq_printf(seq, " %s", mddev->pers->name);
6349 }
6350
6351 sectors = 0;
6352 list_for_each_entry(rdev, &mddev->disks, same_set) {
6353 char b[BDEVNAME_SIZE];
6354 seq_printf(seq, " %s[%d]",
6355 bdevname(rdev->bdev,b), rdev->desc_nr);
6356 if (test_bit(WriteMostly, &rdev->flags))
6357 seq_printf(seq, "(W)");
6358 if (test_bit(Faulty, &rdev->flags)) {
6359 seq_printf(seq, "(F)");
6360 continue;
6361 } else if (rdev->raid_disk < 0)
6362 seq_printf(seq, "(S)"); /* spare */
6363 sectors += rdev->sectors;
6364 }
6365
6366 if (!list_empty(&mddev->disks)) {
6367 if (mddev->pers)
6368 seq_printf(seq, "\n %llu blocks",
6369 (unsigned long long)
6370 mddev->array_sectors / 2);
6371 else
6372 seq_printf(seq, "\n %llu blocks",
6373 (unsigned long long)sectors / 2);
6374 }
6375 if (mddev->persistent) {
6376 if (mddev->major_version != 0 ||
6377 mddev->minor_version != 90) {
6378 seq_printf(seq," super %d.%d",
6379 mddev->major_version,
6380 mddev->minor_version);
6381 }
6382 } else if (mddev->external)
6383 seq_printf(seq, " super external:%s",
6384 mddev->metadata_type);
6385 else
6386 seq_printf(seq, " super non-persistent");
6387
6388 if (mddev->pers) {
6389 mddev->pers->status(seq, mddev);
6390 seq_printf(seq, "\n ");
6391 if (mddev->pers->sync_request) {
6392 if (mddev->curr_resync > 2) {
6393 status_resync(seq, mddev);
6394 seq_printf(seq, "\n ");
6395 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6396 seq_printf(seq, "\tresync=DELAYED\n ");
6397 else if (mddev->recovery_cp < MaxSector)
6398 seq_printf(seq, "\tresync=PENDING\n ");
6399 }
6400 } else
6401 seq_printf(seq, "\n ");
6402
6403 if ((bitmap = mddev->bitmap)) {
6404 unsigned long chunk_kb;
6405 unsigned long flags;
6406 spin_lock_irqsave(&bitmap->lock, flags);
6407 chunk_kb = mddev->bitmap_info.chunksize >> 10;
6408 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6409 "%lu%s chunk",
6410 bitmap->pages - bitmap->missing_pages,
6411 bitmap->pages,
6412 (bitmap->pages - bitmap->missing_pages)
6413 << (PAGE_SHIFT - 10),
6414 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6415 chunk_kb ? "KB" : "B");
6416 if (bitmap->file) {
6417 seq_printf(seq, ", file: ");
6418 seq_path(seq, &bitmap->file->f_path, " \t\n");
6419 }
6420
6421 seq_printf(seq, "\n");
6422 spin_unlock_irqrestore(&bitmap->lock, flags);
6423 }
6424
6425 seq_printf(seq, "\n");
6426 }
6427 mddev_unlock(mddev);
6428
6429 return 0;
6430 }
6431
6432 static const struct seq_operations md_seq_ops = {
6433 .start = md_seq_start,
6434 .next = md_seq_next,
6435 .stop = md_seq_stop,
6436 .show = md_seq_show,
6437 };
6438
6439 static int md_seq_open(struct inode *inode, struct file *file)
6440 {
6441 int error;
6442 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6443 if (mi == NULL)
6444 return -ENOMEM;
6445
6446 error = seq_open(file, &md_seq_ops);
6447 if (error)
6448 kfree(mi);
6449 else {
6450 struct seq_file *p = file->private_data;
6451 p->private = mi;
6452 mi->event = atomic_read(&md_event_count);
6453 }
6454 return error;
6455 }
6456
6457 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6458 {
6459 struct seq_file *m = filp->private_data;
6460 struct mdstat_info *mi = m->private;
6461 int mask;
6462
6463 poll_wait(filp, &md_event_waiters, wait);
6464
6465 /* always allow read */
6466 mask = POLLIN | POLLRDNORM;
6467
6468 if (mi->event != atomic_read(&md_event_count))
6469 mask |= POLLERR | POLLPRI;
6470 return mask;
6471 }
6472
6473 static const struct file_operations md_seq_fops = {
6474 .owner = THIS_MODULE,
6475 .open = md_seq_open,
6476 .read = seq_read,
6477 .llseek = seq_lseek,
6478 .release = seq_release_private,
6479 .poll = mdstat_poll,
6480 };
6481
6482 int register_md_personality(struct mdk_personality *p)
6483 {
6484 spin_lock(&pers_lock);
6485 list_add_tail(&p->list, &pers_list);
6486 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6487 spin_unlock(&pers_lock);
6488 return 0;
6489 }
6490
6491 int unregister_md_personality(struct mdk_personality *p)
6492 {
6493 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6494 spin_lock(&pers_lock);
6495 list_del_init(&p->list);
6496 spin_unlock(&pers_lock);
6497 return 0;
6498 }
6499
6500 static int is_mddev_idle(mddev_t *mddev, int init)
6501 {
6502 mdk_rdev_t * rdev;
6503 int idle;
6504 int curr_events;
6505
6506 idle = 1;
6507 rcu_read_lock();
6508 rdev_for_each_rcu(rdev, mddev) {
6509 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6510 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6511 (int)part_stat_read(&disk->part0, sectors[1]) -
6512 atomic_read(&disk->sync_io);
6513 /* sync IO will cause sync_io to increase before the disk_stats
6514 * as sync_io is counted when a request starts, and
6515 * disk_stats is counted when it completes.
6516 * So resync activity will cause curr_events to be smaller than
6517 * when there was no such activity.
6518 * non-sync IO will cause disk_stat to increase without
6519 * increasing sync_io so curr_events will (eventually)
6520 * be larger than it was before. Once it becomes
6521 * substantially larger, the test below will cause
6522 * the array to appear non-idle, and resync will slow
6523 * down.
6524 * If there is a lot of outstanding resync activity when
6525 * we set last_event to curr_events, then all that activity
6526 * completing might cause the array to appear non-idle
6527 * and resync will be slowed down even though there might
6528 * not have been non-resync activity. This will only
6529 * happen once though. 'last_events' will soon reflect
6530 * the state where there is little or no outstanding
6531 * resync requests, and further resync activity will
6532 * always make curr_events less than last_events.
6533 *
6534 */
6535 if (init || curr_events - rdev->last_events > 64) {
6536 rdev->last_events = curr_events;
6537 idle = 0;
6538 }
6539 }
6540 rcu_read_unlock();
6541 return idle;
6542 }
6543
6544 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6545 {
6546 /* another "blocks" (512byte) blocks have been synced */
6547 atomic_sub(blocks, &mddev->recovery_active);
6548 wake_up(&mddev->recovery_wait);
6549 if (!ok) {
6550 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6551 md_wakeup_thread(mddev->thread);
6552 // stop recovery, signal do_sync ....
6553 }
6554 }
6555
6556
6557 /* md_write_start(mddev, bi)
6558 * If we need to update some array metadata (e.g. 'active' flag
6559 * in superblock) before writing, schedule a superblock update
6560 * and wait for it to complete.
6561 */
6562 void md_write_start(mddev_t *mddev, struct bio *bi)
6563 {
6564 int did_change = 0;
6565 if (bio_data_dir(bi) != WRITE)
6566 return;
6567
6568 BUG_ON(mddev->ro == 1);
6569 if (mddev->ro == 2) {
6570 /* need to switch to read/write */
6571 mddev->ro = 0;
6572 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6573 md_wakeup_thread(mddev->thread);
6574 md_wakeup_thread(mddev->sync_thread);
6575 did_change = 1;
6576 }
6577 atomic_inc(&mddev->writes_pending);
6578 if (mddev->safemode == 1)
6579 mddev->safemode = 0;
6580 if (mddev->in_sync) {
6581 spin_lock_irq(&mddev->write_lock);
6582 if (mddev->in_sync) {
6583 mddev->in_sync = 0;
6584 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6585 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6586 md_wakeup_thread(mddev->thread);
6587 did_change = 1;
6588 }
6589 spin_unlock_irq(&mddev->write_lock);
6590 }
6591 if (did_change)
6592 sysfs_notify_dirent_safe(mddev->sysfs_state);
6593 wait_event(mddev->sb_wait,
6594 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6595 }
6596
6597 void md_write_end(mddev_t *mddev)
6598 {
6599 if (atomic_dec_and_test(&mddev->writes_pending)) {
6600 if (mddev->safemode == 2)
6601 md_wakeup_thread(mddev->thread);
6602 else if (mddev->safemode_delay)
6603 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6604 }
6605 }
6606
6607 /* md_allow_write(mddev)
6608 * Calling this ensures that the array is marked 'active' so that writes
6609 * may proceed without blocking. It is important to call this before
6610 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6611 * Must be called with mddev_lock held.
6612 *
6613 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6614 * is dropped, so return -EAGAIN after notifying userspace.
6615 */
6616 int md_allow_write(mddev_t *mddev)
6617 {
6618 if (!mddev->pers)
6619 return 0;
6620 if (mddev->ro)
6621 return 0;
6622 if (!mddev->pers->sync_request)
6623 return 0;
6624
6625 spin_lock_irq(&mddev->write_lock);
6626 if (mddev->in_sync) {
6627 mddev->in_sync = 0;
6628 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6629 set_bit(MD_CHANGE_PENDING, &mddev->flags);
6630 if (mddev->safemode_delay &&
6631 mddev->safemode == 0)
6632 mddev->safemode = 1;
6633 spin_unlock_irq(&mddev->write_lock);
6634 md_update_sb(mddev, 0);
6635 sysfs_notify_dirent_safe(mddev->sysfs_state);
6636 } else
6637 spin_unlock_irq(&mddev->write_lock);
6638
6639 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6640 return -EAGAIN;
6641 else
6642 return 0;
6643 }
6644 EXPORT_SYMBOL_GPL(md_allow_write);
6645
6646 void md_unplug(mddev_t *mddev)
6647 {
6648 if (mddev->queue)
6649 blk_unplug(mddev->queue);
6650 if (mddev->plug)
6651 mddev->plug->unplug_fn(mddev->plug);
6652 }
6653
6654 #define SYNC_MARKS 10
6655 #define SYNC_MARK_STEP (3*HZ)
6656 void md_do_sync(mddev_t *mddev)
6657 {
6658 mddev_t *mddev2;
6659 unsigned int currspeed = 0,
6660 window;
6661 sector_t max_sectors,j, io_sectors;
6662 unsigned long mark[SYNC_MARKS];
6663 sector_t mark_cnt[SYNC_MARKS];
6664 int last_mark,m;
6665 struct list_head *tmp;
6666 sector_t last_check;
6667 int skipped = 0;
6668 mdk_rdev_t *rdev;
6669 char *desc;
6670
6671 /* just incase thread restarts... */
6672 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6673 return;
6674 if (mddev->ro) /* never try to sync a read-only array */
6675 return;
6676
6677 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6678 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6679 desc = "data-check";
6680 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6681 desc = "requested-resync";
6682 else
6683 desc = "resync";
6684 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6685 desc = "reshape";
6686 else
6687 desc = "recovery";
6688
6689 /* we overload curr_resync somewhat here.
6690 * 0 == not engaged in resync at all
6691 * 2 == checking that there is no conflict with another sync
6692 * 1 == like 2, but have yielded to allow conflicting resync to
6693 * commense
6694 * other == active in resync - this many blocks
6695 *
6696 * Before starting a resync we must have set curr_resync to
6697 * 2, and then checked that every "conflicting" array has curr_resync
6698 * less than ours. When we find one that is the same or higher
6699 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6700 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6701 * This will mean we have to start checking from the beginning again.
6702 *
6703 */
6704
6705 do {
6706 mddev->curr_resync = 2;
6707
6708 try_again:
6709 if (kthread_should_stop())
6710 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6711
6712 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6713 goto skip;
6714 for_each_mddev(mddev2, tmp) {
6715 if (mddev2 == mddev)
6716 continue;
6717 if (!mddev->parallel_resync
6718 && mddev2->curr_resync
6719 && match_mddev_units(mddev, mddev2)) {
6720 DEFINE_WAIT(wq);
6721 if (mddev < mddev2 && mddev->curr_resync == 2) {
6722 /* arbitrarily yield */
6723 mddev->curr_resync = 1;
6724 wake_up(&resync_wait);
6725 }
6726 if (mddev > mddev2 && mddev->curr_resync == 1)
6727 /* no need to wait here, we can wait the next
6728 * time 'round when curr_resync == 2
6729 */
6730 continue;
6731 /* We need to wait 'interruptible' so as not to
6732 * contribute to the load average, and not to
6733 * be caught by 'softlockup'
6734 */
6735 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6736 if (!kthread_should_stop() &&
6737 mddev2->curr_resync >= mddev->curr_resync) {
6738 printk(KERN_INFO "md: delaying %s of %s"
6739 " until %s has finished (they"
6740 " share one or more physical units)\n",
6741 desc, mdname(mddev), mdname(mddev2));
6742 mddev_put(mddev2);
6743 if (signal_pending(current))
6744 flush_signals(current);
6745 schedule();
6746 finish_wait(&resync_wait, &wq);
6747 goto try_again;
6748 }
6749 finish_wait(&resync_wait, &wq);
6750 }
6751 }
6752 } while (mddev->curr_resync < 2);
6753
6754 j = 0;
6755 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6756 /* resync follows the size requested by the personality,
6757 * which defaults to physical size, but can be virtual size
6758 */
6759 max_sectors = mddev->resync_max_sectors;
6760 mddev->resync_mismatches = 0;
6761 /* we don't use the checkpoint if there's a bitmap */
6762 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6763 j = mddev->resync_min;
6764 else if (!mddev->bitmap)
6765 j = mddev->recovery_cp;
6766
6767 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6768 max_sectors = mddev->dev_sectors;
6769 else {
6770 /* recovery follows the physical size of devices */
6771 max_sectors = mddev->dev_sectors;
6772 j = MaxSector;
6773 rcu_read_lock();
6774 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6775 if (rdev->raid_disk >= 0 &&
6776 !test_bit(Faulty, &rdev->flags) &&
6777 !test_bit(In_sync, &rdev->flags) &&
6778 rdev->recovery_offset < j)
6779 j = rdev->recovery_offset;
6780 rcu_read_unlock();
6781 }
6782
6783 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6784 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6785 " %d KB/sec/disk.\n", speed_min(mddev));
6786 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6787 "(but not more than %d KB/sec) for %s.\n",
6788 speed_max(mddev), desc);
6789
6790 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6791
6792 io_sectors = 0;
6793 for (m = 0; m < SYNC_MARKS; m++) {
6794 mark[m] = jiffies;
6795 mark_cnt[m] = io_sectors;
6796 }
6797 last_mark = 0;
6798 mddev->resync_mark = mark[last_mark];
6799 mddev->resync_mark_cnt = mark_cnt[last_mark];
6800
6801 /*
6802 * Tune reconstruction:
6803 */
6804 window = 32*(PAGE_SIZE/512);
6805 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6806 window/2,(unsigned long long) max_sectors/2);
6807
6808 atomic_set(&mddev->recovery_active, 0);
6809 last_check = 0;
6810
6811 if (j>2) {
6812 printk(KERN_INFO
6813 "md: resuming %s of %s from checkpoint.\n",
6814 desc, mdname(mddev));
6815 mddev->curr_resync = j;
6816 }
6817 mddev->curr_resync_completed = mddev->curr_resync;
6818
6819 while (j < max_sectors) {
6820 sector_t sectors;
6821
6822 skipped = 0;
6823
6824 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6825 ((mddev->curr_resync > mddev->curr_resync_completed &&
6826 (mddev->curr_resync - mddev->curr_resync_completed)
6827 > (max_sectors >> 4)) ||
6828 (j - mddev->curr_resync_completed)*2
6829 >= mddev->resync_max - mddev->curr_resync_completed
6830 )) {
6831 /* time to update curr_resync_completed */
6832 md_unplug(mddev);
6833 wait_event(mddev->recovery_wait,
6834 atomic_read(&mddev->recovery_active) == 0);
6835 mddev->curr_resync_completed =
6836 mddev->curr_resync;
6837 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6838 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6839 }
6840
6841 while (j >= mddev->resync_max && !kthread_should_stop()) {
6842 /* As this condition is controlled by user-space,
6843 * we can block indefinitely, so use '_interruptible'
6844 * to avoid triggering warnings.
6845 */
6846 flush_signals(current); /* just in case */
6847 wait_event_interruptible(mddev->recovery_wait,
6848 mddev->resync_max > j
6849 || kthread_should_stop());
6850 }
6851
6852 if (kthread_should_stop())
6853 goto interrupted;
6854
6855 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6856 currspeed < speed_min(mddev));
6857 if (sectors == 0) {
6858 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6859 goto out;
6860 }
6861
6862 if (!skipped) { /* actual IO requested */
6863 io_sectors += sectors;
6864 atomic_add(sectors, &mddev->recovery_active);
6865 }
6866
6867 j += sectors;
6868 if (j>1) mddev->curr_resync = j;
6869 mddev->curr_mark_cnt = io_sectors;
6870 if (last_check == 0)
6871 /* this is the earliers that rebuilt will be
6872 * visible in /proc/mdstat
6873 */
6874 md_new_event(mddev);
6875
6876 if (last_check + window > io_sectors || j == max_sectors)
6877 continue;
6878
6879 last_check = io_sectors;
6880
6881 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6882 break;
6883
6884 repeat:
6885 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6886 /* step marks */
6887 int next = (last_mark+1) % SYNC_MARKS;
6888
6889 mddev->resync_mark = mark[next];
6890 mddev->resync_mark_cnt = mark_cnt[next];
6891 mark[next] = jiffies;
6892 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6893 last_mark = next;
6894 }
6895
6896
6897 if (kthread_should_stop())
6898 goto interrupted;
6899
6900
6901 /*
6902 * this loop exits only if either when we are slower than
6903 * the 'hard' speed limit, or the system was IO-idle for
6904 * a jiffy.
6905 * the system might be non-idle CPU-wise, but we only care
6906 * about not overloading the IO subsystem. (things like an
6907 * e2fsck being done on the RAID array should execute fast)
6908 */
6909 md_unplug(mddev);
6910 cond_resched();
6911
6912 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6913 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6914
6915 if (currspeed > speed_min(mddev)) {
6916 if ((currspeed > speed_max(mddev)) ||
6917 !is_mddev_idle(mddev, 0)) {
6918 msleep(500);
6919 goto repeat;
6920 }
6921 }
6922 }
6923 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6924 /*
6925 * this also signals 'finished resyncing' to md_stop
6926 */
6927 out:
6928 md_unplug(mddev);
6929
6930 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6931
6932 /* tell personality that we are finished */
6933 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6934
6935 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6936 mddev->curr_resync > 2) {
6937 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6938 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6939 if (mddev->curr_resync >= mddev->recovery_cp) {
6940 printk(KERN_INFO
6941 "md: checkpointing %s of %s.\n",
6942 desc, mdname(mddev));
6943 mddev->recovery_cp = mddev->curr_resync;
6944 }
6945 } else
6946 mddev->recovery_cp = MaxSector;
6947 } else {
6948 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6949 mddev->curr_resync = MaxSector;
6950 rcu_read_lock();
6951 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
6952 if (rdev->raid_disk >= 0 &&
6953 mddev->delta_disks >= 0 &&
6954 !test_bit(Faulty, &rdev->flags) &&
6955 !test_bit(In_sync, &rdev->flags) &&
6956 rdev->recovery_offset < mddev->curr_resync)
6957 rdev->recovery_offset = mddev->curr_resync;
6958 rcu_read_unlock();
6959 }
6960 }
6961 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6962
6963 skip:
6964 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6965 /* We completed so min/max setting can be forgotten if used. */
6966 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6967 mddev->resync_min = 0;
6968 mddev->resync_max = MaxSector;
6969 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6970 mddev->resync_min = mddev->curr_resync_completed;
6971 mddev->curr_resync = 0;
6972 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6973 mddev->curr_resync_completed = 0;
6974 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6975 wake_up(&resync_wait);
6976 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6977 md_wakeup_thread(mddev->thread);
6978 return;
6979
6980 interrupted:
6981 /*
6982 * got a signal, exit.
6983 */
6984 printk(KERN_INFO
6985 "md: md_do_sync() got signal ... exiting\n");
6986 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6987 goto out;
6988
6989 }
6990 EXPORT_SYMBOL_GPL(md_do_sync);
6991
6992
6993 static int remove_and_add_spares(mddev_t *mddev)
6994 {
6995 mdk_rdev_t *rdev;
6996 int spares = 0;
6997
6998 mddev->curr_resync_completed = 0;
6999
7000 list_for_each_entry(rdev, &mddev->disks, same_set)
7001 if (rdev->raid_disk >= 0 &&
7002 !test_bit(Blocked, &rdev->flags) &&
7003 (test_bit(Faulty, &rdev->flags) ||
7004 ! test_bit(In_sync, &rdev->flags)) &&
7005 atomic_read(&rdev->nr_pending)==0) {
7006 if (mddev->pers->hot_remove_disk(
7007 mddev, rdev->raid_disk)==0) {
7008 char nm[20];
7009 sprintf(nm,"rd%d", rdev->raid_disk);
7010 sysfs_remove_link(&mddev->kobj, nm);
7011 rdev->raid_disk = -1;
7012 }
7013 }
7014
7015 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
7016 list_for_each_entry(rdev, &mddev->disks, same_set) {
7017 if (rdev->raid_disk >= 0 &&
7018 !test_bit(In_sync, &rdev->flags) &&
7019 !test_bit(Blocked, &rdev->flags))
7020 spares++;
7021 if (rdev->raid_disk < 0
7022 && !test_bit(Faulty, &rdev->flags)) {
7023 rdev->recovery_offset = 0;
7024 if (mddev->pers->
7025 hot_add_disk(mddev, rdev) == 0) {
7026 char nm[20];
7027 sprintf(nm, "rd%d", rdev->raid_disk);
7028 if (sysfs_create_link(&mddev->kobj,
7029 &rdev->kobj, nm))
7030 /* failure here is OK */;
7031 spares++;
7032 md_new_event(mddev);
7033 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7034 } else
7035 break;
7036 }
7037 }
7038 }
7039 return spares;
7040 }
7041 /*
7042 * This routine is regularly called by all per-raid-array threads to
7043 * deal with generic issues like resync and super-block update.
7044 * Raid personalities that don't have a thread (linear/raid0) do not
7045 * need this as they never do any recovery or update the superblock.
7046 *
7047 * It does not do any resync itself, but rather "forks" off other threads
7048 * to do that as needed.
7049 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7050 * "->recovery" and create a thread at ->sync_thread.
7051 * When the thread finishes it sets MD_RECOVERY_DONE
7052 * and wakeups up this thread which will reap the thread and finish up.
7053 * This thread also removes any faulty devices (with nr_pending == 0).
7054 *
7055 * The overall approach is:
7056 * 1/ if the superblock needs updating, update it.
7057 * 2/ If a recovery thread is running, don't do anything else.
7058 * 3/ If recovery has finished, clean up, possibly marking spares active.
7059 * 4/ If there are any faulty devices, remove them.
7060 * 5/ If array is degraded, try to add spares devices
7061 * 6/ If array has spares or is not in-sync, start a resync thread.
7062 */
7063 void md_check_recovery(mddev_t *mddev)
7064 {
7065 mdk_rdev_t *rdev;
7066
7067
7068 if (mddev->bitmap)
7069 bitmap_daemon_work(mddev);
7070
7071 if (mddev->ro)
7072 return;
7073
7074 if (signal_pending(current)) {
7075 if (mddev->pers->sync_request && !mddev->external) {
7076 printk(KERN_INFO "md: %s in immediate safe mode\n",
7077 mdname(mddev));
7078 mddev->safemode = 2;
7079 }
7080 flush_signals(current);
7081 }
7082
7083 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7084 return;
7085 if ( ! (
7086 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7087 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7088 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7089 (mddev->external == 0 && mddev->safemode == 1) ||
7090 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7091 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7092 ))
7093 return;
7094
7095 if (mddev_trylock(mddev)) {
7096 int spares = 0;
7097
7098 if (mddev->ro) {
7099 /* Only thing we do on a ro array is remove
7100 * failed devices.
7101 */
7102 remove_and_add_spares(mddev);
7103 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7104 goto unlock;
7105 }
7106
7107 if (!mddev->external) {
7108 int did_change = 0;
7109 spin_lock_irq(&mddev->write_lock);
7110 if (mddev->safemode &&
7111 !atomic_read(&mddev->writes_pending) &&
7112 !mddev->in_sync &&
7113 mddev->recovery_cp == MaxSector) {
7114 mddev->in_sync = 1;
7115 did_change = 1;
7116 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7117 }
7118 if (mddev->safemode == 1)
7119 mddev->safemode = 0;
7120 spin_unlock_irq(&mddev->write_lock);
7121 if (did_change)
7122 sysfs_notify_dirent_safe(mddev->sysfs_state);
7123 }
7124
7125 if (mddev->flags)
7126 md_update_sb(mddev, 0);
7127
7128 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7129 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7130 /* resync/recovery still happening */
7131 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7132 goto unlock;
7133 }
7134 if (mddev->sync_thread) {
7135 /* resync has finished, collect result */
7136 md_unregister_thread(mddev->sync_thread);
7137 mddev->sync_thread = NULL;
7138 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7139 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7140 /* success...*/
7141 /* activate any spares */
7142 if (mddev->pers->spare_active(mddev))
7143 sysfs_notify(&mddev->kobj, NULL,
7144 "degraded");
7145 }
7146 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7147 mddev->pers->finish_reshape)
7148 mddev->pers->finish_reshape(mddev);
7149 md_update_sb(mddev, 1);
7150
7151 /* if array is no-longer degraded, then any saved_raid_disk
7152 * information must be scrapped
7153 */
7154 if (!mddev->degraded)
7155 list_for_each_entry(rdev, &mddev->disks, same_set)
7156 rdev->saved_raid_disk = -1;
7157
7158 mddev->recovery = 0;
7159 /* flag recovery needed just to double check */
7160 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7161 sysfs_notify_dirent_safe(mddev->sysfs_action);
7162 md_new_event(mddev);
7163 goto unlock;
7164 }
7165 /* Set RUNNING before clearing NEEDED to avoid
7166 * any transients in the value of "sync_action".
7167 */
7168 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7169 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7170 /* Clear some bits that don't mean anything, but
7171 * might be left set
7172 */
7173 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7174 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7175
7176 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7177 goto unlock;
7178 /* no recovery is running.
7179 * remove any failed drives, then
7180 * add spares if possible.
7181 * Spare are also removed and re-added, to allow
7182 * the personality to fail the re-add.
7183 */
7184
7185 if (mddev->reshape_position != MaxSector) {
7186 if (mddev->pers->check_reshape == NULL ||
7187 mddev->pers->check_reshape(mddev) != 0)
7188 /* Cannot proceed */
7189 goto unlock;
7190 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7191 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7192 } else if ((spares = remove_and_add_spares(mddev))) {
7193 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7194 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7195 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7196 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7197 } else if (mddev->recovery_cp < MaxSector) {
7198 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7199 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7200 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7201 /* nothing to be done ... */
7202 goto unlock;
7203
7204 if (mddev->pers->sync_request) {
7205 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7206 /* We are adding a device or devices to an array
7207 * which has the bitmap stored on all devices.
7208 * So make sure all bitmap pages get written
7209 */
7210 bitmap_write_all(mddev->bitmap);
7211 }
7212 mddev->sync_thread = md_register_thread(md_do_sync,
7213 mddev,
7214 "resync");
7215 if (!mddev->sync_thread) {
7216 printk(KERN_ERR "%s: could not start resync"
7217 " thread...\n",
7218 mdname(mddev));
7219 /* leave the spares where they are, it shouldn't hurt */
7220 mddev->recovery = 0;
7221 } else
7222 md_wakeup_thread(mddev->sync_thread);
7223 sysfs_notify_dirent_safe(mddev->sysfs_action);
7224 md_new_event(mddev);
7225 }
7226 unlock:
7227 if (!mddev->sync_thread) {
7228 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7229 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7230 &mddev->recovery))
7231 if (mddev->sysfs_action)
7232 sysfs_notify_dirent_safe(mddev->sysfs_action);
7233 }
7234 mddev_unlock(mddev);
7235 }
7236 }
7237
7238 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7239 {
7240 sysfs_notify_dirent_safe(rdev->sysfs_state);
7241 wait_event_timeout(rdev->blocked_wait,
7242 !test_bit(Blocked, &rdev->flags),
7243 msecs_to_jiffies(5000));
7244 rdev_dec_pending(rdev, mddev);
7245 }
7246 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7247
7248 static int md_notify_reboot(struct notifier_block *this,
7249 unsigned long code, void *x)
7250 {
7251 struct list_head *tmp;
7252 mddev_t *mddev;
7253
7254 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7255
7256 printk(KERN_INFO "md: stopping all md devices.\n");
7257
7258 for_each_mddev(mddev, tmp)
7259 if (mddev_trylock(mddev)) {
7260 /* Force a switch to readonly even array
7261 * appears to still be in use. Hence
7262 * the '100'.
7263 */
7264 md_set_readonly(mddev, 100);
7265 mddev_unlock(mddev);
7266 }
7267 /*
7268 * certain more exotic SCSI devices are known to be
7269 * volatile wrt too early system reboots. While the
7270 * right place to handle this issue is the given
7271 * driver, we do want to have a safe RAID driver ...
7272 */
7273 mdelay(1000*1);
7274 }
7275 return NOTIFY_DONE;
7276 }
7277
7278 static struct notifier_block md_notifier = {
7279 .notifier_call = md_notify_reboot,
7280 .next = NULL,
7281 .priority = INT_MAX, /* before any real devices */
7282 };
7283
7284 static void md_geninit(void)
7285 {
7286 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7287
7288 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
7289 }
7290
7291 static int __init md_init(void)
7292 {
7293 if (register_blkdev(MD_MAJOR, "md"))
7294 return -1;
7295 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
7296 unregister_blkdev(MD_MAJOR, "md");
7297 return -1;
7298 }
7299 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
7300 md_probe, NULL, NULL);
7301 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
7302 md_probe, NULL, NULL);
7303
7304 register_reboot_notifier(&md_notifier);
7305 raid_table_header = register_sysctl_table(raid_root_table);
7306
7307 md_geninit();
7308 return 0;
7309 }
7310
7311
7312 #ifndef MODULE
7313
7314 /*
7315 * Searches all registered partitions for autorun RAID arrays
7316 * at boot time.
7317 */
7318
7319 static LIST_HEAD(all_detected_devices);
7320 struct detected_devices_node {
7321 struct list_head list;
7322 dev_t dev;
7323 };
7324
7325 void md_autodetect_dev(dev_t dev)
7326 {
7327 struct detected_devices_node *node_detected_dev;
7328
7329 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7330 if (node_detected_dev) {
7331 node_detected_dev->dev = dev;
7332 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7333 } else {
7334 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7335 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7336 }
7337 }
7338
7339
7340 static void autostart_arrays(int part)
7341 {
7342 mdk_rdev_t *rdev;
7343 struct detected_devices_node *node_detected_dev;
7344 dev_t dev;
7345 int i_scanned, i_passed;
7346
7347 i_scanned = 0;
7348 i_passed = 0;
7349
7350 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
7351
7352 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7353 i_scanned++;
7354 node_detected_dev = list_entry(all_detected_devices.next,
7355 struct detected_devices_node, list);
7356 list_del(&node_detected_dev->list);
7357 dev = node_detected_dev->dev;
7358 kfree(node_detected_dev);
7359 rdev = md_import_device(dev,0, 90);
7360 if (IS_ERR(rdev))
7361 continue;
7362
7363 if (test_bit(Faulty, &rdev->flags)) {
7364 MD_BUG();
7365 continue;
7366 }
7367 set_bit(AutoDetected, &rdev->flags);
7368 list_add(&rdev->same_set, &pending_raid_disks);
7369 i_passed++;
7370 }
7371
7372 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7373 i_scanned, i_passed);
7374
7375 autorun_devices(part);
7376 }
7377
7378 #endif /* !MODULE */
7379
7380 static __exit void md_exit(void)
7381 {
7382 mddev_t *mddev;
7383 struct list_head *tmp;
7384
7385 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
7386 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
7387
7388 unregister_blkdev(MD_MAJOR,"md");
7389 unregister_blkdev(mdp_major, "mdp");
7390 unregister_reboot_notifier(&md_notifier);
7391 unregister_sysctl_table(raid_table_header);
7392 remove_proc_entry("mdstat", NULL);
7393 for_each_mddev(mddev, tmp) {
7394 export_array(mddev);
7395 mddev->hold_active = 0;
7396 }
7397 }
7398
7399 subsys_initcall(md_init);
7400 module_exit(md_exit)
7401
7402 static int get_ro(char *buffer, struct kernel_param *kp)
7403 {
7404 return sprintf(buffer, "%d", start_readonly);
7405 }
7406 static int set_ro(const char *val, struct kernel_param *kp)
7407 {
7408 char *e;
7409 int num = simple_strtoul(val, &e, 10);
7410 if (*val && (*e == '\0' || *e == '\n')) {
7411 start_readonly = num;
7412 return 0;
7413 }
7414 return -EINVAL;
7415 }
7416
7417 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7418 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
7419
7420 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
7421
7422 EXPORT_SYMBOL(register_md_personality);
7423 EXPORT_SYMBOL(unregister_md_personality);
7424 EXPORT_SYMBOL(md_error);
7425 EXPORT_SYMBOL(md_done_sync);
7426 EXPORT_SYMBOL(md_write_start);
7427 EXPORT_SYMBOL(md_write_end);
7428 EXPORT_SYMBOL(md_register_thread);
7429 EXPORT_SYMBOL(md_unregister_thread);
7430 EXPORT_SYMBOL(md_wakeup_thread);
7431 EXPORT_SYMBOL(md_check_recovery);
7432 MODULE_LICENSE("GPL");
7433 MODULE_DESCRIPTION("MD RAID framework");
7434 MODULE_ALIAS("md");
7435 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Tomato firmware and modifications.