?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mmc: host: sh_mobile_sdhi: move card_busy from tmio to sdhi
[linux/fpc-iii.git] / drivers / md / md.c
blob915e84d631a22111c2256ab2f4676875a2cc61b1
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/badblocks.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
40 #include <linux/fs.h>
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/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
55 #include "md.h"
56 #include "bitmap.h"
57 #include "md-cluster.h"
58
59 #ifndef MODULE
60 static void autostart_arrays(int part);
61 #endif
62
63 /* pers_list is a list of registered personalities protected
64 * by pers_lock.
65 * pers_lock does extra service to protect accesses to
66 * mddev->thread when the mutex cannot be held.
67 */
68 static LIST_HEAD(pers_list);
69 static DEFINE_SPINLOCK(pers_lock);
70
71 struct md_cluster_operations *md_cluster_ops;
72 EXPORT_SYMBOL(md_cluster_ops);
73 struct module *md_cluster_mod;
74 EXPORT_SYMBOL(md_cluster_mod);
75
76 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
77 static struct workqueue_struct *md_wq;
78 static struct workqueue_struct *md_misc_wq;
79
80 static int remove_and_add_spares(struct mddev *mddev,
81 struct md_rdev *this);
82 static void mddev_detach(struct mddev *mddev);
83
84 /*
85 * Default number of read corrections we'll attempt on an rdev
86 * before ejecting it from the array. We divide the read error
87 * count by 2 for every hour elapsed between read errors.
88 */
89 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
90 /*
91 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
92 * is 1000 KB/sec, so the extra system load does not show up that much.
93 * Increase it if you want to have more _guaranteed_ speed. Note that
94 * the RAID driver will use the maximum available bandwidth if the IO
95 * subsystem is idle. There is also an 'absolute maximum' reconstruction
96 * speed limit - in case reconstruction slows down your system despite
97 * idle IO detection.
98 *
99 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
100 * or /sys/block/mdX/md/sync_speed_{min,max}
101 */
102
103 static int sysctl_speed_limit_min = 1000;
104 static int sysctl_speed_limit_max = 200000;
105 static inline int speed_min(struct mddev *mddev)
106 {
107 return mddev->sync_speed_min ?
108 mddev->sync_speed_min : sysctl_speed_limit_min;
109 }
110
111 static inline int speed_max(struct mddev *mddev)
112 {
113 return mddev->sync_speed_max ?
114 mddev->sync_speed_max : sysctl_speed_limit_max;
115 }
116
117 static struct ctl_table_header *raid_table_header;
118
119 static struct ctl_table raid_table[] = {
120 {
121 .procname = "speed_limit_min",
122 .data = &sysctl_speed_limit_min,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
126 },
127 {
128 .procname = "speed_limit_max",
129 .data = &sysctl_speed_limit_max,
130 .maxlen = sizeof(int),
131 .mode = S_IRUGO|S_IWUSR,
132 .proc_handler = proc_dointvec,
133 },
134 { }
135 };
136
137 static struct ctl_table raid_dir_table[] = {
138 {
139 .procname = "raid",
140 .maxlen = 0,
141 .mode = S_IRUGO|S_IXUGO,
142 .child = raid_table,
143 },
144 { }
145 };
146
147 static struct ctl_table raid_root_table[] = {
148 {
149 .procname = "dev",
150 .maxlen = 0,
151 .mode = 0555,
152 .child = raid_dir_table,
153 },
154 { }
155 };
156
157 static const struct block_device_operations md_fops;
158
159 static int start_readonly;
160
161 /* bio_clone_mddev
162 * like bio_clone, but with a local bio set
163 */
164
165 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
166 struct mddev *mddev)
167 {
168 struct bio *b;
169
170 if (!mddev || !mddev->bio_set)
171 return bio_alloc(gfp_mask, nr_iovecs);
172
173 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
174 if (!b)
175 return NULL;
176 return b;
177 }
178 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
179
180 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
181 struct mddev *mddev)
182 {
183 if (!mddev || !mddev->bio_set)
184 return bio_clone(bio, gfp_mask);
185
186 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
187 }
188 EXPORT_SYMBOL_GPL(bio_clone_mddev);
189
190 /*
191 * We have a system wide 'event count' that is incremented
192 * on any 'interesting' event, and readers of /proc/mdstat
193 * can use 'poll' or 'select' to find out when the event
194 * count increases.
195 *
196 * Events are:
197 * start array, stop array, error, add device, remove device,
198 * start build, activate spare
199 */
200 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
201 static atomic_t md_event_count;
202 void md_new_event(struct mddev *mddev)
203 {
204 atomic_inc(&md_event_count);
205 wake_up(&md_event_waiters);
206 }
207 EXPORT_SYMBOL_GPL(md_new_event);
208
209 /*
210 * Enables to iterate over all existing md arrays
211 * all_mddevs_lock protects this list.
212 */
213 static LIST_HEAD(all_mddevs);
214 static DEFINE_SPINLOCK(all_mddevs_lock);
215
216 /*
217 * iterates through all used mddevs in the system.
218 * We take care to grab the all_mddevs_lock whenever navigating
219 * the list, and to always hold a refcount when unlocked.
220 * Any code which breaks out of this loop while own
221 * a reference to the current mddev and must mddev_put it.
222 */
223 #define for_each_mddev(_mddev,_tmp) \
224 \
225 for (({ spin_lock(&all_mddevs_lock); \
226 _tmp = all_mddevs.next; \
227 _mddev = NULL;}); \
228 ({ if (_tmp != &all_mddevs) \
229 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
230 spin_unlock(&all_mddevs_lock); \
231 if (_mddev) mddev_put(_mddev); \
232 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
233 _tmp != &all_mddevs;}); \
234 ({ spin_lock(&all_mddevs_lock); \
235 _tmp = _tmp->next;}) \
236 )
237
238 /* Rather than calling directly into the personality make_request function,
239 * IO requests come here first so that we can check if the device is
240 * being suspended pending a reconfiguration.
241 * We hold a refcount over the call to ->make_request. By the time that
242 * call has finished, the bio has been linked into some internal structure
243 * and so is visible to ->quiesce(), so we don't need the refcount any more.
244 */
245 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
246 {
247 const int rw = bio_data_dir(bio);
248 struct mddev *mddev = q->queuedata;
249 unsigned int sectors;
250 int cpu;
251
252 blk_queue_split(q, &bio, q->bio_split);
253
254 if (mddev == NULL || mddev->pers == NULL) {
255 bio_io_error(bio);
256 return BLK_QC_T_NONE;
257 }
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 if (bio_sectors(bio) != 0)
260 bio->bi_error = -EROFS;
261 bio_endio(bio);
262 return BLK_QC_T_NONE;
263 }
264 smp_rmb(); /* Ensure implications of 'active' are visible */
265 rcu_read_lock();
266 if (mddev->suspended) {
267 DEFINE_WAIT(__wait);
268 for (;;) {
269 prepare_to_wait(&mddev->sb_wait, &__wait,
270 TASK_UNINTERRUPTIBLE);
271 if (!mddev->suspended)
272 break;
273 rcu_read_unlock();
274 schedule();
275 rcu_read_lock();
276 }
277 finish_wait(&mddev->sb_wait, &__wait);
278 }
279 atomic_inc(&mddev->active_io);
280 rcu_read_unlock();
281
282 /*
283 * save the sectors now since our bio can
284 * go away inside make_request
285 */
286 sectors = bio_sectors(bio);
287 /* bio could be mergeable after passing to underlayer */
288 bio->bi_opf &= ~REQ_NOMERGE;
289 mddev->pers->make_request(mddev, bio);
290
291 cpu = part_stat_lock();
292 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
293 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
294 part_stat_unlock();
295
296 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
297 wake_up(&mddev->sb_wait);
298
299 return BLK_QC_T_NONE;
300 }
301
302 /* mddev_suspend makes sure no new requests are submitted
303 * to the device, and that any requests that have been submitted
304 * are completely handled.
305 * Once mddev_detach() is called and completes, the module will be
306 * completely unused.
307 */
308 void mddev_suspend(struct mddev *mddev)
309 {
310 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
311 if (mddev->suspended++)
312 return;
313 synchronize_rcu();
314 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
315 mddev->pers->quiesce(mddev, 1);
316
317 del_timer_sync(&mddev->safemode_timer);
318 }
319 EXPORT_SYMBOL_GPL(mddev_suspend);
320
321 void mddev_resume(struct mddev *mddev)
322 {
323 if (--mddev->suspended)
324 return;
325 wake_up(&mddev->sb_wait);
326 mddev->pers->quiesce(mddev, 0);
327
328 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
329 md_wakeup_thread(mddev->thread);
330 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
331 }
332 EXPORT_SYMBOL_GPL(mddev_resume);
333
334 int mddev_congested(struct mddev *mddev, int bits)
335 {
336 struct md_personality *pers = mddev->pers;
337 int ret = 0;
338
339 rcu_read_lock();
340 if (mddev->suspended)
341 ret = 1;
342 else if (pers && pers->congested)
343 ret = pers->congested(mddev, bits);
344 rcu_read_unlock();
345 return ret;
346 }
347 EXPORT_SYMBOL_GPL(mddev_congested);
348 static int md_congested(void *data, int bits)
349 {
350 struct mddev *mddev = data;
351 return mddev_congested(mddev, bits);
352 }
353
354 /*
355 * Generic flush handling for md
356 */
357
358 static void md_end_flush(struct bio *bio)
359 {
360 struct md_rdev *rdev = bio->bi_private;
361 struct mddev *mddev = rdev->mddev;
362
363 rdev_dec_pending(rdev, mddev);
364
365 if (atomic_dec_and_test(&mddev->flush_pending)) {
366 /* The pre-request flush has finished */
367 queue_work(md_wq, &mddev->flush_work);
368 }
369 bio_put(bio);
370 }
371
372 static void md_submit_flush_data(struct work_struct *ws);
373
374 static void submit_flushes(struct work_struct *ws)
375 {
376 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
377 struct md_rdev *rdev;
378
379 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
380 atomic_set(&mddev->flush_pending, 1);
381 rcu_read_lock();
382 rdev_for_each_rcu(rdev, mddev)
383 if (rdev->raid_disk >= 0 &&
384 !test_bit(Faulty, &rdev->flags)) {
385 /* Take two references, one is dropped
386 * when request finishes, one after
387 * we reclaim rcu_read_lock
388 */
389 struct bio *bi;
390 atomic_inc(&rdev->nr_pending);
391 atomic_inc(&rdev->nr_pending);
392 rcu_read_unlock();
393 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
394 bi->bi_end_io = md_end_flush;
395 bi->bi_private = rdev;
396 bi->bi_bdev = rdev->bdev;
397 bio_set_op_attrs(bi, REQ_OP_WRITE, WRITE_FLUSH);
398 atomic_inc(&mddev->flush_pending);
399 submit_bio(bi);
400 rcu_read_lock();
401 rdev_dec_pending(rdev, mddev);
402 }
403 rcu_read_unlock();
404 if (atomic_dec_and_test(&mddev->flush_pending))
405 queue_work(md_wq, &mddev->flush_work);
406 }
407
408 static void md_submit_flush_data(struct work_struct *ws)
409 {
410 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
411 struct bio *bio = mddev->flush_bio;
412
413 if (bio->bi_iter.bi_size == 0)
414 /* an empty barrier - all done */
415 bio_endio(bio);
416 else {
417 bio->bi_opf &= ~REQ_PREFLUSH;
418 mddev->pers->make_request(mddev, bio);
419 }
420
421 mddev->flush_bio = NULL;
422 wake_up(&mddev->sb_wait);
423 }
424
425 void md_flush_request(struct mddev *mddev, struct bio *bio)
426 {
427 spin_lock_irq(&mddev->lock);
428 wait_event_lock_irq(mddev->sb_wait,
429 !mddev->flush_bio,
430 mddev->lock);
431 mddev->flush_bio = bio;
432 spin_unlock_irq(&mddev->lock);
433
434 INIT_WORK(&mddev->flush_work, submit_flushes);
435 queue_work(md_wq, &mddev->flush_work);
436 }
437 EXPORT_SYMBOL(md_flush_request);
438
439 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
440 {
441 struct mddev *mddev = cb->data;
442 md_wakeup_thread(mddev->thread);
443 kfree(cb);
444 }
445 EXPORT_SYMBOL(md_unplug);
446
447 static inline struct mddev *mddev_get(struct mddev *mddev)
448 {
449 atomic_inc(&mddev->active);
450 return mddev;
451 }
452
453 static void mddev_delayed_delete(struct work_struct *ws);
454
455 static void mddev_put(struct mddev *mddev)
456 {
457 struct bio_set *bs = NULL;
458
459 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
460 return;
461 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
462 mddev->ctime == 0 && !mddev->hold_active) {
463 /* Array is not configured at all, and not held active,
464 * so destroy it */
465 list_del_init(&mddev->all_mddevs);
466 bs = mddev->bio_set;
467 mddev->bio_set = NULL;
468 if (mddev->gendisk) {
469 /* We did a probe so need to clean up. Call
470 * queue_work inside the spinlock so that
471 * flush_workqueue() after mddev_find will
472 * succeed in waiting for the work to be done.
473 */
474 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
475 queue_work(md_misc_wq, &mddev->del_work);
476 } else
477 kfree(mddev);
478 }
479 spin_unlock(&all_mddevs_lock);
480 if (bs)
481 bioset_free(bs);
482 }
483
484 static void md_safemode_timeout(unsigned long data);
485
486 void mddev_init(struct mddev *mddev)
487 {
488 mutex_init(&mddev->open_mutex);
489 mutex_init(&mddev->reconfig_mutex);
490 mutex_init(&mddev->bitmap_info.mutex);
491 INIT_LIST_HEAD(&mddev->disks);
492 INIT_LIST_HEAD(&mddev->all_mddevs);
493 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
494 (unsigned long) mddev);
495 atomic_set(&mddev->active, 1);
496 atomic_set(&mddev->openers, 0);
497 atomic_set(&mddev->active_io, 0);
498 spin_lock_init(&mddev->lock);
499 atomic_set(&mddev->flush_pending, 0);
500 init_waitqueue_head(&mddev->sb_wait);
501 init_waitqueue_head(&mddev->recovery_wait);
502 mddev->reshape_position = MaxSector;
503 mddev->reshape_backwards = 0;
504 mddev->last_sync_action = "none";
505 mddev->resync_min = 0;
506 mddev->resync_max = MaxSector;
507 mddev->level = LEVEL_NONE;
508 }
509 EXPORT_SYMBOL_GPL(mddev_init);
510
511 static struct mddev *mddev_find(dev_t unit)
512 {
513 struct mddev *mddev, *new = NULL;
514
515 if (unit && MAJOR(unit) != MD_MAJOR)
516 unit &= ~((1<<MdpMinorShift)-1);
517
518 retry:
519 spin_lock(&all_mddevs_lock);
520
521 if (unit) {
522 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
523 if (mddev->unit == unit) {
524 mddev_get(mddev);
525 spin_unlock(&all_mddevs_lock);
526 kfree(new);
527 return mddev;
528 }
529
530 if (new) {
531 list_add(&new->all_mddevs, &all_mddevs);
532 spin_unlock(&all_mddevs_lock);
533 new->hold_active = UNTIL_IOCTL;
534 return new;
535 }
536 } else if (new) {
537 /* find an unused unit number */
538 static int next_minor = 512;
539 int start = next_minor;
540 int is_free = 0;
541 int dev = 0;
542 while (!is_free) {
543 dev = MKDEV(MD_MAJOR, next_minor);
544 next_minor++;
545 if (next_minor > MINORMASK)
546 next_minor = 0;
547 if (next_minor == start) {
548 /* Oh dear, all in use. */
549 spin_unlock(&all_mddevs_lock);
550 kfree(new);
551 return NULL;
552 }
553
554 is_free = 1;
555 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
556 if (mddev->unit == dev) {
557 is_free = 0;
558 break;
559 }
560 }
561 new->unit = dev;
562 new->md_minor = MINOR(dev);
563 new->hold_active = UNTIL_STOP;
564 list_add(&new->all_mddevs, &all_mddevs);
565 spin_unlock(&all_mddevs_lock);
566 return new;
567 }
568 spin_unlock(&all_mddevs_lock);
569
570 new = kzalloc(sizeof(*new), GFP_KERNEL);
571 if (!new)
572 return NULL;
573
574 new->unit = unit;
575 if (MAJOR(unit) == MD_MAJOR)
576 new->md_minor = MINOR(unit);
577 else
578 new->md_minor = MINOR(unit) >> MdpMinorShift;
579
580 mddev_init(new);
581
582 goto retry;
583 }
584
585 static struct attribute_group md_redundancy_group;
586
587 void mddev_unlock(struct mddev *mddev)
588 {
589 if (mddev->to_remove) {
590 /* These cannot be removed under reconfig_mutex as
591 * an access to the files will try to take reconfig_mutex
592 * while holding the file unremovable, which leads to
593 * a deadlock.
594 * So hold set sysfs_active while the remove in happeing,
595 * and anything else which might set ->to_remove or my
596 * otherwise change the sysfs namespace will fail with
597 * -EBUSY if sysfs_active is still set.
598 * We set sysfs_active under reconfig_mutex and elsewhere
599 * test it under the same mutex to ensure its correct value
600 * is seen.
601 */
602 struct attribute_group *to_remove = mddev->to_remove;
603 mddev->to_remove = NULL;
604 mddev->sysfs_active = 1;
605 mutex_unlock(&mddev->reconfig_mutex);
606
607 if (mddev->kobj.sd) {
608 if (to_remove != &md_redundancy_group)
609 sysfs_remove_group(&mddev->kobj, to_remove);
610 if (mddev->pers == NULL ||
611 mddev->pers->sync_request == NULL) {
612 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
613 if (mddev->sysfs_action)
614 sysfs_put(mddev->sysfs_action);
615 mddev->sysfs_action = NULL;
616 }
617 }
618 mddev->sysfs_active = 0;
619 } else
620 mutex_unlock(&mddev->reconfig_mutex);
621
622 /* As we've dropped the mutex we need a spinlock to
623 * make sure the thread doesn't disappear
624 */
625 spin_lock(&pers_lock);
626 md_wakeup_thread(mddev->thread);
627 spin_unlock(&pers_lock);
628 }
629 EXPORT_SYMBOL_GPL(mddev_unlock);
630
631 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
632 {
633 struct md_rdev *rdev;
634
635 rdev_for_each_rcu(rdev, mddev)
636 if (rdev->desc_nr == nr)
637 return rdev;
638
639 return NULL;
640 }
641 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
642
643 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
644 {
645 struct md_rdev *rdev;
646
647 rdev_for_each(rdev, mddev)
648 if (rdev->bdev->bd_dev == dev)
649 return rdev;
650
651 return NULL;
652 }
653
654 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
655 {
656 struct md_rdev *rdev;
657
658 rdev_for_each_rcu(rdev, mddev)
659 if (rdev->bdev->bd_dev == dev)
660 return rdev;
661
662 return NULL;
663 }
664
665 static struct md_personality *find_pers(int level, char *clevel)
666 {
667 struct md_personality *pers;
668 list_for_each_entry(pers, &pers_list, list) {
669 if (level != LEVEL_NONE && pers->level == level)
670 return pers;
671 if (strcmp(pers->name, clevel)==0)
672 return pers;
673 }
674 return NULL;
675 }
676
677 /* return the offset of the super block in 512byte sectors */
678 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
679 {
680 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
681 return MD_NEW_SIZE_SECTORS(num_sectors);
682 }
683
684 static int alloc_disk_sb(struct md_rdev *rdev)
685 {
686 rdev->sb_page = alloc_page(GFP_KERNEL);
687 if (!rdev->sb_page) {
688 printk(KERN_ALERT "md: out of memory.\n");
689 return -ENOMEM;
690 }
691
692 return 0;
693 }
694
695 void md_rdev_clear(struct md_rdev *rdev)
696 {
697 if (rdev->sb_page) {
698 put_page(rdev->sb_page);
699 rdev->sb_loaded = 0;
700 rdev->sb_page = NULL;
701 rdev->sb_start = 0;
702 rdev->sectors = 0;
703 }
704 if (rdev->bb_page) {
705 put_page(rdev->bb_page);
706 rdev->bb_page = NULL;
707 }
708 badblocks_exit(&rdev->badblocks);
709 }
710 EXPORT_SYMBOL_GPL(md_rdev_clear);
711
712 static void super_written(struct bio *bio)
713 {
714 struct md_rdev *rdev = bio->bi_private;
715 struct mddev *mddev = rdev->mddev;
716
717 if (bio->bi_error) {
718 printk("md: super_written gets error=%d\n", bio->bi_error);
719 md_error(mddev, rdev);
720 }
721
722 if (atomic_dec_and_test(&mddev->pending_writes))
723 wake_up(&mddev->sb_wait);
724 rdev_dec_pending(rdev, mddev);
725 bio_put(bio);
726 }
727
728 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
729 sector_t sector, int size, struct page *page)
730 {
731 /* write first size bytes of page to sector of rdev
732 * Increment mddev->pending_writes before returning
733 * and decrement it on completion, waking up sb_wait
734 * if zero is reached.
735 * If an error occurred, call md_error
736 */
737 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
738
739 atomic_inc(&rdev->nr_pending);
740
741 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
742 bio->bi_iter.bi_sector = sector;
743 bio_add_page(bio, page, size, 0);
744 bio->bi_private = rdev;
745 bio->bi_end_io = super_written;
746 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH_FUA);
747
748 atomic_inc(&mddev->pending_writes);
749 submit_bio(bio);
750 }
751
752 void md_super_wait(struct mddev *mddev)
753 {
754 /* wait for all superblock writes that were scheduled to complete */
755 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
756 }
757
758 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
759 struct page *page, int op, int op_flags, bool metadata_op)
760 {
761 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
762 int ret;
763
764 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
765 rdev->meta_bdev : rdev->bdev;
766 bio_set_op_attrs(bio, op, op_flags);
767 if (metadata_op)
768 bio->bi_iter.bi_sector = sector + rdev->sb_start;
769 else if (rdev->mddev->reshape_position != MaxSector &&
770 (rdev->mddev->reshape_backwards ==
771 (sector >= rdev->mddev->reshape_position)))
772 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
773 else
774 bio->bi_iter.bi_sector = sector + rdev->data_offset;
775 bio_add_page(bio, page, size, 0);
776
777 submit_bio_wait(bio);
778
779 ret = !bio->bi_error;
780 bio_put(bio);
781 return ret;
782 }
783 EXPORT_SYMBOL_GPL(sync_page_io);
784
785 static int read_disk_sb(struct md_rdev *rdev, int size)
786 {
787 char b[BDEVNAME_SIZE];
788
789 if (rdev->sb_loaded)
790 return 0;
791
792 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
793 goto fail;
794 rdev->sb_loaded = 1;
795 return 0;
796
797 fail:
798 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
799 bdevname(rdev->bdev,b));
800 return -EINVAL;
801 }
802
803 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
804 {
805 return sb1->set_uuid0 == sb2->set_uuid0 &&
806 sb1->set_uuid1 == sb2->set_uuid1 &&
807 sb1->set_uuid2 == sb2->set_uuid2 &&
808 sb1->set_uuid3 == sb2->set_uuid3;
809 }
810
811 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
812 {
813 int ret;
814 mdp_super_t *tmp1, *tmp2;
815
816 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
817 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
818
819 if (!tmp1 || !tmp2) {
820 ret = 0;
821 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
822 goto abort;
823 }
824
825 *tmp1 = *sb1;
826 *tmp2 = *sb2;
827
828 /*
829 * nr_disks is not constant
830 */
831 tmp1->nr_disks = 0;
832 tmp2->nr_disks = 0;
833
834 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
835 abort:
836 kfree(tmp1);
837 kfree(tmp2);
838 return ret;
839 }
840
841 static u32 md_csum_fold(u32 csum)
842 {
843 csum = (csum & 0xffff) + (csum >> 16);
844 return (csum & 0xffff) + (csum >> 16);
845 }
846
847 static unsigned int calc_sb_csum(mdp_super_t *sb)
848 {
849 u64 newcsum = 0;
850 u32 *sb32 = (u32*)sb;
851 int i;
852 unsigned int disk_csum, csum;
853
854 disk_csum = sb->sb_csum;
855 sb->sb_csum = 0;
856
857 for (i = 0; i < MD_SB_BYTES/4 ; i++)
858 newcsum += sb32[i];
859 csum = (newcsum & 0xffffffff) + (newcsum>>32);
860
861 #ifdef CONFIG_ALPHA
862 /* This used to use csum_partial, which was wrong for several
863 * reasons including that different results are returned on
864 * different architectures. It isn't critical that we get exactly
865 * the same return value as before (we always csum_fold before
866 * testing, and that removes any differences). However as we
867 * know that csum_partial always returned a 16bit value on
868 * alphas, do a fold to maximise conformity to previous behaviour.
869 */
870 sb->sb_csum = md_csum_fold(disk_csum);
871 #else
872 sb->sb_csum = disk_csum;
873 #endif
874 return csum;
875 }
876
877 /*
878 * Handle superblock details.
879 * We want to be able to handle multiple superblock formats
880 * so we have a common interface to them all, and an array of
881 * different handlers.
882 * We rely on user-space to write the initial superblock, and support
883 * reading and updating of superblocks.
884 * Interface methods are:
885 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
886 * loads and validates a superblock on dev.
887 * if refdev != NULL, compare superblocks on both devices
888 * Return:
889 * 0 - dev has a superblock that is compatible with refdev
890 * 1 - dev has a superblock that is compatible and newer than refdev
891 * so dev should be used as the refdev in future
892 * -EINVAL superblock incompatible or invalid
893 * -othererror e.g. -EIO
894 *
895 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
896 * Verify that dev is acceptable into mddev.
897 * The first time, mddev->raid_disks will be 0, and data from
898 * dev should be merged in. Subsequent calls check that dev
899 * is new enough. Return 0 or -EINVAL
900 *
901 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
902 * Update the superblock for rdev with data in mddev
903 * This does not write to disc.
904 *
905 */
906
907 struct super_type {
908 char *name;
909 struct module *owner;
910 int (*load_super)(struct md_rdev *rdev,
911 struct md_rdev *refdev,
912 int minor_version);
913 int (*validate_super)(struct mddev *mddev,
914 struct md_rdev *rdev);
915 void (*sync_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
918 sector_t num_sectors);
919 int (*allow_new_offset)(struct md_rdev *rdev,
920 unsigned long long new_offset);
921 };
922
923 /*
924 * Check that the given mddev has no bitmap.
925 *
926 * This function is called from the run method of all personalities that do not
927 * support bitmaps. It prints an error message and returns non-zero if mddev
928 * has a bitmap. Otherwise, it returns 0.
929 *
930 */
931 int md_check_no_bitmap(struct mddev *mddev)
932 {
933 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
934 return 0;
935 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
936 mdname(mddev), mddev->pers->name);
937 return 1;
938 }
939 EXPORT_SYMBOL(md_check_no_bitmap);
940
941 /*
942 * load_super for 0.90.0
943 */
944 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
945 {
946 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
947 mdp_super_t *sb;
948 int ret;
949
950 /*
951 * Calculate the position of the superblock (512byte sectors),
952 * it's at the end of the disk.
953 *
954 * It also happens to be a multiple of 4Kb.
955 */
956 rdev->sb_start = calc_dev_sboffset(rdev);
957
958 ret = read_disk_sb(rdev, MD_SB_BYTES);
959 if (ret) return ret;
960
961 ret = -EINVAL;
962
963 bdevname(rdev->bdev, b);
964 sb = page_address(rdev->sb_page);
965
966 if (sb->md_magic != MD_SB_MAGIC) {
967 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
968 b);
969 goto abort;
970 }
971
972 if (sb->major_version != 0 ||
973 sb->minor_version < 90 ||
974 sb->minor_version > 91) {
975 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
976 sb->major_version, sb->minor_version,
977 b);
978 goto abort;
979 }
980
981 if (sb->raid_disks <= 0)
982 goto abort;
983
984 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
985 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
986 b);
987 goto abort;
988 }
989
990 rdev->preferred_minor = sb->md_minor;
991 rdev->data_offset = 0;
992 rdev->new_data_offset = 0;
993 rdev->sb_size = MD_SB_BYTES;
994 rdev->badblocks.shift = -1;
995
996 if (sb->level == LEVEL_MULTIPATH)
997 rdev->desc_nr = -1;
998 else
999 rdev->desc_nr = sb->this_disk.number;
1000
1001 if (!refdev) {
1002 ret = 1;
1003 } else {
1004 __u64 ev1, ev2;
1005 mdp_super_t *refsb = page_address(refdev->sb_page);
1006 if (!uuid_equal(refsb, sb)) {
1007 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1008 b, bdevname(refdev->bdev,b2));
1009 goto abort;
1010 }
1011 if (!sb_equal(refsb, sb)) {
1012 printk(KERN_WARNING "md: %s has same UUID"
1013 " but different superblock to %s\n",
1014 b, bdevname(refdev->bdev, b2));
1015 goto abort;
1016 }
1017 ev1 = md_event(sb);
1018 ev2 = md_event(refsb);
1019 if (ev1 > ev2)
1020 ret = 1;
1021 else
1022 ret = 0;
1023 }
1024 rdev->sectors = rdev->sb_start;
1025 /* Limit to 4TB as metadata cannot record more than that.
1026 * (not needed for Linear and RAID0 as metadata doesn't
1027 * record this size)
1028 */
1029 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1030 sb->level >= 1)
1031 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1032
1033 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1034 /* "this cannot possibly happen" ... */
1035 ret = -EINVAL;
1036
1037 abort:
1038 return ret;
1039 }
1040
1041 /*
1042 * validate_super for 0.90.0
1043 */
1044 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1045 {
1046 mdp_disk_t *desc;
1047 mdp_super_t *sb = page_address(rdev->sb_page);
1048 __u64 ev1 = md_event(sb);
1049
1050 rdev->raid_disk = -1;
1051 clear_bit(Faulty, &rdev->flags);
1052 clear_bit(In_sync, &rdev->flags);
1053 clear_bit(Bitmap_sync, &rdev->flags);
1054 clear_bit(WriteMostly, &rdev->flags);
1055
1056 if (mddev->raid_disks == 0) {
1057 mddev->major_version = 0;
1058 mddev->minor_version = sb->minor_version;
1059 mddev->patch_version = sb->patch_version;
1060 mddev->external = 0;
1061 mddev->chunk_sectors = sb->chunk_size >> 9;
1062 mddev->ctime = sb->ctime;
1063 mddev->utime = sb->utime;
1064 mddev->level = sb->level;
1065 mddev->clevel[0] = 0;
1066 mddev->layout = sb->layout;
1067 mddev->raid_disks = sb->raid_disks;
1068 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1069 mddev->events = ev1;
1070 mddev->bitmap_info.offset = 0;
1071 mddev->bitmap_info.space = 0;
1072 /* bitmap can use 60 K after the 4K superblocks */
1073 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1074 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1075 mddev->reshape_backwards = 0;
1076
1077 if (mddev->minor_version >= 91) {
1078 mddev->reshape_position = sb->reshape_position;
1079 mddev->delta_disks = sb->delta_disks;
1080 mddev->new_level = sb->new_level;
1081 mddev->new_layout = sb->new_layout;
1082 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1083 if (mddev->delta_disks < 0)
1084 mddev->reshape_backwards = 1;
1085 } else {
1086 mddev->reshape_position = MaxSector;
1087 mddev->delta_disks = 0;
1088 mddev->new_level = mddev->level;
1089 mddev->new_layout = mddev->layout;
1090 mddev->new_chunk_sectors = mddev->chunk_sectors;
1091 }
1092
1093 if (sb->state & (1<<MD_SB_CLEAN))
1094 mddev->recovery_cp = MaxSector;
1095 else {
1096 if (sb->events_hi == sb->cp_events_hi &&
1097 sb->events_lo == sb->cp_events_lo) {
1098 mddev->recovery_cp = sb->recovery_cp;
1099 } else
1100 mddev->recovery_cp = 0;
1101 }
1102
1103 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1104 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1105 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1106 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1107
1108 mddev->max_disks = MD_SB_DISKS;
1109
1110 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1111 mddev->bitmap_info.file == NULL) {
1112 mddev->bitmap_info.offset =
1113 mddev->bitmap_info.default_offset;
1114 mddev->bitmap_info.space =
1115 mddev->bitmap_info.default_space;
1116 }
1117
1118 } else if (mddev->pers == NULL) {
1119 /* Insist on good event counter while assembling, except
1120 * for spares (which don't need an event count) */
1121 ++ev1;
1122 if (sb->disks[rdev->desc_nr].state & (
1123 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1124 if (ev1 < mddev->events)
1125 return -EINVAL;
1126 } else if (mddev->bitmap) {
1127 /* if adding to array with a bitmap, then we can accept an
1128 * older device ... but not too old.
1129 */
1130 if (ev1 < mddev->bitmap->events_cleared)
1131 return 0;
1132 if (ev1 < mddev->events)
1133 set_bit(Bitmap_sync, &rdev->flags);
1134 } else {
1135 if (ev1 < mddev->events)
1136 /* just a hot-add of a new device, leave raid_disk at -1 */
1137 return 0;
1138 }
1139
1140 if (mddev->level != LEVEL_MULTIPATH) {
1141 desc = sb->disks + rdev->desc_nr;
1142
1143 if (desc->state & (1<<MD_DISK_FAULTY))
1144 set_bit(Faulty, &rdev->flags);
1145 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1146 desc->raid_disk < mddev->raid_disks */) {
1147 set_bit(In_sync, &rdev->flags);
1148 rdev->raid_disk = desc->raid_disk;
1149 rdev->saved_raid_disk = desc->raid_disk;
1150 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1151 /* active but not in sync implies recovery up to
1152 * reshape position. We don't know exactly where
1153 * that is, so set to zero for now */
1154 if (mddev->minor_version >= 91) {
1155 rdev->recovery_offset = 0;
1156 rdev->raid_disk = desc->raid_disk;
1157 }
1158 }
1159 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1160 set_bit(WriteMostly, &rdev->flags);
1161 } else /* MULTIPATH are always insync */
1162 set_bit(In_sync, &rdev->flags);
1163 return 0;
1164 }
1165
1166 /*
1167 * sync_super for 0.90.0
1168 */
1169 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1170 {
1171 mdp_super_t *sb;
1172 struct md_rdev *rdev2;
1173 int next_spare = mddev->raid_disks;
1174
1175 /* make rdev->sb match mddev data..
1176 *
1177 * 1/ zero out disks
1178 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1179 * 3/ any empty disks < next_spare become removed
1180 *
1181 * disks[0] gets initialised to REMOVED because
1182 * we cannot be sure from other fields if it has
1183 * been initialised or not.
1184 */
1185 int i;
1186 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1187
1188 rdev->sb_size = MD_SB_BYTES;
1189
1190 sb = page_address(rdev->sb_page);
1191
1192 memset(sb, 0, sizeof(*sb));
1193
1194 sb->md_magic = MD_SB_MAGIC;
1195 sb->major_version = mddev->major_version;
1196 sb->patch_version = mddev->patch_version;
1197 sb->gvalid_words = 0; /* ignored */
1198 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1199 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1200 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1201 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1202
1203 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1204 sb->level = mddev->level;
1205 sb->size = mddev->dev_sectors / 2;
1206 sb->raid_disks = mddev->raid_disks;
1207 sb->md_minor = mddev->md_minor;
1208 sb->not_persistent = 0;
1209 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1210 sb->state = 0;
1211 sb->events_hi = (mddev->events>>32);
1212 sb->events_lo = (u32)mddev->events;
1213
1214 if (mddev->reshape_position == MaxSector)
1215 sb->minor_version = 90;
1216 else {
1217 sb->minor_version = 91;
1218 sb->reshape_position = mddev->reshape_position;
1219 sb->new_level = mddev->new_level;
1220 sb->delta_disks = mddev->delta_disks;
1221 sb->new_layout = mddev->new_layout;
1222 sb->new_chunk = mddev->new_chunk_sectors << 9;
1223 }
1224 mddev->minor_version = sb->minor_version;
1225 if (mddev->in_sync)
1226 {
1227 sb->recovery_cp = mddev->recovery_cp;
1228 sb->cp_events_hi = (mddev->events>>32);
1229 sb->cp_events_lo = (u32)mddev->events;
1230 if (mddev->recovery_cp == MaxSector)
1231 sb->state = (1<< MD_SB_CLEAN);
1232 } else
1233 sb->recovery_cp = 0;
1234
1235 sb->layout = mddev->layout;
1236 sb->chunk_size = mddev->chunk_sectors << 9;
1237
1238 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1239 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1240
1241 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1242 rdev_for_each(rdev2, mddev) {
1243 mdp_disk_t *d;
1244 int desc_nr;
1245 int is_active = test_bit(In_sync, &rdev2->flags);
1246
1247 if (rdev2->raid_disk >= 0 &&
1248 sb->minor_version >= 91)
1249 /* we have nowhere to store the recovery_offset,
1250 * but if it is not below the reshape_position,
1251 * we can piggy-back on that.
1252 */
1253 is_active = 1;
1254 if (rdev2->raid_disk < 0 ||
1255 test_bit(Faulty, &rdev2->flags))
1256 is_active = 0;
1257 if (is_active)
1258 desc_nr = rdev2->raid_disk;
1259 else
1260 desc_nr = next_spare++;
1261 rdev2->desc_nr = desc_nr;
1262 d = &sb->disks[rdev2->desc_nr];
1263 nr_disks++;
1264 d->number = rdev2->desc_nr;
1265 d->major = MAJOR(rdev2->bdev->bd_dev);
1266 d->minor = MINOR(rdev2->bdev->bd_dev);
1267 if (is_active)
1268 d->raid_disk = rdev2->raid_disk;
1269 else
1270 d->raid_disk = rdev2->desc_nr; /* compatibility */
1271 if (test_bit(Faulty, &rdev2->flags))
1272 d->state = (1<<MD_DISK_FAULTY);
1273 else if (is_active) {
1274 d->state = (1<<MD_DISK_ACTIVE);
1275 if (test_bit(In_sync, &rdev2->flags))
1276 d->state |= (1<<MD_DISK_SYNC);
1277 active++;
1278 working++;
1279 } else {
1280 d->state = 0;
1281 spare++;
1282 working++;
1283 }
1284 if (test_bit(WriteMostly, &rdev2->flags))
1285 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1286 }
1287 /* now set the "removed" and "faulty" bits on any missing devices */
1288 for (i=0 ; i < mddev->raid_disks ; i++) {
1289 mdp_disk_t *d = &sb->disks[i];
1290 if (d->state == 0 && d->number == 0) {
1291 d->number = i;
1292 d->raid_disk = i;
1293 d->state = (1<<MD_DISK_REMOVED);
1294 d->state |= (1<<MD_DISK_FAULTY);
1295 failed++;
1296 }
1297 }
1298 sb->nr_disks = nr_disks;
1299 sb->active_disks = active;
1300 sb->working_disks = working;
1301 sb->failed_disks = failed;
1302 sb->spare_disks = spare;
1303
1304 sb->this_disk = sb->disks[rdev->desc_nr];
1305 sb->sb_csum = calc_sb_csum(sb);
1306 }
1307
1308 /*
1309 * rdev_size_change for 0.90.0
1310 */
1311 static unsigned long long
1312 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1313 {
1314 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1315 return 0; /* component must fit device */
1316 if (rdev->mddev->bitmap_info.offset)
1317 return 0; /* can't move bitmap */
1318 rdev->sb_start = calc_dev_sboffset(rdev);
1319 if (!num_sectors || num_sectors > rdev->sb_start)
1320 num_sectors = rdev->sb_start;
1321 /* Limit to 4TB as metadata cannot record more than that.
1322 * 4TB == 2^32 KB, or 2*2^32 sectors.
1323 */
1324 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1325 rdev->mddev->level >= 1)
1326 num_sectors = (sector_t)(2ULL << 32) - 2;
1327 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1328 rdev->sb_page);
1329 md_super_wait(rdev->mddev);
1330 return num_sectors;
1331 }
1332
1333 static int
1334 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1335 {
1336 /* non-zero offset changes not possible with v0.90 */
1337 return new_offset == 0;
1338 }
1339
1340 /*
1341 * version 1 superblock
1342 */
1343
1344 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1345 {
1346 __le32 disk_csum;
1347 u32 csum;
1348 unsigned long long newcsum;
1349 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1350 __le32 *isuper = (__le32*)sb;
1351
1352 disk_csum = sb->sb_csum;
1353 sb->sb_csum = 0;
1354 newcsum = 0;
1355 for (; size >= 4; size -= 4)
1356 newcsum += le32_to_cpu(*isuper++);
1357
1358 if (size == 2)
1359 newcsum += le16_to_cpu(*(__le16*) isuper);
1360
1361 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1362 sb->sb_csum = disk_csum;
1363 return cpu_to_le32(csum);
1364 }
1365
1366 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1367 {
1368 struct mdp_superblock_1 *sb;
1369 int ret;
1370 sector_t sb_start;
1371 sector_t sectors;
1372 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1373 int bmask;
1374
1375 /*
1376 * Calculate the position of the superblock in 512byte sectors.
1377 * It is always aligned to a 4K boundary and
1378 * depeding on minor_version, it can be:
1379 * 0: At least 8K, but less than 12K, from end of device
1380 * 1: At start of device
1381 * 2: 4K from start of device.
1382 */
1383 switch(minor_version) {
1384 case 0:
1385 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1386 sb_start -= 8*2;
1387 sb_start &= ~(sector_t)(4*2-1);
1388 break;
1389 case 1:
1390 sb_start = 0;
1391 break;
1392 case 2:
1393 sb_start = 8;
1394 break;
1395 default:
1396 return -EINVAL;
1397 }
1398 rdev->sb_start = sb_start;
1399
1400 /* superblock is rarely larger than 1K, but it can be larger,
1401 * and it is safe to read 4k, so we do that
1402 */
1403 ret = read_disk_sb(rdev, 4096);
1404 if (ret) return ret;
1405
1406 sb = page_address(rdev->sb_page);
1407
1408 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1409 sb->major_version != cpu_to_le32(1) ||
1410 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1411 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1412 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1413 return -EINVAL;
1414
1415 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1416 printk("md: invalid superblock checksum on %s\n",
1417 bdevname(rdev->bdev,b));
1418 return -EINVAL;
1419 }
1420 if (le64_to_cpu(sb->data_size) < 10) {
1421 printk("md: data_size too small on %s\n",
1422 bdevname(rdev->bdev,b));
1423 return -EINVAL;
1424 }
1425 if (sb->pad0 ||
1426 sb->pad3[0] ||
1427 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1428 /* Some padding is non-zero, might be a new feature */
1429 return -EINVAL;
1430
1431 rdev->preferred_minor = 0xffff;
1432 rdev->data_offset = le64_to_cpu(sb->data_offset);
1433 rdev->new_data_offset = rdev->data_offset;
1434 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1435 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1436 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1437 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1438
1439 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1440 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1441 if (rdev->sb_size & bmask)
1442 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1443
1444 if (minor_version
1445 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1446 return -EINVAL;
1447 if (minor_version
1448 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1449 return -EINVAL;
1450
1451 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1452 rdev->desc_nr = -1;
1453 else
1454 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1455
1456 if (!rdev->bb_page) {
1457 rdev->bb_page = alloc_page(GFP_KERNEL);
1458 if (!rdev->bb_page)
1459 return -ENOMEM;
1460 }
1461 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1462 rdev->badblocks.count == 0) {
1463 /* need to load the bad block list.
1464 * Currently we limit it to one page.
1465 */
1466 s32 offset;
1467 sector_t bb_sector;
1468 u64 *bbp;
1469 int i;
1470 int sectors = le16_to_cpu(sb->bblog_size);
1471 if (sectors > (PAGE_SIZE / 512))
1472 return -EINVAL;
1473 offset = le32_to_cpu(sb->bblog_offset);
1474 if (offset == 0)
1475 return -EINVAL;
1476 bb_sector = (long long)offset;
1477 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1478 rdev->bb_page, REQ_OP_READ, 0, true))
1479 return -EIO;
1480 bbp = (u64 *)page_address(rdev->bb_page);
1481 rdev->badblocks.shift = sb->bblog_shift;
1482 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1483 u64 bb = le64_to_cpu(*bbp);
1484 int count = bb & (0x3ff);
1485 u64 sector = bb >> 10;
1486 sector <<= sb->bblog_shift;
1487 count <<= sb->bblog_shift;
1488 if (bb + 1 == 0)
1489 break;
1490 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1491 return -EINVAL;
1492 }
1493 } else if (sb->bblog_offset != 0)
1494 rdev->badblocks.shift = 0;
1495
1496 if (!refdev) {
1497 ret = 1;
1498 } else {
1499 __u64 ev1, ev2;
1500 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1501
1502 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1503 sb->level != refsb->level ||
1504 sb->layout != refsb->layout ||
1505 sb->chunksize != refsb->chunksize) {
1506 printk(KERN_WARNING "md: %s has strangely different"
1507 " superblock to %s\n",
1508 bdevname(rdev->bdev,b),
1509 bdevname(refdev->bdev,b2));
1510 return -EINVAL;
1511 }
1512 ev1 = le64_to_cpu(sb->events);
1513 ev2 = le64_to_cpu(refsb->events);
1514
1515 if (ev1 > ev2)
1516 ret = 1;
1517 else
1518 ret = 0;
1519 }
1520 if (minor_version) {
1521 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1522 sectors -= rdev->data_offset;
1523 } else
1524 sectors = rdev->sb_start;
1525 if (sectors < le64_to_cpu(sb->data_size))
1526 return -EINVAL;
1527 rdev->sectors = le64_to_cpu(sb->data_size);
1528 return ret;
1529 }
1530
1531 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1532 {
1533 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1534 __u64 ev1 = le64_to_cpu(sb->events);
1535
1536 rdev->raid_disk = -1;
1537 clear_bit(Faulty, &rdev->flags);
1538 clear_bit(In_sync, &rdev->flags);
1539 clear_bit(Bitmap_sync, &rdev->flags);
1540 clear_bit(WriteMostly, &rdev->flags);
1541
1542 if (mddev->raid_disks == 0) {
1543 mddev->major_version = 1;
1544 mddev->patch_version = 0;
1545 mddev->external = 0;
1546 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1547 mddev->ctime = le64_to_cpu(sb->ctime);
1548 mddev->utime = le64_to_cpu(sb->utime);
1549 mddev->level = le32_to_cpu(sb->level);
1550 mddev->clevel[0] = 0;
1551 mddev->layout = le32_to_cpu(sb->layout);
1552 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1553 mddev->dev_sectors = le64_to_cpu(sb->size);
1554 mddev->events = ev1;
1555 mddev->bitmap_info.offset = 0;
1556 mddev->bitmap_info.space = 0;
1557 /* Default location for bitmap is 1K after superblock
1558 * using 3K - total of 4K
1559 */
1560 mddev->bitmap_info.default_offset = 1024 >> 9;
1561 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1562 mddev->reshape_backwards = 0;
1563
1564 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1565 memcpy(mddev->uuid, sb->set_uuid, 16);
1566
1567 mddev->max_disks = (4096-256)/2;
1568
1569 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1570 mddev->bitmap_info.file == NULL) {
1571 mddev->bitmap_info.offset =
1572 (__s32)le32_to_cpu(sb->bitmap_offset);
1573 /* Metadata doesn't record how much space is available.
1574 * For 1.0, we assume we can use up to the superblock
1575 * if before, else to 4K beyond superblock.
1576 * For others, assume no change is possible.
1577 */
1578 if (mddev->minor_version > 0)
1579 mddev->bitmap_info.space = 0;
1580 else if (mddev->bitmap_info.offset > 0)
1581 mddev->bitmap_info.space =
1582 8 - mddev->bitmap_info.offset;
1583 else
1584 mddev->bitmap_info.space =
1585 -mddev->bitmap_info.offset;
1586 }
1587
1588 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1589 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1590 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1591 mddev->new_level = le32_to_cpu(sb->new_level);
1592 mddev->new_layout = le32_to_cpu(sb->new_layout);
1593 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1594 if (mddev->delta_disks < 0 ||
1595 (mddev->delta_disks == 0 &&
1596 (le32_to_cpu(sb->feature_map)
1597 & MD_FEATURE_RESHAPE_BACKWARDS)))
1598 mddev->reshape_backwards = 1;
1599 } else {
1600 mddev->reshape_position = MaxSector;
1601 mddev->delta_disks = 0;
1602 mddev->new_level = mddev->level;
1603 mddev->new_layout = mddev->layout;
1604 mddev->new_chunk_sectors = mddev->chunk_sectors;
1605 }
1606
1607 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1608 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1609 } else if (mddev->pers == NULL) {
1610 /* Insist of good event counter while assembling, except for
1611 * spares (which don't need an event count) */
1612 ++ev1;
1613 if (rdev->desc_nr >= 0 &&
1614 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1615 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1616 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1617 if (ev1 < mddev->events)
1618 return -EINVAL;
1619 } else if (mddev->bitmap) {
1620 /* If adding to array with a bitmap, then we can accept an
1621 * older device, but not too old.
1622 */
1623 if (ev1 < mddev->bitmap->events_cleared)
1624 return 0;
1625 if (ev1 < mddev->events)
1626 set_bit(Bitmap_sync, &rdev->flags);
1627 } else {
1628 if (ev1 < mddev->events)
1629 /* just a hot-add of a new device, leave raid_disk at -1 */
1630 return 0;
1631 }
1632 if (mddev->level != LEVEL_MULTIPATH) {
1633 int role;
1634 if (rdev->desc_nr < 0 ||
1635 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1636 role = MD_DISK_ROLE_SPARE;
1637 rdev->desc_nr = -1;
1638 } else
1639 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1640 switch(role) {
1641 case MD_DISK_ROLE_SPARE: /* spare */
1642 break;
1643 case MD_DISK_ROLE_FAULTY: /* faulty */
1644 set_bit(Faulty, &rdev->flags);
1645 break;
1646 case MD_DISK_ROLE_JOURNAL: /* journal device */
1647 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1648 /* journal device without journal feature */
1649 printk(KERN_WARNING
1650 "md: journal device provided without journal feature, ignoring the device\n");
1651 return -EINVAL;
1652 }
1653 set_bit(Journal, &rdev->flags);
1654 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1655 rdev->raid_disk = 0;
1656 break;
1657 default:
1658 rdev->saved_raid_disk = role;
1659 if ((le32_to_cpu(sb->feature_map) &
1660 MD_FEATURE_RECOVERY_OFFSET)) {
1661 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1662 if (!(le32_to_cpu(sb->feature_map) &
1663 MD_FEATURE_RECOVERY_BITMAP))
1664 rdev->saved_raid_disk = -1;
1665 } else
1666 set_bit(In_sync, &rdev->flags);
1667 rdev->raid_disk = role;
1668 break;
1669 }
1670 if (sb->devflags & WriteMostly1)
1671 set_bit(WriteMostly, &rdev->flags);
1672 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1673 set_bit(Replacement, &rdev->flags);
1674 } else /* MULTIPATH are always insync */
1675 set_bit(In_sync, &rdev->flags);
1676
1677 return 0;
1678 }
1679
1680 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1681 {
1682 struct mdp_superblock_1 *sb;
1683 struct md_rdev *rdev2;
1684 int max_dev, i;
1685 /* make rdev->sb match mddev and rdev data. */
1686
1687 sb = page_address(rdev->sb_page);
1688
1689 sb->feature_map = 0;
1690 sb->pad0 = 0;
1691 sb->recovery_offset = cpu_to_le64(0);
1692 memset(sb->pad3, 0, sizeof(sb->pad3));
1693
1694 sb->utime = cpu_to_le64((__u64)mddev->utime);
1695 sb->events = cpu_to_le64(mddev->events);
1696 if (mddev->in_sync)
1697 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1698 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1699 sb->resync_offset = cpu_to_le64(MaxSector);
1700 else
1701 sb->resync_offset = cpu_to_le64(0);
1702
1703 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1704
1705 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1706 sb->size = cpu_to_le64(mddev->dev_sectors);
1707 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1708 sb->level = cpu_to_le32(mddev->level);
1709 sb->layout = cpu_to_le32(mddev->layout);
1710
1711 if (test_bit(WriteMostly, &rdev->flags))
1712 sb->devflags |= WriteMostly1;
1713 else
1714 sb->devflags &= ~WriteMostly1;
1715 sb->data_offset = cpu_to_le64(rdev->data_offset);
1716 sb->data_size = cpu_to_le64(rdev->sectors);
1717
1718 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1719 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1720 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1721 }
1722
1723 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1724 !test_bit(In_sync, &rdev->flags)) {
1725 sb->feature_map |=
1726 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1727 sb->recovery_offset =
1728 cpu_to_le64(rdev->recovery_offset);
1729 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1730 sb->feature_map |=
1731 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1732 }
1733 /* Note: recovery_offset and journal_tail share space */
1734 if (test_bit(Journal, &rdev->flags))
1735 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1736 if (test_bit(Replacement, &rdev->flags))
1737 sb->feature_map |=
1738 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1739
1740 if (mddev->reshape_position != MaxSector) {
1741 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1742 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1743 sb->new_layout = cpu_to_le32(mddev->new_layout);
1744 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1745 sb->new_level = cpu_to_le32(mddev->new_level);
1746 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1747 if (mddev->delta_disks == 0 &&
1748 mddev->reshape_backwards)
1749 sb->feature_map
1750 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1751 if (rdev->new_data_offset != rdev->data_offset) {
1752 sb->feature_map
1753 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1754 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1755 - rdev->data_offset));
1756 }
1757 }
1758
1759 if (mddev_is_clustered(mddev))
1760 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1761
1762 if (rdev->badblocks.count == 0)
1763 /* Nothing to do for bad blocks*/ ;
1764 else if (sb->bblog_offset == 0)
1765 /* Cannot record bad blocks on this device */
1766 md_error(mddev, rdev);
1767 else {
1768 struct badblocks *bb = &rdev->badblocks;
1769 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1770 u64 *p = bb->page;
1771 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1772 if (bb->changed) {
1773 unsigned seq;
1774
1775 retry:
1776 seq = read_seqbegin(&bb->lock);
1777
1778 memset(bbp, 0xff, PAGE_SIZE);
1779
1780 for (i = 0 ; i < bb->count ; i++) {
1781 u64 internal_bb = p[i];
1782 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1783 | BB_LEN(internal_bb));
1784 bbp[i] = cpu_to_le64(store_bb);
1785 }
1786 bb->changed = 0;
1787 if (read_seqretry(&bb->lock, seq))
1788 goto retry;
1789
1790 bb->sector = (rdev->sb_start +
1791 (int)le32_to_cpu(sb->bblog_offset));
1792 bb->size = le16_to_cpu(sb->bblog_size);
1793 }
1794 }
1795
1796 max_dev = 0;
1797 rdev_for_each(rdev2, mddev)
1798 if (rdev2->desc_nr+1 > max_dev)
1799 max_dev = rdev2->desc_nr+1;
1800
1801 if (max_dev > le32_to_cpu(sb->max_dev)) {
1802 int bmask;
1803 sb->max_dev = cpu_to_le32(max_dev);
1804 rdev->sb_size = max_dev * 2 + 256;
1805 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1806 if (rdev->sb_size & bmask)
1807 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1808 } else
1809 max_dev = le32_to_cpu(sb->max_dev);
1810
1811 for (i=0; i<max_dev;i++)
1812 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1813
1814 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1815 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1816
1817 rdev_for_each(rdev2, mddev) {
1818 i = rdev2->desc_nr;
1819 if (test_bit(Faulty, &rdev2->flags))
1820 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1821 else if (test_bit(In_sync, &rdev2->flags))
1822 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1823 else if (test_bit(Journal, &rdev2->flags))
1824 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1825 else if (rdev2->raid_disk >= 0)
1826 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1827 else
1828 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1829 }
1830
1831 sb->sb_csum = calc_sb_1_csum(sb);
1832 }
1833
1834 static unsigned long long
1835 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1836 {
1837 struct mdp_superblock_1 *sb;
1838 sector_t max_sectors;
1839 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1840 return 0; /* component must fit device */
1841 if (rdev->data_offset != rdev->new_data_offset)
1842 return 0; /* too confusing */
1843 if (rdev->sb_start < rdev->data_offset) {
1844 /* minor versions 1 and 2; superblock before data */
1845 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1846 max_sectors -= rdev->data_offset;
1847 if (!num_sectors || num_sectors > max_sectors)
1848 num_sectors = max_sectors;
1849 } else if (rdev->mddev->bitmap_info.offset) {
1850 /* minor version 0 with bitmap we can't move */
1851 return 0;
1852 } else {
1853 /* minor version 0; superblock after data */
1854 sector_t sb_start;
1855 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1856 sb_start &= ~(sector_t)(4*2 - 1);
1857 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1858 if (!num_sectors || num_sectors > max_sectors)
1859 num_sectors = max_sectors;
1860 rdev->sb_start = sb_start;
1861 }
1862 sb = page_address(rdev->sb_page);
1863 sb->data_size = cpu_to_le64(num_sectors);
1864 sb->super_offset = rdev->sb_start;
1865 sb->sb_csum = calc_sb_1_csum(sb);
1866 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1867 rdev->sb_page);
1868 md_super_wait(rdev->mddev);
1869 return num_sectors;
1870
1871 }
1872
1873 static int
1874 super_1_allow_new_offset(struct md_rdev *rdev,
1875 unsigned long long new_offset)
1876 {
1877 /* All necessary checks on new >= old have been done */
1878 struct bitmap *bitmap;
1879 if (new_offset >= rdev->data_offset)
1880 return 1;
1881
1882 /* with 1.0 metadata, there is no metadata to tread on
1883 * so we can always move back */
1884 if (rdev->mddev->minor_version == 0)
1885 return 1;
1886
1887 /* otherwise we must be sure not to step on
1888 * any metadata, so stay:
1889 * 36K beyond start of superblock
1890 * beyond end of badblocks
1891 * beyond write-intent bitmap
1892 */
1893 if (rdev->sb_start + (32+4)*2 > new_offset)
1894 return 0;
1895 bitmap = rdev->mddev->bitmap;
1896 if (bitmap && !rdev->mddev->bitmap_info.file &&
1897 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1898 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1899 return 0;
1900 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1901 return 0;
1902
1903 return 1;
1904 }
1905
1906 static struct super_type super_types[] = {
1907 [0] = {
1908 .name = "0.90.0",
1909 .owner = THIS_MODULE,
1910 .load_super = super_90_load,
1911 .validate_super = super_90_validate,
1912 .sync_super = super_90_sync,
1913 .rdev_size_change = super_90_rdev_size_change,
1914 .allow_new_offset = super_90_allow_new_offset,
1915 },
1916 [1] = {
1917 .name = "md-1",
1918 .owner = THIS_MODULE,
1919 .load_super = super_1_load,
1920 .validate_super = super_1_validate,
1921 .sync_super = super_1_sync,
1922 .rdev_size_change = super_1_rdev_size_change,
1923 .allow_new_offset = super_1_allow_new_offset,
1924 },
1925 };
1926
1927 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1928 {
1929 if (mddev->sync_super) {
1930 mddev->sync_super(mddev, rdev);
1931 return;
1932 }
1933
1934 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1935
1936 super_types[mddev->major_version].sync_super(mddev, rdev);
1937 }
1938
1939 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1940 {
1941 struct md_rdev *rdev, *rdev2;
1942
1943 rcu_read_lock();
1944 rdev_for_each_rcu(rdev, mddev1) {
1945 if (test_bit(Faulty, &rdev->flags) ||
1946 test_bit(Journal, &rdev->flags) ||
1947 rdev->raid_disk == -1)
1948 continue;
1949 rdev_for_each_rcu(rdev2, mddev2) {
1950 if (test_bit(Faulty, &rdev2->flags) ||
1951 test_bit(Journal, &rdev2->flags) ||
1952 rdev2->raid_disk == -1)
1953 continue;
1954 if (rdev->bdev->bd_contains ==
1955 rdev2->bdev->bd_contains) {
1956 rcu_read_unlock();
1957 return 1;
1958 }
1959 }
1960 }
1961 rcu_read_unlock();
1962 return 0;
1963 }
1964
1965 static LIST_HEAD(pending_raid_disks);
1966
1967 /*
1968 * Try to register data integrity profile for an mddev
1969 *
1970 * This is called when an array is started and after a disk has been kicked
1971 * from the array. It only succeeds if all working and active component devices
1972 * are integrity capable with matching profiles.
1973 */
1974 int md_integrity_register(struct mddev *mddev)
1975 {
1976 struct md_rdev *rdev, *reference = NULL;
1977
1978 if (list_empty(&mddev->disks))
1979 return 0; /* nothing to do */
1980 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1981 return 0; /* shouldn't register, or already is */
1982 rdev_for_each(rdev, mddev) {
1983 /* skip spares and non-functional disks */
1984 if (test_bit(Faulty, &rdev->flags))
1985 continue;
1986 if (rdev->raid_disk < 0)
1987 continue;
1988 if (!reference) {
1989 /* Use the first rdev as the reference */
1990 reference = rdev;
1991 continue;
1992 }
1993 /* does this rdev's profile match the reference profile? */
1994 if (blk_integrity_compare(reference->bdev->bd_disk,
1995 rdev->bdev->bd_disk) < 0)
1996 return -EINVAL;
1997 }
1998 if (!reference || !bdev_get_integrity(reference->bdev))
1999 return 0;
2000 /*
2001 * All component devices are integrity capable and have matching
2002 * profiles, register the common profile for the md device.
2003 */
2004 blk_integrity_register(mddev->gendisk,
2005 bdev_get_integrity(reference->bdev));
2006
2007 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2008 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2009 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2010 mdname(mddev));
2011 return -EINVAL;
2012 }
2013 return 0;
2014 }
2015 EXPORT_SYMBOL(md_integrity_register);
2016
2017 /*
2018 * Attempt to add an rdev, but only if it is consistent with the current
2019 * integrity profile
2020 */
2021 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2022 {
2023 struct blk_integrity *bi_rdev;
2024 struct blk_integrity *bi_mddev;
2025 char name[BDEVNAME_SIZE];
2026
2027 if (!mddev->gendisk)
2028 return 0;
2029
2030 bi_rdev = bdev_get_integrity(rdev->bdev);
2031 bi_mddev = blk_get_integrity(mddev->gendisk);
2032
2033 if (!bi_mddev) /* nothing to do */
2034 return 0;
2035
2036 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2037 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2038 mdname(mddev), bdevname(rdev->bdev, name));
2039 return -ENXIO;
2040 }
2041
2042 return 0;
2043 }
2044 EXPORT_SYMBOL(md_integrity_add_rdev);
2045
2046 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2047 {
2048 char b[BDEVNAME_SIZE];
2049 struct kobject *ko;
2050 int err;
2051
2052 /* prevent duplicates */
2053 if (find_rdev(mddev, rdev->bdev->bd_dev))
2054 return -EEXIST;
2055
2056 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2057 if (!test_bit(Journal, &rdev->flags) &&
2058 rdev->sectors &&
2059 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2060 if (mddev->pers) {
2061 /* Cannot change size, so fail
2062 * If mddev->level <= 0, then we don't care
2063 * about aligning sizes (e.g. linear)
2064 */
2065 if (mddev->level > 0)
2066 return -ENOSPC;
2067 } else
2068 mddev->dev_sectors = rdev->sectors;
2069 }
2070
2071 /* Verify rdev->desc_nr is unique.
2072 * If it is -1, assign a free number, else
2073 * check number is not in use
2074 */
2075 rcu_read_lock();
2076 if (rdev->desc_nr < 0) {
2077 int choice = 0;
2078 if (mddev->pers)
2079 choice = mddev->raid_disks;
2080 while (md_find_rdev_nr_rcu(mddev, choice))
2081 choice++;
2082 rdev->desc_nr = choice;
2083 } else {
2084 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2085 rcu_read_unlock();
2086 return -EBUSY;
2087 }
2088 }
2089 rcu_read_unlock();
2090 if (!test_bit(Journal, &rdev->flags) &&
2091 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2092 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2093 mdname(mddev), mddev->max_disks);
2094 return -EBUSY;
2095 }
2096 bdevname(rdev->bdev,b);
2097 strreplace(b, '/', '!');
2098
2099 rdev->mddev = mddev;
2100 printk(KERN_INFO "md: bind<%s>\n", b);
2101
2102 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2103 goto fail;
2104
2105 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2106 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2107 /* failure here is OK */;
2108 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2109
2110 list_add_rcu(&rdev->same_set, &mddev->disks);
2111 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2112
2113 /* May as well allow recovery to be retried once */
2114 mddev->recovery_disabled++;
2115
2116 return 0;
2117
2118 fail:
2119 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2120 b, mdname(mddev));
2121 return err;
2122 }
2123
2124 static void md_delayed_delete(struct work_struct *ws)
2125 {
2126 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2127 kobject_del(&rdev->kobj);
2128 kobject_put(&rdev->kobj);
2129 }
2130
2131 static void unbind_rdev_from_array(struct md_rdev *rdev)
2132 {
2133 char b[BDEVNAME_SIZE];
2134
2135 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2136 list_del_rcu(&rdev->same_set);
2137 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2138 rdev->mddev = NULL;
2139 sysfs_remove_link(&rdev->kobj, "block");
2140 sysfs_put(rdev->sysfs_state);
2141 rdev->sysfs_state = NULL;
2142 rdev->badblocks.count = 0;
2143 /* We need to delay this, otherwise we can deadlock when
2144 * writing to 'remove' to "dev/state". We also need
2145 * to delay it due to rcu usage.
2146 */
2147 synchronize_rcu();
2148 INIT_WORK(&rdev->del_work, md_delayed_delete);
2149 kobject_get(&rdev->kobj);
2150 queue_work(md_misc_wq, &rdev->del_work);
2151 }
2152
2153 /*
2154 * prevent the device from being mounted, repartitioned or
2155 * otherwise reused by a RAID array (or any other kernel
2156 * subsystem), by bd_claiming the device.
2157 */
2158 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2159 {
2160 int err = 0;
2161 struct block_device *bdev;
2162 char b[BDEVNAME_SIZE];
2163
2164 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2165 shared ? (struct md_rdev *)lock_rdev : rdev);
2166 if (IS_ERR(bdev)) {
2167 printk(KERN_ERR "md: could not open %s.\n",
2168 __bdevname(dev, b));
2169 return PTR_ERR(bdev);
2170 }
2171 rdev->bdev = bdev;
2172 return err;
2173 }
2174
2175 static void unlock_rdev(struct md_rdev *rdev)
2176 {
2177 struct block_device *bdev = rdev->bdev;
2178 rdev->bdev = NULL;
2179 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2180 }
2181
2182 void md_autodetect_dev(dev_t dev);
2183
2184 static void export_rdev(struct md_rdev *rdev)
2185 {
2186 char b[BDEVNAME_SIZE];
2187
2188 printk(KERN_INFO "md: export_rdev(%s)\n",
2189 bdevname(rdev->bdev,b));
2190 md_rdev_clear(rdev);
2191 #ifndef MODULE
2192 if (test_bit(AutoDetected, &rdev->flags))
2193 md_autodetect_dev(rdev->bdev->bd_dev);
2194 #endif
2195 unlock_rdev(rdev);
2196 kobject_put(&rdev->kobj);
2197 }
2198
2199 void md_kick_rdev_from_array(struct md_rdev *rdev)
2200 {
2201 unbind_rdev_from_array(rdev);
2202 export_rdev(rdev);
2203 }
2204 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2205
2206 static void export_array(struct mddev *mddev)
2207 {
2208 struct md_rdev *rdev;
2209
2210 while (!list_empty(&mddev->disks)) {
2211 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2212 same_set);
2213 md_kick_rdev_from_array(rdev);
2214 }
2215 mddev->raid_disks = 0;
2216 mddev->major_version = 0;
2217 }
2218
2219 static void sync_sbs(struct mddev *mddev, int nospares)
2220 {
2221 /* Update each superblock (in-memory image), but
2222 * if we are allowed to, skip spares which already
2223 * have the right event counter, or have one earlier
2224 * (which would mean they aren't being marked as dirty
2225 * with the rest of the array)
2226 */
2227 struct md_rdev *rdev;
2228 rdev_for_each(rdev, mddev) {
2229 if (rdev->sb_events == mddev->events ||
2230 (nospares &&
2231 rdev->raid_disk < 0 &&
2232 rdev->sb_events+1 == mddev->events)) {
2233 /* Don't update this superblock */
2234 rdev->sb_loaded = 2;
2235 } else {
2236 sync_super(mddev, rdev);
2237 rdev->sb_loaded = 1;
2238 }
2239 }
2240 }
2241
2242 static bool does_sb_need_changing(struct mddev *mddev)
2243 {
2244 struct md_rdev *rdev;
2245 struct mdp_superblock_1 *sb;
2246 int role;
2247
2248 /* Find a good rdev */
2249 rdev_for_each(rdev, mddev)
2250 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2251 break;
2252
2253 /* No good device found. */
2254 if (!rdev)
2255 return false;
2256
2257 sb = page_address(rdev->sb_page);
2258 /* Check if a device has become faulty or a spare become active */
2259 rdev_for_each(rdev, mddev) {
2260 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2261 /* Device activated? */
2262 if (role == 0xffff && rdev->raid_disk >=0 &&
2263 !test_bit(Faulty, &rdev->flags))
2264 return true;
2265 /* Device turned faulty? */
2266 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2267 return true;
2268 }
2269
2270 /* Check if any mddev parameters have changed */
2271 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2272 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2273 (mddev->layout != le64_to_cpu(sb->layout)) ||
2274 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2275 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2276 return true;
2277
2278 return false;
2279 }
2280
2281 void md_update_sb(struct mddev *mddev, int force_change)
2282 {
2283 struct md_rdev *rdev;
2284 int sync_req;
2285 int nospares = 0;
2286 int any_badblocks_changed = 0;
2287 int ret = -1;
2288
2289 if (mddev->ro) {
2290 if (force_change)
2291 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2292 return;
2293 }
2294
2295 repeat:
2296 if (mddev_is_clustered(mddev)) {
2297 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2298 force_change = 1;
2299 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2300 nospares = 1;
2301 ret = md_cluster_ops->metadata_update_start(mddev);
2302 /* Has someone else has updated the sb */
2303 if (!does_sb_need_changing(mddev)) {
2304 if (ret == 0)
2305 md_cluster_ops->metadata_update_cancel(mddev);
2306 bit_clear_unless(&mddev->flags, BIT(MD_CHANGE_PENDING),
2307 BIT(MD_CHANGE_DEVS) |
2308 BIT(MD_CHANGE_CLEAN));
2309 return;
2310 }
2311 }
2312
2313 /* First make sure individual recovery_offsets are correct */
2314 rdev_for_each(rdev, mddev) {
2315 if (rdev->raid_disk >= 0 &&
2316 mddev->delta_disks >= 0 &&
2317 !test_bit(Journal, &rdev->flags) &&
2318 !test_bit(In_sync, &rdev->flags) &&
2319 mddev->curr_resync_completed > rdev->recovery_offset)
2320 rdev->recovery_offset = mddev->curr_resync_completed;
2321
2322 }
2323 if (!mddev->persistent) {
2324 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2325 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2326 if (!mddev->external) {
2327 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2328 rdev_for_each(rdev, mddev) {
2329 if (rdev->badblocks.changed) {
2330 rdev->badblocks.changed = 0;
2331 ack_all_badblocks(&rdev->badblocks);
2332 md_error(mddev, rdev);
2333 }
2334 clear_bit(Blocked, &rdev->flags);
2335 clear_bit(BlockedBadBlocks, &rdev->flags);
2336 wake_up(&rdev->blocked_wait);
2337 }
2338 }
2339 wake_up(&mddev->sb_wait);
2340 return;
2341 }
2342
2343 spin_lock(&mddev->lock);
2344
2345 mddev->utime = ktime_get_real_seconds();
2346
2347 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2348 force_change = 1;
2349 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2350 /* just a clean<-> dirty transition, possibly leave spares alone,
2351 * though if events isn't the right even/odd, we will have to do
2352 * spares after all
2353 */
2354 nospares = 1;
2355 if (force_change)
2356 nospares = 0;
2357 if (mddev->degraded)
2358 /* If the array is degraded, then skipping spares is both
2359 * dangerous and fairly pointless.
2360 * Dangerous because a device that was removed from the array
2361 * might have a event_count that still looks up-to-date,
2362 * so it can be re-added without a resync.
2363 * Pointless because if there are any spares to skip,
2364 * then a recovery will happen and soon that array won't
2365 * be degraded any more and the spare can go back to sleep then.
2366 */
2367 nospares = 0;
2368
2369 sync_req = mddev->in_sync;
2370
2371 /* If this is just a dirty<->clean transition, and the array is clean
2372 * and 'events' is odd, we can roll back to the previous clean state */
2373 if (nospares
2374 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2375 && mddev->can_decrease_events
2376 && mddev->events != 1) {
2377 mddev->events--;
2378 mddev->can_decrease_events = 0;
2379 } else {
2380 /* otherwise we have to go forward and ... */
2381 mddev->events ++;
2382 mddev->can_decrease_events = nospares;
2383 }
2384
2385 /*
2386 * This 64-bit counter should never wrap.
2387 * Either we are in around ~1 trillion A.C., assuming
2388 * 1 reboot per second, or we have a bug...
2389 */
2390 WARN_ON(mddev->events == 0);
2391
2392 rdev_for_each(rdev, mddev) {
2393 if (rdev->badblocks.changed)
2394 any_badblocks_changed++;
2395 if (test_bit(Faulty, &rdev->flags))
2396 set_bit(FaultRecorded, &rdev->flags);
2397 }
2398
2399 sync_sbs(mddev, nospares);
2400 spin_unlock(&mddev->lock);
2401
2402 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2403 mdname(mddev), mddev->in_sync);
2404
2405 bitmap_update_sb(mddev->bitmap);
2406 rdev_for_each(rdev, mddev) {
2407 char b[BDEVNAME_SIZE];
2408
2409 if (rdev->sb_loaded != 1)
2410 continue; /* no noise on spare devices */
2411
2412 if (!test_bit(Faulty, &rdev->flags)) {
2413 md_super_write(mddev,rdev,
2414 rdev->sb_start, rdev->sb_size,
2415 rdev->sb_page);
2416 pr_debug("md: (write) %s's sb offset: %llu\n",
2417 bdevname(rdev->bdev, b),
2418 (unsigned long long)rdev->sb_start);
2419 rdev->sb_events = mddev->events;
2420 if (rdev->badblocks.size) {
2421 md_super_write(mddev, rdev,
2422 rdev->badblocks.sector,
2423 rdev->badblocks.size << 9,
2424 rdev->bb_page);
2425 rdev->badblocks.size = 0;
2426 }
2427
2428 } else
2429 pr_debug("md: %s (skipping faulty)\n",
2430 bdevname(rdev->bdev, b));
2431
2432 if (mddev->level == LEVEL_MULTIPATH)
2433 /* only need to write one superblock... */
2434 break;
2435 }
2436 md_super_wait(mddev);
2437 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2438
2439 if (mddev_is_clustered(mddev) && ret == 0)
2440 md_cluster_ops->metadata_update_finish(mddev);
2441
2442 if (mddev->in_sync != sync_req ||
2443 !bit_clear_unless(&mddev->flags, BIT(MD_CHANGE_PENDING),
2444 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_CLEAN)))
2445 /* have to write it out again */
2446 goto repeat;
2447 wake_up(&mddev->sb_wait);
2448 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2449 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2450
2451 rdev_for_each(rdev, mddev) {
2452 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2453 clear_bit(Blocked, &rdev->flags);
2454
2455 if (any_badblocks_changed)
2456 ack_all_badblocks(&rdev->badblocks);
2457 clear_bit(BlockedBadBlocks, &rdev->flags);
2458 wake_up(&rdev->blocked_wait);
2459 }
2460 }
2461 EXPORT_SYMBOL(md_update_sb);
2462
2463 static int add_bound_rdev(struct md_rdev *rdev)
2464 {
2465 struct mddev *mddev = rdev->mddev;
2466 int err = 0;
2467 bool add_journal = test_bit(Journal, &rdev->flags);
2468
2469 if (!mddev->pers->hot_remove_disk || add_journal) {
2470 /* If there is hot_add_disk but no hot_remove_disk
2471 * then added disks for geometry changes,
2472 * and should be added immediately.
2473 */
2474 super_types[mddev->major_version].
2475 validate_super(mddev, rdev);
2476 if (add_journal)
2477 mddev_suspend(mddev);
2478 err = mddev->pers->hot_add_disk(mddev, rdev);
2479 if (add_journal)
2480 mddev_resume(mddev);
2481 if (err) {
2482 md_kick_rdev_from_array(rdev);
2483 return err;
2484 }
2485 }
2486 sysfs_notify_dirent_safe(rdev->sysfs_state);
2487
2488 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2489 if (mddev->degraded)
2490 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2491 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2492 md_new_event(mddev);
2493 md_wakeup_thread(mddev->thread);
2494 return 0;
2495 }
2496
2497 /* words written to sysfs files may, or may not, be \n terminated.
2498 * We want to accept with case. For this we use cmd_match.
2499 */
2500 static int cmd_match(const char *cmd, const char *str)
2501 {
2502 /* See if cmd, written into a sysfs file, matches
2503 * str. They must either be the same, or cmd can
2504 * have a trailing newline
2505 */
2506 while (*cmd && *str && *cmd == *str) {
2507 cmd++;
2508 str++;
2509 }
2510 if (*cmd == '\n')
2511 cmd++;
2512 if (*str || *cmd)
2513 return 0;
2514 return 1;
2515 }
2516
2517 struct rdev_sysfs_entry {
2518 struct attribute attr;
2519 ssize_t (*show)(struct md_rdev *, char *);
2520 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2521 };
2522
2523 static ssize_t
2524 state_show(struct md_rdev *rdev, char *page)
2525 {
2526 char *sep = "";
2527 size_t len = 0;
2528 unsigned long flags = ACCESS_ONCE(rdev->flags);
2529
2530 if (test_bit(Faulty, &flags) ||
2531 rdev->badblocks.unacked_exist) {
2532 len+= sprintf(page+len, "%sfaulty",sep);
2533 sep = ",";
2534 }
2535 if (test_bit(In_sync, &flags)) {
2536 len += sprintf(page+len, "%sin_sync",sep);
2537 sep = ",";
2538 }
2539 if (test_bit(Journal, &flags)) {
2540 len += sprintf(page+len, "%sjournal",sep);
2541 sep = ",";
2542 }
2543 if (test_bit(WriteMostly, &flags)) {
2544 len += sprintf(page+len, "%swrite_mostly",sep);
2545 sep = ",";
2546 }
2547 if (test_bit(Blocked, &flags) ||
2548 (rdev->badblocks.unacked_exist
2549 && !test_bit(Faulty, &flags))) {
2550 len += sprintf(page+len, "%sblocked", sep);
2551 sep = ",";
2552 }
2553 if (!test_bit(Faulty, &flags) &&
2554 !test_bit(Journal, &flags) &&
2555 !test_bit(In_sync, &flags)) {
2556 len += sprintf(page+len, "%sspare", sep);
2557 sep = ",";
2558 }
2559 if (test_bit(WriteErrorSeen, &flags)) {
2560 len += sprintf(page+len, "%swrite_error", sep);
2561 sep = ",";
2562 }
2563 if (test_bit(WantReplacement, &flags)) {
2564 len += sprintf(page+len, "%swant_replacement", sep);
2565 sep = ",";
2566 }
2567 if (test_bit(Replacement, &flags)) {
2568 len += sprintf(page+len, "%sreplacement", sep);
2569 sep = ",";
2570 }
2571
2572 return len+sprintf(page+len, "\n");
2573 }
2574
2575 static ssize_t
2576 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2577 {
2578 /* can write
2579 * faulty - simulates an error
2580 * remove - disconnects the device
2581 * writemostly - sets write_mostly
2582 * -writemostly - clears write_mostly
2583 * blocked - sets the Blocked flags
2584 * -blocked - clears the Blocked and possibly simulates an error
2585 * insync - sets Insync providing device isn't active
2586 * -insync - clear Insync for a device with a slot assigned,
2587 * so that it gets rebuilt based on bitmap
2588 * write_error - sets WriteErrorSeen
2589 * -write_error - clears WriteErrorSeen
2590 */
2591 int err = -EINVAL;
2592 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2593 md_error(rdev->mddev, rdev);
2594 if (test_bit(Faulty, &rdev->flags))
2595 err = 0;
2596 else
2597 err = -EBUSY;
2598 } else if (cmd_match(buf, "remove")) {
2599 if (rdev->mddev->pers) {
2600 clear_bit(Blocked, &rdev->flags);
2601 remove_and_add_spares(rdev->mddev, rdev);
2602 }
2603 if (rdev->raid_disk >= 0)
2604 err = -EBUSY;
2605 else {
2606 struct mddev *mddev = rdev->mddev;
2607 err = 0;
2608 if (mddev_is_clustered(mddev))
2609 err = md_cluster_ops->remove_disk(mddev, rdev);
2610
2611 if (err == 0) {
2612 md_kick_rdev_from_array(rdev);
2613 if (mddev->pers)
2614 md_update_sb(mddev, 1);
2615 md_new_event(mddev);
2616 }
2617 }
2618 } else if (cmd_match(buf, "writemostly")) {
2619 set_bit(WriteMostly, &rdev->flags);
2620 err = 0;
2621 } else if (cmd_match(buf, "-writemostly")) {
2622 clear_bit(WriteMostly, &rdev->flags);
2623 err = 0;
2624 } else if (cmd_match(buf, "blocked")) {
2625 set_bit(Blocked, &rdev->flags);
2626 err = 0;
2627 } else if (cmd_match(buf, "-blocked")) {
2628 if (!test_bit(Faulty, &rdev->flags) &&
2629 rdev->badblocks.unacked_exist) {
2630 /* metadata handler doesn't understand badblocks,
2631 * so we need to fail the device
2632 */
2633 md_error(rdev->mddev, rdev);
2634 }
2635 clear_bit(Blocked, &rdev->flags);
2636 clear_bit(BlockedBadBlocks, &rdev->flags);
2637 wake_up(&rdev->blocked_wait);
2638 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2639 md_wakeup_thread(rdev->mddev->thread);
2640
2641 err = 0;
2642 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2643 set_bit(In_sync, &rdev->flags);
2644 err = 0;
2645 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2646 !test_bit(Journal, &rdev->flags)) {
2647 if (rdev->mddev->pers == NULL) {
2648 clear_bit(In_sync, &rdev->flags);
2649 rdev->saved_raid_disk = rdev->raid_disk;
2650 rdev->raid_disk = -1;
2651 err = 0;
2652 }
2653 } else if (cmd_match(buf, "write_error")) {
2654 set_bit(WriteErrorSeen, &rdev->flags);
2655 err = 0;
2656 } else if (cmd_match(buf, "-write_error")) {
2657 clear_bit(WriteErrorSeen, &rdev->flags);
2658 err = 0;
2659 } else if (cmd_match(buf, "want_replacement")) {
2660 /* Any non-spare device that is not a replacement can
2661 * become want_replacement at any time, but we then need to
2662 * check if recovery is needed.
2663 */
2664 if (rdev->raid_disk >= 0 &&
2665 !test_bit(Journal, &rdev->flags) &&
2666 !test_bit(Replacement, &rdev->flags))
2667 set_bit(WantReplacement, &rdev->flags);
2668 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2669 md_wakeup_thread(rdev->mddev->thread);
2670 err = 0;
2671 } else if (cmd_match(buf, "-want_replacement")) {
2672 /* Clearing 'want_replacement' is always allowed.
2673 * Once replacements starts it is too late though.
2674 */
2675 err = 0;
2676 clear_bit(WantReplacement, &rdev->flags);
2677 } else if (cmd_match(buf, "replacement")) {
2678 /* Can only set a device as a replacement when array has not
2679 * yet been started. Once running, replacement is automatic
2680 * from spares, or by assigning 'slot'.
2681 */
2682 if (rdev->mddev->pers)
2683 err = -EBUSY;
2684 else {
2685 set_bit(Replacement, &rdev->flags);
2686 err = 0;
2687 }
2688 } else if (cmd_match(buf, "-replacement")) {
2689 /* Similarly, can only clear Replacement before start */
2690 if (rdev->mddev->pers)
2691 err = -EBUSY;
2692 else {
2693 clear_bit(Replacement, &rdev->flags);
2694 err = 0;
2695 }
2696 } else if (cmd_match(buf, "re-add")) {
2697 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2698 /* clear_bit is performed _after_ all the devices
2699 * have their local Faulty bit cleared. If any writes
2700 * happen in the meantime in the local node, they
2701 * will land in the local bitmap, which will be synced
2702 * by this node eventually
2703 */
2704 if (!mddev_is_clustered(rdev->mddev) ||
2705 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2706 clear_bit(Faulty, &rdev->flags);
2707 err = add_bound_rdev(rdev);
2708 }
2709 } else
2710 err = -EBUSY;
2711 }
2712 if (!err)
2713 sysfs_notify_dirent_safe(rdev->sysfs_state);
2714 return err ? err : len;
2715 }
2716 static struct rdev_sysfs_entry rdev_state =
2717 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2718
2719 static ssize_t
2720 errors_show(struct md_rdev *rdev, char *page)
2721 {
2722 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2723 }
2724
2725 static ssize_t
2726 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2727 {
2728 unsigned int n;
2729 int rv;
2730
2731 rv = kstrtouint(buf, 10, &n);
2732 if (rv < 0)
2733 return rv;
2734 atomic_set(&rdev->corrected_errors, n);
2735 return len;
2736 }
2737 static struct rdev_sysfs_entry rdev_errors =
2738 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2739
2740 static ssize_t
2741 slot_show(struct md_rdev *rdev, char *page)
2742 {
2743 if (test_bit(Journal, &rdev->flags))
2744 return sprintf(page, "journal\n");
2745 else if (rdev->raid_disk < 0)
2746 return sprintf(page, "none\n");
2747 else
2748 return sprintf(page, "%d\n", rdev->raid_disk);
2749 }
2750
2751 static ssize_t
2752 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2753 {
2754 int slot;
2755 int err;
2756
2757 if (test_bit(Journal, &rdev->flags))
2758 return -EBUSY;
2759 if (strncmp(buf, "none", 4)==0)
2760 slot = -1;
2761 else {
2762 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2763 if (err < 0)
2764 return err;
2765 }
2766 if (rdev->mddev->pers && slot == -1) {
2767 /* Setting 'slot' on an active array requires also
2768 * updating the 'rd%d' link, and communicating
2769 * with the personality with ->hot_*_disk.
2770 * For now we only support removing
2771 * failed/spare devices. This normally happens automatically,
2772 * but not when the metadata is externally managed.
2773 */
2774 if (rdev->raid_disk == -1)
2775 return -EEXIST;
2776 /* personality does all needed checks */
2777 if (rdev->mddev->pers->hot_remove_disk == NULL)
2778 return -EINVAL;
2779 clear_bit(Blocked, &rdev->flags);
2780 remove_and_add_spares(rdev->mddev, rdev);
2781 if (rdev->raid_disk >= 0)
2782 return -EBUSY;
2783 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2784 md_wakeup_thread(rdev->mddev->thread);
2785 } else if (rdev->mddev->pers) {
2786 /* Activating a spare .. or possibly reactivating
2787 * if we ever get bitmaps working here.
2788 */
2789 int err;
2790
2791 if (rdev->raid_disk != -1)
2792 return -EBUSY;
2793
2794 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2795 return -EBUSY;
2796
2797 if (rdev->mddev->pers->hot_add_disk == NULL)
2798 return -EINVAL;
2799
2800 if (slot >= rdev->mddev->raid_disks &&
2801 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2802 return -ENOSPC;
2803
2804 rdev->raid_disk = slot;
2805 if (test_bit(In_sync, &rdev->flags))
2806 rdev->saved_raid_disk = slot;
2807 else
2808 rdev->saved_raid_disk = -1;
2809 clear_bit(In_sync, &rdev->flags);
2810 clear_bit(Bitmap_sync, &rdev->flags);
2811 err = rdev->mddev->pers->
2812 hot_add_disk(rdev->mddev, rdev);
2813 if (err) {
2814 rdev->raid_disk = -1;
2815 return err;
2816 } else
2817 sysfs_notify_dirent_safe(rdev->sysfs_state);
2818 if (sysfs_link_rdev(rdev->mddev, rdev))
2819 /* failure here is OK */;
2820 /* don't wakeup anyone, leave that to userspace. */
2821 } else {
2822 if (slot >= rdev->mddev->raid_disks &&
2823 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2824 return -ENOSPC;
2825 rdev->raid_disk = slot;
2826 /* assume it is working */
2827 clear_bit(Faulty, &rdev->flags);
2828 clear_bit(WriteMostly, &rdev->flags);
2829 set_bit(In_sync, &rdev->flags);
2830 sysfs_notify_dirent_safe(rdev->sysfs_state);
2831 }
2832 return len;
2833 }
2834
2835 static struct rdev_sysfs_entry rdev_slot =
2836 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2837
2838 static ssize_t
2839 offset_show(struct md_rdev *rdev, char *page)
2840 {
2841 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2842 }
2843
2844 static ssize_t
2845 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2846 {
2847 unsigned long long offset;
2848 if (kstrtoull(buf, 10, &offset) < 0)
2849 return -EINVAL;
2850 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2851 return -EBUSY;
2852 if (rdev->sectors && rdev->mddev->external)
2853 /* Must set offset before size, so overlap checks
2854 * can be sane */
2855 return -EBUSY;
2856 rdev->data_offset = offset;
2857 rdev->new_data_offset = offset;
2858 return len;
2859 }
2860
2861 static struct rdev_sysfs_entry rdev_offset =
2862 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2863
2864 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2865 {
2866 return sprintf(page, "%llu\n",
2867 (unsigned long long)rdev->new_data_offset);
2868 }
2869
2870 static ssize_t new_offset_store(struct md_rdev *rdev,
2871 const char *buf, size_t len)
2872 {
2873 unsigned long long new_offset;
2874 struct mddev *mddev = rdev->mddev;
2875
2876 if (kstrtoull(buf, 10, &new_offset) < 0)
2877 return -EINVAL;
2878
2879 if (mddev->sync_thread ||
2880 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2881 return -EBUSY;
2882 if (new_offset == rdev->data_offset)
2883 /* reset is always permitted */
2884 ;
2885 else if (new_offset > rdev->data_offset) {
2886 /* must not push array size beyond rdev_sectors */
2887 if (new_offset - rdev->data_offset
2888 + mddev->dev_sectors > rdev->sectors)
2889 return -E2BIG;
2890 }
2891 /* Metadata worries about other space details. */
2892
2893 /* decreasing the offset is inconsistent with a backwards
2894 * reshape.
2895 */
2896 if (new_offset < rdev->data_offset &&
2897 mddev->reshape_backwards)
2898 return -EINVAL;
2899 /* Increasing offset is inconsistent with forwards
2900 * reshape. reshape_direction should be set to
2901 * 'backwards' first.
2902 */
2903 if (new_offset > rdev->data_offset &&
2904 !mddev->reshape_backwards)
2905 return -EINVAL;
2906
2907 if (mddev->pers && mddev->persistent &&
2908 !super_types[mddev->major_version]
2909 .allow_new_offset(rdev, new_offset))
2910 return -E2BIG;
2911 rdev->new_data_offset = new_offset;
2912 if (new_offset > rdev->data_offset)
2913 mddev->reshape_backwards = 1;
2914 else if (new_offset < rdev->data_offset)
2915 mddev->reshape_backwards = 0;
2916
2917 return len;
2918 }
2919 static struct rdev_sysfs_entry rdev_new_offset =
2920 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2921
2922 static ssize_t
2923 rdev_size_show(struct md_rdev *rdev, char *page)
2924 {
2925 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2926 }
2927
2928 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2929 {
2930 /* check if two start/length pairs overlap */
2931 if (s1+l1 <= s2)
2932 return 0;
2933 if (s2+l2 <= s1)
2934 return 0;
2935 return 1;
2936 }
2937
2938 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2939 {
2940 unsigned long long blocks;
2941 sector_t new;
2942
2943 if (kstrtoull(buf, 10, &blocks) < 0)
2944 return -EINVAL;
2945
2946 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2947 return -EINVAL; /* sector conversion overflow */
2948
2949 new = blocks * 2;
2950 if (new != blocks * 2)
2951 return -EINVAL; /* unsigned long long to sector_t overflow */
2952
2953 *sectors = new;
2954 return 0;
2955 }
2956
2957 static ssize_t
2958 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2959 {
2960 struct mddev *my_mddev = rdev->mddev;
2961 sector_t oldsectors = rdev->sectors;
2962 sector_t sectors;
2963
2964 if (test_bit(Journal, &rdev->flags))
2965 return -EBUSY;
2966 if (strict_blocks_to_sectors(buf, &sectors) < 0)
2967 return -EINVAL;
2968 if (rdev->data_offset != rdev->new_data_offset)
2969 return -EINVAL; /* too confusing */
2970 if (my_mddev->pers && rdev->raid_disk >= 0) {
2971 if (my_mddev->persistent) {
2972 sectors = super_types[my_mddev->major_version].
2973 rdev_size_change(rdev, sectors);
2974 if (!sectors)
2975 return -EBUSY;
2976 } else if (!sectors)
2977 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2978 rdev->data_offset;
2979 if (!my_mddev->pers->resize)
2980 /* Cannot change size for RAID0 or Linear etc */
2981 return -EINVAL;
2982 }
2983 if (sectors < my_mddev->dev_sectors)
2984 return -EINVAL; /* component must fit device */
2985
2986 rdev->sectors = sectors;
2987 if (sectors > oldsectors && my_mddev->external) {
2988 /* Need to check that all other rdevs with the same
2989 * ->bdev do not overlap. 'rcu' is sufficient to walk
2990 * the rdev lists safely.
2991 * This check does not provide a hard guarantee, it
2992 * just helps avoid dangerous mistakes.
2993 */
2994 struct mddev *mddev;
2995 int overlap = 0;
2996 struct list_head *tmp;
2997
2998 rcu_read_lock();
2999 for_each_mddev(mddev, tmp) {
3000 struct md_rdev *rdev2;
3001
3002 rdev_for_each(rdev2, mddev)
3003 if (rdev->bdev == rdev2->bdev &&
3004 rdev != rdev2 &&
3005 overlaps(rdev->data_offset, rdev->sectors,
3006 rdev2->data_offset,
3007 rdev2->sectors)) {
3008 overlap = 1;
3009 break;
3010 }
3011 if (overlap) {
3012 mddev_put(mddev);
3013 break;
3014 }
3015 }
3016 rcu_read_unlock();
3017 if (overlap) {
3018 /* Someone else could have slipped in a size
3019 * change here, but doing so is just silly.
3020 * We put oldsectors back because we *know* it is
3021 * safe, and trust userspace not to race with
3022 * itself
3023 */
3024 rdev->sectors = oldsectors;
3025 return -EBUSY;
3026 }
3027 }
3028 return len;
3029 }
3030
3031 static struct rdev_sysfs_entry rdev_size =
3032 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3033
3034 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3035 {
3036 unsigned long long recovery_start = rdev->recovery_offset;
3037
3038 if (test_bit(In_sync, &rdev->flags) ||
3039 recovery_start == MaxSector)
3040 return sprintf(page, "none\n");
3041
3042 return sprintf(page, "%llu\n", recovery_start);
3043 }
3044
3045 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3046 {
3047 unsigned long long recovery_start;
3048
3049 if (cmd_match(buf, "none"))
3050 recovery_start = MaxSector;
3051 else if (kstrtoull(buf, 10, &recovery_start))
3052 return -EINVAL;
3053
3054 if (rdev->mddev->pers &&
3055 rdev->raid_disk >= 0)
3056 return -EBUSY;
3057
3058 rdev->recovery_offset = recovery_start;
3059 if (recovery_start == MaxSector)
3060 set_bit(In_sync, &rdev->flags);
3061 else
3062 clear_bit(In_sync, &rdev->flags);
3063 return len;
3064 }
3065
3066 static struct rdev_sysfs_entry rdev_recovery_start =
3067 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3068
3069 /* sysfs access to bad-blocks list.
3070 * We present two files.
3071 * 'bad-blocks' lists sector numbers and lengths of ranges that
3072 * are recorded as bad. The list is truncated to fit within
3073 * the one-page limit of sysfs.
3074 * Writing "sector length" to this file adds an acknowledged
3075 * bad block list.
3076 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3077 * been acknowledged. Writing to this file adds bad blocks
3078 * without acknowledging them. This is largely for testing.
3079 */
3080 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3081 {
3082 return badblocks_show(&rdev->badblocks, page, 0);
3083 }
3084 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3085 {
3086 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3087 /* Maybe that ack was all we needed */
3088 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3089 wake_up(&rdev->blocked_wait);
3090 return rv;
3091 }
3092 static struct rdev_sysfs_entry rdev_bad_blocks =
3093 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3094
3095 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3096 {
3097 return badblocks_show(&rdev->badblocks, page, 1);
3098 }
3099 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3100 {
3101 return badblocks_store(&rdev->badblocks, page, len, 1);
3102 }
3103 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3104 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3105
3106 static struct attribute *rdev_default_attrs[] = {
3107 &rdev_state.attr,
3108 &rdev_errors.attr,
3109 &rdev_slot.attr,
3110 &rdev_offset.attr,
3111 &rdev_new_offset.attr,
3112 &rdev_size.attr,
3113 &rdev_recovery_start.attr,
3114 &rdev_bad_blocks.attr,
3115 &rdev_unack_bad_blocks.attr,
3116 NULL,
3117 };
3118 static ssize_t
3119 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3120 {
3121 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3122 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3123
3124 if (!entry->show)
3125 return -EIO;
3126 if (!rdev->mddev)
3127 return -EBUSY;
3128 return entry->show(rdev, page);
3129 }
3130
3131 static ssize_t
3132 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3133 const char *page, size_t length)
3134 {
3135 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3136 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3137 ssize_t rv;
3138 struct mddev *mddev = rdev->mddev;
3139
3140 if (!entry->store)
3141 return -EIO;
3142 if (!capable(CAP_SYS_ADMIN))
3143 return -EACCES;
3144 rv = mddev ? mddev_lock(mddev): -EBUSY;
3145 if (!rv) {
3146 if (rdev->mddev == NULL)
3147 rv = -EBUSY;
3148 else
3149 rv = entry->store(rdev, page, length);
3150 mddev_unlock(mddev);
3151 }
3152 return rv;
3153 }
3154
3155 static void rdev_free(struct kobject *ko)
3156 {
3157 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3158 kfree(rdev);
3159 }
3160 static const struct sysfs_ops rdev_sysfs_ops = {
3161 .show = rdev_attr_show,
3162 .store = rdev_attr_store,
3163 };
3164 static struct kobj_type rdev_ktype = {
3165 .release = rdev_free,
3166 .sysfs_ops = &rdev_sysfs_ops,
3167 .default_attrs = rdev_default_attrs,
3168 };
3169
3170 int md_rdev_init(struct md_rdev *rdev)
3171 {
3172 rdev->desc_nr = -1;
3173 rdev->saved_raid_disk = -1;
3174 rdev->raid_disk = -1;
3175 rdev->flags = 0;
3176 rdev->data_offset = 0;
3177 rdev->new_data_offset = 0;
3178 rdev->sb_events = 0;
3179 rdev->last_read_error = 0;
3180 rdev->sb_loaded = 0;
3181 rdev->bb_page = NULL;
3182 atomic_set(&rdev->nr_pending, 0);
3183 atomic_set(&rdev->read_errors, 0);
3184 atomic_set(&rdev->corrected_errors, 0);
3185
3186 INIT_LIST_HEAD(&rdev->same_set);
3187 init_waitqueue_head(&rdev->blocked_wait);
3188
3189 /* Add space to store bad block list.
3190 * This reserves the space even on arrays where it cannot
3191 * be used - I wonder if that matters
3192 */
3193 return badblocks_init(&rdev->badblocks, 0);
3194 }
3195 EXPORT_SYMBOL_GPL(md_rdev_init);
3196 /*
3197 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3198 *
3199 * mark the device faulty if:
3200 *
3201 * - the device is nonexistent (zero size)
3202 * - the device has no valid superblock
3203 *
3204 * a faulty rdev _never_ has rdev->sb set.
3205 */
3206 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3207 {
3208 char b[BDEVNAME_SIZE];
3209 int err;
3210 struct md_rdev *rdev;
3211 sector_t size;
3212
3213 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3214 if (!rdev) {
3215 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3216 return ERR_PTR(-ENOMEM);
3217 }
3218
3219 err = md_rdev_init(rdev);
3220 if (err)
3221 goto abort_free;
3222 err = alloc_disk_sb(rdev);
3223 if (err)
3224 goto abort_free;
3225
3226 err = lock_rdev(rdev, newdev, super_format == -2);
3227 if (err)
3228 goto abort_free;
3229
3230 kobject_init(&rdev->kobj, &rdev_ktype);
3231
3232 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3233 if (!size) {
3234 printk(KERN_WARNING
3235 "md: %s has zero or unknown size, marking faulty!\n",
3236 bdevname(rdev->bdev,b));
3237 err = -EINVAL;
3238 goto abort_free;
3239 }
3240
3241 if (super_format >= 0) {
3242 err = super_types[super_format].
3243 load_super(rdev, NULL, super_minor);
3244 if (err == -EINVAL) {
3245 printk(KERN_WARNING
3246 "md: %s does not have a valid v%d.%d "
3247 "superblock, not importing!\n",
3248 bdevname(rdev->bdev,b),
3249 super_format, super_minor);
3250 goto abort_free;
3251 }
3252 if (err < 0) {
3253 printk(KERN_WARNING
3254 "md: could not read %s's sb, not importing!\n",
3255 bdevname(rdev->bdev,b));
3256 goto abort_free;
3257 }
3258 }
3259
3260 return rdev;
3261
3262 abort_free:
3263 if (rdev->bdev)
3264 unlock_rdev(rdev);
3265 md_rdev_clear(rdev);
3266 kfree(rdev);
3267 return ERR_PTR(err);
3268 }
3269
3270 /*
3271 * Check a full RAID array for plausibility
3272 */
3273
3274 static void analyze_sbs(struct mddev *mddev)
3275 {
3276 int i;
3277 struct md_rdev *rdev, *freshest, *tmp;
3278 char b[BDEVNAME_SIZE];
3279
3280 freshest = NULL;
3281 rdev_for_each_safe(rdev, tmp, mddev)
3282 switch (super_types[mddev->major_version].
3283 load_super(rdev, freshest, mddev->minor_version)) {
3284 case 1:
3285 freshest = rdev;
3286 break;
3287 case 0:
3288 break;
3289 default:
3290 printk( KERN_ERR \
3291 "md: fatal superblock inconsistency in %s"
3292 " -- removing from array\n",
3293 bdevname(rdev->bdev,b));
3294 md_kick_rdev_from_array(rdev);
3295 }
3296
3297 super_types[mddev->major_version].
3298 validate_super(mddev, freshest);
3299
3300 i = 0;
3301 rdev_for_each_safe(rdev, tmp, mddev) {
3302 if (mddev->max_disks &&
3303 (rdev->desc_nr >= mddev->max_disks ||
3304 i > mddev->max_disks)) {
3305 printk(KERN_WARNING
3306 "md: %s: %s: only %d devices permitted\n",
3307 mdname(mddev), bdevname(rdev->bdev, b),
3308 mddev->max_disks);
3309 md_kick_rdev_from_array(rdev);
3310 continue;
3311 }
3312 if (rdev != freshest) {
3313 if (super_types[mddev->major_version].
3314 validate_super(mddev, rdev)) {
3315 printk(KERN_WARNING "md: kicking non-fresh %s"
3316 " from array!\n",
3317 bdevname(rdev->bdev,b));
3318 md_kick_rdev_from_array(rdev);
3319 continue;
3320 }
3321 }
3322 if (mddev->level == LEVEL_MULTIPATH) {
3323 rdev->desc_nr = i++;
3324 rdev->raid_disk = rdev->desc_nr;
3325 set_bit(In_sync, &rdev->flags);
3326 } else if (rdev->raid_disk >=
3327 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3328 !test_bit(Journal, &rdev->flags)) {
3329 rdev->raid_disk = -1;
3330 clear_bit(In_sync, &rdev->flags);
3331 }
3332 }
3333 }
3334
3335 /* Read a fixed-point number.
3336 * Numbers in sysfs attributes should be in "standard" units where
3337 * possible, so time should be in seconds.
3338 * However we internally use a a much smaller unit such as
3339 * milliseconds or jiffies.
3340 * This function takes a decimal number with a possible fractional
3341 * component, and produces an integer which is the result of
3342 * multiplying that number by 10^'scale'.
3343 * all without any floating-point arithmetic.
3344 */
3345 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3346 {
3347 unsigned long result = 0;
3348 long decimals = -1;
3349 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3350 if (*cp == '.')
3351 decimals = 0;
3352 else if (decimals < scale) {
3353 unsigned int value;
3354 value = *cp - '0';
3355 result = result * 10 + value;
3356 if (decimals >= 0)
3357 decimals++;
3358 }
3359 cp++;
3360 }
3361 if (*cp == '\n')
3362 cp++;
3363 if (*cp)
3364 return -EINVAL;
3365 if (decimals < 0)
3366 decimals = 0;
3367 while (decimals < scale) {
3368 result *= 10;
3369 decimals ++;
3370 }
3371 *res = result;
3372 return 0;
3373 }
3374
3375 static ssize_t
3376 safe_delay_show(struct mddev *mddev, char *page)
3377 {
3378 int msec = (mddev->safemode_delay*1000)/HZ;
3379 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3380 }
3381 static ssize_t
3382 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3383 {
3384 unsigned long msec;
3385
3386 if (mddev_is_clustered(mddev)) {
3387 pr_info("md: Safemode is disabled for clustered mode\n");
3388 return -EINVAL;
3389 }
3390
3391 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3392 return -EINVAL;
3393 if (msec == 0)
3394 mddev->safemode_delay = 0;
3395 else {
3396 unsigned long old_delay = mddev->safemode_delay;
3397 unsigned long new_delay = (msec*HZ)/1000;
3398
3399 if (new_delay == 0)
3400 new_delay = 1;
3401 mddev->safemode_delay = new_delay;
3402 if (new_delay < old_delay || old_delay == 0)
3403 mod_timer(&mddev->safemode_timer, jiffies+1);
3404 }
3405 return len;
3406 }
3407 static struct md_sysfs_entry md_safe_delay =
3408 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3409
3410 static ssize_t
3411 level_show(struct mddev *mddev, char *page)
3412 {
3413 struct md_personality *p;
3414 int ret;
3415 spin_lock(&mddev->lock);
3416 p = mddev->pers;
3417 if (p)
3418 ret = sprintf(page, "%s\n", p->name);
3419 else if (mddev->clevel[0])
3420 ret = sprintf(page, "%s\n", mddev->clevel);
3421 else if (mddev->level != LEVEL_NONE)
3422 ret = sprintf(page, "%d\n", mddev->level);
3423 else
3424 ret = 0;
3425 spin_unlock(&mddev->lock);
3426 return ret;
3427 }
3428
3429 static ssize_t
3430 level_store(struct mddev *mddev, const char *buf, size_t len)
3431 {
3432 char clevel[16];
3433 ssize_t rv;
3434 size_t slen = len;
3435 struct md_personality *pers, *oldpers;
3436 long level;
3437 void *priv, *oldpriv;
3438 struct md_rdev *rdev;
3439
3440 if (slen == 0 || slen >= sizeof(clevel))
3441 return -EINVAL;
3442
3443 rv = mddev_lock(mddev);
3444 if (rv)
3445 return rv;
3446
3447 if (mddev->pers == NULL) {
3448 strncpy(mddev->clevel, buf, slen);
3449 if (mddev->clevel[slen-1] == '\n')
3450 slen--;
3451 mddev->clevel[slen] = 0;
3452 mddev->level = LEVEL_NONE;
3453 rv = len;
3454 goto out_unlock;
3455 }
3456 rv = -EROFS;
3457 if (mddev->ro)
3458 goto out_unlock;
3459
3460 /* request to change the personality. Need to ensure:
3461 * - array is not engaged in resync/recovery/reshape
3462 * - old personality can be suspended
3463 * - new personality will access other array.
3464 */
3465
3466 rv = -EBUSY;
3467 if (mddev->sync_thread ||
3468 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3469 mddev->reshape_position != MaxSector ||
3470 mddev->sysfs_active)
3471 goto out_unlock;
3472
3473 rv = -EINVAL;
3474 if (!mddev->pers->quiesce) {
3475 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3476 mdname(mddev), mddev->pers->name);
3477 goto out_unlock;
3478 }
3479
3480 /* Now find the new personality */
3481 strncpy(clevel, buf, slen);
3482 if (clevel[slen-1] == '\n')
3483 slen--;
3484 clevel[slen] = 0;
3485 if (kstrtol(clevel, 10, &level))
3486 level = LEVEL_NONE;
3487
3488 if (request_module("md-%s", clevel) != 0)
3489 request_module("md-level-%s", clevel);
3490 spin_lock(&pers_lock);
3491 pers = find_pers(level, clevel);
3492 if (!pers || !try_module_get(pers->owner)) {
3493 spin_unlock(&pers_lock);
3494 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3495 rv = -EINVAL;
3496 goto out_unlock;
3497 }
3498 spin_unlock(&pers_lock);
3499
3500 if (pers == mddev->pers) {
3501 /* Nothing to do! */
3502 module_put(pers->owner);
3503 rv = len;
3504 goto out_unlock;
3505 }
3506 if (!pers->takeover) {
3507 module_put(pers->owner);
3508 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3509 mdname(mddev), clevel);
3510 rv = -EINVAL;
3511 goto out_unlock;
3512 }
3513
3514 rdev_for_each(rdev, mddev)
3515 rdev->new_raid_disk = rdev->raid_disk;
3516
3517 /* ->takeover must set new_* and/or delta_disks
3518 * if it succeeds, and may set them when it fails.
3519 */
3520 priv = pers->takeover(mddev);
3521 if (IS_ERR(priv)) {
3522 mddev->new_level = mddev->level;
3523 mddev->new_layout = mddev->layout;
3524 mddev->new_chunk_sectors = mddev->chunk_sectors;
3525 mddev->raid_disks -= mddev->delta_disks;
3526 mddev->delta_disks = 0;
3527 mddev->reshape_backwards = 0;
3528 module_put(pers->owner);
3529 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3530 mdname(mddev), clevel);
3531 rv = PTR_ERR(priv);
3532 goto out_unlock;
3533 }
3534
3535 /* Looks like we have a winner */
3536 mddev_suspend(mddev);
3537 mddev_detach(mddev);
3538
3539 spin_lock(&mddev->lock);
3540 oldpers = mddev->pers;
3541 oldpriv = mddev->private;
3542 mddev->pers = pers;
3543 mddev->private = priv;
3544 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3545 mddev->level = mddev->new_level;
3546 mddev->layout = mddev->new_layout;
3547 mddev->chunk_sectors = mddev->new_chunk_sectors;
3548 mddev->delta_disks = 0;
3549 mddev->reshape_backwards = 0;
3550 mddev->degraded = 0;
3551 spin_unlock(&mddev->lock);
3552
3553 if (oldpers->sync_request == NULL &&
3554 mddev->external) {
3555 /* We are converting from a no-redundancy array
3556 * to a redundancy array and metadata is managed
3557 * externally so we need to be sure that writes
3558 * won't block due to a need to transition
3559 * clean->dirty
3560 * until external management is started.
3561 */
3562 mddev->in_sync = 0;
3563 mddev->safemode_delay = 0;
3564 mddev->safemode = 0;
3565 }
3566
3567 oldpers->free(mddev, oldpriv);
3568
3569 if (oldpers->sync_request == NULL &&
3570 pers->sync_request != NULL) {
3571 /* need to add the md_redundancy_group */
3572 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3573 printk(KERN_WARNING
3574 "md: cannot register extra attributes for %s\n",
3575 mdname(mddev));
3576 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3577 }
3578 if (oldpers->sync_request != NULL &&
3579 pers->sync_request == NULL) {
3580 /* need to remove the md_redundancy_group */
3581 if (mddev->to_remove == NULL)
3582 mddev->to_remove = &md_redundancy_group;
3583 }
3584
3585 module_put(oldpers->owner);
3586
3587 rdev_for_each(rdev, mddev) {
3588 if (rdev->raid_disk < 0)
3589 continue;
3590 if (rdev->new_raid_disk >= mddev->raid_disks)
3591 rdev->new_raid_disk = -1;
3592 if (rdev->new_raid_disk == rdev->raid_disk)
3593 continue;
3594 sysfs_unlink_rdev(mddev, rdev);
3595 }
3596 rdev_for_each(rdev, mddev) {
3597 if (rdev->raid_disk < 0)
3598 continue;
3599 if (rdev->new_raid_disk == rdev->raid_disk)
3600 continue;
3601 rdev->raid_disk = rdev->new_raid_disk;
3602 if (rdev->raid_disk < 0)
3603 clear_bit(In_sync, &rdev->flags);
3604 else {
3605 if (sysfs_link_rdev(mddev, rdev))
3606 printk(KERN_WARNING "md: cannot register rd%d"
3607 " for %s after level change\n",
3608 rdev->raid_disk, mdname(mddev));
3609 }
3610 }
3611
3612 if (pers->sync_request == NULL) {
3613 /* this is now an array without redundancy, so
3614 * it must always be in_sync
3615 */
3616 mddev->in_sync = 1;
3617 del_timer_sync(&mddev->safemode_timer);
3618 }
3619 blk_set_stacking_limits(&mddev->queue->limits);
3620 pers->run(mddev);
3621 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3622 mddev_resume(mddev);
3623 if (!mddev->thread)
3624 md_update_sb(mddev, 1);
3625 sysfs_notify(&mddev->kobj, NULL, "level");
3626 md_new_event(mddev);
3627 rv = len;
3628 out_unlock:
3629 mddev_unlock(mddev);
3630 return rv;
3631 }
3632
3633 static struct md_sysfs_entry md_level =
3634 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3635
3636 static ssize_t
3637 layout_show(struct mddev *mddev, char *page)
3638 {
3639 /* just a number, not meaningful for all levels */
3640 if (mddev->reshape_position != MaxSector &&
3641 mddev->layout != mddev->new_layout)
3642 return sprintf(page, "%d (%d)\n",
3643 mddev->new_layout, mddev->layout);
3644 return sprintf(page, "%d\n", mddev->layout);
3645 }
3646
3647 static ssize_t
3648 layout_store(struct mddev *mddev, const char *buf, size_t len)
3649 {
3650 unsigned int n;
3651 int err;
3652
3653 err = kstrtouint(buf, 10, &n);
3654 if (err < 0)
3655 return err;
3656 err = mddev_lock(mddev);
3657 if (err)
3658 return err;
3659
3660 if (mddev->pers) {
3661 if (mddev->pers->check_reshape == NULL)
3662 err = -EBUSY;
3663 else if (mddev->ro)
3664 err = -EROFS;
3665 else {
3666 mddev->new_layout = n;
3667 err = mddev->pers->check_reshape(mddev);
3668 if (err)
3669 mddev->new_layout = mddev->layout;
3670 }
3671 } else {
3672 mddev->new_layout = n;
3673 if (mddev->reshape_position == MaxSector)
3674 mddev->layout = n;
3675 }
3676 mddev_unlock(mddev);
3677 return err ?: len;
3678 }
3679 static struct md_sysfs_entry md_layout =
3680 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3681
3682 static ssize_t
3683 raid_disks_show(struct mddev *mddev, char *page)
3684 {
3685 if (mddev->raid_disks == 0)
3686 return 0;
3687 if (mddev->reshape_position != MaxSector &&
3688 mddev->delta_disks != 0)
3689 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3690 mddev->raid_disks - mddev->delta_disks);
3691 return sprintf(page, "%d\n", mddev->raid_disks);
3692 }
3693
3694 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3695
3696 static ssize_t
3697 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3698 {
3699 unsigned int n;
3700 int err;
3701
3702 err = kstrtouint(buf, 10, &n);
3703 if (err < 0)
3704 return err;
3705
3706 err = mddev_lock(mddev);
3707 if (err)
3708 return err;
3709 if (mddev->pers)
3710 err = update_raid_disks(mddev, n);
3711 else if (mddev->reshape_position != MaxSector) {
3712 struct md_rdev *rdev;
3713 int olddisks = mddev->raid_disks - mddev->delta_disks;
3714
3715 err = -EINVAL;
3716 rdev_for_each(rdev, mddev) {
3717 if (olddisks < n &&
3718 rdev->data_offset < rdev->new_data_offset)
3719 goto out_unlock;
3720 if (olddisks > n &&
3721 rdev->data_offset > rdev->new_data_offset)
3722 goto out_unlock;
3723 }
3724 err = 0;
3725 mddev->delta_disks = n - olddisks;
3726 mddev->raid_disks = n;
3727 mddev->reshape_backwards = (mddev->delta_disks < 0);
3728 } else
3729 mddev->raid_disks = n;
3730 out_unlock:
3731 mddev_unlock(mddev);
3732 return err ? err : len;
3733 }
3734 static struct md_sysfs_entry md_raid_disks =
3735 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3736
3737 static ssize_t
3738 chunk_size_show(struct mddev *mddev, char *page)
3739 {
3740 if (mddev->reshape_position != MaxSector &&
3741 mddev->chunk_sectors != mddev->new_chunk_sectors)
3742 return sprintf(page, "%d (%d)\n",
3743 mddev->new_chunk_sectors << 9,
3744 mddev->chunk_sectors << 9);
3745 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3746 }
3747
3748 static ssize_t
3749 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3750 {
3751 unsigned long n;
3752 int err;
3753
3754 err = kstrtoul(buf, 10, &n);
3755 if (err < 0)
3756 return err;
3757
3758 err = mddev_lock(mddev);
3759 if (err)
3760 return err;
3761 if (mddev->pers) {
3762 if (mddev->pers->check_reshape == NULL)
3763 err = -EBUSY;
3764 else if (mddev->ro)
3765 err = -EROFS;
3766 else {
3767 mddev->new_chunk_sectors = n >> 9;
3768 err = mddev->pers->check_reshape(mddev);
3769 if (err)
3770 mddev->new_chunk_sectors = mddev->chunk_sectors;
3771 }
3772 } else {
3773 mddev->new_chunk_sectors = n >> 9;
3774 if (mddev->reshape_position == MaxSector)
3775 mddev->chunk_sectors = n >> 9;
3776 }
3777 mddev_unlock(mddev);
3778 return err ?: len;
3779 }
3780 static struct md_sysfs_entry md_chunk_size =
3781 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3782
3783 static ssize_t
3784 resync_start_show(struct mddev *mddev, char *page)
3785 {
3786 if (mddev->recovery_cp == MaxSector)
3787 return sprintf(page, "none\n");
3788 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3789 }
3790
3791 static ssize_t
3792 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3793 {
3794 unsigned long long n;
3795 int err;
3796
3797 if (cmd_match(buf, "none"))
3798 n = MaxSector;
3799 else {
3800 err = kstrtoull(buf, 10, &n);
3801 if (err < 0)
3802 return err;
3803 if (n != (sector_t)n)
3804 return -EINVAL;
3805 }
3806
3807 err = mddev_lock(mddev);
3808 if (err)
3809 return err;
3810 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3811 err = -EBUSY;
3812
3813 if (!err) {
3814 mddev->recovery_cp = n;
3815 if (mddev->pers)
3816 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3817 }
3818 mddev_unlock(mddev);
3819 return err ?: len;
3820 }
3821 static struct md_sysfs_entry md_resync_start =
3822 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3823 resync_start_show, resync_start_store);
3824
3825 /*
3826 * The array state can be:
3827 *
3828 * clear
3829 * No devices, no size, no level
3830 * Equivalent to STOP_ARRAY ioctl
3831 * inactive
3832 * May have some settings, but array is not active
3833 * all IO results in error
3834 * When written, doesn't tear down array, but just stops it
3835 * suspended (not supported yet)
3836 * All IO requests will block. The array can be reconfigured.
3837 * Writing this, if accepted, will block until array is quiescent
3838 * readonly
3839 * no resync can happen. no superblocks get written.
3840 * write requests fail
3841 * read-auto
3842 * like readonly, but behaves like 'clean' on a write request.
3843 *
3844 * clean - no pending writes, but otherwise active.
3845 * When written to inactive array, starts without resync
3846 * If a write request arrives then
3847 * if metadata is known, mark 'dirty' and switch to 'active'.
3848 * if not known, block and switch to write-pending
3849 * If written to an active array that has pending writes, then fails.
3850 * active
3851 * fully active: IO and resync can be happening.
3852 * When written to inactive array, starts with resync
3853 *
3854 * write-pending
3855 * clean, but writes are blocked waiting for 'active' to be written.
3856 *
3857 * active-idle
3858 * like active, but no writes have been seen for a while (100msec).
3859 *
3860 */
3861 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3862 write_pending, active_idle, bad_word};
3863 static char *array_states[] = {
3864 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3865 "write-pending", "active-idle", NULL };
3866
3867 static int match_word(const char *word, char **list)
3868 {
3869 int n;
3870 for (n=0; list[n]; n++)
3871 if (cmd_match(word, list[n]))
3872 break;
3873 return n;
3874 }
3875
3876 static ssize_t
3877 array_state_show(struct mddev *mddev, char *page)
3878 {
3879 enum array_state st = inactive;
3880
3881 if (mddev->pers)
3882 switch(mddev->ro) {
3883 case 1:
3884 st = readonly;
3885 break;
3886 case 2:
3887 st = read_auto;
3888 break;
3889 case 0:
3890 if (mddev->in_sync)
3891 st = clean;
3892 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3893 st = write_pending;
3894 else if (mddev->safemode)
3895 st = active_idle;
3896 else
3897 st = active;
3898 }
3899 else {
3900 if (list_empty(&mddev->disks) &&
3901 mddev->raid_disks == 0 &&
3902 mddev->dev_sectors == 0)
3903 st = clear;
3904 else
3905 st = inactive;
3906 }
3907 return sprintf(page, "%s\n", array_states[st]);
3908 }
3909
3910 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3911 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3912 static int do_md_run(struct mddev *mddev);
3913 static int restart_array(struct mddev *mddev);
3914
3915 static ssize_t
3916 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3917 {
3918 int err;
3919 enum array_state st = match_word(buf, array_states);
3920
3921 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3922 /* don't take reconfig_mutex when toggling between
3923 * clean and active
3924 */
3925 spin_lock(&mddev->lock);
3926 if (st == active) {
3927 restart_array(mddev);
3928 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3929 wake_up(&mddev->sb_wait);
3930 err = 0;
3931 } else /* st == clean */ {
3932 restart_array(mddev);
3933 if (atomic_read(&mddev->writes_pending) == 0) {
3934 if (mddev->in_sync == 0) {
3935 mddev->in_sync = 1;
3936 if (mddev->safemode == 1)
3937 mddev->safemode = 0;
3938 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3939 }
3940 err = 0;
3941 } else
3942 err = -EBUSY;
3943 }
3944 if (!err)
3945 sysfs_notify_dirent_safe(mddev->sysfs_state);
3946 spin_unlock(&mddev->lock);
3947 return err ?: len;
3948 }
3949 err = mddev_lock(mddev);
3950 if (err)
3951 return err;
3952 err = -EINVAL;
3953 switch(st) {
3954 case bad_word:
3955 break;
3956 case clear:
3957 /* stopping an active array */
3958 err = do_md_stop(mddev, 0, NULL);
3959 break;
3960 case inactive:
3961 /* stopping an active array */
3962 if (mddev->pers)
3963 err = do_md_stop(mddev, 2, NULL);
3964 else
3965 err = 0; /* already inactive */
3966 break;
3967 case suspended:
3968 break; /* not supported yet */
3969 case readonly:
3970 if (mddev->pers)
3971 err = md_set_readonly(mddev, NULL);
3972 else {
3973 mddev->ro = 1;
3974 set_disk_ro(mddev->gendisk, 1);
3975 err = do_md_run(mddev);
3976 }
3977 break;
3978 case read_auto:
3979 if (mddev->pers) {
3980 if (mddev->ro == 0)
3981 err = md_set_readonly(mddev, NULL);
3982 else if (mddev->ro == 1)
3983 err = restart_array(mddev);
3984 if (err == 0) {
3985 mddev->ro = 2;
3986 set_disk_ro(mddev->gendisk, 0);
3987 }
3988 } else {
3989 mddev->ro = 2;
3990 err = do_md_run(mddev);
3991 }
3992 break;
3993 case clean:
3994 if (mddev->pers) {
3995 err = restart_array(mddev);
3996 if (err)
3997 break;
3998 spin_lock(&mddev->lock);
3999 if (atomic_read(&mddev->writes_pending) == 0) {
4000 if (mddev->in_sync == 0) {
4001 mddev->in_sync = 1;
4002 if (mddev->safemode == 1)
4003 mddev->safemode = 0;
4004 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
4005 }
4006 err = 0;
4007 } else
4008 err = -EBUSY;
4009 spin_unlock(&mddev->lock);
4010 } else
4011 err = -EINVAL;
4012 break;
4013 case active:
4014 if (mddev->pers) {
4015 err = restart_array(mddev);
4016 if (err)
4017 break;
4018 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4019 wake_up(&mddev->sb_wait);
4020 err = 0;
4021 } else {
4022 mddev->ro = 0;
4023 set_disk_ro(mddev->gendisk, 0);
4024 err = do_md_run(mddev);
4025 }
4026 break;
4027 case write_pending:
4028 case active_idle:
4029 /* these cannot be set */
4030 break;
4031 }
4032
4033 if (!err) {
4034 if (mddev->hold_active == UNTIL_IOCTL)
4035 mddev->hold_active = 0;
4036 sysfs_notify_dirent_safe(mddev->sysfs_state);
4037 }
4038 mddev_unlock(mddev);
4039 return err ?: len;
4040 }
4041 static struct md_sysfs_entry md_array_state =
4042 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4043
4044 static ssize_t
4045 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4046 return sprintf(page, "%d\n",
4047 atomic_read(&mddev->max_corr_read_errors));
4048 }
4049
4050 static ssize_t
4051 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4052 {
4053 unsigned int n;
4054 int rv;
4055
4056 rv = kstrtouint(buf, 10, &n);
4057 if (rv < 0)
4058 return rv;
4059 atomic_set(&mddev->max_corr_read_errors, n);
4060 return len;
4061 }
4062
4063 static struct md_sysfs_entry max_corr_read_errors =
4064 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4065 max_corrected_read_errors_store);
4066
4067 static ssize_t
4068 null_show(struct mddev *mddev, char *page)
4069 {
4070 return -EINVAL;
4071 }
4072
4073 static ssize_t
4074 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4075 {
4076 /* buf must be %d:%d\n? giving major and minor numbers */
4077 /* The new device is added to the array.
4078 * If the array has a persistent superblock, we read the
4079 * superblock to initialise info and check validity.
4080 * Otherwise, only checking done is that in bind_rdev_to_array,
4081 * which mainly checks size.
4082 */
4083 char *e;
4084 int major = simple_strtoul(buf, &e, 10);
4085 int minor;
4086 dev_t dev;
4087 struct md_rdev *rdev;
4088 int err;
4089
4090 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4091 return -EINVAL;
4092 minor = simple_strtoul(e+1, &e, 10);
4093 if (*e && *e != '\n')
4094 return -EINVAL;
4095 dev = MKDEV(major, minor);
4096 if (major != MAJOR(dev) ||
4097 minor != MINOR(dev))
4098 return -EOVERFLOW;
4099
4100 flush_workqueue(md_misc_wq);
4101
4102 err = mddev_lock(mddev);
4103 if (err)
4104 return err;
4105 if (mddev->persistent) {
4106 rdev = md_import_device(dev, mddev->major_version,
4107 mddev->minor_version);
4108 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4109 struct md_rdev *rdev0
4110 = list_entry(mddev->disks.next,
4111 struct md_rdev, same_set);
4112 err = super_types[mddev->major_version]
4113 .load_super(rdev, rdev0, mddev->minor_version);
4114 if (err < 0)
4115 goto out;
4116 }
4117 } else if (mddev->external)
4118 rdev = md_import_device(dev, -2, -1);
4119 else
4120 rdev = md_import_device(dev, -1, -1);
4121
4122 if (IS_ERR(rdev)) {
4123 mddev_unlock(mddev);
4124 return PTR_ERR(rdev);
4125 }
4126 err = bind_rdev_to_array(rdev, mddev);
4127 out:
4128 if (err)
4129 export_rdev(rdev);
4130 mddev_unlock(mddev);
4131 return err ? err : len;
4132 }
4133
4134 static struct md_sysfs_entry md_new_device =
4135 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4136
4137 static ssize_t
4138 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4139 {
4140 char *end;
4141 unsigned long chunk, end_chunk;
4142 int err;
4143
4144 err = mddev_lock(mddev);
4145 if (err)
4146 return err;
4147 if (!mddev->bitmap)
4148 goto out;
4149 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4150 while (*buf) {
4151 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4152 if (buf == end) break;
4153 if (*end == '-') { /* range */
4154 buf = end + 1;
4155 end_chunk = simple_strtoul(buf, &end, 0);
4156 if (buf == end) break;
4157 }
4158 if (*end && !isspace(*end)) break;
4159 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4160 buf = skip_spaces(end);
4161 }
4162 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4163 out:
4164 mddev_unlock(mddev);
4165 return len;
4166 }
4167
4168 static struct md_sysfs_entry md_bitmap =
4169 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4170
4171 static ssize_t
4172 size_show(struct mddev *mddev, char *page)
4173 {
4174 return sprintf(page, "%llu\n",
4175 (unsigned long long)mddev->dev_sectors / 2);
4176 }
4177
4178 static int update_size(struct mddev *mddev, sector_t num_sectors);
4179
4180 static ssize_t
4181 size_store(struct mddev *mddev, const char *buf, size_t len)
4182 {
4183 /* If array is inactive, we can reduce the component size, but
4184 * not increase it (except from 0).
4185 * If array is active, we can try an on-line resize
4186 */
4187 sector_t sectors;
4188 int err = strict_blocks_to_sectors(buf, &sectors);
4189
4190 if (err < 0)
4191 return err;
4192 err = mddev_lock(mddev);
4193 if (err)
4194 return err;
4195 if (mddev->pers) {
4196 err = update_size(mddev, sectors);
4197 if (err == 0)
4198 md_update_sb(mddev, 1);
4199 } else {
4200 if (mddev->dev_sectors == 0 ||
4201 mddev->dev_sectors > sectors)
4202 mddev->dev_sectors = sectors;
4203 else
4204 err = -ENOSPC;
4205 }
4206 mddev_unlock(mddev);
4207 return err ? err : len;
4208 }
4209
4210 static struct md_sysfs_entry md_size =
4211 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4212
4213 /* Metadata version.
4214 * This is one of
4215 * 'none' for arrays with no metadata (good luck...)
4216 * 'external' for arrays with externally managed metadata,
4217 * or N.M for internally known formats
4218 */
4219 static ssize_t
4220 metadata_show(struct mddev *mddev, char *page)
4221 {
4222 if (mddev->persistent)
4223 return sprintf(page, "%d.%d\n",
4224 mddev->major_version, mddev->minor_version);
4225 else if (mddev->external)
4226 return sprintf(page, "external:%s\n", mddev->metadata_type);
4227 else
4228 return sprintf(page, "none\n");
4229 }
4230
4231 static ssize_t
4232 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4233 {
4234 int major, minor;
4235 char *e;
4236 int err;
4237 /* Changing the details of 'external' metadata is
4238 * always permitted. Otherwise there must be
4239 * no devices attached to the array.
4240 */
4241
4242 err = mddev_lock(mddev);
4243 if (err)
4244 return err;
4245 err = -EBUSY;
4246 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4247 ;
4248 else if (!list_empty(&mddev->disks))
4249 goto out_unlock;
4250
4251 err = 0;
4252 if (cmd_match(buf, "none")) {
4253 mddev->persistent = 0;
4254 mddev->external = 0;
4255 mddev->major_version = 0;
4256 mddev->minor_version = 90;
4257 goto out_unlock;
4258 }
4259 if (strncmp(buf, "external:", 9) == 0) {
4260 size_t namelen = len-9;
4261 if (namelen >= sizeof(mddev->metadata_type))
4262 namelen = sizeof(mddev->metadata_type)-1;
4263 strncpy(mddev->metadata_type, buf+9, namelen);
4264 mddev->metadata_type[namelen] = 0;
4265 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4266 mddev->metadata_type[--namelen] = 0;
4267 mddev->persistent = 0;
4268 mddev->external = 1;
4269 mddev->major_version = 0;
4270 mddev->minor_version = 90;
4271 goto out_unlock;
4272 }
4273 major = simple_strtoul(buf, &e, 10);
4274 err = -EINVAL;
4275 if (e==buf || *e != '.')
4276 goto out_unlock;
4277 buf = e+1;
4278 minor = simple_strtoul(buf, &e, 10);
4279 if (e==buf || (*e && *e != '\n') )
4280 goto out_unlock;
4281 err = -ENOENT;
4282 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4283 goto out_unlock;
4284 mddev->major_version = major;
4285 mddev->minor_version = minor;
4286 mddev->persistent = 1;
4287 mddev->external = 0;
4288 err = 0;
4289 out_unlock:
4290 mddev_unlock(mddev);
4291 return err ?: len;
4292 }
4293
4294 static struct md_sysfs_entry md_metadata =
4295 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4296
4297 static ssize_t
4298 action_show(struct mddev *mddev, char *page)
4299 {
4300 char *type = "idle";
4301 unsigned long recovery = mddev->recovery;
4302 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4303 type = "frozen";
4304 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4305 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4306 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4307 type = "reshape";
4308 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4309 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4310 type = "resync";
4311 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4312 type = "check";
4313 else
4314 type = "repair";
4315 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4316 type = "recover";
4317 else if (mddev->reshape_position != MaxSector)
4318 type = "reshape";
4319 }
4320 return sprintf(page, "%s\n", type);
4321 }
4322
4323 static ssize_t
4324 action_store(struct mddev *mddev, const char *page, size_t len)
4325 {
4326 if (!mddev->pers || !mddev->pers->sync_request)
4327 return -EINVAL;
4328
4329
4330 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4331 if (cmd_match(page, "frozen"))
4332 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4333 else
4334 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4335 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4336 mddev_lock(mddev) == 0) {
4337 flush_workqueue(md_misc_wq);
4338 if (mddev->sync_thread) {
4339 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4340 md_reap_sync_thread(mddev);
4341 }
4342 mddev_unlock(mddev);
4343 }
4344 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4345 return -EBUSY;
4346 else if (cmd_match(page, "resync"))
4347 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4348 else if (cmd_match(page, "recover")) {
4349 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4350 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4351 } else if (cmd_match(page, "reshape")) {
4352 int err;
4353 if (mddev->pers->start_reshape == NULL)
4354 return -EINVAL;
4355 err = mddev_lock(mddev);
4356 if (!err) {
4357 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4358 err = -EBUSY;
4359 else {
4360 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4361 err = mddev->pers->start_reshape(mddev);
4362 }
4363 mddev_unlock(mddev);
4364 }
4365 if (err)
4366 return err;
4367 sysfs_notify(&mddev->kobj, NULL, "degraded");
4368 } else {
4369 if (cmd_match(page, "check"))
4370 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4371 else if (!cmd_match(page, "repair"))
4372 return -EINVAL;
4373 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4374 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4375 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4376 }
4377 if (mddev->ro == 2) {
4378 /* A write to sync_action is enough to justify
4379 * canceling read-auto mode
4380 */
4381 mddev->ro = 0;
4382 md_wakeup_thread(mddev->sync_thread);
4383 }
4384 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4385 md_wakeup_thread(mddev->thread);
4386 sysfs_notify_dirent_safe(mddev->sysfs_action);
4387 return len;
4388 }
4389
4390 static struct md_sysfs_entry md_scan_mode =
4391 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4392
4393 static ssize_t
4394 last_sync_action_show(struct mddev *mddev, char *page)
4395 {
4396 return sprintf(page, "%s\n", mddev->last_sync_action);
4397 }
4398
4399 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4400
4401 static ssize_t
4402 mismatch_cnt_show(struct mddev *mddev, char *page)
4403 {
4404 return sprintf(page, "%llu\n",
4405 (unsigned long long)
4406 atomic64_read(&mddev->resync_mismatches));
4407 }
4408
4409 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4410
4411 static ssize_t
4412 sync_min_show(struct mddev *mddev, char *page)
4413 {
4414 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4415 mddev->sync_speed_min ? "local": "system");
4416 }
4417
4418 static ssize_t
4419 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4420 {
4421 unsigned int min;
4422 int rv;
4423
4424 if (strncmp(buf, "system", 6)==0) {
4425 min = 0;
4426 } else {
4427 rv = kstrtouint(buf, 10, &min);
4428 if (rv < 0)
4429 return rv;
4430 if (min == 0)
4431 return -EINVAL;
4432 }
4433 mddev->sync_speed_min = min;
4434 return len;
4435 }
4436
4437 static struct md_sysfs_entry md_sync_min =
4438 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4439
4440 static ssize_t
4441 sync_max_show(struct mddev *mddev, char *page)
4442 {
4443 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4444 mddev->sync_speed_max ? "local": "system");
4445 }
4446
4447 static ssize_t
4448 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4449 {
4450 unsigned int max;
4451 int rv;
4452
4453 if (strncmp(buf, "system", 6)==0) {
4454 max = 0;
4455 } else {
4456 rv = kstrtouint(buf, 10, &max);
4457 if (rv < 0)
4458 return rv;
4459 if (max == 0)
4460 return -EINVAL;
4461 }
4462 mddev->sync_speed_max = max;
4463 return len;
4464 }
4465
4466 static struct md_sysfs_entry md_sync_max =
4467 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4468
4469 static ssize_t
4470 degraded_show(struct mddev *mddev, char *page)
4471 {
4472 return sprintf(page, "%d\n", mddev->degraded);
4473 }
4474 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4475
4476 static ssize_t
4477 sync_force_parallel_show(struct mddev *mddev, char *page)
4478 {
4479 return sprintf(page, "%d\n", mddev->parallel_resync);
4480 }
4481
4482 static ssize_t
4483 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4484 {
4485 long n;
4486
4487 if (kstrtol(buf, 10, &n))
4488 return -EINVAL;
4489
4490 if (n != 0 && n != 1)
4491 return -EINVAL;
4492
4493 mddev->parallel_resync = n;
4494
4495 if (mddev->sync_thread)
4496 wake_up(&resync_wait);
4497
4498 return len;
4499 }
4500
4501 /* force parallel resync, even with shared block devices */
4502 static struct md_sysfs_entry md_sync_force_parallel =
4503 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4504 sync_force_parallel_show, sync_force_parallel_store);
4505
4506 static ssize_t
4507 sync_speed_show(struct mddev *mddev, char *page)
4508 {
4509 unsigned long resync, dt, db;
4510 if (mddev->curr_resync == 0)
4511 return sprintf(page, "none\n");
4512 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4513 dt = (jiffies - mddev->resync_mark) / HZ;
4514 if (!dt) dt++;
4515 db = resync - mddev->resync_mark_cnt;
4516 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4517 }
4518
4519 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4520
4521 static ssize_t
4522 sync_completed_show(struct mddev *mddev, char *page)
4523 {
4524 unsigned long long max_sectors, resync;
4525
4526 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4527 return sprintf(page, "none\n");
4528
4529 if (mddev->curr_resync == 1 ||
4530 mddev->curr_resync == 2)
4531 return sprintf(page, "delayed\n");
4532
4533 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4534 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4535 max_sectors = mddev->resync_max_sectors;
4536 else
4537 max_sectors = mddev->dev_sectors;
4538
4539 resync = mddev->curr_resync_completed;
4540 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4541 }
4542
4543 static struct md_sysfs_entry md_sync_completed =
4544 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4545
4546 static ssize_t
4547 min_sync_show(struct mddev *mddev, char *page)
4548 {
4549 return sprintf(page, "%llu\n",
4550 (unsigned long long)mddev->resync_min);
4551 }
4552 static ssize_t
4553 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4554 {
4555 unsigned long long min;
4556 int err;
4557
4558 if (kstrtoull(buf, 10, &min))
4559 return -EINVAL;
4560
4561 spin_lock(&mddev->lock);
4562 err = -EINVAL;
4563 if (min > mddev->resync_max)
4564 goto out_unlock;
4565
4566 err = -EBUSY;
4567 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4568 goto out_unlock;
4569
4570 /* Round down to multiple of 4K for safety */
4571 mddev->resync_min = round_down(min, 8);
4572 err = 0;
4573
4574 out_unlock:
4575 spin_unlock(&mddev->lock);
4576 return err ?: len;
4577 }
4578
4579 static struct md_sysfs_entry md_min_sync =
4580 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4581
4582 static ssize_t
4583 max_sync_show(struct mddev *mddev, char *page)
4584 {
4585 if (mddev->resync_max == MaxSector)
4586 return sprintf(page, "max\n");
4587 else
4588 return sprintf(page, "%llu\n",
4589 (unsigned long long)mddev->resync_max);
4590 }
4591 static ssize_t
4592 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4593 {
4594 int err;
4595 spin_lock(&mddev->lock);
4596 if (strncmp(buf, "max", 3) == 0)
4597 mddev->resync_max = MaxSector;
4598 else {
4599 unsigned long long max;
4600 int chunk;
4601
4602 err = -EINVAL;
4603 if (kstrtoull(buf, 10, &max))
4604 goto out_unlock;
4605 if (max < mddev->resync_min)
4606 goto out_unlock;
4607
4608 err = -EBUSY;
4609 if (max < mddev->resync_max &&
4610 mddev->ro == 0 &&
4611 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4612 goto out_unlock;
4613
4614 /* Must be a multiple of chunk_size */
4615 chunk = mddev->chunk_sectors;
4616 if (chunk) {
4617 sector_t temp = max;
4618
4619 err = -EINVAL;
4620 if (sector_div(temp, chunk))
4621 goto out_unlock;
4622 }
4623 mddev->resync_max = max;
4624 }
4625 wake_up(&mddev->recovery_wait);
4626 err = 0;
4627 out_unlock:
4628 spin_unlock(&mddev->lock);
4629 return err ?: len;
4630 }
4631
4632 static struct md_sysfs_entry md_max_sync =
4633 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4634
4635 static ssize_t
4636 suspend_lo_show(struct mddev *mddev, char *page)
4637 {
4638 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4639 }
4640
4641 static ssize_t
4642 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4643 {
4644 unsigned long long old, new;
4645 int err;
4646
4647 err = kstrtoull(buf, 10, &new);
4648 if (err < 0)
4649 return err;
4650 if (new != (sector_t)new)
4651 return -EINVAL;
4652
4653 err = mddev_lock(mddev);
4654 if (err)
4655 return err;
4656 err = -EINVAL;
4657 if (mddev->pers == NULL ||
4658 mddev->pers->quiesce == NULL)
4659 goto unlock;
4660 old = mddev->suspend_lo;
4661 mddev->suspend_lo = new;
4662 if (new >= old)
4663 /* Shrinking suspended region */
4664 mddev->pers->quiesce(mddev, 2);
4665 else {
4666 /* Expanding suspended region - need to wait */
4667 mddev->pers->quiesce(mddev, 1);
4668 mddev->pers->quiesce(mddev, 0);
4669 }
4670 err = 0;
4671 unlock:
4672 mddev_unlock(mddev);
4673 return err ?: len;
4674 }
4675 static struct md_sysfs_entry md_suspend_lo =
4676 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4677
4678 static ssize_t
4679 suspend_hi_show(struct mddev *mddev, char *page)
4680 {
4681 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4682 }
4683
4684 static ssize_t
4685 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4686 {
4687 unsigned long long old, new;
4688 int err;
4689
4690 err = kstrtoull(buf, 10, &new);
4691 if (err < 0)
4692 return err;
4693 if (new != (sector_t)new)
4694 return -EINVAL;
4695
4696 err = mddev_lock(mddev);
4697 if (err)
4698 return err;
4699 err = -EINVAL;
4700 if (mddev->pers == NULL ||
4701 mddev->pers->quiesce == NULL)
4702 goto unlock;
4703 old = mddev->suspend_hi;
4704 mddev->suspend_hi = new;
4705 if (new <= old)
4706 /* Shrinking suspended region */
4707 mddev->pers->quiesce(mddev, 2);
4708 else {
4709 /* Expanding suspended region - need to wait */
4710 mddev->pers->quiesce(mddev, 1);
4711 mddev->pers->quiesce(mddev, 0);
4712 }
4713 err = 0;
4714 unlock:
4715 mddev_unlock(mddev);
4716 return err ?: len;
4717 }
4718 static struct md_sysfs_entry md_suspend_hi =
4719 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4720
4721 static ssize_t
4722 reshape_position_show(struct mddev *mddev, char *page)
4723 {
4724 if (mddev->reshape_position != MaxSector)
4725 return sprintf(page, "%llu\n",
4726 (unsigned long long)mddev->reshape_position);
4727 strcpy(page, "none\n");
4728 return 5;
4729 }
4730
4731 static ssize_t
4732 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4733 {
4734 struct md_rdev *rdev;
4735 unsigned long long new;
4736 int err;
4737
4738 err = kstrtoull(buf, 10, &new);
4739 if (err < 0)
4740 return err;
4741 if (new != (sector_t)new)
4742 return -EINVAL;
4743 err = mddev_lock(mddev);
4744 if (err)
4745 return err;
4746 err = -EBUSY;
4747 if (mddev->pers)
4748 goto unlock;
4749 mddev->reshape_position = new;
4750 mddev->delta_disks = 0;
4751 mddev->reshape_backwards = 0;
4752 mddev->new_level = mddev->level;
4753 mddev->new_layout = mddev->layout;
4754 mddev->new_chunk_sectors = mddev->chunk_sectors;
4755 rdev_for_each(rdev, mddev)
4756 rdev->new_data_offset = rdev->data_offset;
4757 err = 0;
4758 unlock:
4759 mddev_unlock(mddev);
4760 return err ?: len;
4761 }
4762
4763 static struct md_sysfs_entry md_reshape_position =
4764 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4765 reshape_position_store);
4766
4767 static ssize_t
4768 reshape_direction_show(struct mddev *mddev, char *page)
4769 {
4770 return sprintf(page, "%s\n",
4771 mddev->reshape_backwards ? "backwards" : "forwards");
4772 }
4773
4774 static ssize_t
4775 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4776 {
4777 int backwards = 0;
4778 int err;
4779
4780 if (cmd_match(buf, "forwards"))
4781 backwards = 0;
4782 else if (cmd_match(buf, "backwards"))
4783 backwards = 1;
4784 else
4785 return -EINVAL;
4786 if (mddev->reshape_backwards == backwards)
4787 return len;
4788
4789 err = mddev_lock(mddev);
4790 if (err)
4791 return err;
4792 /* check if we are allowed to change */
4793 if (mddev->delta_disks)
4794 err = -EBUSY;
4795 else if (mddev->persistent &&
4796 mddev->major_version == 0)
4797 err = -EINVAL;
4798 else
4799 mddev->reshape_backwards = backwards;
4800 mddev_unlock(mddev);
4801 return err ?: len;
4802 }
4803
4804 static struct md_sysfs_entry md_reshape_direction =
4805 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4806 reshape_direction_store);
4807
4808 static ssize_t
4809 array_size_show(struct mddev *mddev, char *page)
4810 {
4811 if (mddev->external_size)
4812 return sprintf(page, "%llu\n",
4813 (unsigned long long)mddev->array_sectors/2);
4814 else
4815 return sprintf(page, "default\n");
4816 }
4817
4818 static ssize_t
4819 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4820 {
4821 sector_t sectors;
4822 int err;
4823
4824 err = mddev_lock(mddev);
4825 if (err)
4826 return err;
4827
4828 /* cluster raid doesn't support change array_sectors */
4829 if (mddev_is_clustered(mddev))
4830 return -EINVAL;
4831
4832 if (strncmp(buf, "default", 7) == 0) {
4833 if (mddev->pers)
4834 sectors = mddev->pers->size(mddev, 0, 0);
4835 else
4836 sectors = mddev->array_sectors;
4837
4838 mddev->external_size = 0;
4839 } else {
4840 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4841 err = -EINVAL;
4842 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4843 err = -E2BIG;
4844 else
4845 mddev->external_size = 1;
4846 }
4847
4848 if (!err) {
4849 mddev->array_sectors = sectors;
4850 if (mddev->pers) {
4851 set_capacity(mddev->gendisk, mddev->array_sectors);
4852 revalidate_disk(mddev->gendisk);
4853 }
4854 }
4855 mddev_unlock(mddev);
4856 return err ?: len;
4857 }
4858
4859 static struct md_sysfs_entry md_array_size =
4860 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4861 array_size_store);
4862
4863 static struct attribute *md_default_attrs[] = {
4864 &md_level.attr,
4865 &md_layout.attr,
4866 &md_raid_disks.attr,
4867 &md_chunk_size.attr,
4868 &md_size.attr,
4869 &md_resync_start.attr,
4870 &md_metadata.attr,
4871 &md_new_device.attr,
4872 &md_safe_delay.attr,
4873 &md_array_state.attr,
4874 &md_reshape_position.attr,
4875 &md_reshape_direction.attr,
4876 &md_array_size.attr,
4877 &max_corr_read_errors.attr,
4878 NULL,
4879 };
4880
4881 static struct attribute *md_redundancy_attrs[] = {
4882 &md_scan_mode.attr,
4883 &md_last_scan_mode.attr,
4884 &md_mismatches.attr,
4885 &md_sync_min.attr,
4886 &md_sync_max.attr,
4887 &md_sync_speed.attr,
4888 &md_sync_force_parallel.attr,
4889 &md_sync_completed.attr,
4890 &md_min_sync.attr,
4891 &md_max_sync.attr,
4892 &md_suspend_lo.attr,
4893 &md_suspend_hi.attr,
4894 &md_bitmap.attr,
4895 &md_degraded.attr,
4896 NULL,
4897 };
4898 static struct attribute_group md_redundancy_group = {
4899 .name = NULL,
4900 .attrs = md_redundancy_attrs,
4901 };
4902
4903 static ssize_t
4904 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4905 {
4906 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4907 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4908 ssize_t rv;
4909
4910 if (!entry->show)
4911 return -EIO;
4912 spin_lock(&all_mddevs_lock);
4913 if (list_empty(&mddev->all_mddevs)) {
4914 spin_unlock(&all_mddevs_lock);
4915 return -EBUSY;
4916 }
4917 mddev_get(mddev);
4918 spin_unlock(&all_mddevs_lock);
4919
4920 rv = entry->show(mddev, page);
4921 mddev_put(mddev);
4922 return rv;
4923 }
4924
4925 static ssize_t
4926 md_attr_store(struct kobject *kobj, struct attribute *attr,
4927 const char *page, size_t length)
4928 {
4929 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4930 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4931 ssize_t rv;
4932
4933 if (!entry->store)
4934 return -EIO;
4935 if (!capable(CAP_SYS_ADMIN))
4936 return -EACCES;
4937 spin_lock(&all_mddevs_lock);
4938 if (list_empty(&mddev->all_mddevs)) {
4939 spin_unlock(&all_mddevs_lock);
4940 return -EBUSY;
4941 }
4942 mddev_get(mddev);
4943 spin_unlock(&all_mddevs_lock);
4944 rv = entry->store(mddev, page, length);
4945 mddev_put(mddev);
4946 return rv;
4947 }
4948
4949 static void md_free(struct kobject *ko)
4950 {
4951 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4952
4953 if (mddev->sysfs_state)
4954 sysfs_put(mddev->sysfs_state);
4955
4956 if (mddev->queue)
4957 blk_cleanup_queue(mddev->queue);
4958 if (mddev->gendisk) {
4959 del_gendisk(mddev->gendisk);
4960 put_disk(mddev->gendisk);
4961 }
4962
4963 kfree(mddev);
4964 }
4965
4966 static const struct sysfs_ops md_sysfs_ops = {
4967 .show = md_attr_show,
4968 .store = md_attr_store,
4969 };
4970 static struct kobj_type md_ktype = {
4971 .release = md_free,
4972 .sysfs_ops = &md_sysfs_ops,
4973 .default_attrs = md_default_attrs,
4974 };
4975
4976 int mdp_major = 0;
4977
4978 static void mddev_delayed_delete(struct work_struct *ws)
4979 {
4980 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4981
4982 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4983 kobject_del(&mddev->kobj);
4984 kobject_put(&mddev->kobj);
4985 }
4986
4987 static int md_alloc(dev_t dev, char *name)
4988 {
4989 static DEFINE_MUTEX(disks_mutex);
4990 struct mddev *mddev = mddev_find(dev);
4991 struct gendisk *disk;
4992 int partitioned;
4993 int shift;
4994 int unit;
4995 int error;
4996
4997 if (!mddev)
4998 return -ENODEV;
4999
5000 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5001 shift = partitioned ? MdpMinorShift : 0;
5002 unit = MINOR(mddev->unit) >> shift;
5003
5004 /* wait for any previous instance of this device to be
5005 * completely removed (mddev_delayed_delete).
5006 */
5007 flush_workqueue(md_misc_wq);
5008
5009 mutex_lock(&disks_mutex);
5010 error = -EEXIST;
5011 if (mddev->gendisk)
5012 goto abort;
5013
5014 if (name) {
5015 /* Need to ensure that 'name' is not a duplicate.
5016 */
5017 struct mddev *mddev2;
5018 spin_lock(&all_mddevs_lock);
5019
5020 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5021 if (mddev2->gendisk &&
5022 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5023 spin_unlock(&all_mddevs_lock);
5024 goto abort;
5025 }
5026 spin_unlock(&all_mddevs_lock);
5027 }
5028
5029 error = -ENOMEM;
5030 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5031 if (!mddev->queue)
5032 goto abort;
5033 mddev->queue->queuedata = mddev;
5034
5035 blk_queue_make_request(mddev->queue, md_make_request);
5036 blk_set_stacking_limits(&mddev->queue->limits);
5037
5038 disk = alloc_disk(1 << shift);
5039 if (!disk) {
5040 blk_cleanup_queue(mddev->queue);
5041 mddev->queue = NULL;
5042 goto abort;
5043 }
5044 disk->major = MAJOR(mddev->unit);
5045 disk->first_minor = unit << shift;
5046 if (name)
5047 strcpy(disk->disk_name, name);
5048 else if (partitioned)
5049 sprintf(disk->disk_name, "md_d%d", unit);
5050 else
5051 sprintf(disk->disk_name, "md%d", unit);
5052 disk->fops = &md_fops;
5053 disk->private_data = mddev;
5054 disk->queue = mddev->queue;
5055 blk_queue_write_cache(mddev->queue, true, true);
5056 /* Allow extended partitions. This makes the
5057 * 'mdp' device redundant, but we can't really
5058 * remove it now.
5059 */
5060 disk->flags |= GENHD_FL_EXT_DEVT;
5061 mddev->gendisk = disk;
5062 /* As soon as we call add_disk(), another thread could get
5063 * through to md_open, so make sure it doesn't get too far
5064 */
5065 mutex_lock(&mddev->open_mutex);
5066 add_disk(disk);
5067
5068 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5069 &disk_to_dev(disk)->kobj, "%s", "md");
5070 if (error) {
5071 /* This isn't possible, but as kobject_init_and_add is marked
5072 * __must_check, we must do something with the result
5073 */
5074 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5075 disk->disk_name);
5076 error = 0;
5077 }
5078 if (mddev->kobj.sd &&
5079 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5080 printk(KERN_DEBUG "pointless warning\n");
5081 mutex_unlock(&mddev->open_mutex);
5082 abort:
5083 mutex_unlock(&disks_mutex);
5084 if (!error && mddev->kobj.sd) {
5085 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5086 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5087 }
5088 mddev_put(mddev);
5089 return error;
5090 }
5091
5092 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5093 {
5094 md_alloc(dev, NULL);
5095 return NULL;
5096 }
5097
5098 static int add_named_array(const char *val, struct kernel_param *kp)
5099 {
5100 /* val must be "md_*" where * is not all digits.
5101 * We allocate an array with a large free minor number, and
5102 * set the name to val. val must not already be an active name.
5103 */
5104 int len = strlen(val);
5105 char buf[DISK_NAME_LEN];
5106
5107 while (len && val[len-1] == '\n')
5108 len--;
5109 if (len >= DISK_NAME_LEN)
5110 return -E2BIG;
5111 strlcpy(buf, val, len+1);
5112 if (strncmp(buf, "md_", 3) != 0)
5113 return -EINVAL;
5114 return md_alloc(0, buf);
5115 }
5116
5117 static void md_safemode_timeout(unsigned long data)
5118 {
5119 struct mddev *mddev = (struct mddev *) data;
5120
5121 if (!atomic_read(&mddev->writes_pending)) {
5122 mddev->safemode = 1;
5123 if (mddev->external)
5124 sysfs_notify_dirent_safe(mddev->sysfs_state);
5125 }
5126 md_wakeup_thread(mddev->thread);
5127 }
5128
5129 static int start_dirty_degraded;
5130
5131 int md_run(struct mddev *mddev)
5132 {
5133 int err;
5134 struct md_rdev *rdev;
5135 struct md_personality *pers;
5136
5137 if (list_empty(&mddev->disks))
5138 /* cannot run an array with no devices.. */
5139 return -EINVAL;
5140
5141 if (mddev->pers)
5142 return -EBUSY;
5143 /* Cannot run until previous stop completes properly */
5144 if (mddev->sysfs_active)
5145 return -EBUSY;
5146
5147 /*
5148 * Analyze all RAID superblock(s)
5149 */
5150 if (!mddev->raid_disks) {
5151 if (!mddev->persistent)
5152 return -EINVAL;
5153 analyze_sbs(mddev);
5154 }
5155
5156 if (mddev->level != LEVEL_NONE)
5157 request_module("md-level-%d", mddev->level);
5158 else if (mddev->clevel[0])
5159 request_module("md-%s", mddev->clevel);
5160
5161 /*
5162 * Drop all container device buffers, from now on
5163 * the only valid external interface is through the md
5164 * device.
5165 */
5166 rdev_for_each(rdev, mddev) {
5167 if (test_bit(Faulty, &rdev->flags))
5168 continue;
5169 sync_blockdev(rdev->bdev);
5170 invalidate_bdev(rdev->bdev);
5171
5172 /* perform some consistency tests on the device.
5173 * We don't want the data to overlap the metadata,
5174 * Internal Bitmap issues have been handled elsewhere.
5175 */
5176 if (rdev->meta_bdev) {
5177 /* Nothing to check */;
5178 } else if (rdev->data_offset < rdev->sb_start) {
5179 if (mddev->dev_sectors &&
5180 rdev->data_offset + mddev->dev_sectors
5181 > rdev->sb_start) {
5182 printk("md: %s: data overlaps metadata\n",
5183 mdname(mddev));
5184 return -EINVAL;
5185 }
5186 } else {
5187 if (rdev->sb_start + rdev->sb_size/512
5188 > rdev->data_offset) {
5189 printk("md: %s: metadata overlaps data\n",
5190 mdname(mddev));
5191 return -EINVAL;
5192 }
5193 }
5194 sysfs_notify_dirent_safe(rdev->sysfs_state);
5195 }
5196
5197 if (mddev->bio_set == NULL)
5198 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5199
5200 spin_lock(&pers_lock);
5201 pers = find_pers(mddev->level, mddev->clevel);
5202 if (!pers || !try_module_get(pers->owner)) {
5203 spin_unlock(&pers_lock);
5204 if (mddev->level != LEVEL_NONE)
5205 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5206 mddev->level);
5207 else
5208 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5209 mddev->clevel);
5210 return -EINVAL;
5211 }
5212 spin_unlock(&pers_lock);
5213 if (mddev->level != pers->level) {
5214 mddev->level = pers->level;
5215 mddev->new_level = pers->level;
5216 }
5217 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5218
5219 if (mddev->reshape_position != MaxSector &&
5220 pers->start_reshape == NULL) {
5221 /* This personality cannot handle reshaping... */
5222 module_put(pers->owner);
5223 return -EINVAL;
5224 }
5225
5226 if (pers->sync_request) {
5227 /* Warn if this is a potentially silly
5228 * configuration.
5229 */
5230 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5231 struct md_rdev *rdev2;
5232 int warned = 0;
5233
5234 rdev_for_each(rdev, mddev)
5235 rdev_for_each(rdev2, mddev) {
5236 if (rdev < rdev2 &&
5237 rdev->bdev->bd_contains ==
5238 rdev2->bdev->bd_contains) {
5239 printk(KERN_WARNING
5240 "%s: WARNING: %s appears to be"
5241 " on the same physical disk as"
5242 " %s.\n",
5243 mdname(mddev),
5244 bdevname(rdev->bdev,b),
5245 bdevname(rdev2->bdev,b2));
5246 warned = 1;
5247 }
5248 }
5249
5250 if (warned)
5251 printk(KERN_WARNING
5252 "True protection against single-disk"
5253 " failure might be compromised.\n");
5254 }
5255
5256 mddev->recovery = 0;
5257 /* may be over-ridden by personality */
5258 mddev->resync_max_sectors = mddev->dev_sectors;
5259
5260 mddev->ok_start_degraded = start_dirty_degraded;
5261
5262 if (start_readonly && mddev->ro == 0)
5263 mddev->ro = 2; /* read-only, but switch on first write */
5264
5265 err = pers->run(mddev);
5266 if (err)
5267 printk(KERN_ERR "md: pers->run() failed ...\n");
5268 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5269 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5270 " but 'external_size' not in effect?\n", __func__);
5271 printk(KERN_ERR
5272 "md: invalid array_size %llu > default size %llu\n",
5273 (unsigned long long)mddev->array_sectors / 2,
5274 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5275 err = -EINVAL;
5276 }
5277 if (err == 0 && pers->sync_request &&
5278 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5279 struct bitmap *bitmap;
5280
5281 bitmap = bitmap_create(mddev, -1);
5282 if (IS_ERR(bitmap)) {
5283 err = PTR_ERR(bitmap);
5284 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5285 mdname(mddev), err);
5286 } else
5287 mddev->bitmap = bitmap;
5288
5289 }
5290 if (err) {
5291 mddev_detach(mddev);
5292 if (mddev->private)
5293 pers->free(mddev, mddev->private);
5294 mddev->private = NULL;
5295 module_put(pers->owner);
5296 bitmap_destroy(mddev);
5297 return err;
5298 }
5299 if (mddev->queue) {
5300 mddev->queue->backing_dev_info.congested_data = mddev;
5301 mddev->queue->backing_dev_info.congested_fn = md_congested;
5302 }
5303 if (pers->sync_request) {
5304 if (mddev->kobj.sd &&
5305 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5306 printk(KERN_WARNING
5307 "md: cannot register extra attributes for %s\n",
5308 mdname(mddev));
5309 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5310 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5311 mddev->ro = 0;
5312
5313 atomic_set(&mddev->writes_pending,0);
5314 atomic_set(&mddev->max_corr_read_errors,
5315 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5316 mddev->safemode = 0;
5317 if (mddev_is_clustered(mddev))
5318 mddev->safemode_delay = 0;
5319 else
5320 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5321 mddev->in_sync = 1;
5322 smp_wmb();
5323 spin_lock(&mddev->lock);
5324 mddev->pers = pers;
5325 spin_unlock(&mddev->lock);
5326 rdev_for_each(rdev, mddev)
5327 if (rdev->raid_disk >= 0)
5328 if (sysfs_link_rdev(mddev, rdev))
5329 /* failure here is OK */;
5330
5331 if (mddev->degraded && !mddev->ro)
5332 /* This ensures that recovering status is reported immediately
5333 * via sysfs - until a lack of spares is confirmed.
5334 */
5335 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5336 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5337
5338 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5339 md_update_sb(mddev, 0);
5340
5341 md_new_event(mddev);
5342 sysfs_notify_dirent_safe(mddev->sysfs_state);
5343 sysfs_notify_dirent_safe(mddev->sysfs_action);
5344 sysfs_notify(&mddev->kobj, NULL, "degraded");
5345 return 0;
5346 }
5347 EXPORT_SYMBOL_GPL(md_run);
5348
5349 static int do_md_run(struct mddev *mddev)
5350 {
5351 int err;
5352
5353 err = md_run(mddev);
5354 if (err)
5355 goto out;
5356 err = bitmap_load(mddev);
5357 if (err) {
5358 bitmap_destroy(mddev);
5359 goto out;
5360 }
5361
5362 if (mddev_is_clustered(mddev))
5363 md_allow_write(mddev);
5364
5365 md_wakeup_thread(mddev->thread);
5366 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5367
5368 set_capacity(mddev->gendisk, mddev->array_sectors);
5369 revalidate_disk(mddev->gendisk);
5370 mddev->changed = 1;
5371 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5372 out:
5373 return err;
5374 }
5375
5376 static int restart_array(struct mddev *mddev)
5377 {
5378 struct gendisk *disk = mddev->gendisk;
5379
5380 /* Complain if it has no devices */
5381 if (list_empty(&mddev->disks))
5382 return -ENXIO;
5383 if (!mddev->pers)
5384 return -EINVAL;
5385 if (!mddev->ro)
5386 return -EBUSY;
5387 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5388 struct md_rdev *rdev;
5389 bool has_journal = false;
5390
5391 rcu_read_lock();
5392 rdev_for_each_rcu(rdev, mddev) {
5393 if (test_bit(Journal, &rdev->flags) &&
5394 !test_bit(Faulty, &rdev->flags)) {
5395 has_journal = true;
5396 break;
5397 }
5398 }
5399 rcu_read_unlock();
5400
5401 /* Don't restart rw with journal missing/faulty */
5402 if (!has_journal)
5403 return -EINVAL;
5404 }
5405
5406 mddev->safemode = 0;
5407 mddev->ro = 0;
5408 set_disk_ro(disk, 0);
5409 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5410 mdname(mddev));
5411 /* Kick recovery or resync if necessary */
5412 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5413 md_wakeup_thread(mddev->thread);
5414 md_wakeup_thread(mddev->sync_thread);
5415 sysfs_notify_dirent_safe(mddev->sysfs_state);
5416 return 0;
5417 }
5418
5419 static void md_clean(struct mddev *mddev)
5420 {
5421 mddev->array_sectors = 0;
5422 mddev->external_size = 0;
5423 mddev->dev_sectors = 0;
5424 mddev->raid_disks = 0;
5425 mddev->recovery_cp = 0;
5426 mddev->resync_min = 0;
5427 mddev->resync_max = MaxSector;
5428 mddev->reshape_position = MaxSector;
5429 mddev->external = 0;
5430 mddev->persistent = 0;
5431 mddev->level = LEVEL_NONE;
5432 mddev->clevel[0] = 0;
5433 mddev->flags = 0;
5434 mddev->ro = 0;
5435 mddev->metadata_type[0] = 0;
5436 mddev->chunk_sectors = 0;
5437 mddev->ctime = mddev->utime = 0;
5438 mddev->layout = 0;
5439 mddev->max_disks = 0;
5440 mddev->events = 0;
5441 mddev->can_decrease_events = 0;
5442 mddev->delta_disks = 0;
5443 mddev->reshape_backwards = 0;
5444 mddev->new_level = LEVEL_NONE;
5445 mddev->new_layout = 0;
5446 mddev->new_chunk_sectors = 0;
5447 mddev->curr_resync = 0;
5448 atomic64_set(&mddev->resync_mismatches, 0);
5449 mddev->suspend_lo = mddev->suspend_hi = 0;
5450 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5451 mddev->recovery = 0;
5452 mddev->in_sync = 0;
5453 mddev->changed = 0;
5454 mddev->degraded = 0;
5455 mddev->safemode = 0;
5456 mddev->private = NULL;
5457 mddev->bitmap_info.offset = 0;
5458 mddev->bitmap_info.default_offset = 0;
5459 mddev->bitmap_info.default_space = 0;
5460 mddev->bitmap_info.chunksize = 0;
5461 mddev->bitmap_info.daemon_sleep = 0;
5462 mddev->bitmap_info.max_write_behind = 0;
5463 }
5464
5465 static void __md_stop_writes(struct mddev *mddev)
5466 {
5467 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5468 flush_workqueue(md_misc_wq);
5469 if (mddev->sync_thread) {
5470 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5471 md_reap_sync_thread(mddev);
5472 }
5473
5474 del_timer_sync(&mddev->safemode_timer);
5475
5476 bitmap_flush(mddev);
5477 md_super_wait(mddev);
5478
5479 if (mddev->ro == 0 &&
5480 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5481 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5482 /* mark array as shutdown cleanly */
5483 if (!mddev_is_clustered(mddev))
5484 mddev->in_sync = 1;
5485 md_update_sb(mddev, 1);
5486 }
5487 }
5488
5489 void md_stop_writes(struct mddev *mddev)
5490 {
5491 mddev_lock_nointr(mddev);
5492 __md_stop_writes(mddev);
5493 mddev_unlock(mddev);
5494 }
5495 EXPORT_SYMBOL_GPL(md_stop_writes);
5496
5497 static void mddev_detach(struct mddev *mddev)
5498 {
5499 struct bitmap *bitmap = mddev->bitmap;
5500 /* wait for behind writes to complete */
5501 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5502 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5503 mdname(mddev));
5504 /* need to kick something here to make sure I/O goes? */
5505 wait_event(bitmap->behind_wait,
5506 atomic_read(&bitmap->behind_writes) == 0);
5507 }
5508 if (mddev->pers && mddev->pers->quiesce) {
5509 mddev->pers->quiesce(mddev, 1);
5510 mddev->pers->quiesce(mddev, 0);
5511 }
5512 md_unregister_thread(&mddev->thread);
5513 if (mddev->queue)
5514 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5515 }
5516
5517 static void __md_stop(struct mddev *mddev)
5518 {
5519 struct md_personality *pers = mddev->pers;
5520 mddev_detach(mddev);
5521 /* Ensure ->event_work is done */
5522 flush_workqueue(md_misc_wq);
5523 spin_lock(&mddev->lock);
5524 mddev->pers = NULL;
5525 spin_unlock(&mddev->lock);
5526 pers->free(mddev, mddev->private);
5527 mddev->private = NULL;
5528 if (pers->sync_request && mddev->to_remove == NULL)
5529 mddev->to_remove = &md_redundancy_group;
5530 module_put(pers->owner);
5531 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5532 }
5533
5534 void md_stop(struct mddev *mddev)
5535 {
5536 /* stop the array and free an attached data structures.
5537 * This is called from dm-raid
5538 */
5539 __md_stop(mddev);
5540 bitmap_destroy(mddev);
5541 if (mddev->bio_set)
5542 bioset_free(mddev->bio_set);
5543 }
5544
5545 EXPORT_SYMBOL_GPL(md_stop);
5546
5547 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5548 {
5549 int err = 0;
5550 int did_freeze = 0;
5551
5552 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5553 did_freeze = 1;
5554 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5555 md_wakeup_thread(mddev->thread);
5556 }
5557 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5558 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5559 if (mddev->sync_thread)
5560 /* Thread might be blocked waiting for metadata update
5561 * which will now never happen */
5562 wake_up_process(mddev->sync_thread->tsk);
5563
5564 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5565 return -EBUSY;
5566 mddev_unlock(mddev);
5567 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5568 &mddev->recovery));
5569 wait_event(mddev->sb_wait,
5570 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5571 mddev_lock_nointr(mddev);
5572
5573 mutex_lock(&mddev->open_mutex);
5574 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5575 mddev->sync_thread ||
5576 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5577 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5578 printk("md: %s still in use.\n",mdname(mddev));
5579 if (did_freeze) {
5580 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5581 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5582 md_wakeup_thread(mddev->thread);
5583 }
5584 err = -EBUSY;
5585 goto out;
5586 }
5587 if (mddev->pers) {
5588 __md_stop_writes(mddev);
5589
5590 err = -ENXIO;
5591 if (mddev->ro==1)
5592 goto out;
5593 mddev->ro = 1;
5594 set_disk_ro(mddev->gendisk, 1);
5595 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5596 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5597 md_wakeup_thread(mddev->thread);
5598 sysfs_notify_dirent_safe(mddev->sysfs_state);
5599 err = 0;
5600 }
5601 out:
5602 mutex_unlock(&mddev->open_mutex);
5603 return err;
5604 }
5605
5606 /* mode:
5607 * 0 - completely stop and dis-assemble array
5608 * 2 - stop but do not disassemble array
5609 */
5610 static int do_md_stop(struct mddev *mddev, int mode,
5611 struct block_device *bdev)
5612 {
5613 struct gendisk *disk = mddev->gendisk;
5614 struct md_rdev *rdev;
5615 int did_freeze = 0;
5616
5617 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5618 did_freeze = 1;
5619 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5620 md_wakeup_thread(mddev->thread);
5621 }
5622 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5623 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5624 if (mddev->sync_thread)
5625 /* Thread might be blocked waiting for metadata update
5626 * which will now never happen */
5627 wake_up_process(mddev->sync_thread->tsk);
5628
5629 mddev_unlock(mddev);
5630 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5631 !test_bit(MD_RECOVERY_RUNNING,
5632 &mddev->recovery)));
5633 mddev_lock_nointr(mddev);
5634
5635 mutex_lock(&mddev->open_mutex);
5636 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5637 mddev->sysfs_active ||
5638 mddev->sync_thread ||
5639 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5640 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5641 printk("md: %s still in use.\n",mdname(mddev));
5642 mutex_unlock(&mddev->open_mutex);
5643 if (did_freeze) {
5644 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5645 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5646 md_wakeup_thread(mddev->thread);
5647 }
5648 return -EBUSY;
5649 }
5650 if (mddev->pers) {
5651 if (mddev->ro)
5652 set_disk_ro(disk, 0);
5653
5654 __md_stop_writes(mddev);
5655 __md_stop(mddev);
5656 mddev->queue->backing_dev_info.congested_fn = NULL;
5657
5658 /* tell userspace to handle 'inactive' */
5659 sysfs_notify_dirent_safe(mddev->sysfs_state);
5660
5661 rdev_for_each(rdev, mddev)
5662 if (rdev->raid_disk >= 0)
5663 sysfs_unlink_rdev(mddev, rdev);
5664
5665 set_capacity(disk, 0);
5666 mutex_unlock(&mddev->open_mutex);
5667 mddev->changed = 1;
5668 revalidate_disk(disk);
5669
5670 if (mddev->ro)
5671 mddev->ro = 0;
5672 } else
5673 mutex_unlock(&mddev->open_mutex);
5674 /*
5675 * Free resources if final stop
5676 */
5677 if (mode == 0) {
5678 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5679
5680 bitmap_destroy(mddev);
5681 if (mddev->bitmap_info.file) {
5682 struct file *f = mddev->bitmap_info.file;
5683 spin_lock(&mddev->lock);
5684 mddev->bitmap_info.file = NULL;
5685 spin_unlock(&mddev->lock);
5686 fput(f);
5687 }
5688 mddev->bitmap_info.offset = 0;
5689
5690 export_array(mddev);
5691
5692 md_clean(mddev);
5693 if (mddev->hold_active == UNTIL_STOP)
5694 mddev->hold_active = 0;
5695 }
5696 md_new_event(mddev);
5697 sysfs_notify_dirent_safe(mddev->sysfs_state);
5698 return 0;
5699 }
5700
5701 #ifndef MODULE
5702 static void autorun_array(struct mddev *mddev)
5703 {
5704 struct md_rdev *rdev;
5705 int err;
5706
5707 if (list_empty(&mddev->disks))
5708 return;
5709
5710 printk(KERN_INFO "md: running: ");
5711
5712 rdev_for_each(rdev, mddev) {
5713 char b[BDEVNAME_SIZE];
5714 printk("<%s>", bdevname(rdev->bdev,b));
5715 }
5716 printk("\n");
5717
5718 err = do_md_run(mddev);
5719 if (err) {
5720 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5721 do_md_stop(mddev, 0, NULL);
5722 }
5723 }
5724
5725 /*
5726 * lets try to run arrays based on all disks that have arrived
5727 * until now. (those are in pending_raid_disks)
5728 *
5729 * the method: pick the first pending disk, collect all disks with
5730 * the same UUID, remove all from the pending list and put them into
5731 * the 'same_array' list. Then order this list based on superblock
5732 * update time (freshest comes first), kick out 'old' disks and
5733 * compare superblocks. If everything's fine then run it.
5734 *
5735 * If "unit" is allocated, then bump its reference count
5736 */
5737 static void autorun_devices(int part)
5738 {
5739 struct md_rdev *rdev0, *rdev, *tmp;
5740 struct mddev *mddev;
5741 char b[BDEVNAME_SIZE];
5742
5743 printk(KERN_INFO "md: autorun ...\n");
5744 while (!list_empty(&pending_raid_disks)) {
5745 int unit;
5746 dev_t dev;
5747 LIST_HEAD(candidates);
5748 rdev0 = list_entry(pending_raid_disks.next,
5749 struct md_rdev, same_set);
5750
5751 printk(KERN_INFO "md: considering %s ...\n",
5752 bdevname(rdev0->bdev,b));
5753 INIT_LIST_HEAD(&candidates);
5754 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5755 if (super_90_load(rdev, rdev0, 0) >= 0) {
5756 printk(KERN_INFO "md: adding %s ...\n",
5757 bdevname(rdev->bdev,b));
5758 list_move(&rdev->same_set, &candidates);
5759 }
5760 /*
5761 * now we have a set of devices, with all of them having
5762 * mostly sane superblocks. It's time to allocate the
5763 * mddev.
5764 */
5765 if (part) {
5766 dev = MKDEV(mdp_major,
5767 rdev0->preferred_minor << MdpMinorShift);
5768 unit = MINOR(dev) >> MdpMinorShift;
5769 } else {
5770 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5771 unit = MINOR(dev);
5772 }
5773 if (rdev0->preferred_minor != unit) {
5774 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5775 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5776 break;
5777 }
5778
5779 md_probe(dev, NULL, NULL);
5780 mddev = mddev_find(dev);
5781 if (!mddev || !mddev->gendisk) {
5782 if (mddev)
5783 mddev_put(mddev);
5784 printk(KERN_ERR
5785 "md: cannot allocate memory for md drive.\n");
5786 break;
5787 }
5788 if (mddev_lock(mddev))
5789 printk(KERN_WARNING "md: %s locked, cannot run\n",
5790 mdname(mddev));
5791 else if (mddev->raid_disks || mddev->major_version
5792 || !list_empty(&mddev->disks)) {
5793 printk(KERN_WARNING
5794 "md: %s already running, cannot run %s\n",
5795 mdname(mddev), bdevname(rdev0->bdev,b));
5796 mddev_unlock(mddev);
5797 } else {
5798 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5799 mddev->persistent = 1;
5800 rdev_for_each_list(rdev, tmp, &candidates) {
5801 list_del_init(&rdev->same_set);
5802 if (bind_rdev_to_array(rdev, mddev))
5803 export_rdev(rdev);
5804 }
5805 autorun_array(mddev);
5806 mddev_unlock(mddev);
5807 }
5808 /* on success, candidates will be empty, on error
5809 * it won't...
5810 */
5811 rdev_for_each_list(rdev, tmp, &candidates) {
5812 list_del_init(&rdev->same_set);
5813 export_rdev(rdev);
5814 }
5815 mddev_put(mddev);
5816 }
5817 printk(KERN_INFO "md: ... autorun DONE.\n");
5818 }
5819 #endif /* !MODULE */
5820
5821 static int get_version(void __user *arg)
5822 {
5823 mdu_version_t ver;
5824
5825 ver.major = MD_MAJOR_VERSION;
5826 ver.minor = MD_MINOR_VERSION;
5827 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5828
5829 if (copy_to_user(arg, &ver, sizeof(ver)))
5830 return -EFAULT;
5831
5832 return 0;
5833 }
5834
5835 static int get_array_info(struct mddev *mddev, void __user *arg)
5836 {
5837 mdu_array_info_t info;
5838 int nr,working,insync,failed,spare;
5839 struct md_rdev *rdev;
5840
5841 nr = working = insync = failed = spare = 0;
5842 rcu_read_lock();
5843 rdev_for_each_rcu(rdev, mddev) {
5844 nr++;
5845 if (test_bit(Faulty, &rdev->flags))
5846 failed++;
5847 else {
5848 working++;
5849 if (test_bit(In_sync, &rdev->flags))
5850 insync++;
5851 else if (test_bit(Journal, &rdev->flags))
5852 /* TODO: add journal count to md_u.h */
5853 ;
5854 else
5855 spare++;
5856 }
5857 }
5858 rcu_read_unlock();
5859
5860 info.major_version = mddev->major_version;
5861 info.minor_version = mddev->minor_version;
5862 info.patch_version = MD_PATCHLEVEL_VERSION;
5863 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5864 info.level = mddev->level;
5865 info.size = mddev->dev_sectors / 2;
5866 if (info.size != mddev->dev_sectors / 2) /* overflow */
5867 info.size = -1;
5868 info.nr_disks = nr;
5869 info.raid_disks = mddev->raid_disks;
5870 info.md_minor = mddev->md_minor;
5871 info.not_persistent= !mddev->persistent;
5872
5873 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5874 info.state = 0;
5875 if (mddev->in_sync)
5876 info.state = (1<<MD_SB_CLEAN);
5877 if (mddev->bitmap && mddev->bitmap_info.offset)
5878 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5879 if (mddev_is_clustered(mddev))
5880 info.state |= (1<<MD_SB_CLUSTERED);
5881 info.active_disks = insync;
5882 info.working_disks = working;
5883 info.failed_disks = failed;
5884 info.spare_disks = spare;
5885
5886 info.layout = mddev->layout;
5887 info.chunk_size = mddev->chunk_sectors << 9;
5888
5889 if (copy_to_user(arg, &info, sizeof(info)))
5890 return -EFAULT;
5891
5892 return 0;
5893 }
5894
5895 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5896 {
5897 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5898 char *ptr;
5899 int err;
5900
5901 file = kzalloc(sizeof(*file), GFP_NOIO);
5902 if (!file)
5903 return -ENOMEM;
5904
5905 err = 0;
5906 spin_lock(&mddev->lock);
5907 /* bitmap enabled */
5908 if (mddev->bitmap_info.file) {
5909 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5910 sizeof(file->pathname));
5911 if (IS_ERR(ptr))
5912 err = PTR_ERR(ptr);
5913 else
5914 memmove(file->pathname, ptr,
5915 sizeof(file->pathname)-(ptr-file->pathname));
5916 }
5917 spin_unlock(&mddev->lock);
5918
5919 if (err == 0 &&
5920 copy_to_user(arg, file, sizeof(*file)))
5921 err = -EFAULT;
5922
5923 kfree(file);
5924 return err;
5925 }
5926
5927 static int get_disk_info(struct mddev *mddev, void __user * arg)
5928 {
5929 mdu_disk_info_t info;
5930 struct md_rdev *rdev;
5931
5932 if (copy_from_user(&info, arg, sizeof(info)))
5933 return -EFAULT;
5934
5935 rcu_read_lock();
5936 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5937 if (rdev) {
5938 info.major = MAJOR(rdev->bdev->bd_dev);
5939 info.minor = MINOR(rdev->bdev->bd_dev);
5940 info.raid_disk = rdev->raid_disk;
5941 info.state = 0;
5942 if (test_bit(Faulty, &rdev->flags))
5943 info.state |= (1<<MD_DISK_FAULTY);
5944 else if (test_bit(In_sync, &rdev->flags)) {
5945 info.state |= (1<<MD_DISK_ACTIVE);
5946 info.state |= (1<<MD_DISK_SYNC);
5947 }
5948 if (test_bit(Journal, &rdev->flags))
5949 info.state |= (1<<MD_DISK_JOURNAL);
5950 if (test_bit(WriteMostly, &rdev->flags))
5951 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5952 } else {
5953 info.major = info.minor = 0;
5954 info.raid_disk = -1;
5955 info.state = (1<<MD_DISK_REMOVED);
5956 }
5957 rcu_read_unlock();
5958
5959 if (copy_to_user(arg, &info, sizeof(info)))
5960 return -EFAULT;
5961
5962 return 0;
5963 }
5964
5965 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5966 {
5967 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5968 struct md_rdev *rdev;
5969 dev_t dev = MKDEV(info->major,info->minor);
5970
5971 if (mddev_is_clustered(mddev) &&
5972 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5973 pr_err("%s: Cannot add to clustered mddev.\n",
5974 mdname(mddev));
5975 return -EINVAL;
5976 }
5977
5978 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5979 return -EOVERFLOW;
5980
5981 if (!mddev->raid_disks) {
5982 int err;
5983 /* expecting a device which has a superblock */
5984 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5985 if (IS_ERR(rdev)) {
5986 printk(KERN_WARNING
5987 "md: md_import_device returned %ld\n",
5988 PTR_ERR(rdev));
5989 return PTR_ERR(rdev);
5990 }
5991 if (!list_empty(&mddev->disks)) {
5992 struct md_rdev *rdev0
5993 = list_entry(mddev->disks.next,
5994 struct md_rdev, same_set);
5995 err = super_types[mddev->major_version]
5996 .load_super(rdev, rdev0, mddev->minor_version);
5997 if (err < 0) {
5998 printk(KERN_WARNING
5999 "md: %s has different UUID to %s\n",
6000 bdevname(rdev->bdev,b),
6001 bdevname(rdev0->bdev,b2));
6002 export_rdev(rdev);
6003 return -EINVAL;
6004 }
6005 }
6006 err = bind_rdev_to_array(rdev, mddev);
6007 if (err)
6008 export_rdev(rdev);
6009 return err;
6010 }
6011
6012 /*
6013 * add_new_disk can be used once the array is assembled
6014 * to add "hot spares". They must already have a superblock
6015 * written
6016 */
6017 if (mddev->pers) {
6018 int err;
6019 if (!mddev->pers->hot_add_disk) {
6020 printk(KERN_WARNING
6021 "%s: personality does not support diskops!\n",
6022 mdname(mddev));
6023 return -EINVAL;
6024 }
6025 if (mddev->persistent)
6026 rdev = md_import_device(dev, mddev->major_version,
6027 mddev->minor_version);
6028 else
6029 rdev = md_import_device(dev, -1, -1);
6030 if (IS_ERR(rdev)) {
6031 printk(KERN_WARNING
6032 "md: md_import_device returned %ld\n",
6033 PTR_ERR(rdev));
6034 return PTR_ERR(rdev);
6035 }
6036 /* set saved_raid_disk if appropriate */
6037 if (!mddev->persistent) {
6038 if (info->state & (1<<MD_DISK_SYNC) &&
6039 info->raid_disk < mddev->raid_disks) {
6040 rdev->raid_disk = info->raid_disk;
6041 set_bit(In_sync, &rdev->flags);
6042 clear_bit(Bitmap_sync, &rdev->flags);
6043 } else
6044 rdev->raid_disk = -1;
6045 rdev->saved_raid_disk = rdev->raid_disk;
6046 } else
6047 super_types[mddev->major_version].
6048 validate_super(mddev, rdev);
6049 if ((info->state & (1<<MD_DISK_SYNC)) &&
6050 rdev->raid_disk != info->raid_disk) {
6051 /* This was a hot-add request, but events doesn't
6052 * match, so reject it.
6053 */
6054 export_rdev(rdev);
6055 return -EINVAL;
6056 }
6057
6058 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6059 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6060 set_bit(WriteMostly, &rdev->flags);
6061 else
6062 clear_bit(WriteMostly, &rdev->flags);
6063
6064 if (info->state & (1<<MD_DISK_JOURNAL)) {
6065 struct md_rdev *rdev2;
6066 bool has_journal = false;
6067
6068 /* make sure no existing journal disk */
6069 rdev_for_each(rdev2, mddev) {
6070 if (test_bit(Journal, &rdev2->flags)) {
6071 has_journal = true;
6072 break;
6073 }
6074 }
6075 if (has_journal) {
6076 export_rdev(rdev);
6077 return -EBUSY;
6078 }
6079 set_bit(Journal, &rdev->flags);
6080 }
6081 /*
6082 * check whether the device shows up in other nodes
6083 */
6084 if (mddev_is_clustered(mddev)) {
6085 if (info->state & (1 << MD_DISK_CANDIDATE))
6086 set_bit(Candidate, &rdev->flags);
6087 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6088 /* --add initiated by this node */
6089 err = md_cluster_ops->add_new_disk(mddev, rdev);
6090 if (err) {
6091 export_rdev(rdev);
6092 return err;
6093 }
6094 }
6095 }
6096
6097 rdev->raid_disk = -1;
6098 err = bind_rdev_to_array(rdev, mddev);
6099
6100 if (err)
6101 export_rdev(rdev);
6102
6103 if (mddev_is_clustered(mddev)) {
6104 if (info->state & (1 << MD_DISK_CANDIDATE))
6105 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6106 else {
6107 if (err)
6108 md_cluster_ops->add_new_disk_cancel(mddev);
6109 else
6110 err = add_bound_rdev(rdev);
6111 }
6112
6113 } else if (!err)
6114 err = add_bound_rdev(rdev);
6115
6116 return err;
6117 }
6118
6119 /* otherwise, add_new_disk is only allowed
6120 * for major_version==0 superblocks
6121 */
6122 if (mddev->major_version != 0) {
6123 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6124 mdname(mddev));
6125 return -EINVAL;
6126 }
6127
6128 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6129 int err;
6130 rdev = md_import_device(dev, -1, 0);
6131 if (IS_ERR(rdev)) {
6132 printk(KERN_WARNING
6133 "md: error, md_import_device() returned %ld\n",
6134 PTR_ERR(rdev));
6135 return PTR_ERR(rdev);
6136 }
6137 rdev->desc_nr = info->number;
6138 if (info->raid_disk < mddev->raid_disks)
6139 rdev->raid_disk = info->raid_disk;
6140 else
6141 rdev->raid_disk = -1;
6142
6143 if (rdev->raid_disk < mddev->raid_disks)
6144 if (info->state & (1<<MD_DISK_SYNC))
6145 set_bit(In_sync, &rdev->flags);
6146
6147 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6148 set_bit(WriteMostly, &rdev->flags);
6149
6150 if (!mddev->persistent) {
6151 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6152 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6153 } else
6154 rdev->sb_start = calc_dev_sboffset(rdev);
6155 rdev->sectors = rdev->sb_start;
6156
6157 err = bind_rdev_to_array(rdev, mddev);
6158 if (err) {
6159 export_rdev(rdev);
6160 return err;
6161 }
6162 }
6163
6164 return 0;
6165 }
6166
6167 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6168 {
6169 char b[BDEVNAME_SIZE];
6170 struct md_rdev *rdev;
6171
6172 rdev = find_rdev(mddev, dev);
6173 if (!rdev)
6174 return -ENXIO;
6175
6176 if (rdev->raid_disk < 0)
6177 goto kick_rdev;
6178
6179 clear_bit(Blocked, &rdev->flags);
6180 remove_and_add_spares(mddev, rdev);
6181
6182 if (rdev->raid_disk >= 0)
6183 goto busy;
6184
6185 kick_rdev:
6186 if (mddev_is_clustered(mddev))
6187 md_cluster_ops->remove_disk(mddev, rdev);
6188
6189 md_kick_rdev_from_array(rdev);
6190 md_update_sb(mddev, 1);
6191 md_new_event(mddev);
6192
6193 return 0;
6194 busy:
6195 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6196 bdevname(rdev->bdev,b), mdname(mddev));
6197 return -EBUSY;
6198 }
6199
6200 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6201 {
6202 char b[BDEVNAME_SIZE];
6203 int err;
6204 struct md_rdev *rdev;
6205
6206 if (!mddev->pers)
6207 return -ENODEV;
6208
6209 if (mddev->major_version != 0) {
6210 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6211 " version-0 superblocks.\n",
6212 mdname(mddev));
6213 return -EINVAL;
6214 }
6215 if (!mddev->pers->hot_add_disk) {
6216 printk(KERN_WARNING
6217 "%s: personality does not support diskops!\n",
6218 mdname(mddev));
6219 return -EINVAL;
6220 }
6221
6222 rdev = md_import_device(dev, -1, 0);
6223 if (IS_ERR(rdev)) {
6224 printk(KERN_WARNING
6225 "md: error, md_import_device() returned %ld\n",
6226 PTR_ERR(rdev));
6227 return -EINVAL;
6228 }
6229
6230 if (mddev->persistent)
6231 rdev->sb_start = calc_dev_sboffset(rdev);
6232 else
6233 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6234
6235 rdev->sectors = rdev->sb_start;
6236
6237 if (test_bit(Faulty, &rdev->flags)) {
6238 printk(KERN_WARNING
6239 "md: can not hot-add faulty %s disk to %s!\n",
6240 bdevname(rdev->bdev,b), mdname(mddev));
6241 err = -EINVAL;
6242 goto abort_export;
6243 }
6244
6245 clear_bit(In_sync, &rdev->flags);
6246 rdev->desc_nr = -1;
6247 rdev->saved_raid_disk = -1;
6248 err = bind_rdev_to_array(rdev, mddev);
6249 if (err)
6250 goto abort_export;
6251
6252 /*
6253 * The rest should better be atomic, we can have disk failures
6254 * noticed in interrupt contexts ...
6255 */
6256
6257 rdev->raid_disk = -1;
6258
6259 md_update_sb(mddev, 1);
6260 /*
6261 * Kick recovery, maybe this spare has to be added to the
6262 * array immediately.
6263 */
6264 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6265 md_wakeup_thread(mddev->thread);
6266 md_new_event(mddev);
6267 return 0;
6268
6269 abort_export:
6270 export_rdev(rdev);
6271 return err;
6272 }
6273
6274 static int set_bitmap_file(struct mddev *mddev, int fd)
6275 {
6276 int err = 0;
6277
6278 if (mddev->pers) {
6279 if (!mddev->pers->quiesce || !mddev->thread)
6280 return -EBUSY;
6281 if (mddev->recovery || mddev->sync_thread)
6282 return -EBUSY;
6283 /* we should be able to change the bitmap.. */
6284 }
6285
6286 if (fd >= 0) {
6287 struct inode *inode;
6288 struct file *f;
6289
6290 if (mddev->bitmap || mddev->bitmap_info.file)
6291 return -EEXIST; /* cannot add when bitmap is present */
6292 f = fget(fd);
6293
6294 if (f == NULL) {
6295 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6296 mdname(mddev));
6297 return -EBADF;
6298 }
6299
6300 inode = f->f_mapping->host;
6301 if (!S_ISREG(inode->i_mode)) {
6302 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6303 mdname(mddev));
6304 err = -EBADF;
6305 } else if (!(f->f_mode & FMODE_WRITE)) {
6306 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6307 mdname(mddev));
6308 err = -EBADF;
6309 } else if (atomic_read(&inode->i_writecount) != 1) {
6310 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6311 mdname(mddev));
6312 err = -EBUSY;
6313 }
6314 if (err) {
6315 fput(f);
6316 return err;
6317 }
6318 mddev->bitmap_info.file = f;
6319 mddev->bitmap_info.offset = 0; /* file overrides offset */
6320 } else if (mddev->bitmap == NULL)
6321 return -ENOENT; /* cannot remove what isn't there */
6322 err = 0;
6323 if (mddev->pers) {
6324 mddev->pers->quiesce(mddev, 1);
6325 if (fd >= 0) {
6326 struct bitmap *bitmap;
6327
6328 bitmap = bitmap_create(mddev, -1);
6329 if (!IS_ERR(bitmap)) {
6330 mddev->bitmap = bitmap;
6331 err = bitmap_load(mddev);
6332 } else
6333 err = PTR_ERR(bitmap);
6334 }
6335 if (fd < 0 || err) {
6336 bitmap_destroy(mddev);
6337 fd = -1; /* make sure to put the file */
6338 }
6339 mddev->pers->quiesce(mddev, 0);
6340 }
6341 if (fd < 0) {
6342 struct file *f = mddev->bitmap_info.file;
6343 if (f) {
6344 spin_lock(&mddev->lock);
6345 mddev->bitmap_info.file = NULL;
6346 spin_unlock(&mddev->lock);
6347 fput(f);
6348 }
6349 }
6350
6351 return err;
6352 }
6353
6354 /*
6355 * set_array_info is used two different ways
6356 * The original usage is when creating a new array.
6357 * In this usage, raid_disks is > 0 and it together with
6358 * level, size, not_persistent,layout,chunksize determine the
6359 * shape of the array.
6360 * This will always create an array with a type-0.90.0 superblock.
6361 * The newer usage is when assembling an array.
6362 * In this case raid_disks will be 0, and the major_version field is
6363 * use to determine which style super-blocks are to be found on the devices.
6364 * The minor and patch _version numbers are also kept incase the
6365 * super_block handler wishes to interpret them.
6366 */
6367 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6368 {
6369
6370 if (info->raid_disks == 0) {
6371 /* just setting version number for superblock loading */
6372 if (info->major_version < 0 ||
6373 info->major_version >= ARRAY_SIZE(super_types) ||
6374 super_types[info->major_version].name == NULL) {
6375 /* maybe try to auto-load a module? */
6376 printk(KERN_INFO
6377 "md: superblock version %d not known\n",
6378 info->major_version);
6379 return -EINVAL;
6380 }
6381 mddev->major_version = info->major_version;
6382 mddev->minor_version = info->minor_version;
6383 mddev->patch_version = info->patch_version;
6384 mddev->persistent = !info->not_persistent;
6385 /* ensure mddev_put doesn't delete this now that there
6386 * is some minimal configuration.
6387 */
6388 mddev->ctime = ktime_get_real_seconds();
6389 return 0;
6390 }
6391 mddev->major_version = MD_MAJOR_VERSION;
6392 mddev->minor_version = MD_MINOR_VERSION;
6393 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6394 mddev->ctime = ktime_get_real_seconds();
6395
6396 mddev->level = info->level;
6397 mddev->clevel[0] = 0;
6398 mddev->dev_sectors = 2 * (sector_t)info->size;
6399 mddev->raid_disks = info->raid_disks;
6400 /* don't set md_minor, it is determined by which /dev/md* was
6401 * openned
6402 */
6403 if (info->state & (1<<MD_SB_CLEAN))
6404 mddev->recovery_cp = MaxSector;
6405 else
6406 mddev->recovery_cp = 0;
6407 mddev->persistent = ! info->not_persistent;
6408 mddev->external = 0;
6409
6410 mddev->layout = info->layout;
6411 mddev->chunk_sectors = info->chunk_size >> 9;
6412
6413 mddev->max_disks = MD_SB_DISKS;
6414
6415 if (mddev->persistent)
6416 mddev->flags = 0;
6417 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6418
6419 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6420 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6421 mddev->bitmap_info.offset = 0;
6422
6423 mddev->reshape_position = MaxSector;
6424
6425 /*
6426 * Generate a 128 bit UUID
6427 */
6428 get_random_bytes(mddev->uuid, 16);
6429
6430 mddev->new_level = mddev->level;
6431 mddev->new_chunk_sectors = mddev->chunk_sectors;
6432 mddev->new_layout = mddev->layout;
6433 mddev->delta_disks = 0;
6434 mddev->reshape_backwards = 0;
6435
6436 return 0;
6437 }
6438
6439 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6440 {
6441 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6442
6443 if (mddev->external_size)
6444 return;
6445
6446 mddev->array_sectors = array_sectors;
6447 }
6448 EXPORT_SYMBOL(md_set_array_sectors);
6449
6450 static int update_size(struct mddev *mddev, sector_t num_sectors)
6451 {
6452 struct md_rdev *rdev;
6453 int rv;
6454 int fit = (num_sectors == 0);
6455
6456 /* cluster raid doesn't support update size */
6457 if (mddev_is_clustered(mddev))
6458 return -EINVAL;
6459
6460 if (mddev->pers->resize == NULL)
6461 return -EINVAL;
6462 /* The "num_sectors" is the number of sectors of each device that
6463 * is used. This can only make sense for arrays with redundancy.
6464 * linear and raid0 always use whatever space is available. We can only
6465 * consider changing this number if no resync or reconstruction is
6466 * happening, and if the new size is acceptable. It must fit before the
6467 * sb_start or, if that is <data_offset, it must fit before the size
6468 * of each device. If num_sectors is zero, we find the largest size
6469 * that fits.
6470 */
6471 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6472 mddev->sync_thread)
6473 return -EBUSY;
6474 if (mddev->ro)
6475 return -EROFS;
6476
6477 rdev_for_each(rdev, mddev) {
6478 sector_t avail = rdev->sectors;
6479
6480 if (fit && (num_sectors == 0 || num_sectors > avail))
6481 num_sectors = avail;
6482 if (avail < num_sectors)
6483 return -ENOSPC;
6484 }
6485 rv = mddev->pers->resize(mddev, num_sectors);
6486 if (!rv)
6487 revalidate_disk(mddev->gendisk);
6488 return rv;
6489 }
6490
6491 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6492 {
6493 int rv;
6494 struct md_rdev *rdev;
6495 /* change the number of raid disks */
6496 if (mddev->pers->check_reshape == NULL)
6497 return -EINVAL;
6498 if (mddev->ro)
6499 return -EROFS;
6500 if (raid_disks <= 0 ||
6501 (mddev->max_disks && raid_disks >= mddev->max_disks))
6502 return -EINVAL;
6503 if (mddev->sync_thread ||
6504 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6505 mddev->reshape_position != MaxSector)
6506 return -EBUSY;
6507
6508 rdev_for_each(rdev, mddev) {
6509 if (mddev->raid_disks < raid_disks &&
6510 rdev->data_offset < rdev->new_data_offset)
6511 return -EINVAL;
6512 if (mddev->raid_disks > raid_disks &&
6513 rdev->data_offset > rdev->new_data_offset)
6514 return -EINVAL;
6515 }
6516
6517 mddev->delta_disks = raid_disks - mddev->raid_disks;
6518 if (mddev->delta_disks < 0)
6519 mddev->reshape_backwards = 1;
6520 else if (mddev->delta_disks > 0)
6521 mddev->reshape_backwards = 0;
6522
6523 rv = mddev->pers->check_reshape(mddev);
6524 if (rv < 0) {
6525 mddev->delta_disks = 0;
6526 mddev->reshape_backwards = 0;
6527 }
6528 return rv;
6529 }
6530
6531 /*
6532 * update_array_info is used to change the configuration of an
6533 * on-line array.
6534 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6535 * fields in the info are checked against the array.
6536 * Any differences that cannot be handled will cause an error.
6537 * Normally, only one change can be managed at a time.
6538 */
6539 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6540 {
6541 int rv = 0;
6542 int cnt = 0;
6543 int state = 0;
6544
6545 /* calculate expected state,ignoring low bits */
6546 if (mddev->bitmap && mddev->bitmap_info.offset)
6547 state |= (1 << MD_SB_BITMAP_PRESENT);
6548
6549 if (mddev->major_version != info->major_version ||
6550 mddev->minor_version != info->minor_version ||
6551 /* mddev->patch_version != info->patch_version || */
6552 mddev->ctime != info->ctime ||
6553 mddev->level != info->level ||
6554 /* mddev->layout != info->layout || */
6555 mddev->persistent != !info->not_persistent ||
6556 mddev->chunk_sectors != info->chunk_size >> 9 ||
6557 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6558 ((state^info->state) & 0xfffffe00)
6559 )
6560 return -EINVAL;
6561 /* Check there is only one change */
6562 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6563 cnt++;
6564 if (mddev->raid_disks != info->raid_disks)
6565 cnt++;
6566 if (mddev->layout != info->layout)
6567 cnt++;
6568 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6569 cnt++;
6570 if (cnt == 0)
6571 return 0;
6572 if (cnt > 1)
6573 return -EINVAL;
6574
6575 if (mddev->layout != info->layout) {
6576 /* Change layout
6577 * we don't need to do anything at the md level, the
6578 * personality will take care of it all.
6579 */
6580 if (mddev->pers->check_reshape == NULL)
6581 return -EINVAL;
6582 else {
6583 mddev->new_layout = info->layout;
6584 rv = mddev->pers->check_reshape(mddev);
6585 if (rv)
6586 mddev->new_layout = mddev->layout;
6587 return rv;
6588 }
6589 }
6590 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6591 rv = update_size(mddev, (sector_t)info->size * 2);
6592
6593 if (mddev->raid_disks != info->raid_disks)
6594 rv = update_raid_disks(mddev, info->raid_disks);
6595
6596 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6597 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6598 rv = -EINVAL;
6599 goto err;
6600 }
6601 if (mddev->recovery || mddev->sync_thread) {
6602 rv = -EBUSY;
6603 goto err;
6604 }
6605 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6606 struct bitmap *bitmap;
6607 /* add the bitmap */
6608 if (mddev->bitmap) {
6609 rv = -EEXIST;
6610 goto err;
6611 }
6612 if (mddev->bitmap_info.default_offset == 0) {
6613 rv = -EINVAL;
6614 goto err;
6615 }
6616 mddev->bitmap_info.offset =
6617 mddev->bitmap_info.default_offset;
6618 mddev->bitmap_info.space =
6619 mddev->bitmap_info.default_space;
6620 mddev->pers->quiesce(mddev, 1);
6621 bitmap = bitmap_create(mddev, -1);
6622 if (!IS_ERR(bitmap)) {
6623 mddev->bitmap = bitmap;
6624 rv = bitmap_load(mddev);
6625 } else
6626 rv = PTR_ERR(bitmap);
6627 if (rv)
6628 bitmap_destroy(mddev);
6629 mddev->pers->quiesce(mddev, 0);
6630 } else {
6631 /* remove the bitmap */
6632 if (!mddev->bitmap) {
6633 rv = -ENOENT;
6634 goto err;
6635 }
6636 if (mddev->bitmap->storage.file) {
6637 rv = -EINVAL;
6638 goto err;
6639 }
6640 if (mddev->bitmap_info.nodes) {
6641 /* hold PW on all the bitmap lock */
6642 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6643 printk("md: can't change bitmap to none since the"
6644 " array is in use by more than one node\n");
6645 rv = -EPERM;
6646 md_cluster_ops->unlock_all_bitmaps(mddev);
6647 goto err;
6648 }
6649
6650 mddev->bitmap_info.nodes = 0;
6651 md_cluster_ops->leave(mddev);
6652 }
6653 mddev->pers->quiesce(mddev, 1);
6654 bitmap_destroy(mddev);
6655 mddev->pers->quiesce(mddev, 0);
6656 mddev->bitmap_info.offset = 0;
6657 }
6658 }
6659 md_update_sb(mddev, 1);
6660 return rv;
6661 err:
6662 return rv;
6663 }
6664
6665 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6666 {
6667 struct md_rdev *rdev;
6668 int err = 0;
6669
6670 if (mddev->pers == NULL)
6671 return -ENODEV;
6672
6673 rcu_read_lock();
6674 rdev = find_rdev_rcu(mddev, dev);
6675 if (!rdev)
6676 err = -ENODEV;
6677 else {
6678 md_error(mddev, rdev);
6679 if (!test_bit(Faulty, &rdev->flags))
6680 err = -EBUSY;
6681 }
6682 rcu_read_unlock();
6683 return err;
6684 }
6685
6686 /*
6687 * We have a problem here : there is no easy way to give a CHS
6688 * virtual geometry. We currently pretend that we have a 2 heads
6689 * 4 sectors (with a BIG number of cylinders...). This drives
6690 * dosfs just mad... ;-)
6691 */
6692 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6693 {
6694 struct mddev *mddev = bdev->bd_disk->private_data;
6695
6696 geo->heads = 2;
6697 geo->sectors = 4;
6698 geo->cylinders = mddev->array_sectors / 8;
6699 return 0;
6700 }
6701
6702 static inline bool md_ioctl_valid(unsigned int cmd)
6703 {
6704 switch (cmd) {
6705 case ADD_NEW_DISK:
6706 case BLKROSET:
6707 case GET_ARRAY_INFO:
6708 case GET_BITMAP_FILE:
6709 case GET_DISK_INFO:
6710 case HOT_ADD_DISK:
6711 case HOT_REMOVE_DISK:
6712 case RAID_AUTORUN:
6713 case RAID_VERSION:
6714 case RESTART_ARRAY_RW:
6715 case RUN_ARRAY:
6716 case SET_ARRAY_INFO:
6717 case SET_BITMAP_FILE:
6718 case SET_DISK_FAULTY:
6719 case STOP_ARRAY:
6720 case STOP_ARRAY_RO:
6721 case CLUSTERED_DISK_NACK:
6722 return true;
6723 default:
6724 return false;
6725 }
6726 }
6727
6728 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6729 unsigned int cmd, unsigned long arg)
6730 {
6731 int err = 0;
6732 void __user *argp = (void __user *)arg;
6733 struct mddev *mddev = NULL;
6734 int ro;
6735
6736 if (!md_ioctl_valid(cmd))
6737 return -ENOTTY;
6738
6739 switch (cmd) {
6740 case RAID_VERSION:
6741 case GET_ARRAY_INFO:
6742 case GET_DISK_INFO:
6743 break;
6744 default:
6745 if (!capable(CAP_SYS_ADMIN))
6746 return -EACCES;
6747 }
6748
6749 /*
6750 * Commands dealing with the RAID driver but not any
6751 * particular array:
6752 */
6753 switch (cmd) {
6754 case RAID_VERSION:
6755 err = get_version(argp);
6756 goto out;
6757
6758 #ifndef MODULE
6759 case RAID_AUTORUN:
6760 err = 0;
6761 autostart_arrays(arg);
6762 goto out;
6763 #endif
6764 default:;
6765 }
6766
6767 /*
6768 * Commands creating/starting a new array:
6769 */
6770
6771 mddev = bdev->bd_disk->private_data;
6772
6773 if (!mddev) {
6774 BUG();
6775 goto out;
6776 }
6777
6778 /* Some actions do not requires the mutex */
6779 switch (cmd) {
6780 case GET_ARRAY_INFO:
6781 if (!mddev->raid_disks && !mddev->external)
6782 err = -ENODEV;
6783 else
6784 err = get_array_info(mddev, argp);
6785 goto out;
6786
6787 case GET_DISK_INFO:
6788 if (!mddev->raid_disks && !mddev->external)
6789 err = -ENODEV;
6790 else
6791 err = get_disk_info(mddev, argp);
6792 goto out;
6793
6794 case SET_DISK_FAULTY:
6795 err = set_disk_faulty(mddev, new_decode_dev(arg));
6796 goto out;
6797
6798 case GET_BITMAP_FILE:
6799 err = get_bitmap_file(mddev, argp);
6800 goto out;
6801
6802 }
6803
6804 if (cmd == ADD_NEW_DISK)
6805 /* need to ensure md_delayed_delete() has completed */
6806 flush_workqueue(md_misc_wq);
6807
6808 if (cmd == HOT_REMOVE_DISK)
6809 /* need to ensure recovery thread has run */
6810 wait_event_interruptible_timeout(mddev->sb_wait,
6811 !test_bit(MD_RECOVERY_NEEDED,
6812 &mddev->flags),
6813 msecs_to_jiffies(5000));
6814 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6815 /* Need to flush page cache, and ensure no-one else opens
6816 * and writes
6817 */
6818 mutex_lock(&mddev->open_mutex);
6819 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6820 mutex_unlock(&mddev->open_mutex);
6821 err = -EBUSY;
6822 goto out;
6823 }
6824 set_bit(MD_STILL_CLOSED, &mddev->flags);
6825 mutex_unlock(&mddev->open_mutex);
6826 sync_blockdev(bdev);
6827 }
6828 err = mddev_lock(mddev);
6829 if (err) {
6830 printk(KERN_INFO
6831 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6832 err, cmd);
6833 goto out;
6834 }
6835
6836 if (cmd == SET_ARRAY_INFO) {
6837 mdu_array_info_t info;
6838 if (!arg)
6839 memset(&info, 0, sizeof(info));
6840 else if (copy_from_user(&info, argp, sizeof(info))) {
6841 err = -EFAULT;
6842 goto unlock;
6843 }
6844 if (mddev->pers) {
6845 err = update_array_info(mddev, &info);
6846 if (err) {
6847 printk(KERN_WARNING "md: couldn't update"
6848 " array info. %d\n", err);
6849 goto unlock;
6850 }
6851 goto unlock;
6852 }
6853 if (!list_empty(&mddev->disks)) {
6854 printk(KERN_WARNING
6855 "md: array %s already has disks!\n",
6856 mdname(mddev));
6857 err = -EBUSY;
6858 goto unlock;
6859 }
6860 if (mddev->raid_disks) {
6861 printk(KERN_WARNING
6862 "md: array %s already initialised!\n",
6863 mdname(mddev));
6864 err = -EBUSY;
6865 goto unlock;
6866 }
6867 err = set_array_info(mddev, &info);
6868 if (err) {
6869 printk(KERN_WARNING "md: couldn't set"
6870 " array info. %d\n", err);
6871 goto unlock;
6872 }
6873 goto unlock;
6874 }
6875
6876 /*
6877 * Commands querying/configuring an existing array:
6878 */
6879 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6880 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6881 if ((!mddev->raid_disks && !mddev->external)
6882 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6883 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6884 && cmd != GET_BITMAP_FILE) {
6885 err = -ENODEV;
6886 goto unlock;
6887 }
6888
6889 /*
6890 * Commands even a read-only array can execute:
6891 */
6892 switch (cmd) {
6893 case RESTART_ARRAY_RW:
6894 err = restart_array(mddev);
6895 goto unlock;
6896
6897 case STOP_ARRAY:
6898 err = do_md_stop(mddev, 0, bdev);
6899 goto unlock;
6900
6901 case STOP_ARRAY_RO:
6902 err = md_set_readonly(mddev, bdev);
6903 goto unlock;
6904
6905 case HOT_REMOVE_DISK:
6906 err = hot_remove_disk(mddev, new_decode_dev(arg));
6907 goto unlock;
6908
6909 case ADD_NEW_DISK:
6910 /* We can support ADD_NEW_DISK on read-only arrays
6911 * only if we are re-adding a preexisting device.
6912 * So require mddev->pers and MD_DISK_SYNC.
6913 */
6914 if (mddev->pers) {
6915 mdu_disk_info_t info;
6916 if (copy_from_user(&info, argp, sizeof(info)))
6917 err = -EFAULT;
6918 else if (!(info.state & (1<<MD_DISK_SYNC)))
6919 /* Need to clear read-only for this */
6920 break;
6921 else
6922 err = add_new_disk(mddev, &info);
6923 goto unlock;
6924 }
6925 break;
6926
6927 case BLKROSET:
6928 if (get_user(ro, (int __user *)(arg))) {
6929 err = -EFAULT;
6930 goto unlock;
6931 }
6932 err = -EINVAL;
6933
6934 /* if the bdev is going readonly the value of mddev->ro
6935 * does not matter, no writes are coming
6936 */
6937 if (ro)
6938 goto unlock;
6939
6940 /* are we are already prepared for writes? */
6941 if (mddev->ro != 1)
6942 goto unlock;
6943
6944 /* transitioning to readauto need only happen for
6945 * arrays that call md_write_start
6946 */
6947 if (mddev->pers) {
6948 err = restart_array(mddev);
6949 if (err == 0) {
6950 mddev->ro = 2;
6951 set_disk_ro(mddev->gendisk, 0);
6952 }
6953 }
6954 goto unlock;
6955 }
6956
6957 /*
6958 * The remaining ioctls are changing the state of the
6959 * superblock, so we do not allow them on read-only arrays.
6960 */
6961 if (mddev->ro && mddev->pers) {
6962 if (mddev->ro == 2) {
6963 mddev->ro = 0;
6964 sysfs_notify_dirent_safe(mddev->sysfs_state);
6965 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6966 /* mddev_unlock will wake thread */
6967 /* If a device failed while we were read-only, we
6968 * need to make sure the metadata is updated now.
6969 */
6970 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6971 mddev_unlock(mddev);
6972 wait_event(mddev->sb_wait,
6973 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6974 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6975 mddev_lock_nointr(mddev);
6976 }
6977 } else {
6978 err = -EROFS;
6979 goto unlock;
6980 }
6981 }
6982
6983 switch (cmd) {
6984 case ADD_NEW_DISK:
6985 {
6986 mdu_disk_info_t info;
6987 if (copy_from_user(&info, argp, sizeof(info)))
6988 err = -EFAULT;
6989 else
6990 err = add_new_disk(mddev, &info);
6991 goto unlock;
6992 }
6993
6994 case CLUSTERED_DISK_NACK:
6995 if (mddev_is_clustered(mddev))
6996 md_cluster_ops->new_disk_ack(mddev, false);
6997 else
6998 err = -EINVAL;
6999 goto unlock;
7000
7001 case HOT_ADD_DISK:
7002 err = hot_add_disk(mddev, new_decode_dev(arg));
7003 goto unlock;
7004
7005 case RUN_ARRAY:
7006 err = do_md_run(mddev);
7007 goto unlock;
7008
7009 case SET_BITMAP_FILE:
7010 err = set_bitmap_file(mddev, (int)arg);
7011 goto unlock;
7012
7013 default:
7014 err = -EINVAL;
7015 goto unlock;
7016 }
7017
7018 unlock:
7019 if (mddev->hold_active == UNTIL_IOCTL &&
7020 err != -EINVAL)
7021 mddev->hold_active = 0;
7022 mddev_unlock(mddev);
7023 out:
7024 return err;
7025 }
7026 #ifdef CONFIG_COMPAT
7027 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7028 unsigned int cmd, unsigned long arg)
7029 {
7030 switch (cmd) {
7031 case HOT_REMOVE_DISK:
7032 case HOT_ADD_DISK:
7033 case SET_DISK_FAULTY:
7034 case SET_BITMAP_FILE:
7035 /* These take in integer arg, do not convert */
7036 break;
7037 default:
7038 arg = (unsigned long)compat_ptr(arg);
7039 break;
7040 }
7041
7042 return md_ioctl(bdev, mode, cmd, arg);
7043 }
7044 #endif /* CONFIG_COMPAT */
7045
7046 static int md_open(struct block_device *bdev, fmode_t mode)
7047 {
7048 /*
7049 * Succeed if we can lock the mddev, which confirms that
7050 * it isn't being stopped right now.
7051 */
7052 struct mddev *mddev = mddev_find(bdev->bd_dev);
7053 int err;
7054
7055 if (!mddev)
7056 return -ENODEV;
7057
7058 if (mddev->gendisk != bdev->bd_disk) {
7059 /* we are racing with mddev_put which is discarding this
7060 * bd_disk.
7061 */
7062 mddev_put(mddev);
7063 /* Wait until bdev->bd_disk is definitely gone */
7064 flush_workqueue(md_misc_wq);
7065 /* Then retry the open from the top */
7066 return -ERESTARTSYS;
7067 }
7068 BUG_ON(mddev != bdev->bd_disk->private_data);
7069
7070 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7071 goto out;
7072
7073 err = 0;
7074 atomic_inc(&mddev->openers);
7075 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7076 mutex_unlock(&mddev->open_mutex);
7077
7078 check_disk_change(bdev);
7079 out:
7080 return err;
7081 }
7082
7083 static void md_release(struct gendisk *disk, fmode_t mode)
7084 {
7085 struct mddev *mddev = disk->private_data;
7086
7087 BUG_ON(!mddev);
7088 atomic_dec(&mddev->openers);
7089 mddev_put(mddev);
7090 }
7091
7092 static int md_media_changed(struct gendisk *disk)
7093 {
7094 struct mddev *mddev = disk->private_data;
7095
7096 return mddev->changed;
7097 }
7098
7099 static int md_revalidate(struct gendisk *disk)
7100 {
7101 struct mddev *mddev = disk->private_data;
7102
7103 mddev->changed = 0;
7104 return 0;
7105 }
7106 static const struct block_device_operations md_fops =
7107 {
7108 .owner = THIS_MODULE,
7109 .open = md_open,
7110 .release = md_release,
7111 .ioctl = md_ioctl,
7112 #ifdef CONFIG_COMPAT
7113 .compat_ioctl = md_compat_ioctl,
7114 #endif
7115 .getgeo = md_getgeo,
7116 .media_changed = md_media_changed,
7117 .revalidate_disk= md_revalidate,
7118 };
7119
7120 static int md_thread(void *arg)
7121 {
7122 struct md_thread *thread = arg;
7123
7124 /*
7125 * md_thread is a 'system-thread', it's priority should be very
7126 * high. We avoid resource deadlocks individually in each
7127 * raid personality. (RAID5 does preallocation) We also use RR and
7128 * the very same RT priority as kswapd, thus we will never get
7129 * into a priority inversion deadlock.
7130 *
7131 * we definitely have to have equal or higher priority than
7132 * bdflush, otherwise bdflush will deadlock if there are too
7133 * many dirty RAID5 blocks.
7134 */
7135
7136 allow_signal(SIGKILL);
7137 while (!kthread_should_stop()) {
7138
7139 /* We need to wait INTERRUPTIBLE so that
7140 * we don't add to the load-average.
7141 * That means we need to be sure no signals are
7142 * pending
7143 */
7144 if (signal_pending(current))
7145 flush_signals(current);
7146
7147 wait_event_interruptible_timeout
7148 (thread->wqueue,
7149 test_bit(THREAD_WAKEUP, &thread->flags)
7150 || kthread_should_stop(),
7151 thread->timeout);
7152
7153 clear_bit(THREAD_WAKEUP, &thread->flags);
7154 if (!kthread_should_stop())
7155 thread->run(thread);
7156 }
7157
7158 return 0;
7159 }
7160
7161 void md_wakeup_thread(struct md_thread *thread)
7162 {
7163 if (thread) {
7164 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7165 set_bit(THREAD_WAKEUP, &thread->flags);
7166 wake_up(&thread->wqueue);
7167 }
7168 }
7169 EXPORT_SYMBOL(md_wakeup_thread);
7170
7171 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7172 struct mddev *mddev, const char *name)
7173 {
7174 struct md_thread *thread;
7175
7176 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7177 if (!thread)
7178 return NULL;
7179
7180 init_waitqueue_head(&thread->wqueue);
7181
7182 thread->run = run;
7183 thread->mddev = mddev;
7184 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7185 thread->tsk = kthread_run(md_thread, thread,
7186 "%s_%s",
7187 mdname(thread->mddev),
7188 name);
7189 if (IS_ERR(thread->tsk)) {
7190 kfree(thread);
7191 return NULL;
7192 }
7193 return thread;
7194 }
7195 EXPORT_SYMBOL(md_register_thread);
7196
7197 void md_unregister_thread(struct md_thread **threadp)
7198 {
7199 struct md_thread *thread = *threadp;
7200 if (!thread)
7201 return;
7202 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7203 /* Locking ensures that mddev_unlock does not wake_up a
7204 * non-existent thread
7205 */
7206 spin_lock(&pers_lock);
7207 *threadp = NULL;
7208 spin_unlock(&pers_lock);
7209
7210 kthread_stop(thread->tsk);
7211 kfree(thread);
7212 }
7213 EXPORT_SYMBOL(md_unregister_thread);
7214
7215 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7216 {
7217 if (!rdev || test_bit(Faulty, &rdev->flags))
7218 return;
7219
7220 if (!mddev->pers || !mddev->pers->error_handler)
7221 return;
7222 mddev->pers->error_handler(mddev,rdev);
7223 if (mddev->degraded)
7224 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7225 sysfs_notify_dirent_safe(rdev->sysfs_state);
7226 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7228 md_wakeup_thread(mddev->thread);
7229 if (mddev->event_work.func)
7230 queue_work(md_misc_wq, &mddev->event_work);
7231 md_new_event(mddev);
7232 }
7233 EXPORT_SYMBOL(md_error);
7234
7235 /* seq_file implementation /proc/mdstat */
7236
7237 static void status_unused(struct seq_file *seq)
7238 {
7239 int i = 0;
7240 struct md_rdev *rdev;
7241
7242 seq_printf(seq, "unused devices: ");
7243
7244 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7245 char b[BDEVNAME_SIZE];
7246 i++;
7247 seq_printf(seq, "%s ",
7248 bdevname(rdev->bdev,b));
7249 }
7250 if (!i)
7251 seq_printf(seq, "<none>");
7252
7253 seq_printf(seq, "\n");
7254 }
7255
7256 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7257 {
7258 sector_t max_sectors, resync, res;
7259 unsigned long dt, db;
7260 sector_t rt;
7261 int scale;
7262 unsigned int per_milli;
7263
7264 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7265 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7266 max_sectors = mddev->resync_max_sectors;
7267 else
7268 max_sectors = mddev->dev_sectors;
7269
7270 resync = mddev->curr_resync;
7271 if (resync <= 3) {
7272 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7273 /* Still cleaning up */
7274 resync = max_sectors;
7275 } else
7276 resync -= atomic_read(&mddev->recovery_active);
7277
7278 if (resync == 0) {
7279 if (mddev->recovery_cp < MaxSector) {
7280 seq_printf(seq, "\tresync=PENDING");
7281 return 1;
7282 }
7283 return 0;
7284 }
7285 if (resync < 3) {
7286 seq_printf(seq, "\tresync=DELAYED");
7287 return 1;
7288 }
7289
7290 WARN_ON(max_sectors == 0);
7291 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7292 * in a sector_t, and (max_sectors>>scale) will fit in a
7293 * u32, as those are the requirements for sector_div.
7294 * Thus 'scale' must be at least 10
7295 */
7296 scale = 10;
7297 if (sizeof(sector_t) > sizeof(unsigned long)) {
7298 while ( max_sectors/2 > (1ULL<<(scale+32)))
7299 scale++;
7300 }
7301 res = (resync>>scale)*1000;
7302 sector_div(res, (u32)((max_sectors>>scale)+1));
7303
7304 per_milli = res;
7305 {
7306 int i, x = per_milli/50, y = 20-x;
7307 seq_printf(seq, "[");
7308 for (i = 0; i < x; i++)
7309 seq_printf(seq, "=");
7310 seq_printf(seq, ">");
7311 for (i = 0; i < y; i++)
7312 seq_printf(seq, ".");
7313 seq_printf(seq, "] ");
7314 }
7315 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7316 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7317 "reshape" :
7318 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7319 "check" :
7320 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7321 "resync" : "recovery"))),
7322 per_milli/10, per_milli % 10,
7323 (unsigned long long) resync/2,
7324 (unsigned long long) max_sectors/2);
7325
7326 /*
7327 * dt: time from mark until now
7328 * db: blocks written from mark until now
7329 * rt: remaining time
7330 *
7331 * rt is a sector_t, so could be 32bit or 64bit.
7332 * So we divide before multiply in case it is 32bit and close
7333 * to the limit.
7334 * We scale the divisor (db) by 32 to avoid losing precision
7335 * near the end of resync when the number of remaining sectors
7336 * is close to 'db'.
7337 * We then divide rt by 32 after multiplying by db to compensate.
7338 * The '+1' avoids division by zero if db is very small.
7339 */
7340 dt = ((jiffies - mddev->resync_mark) / HZ);
7341 if (!dt) dt++;
7342 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7343 - mddev->resync_mark_cnt;
7344
7345 rt = max_sectors - resync; /* number of remaining sectors */
7346 sector_div(rt, db/32+1);
7347 rt *= dt;
7348 rt >>= 5;
7349
7350 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7351 ((unsigned long)rt % 60)/6);
7352
7353 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7354 return 1;
7355 }
7356
7357 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7358 {
7359 struct list_head *tmp;
7360 loff_t l = *pos;
7361 struct mddev *mddev;
7362
7363 if (l >= 0x10000)
7364 return NULL;
7365 if (!l--)
7366 /* header */
7367 return (void*)1;
7368
7369 spin_lock(&all_mddevs_lock);
7370 list_for_each(tmp,&all_mddevs)
7371 if (!l--) {
7372 mddev = list_entry(tmp, struct mddev, all_mddevs);
7373 mddev_get(mddev);
7374 spin_unlock(&all_mddevs_lock);
7375 return mddev;
7376 }
7377 spin_unlock(&all_mddevs_lock);
7378 if (!l--)
7379 return (void*)2;/* tail */
7380 return NULL;
7381 }
7382
7383 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7384 {
7385 struct list_head *tmp;
7386 struct mddev *next_mddev, *mddev = v;
7387
7388 ++*pos;
7389 if (v == (void*)2)
7390 return NULL;
7391
7392 spin_lock(&all_mddevs_lock);
7393 if (v == (void*)1)
7394 tmp = all_mddevs.next;
7395 else
7396 tmp = mddev->all_mddevs.next;
7397 if (tmp != &all_mddevs)
7398 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7399 else {
7400 next_mddev = (void*)2;
7401 *pos = 0x10000;
7402 }
7403 spin_unlock(&all_mddevs_lock);
7404
7405 if (v != (void*)1)
7406 mddev_put(mddev);
7407 return next_mddev;
7408
7409 }
7410
7411 static void md_seq_stop(struct seq_file *seq, void *v)
7412 {
7413 struct mddev *mddev = v;
7414
7415 if (mddev && v != (void*)1 && v != (void*)2)
7416 mddev_put(mddev);
7417 }
7418
7419 static int md_seq_show(struct seq_file *seq, void *v)
7420 {
7421 struct mddev *mddev = v;
7422 sector_t sectors;
7423 struct md_rdev *rdev;
7424
7425 if (v == (void*)1) {
7426 struct md_personality *pers;
7427 seq_printf(seq, "Personalities : ");
7428 spin_lock(&pers_lock);
7429 list_for_each_entry(pers, &pers_list, list)
7430 seq_printf(seq, "[%s] ", pers->name);
7431
7432 spin_unlock(&pers_lock);
7433 seq_printf(seq, "\n");
7434 seq->poll_event = atomic_read(&md_event_count);
7435 return 0;
7436 }
7437 if (v == (void*)2) {
7438 status_unused(seq);
7439 return 0;
7440 }
7441
7442 spin_lock(&mddev->lock);
7443 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7444 seq_printf(seq, "%s : %sactive", mdname(mddev),
7445 mddev->pers ? "" : "in");
7446 if (mddev->pers) {
7447 if (mddev->ro==1)
7448 seq_printf(seq, " (read-only)");
7449 if (mddev->ro==2)
7450 seq_printf(seq, " (auto-read-only)");
7451 seq_printf(seq, " %s", mddev->pers->name);
7452 }
7453
7454 sectors = 0;
7455 rcu_read_lock();
7456 rdev_for_each_rcu(rdev, mddev) {
7457 char b[BDEVNAME_SIZE];
7458 seq_printf(seq, " %s[%d]",
7459 bdevname(rdev->bdev,b), rdev->desc_nr);
7460 if (test_bit(WriteMostly, &rdev->flags))
7461 seq_printf(seq, "(W)");
7462 if (test_bit(Journal, &rdev->flags))
7463 seq_printf(seq, "(J)");
7464 if (test_bit(Faulty, &rdev->flags)) {
7465 seq_printf(seq, "(F)");
7466 continue;
7467 }
7468 if (rdev->raid_disk < 0)
7469 seq_printf(seq, "(S)"); /* spare */
7470 if (test_bit(Replacement, &rdev->flags))
7471 seq_printf(seq, "(R)");
7472 sectors += rdev->sectors;
7473 }
7474 rcu_read_unlock();
7475
7476 if (!list_empty(&mddev->disks)) {
7477 if (mddev->pers)
7478 seq_printf(seq, "\n %llu blocks",
7479 (unsigned long long)
7480 mddev->array_sectors / 2);
7481 else
7482 seq_printf(seq, "\n %llu blocks",
7483 (unsigned long long)sectors / 2);
7484 }
7485 if (mddev->persistent) {
7486 if (mddev->major_version != 0 ||
7487 mddev->minor_version != 90) {
7488 seq_printf(seq," super %d.%d",
7489 mddev->major_version,
7490 mddev->minor_version);
7491 }
7492 } else if (mddev->external)
7493 seq_printf(seq, " super external:%s",
7494 mddev->metadata_type);
7495 else
7496 seq_printf(seq, " super non-persistent");
7497
7498 if (mddev->pers) {
7499 mddev->pers->status(seq, mddev);
7500 seq_printf(seq, "\n ");
7501 if (mddev->pers->sync_request) {
7502 if (status_resync(seq, mddev))
7503 seq_printf(seq, "\n ");
7504 }
7505 } else
7506 seq_printf(seq, "\n ");
7507
7508 bitmap_status(seq, mddev->bitmap);
7509
7510 seq_printf(seq, "\n");
7511 }
7512 spin_unlock(&mddev->lock);
7513
7514 return 0;
7515 }
7516
7517 static const struct seq_operations md_seq_ops = {
7518 .start = md_seq_start,
7519 .next = md_seq_next,
7520 .stop = md_seq_stop,
7521 .show = md_seq_show,
7522 };
7523
7524 static int md_seq_open(struct inode *inode, struct file *file)
7525 {
7526 struct seq_file *seq;
7527 int error;
7528
7529 error = seq_open(file, &md_seq_ops);
7530 if (error)
7531 return error;
7532
7533 seq = file->private_data;
7534 seq->poll_event = atomic_read(&md_event_count);
7535 return error;
7536 }
7537
7538 static int md_unloading;
7539 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7540 {
7541 struct seq_file *seq = filp->private_data;
7542 int mask;
7543
7544 if (md_unloading)
7545 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7546 poll_wait(filp, &md_event_waiters, wait);
7547
7548 /* always allow read */
7549 mask = POLLIN | POLLRDNORM;
7550
7551 if (seq->poll_event != atomic_read(&md_event_count))
7552 mask |= POLLERR | POLLPRI;
7553 return mask;
7554 }
7555
7556 static const struct file_operations md_seq_fops = {
7557 .owner = THIS_MODULE,
7558 .open = md_seq_open,
7559 .read = seq_read,
7560 .llseek = seq_lseek,
7561 .release = seq_release_private,
7562 .poll = mdstat_poll,
7563 };
7564
7565 int register_md_personality(struct md_personality *p)
7566 {
7567 printk(KERN_INFO "md: %s personality registered for level %d\n",
7568 p->name, p->level);
7569 spin_lock(&pers_lock);
7570 list_add_tail(&p->list, &pers_list);
7571 spin_unlock(&pers_lock);
7572 return 0;
7573 }
7574 EXPORT_SYMBOL(register_md_personality);
7575
7576 int unregister_md_personality(struct md_personality *p)
7577 {
7578 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7579 spin_lock(&pers_lock);
7580 list_del_init(&p->list);
7581 spin_unlock(&pers_lock);
7582 return 0;
7583 }
7584 EXPORT_SYMBOL(unregister_md_personality);
7585
7586 int register_md_cluster_operations(struct md_cluster_operations *ops,
7587 struct module *module)
7588 {
7589 int ret = 0;
7590 spin_lock(&pers_lock);
7591 if (md_cluster_ops != NULL)
7592 ret = -EALREADY;
7593 else {
7594 md_cluster_ops = ops;
7595 md_cluster_mod = module;
7596 }
7597 spin_unlock(&pers_lock);
7598 return ret;
7599 }
7600 EXPORT_SYMBOL(register_md_cluster_operations);
7601
7602 int unregister_md_cluster_operations(void)
7603 {
7604 spin_lock(&pers_lock);
7605 md_cluster_ops = NULL;
7606 spin_unlock(&pers_lock);
7607 return 0;
7608 }
7609 EXPORT_SYMBOL(unregister_md_cluster_operations);
7610
7611 int md_setup_cluster(struct mddev *mddev, int nodes)
7612 {
7613 if (!md_cluster_ops)
7614 request_module("md-cluster");
7615 spin_lock(&pers_lock);
7616 /* ensure module won't be unloaded */
7617 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7618 pr_err("can't find md-cluster module or get it's reference.\n");
7619 spin_unlock(&pers_lock);
7620 return -ENOENT;
7621 }
7622 spin_unlock(&pers_lock);
7623
7624 return md_cluster_ops->join(mddev, nodes);
7625 }
7626
7627 void md_cluster_stop(struct mddev *mddev)
7628 {
7629 if (!md_cluster_ops)
7630 return;
7631 md_cluster_ops->leave(mddev);
7632 module_put(md_cluster_mod);
7633 }
7634
7635 static int is_mddev_idle(struct mddev *mddev, int init)
7636 {
7637 struct md_rdev *rdev;
7638 int idle;
7639 int curr_events;
7640
7641 idle = 1;
7642 rcu_read_lock();
7643 rdev_for_each_rcu(rdev, mddev) {
7644 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7645 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7646 (int)part_stat_read(&disk->part0, sectors[1]) -
7647 atomic_read(&disk->sync_io);
7648 /* sync IO will cause sync_io to increase before the disk_stats
7649 * as sync_io is counted when a request starts, and
7650 * disk_stats is counted when it completes.
7651 * So resync activity will cause curr_events to be smaller than
7652 * when there was no such activity.
7653 * non-sync IO will cause disk_stat to increase without
7654 * increasing sync_io so curr_events will (eventually)
7655 * be larger than it was before. Once it becomes
7656 * substantially larger, the test below will cause
7657 * the array to appear non-idle, and resync will slow
7658 * down.
7659 * If there is a lot of outstanding resync activity when
7660 * we set last_event to curr_events, then all that activity
7661 * completing might cause the array to appear non-idle
7662 * and resync will be slowed down even though there might
7663 * not have been non-resync activity. This will only
7664 * happen once though. 'last_events' will soon reflect
7665 * the state where there is little or no outstanding
7666 * resync requests, and further resync activity will
7667 * always make curr_events less than last_events.
7668 *
7669 */
7670 if (init || curr_events - rdev->last_events > 64) {
7671 rdev->last_events = curr_events;
7672 idle = 0;
7673 }
7674 }
7675 rcu_read_unlock();
7676 return idle;
7677 }
7678
7679 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7680 {
7681 /* another "blocks" (512byte) blocks have been synced */
7682 atomic_sub(blocks, &mddev->recovery_active);
7683 wake_up(&mddev->recovery_wait);
7684 if (!ok) {
7685 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7686 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7687 md_wakeup_thread(mddev->thread);
7688 // stop recovery, signal do_sync ....
7689 }
7690 }
7691 EXPORT_SYMBOL(md_done_sync);
7692
7693 /* md_write_start(mddev, bi)
7694 * If we need to update some array metadata (e.g. 'active' flag
7695 * in superblock) before writing, schedule a superblock update
7696 * and wait for it to complete.
7697 */
7698 void md_write_start(struct mddev *mddev, struct bio *bi)
7699 {
7700 int did_change = 0;
7701 if (bio_data_dir(bi) != WRITE)
7702 return;
7703
7704 BUG_ON(mddev->ro == 1);
7705 if (mddev->ro == 2) {
7706 /* need to switch to read/write */
7707 mddev->ro = 0;
7708 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7709 md_wakeup_thread(mddev->thread);
7710 md_wakeup_thread(mddev->sync_thread);
7711 did_change = 1;
7712 }
7713 atomic_inc(&mddev->writes_pending);
7714 if (mddev->safemode == 1)
7715 mddev->safemode = 0;
7716 if (mddev->in_sync) {
7717 spin_lock(&mddev->lock);
7718 if (mddev->in_sync) {
7719 mddev->in_sync = 0;
7720 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7721 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7722 md_wakeup_thread(mddev->thread);
7723 did_change = 1;
7724 }
7725 spin_unlock(&mddev->lock);
7726 }
7727 if (did_change)
7728 sysfs_notify_dirent_safe(mddev->sysfs_state);
7729 wait_event(mddev->sb_wait,
7730 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7731 }
7732 EXPORT_SYMBOL(md_write_start);
7733
7734 void md_write_end(struct mddev *mddev)
7735 {
7736 if (atomic_dec_and_test(&mddev->writes_pending)) {
7737 if (mddev->safemode == 2)
7738 md_wakeup_thread(mddev->thread);
7739 else if (mddev->safemode_delay)
7740 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7741 }
7742 }
7743 EXPORT_SYMBOL(md_write_end);
7744
7745 /* md_allow_write(mddev)
7746 * Calling this ensures that the array is marked 'active' so that writes
7747 * may proceed without blocking. It is important to call this before
7748 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7749 * Must be called with mddev_lock held.
7750 *
7751 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7752 * is dropped, so return -EAGAIN after notifying userspace.
7753 */
7754 int md_allow_write(struct mddev *mddev)
7755 {
7756 if (!mddev->pers)
7757 return 0;
7758 if (mddev->ro)
7759 return 0;
7760 if (!mddev->pers->sync_request)
7761 return 0;
7762
7763 spin_lock(&mddev->lock);
7764 if (mddev->in_sync) {
7765 mddev->in_sync = 0;
7766 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7767 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7768 if (mddev->safemode_delay &&
7769 mddev->safemode == 0)
7770 mddev->safemode = 1;
7771 spin_unlock(&mddev->lock);
7772 md_update_sb(mddev, 0);
7773 sysfs_notify_dirent_safe(mddev->sysfs_state);
7774 } else
7775 spin_unlock(&mddev->lock);
7776
7777 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7778 return -EAGAIN;
7779 else
7780 return 0;
7781 }
7782 EXPORT_SYMBOL_GPL(md_allow_write);
7783
7784 #define SYNC_MARKS 10
7785 #define SYNC_MARK_STEP (3*HZ)
7786 #define UPDATE_FREQUENCY (5*60*HZ)
7787 void md_do_sync(struct md_thread *thread)
7788 {
7789 struct mddev *mddev = thread->mddev;
7790 struct mddev *mddev2;
7791 unsigned int currspeed = 0,
7792 window;
7793 sector_t max_sectors,j, io_sectors, recovery_done;
7794 unsigned long mark[SYNC_MARKS];
7795 unsigned long update_time;
7796 sector_t mark_cnt[SYNC_MARKS];
7797 int last_mark,m;
7798 struct list_head *tmp;
7799 sector_t last_check;
7800 int skipped = 0;
7801 struct md_rdev *rdev;
7802 char *desc, *action = NULL;
7803 struct blk_plug plug;
7804 int ret;
7805
7806 /* just incase thread restarts... */
7807 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7808 return;
7809 if (mddev->ro) {/* never try to sync a read-only array */
7810 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7811 return;
7812 }
7813
7814 if (mddev_is_clustered(mddev)) {
7815 ret = md_cluster_ops->resync_start(mddev);
7816 if (ret)
7817 goto skip;
7818
7819 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7820 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7821 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7822 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7823 && ((unsigned long long)mddev->curr_resync_completed
7824 < (unsigned long long)mddev->resync_max_sectors))
7825 goto skip;
7826 }
7827
7828 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7829 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7830 desc = "data-check";
7831 action = "check";
7832 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7833 desc = "requested-resync";
7834 action = "repair";
7835 } else
7836 desc = "resync";
7837 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7838 desc = "reshape";
7839 else
7840 desc = "recovery";
7841
7842 mddev->last_sync_action = action ?: desc;
7843
7844 /* we overload curr_resync somewhat here.
7845 * 0 == not engaged in resync at all
7846 * 2 == checking that there is no conflict with another sync
7847 * 1 == like 2, but have yielded to allow conflicting resync to
7848 * commense
7849 * other == active in resync - this many blocks
7850 *
7851 * Before starting a resync we must have set curr_resync to
7852 * 2, and then checked that every "conflicting" array has curr_resync
7853 * less than ours. When we find one that is the same or higher
7854 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7855 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7856 * This will mean we have to start checking from the beginning again.
7857 *
7858 */
7859
7860 do {
7861 int mddev2_minor = -1;
7862 mddev->curr_resync = 2;
7863
7864 try_again:
7865 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7866 goto skip;
7867 for_each_mddev(mddev2, tmp) {
7868 if (mddev2 == mddev)
7869 continue;
7870 if (!mddev->parallel_resync
7871 && mddev2->curr_resync
7872 && match_mddev_units(mddev, mddev2)) {
7873 DEFINE_WAIT(wq);
7874 if (mddev < mddev2 && mddev->curr_resync == 2) {
7875 /* arbitrarily yield */
7876 mddev->curr_resync = 1;
7877 wake_up(&resync_wait);
7878 }
7879 if (mddev > mddev2 && mddev->curr_resync == 1)
7880 /* no need to wait here, we can wait the next
7881 * time 'round when curr_resync == 2
7882 */
7883 continue;
7884 /* We need to wait 'interruptible' so as not to
7885 * contribute to the load average, and not to
7886 * be caught by 'softlockup'
7887 */
7888 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7889 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7890 mddev2->curr_resync >= mddev->curr_resync) {
7891 if (mddev2_minor != mddev2->md_minor) {
7892 mddev2_minor = mddev2->md_minor;
7893 printk(KERN_INFO "md: delaying %s of %s"
7894 " until %s has finished (they"
7895 " share one or more physical units)\n",
7896 desc, mdname(mddev),
7897 mdname(mddev2));
7898 }
7899 mddev_put(mddev2);
7900 if (signal_pending(current))
7901 flush_signals(current);
7902 schedule();
7903 finish_wait(&resync_wait, &wq);
7904 goto try_again;
7905 }
7906 finish_wait(&resync_wait, &wq);
7907 }
7908 }
7909 } while (mddev->curr_resync < 2);
7910
7911 j = 0;
7912 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7913 /* resync follows the size requested by the personality,
7914 * which defaults to physical size, but can be virtual size
7915 */
7916 max_sectors = mddev->resync_max_sectors;
7917 atomic64_set(&mddev->resync_mismatches, 0);
7918 /* we don't use the checkpoint if there's a bitmap */
7919 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7920 j = mddev->resync_min;
7921 else if (!mddev->bitmap)
7922 j = mddev->recovery_cp;
7923
7924 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7925 max_sectors = mddev->resync_max_sectors;
7926 else {
7927 /* recovery follows the physical size of devices */
7928 max_sectors = mddev->dev_sectors;
7929 j = MaxSector;
7930 rcu_read_lock();
7931 rdev_for_each_rcu(rdev, mddev)
7932 if (rdev->raid_disk >= 0 &&
7933 !test_bit(Journal, &rdev->flags) &&
7934 !test_bit(Faulty, &rdev->flags) &&
7935 !test_bit(In_sync, &rdev->flags) &&
7936 rdev->recovery_offset < j)
7937 j = rdev->recovery_offset;
7938 rcu_read_unlock();
7939
7940 /* If there is a bitmap, we need to make sure all
7941 * writes that started before we added a spare
7942 * complete before we start doing a recovery.
7943 * Otherwise the write might complete and (via
7944 * bitmap_endwrite) set a bit in the bitmap after the
7945 * recovery has checked that bit and skipped that
7946 * region.
7947 */
7948 if (mddev->bitmap) {
7949 mddev->pers->quiesce(mddev, 1);
7950 mddev->pers->quiesce(mddev, 0);
7951 }
7952 }
7953
7954 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7955 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7956 " %d KB/sec/disk.\n", speed_min(mddev));
7957 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7958 "(but not more than %d KB/sec) for %s.\n",
7959 speed_max(mddev), desc);
7960
7961 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7962
7963 io_sectors = 0;
7964 for (m = 0; m < SYNC_MARKS; m++) {
7965 mark[m] = jiffies;
7966 mark_cnt[m] = io_sectors;
7967 }
7968 last_mark = 0;
7969 mddev->resync_mark = mark[last_mark];
7970 mddev->resync_mark_cnt = mark_cnt[last_mark];
7971
7972 /*
7973 * Tune reconstruction:
7974 */
7975 window = 32*(PAGE_SIZE/512);
7976 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7977 window/2, (unsigned long long)max_sectors/2);
7978
7979 atomic_set(&mddev->recovery_active, 0);
7980 last_check = 0;
7981
7982 if (j>2) {
7983 printk(KERN_INFO
7984 "md: resuming %s of %s from checkpoint.\n",
7985 desc, mdname(mddev));
7986 mddev->curr_resync = j;
7987 } else
7988 mddev->curr_resync = 3; /* no longer delayed */
7989 mddev->curr_resync_completed = j;
7990 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7991 md_new_event(mddev);
7992 update_time = jiffies;
7993
7994 blk_start_plug(&plug);
7995 while (j < max_sectors) {
7996 sector_t sectors;
7997
7998 skipped = 0;
7999
8000 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8001 ((mddev->curr_resync > mddev->curr_resync_completed &&
8002 (mddev->curr_resync - mddev->curr_resync_completed)
8003 > (max_sectors >> 4)) ||
8004 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8005 (j - mddev->curr_resync_completed)*2
8006 >= mddev->resync_max - mddev->curr_resync_completed ||
8007 mddev->curr_resync_completed > mddev->resync_max
8008 )) {
8009 /* time to update curr_resync_completed */
8010 wait_event(mddev->recovery_wait,
8011 atomic_read(&mddev->recovery_active) == 0);
8012 mddev->curr_resync_completed = j;
8013 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8014 j > mddev->recovery_cp)
8015 mddev->recovery_cp = j;
8016 update_time = jiffies;
8017 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8018 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8019 }
8020
8021 while (j >= mddev->resync_max &&
8022 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8023 /* As this condition is controlled by user-space,
8024 * we can block indefinitely, so use '_interruptible'
8025 * to avoid triggering warnings.
8026 */
8027 flush_signals(current); /* just in case */
8028 wait_event_interruptible(mddev->recovery_wait,
8029 mddev->resync_max > j
8030 || test_bit(MD_RECOVERY_INTR,
8031 &mddev->recovery));
8032 }
8033
8034 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8035 break;
8036
8037 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8038 if (sectors == 0) {
8039 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8040 break;
8041 }
8042
8043 if (!skipped) { /* actual IO requested */
8044 io_sectors += sectors;
8045 atomic_add(sectors, &mddev->recovery_active);
8046 }
8047
8048 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8049 break;
8050
8051 j += sectors;
8052 if (j > max_sectors)
8053 /* when skipping, extra large numbers can be returned. */
8054 j = max_sectors;
8055 if (j > 2)
8056 mddev->curr_resync = j;
8057 mddev->curr_mark_cnt = io_sectors;
8058 if (last_check == 0)
8059 /* this is the earliest that rebuild will be
8060 * visible in /proc/mdstat
8061 */
8062 md_new_event(mddev);
8063
8064 if (last_check + window > io_sectors || j == max_sectors)
8065 continue;
8066
8067 last_check = io_sectors;
8068 repeat:
8069 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8070 /* step marks */
8071 int next = (last_mark+1) % SYNC_MARKS;
8072
8073 mddev->resync_mark = mark[next];
8074 mddev->resync_mark_cnt = mark_cnt[next];
8075 mark[next] = jiffies;
8076 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8077 last_mark = next;
8078 }
8079
8080 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8081 break;
8082
8083 /*
8084 * this loop exits only if either when we are slower than
8085 * the 'hard' speed limit, or the system was IO-idle for
8086 * a jiffy.
8087 * the system might be non-idle CPU-wise, but we only care
8088 * about not overloading the IO subsystem. (things like an
8089 * e2fsck being done on the RAID array should execute fast)
8090 */
8091 cond_resched();
8092
8093 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8094 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8095 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8096
8097 if (currspeed > speed_min(mddev)) {
8098 if (currspeed > speed_max(mddev)) {
8099 msleep(500);
8100 goto repeat;
8101 }
8102 if (!is_mddev_idle(mddev, 0)) {
8103 /*
8104 * Give other IO more of a chance.
8105 * The faster the devices, the less we wait.
8106 */
8107 wait_event(mddev->recovery_wait,
8108 !atomic_read(&mddev->recovery_active));
8109 }
8110 }
8111 }
8112 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8113 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8114 ? "interrupted" : "done");
8115 /*
8116 * this also signals 'finished resyncing' to md_stop
8117 */
8118 blk_finish_plug(&plug);
8119 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8120
8121 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8122 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8123 mddev->curr_resync > 2) {
8124 mddev->curr_resync_completed = mddev->curr_resync;
8125 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8126 }
8127 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8128
8129 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8130 mddev->curr_resync > 2) {
8131 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8132 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8133 if (mddev->curr_resync >= mddev->recovery_cp) {
8134 printk(KERN_INFO
8135 "md: checkpointing %s of %s.\n",
8136 desc, mdname(mddev));
8137 if (test_bit(MD_RECOVERY_ERROR,
8138 &mddev->recovery))
8139 mddev->recovery_cp =
8140 mddev->curr_resync_completed;
8141 else
8142 mddev->recovery_cp =
8143 mddev->curr_resync;
8144 }
8145 } else
8146 mddev->recovery_cp = MaxSector;
8147 } else {
8148 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8149 mddev->curr_resync = MaxSector;
8150 rcu_read_lock();
8151 rdev_for_each_rcu(rdev, mddev)
8152 if (rdev->raid_disk >= 0 &&
8153 mddev->delta_disks >= 0 &&
8154 !test_bit(Journal, &rdev->flags) &&
8155 !test_bit(Faulty, &rdev->flags) &&
8156 !test_bit(In_sync, &rdev->flags) &&
8157 rdev->recovery_offset < mddev->curr_resync)
8158 rdev->recovery_offset = mddev->curr_resync;
8159 rcu_read_unlock();
8160 }
8161 }
8162 skip:
8163 /* set CHANGE_PENDING here since maybe another update is needed,
8164 * so other nodes are informed. It should be harmless for normal
8165 * raid */
8166 set_mask_bits(&mddev->flags, 0,
8167 BIT(MD_CHANGE_PENDING) | BIT(MD_CHANGE_DEVS));
8168
8169 spin_lock(&mddev->lock);
8170 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8171 /* We completed so min/max setting can be forgotten if used. */
8172 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8173 mddev->resync_min = 0;
8174 mddev->resync_max = MaxSector;
8175 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8176 mddev->resync_min = mddev->curr_resync_completed;
8177 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8178 mddev->curr_resync = 0;
8179 spin_unlock(&mddev->lock);
8180
8181 wake_up(&resync_wait);
8182 md_wakeup_thread(mddev->thread);
8183 return;
8184 }
8185 EXPORT_SYMBOL_GPL(md_do_sync);
8186
8187 static int remove_and_add_spares(struct mddev *mddev,
8188 struct md_rdev *this)
8189 {
8190 struct md_rdev *rdev;
8191 int spares = 0;
8192 int removed = 0;
8193 bool remove_some = false;
8194
8195 rdev_for_each(rdev, mddev) {
8196 if ((this == NULL || rdev == this) &&
8197 rdev->raid_disk >= 0 &&
8198 !test_bit(Blocked, &rdev->flags) &&
8199 test_bit(Faulty, &rdev->flags) &&
8200 atomic_read(&rdev->nr_pending)==0) {
8201 /* Faulty non-Blocked devices with nr_pending == 0
8202 * never get nr_pending incremented,
8203 * never get Faulty cleared, and never get Blocked set.
8204 * So we can synchronize_rcu now rather than once per device
8205 */
8206 remove_some = true;
8207 set_bit(RemoveSynchronized, &rdev->flags);
8208 }
8209 }
8210
8211 if (remove_some)
8212 synchronize_rcu();
8213 rdev_for_each(rdev, mddev) {
8214 if ((this == NULL || rdev == this) &&
8215 rdev->raid_disk >= 0 &&
8216 !test_bit(Blocked, &rdev->flags) &&
8217 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8218 (!test_bit(In_sync, &rdev->flags) &&
8219 !test_bit(Journal, &rdev->flags))) &&
8220 atomic_read(&rdev->nr_pending)==0)) {
8221 if (mddev->pers->hot_remove_disk(
8222 mddev, rdev) == 0) {
8223 sysfs_unlink_rdev(mddev, rdev);
8224 rdev->raid_disk = -1;
8225 removed++;
8226 }
8227 }
8228 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8229 clear_bit(RemoveSynchronized, &rdev->flags);
8230 }
8231
8232 if (removed && mddev->kobj.sd)
8233 sysfs_notify(&mddev->kobj, NULL, "degraded");
8234
8235 if (this && removed)
8236 goto no_add;
8237
8238 rdev_for_each(rdev, mddev) {
8239 if (this && this != rdev)
8240 continue;
8241 if (test_bit(Candidate, &rdev->flags))
8242 continue;
8243 if (rdev->raid_disk >= 0 &&
8244 !test_bit(In_sync, &rdev->flags) &&
8245 !test_bit(Journal, &rdev->flags) &&
8246 !test_bit(Faulty, &rdev->flags))
8247 spares++;
8248 if (rdev->raid_disk >= 0)
8249 continue;
8250 if (test_bit(Faulty, &rdev->flags))
8251 continue;
8252 if (!test_bit(Journal, &rdev->flags)) {
8253 if (mddev->ro &&
8254 ! (rdev->saved_raid_disk >= 0 &&
8255 !test_bit(Bitmap_sync, &rdev->flags)))
8256 continue;
8257
8258 rdev->recovery_offset = 0;
8259 }
8260 if (mddev->pers->
8261 hot_add_disk(mddev, rdev) == 0) {
8262 if (sysfs_link_rdev(mddev, rdev))
8263 /* failure here is OK */;
8264 if (!test_bit(Journal, &rdev->flags))
8265 spares++;
8266 md_new_event(mddev);
8267 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8268 }
8269 }
8270 no_add:
8271 if (removed)
8272 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8273 return spares;
8274 }
8275
8276 static void md_start_sync(struct work_struct *ws)
8277 {
8278 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8279
8280 mddev->sync_thread = md_register_thread(md_do_sync,
8281 mddev,
8282 "resync");
8283 if (!mddev->sync_thread) {
8284 printk(KERN_ERR "%s: could not start resync thread...\n",
8285 mdname(mddev));
8286 /* leave the spares where they are, it shouldn't hurt */
8287 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8288 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8289 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8290 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8291 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8292 wake_up(&resync_wait);
8293 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8294 &mddev->recovery))
8295 if (mddev->sysfs_action)
8296 sysfs_notify_dirent_safe(mddev->sysfs_action);
8297 } else
8298 md_wakeup_thread(mddev->sync_thread);
8299 sysfs_notify_dirent_safe(mddev->sysfs_action);
8300 md_new_event(mddev);
8301 }
8302
8303 /*
8304 * This routine is regularly called by all per-raid-array threads to
8305 * deal with generic issues like resync and super-block update.
8306 * Raid personalities that don't have a thread (linear/raid0) do not
8307 * need this as they never do any recovery or update the superblock.
8308 *
8309 * It does not do any resync itself, but rather "forks" off other threads
8310 * to do that as needed.
8311 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8312 * "->recovery" and create a thread at ->sync_thread.
8313 * When the thread finishes it sets MD_RECOVERY_DONE
8314 * and wakeups up this thread which will reap the thread and finish up.
8315 * This thread also removes any faulty devices (with nr_pending == 0).
8316 *
8317 * The overall approach is:
8318 * 1/ if the superblock needs updating, update it.
8319 * 2/ If a recovery thread is running, don't do anything else.
8320 * 3/ If recovery has finished, clean up, possibly marking spares active.
8321 * 4/ If there are any faulty devices, remove them.
8322 * 5/ If array is degraded, try to add spares devices
8323 * 6/ If array has spares or is not in-sync, start a resync thread.
8324 */
8325 void md_check_recovery(struct mddev *mddev)
8326 {
8327 if (mddev->suspended)
8328 return;
8329
8330 if (mddev->bitmap)
8331 bitmap_daemon_work(mddev);
8332
8333 if (signal_pending(current)) {
8334 if (mddev->pers->sync_request && !mddev->external) {
8335 printk(KERN_INFO "md: %s in immediate safe mode\n",
8336 mdname(mddev));
8337 mddev->safemode = 2;
8338 }
8339 flush_signals(current);
8340 }
8341
8342 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8343 return;
8344 if ( ! (
8345 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8346 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8347 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8348 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8349 (mddev->external == 0 && mddev->safemode == 1) ||
8350 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8351 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8352 ))
8353 return;
8354
8355 if (mddev_trylock(mddev)) {
8356 int spares = 0;
8357
8358 if (mddev->ro) {
8359 struct md_rdev *rdev;
8360 if (!mddev->external && mddev->in_sync)
8361 /* 'Blocked' flag not needed as failed devices
8362 * will be recorded if array switched to read/write.
8363 * Leaving it set will prevent the device
8364 * from being removed.
8365 */
8366 rdev_for_each(rdev, mddev)
8367 clear_bit(Blocked, &rdev->flags);
8368 /* On a read-only array we can:
8369 * - remove failed devices
8370 * - add already-in_sync devices if the array itself
8371 * is in-sync.
8372 * As we only add devices that are already in-sync,
8373 * we can activate the spares immediately.
8374 */
8375 remove_and_add_spares(mddev, NULL);
8376 /* There is no thread, but we need to call
8377 * ->spare_active and clear saved_raid_disk
8378 */
8379 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8380 md_reap_sync_thread(mddev);
8381 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8382 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8383 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8384 goto unlock;
8385 }
8386
8387 if (mddev_is_clustered(mddev)) {
8388 struct md_rdev *rdev;
8389 /* kick the device if another node issued a
8390 * remove disk.
8391 */
8392 rdev_for_each(rdev, mddev) {
8393 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8394 rdev->raid_disk < 0)
8395 md_kick_rdev_from_array(rdev);
8396 }
8397
8398 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8399 md_reload_sb(mddev, mddev->good_device_nr);
8400 }
8401
8402 if (!mddev->external) {
8403 int did_change = 0;
8404 spin_lock(&mddev->lock);
8405 if (mddev->safemode &&
8406 !atomic_read(&mddev->writes_pending) &&
8407 !mddev->in_sync &&
8408 mddev->recovery_cp == MaxSector) {
8409 mddev->in_sync = 1;
8410 did_change = 1;
8411 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8412 }
8413 if (mddev->safemode == 1)
8414 mddev->safemode = 0;
8415 spin_unlock(&mddev->lock);
8416 if (did_change)
8417 sysfs_notify_dirent_safe(mddev->sysfs_state);
8418 }
8419
8420 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8421 md_update_sb(mddev, 0);
8422
8423 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8424 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8425 /* resync/recovery still happening */
8426 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8427 goto unlock;
8428 }
8429 if (mddev->sync_thread) {
8430 md_reap_sync_thread(mddev);
8431 goto unlock;
8432 }
8433 /* Set RUNNING before clearing NEEDED to avoid
8434 * any transients in the value of "sync_action".
8435 */
8436 mddev->curr_resync_completed = 0;
8437 spin_lock(&mddev->lock);
8438 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8439 spin_unlock(&mddev->lock);
8440 /* Clear some bits that don't mean anything, but
8441 * might be left set
8442 */
8443 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8444 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8445
8446 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8447 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8448 goto not_running;
8449 /* no recovery is running.
8450 * remove any failed drives, then
8451 * add spares if possible.
8452 * Spares are also removed and re-added, to allow
8453 * the personality to fail the re-add.
8454 */
8455
8456 if (mddev->reshape_position != MaxSector) {
8457 if (mddev->pers->check_reshape == NULL ||
8458 mddev->pers->check_reshape(mddev) != 0)
8459 /* Cannot proceed */
8460 goto not_running;
8461 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8462 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8463 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8464 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8465 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8466 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8467 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8468 } else if (mddev->recovery_cp < MaxSector) {
8469 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8470 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8471 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8472 /* nothing to be done ... */
8473 goto not_running;
8474
8475 if (mddev->pers->sync_request) {
8476 if (spares) {
8477 /* We are adding a device or devices to an array
8478 * which has the bitmap stored on all devices.
8479 * So make sure all bitmap pages get written
8480 */
8481 bitmap_write_all(mddev->bitmap);
8482 }
8483 INIT_WORK(&mddev->del_work, md_start_sync);
8484 queue_work(md_misc_wq, &mddev->del_work);
8485 goto unlock;
8486 }
8487 not_running:
8488 if (!mddev->sync_thread) {
8489 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8490 wake_up(&resync_wait);
8491 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8492 &mddev->recovery))
8493 if (mddev->sysfs_action)
8494 sysfs_notify_dirent_safe(mddev->sysfs_action);
8495 }
8496 unlock:
8497 wake_up(&mddev->sb_wait);
8498 mddev_unlock(mddev);
8499 }
8500 }
8501 EXPORT_SYMBOL(md_check_recovery);
8502
8503 void md_reap_sync_thread(struct mddev *mddev)
8504 {
8505 struct md_rdev *rdev;
8506
8507 /* resync has finished, collect result */
8508 md_unregister_thread(&mddev->sync_thread);
8509 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8510 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8511 /* success...*/
8512 /* activate any spares */
8513 if (mddev->pers->spare_active(mddev)) {
8514 sysfs_notify(&mddev->kobj, NULL,
8515 "degraded");
8516 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8517 }
8518 }
8519 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8520 mddev->pers->finish_reshape)
8521 mddev->pers->finish_reshape(mddev);
8522
8523 /* If array is no-longer degraded, then any saved_raid_disk
8524 * information must be scrapped.
8525 */
8526 if (!mddev->degraded)
8527 rdev_for_each(rdev, mddev)
8528 rdev->saved_raid_disk = -1;
8529
8530 md_update_sb(mddev, 1);
8531 /* MD_CHANGE_PENDING should be cleared by md_update_sb, so we can
8532 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8533 * clustered raid */
8534 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8535 md_cluster_ops->resync_finish(mddev);
8536 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8537 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8538 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8539 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8540 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8541 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8542 wake_up(&resync_wait);
8543 /* flag recovery needed just to double check */
8544 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8545 sysfs_notify_dirent_safe(mddev->sysfs_action);
8546 md_new_event(mddev);
8547 if (mddev->event_work.func)
8548 queue_work(md_misc_wq, &mddev->event_work);
8549 }
8550 EXPORT_SYMBOL(md_reap_sync_thread);
8551
8552 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8553 {
8554 sysfs_notify_dirent_safe(rdev->sysfs_state);
8555 wait_event_timeout(rdev->blocked_wait,
8556 !test_bit(Blocked, &rdev->flags) &&
8557 !test_bit(BlockedBadBlocks, &rdev->flags),
8558 msecs_to_jiffies(5000));
8559 rdev_dec_pending(rdev, mddev);
8560 }
8561 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8562
8563 void md_finish_reshape(struct mddev *mddev)
8564 {
8565 /* called be personality module when reshape completes. */
8566 struct md_rdev *rdev;
8567
8568 rdev_for_each(rdev, mddev) {
8569 if (rdev->data_offset > rdev->new_data_offset)
8570 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8571 else
8572 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8573 rdev->data_offset = rdev->new_data_offset;
8574 }
8575 }
8576 EXPORT_SYMBOL(md_finish_reshape);
8577
8578 /* Bad block management */
8579
8580 /* Returns 1 on success, 0 on failure */
8581 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8582 int is_new)
8583 {
8584 struct mddev *mddev = rdev->mddev;
8585 int rv;
8586 if (is_new)
8587 s += rdev->new_data_offset;
8588 else
8589 s += rdev->data_offset;
8590 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8591 if (rv == 0) {
8592 /* Make sure they get written out promptly */
8593 sysfs_notify_dirent_safe(rdev->sysfs_state);
8594 set_mask_bits(&mddev->flags, 0,
8595 BIT(MD_CHANGE_CLEAN) | BIT(MD_CHANGE_PENDING));
8596 md_wakeup_thread(rdev->mddev->thread);
8597 return 1;
8598 } else
8599 return 0;
8600 }
8601 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8602
8603 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8604 int is_new)
8605 {
8606 if (is_new)
8607 s += rdev->new_data_offset;
8608 else
8609 s += rdev->data_offset;
8610 return badblocks_clear(&rdev->badblocks,
8611 s, sectors);
8612 }
8613 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8614
8615 static int md_notify_reboot(struct notifier_block *this,
8616 unsigned long code, void *x)
8617 {
8618 struct list_head *tmp;
8619 struct mddev *mddev;
8620 int need_delay = 0;
8621
8622 for_each_mddev(mddev, tmp) {
8623 if (mddev_trylock(mddev)) {
8624 if (mddev->pers)
8625 __md_stop_writes(mddev);
8626 if (mddev->persistent)
8627 mddev->safemode = 2;
8628 mddev_unlock(mddev);
8629 }
8630 need_delay = 1;
8631 }
8632 /*
8633 * certain more exotic SCSI devices are known to be
8634 * volatile wrt too early system reboots. While the
8635 * right place to handle this issue is the given
8636 * driver, we do want to have a safe RAID driver ...
8637 */
8638 if (need_delay)
8639 mdelay(1000*1);
8640
8641 return NOTIFY_DONE;
8642 }
8643
8644 static struct notifier_block md_notifier = {
8645 .notifier_call = md_notify_reboot,
8646 .next = NULL,
8647 .priority = INT_MAX, /* before any real devices */
8648 };
8649
8650 static void md_geninit(void)
8651 {
8652 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8653
8654 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8655 }
8656
8657 static int __init md_init(void)
8658 {
8659 int ret = -ENOMEM;
8660
8661 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8662 if (!md_wq)
8663 goto err_wq;
8664
8665 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8666 if (!md_misc_wq)
8667 goto err_misc_wq;
8668
8669 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8670 goto err_md;
8671
8672 if ((ret = register_blkdev(0, "mdp")) < 0)
8673 goto err_mdp;
8674 mdp_major = ret;
8675
8676 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8677 md_probe, NULL, NULL);
8678 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8679 md_probe, NULL, NULL);
8680
8681 register_reboot_notifier(&md_notifier);
8682 raid_table_header = register_sysctl_table(raid_root_table);
8683
8684 md_geninit();
8685 return 0;
8686
8687 err_mdp:
8688 unregister_blkdev(MD_MAJOR, "md");
8689 err_md:
8690 destroy_workqueue(md_misc_wq);
8691 err_misc_wq:
8692 destroy_workqueue(md_wq);
8693 err_wq:
8694 return ret;
8695 }
8696
8697 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8698 {
8699 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8700 struct md_rdev *rdev2;
8701 int role, ret;
8702 char b[BDEVNAME_SIZE];
8703
8704 /* Check for change of roles in the active devices */
8705 rdev_for_each(rdev2, mddev) {
8706 if (test_bit(Faulty, &rdev2->flags))
8707 continue;
8708
8709 /* Check if the roles changed */
8710 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8711
8712 if (test_bit(Candidate, &rdev2->flags)) {
8713 if (role == 0xfffe) {
8714 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8715 md_kick_rdev_from_array(rdev2);
8716 continue;
8717 }
8718 else
8719 clear_bit(Candidate, &rdev2->flags);
8720 }
8721
8722 if (role != rdev2->raid_disk) {
8723 /* got activated */
8724 if (rdev2->raid_disk == -1 && role != 0xffff) {
8725 rdev2->saved_raid_disk = role;
8726 ret = remove_and_add_spares(mddev, rdev2);
8727 pr_info("Activated spare: %s\n",
8728 bdevname(rdev2->bdev,b));
8729 /* wakeup mddev->thread here, so array could
8730 * perform resync with the new activated disk */
8731 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8732 md_wakeup_thread(mddev->thread);
8733
8734 }
8735 /* device faulty
8736 * We just want to do the minimum to mark the disk
8737 * as faulty. The recovery is performed by the
8738 * one who initiated the error.
8739 */
8740 if ((role == 0xfffe) || (role == 0xfffd)) {
8741 md_error(mddev, rdev2);
8742 clear_bit(Blocked, &rdev2->flags);
8743 }
8744 }
8745 }
8746
8747 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8748 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8749
8750 /* Finally set the event to be up to date */
8751 mddev->events = le64_to_cpu(sb->events);
8752 }
8753
8754 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8755 {
8756 int err;
8757 struct page *swapout = rdev->sb_page;
8758 struct mdp_superblock_1 *sb;
8759
8760 /* Store the sb page of the rdev in the swapout temporary
8761 * variable in case we err in the future
8762 */
8763 rdev->sb_page = NULL;
8764 alloc_disk_sb(rdev);
8765 ClearPageUptodate(rdev->sb_page);
8766 rdev->sb_loaded = 0;
8767 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
8768
8769 if (err < 0) {
8770 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8771 __func__, __LINE__, rdev->desc_nr, err);
8772 put_page(rdev->sb_page);
8773 rdev->sb_page = swapout;
8774 rdev->sb_loaded = 1;
8775 return err;
8776 }
8777
8778 sb = page_address(rdev->sb_page);
8779 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8780 * is not set
8781 */
8782
8783 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8784 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8785
8786 /* The other node finished recovery, call spare_active to set
8787 * device In_sync and mddev->degraded
8788 */
8789 if (rdev->recovery_offset == MaxSector &&
8790 !test_bit(In_sync, &rdev->flags) &&
8791 mddev->pers->spare_active(mddev))
8792 sysfs_notify(&mddev->kobj, NULL, "degraded");
8793
8794 put_page(swapout);
8795 return 0;
8796 }
8797
8798 void md_reload_sb(struct mddev *mddev, int nr)
8799 {
8800 struct md_rdev *rdev;
8801 int err;
8802
8803 /* Find the rdev */
8804 rdev_for_each_rcu(rdev, mddev) {
8805 if (rdev->desc_nr == nr)
8806 break;
8807 }
8808
8809 if (!rdev || rdev->desc_nr != nr) {
8810 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8811 return;
8812 }
8813
8814 err = read_rdev(mddev, rdev);
8815 if (err < 0)
8816 return;
8817
8818 check_sb_changes(mddev, rdev);
8819
8820 /* Read all rdev's to update recovery_offset */
8821 rdev_for_each_rcu(rdev, mddev)
8822 read_rdev(mddev, rdev);
8823 }
8824 EXPORT_SYMBOL(md_reload_sb);
8825
8826 #ifndef MODULE
8827
8828 /*
8829 * Searches all registered partitions for autorun RAID arrays
8830 * at boot time.
8831 */
8832
8833 static DEFINE_MUTEX(detected_devices_mutex);
8834 static LIST_HEAD(all_detected_devices);
8835 struct detected_devices_node {
8836 struct list_head list;
8837 dev_t dev;
8838 };
8839
8840 void md_autodetect_dev(dev_t dev)
8841 {
8842 struct detected_devices_node *node_detected_dev;
8843
8844 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8845 if (node_detected_dev) {
8846 node_detected_dev->dev = dev;
8847 mutex_lock(&detected_devices_mutex);
8848 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8849 mutex_unlock(&detected_devices_mutex);
8850 } else {
8851 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8852 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8853 }
8854 }
8855
8856 static void autostart_arrays(int part)
8857 {
8858 struct md_rdev *rdev;
8859 struct detected_devices_node *node_detected_dev;
8860 dev_t dev;
8861 int i_scanned, i_passed;
8862
8863 i_scanned = 0;
8864 i_passed = 0;
8865
8866 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8867
8868 mutex_lock(&detected_devices_mutex);
8869 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8870 i_scanned++;
8871 node_detected_dev = list_entry(all_detected_devices.next,
8872 struct detected_devices_node, list);
8873 list_del(&node_detected_dev->list);
8874 dev = node_detected_dev->dev;
8875 kfree(node_detected_dev);
8876 rdev = md_import_device(dev,0, 90);
8877 if (IS_ERR(rdev))
8878 continue;
8879
8880 if (test_bit(Faulty, &rdev->flags))
8881 continue;
8882
8883 set_bit(AutoDetected, &rdev->flags);
8884 list_add(&rdev->same_set, &pending_raid_disks);
8885 i_passed++;
8886 }
8887 mutex_unlock(&detected_devices_mutex);
8888
8889 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8890 i_scanned, i_passed);
8891
8892 autorun_devices(part);
8893 }
8894
8895 #endif /* !MODULE */
8896
8897 static __exit void md_exit(void)
8898 {
8899 struct mddev *mddev;
8900 struct list_head *tmp;
8901 int delay = 1;
8902
8903 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8904 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8905
8906 unregister_blkdev(MD_MAJOR,"md");
8907 unregister_blkdev(mdp_major, "mdp");
8908 unregister_reboot_notifier(&md_notifier);
8909 unregister_sysctl_table(raid_table_header);
8910
8911 /* We cannot unload the modules while some process is
8912 * waiting for us in select() or poll() - wake them up
8913 */
8914 md_unloading = 1;
8915 while (waitqueue_active(&md_event_waiters)) {
8916 /* not safe to leave yet */
8917 wake_up(&md_event_waiters);
8918 msleep(delay);
8919 delay += delay;
8920 }
8921 remove_proc_entry("mdstat", NULL);
8922
8923 for_each_mddev(mddev, tmp) {
8924 export_array(mddev);
8925 mddev->hold_active = 0;
8926 }
8927 destroy_workqueue(md_misc_wq);
8928 destroy_workqueue(md_wq);
8929 }
8930
8931 subsys_initcall(md_init);
8932 module_exit(md_exit)
8933
8934 static int get_ro(char *buffer, struct kernel_param *kp)
8935 {
8936 return sprintf(buffer, "%d", start_readonly);
8937 }
8938 static int set_ro(const char *val, struct kernel_param *kp)
8939 {
8940 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
8941 }
8942
8943 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8944 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8945 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8946
8947 MODULE_LICENSE("GPL");
8948 MODULE_DESCRIPTION("MD RAID framework");
8949 MODULE_ALIAS("md");
8950 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux with added FPC-III support