]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame_incremental - drivers/md/md.c
projects / mirror_ubuntu-jammy-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
proc: convert everything to "struct proc_ops"
[mirror_ubuntu-jammy-kernel.git] / drivers / md / md.c
... / ...
CommitLineData
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5
6 completely rewritten, based on the MD driver code from Marc Zyngier
7
8 Changes:
9
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
20
21 Neil Brown <neilb@cse.unsw.edu.au>.
22
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25
26
27 Errors, Warnings, etc.
28 Please use:
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
37
38*/
39
40#include <linux/sched/mm.h>
41#include <linux/sched/signal.h>
42#include <linux/kthread.h>
43#include <linux/blkdev.h>
44#include <linux/badblocks.h>
45#include <linux/sysctl.h>
46#include <linux/seq_file.h>
47#include <linux/fs.h>
48#include <linux/poll.h>
49#include <linux/ctype.h>
50#include <linux/string.h>
51#include <linux/hdreg.h>
52#include <linux/proc_fs.h>
53#include <linux/random.h>
54#include <linux/module.h>
55#include <linux/reboot.h>
56#include <linux/file.h>
57#include <linux/compat.h>
58#include <linux/delay.h>
59#include <linux/raid/md_p.h>
60#include <linux/raid/md_u.h>
61#include <linux/slab.h>
62#include <linux/percpu-refcount.h>
63
64#include <trace/events/block.h>
65#include "md.h"
66#include "md-bitmap.h"
67#include "md-cluster.h"
68
69#ifndef MODULE
70static void autostart_arrays(int part);
71#endif
72
73/* pers_list is a list of registered personalities protected
74 * by pers_lock.
75 * pers_lock does extra service to protect accesses to
76 * mddev->thread when the mutex cannot be held.
77 */
78static LIST_HEAD(pers_list);
79static DEFINE_SPINLOCK(pers_lock);
80
81static struct kobj_type md_ktype;
82
83struct md_cluster_operations *md_cluster_ops;
84EXPORT_SYMBOL(md_cluster_ops);
85static struct module *md_cluster_mod;
86
87static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
88static struct workqueue_struct *md_wq;
89static struct workqueue_struct *md_misc_wq;
90
91static int remove_and_add_spares(struct mddev *mddev,
92 struct md_rdev *this);
93static void mddev_detach(struct mddev *mddev);
94
95/*
96 * Default number of read corrections we'll attempt on an rdev
97 * before ejecting it from the array. We divide the read error
98 * count by 2 for every hour elapsed between read errors.
99 */
100#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
101/*
102 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
103 * is 1000 KB/sec, so the extra system load does not show up that much.
104 * Increase it if you want to have more _guaranteed_ speed. Note that
105 * the RAID driver will use the maximum available bandwidth if the IO
106 * subsystem is idle. There is also an 'absolute maximum' reconstruction
107 * speed limit - in case reconstruction slows down your system despite
108 * idle IO detection.
109 *
110 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
111 * or /sys/block/mdX/md/sync_speed_{min,max}
112 */
113
114static int sysctl_speed_limit_min = 1000;
115static int sysctl_speed_limit_max = 200000;
116static inline int speed_min(struct mddev *mddev)
117{
118 return mddev->sync_speed_min ?
119 mddev->sync_speed_min : sysctl_speed_limit_min;
120}
121
122static inline int speed_max(struct mddev *mddev)
123{
124 return mddev->sync_speed_max ?
125 mddev->sync_speed_max : sysctl_speed_limit_max;
126}
127
128static void rdev_uninit_serial(struct md_rdev *rdev)
129{
130 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
131 return;
132
133 kvfree(rdev->serial);
134 rdev->serial = NULL;
135}
136
137static void rdevs_uninit_serial(struct mddev *mddev)
138{
139 struct md_rdev *rdev;
140
141 rdev_for_each(rdev, mddev)
142 rdev_uninit_serial(rdev);
143}
144
145static int rdev_init_serial(struct md_rdev *rdev)
146{
147 /* serial_nums equals with BARRIER_BUCKETS_NR */
148 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
149 struct serial_in_rdev *serial = NULL;
150
151 if (test_bit(CollisionCheck, &rdev->flags))
152 return 0;
153
154 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
155 GFP_KERNEL);
156 if (!serial)
157 return -ENOMEM;
158
159 for (i = 0; i < serial_nums; i++) {
160 struct serial_in_rdev *serial_tmp = &serial[i];
161
162 spin_lock_init(&serial_tmp->serial_lock);
163 serial_tmp->serial_rb = RB_ROOT_CACHED;
164 init_waitqueue_head(&serial_tmp->serial_io_wait);
165 }
166
167 rdev->serial = serial;
168 set_bit(CollisionCheck, &rdev->flags);
169
170 return 0;
171}
172
173static int rdevs_init_serial(struct mddev *mddev)
174{
175 struct md_rdev *rdev;
176 int ret = 0;
177
178 rdev_for_each(rdev, mddev) {
179 ret = rdev_init_serial(rdev);
180 if (ret)
181 break;
182 }
183
184 /* Free all resources if pool is not existed */
185 if (ret && !mddev->serial_info_pool)
186 rdevs_uninit_serial(mddev);
187
188 return ret;
189}
190
191/*
192 * rdev needs to enable serial stuffs if it meets the conditions:
193 * 1. it is multi-queue device flaged with writemostly.
194 * 2. the write-behind mode is enabled.
195 */
196static int rdev_need_serial(struct md_rdev *rdev)
197{
198 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
199 rdev->bdev->bd_queue->nr_hw_queues != 1 &&
200 test_bit(WriteMostly, &rdev->flags));
201}
202
203/*
204 * Init resource for rdev(s), then create serial_info_pool if:
205 * 1. rdev is the first device which return true from rdev_enable_serial.
206 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
207 */
208void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
209 bool is_suspend)
210{
211 int ret = 0;
212
213 if (rdev && !rdev_need_serial(rdev) &&
214 !test_bit(CollisionCheck, &rdev->flags))
215 return;
216
217 if (!is_suspend)
218 mddev_suspend(mddev);
219
220 if (!rdev)
221 ret = rdevs_init_serial(mddev);
222 else
223 ret = rdev_init_serial(rdev);
224 if (ret)
225 goto abort;
226
227 if (mddev->serial_info_pool == NULL) {
228 unsigned int noio_flag;
229
230 noio_flag = memalloc_noio_save();
231 mddev->serial_info_pool =
232 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
233 sizeof(struct serial_info));
234 memalloc_noio_restore(noio_flag);
235 if (!mddev->serial_info_pool) {
236 rdevs_uninit_serial(mddev);
237 pr_err("can't alloc memory pool for serialization\n");
238 }
239 }
240
241abort:
242 if (!is_suspend)
243 mddev_resume(mddev);
244}
245
246/*
247 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
248 * 1. rdev is the last device flaged with CollisionCheck.
249 * 2. when bitmap is destroyed while policy is not enabled.
250 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
251 */
252void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
253 bool is_suspend)
254{
255 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
256 return;
257
258 if (mddev->serial_info_pool) {
259 struct md_rdev *temp;
260 int num = 0; /* used to track if other rdevs need the pool */
261
262 if (!is_suspend)
263 mddev_suspend(mddev);
264 rdev_for_each(temp, mddev) {
265 if (!rdev) {
266 if (!mddev->serialize_policy ||
267 !rdev_need_serial(temp))
268 rdev_uninit_serial(temp);
269 else
270 num++;
271 } else if (temp != rdev &&
272 test_bit(CollisionCheck, &temp->flags))
273 num++;
274 }
275
276 if (rdev)
277 rdev_uninit_serial(rdev);
278
279 if (num)
280 pr_info("The mempool could be used by other devices\n");
281 else {
282 mempool_destroy(mddev->serial_info_pool);
283 mddev->serial_info_pool = NULL;
284 }
285 if (!is_suspend)
286 mddev_resume(mddev);
287 }
288}
289
290static struct ctl_table_header *raid_table_header;
291
292static struct ctl_table raid_table[] = {
293 {
294 .procname = "speed_limit_min",
295 .data = &sysctl_speed_limit_min,
296 .maxlen = sizeof(int),
297 .mode = S_IRUGO|S_IWUSR,
298 .proc_handler = proc_dointvec,
299 },
300 {
301 .procname = "speed_limit_max",
302 .data = &sysctl_speed_limit_max,
303 .maxlen = sizeof(int),
304 .mode = S_IRUGO|S_IWUSR,
305 .proc_handler = proc_dointvec,
306 },
307 { }
308};
309
310static struct ctl_table raid_dir_table[] = {
311 {
312 .procname = "raid",
313 .maxlen = 0,
314 .mode = S_IRUGO|S_IXUGO,
315 .child = raid_table,
316 },
317 { }
318};
319
320static struct ctl_table raid_root_table[] = {
321 {
322 .procname = "dev",
323 .maxlen = 0,
324 .mode = 0555,
325 .child = raid_dir_table,
326 },
327 { }
328};
329
330static const struct block_device_operations md_fops;
331
332static int start_readonly;
333
334/*
335 * The original mechanism for creating an md device is to create
336 * a device node in /dev and to open it. This causes races with device-close.
337 * The preferred method is to write to the "new_array" module parameter.
338 * This can avoid races.
339 * Setting create_on_open to false disables the original mechanism
340 * so all the races disappear.
341 */
342static bool create_on_open = true;
343
344struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
345 struct mddev *mddev)
346{
347 if (!mddev || !bioset_initialized(&mddev->bio_set))
348 return bio_alloc(gfp_mask, nr_iovecs);
349
350 return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
351}
352EXPORT_SYMBOL_GPL(bio_alloc_mddev);
353
354static struct bio *md_bio_alloc_sync(struct mddev *mddev)
355{
356 if (!mddev || !bioset_initialized(&mddev->sync_set))
357 return bio_alloc(GFP_NOIO, 1);
358
359 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
360}
361
362/*
363 * We have a system wide 'event count' that is incremented
364 * on any 'interesting' event, and readers of /proc/mdstat
365 * can use 'poll' or 'select' to find out when the event
366 * count increases.
367 *
368 * Events are:
369 * start array, stop array, error, add device, remove device,
370 * start build, activate spare
371 */
372static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
373static atomic_t md_event_count;
374void md_new_event(struct mddev *mddev)
375{
376 atomic_inc(&md_event_count);
377 wake_up(&md_event_waiters);
378}
379EXPORT_SYMBOL_GPL(md_new_event);
380
381/*
382 * Enables to iterate over all existing md arrays
383 * all_mddevs_lock protects this list.
384 */
385static LIST_HEAD(all_mddevs);
386static DEFINE_SPINLOCK(all_mddevs_lock);
387
388/*
389 * iterates through all used mddevs in the system.
390 * We take care to grab the all_mddevs_lock whenever navigating
391 * the list, and to always hold a refcount when unlocked.
392 * Any code which breaks out of this loop while own
393 * a reference to the current mddev and must mddev_put it.
394 */
395#define for_each_mddev(_mddev,_tmp) \
396 \
397 for (({ spin_lock(&all_mddevs_lock); \
398 _tmp = all_mddevs.next; \
399 _mddev = NULL;}); \
400 ({ if (_tmp != &all_mddevs) \
401 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
402 spin_unlock(&all_mddevs_lock); \
403 if (_mddev) mddev_put(_mddev); \
404 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
405 _tmp != &all_mddevs;}); \
406 ({ spin_lock(&all_mddevs_lock); \
407 _tmp = _tmp->next;}) \
408 )
409
410/* Rather than calling directly into the personality make_request function,
411 * IO requests come here first so that we can check if the device is
412 * being suspended pending a reconfiguration.
413 * We hold a refcount over the call to ->make_request. By the time that
414 * call has finished, the bio has been linked into some internal structure
415 * and so is visible to ->quiesce(), so we don't need the refcount any more.
416 */
417static bool is_suspended(struct mddev *mddev, struct bio *bio)
418{
419 if (mddev->suspended)
420 return true;
421 if (bio_data_dir(bio) != WRITE)
422 return false;
423 if (mddev->suspend_lo >= mddev->suspend_hi)
424 return false;
425 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
426 return false;
427 if (bio_end_sector(bio) < mddev->suspend_lo)
428 return false;
429 return true;
430}
431
432void md_handle_request(struct mddev *mddev, struct bio *bio)
433{
434check_suspended:
435 rcu_read_lock();
436 if (is_suspended(mddev, bio)) {
437 DEFINE_WAIT(__wait);
438 for (;;) {
439 prepare_to_wait(&mddev->sb_wait, &__wait,
440 TASK_UNINTERRUPTIBLE);
441 if (!is_suspended(mddev, bio))
442 break;
443 rcu_read_unlock();
444 schedule();
445 rcu_read_lock();
446 }
447 finish_wait(&mddev->sb_wait, &__wait);
448 }
449 atomic_inc(&mddev->active_io);
450 rcu_read_unlock();
451
452 if (!mddev->pers->make_request(mddev, bio)) {
453 atomic_dec(&mddev->active_io);
454 wake_up(&mddev->sb_wait);
455 goto check_suspended;
456 }
457
458 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
459 wake_up(&mddev->sb_wait);
460}
461EXPORT_SYMBOL(md_handle_request);
462
463static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
464{
465 const int rw = bio_data_dir(bio);
466 const int sgrp = op_stat_group(bio_op(bio));
467 struct mddev *mddev = q->queuedata;
468 unsigned int sectors;
469
470 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
471 bio_io_error(bio);
472 return BLK_QC_T_NONE;
473 }
474
475 blk_queue_split(q, &bio);
476
477 if (mddev == NULL || mddev->pers == NULL) {
478 bio_io_error(bio);
479 return BLK_QC_T_NONE;
480 }
481 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
482 if (bio_sectors(bio) != 0)
483 bio->bi_status = BLK_STS_IOERR;
484 bio_endio(bio);
485 return BLK_QC_T_NONE;
486 }
487
488 /*
489 * save the sectors now since our bio can
490 * go away inside make_request
491 */
492 sectors = bio_sectors(bio);
493 /* bio could be mergeable after passing to underlayer */
494 bio->bi_opf &= ~REQ_NOMERGE;
495
496 md_handle_request(mddev, bio);
497
498 part_stat_lock();
499 part_stat_inc(&mddev->gendisk->part0, ios[sgrp]);
500 part_stat_add(&mddev->gendisk->part0, sectors[sgrp], sectors);
501 part_stat_unlock();
502
503 return BLK_QC_T_NONE;
504}
505
506/* mddev_suspend makes sure no new requests are submitted
507 * to the device, and that any requests that have been submitted
508 * are completely handled.
509 * Once mddev_detach() is called and completes, the module will be
510 * completely unused.
511 */
512void mddev_suspend(struct mddev *mddev)
513{
514 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
515 lockdep_assert_held(&mddev->reconfig_mutex);
516 if (mddev->suspended++)
517 return;
518 synchronize_rcu();
519 wake_up(&mddev->sb_wait);
520 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
521 smp_mb__after_atomic();
522 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
523 mddev->pers->quiesce(mddev, 1);
524 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
525 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
526
527 del_timer_sync(&mddev->safemode_timer);
528}
529EXPORT_SYMBOL_GPL(mddev_suspend);
530
531void mddev_resume(struct mddev *mddev)
532{
533 lockdep_assert_held(&mddev->reconfig_mutex);
534 if (--mddev->suspended)
535 return;
536 wake_up(&mddev->sb_wait);
537 mddev->pers->quiesce(mddev, 0);
538
539 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
540 md_wakeup_thread(mddev->thread);
541 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
542}
543EXPORT_SYMBOL_GPL(mddev_resume);
544
545int mddev_congested(struct mddev *mddev, int bits)
546{
547 struct md_personality *pers = mddev->pers;
548 int ret = 0;
549
550 rcu_read_lock();
551 if (mddev->suspended)
552 ret = 1;
553 else if (pers && pers->congested)
554 ret = pers->congested(mddev, bits);
555 rcu_read_unlock();
556 return ret;
557}
558EXPORT_SYMBOL_GPL(mddev_congested);
559static int md_congested(void *data, int bits)
560{
561 struct mddev *mddev = data;
562 return mddev_congested(mddev, bits);
563}
564
565/*
566 * Generic flush handling for md
567 */
568
569static void md_end_flush(struct bio *bio)
570{
571 struct md_rdev *rdev = bio->bi_private;
572 struct mddev *mddev = rdev->mddev;
573
574 rdev_dec_pending(rdev, mddev);
575
576 if (atomic_dec_and_test(&mddev->flush_pending)) {
577 /* The pre-request flush has finished */
578 queue_work(md_wq, &mddev->flush_work);
579 }
580 bio_put(bio);
581}
582
583static void md_submit_flush_data(struct work_struct *ws);
584
585static void submit_flushes(struct work_struct *ws)
586{
587 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
588 struct md_rdev *rdev;
589
590 mddev->start_flush = ktime_get_boottime();
591 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
592 atomic_set(&mddev->flush_pending, 1);
593 rcu_read_lock();
594 rdev_for_each_rcu(rdev, mddev)
595 if (rdev->raid_disk >= 0 &&
596 !test_bit(Faulty, &rdev->flags)) {
597 /* Take two references, one is dropped
598 * when request finishes, one after
599 * we reclaim rcu_read_lock
600 */
601 struct bio *bi;
602 atomic_inc(&rdev->nr_pending);
603 atomic_inc(&rdev->nr_pending);
604 rcu_read_unlock();
605 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
606 bi->bi_end_io = md_end_flush;
607 bi->bi_private = rdev;
608 bio_set_dev(bi, rdev->bdev);
609 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
610 atomic_inc(&mddev->flush_pending);
611 submit_bio(bi);
612 rcu_read_lock();
613 rdev_dec_pending(rdev, mddev);
614 }
615 rcu_read_unlock();
616 if (atomic_dec_and_test(&mddev->flush_pending))
617 queue_work(md_wq, &mddev->flush_work);
618}
619
620static void md_submit_flush_data(struct work_struct *ws)
621{
622 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
623 struct bio *bio = mddev->flush_bio;
624
625 /*
626 * must reset flush_bio before calling into md_handle_request to avoid a
627 * deadlock, because other bios passed md_handle_request suspend check
628 * could wait for this and below md_handle_request could wait for those
629 * bios because of suspend check
630 */
631 mddev->last_flush = mddev->start_flush;
632 mddev->flush_bio = NULL;
633 wake_up(&mddev->sb_wait);
634
635 if (bio->bi_iter.bi_size == 0) {
636 /* an empty barrier - all done */
637 bio_endio(bio);
638 } else {
639 bio->bi_opf &= ~REQ_PREFLUSH;
640 md_handle_request(mddev, bio);
641 }
642}
643
644/*
645 * Manages consolidation of flushes and submitting any flushes needed for
646 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
647 * being finished in another context. Returns false if the flushing is
648 * complete but still needs the I/O portion of the bio to be processed.
649 */
650bool md_flush_request(struct mddev *mddev, struct bio *bio)
651{
652 ktime_t start = ktime_get_boottime();
653 spin_lock_irq(&mddev->lock);
654 wait_event_lock_irq(mddev->sb_wait,
655 !mddev->flush_bio ||
656 ktime_after(mddev->last_flush, start),
657 mddev->lock);
658 if (!ktime_after(mddev->last_flush, start)) {
659 WARN_ON(mddev->flush_bio);
660 mddev->flush_bio = bio;
661 bio = NULL;
662 }
663 spin_unlock_irq(&mddev->lock);
664
665 if (!bio) {
666 INIT_WORK(&mddev->flush_work, submit_flushes);
667 queue_work(md_wq, &mddev->flush_work);
668 } else {
669 /* flush was performed for some other bio while we waited. */
670 if (bio->bi_iter.bi_size == 0)
671 /* an empty barrier - all done */
672 bio_endio(bio);
673 else {
674 bio->bi_opf &= ~REQ_PREFLUSH;
675 return false;
676 }
677 }
678 return true;
679}
680EXPORT_SYMBOL(md_flush_request);
681
682static inline struct mddev *mddev_get(struct mddev *mddev)
683{
684 atomic_inc(&mddev->active);
685 return mddev;
686}
687
688static void mddev_delayed_delete(struct work_struct *ws);
689
690static void mddev_put(struct mddev *mddev)
691{
692 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
693 return;
694 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
695 mddev->ctime == 0 && !mddev->hold_active) {
696 /* Array is not configured at all, and not held active,
697 * so destroy it */
698 list_del_init(&mddev->all_mddevs);
699
700 /*
701 * Call queue_work inside the spinlock so that
702 * flush_workqueue() after mddev_find will succeed in waiting
703 * for the work to be done.
704 */
705 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
706 queue_work(md_misc_wq, &mddev->del_work);
707 }
708 spin_unlock(&all_mddevs_lock);
709}
710
711static void md_safemode_timeout(struct timer_list *t);
712
713void mddev_init(struct mddev *mddev)
714{
715 kobject_init(&mddev->kobj, &md_ktype);
716 mutex_init(&mddev->open_mutex);
717 mutex_init(&mddev->reconfig_mutex);
718 mutex_init(&mddev->bitmap_info.mutex);
719 INIT_LIST_HEAD(&mddev->disks);
720 INIT_LIST_HEAD(&mddev->all_mddevs);
721 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
722 atomic_set(&mddev->active, 1);
723 atomic_set(&mddev->openers, 0);
724 atomic_set(&mddev->active_io, 0);
725 spin_lock_init(&mddev->lock);
726 atomic_set(&mddev->flush_pending, 0);
727 init_waitqueue_head(&mddev->sb_wait);
728 init_waitqueue_head(&mddev->recovery_wait);
729 mddev->reshape_position = MaxSector;
730 mddev->reshape_backwards = 0;
731 mddev->last_sync_action = "none";
732 mddev->resync_min = 0;
733 mddev->resync_max = MaxSector;
734 mddev->level = LEVEL_NONE;
735}
736EXPORT_SYMBOL_GPL(mddev_init);
737
738static struct mddev *mddev_find(dev_t unit)
739{
740 struct mddev *mddev, *new = NULL;
741
742 if (unit && MAJOR(unit) != MD_MAJOR)
743 unit &= ~((1<<MdpMinorShift)-1);
744
745 retry:
746 spin_lock(&all_mddevs_lock);
747
748 if (unit) {
749 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
750 if (mddev->unit == unit) {
751 mddev_get(mddev);
752 spin_unlock(&all_mddevs_lock);
753 kfree(new);
754 return mddev;
755 }
756
757 if (new) {
758 list_add(&new->all_mddevs, &all_mddevs);
759 spin_unlock(&all_mddevs_lock);
760 new->hold_active = UNTIL_IOCTL;
761 return new;
762 }
763 } else if (new) {
764 /* find an unused unit number */
765 static int next_minor = 512;
766 int start = next_minor;
767 int is_free = 0;
768 int dev = 0;
769 while (!is_free) {
770 dev = MKDEV(MD_MAJOR, next_minor);
771 next_minor++;
772 if (next_minor > MINORMASK)
773 next_minor = 0;
774 if (next_minor == start) {
775 /* Oh dear, all in use. */
776 spin_unlock(&all_mddevs_lock);
777 kfree(new);
778 return NULL;
779 }
780
781 is_free = 1;
782 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
783 if (mddev->unit == dev) {
784 is_free = 0;
785 break;
786 }
787 }
788 new->unit = dev;
789 new->md_minor = MINOR(dev);
790 new->hold_active = UNTIL_STOP;
791 list_add(&new->all_mddevs, &all_mddevs);
792 spin_unlock(&all_mddevs_lock);
793 return new;
794 }
795 spin_unlock(&all_mddevs_lock);
796
797 new = kzalloc(sizeof(*new), GFP_KERNEL);
798 if (!new)
799 return NULL;
800
801 new->unit = unit;
802 if (MAJOR(unit) == MD_MAJOR)
803 new->md_minor = MINOR(unit);
804 else
805 new->md_minor = MINOR(unit) >> MdpMinorShift;
806
807 mddev_init(new);
808
809 goto retry;
810}
811
812static struct attribute_group md_redundancy_group;
813
814void mddev_unlock(struct mddev *mddev)
815{
816 if (mddev->to_remove) {
817 /* These cannot be removed under reconfig_mutex as
818 * an access to the files will try to take reconfig_mutex
819 * while holding the file unremovable, which leads to
820 * a deadlock.
821 * So hold set sysfs_active while the remove in happeing,
822 * and anything else which might set ->to_remove or my
823 * otherwise change the sysfs namespace will fail with
824 * -EBUSY if sysfs_active is still set.
825 * We set sysfs_active under reconfig_mutex and elsewhere
826 * test it under the same mutex to ensure its correct value
827 * is seen.
828 */
829 struct attribute_group *to_remove = mddev->to_remove;
830 mddev->to_remove = NULL;
831 mddev->sysfs_active = 1;
832 mutex_unlock(&mddev->reconfig_mutex);
833
834 if (mddev->kobj.sd) {
835 if (to_remove != &md_redundancy_group)
836 sysfs_remove_group(&mddev->kobj, to_remove);
837 if (mddev->pers == NULL ||
838 mddev->pers->sync_request == NULL) {
839 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
840 if (mddev->sysfs_action)
841 sysfs_put(mddev->sysfs_action);
842 mddev->sysfs_action = NULL;
843 }
844 }
845 mddev->sysfs_active = 0;
846 } else
847 mutex_unlock(&mddev->reconfig_mutex);
848
849 /* As we've dropped the mutex we need a spinlock to
850 * make sure the thread doesn't disappear
851 */
852 spin_lock(&pers_lock);
853 md_wakeup_thread(mddev->thread);
854 wake_up(&mddev->sb_wait);
855 spin_unlock(&pers_lock);
856}
857EXPORT_SYMBOL_GPL(mddev_unlock);
858
859struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
860{
861 struct md_rdev *rdev;
862
863 rdev_for_each_rcu(rdev, mddev)
864 if (rdev->desc_nr == nr)
865 return rdev;
866
867 return NULL;
868}
869EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
870
871static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
872{
873 struct md_rdev *rdev;
874
875 rdev_for_each(rdev, mddev)
876 if (rdev->bdev->bd_dev == dev)
877 return rdev;
878
879 return NULL;
880}
881
882struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
883{
884 struct md_rdev *rdev;
885
886 rdev_for_each_rcu(rdev, mddev)
887 if (rdev->bdev->bd_dev == dev)
888 return rdev;
889
890 return NULL;
891}
892EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
893
894static struct md_personality *find_pers(int level, char *clevel)
895{
896 struct md_personality *pers;
897 list_for_each_entry(pers, &pers_list, list) {
898 if (level != LEVEL_NONE && pers->level == level)
899 return pers;
900 if (strcmp(pers->name, clevel)==0)
901 return pers;
902 }
903 return NULL;
904}
905
906/* return the offset of the super block in 512byte sectors */
907static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
908{
909 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
910 return MD_NEW_SIZE_SECTORS(num_sectors);
911}
912
913static int alloc_disk_sb(struct md_rdev *rdev)
914{
915 rdev->sb_page = alloc_page(GFP_KERNEL);
916 if (!rdev->sb_page)
917 return -ENOMEM;
918 return 0;
919}
920
921void md_rdev_clear(struct md_rdev *rdev)
922{
923 if (rdev->sb_page) {
924 put_page(rdev->sb_page);
925 rdev->sb_loaded = 0;
926 rdev->sb_page = NULL;
927 rdev->sb_start = 0;
928 rdev->sectors = 0;
929 }
930 if (rdev->bb_page) {
931 put_page(rdev->bb_page);
932 rdev->bb_page = NULL;
933 }
934 badblocks_exit(&rdev->badblocks);
935}
936EXPORT_SYMBOL_GPL(md_rdev_clear);
937
938static void super_written(struct bio *bio)
939{
940 struct md_rdev *rdev = bio->bi_private;
941 struct mddev *mddev = rdev->mddev;
942
943 if (bio->bi_status) {
944 pr_err("md: super_written gets error=%d\n", bio->bi_status);
945 md_error(mddev, rdev);
946 if (!test_bit(Faulty, &rdev->flags)
947 && (bio->bi_opf & MD_FAILFAST)) {
948 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
949 set_bit(LastDev, &rdev->flags);
950 }
951 } else
952 clear_bit(LastDev, &rdev->flags);
953
954 if (atomic_dec_and_test(&mddev->pending_writes))
955 wake_up(&mddev->sb_wait);
956 rdev_dec_pending(rdev, mddev);
957 bio_put(bio);
958}
959
960void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
961 sector_t sector, int size, struct page *page)
962{
963 /* write first size bytes of page to sector of rdev
964 * Increment mddev->pending_writes before returning
965 * and decrement it on completion, waking up sb_wait
966 * if zero is reached.
967 * If an error occurred, call md_error
968 */
969 struct bio *bio;
970 int ff = 0;
971
972 if (!page)
973 return;
974
975 if (test_bit(Faulty, &rdev->flags))
976 return;
977
978 bio = md_bio_alloc_sync(mddev);
979
980 atomic_inc(&rdev->nr_pending);
981
982 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
983 bio->bi_iter.bi_sector = sector;
984 bio_add_page(bio, page, size, 0);
985 bio->bi_private = rdev;
986 bio->bi_end_io = super_written;
987
988 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
989 test_bit(FailFast, &rdev->flags) &&
990 !test_bit(LastDev, &rdev->flags))
991 ff = MD_FAILFAST;
992 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
993
994 atomic_inc(&mddev->pending_writes);
995 submit_bio(bio);
996}
997
998int md_super_wait(struct mddev *mddev)
999{
1000 /* wait for all superblock writes that were scheduled to complete */
1001 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1002 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1003 return -EAGAIN;
1004 return 0;
1005}
1006
1007int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1008 struct page *page, int op, int op_flags, bool metadata_op)
1009{
1010 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
1011 int ret;
1012
1013 if (metadata_op && rdev->meta_bdev)
1014 bio_set_dev(bio, rdev->meta_bdev);
1015 else
1016 bio_set_dev(bio, rdev->bdev);
1017 bio_set_op_attrs(bio, op, op_flags);
1018 if (metadata_op)
1019 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1020 else if (rdev->mddev->reshape_position != MaxSector &&
1021 (rdev->mddev->reshape_backwards ==
1022 (sector >= rdev->mddev->reshape_position)))
1023 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
1024 else
1025 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1026 bio_add_page(bio, page, size, 0);
1027
1028 submit_bio_wait(bio);
1029
1030 ret = !bio->bi_status;
1031 bio_put(bio);
1032 return ret;
1033}
1034EXPORT_SYMBOL_GPL(sync_page_io);
1035
1036static int read_disk_sb(struct md_rdev *rdev, int size)
1037{
1038 char b[BDEVNAME_SIZE];
1039
1040 if (rdev->sb_loaded)
1041 return 0;
1042
1043 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1044 goto fail;
1045 rdev->sb_loaded = 1;
1046 return 0;
1047
1048fail:
1049 pr_err("md: disabled device %s, could not read superblock.\n",
1050 bdevname(rdev->bdev,b));
1051 return -EINVAL;
1052}
1053
1054static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1055{
1056 return sb1->set_uuid0 == sb2->set_uuid0 &&
1057 sb1->set_uuid1 == sb2->set_uuid1 &&
1058 sb1->set_uuid2 == sb2->set_uuid2 &&
1059 sb1->set_uuid3 == sb2->set_uuid3;
1060}
1061
1062static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1063{
1064 int ret;
1065 mdp_super_t *tmp1, *tmp2;
1066
1067 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1068 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1069
1070 if (!tmp1 || !tmp2) {
1071 ret = 0;
1072 goto abort;
1073 }
1074
1075 *tmp1 = *sb1;
1076 *tmp2 = *sb2;
1077
1078 /*
1079 * nr_disks is not constant
1080 */
1081 tmp1->nr_disks = 0;
1082 tmp2->nr_disks = 0;
1083
1084 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1085abort:
1086 kfree(tmp1);
1087 kfree(tmp2);
1088 return ret;
1089}
1090
1091static u32 md_csum_fold(u32 csum)
1092{
1093 csum = (csum & 0xffff) + (csum >> 16);
1094 return (csum & 0xffff) + (csum >> 16);
1095}
1096
1097static unsigned int calc_sb_csum(mdp_super_t *sb)
1098{
1099 u64 newcsum = 0;
1100 u32 *sb32 = (u32*)sb;
1101 int i;
1102 unsigned int disk_csum, csum;
1103
1104 disk_csum = sb->sb_csum;
1105 sb->sb_csum = 0;
1106
1107 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1108 newcsum += sb32[i];
1109 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1110
1111#ifdef CONFIG_ALPHA
1112 /* This used to use csum_partial, which was wrong for several
1113 * reasons including that different results are returned on
1114 * different architectures. It isn't critical that we get exactly
1115 * the same return value as before (we always csum_fold before
1116 * testing, and that removes any differences). However as we
1117 * know that csum_partial always returned a 16bit value on
1118 * alphas, do a fold to maximise conformity to previous behaviour.
1119 */
1120 sb->sb_csum = md_csum_fold(disk_csum);
1121#else
1122 sb->sb_csum = disk_csum;
1123#endif
1124 return csum;
1125}
1126
1127/*
1128 * Handle superblock details.
1129 * We want to be able to handle multiple superblock formats
1130 * so we have a common interface to them all, and an array of
1131 * different handlers.
1132 * We rely on user-space to write the initial superblock, and support
1133 * reading and updating of superblocks.
1134 * Interface methods are:
1135 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1136 * loads and validates a superblock on dev.
1137 * if refdev != NULL, compare superblocks on both devices
1138 * Return:
1139 * 0 - dev has a superblock that is compatible with refdev
1140 * 1 - dev has a superblock that is compatible and newer than refdev
1141 * so dev should be used as the refdev in future
1142 * -EINVAL superblock incompatible or invalid
1143 * -othererror e.g. -EIO
1144 *
1145 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1146 * Verify that dev is acceptable into mddev.
1147 * The first time, mddev->raid_disks will be 0, and data from
1148 * dev should be merged in. Subsequent calls check that dev
1149 * is new enough. Return 0 or -EINVAL
1150 *
1151 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1152 * Update the superblock for rdev with data in mddev
1153 * This does not write to disc.
1154 *
1155 */
1156
1157struct super_type {
1158 char *name;
1159 struct module *owner;
1160 int (*load_super)(struct md_rdev *rdev,
1161 struct md_rdev *refdev,
1162 int minor_version);
1163 int (*validate_super)(struct mddev *mddev,
1164 struct md_rdev *rdev);
1165 void (*sync_super)(struct mddev *mddev,
1166 struct md_rdev *rdev);
1167 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1168 sector_t num_sectors);
1169 int (*allow_new_offset)(struct md_rdev *rdev,
1170 unsigned long long new_offset);
1171};
1172
1173/*
1174 * Check that the given mddev has no bitmap.
1175 *
1176 * This function is called from the run method of all personalities that do not
1177 * support bitmaps. It prints an error message and returns non-zero if mddev
1178 * has a bitmap. Otherwise, it returns 0.
1179 *
1180 */
1181int md_check_no_bitmap(struct mddev *mddev)
1182{
1183 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1184 return 0;
1185 pr_warn("%s: bitmaps are not supported for %s\n",
1186 mdname(mddev), mddev->pers->name);
1187 return 1;
1188}
1189EXPORT_SYMBOL(md_check_no_bitmap);
1190
1191/*
1192 * load_super for 0.90.0
1193 */
1194static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1195{
1196 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1197 mdp_super_t *sb;
1198 int ret;
1199 bool spare_disk = true;
1200
1201 /*
1202 * Calculate the position of the superblock (512byte sectors),
1203 * it's at the end of the disk.
1204 *
1205 * It also happens to be a multiple of 4Kb.
1206 */
1207 rdev->sb_start = calc_dev_sboffset(rdev);
1208
1209 ret = read_disk_sb(rdev, MD_SB_BYTES);
1210 if (ret)
1211 return ret;
1212
1213 ret = -EINVAL;
1214
1215 bdevname(rdev->bdev, b);
1216 sb = page_address(rdev->sb_page);
1217
1218 if (sb->md_magic != MD_SB_MAGIC) {
1219 pr_warn("md: invalid raid superblock magic on %s\n", b);
1220 goto abort;
1221 }
1222
1223 if (sb->major_version != 0 ||
1224 sb->minor_version < 90 ||
1225 sb->minor_version > 91) {
1226 pr_warn("Bad version number %d.%d on %s\n",
1227 sb->major_version, sb->minor_version, b);
1228 goto abort;
1229 }
1230
1231 if (sb->raid_disks <= 0)
1232 goto abort;
1233
1234 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1235 pr_warn("md: invalid superblock checksum on %s\n", b);
1236 goto abort;
1237 }
1238
1239 rdev->preferred_minor = sb->md_minor;
1240 rdev->data_offset = 0;
1241 rdev->new_data_offset = 0;
1242 rdev->sb_size = MD_SB_BYTES;
1243 rdev->badblocks.shift = -1;
1244
1245 if (sb->level == LEVEL_MULTIPATH)
1246 rdev->desc_nr = -1;
1247 else
1248 rdev->desc_nr = sb->this_disk.number;
1249
1250 /* not spare disk, or LEVEL_MULTIPATH */
1251 if (sb->level == LEVEL_MULTIPATH ||
1252 (rdev->desc_nr >= 0 &&
1253 rdev->desc_nr < MD_SB_DISKS &&
1254 sb->disks[rdev->desc_nr].state &
1255 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1256 spare_disk = false;
1257
1258 if (!refdev) {
1259 if (!spare_disk)
1260 ret = 1;
1261 else
1262 ret = 0;
1263 } else {
1264 __u64 ev1, ev2;
1265 mdp_super_t *refsb = page_address(refdev->sb_page);
1266 if (!md_uuid_equal(refsb, sb)) {
1267 pr_warn("md: %s has different UUID to %s\n",
1268 b, bdevname(refdev->bdev,b2));
1269 goto abort;
1270 }
1271 if (!md_sb_equal(refsb, sb)) {
1272 pr_warn("md: %s has same UUID but different superblock to %s\n",
1273 b, bdevname(refdev->bdev, b2));
1274 goto abort;
1275 }
1276 ev1 = md_event(sb);
1277 ev2 = md_event(refsb);
1278
1279 if (!spare_disk && ev1 > ev2)
1280 ret = 1;
1281 else
1282 ret = 0;
1283 }
1284 rdev->sectors = rdev->sb_start;
1285 /* Limit to 4TB as metadata cannot record more than that.
1286 * (not needed for Linear and RAID0 as metadata doesn't
1287 * record this size)
1288 */
1289 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1290 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1291
1292 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1293 /* "this cannot possibly happen" ... */
1294 ret = -EINVAL;
1295
1296 abort:
1297 return ret;
1298}
1299
1300/*
1301 * validate_super for 0.90.0
1302 */
1303static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1304{
1305 mdp_disk_t *desc;
1306 mdp_super_t *sb = page_address(rdev->sb_page);
1307 __u64 ev1 = md_event(sb);
1308
1309 rdev->raid_disk = -1;
1310 clear_bit(Faulty, &rdev->flags);
1311 clear_bit(In_sync, &rdev->flags);
1312 clear_bit(Bitmap_sync, &rdev->flags);
1313 clear_bit(WriteMostly, &rdev->flags);
1314
1315 if (mddev->raid_disks == 0) {
1316 mddev->major_version = 0;
1317 mddev->minor_version = sb->minor_version;
1318 mddev->patch_version = sb->patch_version;
1319 mddev->external = 0;
1320 mddev->chunk_sectors = sb->chunk_size >> 9;
1321 mddev->ctime = sb->ctime;
1322 mddev->utime = sb->utime;
1323 mddev->level = sb->level;
1324 mddev->clevel[0] = 0;
1325 mddev->layout = sb->layout;
1326 mddev->raid_disks = sb->raid_disks;
1327 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1328 mddev->events = ev1;
1329 mddev->bitmap_info.offset = 0;
1330 mddev->bitmap_info.space = 0;
1331 /* bitmap can use 60 K after the 4K superblocks */
1332 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1333 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1334 mddev->reshape_backwards = 0;
1335
1336 if (mddev->minor_version >= 91) {
1337 mddev->reshape_position = sb->reshape_position;
1338 mddev->delta_disks = sb->delta_disks;
1339 mddev->new_level = sb->new_level;
1340 mddev->new_layout = sb->new_layout;
1341 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1342 if (mddev->delta_disks < 0)
1343 mddev->reshape_backwards = 1;
1344 } else {
1345 mddev->reshape_position = MaxSector;
1346 mddev->delta_disks = 0;
1347 mddev->new_level = mddev->level;
1348 mddev->new_layout = mddev->layout;
1349 mddev->new_chunk_sectors = mddev->chunk_sectors;
1350 }
1351 if (mddev->level == 0)
1352 mddev->layout = -1;
1353
1354 if (sb->state & (1<<MD_SB_CLEAN))
1355 mddev->recovery_cp = MaxSector;
1356 else {
1357 if (sb->events_hi == sb->cp_events_hi &&
1358 sb->events_lo == sb->cp_events_lo) {
1359 mddev->recovery_cp = sb->recovery_cp;
1360 } else
1361 mddev->recovery_cp = 0;
1362 }
1363
1364 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1365 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1366 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1367 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1368
1369 mddev->max_disks = MD_SB_DISKS;
1370
1371 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1372 mddev->bitmap_info.file == NULL) {
1373 mddev->bitmap_info.offset =
1374 mddev->bitmap_info.default_offset;
1375 mddev->bitmap_info.space =
1376 mddev->bitmap_info.default_space;
1377 }
1378
1379 } else if (mddev->pers == NULL) {
1380 /* Insist on good event counter while assembling, except
1381 * for spares (which don't need an event count) */
1382 ++ev1;
1383 if (sb->disks[rdev->desc_nr].state & (
1384 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1385 if (ev1 < mddev->events)
1386 return -EINVAL;
1387 } else if (mddev->bitmap) {
1388 /* if adding to array with a bitmap, then we can accept an
1389 * older device ... but not too old.
1390 */
1391 if (ev1 < mddev->bitmap->events_cleared)
1392 return 0;
1393 if (ev1 < mddev->events)
1394 set_bit(Bitmap_sync, &rdev->flags);
1395 } else {
1396 if (ev1 < mddev->events)
1397 /* just a hot-add of a new device, leave raid_disk at -1 */
1398 return 0;
1399 }
1400
1401 if (mddev->level != LEVEL_MULTIPATH) {
1402 desc = sb->disks + rdev->desc_nr;
1403
1404 if (desc->state & (1<<MD_DISK_FAULTY))
1405 set_bit(Faulty, &rdev->flags);
1406 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1407 desc->raid_disk < mddev->raid_disks */) {
1408 set_bit(In_sync, &rdev->flags);
1409 rdev->raid_disk = desc->raid_disk;
1410 rdev->saved_raid_disk = desc->raid_disk;
1411 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1412 /* active but not in sync implies recovery up to
1413 * reshape position. We don't know exactly where
1414 * that is, so set to zero for now */
1415 if (mddev->minor_version >= 91) {
1416 rdev->recovery_offset = 0;
1417 rdev->raid_disk = desc->raid_disk;
1418 }
1419 }
1420 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1421 set_bit(WriteMostly, &rdev->flags);
1422 if (desc->state & (1<<MD_DISK_FAILFAST))
1423 set_bit(FailFast, &rdev->flags);
1424 } else /* MULTIPATH are always insync */
1425 set_bit(In_sync, &rdev->flags);
1426 return 0;
1427}
1428
1429/*
1430 * sync_super for 0.90.0
1431 */
1432static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1433{
1434 mdp_super_t *sb;
1435 struct md_rdev *rdev2;
1436 int next_spare = mddev->raid_disks;
1437
1438 /* make rdev->sb match mddev data..
1439 *
1440 * 1/ zero out disks
1441 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1442 * 3/ any empty disks < next_spare become removed
1443 *
1444 * disks[0] gets initialised to REMOVED because
1445 * we cannot be sure from other fields if it has
1446 * been initialised or not.
1447 */
1448 int i;
1449 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1450
1451 rdev->sb_size = MD_SB_BYTES;
1452
1453 sb = page_address(rdev->sb_page);
1454
1455 memset(sb, 0, sizeof(*sb));
1456
1457 sb->md_magic = MD_SB_MAGIC;
1458 sb->major_version = mddev->major_version;
1459 sb->patch_version = mddev->patch_version;
1460 sb->gvalid_words = 0; /* ignored */
1461 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1462 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1463 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1464 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1465
1466 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1467 sb->level = mddev->level;
1468 sb->size = mddev->dev_sectors / 2;
1469 sb->raid_disks = mddev->raid_disks;
1470 sb->md_minor = mddev->md_minor;
1471 sb->not_persistent = 0;
1472 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1473 sb->state = 0;
1474 sb->events_hi = (mddev->events>>32);
1475 sb->events_lo = (u32)mddev->events;
1476
1477 if (mddev->reshape_position == MaxSector)
1478 sb->minor_version = 90;
1479 else {
1480 sb->minor_version = 91;
1481 sb->reshape_position = mddev->reshape_position;
1482 sb->new_level = mddev->new_level;
1483 sb->delta_disks = mddev->delta_disks;
1484 sb->new_layout = mddev->new_layout;
1485 sb->new_chunk = mddev->new_chunk_sectors << 9;
1486 }
1487 mddev->minor_version = sb->minor_version;
1488 if (mddev->in_sync)
1489 {
1490 sb->recovery_cp = mddev->recovery_cp;
1491 sb->cp_events_hi = (mddev->events>>32);
1492 sb->cp_events_lo = (u32)mddev->events;
1493 if (mddev->recovery_cp == MaxSector)
1494 sb->state = (1<< MD_SB_CLEAN);
1495 } else
1496 sb->recovery_cp = 0;
1497
1498 sb->layout = mddev->layout;
1499 sb->chunk_size = mddev->chunk_sectors << 9;
1500
1501 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1502 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1503
1504 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1505 rdev_for_each(rdev2, mddev) {
1506 mdp_disk_t *d;
1507 int desc_nr;
1508 int is_active = test_bit(In_sync, &rdev2->flags);
1509
1510 if (rdev2->raid_disk >= 0 &&
1511 sb->minor_version >= 91)
1512 /* we have nowhere to store the recovery_offset,
1513 * but if it is not below the reshape_position,
1514 * we can piggy-back on that.
1515 */
1516 is_active = 1;
1517 if (rdev2->raid_disk < 0 ||
1518 test_bit(Faulty, &rdev2->flags))
1519 is_active = 0;
1520 if (is_active)
1521 desc_nr = rdev2->raid_disk;
1522 else
1523 desc_nr = next_spare++;
1524 rdev2->desc_nr = desc_nr;
1525 d = &sb->disks[rdev2->desc_nr];
1526 nr_disks++;
1527 d->number = rdev2->desc_nr;
1528 d->major = MAJOR(rdev2->bdev->bd_dev);
1529 d->minor = MINOR(rdev2->bdev->bd_dev);
1530 if (is_active)
1531 d->raid_disk = rdev2->raid_disk;
1532 else
1533 d->raid_disk = rdev2->desc_nr; /* compatibility */
1534 if (test_bit(Faulty, &rdev2->flags))
1535 d->state = (1<<MD_DISK_FAULTY);
1536 else if (is_active) {
1537 d->state = (1<<MD_DISK_ACTIVE);
1538 if (test_bit(In_sync, &rdev2->flags))
1539 d->state |= (1<<MD_DISK_SYNC);
1540 active++;
1541 working++;
1542 } else {
1543 d->state = 0;
1544 spare++;
1545 working++;
1546 }
1547 if (test_bit(WriteMostly, &rdev2->flags))
1548 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1549 if (test_bit(FailFast, &rdev2->flags))
1550 d->state |= (1<<MD_DISK_FAILFAST);
1551 }
1552 /* now set the "removed" and "faulty" bits on any missing devices */
1553 for (i=0 ; i < mddev->raid_disks ; i++) {
1554 mdp_disk_t *d = &sb->disks[i];
1555 if (d->state == 0 && d->number == 0) {
1556 d->number = i;
1557 d->raid_disk = i;
1558 d->state = (1<<MD_DISK_REMOVED);
1559 d->state |= (1<<MD_DISK_FAULTY);
1560 failed++;
1561 }
1562 }
1563 sb->nr_disks = nr_disks;
1564 sb->active_disks = active;
1565 sb->working_disks = working;
1566 sb->failed_disks = failed;
1567 sb->spare_disks = spare;
1568
1569 sb->this_disk = sb->disks[rdev->desc_nr];
1570 sb->sb_csum = calc_sb_csum(sb);
1571}
1572
1573/*
1574 * rdev_size_change for 0.90.0
1575 */
1576static unsigned long long
1577super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1578{
1579 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1580 return 0; /* component must fit device */
1581 if (rdev->mddev->bitmap_info.offset)
1582 return 0; /* can't move bitmap */
1583 rdev->sb_start = calc_dev_sboffset(rdev);
1584 if (!num_sectors || num_sectors > rdev->sb_start)
1585 num_sectors = rdev->sb_start;
1586 /* Limit to 4TB as metadata cannot record more than that.
1587 * 4TB == 2^32 KB, or 2*2^32 sectors.
1588 */
1589 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1590 num_sectors = (sector_t)(2ULL << 32) - 2;
1591 do {
1592 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1593 rdev->sb_page);
1594 } while (md_super_wait(rdev->mddev) < 0);
1595 return num_sectors;
1596}
1597
1598static int
1599super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1600{
1601 /* non-zero offset changes not possible with v0.90 */
1602 return new_offset == 0;
1603}
1604
1605/*
1606 * version 1 superblock
1607 */
1608
1609static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1610{
1611 __le32 disk_csum;
1612 u32 csum;
1613 unsigned long long newcsum;
1614 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1615 __le32 *isuper = (__le32*)sb;
1616
1617 disk_csum = sb->sb_csum;
1618 sb->sb_csum = 0;
1619 newcsum = 0;
1620 for (; size >= 4; size -= 4)
1621 newcsum += le32_to_cpu(*isuper++);
1622
1623 if (size == 2)
1624 newcsum += le16_to_cpu(*(__le16*) isuper);
1625
1626 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1627 sb->sb_csum = disk_csum;
1628 return cpu_to_le32(csum);
1629}
1630
1631static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1632{
1633 struct mdp_superblock_1 *sb;
1634 int ret;
1635 sector_t sb_start;
1636 sector_t sectors;
1637 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1638 int bmask;
1639 bool spare_disk = true;
1640
1641 /*
1642 * Calculate the position of the superblock in 512byte sectors.
1643 * It is always aligned to a 4K boundary and
1644 * depeding on minor_version, it can be:
1645 * 0: At least 8K, but less than 12K, from end of device
1646 * 1: At start of device
1647 * 2: 4K from start of device.
1648 */
1649 switch(minor_version) {
1650 case 0:
1651 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1652 sb_start -= 8*2;
1653 sb_start &= ~(sector_t)(4*2-1);
1654 break;
1655 case 1:
1656 sb_start = 0;
1657 break;
1658 case 2:
1659 sb_start = 8;
1660 break;
1661 default:
1662 return -EINVAL;
1663 }
1664 rdev->sb_start = sb_start;
1665
1666 /* superblock is rarely larger than 1K, but it can be larger,
1667 * and it is safe to read 4k, so we do that
1668 */
1669 ret = read_disk_sb(rdev, 4096);
1670 if (ret) return ret;
1671
1672 sb = page_address(rdev->sb_page);
1673
1674 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1675 sb->major_version != cpu_to_le32(1) ||
1676 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1677 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1678 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1679 return -EINVAL;
1680
1681 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1682 pr_warn("md: invalid superblock checksum on %s\n",
1683 bdevname(rdev->bdev,b));
1684 return -EINVAL;
1685 }
1686 if (le64_to_cpu(sb->data_size) < 10) {
1687 pr_warn("md: data_size too small on %s\n",
1688 bdevname(rdev->bdev,b));
1689 return -EINVAL;
1690 }
1691 if (sb->pad0 ||
1692 sb->pad3[0] ||
1693 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1694 /* Some padding is non-zero, might be a new feature */
1695 return -EINVAL;
1696
1697 rdev->preferred_minor = 0xffff;
1698 rdev->data_offset = le64_to_cpu(sb->data_offset);
1699 rdev->new_data_offset = rdev->data_offset;
1700 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1701 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1702 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1703 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1704
1705 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1706 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1707 if (rdev->sb_size & bmask)
1708 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1709
1710 if (minor_version
1711 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1712 return -EINVAL;
1713 if (minor_version
1714 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1715 return -EINVAL;
1716
1717 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1718 rdev->desc_nr = -1;
1719 else
1720 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1721
1722 if (!rdev->bb_page) {
1723 rdev->bb_page = alloc_page(GFP_KERNEL);
1724 if (!rdev->bb_page)
1725 return -ENOMEM;
1726 }
1727 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1728 rdev->badblocks.count == 0) {
1729 /* need to load the bad block list.
1730 * Currently we limit it to one page.
1731 */
1732 s32 offset;
1733 sector_t bb_sector;
1734 __le64 *bbp;
1735 int i;
1736 int sectors = le16_to_cpu(sb->bblog_size);
1737 if (sectors > (PAGE_SIZE / 512))
1738 return -EINVAL;
1739 offset = le32_to_cpu(sb->bblog_offset);
1740 if (offset == 0)
1741 return -EINVAL;
1742 bb_sector = (long long)offset;
1743 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1744 rdev->bb_page, REQ_OP_READ, 0, true))
1745 return -EIO;
1746 bbp = (__le64 *)page_address(rdev->bb_page);
1747 rdev->badblocks.shift = sb->bblog_shift;
1748 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1749 u64 bb = le64_to_cpu(*bbp);
1750 int count = bb & (0x3ff);
1751 u64 sector = bb >> 10;
1752 sector <<= sb->bblog_shift;
1753 count <<= sb->bblog_shift;
1754 if (bb + 1 == 0)
1755 break;
1756 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1757 return -EINVAL;
1758 }
1759 } else if (sb->bblog_offset != 0)
1760 rdev->badblocks.shift = 0;
1761
1762 if ((le32_to_cpu(sb->feature_map) &
1763 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1764 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1765 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1766 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1767 }
1768
1769 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1770 sb->level != 0)
1771 return -EINVAL;
1772
1773 /* not spare disk, or LEVEL_MULTIPATH */
1774 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1775 (rdev->desc_nr >= 0 &&
1776 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1777 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1778 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1779 spare_disk = false;
1780
1781 if (!refdev) {
1782 if (!spare_disk)
1783 ret = 1;
1784 else
1785 ret = 0;
1786 } else {
1787 __u64 ev1, ev2;
1788 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1789
1790 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1791 sb->level != refsb->level ||
1792 sb->layout != refsb->layout ||
1793 sb->chunksize != refsb->chunksize) {
1794 pr_warn("md: %s has strangely different superblock to %s\n",
1795 bdevname(rdev->bdev,b),
1796 bdevname(refdev->bdev,b2));
1797 return -EINVAL;
1798 }
1799 ev1 = le64_to_cpu(sb->events);
1800 ev2 = le64_to_cpu(refsb->events);
1801
1802 if (!spare_disk && ev1 > ev2)
1803 ret = 1;
1804 else
1805 ret = 0;
1806 }
1807 if (minor_version) {
1808 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1809 sectors -= rdev->data_offset;
1810 } else
1811 sectors = rdev->sb_start;
1812 if (sectors < le64_to_cpu(sb->data_size))
1813 return -EINVAL;
1814 rdev->sectors = le64_to_cpu(sb->data_size);
1815 return ret;
1816}
1817
1818static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1819{
1820 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1821 __u64 ev1 = le64_to_cpu(sb->events);
1822
1823 rdev->raid_disk = -1;
1824 clear_bit(Faulty, &rdev->flags);
1825 clear_bit(In_sync, &rdev->flags);
1826 clear_bit(Bitmap_sync, &rdev->flags);
1827 clear_bit(WriteMostly, &rdev->flags);
1828
1829 if (mddev->raid_disks == 0) {
1830 mddev->major_version = 1;
1831 mddev->patch_version = 0;
1832 mddev->external = 0;
1833 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1834 mddev->ctime = le64_to_cpu(sb->ctime);
1835 mddev->utime = le64_to_cpu(sb->utime);
1836 mddev->level = le32_to_cpu(sb->level);
1837 mddev->clevel[0] = 0;
1838 mddev->layout = le32_to_cpu(sb->layout);
1839 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1840 mddev->dev_sectors = le64_to_cpu(sb->size);
1841 mddev->events = ev1;
1842 mddev->bitmap_info.offset = 0;
1843 mddev->bitmap_info.space = 0;
1844 /* Default location for bitmap is 1K after superblock
1845 * using 3K - total of 4K
1846 */
1847 mddev->bitmap_info.default_offset = 1024 >> 9;
1848 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1849 mddev->reshape_backwards = 0;
1850
1851 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1852 memcpy(mddev->uuid, sb->set_uuid, 16);
1853
1854 mddev->max_disks = (4096-256)/2;
1855
1856 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1857 mddev->bitmap_info.file == NULL) {
1858 mddev->bitmap_info.offset =
1859 (__s32)le32_to_cpu(sb->bitmap_offset);
1860 /* Metadata doesn't record how much space is available.
1861 * For 1.0, we assume we can use up to the superblock
1862 * if before, else to 4K beyond superblock.
1863 * For others, assume no change is possible.
1864 */
1865 if (mddev->minor_version > 0)
1866 mddev->bitmap_info.space = 0;
1867 else if (mddev->bitmap_info.offset > 0)
1868 mddev->bitmap_info.space =
1869 8 - mddev->bitmap_info.offset;
1870 else
1871 mddev->bitmap_info.space =
1872 -mddev->bitmap_info.offset;
1873 }
1874
1875 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1876 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1877 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1878 mddev->new_level = le32_to_cpu(sb->new_level);
1879 mddev->new_layout = le32_to_cpu(sb->new_layout);
1880 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1881 if (mddev->delta_disks < 0 ||
1882 (mddev->delta_disks == 0 &&
1883 (le32_to_cpu(sb->feature_map)
1884 & MD_FEATURE_RESHAPE_BACKWARDS)))
1885 mddev->reshape_backwards = 1;
1886 } else {
1887 mddev->reshape_position = MaxSector;
1888 mddev->delta_disks = 0;
1889 mddev->new_level = mddev->level;
1890 mddev->new_layout = mddev->layout;
1891 mddev->new_chunk_sectors = mddev->chunk_sectors;
1892 }
1893
1894 if (mddev->level == 0 &&
1895 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1896 mddev->layout = -1;
1897
1898 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1899 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1900
1901 if (le32_to_cpu(sb->feature_map) &
1902 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1903 if (le32_to_cpu(sb->feature_map) &
1904 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1905 return -EINVAL;
1906 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1907 (le32_to_cpu(sb->feature_map) &
1908 MD_FEATURE_MULTIPLE_PPLS))
1909 return -EINVAL;
1910 set_bit(MD_HAS_PPL, &mddev->flags);
1911 }
1912 } else if (mddev->pers == NULL) {
1913 /* Insist of good event counter while assembling, except for
1914 * spares (which don't need an event count) */
1915 ++ev1;
1916 if (rdev->desc_nr >= 0 &&
1917 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1918 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1919 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1920 if (ev1 < mddev->events)
1921 return -EINVAL;
1922 } else if (mddev->bitmap) {
1923 /* If adding to array with a bitmap, then we can accept an
1924 * older device, but not too old.
1925 */
1926 if (ev1 < mddev->bitmap->events_cleared)
1927 return 0;
1928 if (ev1 < mddev->events)
1929 set_bit(Bitmap_sync, &rdev->flags);
1930 } else {
1931 if (ev1 < mddev->events)
1932 /* just a hot-add of a new device, leave raid_disk at -1 */
1933 return 0;
1934 }
1935 if (mddev->level != LEVEL_MULTIPATH) {
1936 int role;
1937 if (rdev->desc_nr < 0 ||
1938 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1939 role = MD_DISK_ROLE_SPARE;
1940 rdev->desc_nr = -1;
1941 } else
1942 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1943 switch(role) {
1944 case MD_DISK_ROLE_SPARE: /* spare */
1945 break;
1946 case MD_DISK_ROLE_FAULTY: /* faulty */
1947 set_bit(Faulty, &rdev->flags);
1948 break;
1949 case MD_DISK_ROLE_JOURNAL: /* journal device */
1950 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1951 /* journal device without journal feature */
1952 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1953 return -EINVAL;
1954 }
1955 set_bit(Journal, &rdev->flags);
1956 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1957 rdev->raid_disk = 0;
1958 break;
1959 default:
1960 rdev->saved_raid_disk = role;
1961 if ((le32_to_cpu(sb->feature_map) &
1962 MD_FEATURE_RECOVERY_OFFSET)) {
1963 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1964 if (!(le32_to_cpu(sb->feature_map) &
1965 MD_FEATURE_RECOVERY_BITMAP))
1966 rdev->saved_raid_disk = -1;
1967 } else {
1968 /*
1969 * If the array is FROZEN, then the device can't
1970 * be in_sync with rest of array.
1971 */
1972 if (!test_bit(MD_RECOVERY_FROZEN,
1973 &mddev->recovery))
1974 set_bit(In_sync, &rdev->flags);
1975 }
1976 rdev->raid_disk = role;
1977 break;
1978 }
1979 if (sb->devflags & WriteMostly1)
1980 set_bit(WriteMostly, &rdev->flags);
1981 if (sb->devflags & FailFast1)
1982 set_bit(FailFast, &rdev->flags);
1983 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1984 set_bit(Replacement, &rdev->flags);
1985 } else /* MULTIPATH are always insync */
1986 set_bit(In_sync, &rdev->flags);
1987
1988 return 0;
1989}
1990
1991static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1992{
1993 struct mdp_superblock_1 *sb;
1994 struct md_rdev *rdev2;
1995 int max_dev, i;
1996 /* make rdev->sb match mddev and rdev data. */
1997
1998 sb = page_address(rdev->sb_page);
1999
2000 sb->feature_map = 0;
2001 sb->pad0 = 0;
2002 sb->recovery_offset = cpu_to_le64(0);
2003 memset(sb->pad3, 0, sizeof(sb->pad3));
2004
2005 sb->utime = cpu_to_le64((__u64)mddev->utime);
2006 sb->events = cpu_to_le64(mddev->events);
2007 if (mddev->in_sync)
2008 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2009 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2010 sb->resync_offset = cpu_to_le64(MaxSector);
2011 else
2012 sb->resync_offset = cpu_to_le64(0);
2013
2014 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2015
2016 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2017 sb->size = cpu_to_le64(mddev->dev_sectors);
2018 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2019 sb->level = cpu_to_le32(mddev->level);
2020 sb->layout = cpu_to_le32(mddev->layout);
2021 if (test_bit(FailFast, &rdev->flags))
2022 sb->devflags |= FailFast1;
2023 else
2024 sb->devflags &= ~FailFast1;
2025
2026 if (test_bit(WriteMostly, &rdev->flags))
2027 sb->devflags |= WriteMostly1;
2028 else
2029 sb->devflags &= ~WriteMostly1;
2030 sb->data_offset = cpu_to_le64(rdev->data_offset);
2031 sb->data_size = cpu_to_le64(rdev->sectors);
2032
2033 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2034 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2035 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2036 }
2037
2038 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2039 !test_bit(In_sync, &rdev->flags)) {
2040 sb->feature_map |=
2041 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2042 sb->recovery_offset =
2043 cpu_to_le64(rdev->recovery_offset);
2044 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2045 sb->feature_map |=
2046 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2047 }
2048 /* Note: recovery_offset and journal_tail share space */
2049 if (test_bit(Journal, &rdev->flags))
2050 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2051 if (test_bit(Replacement, &rdev->flags))
2052 sb->feature_map |=
2053 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2054
2055 if (mddev->reshape_position != MaxSector) {
2056 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2057 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2058 sb->new_layout = cpu_to_le32(mddev->new_layout);
2059 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2060 sb->new_level = cpu_to_le32(mddev->new_level);
2061 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2062 if (mddev->delta_disks == 0 &&
2063 mddev->reshape_backwards)
2064 sb->feature_map
2065 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2066 if (rdev->new_data_offset != rdev->data_offset) {
2067 sb->feature_map
2068 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2069 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2070 - rdev->data_offset));
2071 }
2072 }
2073
2074 if (mddev_is_clustered(mddev))
2075 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2076
2077 if (rdev->badblocks.count == 0)
2078 /* Nothing to do for bad blocks*/ ;
2079 else if (sb->bblog_offset == 0)
2080 /* Cannot record bad blocks on this device */
2081 md_error(mddev, rdev);
2082 else {
2083 struct badblocks *bb = &rdev->badblocks;
2084 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2085 u64 *p = bb->page;
2086 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2087 if (bb->changed) {
2088 unsigned seq;
2089
2090retry:
2091 seq = read_seqbegin(&bb->lock);
2092
2093 memset(bbp, 0xff, PAGE_SIZE);
2094
2095 for (i = 0 ; i < bb->count ; i++) {
2096 u64 internal_bb = p[i];
2097 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2098 | BB_LEN(internal_bb));
2099 bbp[i] = cpu_to_le64(store_bb);
2100 }
2101 bb->changed = 0;
2102 if (read_seqretry(&bb->lock, seq))
2103 goto retry;
2104
2105 bb->sector = (rdev->sb_start +
2106 (int)le32_to_cpu(sb->bblog_offset));
2107 bb->size = le16_to_cpu(sb->bblog_size);
2108 }
2109 }
2110
2111 max_dev = 0;
2112 rdev_for_each(rdev2, mddev)
2113 if (rdev2->desc_nr+1 > max_dev)
2114 max_dev = rdev2->desc_nr+1;
2115
2116 if (max_dev > le32_to_cpu(sb->max_dev)) {
2117 int bmask;
2118 sb->max_dev = cpu_to_le32(max_dev);
2119 rdev->sb_size = max_dev * 2 + 256;
2120 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2121 if (rdev->sb_size & bmask)
2122 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2123 } else
2124 max_dev = le32_to_cpu(sb->max_dev);
2125
2126 for (i=0; i<max_dev;i++)
2127 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2128
2129 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2130 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2131
2132 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2133 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2134 sb->feature_map |=
2135 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2136 else
2137 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2138 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2139 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2140 }
2141
2142 rdev_for_each(rdev2, mddev) {
2143 i = rdev2->desc_nr;
2144 if (test_bit(Faulty, &rdev2->flags))
2145 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2146 else if (test_bit(In_sync, &rdev2->flags))
2147 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2148 else if (test_bit(Journal, &rdev2->flags))
2149 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2150 else if (rdev2->raid_disk >= 0)
2151 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2152 else
2153 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2154 }
2155
2156 sb->sb_csum = calc_sb_1_csum(sb);
2157}
2158
2159static unsigned long long
2160super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2161{
2162 struct mdp_superblock_1 *sb;
2163 sector_t max_sectors;
2164 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2165 return 0; /* component must fit device */
2166 if (rdev->data_offset != rdev->new_data_offset)
2167 return 0; /* too confusing */
2168 if (rdev->sb_start < rdev->data_offset) {
2169 /* minor versions 1 and 2; superblock before data */
2170 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2171 max_sectors -= rdev->data_offset;
2172 if (!num_sectors || num_sectors > max_sectors)
2173 num_sectors = max_sectors;
2174 } else if (rdev->mddev->bitmap_info.offset) {
2175 /* minor version 0 with bitmap we can't move */
2176 return 0;
2177 } else {
2178 /* minor version 0; superblock after data */
2179 sector_t sb_start;
2180 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
2181 sb_start &= ~(sector_t)(4*2 - 1);
2182 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
2183 if (!num_sectors || num_sectors > max_sectors)
2184 num_sectors = max_sectors;
2185 rdev->sb_start = sb_start;
2186 }
2187 sb = page_address(rdev->sb_page);
2188 sb->data_size = cpu_to_le64(num_sectors);
2189 sb->super_offset = cpu_to_le64(rdev->sb_start);
2190 sb->sb_csum = calc_sb_1_csum(sb);
2191 do {
2192 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2193 rdev->sb_page);
2194 } while (md_super_wait(rdev->mddev) < 0);
2195 return num_sectors;
2196
2197}
2198
2199static int
2200super_1_allow_new_offset(struct md_rdev *rdev,
2201 unsigned long long new_offset)
2202{
2203 /* All necessary checks on new >= old have been done */
2204 struct bitmap *bitmap;
2205 if (new_offset >= rdev->data_offset)
2206 return 1;
2207
2208 /* with 1.0 metadata, there is no metadata to tread on
2209 * so we can always move back */
2210 if (rdev->mddev->minor_version == 0)
2211 return 1;
2212
2213 /* otherwise we must be sure not to step on
2214 * any metadata, so stay:
2215 * 36K beyond start of superblock
2216 * beyond end of badblocks
2217 * beyond write-intent bitmap
2218 */
2219 if (rdev->sb_start + (32+4)*2 > new_offset)
2220 return 0;
2221 bitmap = rdev->mddev->bitmap;
2222 if (bitmap && !rdev->mddev->bitmap_info.file &&
2223 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2224 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2225 return 0;
2226 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2227 return 0;
2228
2229 return 1;
2230}
2231
2232static struct super_type super_types[] = {
2233 [0] = {
2234 .name = "0.90.0",
2235 .owner = THIS_MODULE,
2236 .load_super = super_90_load,
2237 .validate_super = super_90_validate,
2238 .sync_super = super_90_sync,
2239 .rdev_size_change = super_90_rdev_size_change,
2240 .allow_new_offset = super_90_allow_new_offset,
2241 },
2242 [1] = {
2243 .name = "md-1",
2244 .owner = THIS_MODULE,
2245 .load_super = super_1_load,
2246 .validate_super = super_1_validate,
2247 .sync_super = super_1_sync,
2248 .rdev_size_change = super_1_rdev_size_change,
2249 .allow_new_offset = super_1_allow_new_offset,
2250 },
2251};
2252
2253static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2254{
2255 if (mddev->sync_super) {
2256 mddev->sync_super(mddev, rdev);
2257 return;
2258 }
2259
2260 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2261
2262 super_types[mddev->major_version].sync_super(mddev, rdev);
2263}
2264
2265static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2266{
2267 struct md_rdev *rdev, *rdev2;
2268
2269 rcu_read_lock();
2270 rdev_for_each_rcu(rdev, mddev1) {
2271 if (test_bit(Faulty, &rdev->flags) ||
2272 test_bit(Journal, &rdev->flags) ||
2273 rdev->raid_disk == -1)
2274 continue;
2275 rdev_for_each_rcu(rdev2, mddev2) {
2276 if (test_bit(Faulty, &rdev2->flags) ||
2277 test_bit(Journal, &rdev2->flags) ||
2278 rdev2->raid_disk == -1)
2279 continue;
2280 if (rdev->bdev->bd_contains ==
2281 rdev2->bdev->bd_contains) {
2282 rcu_read_unlock();
2283 return 1;
2284 }
2285 }
2286 }
2287 rcu_read_unlock();
2288 return 0;
2289}
2290
2291static LIST_HEAD(pending_raid_disks);
2292
2293/*
2294 * Try to register data integrity profile for an mddev
2295 *
2296 * This is called when an array is started and after a disk has been kicked
2297 * from the array. It only succeeds if all working and active component devices
2298 * are integrity capable with matching profiles.
2299 */
2300int md_integrity_register(struct mddev *mddev)
2301{
2302 struct md_rdev *rdev, *reference = NULL;
2303
2304 if (list_empty(&mddev->disks))
2305 return 0; /* nothing to do */
2306 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2307 return 0; /* shouldn't register, or already is */
2308 rdev_for_each(rdev, mddev) {
2309 /* skip spares and non-functional disks */
2310 if (test_bit(Faulty, &rdev->flags))
2311 continue;
2312 if (rdev->raid_disk < 0)
2313 continue;
2314 if (!reference) {
2315 /* Use the first rdev as the reference */
2316 reference = rdev;
2317 continue;
2318 }
2319 /* does this rdev's profile match the reference profile? */
2320 if (blk_integrity_compare(reference->bdev->bd_disk,
2321 rdev->bdev->bd_disk) < 0)
2322 return -EINVAL;
2323 }
2324 if (!reference || !bdev_get_integrity(reference->bdev))
2325 return 0;
2326 /*
2327 * All component devices are integrity capable and have matching
2328 * profiles, register the common profile for the md device.
2329 */
2330 blk_integrity_register(mddev->gendisk,
2331 bdev_get_integrity(reference->bdev));
2332
2333 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2334 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2335 pr_err("md: failed to create integrity pool for %s\n",
2336 mdname(mddev));
2337 return -EINVAL;
2338 }
2339 return 0;
2340}
2341EXPORT_SYMBOL(md_integrity_register);
2342
2343/*
2344 * Attempt to add an rdev, but only if it is consistent with the current
2345 * integrity profile
2346 */
2347int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2348{
2349 struct blk_integrity *bi_mddev;
2350 char name[BDEVNAME_SIZE];
2351
2352 if (!mddev->gendisk)
2353 return 0;
2354
2355 bi_mddev = blk_get_integrity(mddev->gendisk);
2356
2357 if (!bi_mddev) /* nothing to do */
2358 return 0;
2359
2360 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2361 pr_err("%s: incompatible integrity profile for %s\n",
2362 mdname(mddev), bdevname(rdev->bdev, name));
2363 return -ENXIO;
2364 }
2365
2366 return 0;
2367}
2368EXPORT_SYMBOL(md_integrity_add_rdev);
2369
2370static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2371{
2372 char b[BDEVNAME_SIZE];
2373 struct kobject *ko;
2374 int err;
2375
2376 /* prevent duplicates */
2377 if (find_rdev(mddev, rdev->bdev->bd_dev))
2378 return -EEXIST;
2379
2380 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2381 mddev->pers)
2382 return -EROFS;
2383
2384 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2385 if (!test_bit(Journal, &rdev->flags) &&
2386 rdev->sectors &&
2387 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2388 if (mddev->pers) {
2389 /* Cannot change size, so fail
2390 * If mddev->level <= 0, then we don't care
2391 * about aligning sizes (e.g. linear)
2392 */
2393 if (mddev->level > 0)
2394 return -ENOSPC;
2395 } else
2396 mddev->dev_sectors = rdev->sectors;
2397 }
2398
2399 /* Verify rdev->desc_nr is unique.
2400 * If it is -1, assign a free number, else
2401 * check number is not in use
2402 */
2403 rcu_read_lock();
2404 if (rdev->desc_nr < 0) {
2405 int choice = 0;
2406 if (mddev->pers)
2407 choice = mddev->raid_disks;
2408 while (md_find_rdev_nr_rcu(mddev, choice))
2409 choice++;
2410 rdev->desc_nr = choice;
2411 } else {
2412 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2413 rcu_read_unlock();
2414 return -EBUSY;
2415 }
2416 }
2417 rcu_read_unlock();
2418 if (!test_bit(Journal, &rdev->flags) &&
2419 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2420 pr_warn("md: %s: array is limited to %d devices\n",
2421 mdname(mddev), mddev->max_disks);
2422 return -EBUSY;
2423 }
2424 bdevname(rdev->bdev,b);
2425 strreplace(b, '/', '!');
2426
2427 rdev->mddev = mddev;
2428 pr_debug("md: bind<%s>\n", b);
2429
2430 if (mddev->raid_disks)
2431 mddev_create_serial_pool(mddev, rdev, false);
2432
2433 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2434 goto fail;
2435
2436 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2437 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2438 /* failure here is OK */;
2439 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2440
2441 list_add_rcu(&rdev->same_set, &mddev->disks);
2442 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2443
2444 /* May as well allow recovery to be retried once */
2445 mddev->recovery_disabled++;
2446
2447 return 0;
2448
2449 fail:
2450 pr_warn("md: failed to register dev-%s for %s\n",
2451 b, mdname(mddev));
2452 return err;
2453}
2454
2455static void md_delayed_delete(struct work_struct *ws)
2456{
2457 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2458 kobject_del(&rdev->kobj);
2459 kobject_put(&rdev->kobj);
2460}
2461
2462static void unbind_rdev_from_array(struct md_rdev *rdev)
2463{
2464 char b[BDEVNAME_SIZE];
2465
2466 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2467 list_del_rcu(&rdev->same_set);
2468 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2469 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2470 rdev->mddev = NULL;
2471 sysfs_remove_link(&rdev->kobj, "block");
2472 sysfs_put(rdev->sysfs_state);
2473 rdev->sysfs_state = NULL;
2474 rdev->badblocks.count = 0;
2475 /* We need to delay this, otherwise we can deadlock when
2476 * writing to 'remove' to "dev/state". We also need
2477 * to delay it due to rcu usage.
2478 */
2479 synchronize_rcu();
2480 INIT_WORK(&rdev->del_work, md_delayed_delete);
2481 kobject_get(&rdev->kobj);
2482 queue_work(md_misc_wq, &rdev->del_work);
2483}
2484
2485/*
2486 * prevent the device from being mounted, repartitioned or
2487 * otherwise reused by a RAID array (or any other kernel
2488 * subsystem), by bd_claiming the device.
2489 */
2490static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2491{
2492 int err = 0;
2493 struct block_device *bdev;
2494 char b[BDEVNAME_SIZE];
2495
2496 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2497 shared ? (struct md_rdev *)lock_rdev : rdev);
2498 if (IS_ERR(bdev)) {
2499 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2500 return PTR_ERR(bdev);
2501 }
2502 rdev->bdev = bdev;
2503 return err;
2504}
2505
2506static void unlock_rdev(struct md_rdev *rdev)
2507{
2508 struct block_device *bdev = rdev->bdev;
2509 rdev->bdev = NULL;
2510 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2511}
2512
2513void md_autodetect_dev(dev_t dev);
2514
2515static void export_rdev(struct md_rdev *rdev)
2516{
2517 char b[BDEVNAME_SIZE];
2518
2519 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2520 md_rdev_clear(rdev);
2521#ifndef MODULE
2522 if (test_bit(AutoDetected, &rdev->flags))
2523 md_autodetect_dev(rdev->bdev->bd_dev);
2524#endif
2525 unlock_rdev(rdev);
2526 kobject_put(&rdev->kobj);
2527}
2528
2529void md_kick_rdev_from_array(struct md_rdev *rdev)
2530{
2531 unbind_rdev_from_array(rdev);
2532 export_rdev(rdev);
2533}
2534EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2535
2536static void export_array(struct mddev *mddev)
2537{
2538 struct md_rdev *rdev;
2539
2540 while (!list_empty(&mddev->disks)) {
2541 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2542 same_set);
2543 md_kick_rdev_from_array(rdev);
2544 }
2545 mddev->raid_disks = 0;
2546 mddev->major_version = 0;
2547}
2548
2549static bool set_in_sync(struct mddev *mddev)
2550{
2551 lockdep_assert_held(&mddev->lock);
2552 if (!mddev->in_sync) {
2553 mddev->sync_checkers++;
2554 spin_unlock(&mddev->lock);
2555 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2556 spin_lock(&mddev->lock);
2557 if (!mddev->in_sync &&
2558 percpu_ref_is_zero(&mddev->writes_pending)) {
2559 mddev->in_sync = 1;
2560 /*
2561 * Ensure ->in_sync is visible before we clear
2562 * ->sync_checkers.
2563 */
2564 smp_mb();
2565 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2566 sysfs_notify_dirent_safe(mddev->sysfs_state);
2567 }
2568 if (--mddev->sync_checkers == 0)
2569 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2570 }
2571 if (mddev->safemode == 1)
2572 mddev->safemode = 0;
2573 return mddev->in_sync;
2574}
2575
2576static void sync_sbs(struct mddev *mddev, int nospares)
2577{
2578 /* Update each superblock (in-memory image), but
2579 * if we are allowed to, skip spares which already
2580 * have the right event counter, or have one earlier
2581 * (which would mean they aren't being marked as dirty
2582 * with the rest of the array)
2583 */
2584 struct md_rdev *rdev;
2585 rdev_for_each(rdev, mddev) {
2586 if (rdev->sb_events == mddev->events ||
2587 (nospares &&
2588 rdev->raid_disk < 0 &&
2589 rdev->sb_events+1 == mddev->events)) {
2590 /* Don't update this superblock */
2591 rdev->sb_loaded = 2;
2592 } else {
2593 sync_super(mddev, rdev);
2594 rdev->sb_loaded = 1;
2595 }
2596 }
2597}
2598
2599static bool does_sb_need_changing(struct mddev *mddev)
2600{
2601 struct md_rdev *rdev;
2602 struct mdp_superblock_1 *sb;
2603 int role;
2604
2605 /* Find a good rdev */
2606 rdev_for_each(rdev, mddev)
2607 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2608 break;
2609
2610 /* No good device found. */
2611 if (!rdev)
2612 return false;
2613
2614 sb = page_address(rdev->sb_page);
2615 /* Check if a device has become faulty or a spare become active */
2616 rdev_for_each(rdev, mddev) {
2617 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2618 /* Device activated? */
2619 if (role == 0xffff && rdev->raid_disk >=0 &&
2620 !test_bit(Faulty, &rdev->flags))
2621 return true;
2622 /* Device turned faulty? */
2623 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2624 return true;
2625 }
2626
2627 /* Check if any mddev parameters have changed */
2628 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2629 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2630 (mddev->layout != le32_to_cpu(sb->layout)) ||
2631 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2632 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2633 return true;
2634
2635 return false;
2636}
2637
2638void md_update_sb(struct mddev *mddev, int force_change)
2639{
2640 struct md_rdev *rdev;
2641 int sync_req;
2642 int nospares = 0;
2643 int any_badblocks_changed = 0;
2644 int ret = -1;
2645
2646 if (mddev->ro) {
2647 if (force_change)
2648 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2649 return;
2650 }
2651
2652repeat:
2653 if (mddev_is_clustered(mddev)) {
2654 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2655 force_change = 1;
2656 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2657 nospares = 1;
2658 ret = md_cluster_ops->metadata_update_start(mddev);
2659 /* Has someone else has updated the sb */
2660 if (!does_sb_need_changing(mddev)) {
2661 if (ret == 0)
2662 md_cluster_ops->metadata_update_cancel(mddev);
2663 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2664 BIT(MD_SB_CHANGE_DEVS) |
2665 BIT(MD_SB_CHANGE_CLEAN));
2666 return;
2667 }
2668 }
2669
2670 /*
2671 * First make sure individual recovery_offsets are correct
2672 * curr_resync_completed can only be used during recovery.
2673 * During reshape/resync it might use array-addresses rather
2674 * that device addresses.
2675 */
2676 rdev_for_each(rdev, mddev) {
2677 if (rdev->raid_disk >= 0 &&
2678 mddev->delta_disks >= 0 &&
2679 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2680 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2681 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2682 !test_bit(Journal, &rdev->flags) &&
2683 !test_bit(In_sync, &rdev->flags) &&
2684 mddev->curr_resync_completed > rdev->recovery_offset)
2685 rdev->recovery_offset = mddev->curr_resync_completed;
2686
2687 }
2688 if (!mddev->persistent) {
2689 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2690 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2691 if (!mddev->external) {
2692 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2693 rdev_for_each(rdev, mddev) {
2694 if (rdev->badblocks.changed) {
2695 rdev->badblocks.changed = 0;
2696 ack_all_badblocks(&rdev->badblocks);
2697 md_error(mddev, rdev);
2698 }
2699 clear_bit(Blocked, &rdev->flags);
2700 clear_bit(BlockedBadBlocks, &rdev->flags);
2701 wake_up(&rdev->blocked_wait);
2702 }
2703 }
2704 wake_up(&mddev->sb_wait);
2705 return;
2706 }
2707
2708 spin_lock(&mddev->lock);
2709
2710 mddev->utime = ktime_get_real_seconds();
2711
2712 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2713 force_change = 1;
2714 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2715 /* just a clean<-> dirty transition, possibly leave spares alone,
2716 * though if events isn't the right even/odd, we will have to do
2717 * spares after all
2718 */
2719 nospares = 1;
2720 if (force_change)
2721 nospares = 0;
2722 if (mddev->degraded)
2723 /* If the array is degraded, then skipping spares is both
2724 * dangerous and fairly pointless.
2725 * Dangerous because a device that was removed from the array
2726 * might have a event_count that still looks up-to-date,
2727 * so it can be re-added without a resync.
2728 * Pointless because if there are any spares to skip,
2729 * then a recovery will happen and soon that array won't
2730 * be degraded any more and the spare can go back to sleep then.
2731 */
2732 nospares = 0;
2733
2734 sync_req = mddev->in_sync;
2735
2736 /* If this is just a dirty<->clean transition, and the array is clean
2737 * and 'events' is odd, we can roll back to the previous clean state */
2738 if (nospares
2739 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2740 && mddev->can_decrease_events
2741 && mddev->events != 1) {
2742 mddev->events--;
2743 mddev->can_decrease_events = 0;
2744 } else {
2745 /* otherwise we have to go forward and ... */
2746 mddev->events ++;
2747 mddev->can_decrease_events = nospares;
2748 }
2749
2750 /*
2751 * This 64-bit counter should never wrap.
2752 * Either we are in around ~1 trillion A.C., assuming
2753 * 1 reboot per second, or we have a bug...
2754 */
2755 WARN_ON(mddev->events == 0);
2756
2757 rdev_for_each(rdev, mddev) {
2758 if (rdev->badblocks.changed)
2759 any_badblocks_changed++;
2760 if (test_bit(Faulty, &rdev->flags))
2761 set_bit(FaultRecorded, &rdev->flags);
2762 }
2763
2764 sync_sbs(mddev, nospares);
2765 spin_unlock(&mddev->lock);
2766
2767 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2768 mdname(mddev), mddev->in_sync);
2769
2770 if (mddev->queue)
2771 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2772rewrite:
2773 md_bitmap_update_sb(mddev->bitmap);
2774 rdev_for_each(rdev, mddev) {
2775 char b[BDEVNAME_SIZE];
2776
2777 if (rdev->sb_loaded != 1)
2778 continue; /* no noise on spare devices */
2779
2780 if (!test_bit(Faulty, &rdev->flags)) {
2781 md_super_write(mddev,rdev,
2782 rdev->sb_start, rdev->sb_size,
2783 rdev->sb_page);
2784 pr_debug("md: (write) %s's sb offset: %llu\n",
2785 bdevname(rdev->bdev, b),
2786 (unsigned long long)rdev->sb_start);
2787 rdev->sb_events = mddev->events;
2788 if (rdev->badblocks.size) {
2789 md_super_write(mddev, rdev,
2790 rdev->badblocks.sector,
2791 rdev->badblocks.size << 9,
2792 rdev->bb_page);
2793 rdev->badblocks.size = 0;
2794 }
2795
2796 } else
2797 pr_debug("md: %s (skipping faulty)\n",
2798 bdevname(rdev->bdev, b));
2799
2800 if (mddev->level == LEVEL_MULTIPATH)
2801 /* only need to write one superblock... */
2802 break;
2803 }
2804 if (md_super_wait(mddev) < 0)
2805 goto rewrite;
2806 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2807
2808 if (mddev_is_clustered(mddev) && ret == 0)
2809 md_cluster_ops->metadata_update_finish(mddev);
2810
2811 if (mddev->in_sync != sync_req ||
2812 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2813 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2814 /* have to write it out again */
2815 goto repeat;
2816 wake_up(&mddev->sb_wait);
2817 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2818 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2819
2820 rdev_for_each(rdev, mddev) {
2821 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2822 clear_bit(Blocked, &rdev->flags);
2823
2824 if (any_badblocks_changed)
2825 ack_all_badblocks(&rdev->badblocks);
2826 clear_bit(BlockedBadBlocks, &rdev->flags);
2827 wake_up(&rdev->blocked_wait);
2828 }
2829}
2830EXPORT_SYMBOL(md_update_sb);
2831
2832static int add_bound_rdev(struct md_rdev *rdev)
2833{
2834 struct mddev *mddev = rdev->mddev;
2835 int err = 0;
2836 bool add_journal = test_bit(Journal, &rdev->flags);
2837
2838 if (!mddev->pers->hot_remove_disk || add_journal) {
2839 /* If there is hot_add_disk but no hot_remove_disk
2840 * then added disks for geometry changes,
2841 * and should be added immediately.
2842 */
2843 super_types[mddev->major_version].
2844 validate_super(mddev, rdev);
2845 if (add_journal)
2846 mddev_suspend(mddev);
2847 err = mddev->pers->hot_add_disk(mddev, rdev);
2848 if (add_journal)
2849 mddev_resume(mddev);
2850 if (err) {
2851 md_kick_rdev_from_array(rdev);
2852 return err;
2853 }
2854 }
2855 sysfs_notify_dirent_safe(rdev->sysfs_state);
2856
2857 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2858 if (mddev->degraded)
2859 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2860 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2861 md_new_event(mddev);
2862 md_wakeup_thread(mddev->thread);
2863 return 0;
2864}
2865
2866/* words written to sysfs files may, or may not, be \n terminated.
2867 * We want to accept with case. For this we use cmd_match.
2868 */
2869static int cmd_match(const char *cmd, const char *str)
2870{
2871 /* See if cmd, written into a sysfs file, matches
2872 * str. They must either be the same, or cmd can
2873 * have a trailing newline
2874 */
2875 while (*cmd && *str && *cmd == *str) {
2876 cmd++;
2877 str++;
2878 }
2879 if (*cmd == '\n')
2880 cmd++;
2881 if (*str || *cmd)
2882 return 0;
2883 return 1;
2884}
2885
2886struct rdev_sysfs_entry {
2887 struct attribute attr;
2888 ssize_t (*show)(struct md_rdev *, char *);
2889 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2890};
2891
2892static ssize_t
2893state_show(struct md_rdev *rdev, char *page)
2894{
2895 char *sep = ",";
2896 size_t len = 0;
2897 unsigned long flags = READ_ONCE(rdev->flags);
2898
2899 if (test_bit(Faulty, &flags) ||
2900 (!test_bit(ExternalBbl, &flags) &&
2901 rdev->badblocks.unacked_exist))
2902 len += sprintf(page+len, "faulty%s", sep);
2903 if (test_bit(In_sync, &flags))
2904 len += sprintf(page+len, "in_sync%s", sep);
2905 if (test_bit(Journal, &flags))
2906 len += sprintf(page+len, "journal%s", sep);
2907 if (test_bit(WriteMostly, &flags))
2908 len += sprintf(page+len, "write_mostly%s", sep);
2909 if (test_bit(Blocked, &flags) ||
2910 (rdev->badblocks.unacked_exist
2911 && !test_bit(Faulty, &flags)))
2912 len += sprintf(page+len, "blocked%s", sep);
2913 if (!test_bit(Faulty, &flags) &&
2914 !test_bit(Journal, &flags) &&
2915 !test_bit(In_sync, &flags))
2916 len += sprintf(page+len, "spare%s", sep);
2917 if (test_bit(WriteErrorSeen, &flags))
2918 len += sprintf(page+len, "write_error%s", sep);
2919 if (test_bit(WantReplacement, &flags))
2920 len += sprintf(page+len, "want_replacement%s", sep);
2921 if (test_bit(Replacement, &flags))
2922 len += sprintf(page+len, "replacement%s", sep);
2923 if (test_bit(ExternalBbl, &flags))
2924 len += sprintf(page+len, "external_bbl%s", sep);
2925 if (test_bit(FailFast, &flags))
2926 len += sprintf(page+len, "failfast%s", sep);
2927
2928 if (len)
2929 len -= strlen(sep);
2930
2931 return len+sprintf(page+len, "\n");
2932}
2933
2934static ssize_t
2935state_store(struct md_rdev *rdev, const char *buf, size_t len)
2936{
2937 /* can write
2938 * faulty - simulates an error
2939 * remove - disconnects the device
2940 * writemostly - sets write_mostly
2941 * -writemostly - clears write_mostly
2942 * blocked - sets the Blocked flags
2943 * -blocked - clears the Blocked and possibly simulates an error
2944 * insync - sets Insync providing device isn't active
2945 * -insync - clear Insync for a device with a slot assigned,
2946 * so that it gets rebuilt based on bitmap
2947 * write_error - sets WriteErrorSeen
2948 * -write_error - clears WriteErrorSeen
2949 * {,-}failfast - set/clear FailFast
2950 */
2951 int err = -EINVAL;
2952 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2953 md_error(rdev->mddev, rdev);
2954 if (test_bit(Faulty, &rdev->flags))
2955 err = 0;
2956 else
2957 err = -EBUSY;
2958 } else if (cmd_match(buf, "remove")) {
2959 if (rdev->mddev->pers) {
2960 clear_bit(Blocked, &rdev->flags);
2961 remove_and_add_spares(rdev->mddev, rdev);
2962 }
2963 if (rdev->raid_disk >= 0)
2964 err = -EBUSY;
2965 else {
2966 struct mddev *mddev = rdev->mddev;
2967 err = 0;
2968 if (mddev_is_clustered(mddev))
2969 err = md_cluster_ops->remove_disk(mddev, rdev);
2970
2971 if (err == 0) {
2972 md_kick_rdev_from_array(rdev);
2973 if (mddev->pers) {
2974 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2975 md_wakeup_thread(mddev->thread);
2976 }
2977 md_new_event(mddev);
2978 }
2979 }
2980 } else if (cmd_match(buf, "writemostly")) {
2981 set_bit(WriteMostly, &rdev->flags);
2982 mddev_create_serial_pool(rdev->mddev, rdev, false);
2983 err = 0;
2984 } else if (cmd_match(buf, "-writemostly")) {
2985 mddev_destroy_serial_pool(rdev->mddev, rdev, false);
2986 clear_bit(WriteMostly, &rdev->flags);
2987 err = 0;
2988 } else if (cmd_match(buf, "blocked")) {
2989 set_bit(Blocked, &rdev->flags);
2990 err = 0;
2991 } else if (cmd_match(buf, "-blocked")) {
2992 if (!test_bit(Faulty, &rdev->flags) &&
2993 !test_bit(ExternalBbl, &rdev->flags) &&
2994 rdev->badblocks.unacked_exist) {
2995 /* metadata handler doesn't understand badblocks,
2996 * so we need to fail the device
2997 */
2998 md_error(rdev->mddev, rdev);
2999 }
3000 clear_bit(Blocked, &rdev->flags);
3001 clear_bit(BlockedBadBlocks, &rdev->flags);
3002 wake_up(&rdev->blocked_wait);
3003 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3004 md_wakeup_thread(rdev->mddev->thread);
3005
3006 err = 0;
3007 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
3008 set_bit(In_sync, &rdev->flags);
3009 err = 0;
3010 } else if (cmd_match(buf, "failfast")) {
3011 set_bit(FailFast, &rdev->flags);
3012 err = 0;
3013 } else if (cmd_match(buf, "-failfast")) {
3014 clear_bit(FailFast, &rdev->flags);
3015 err = 0;
3016 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3017 !test_bit(Journal, &rdev->flags)) {
3018 if (rdev->mddev->pers == NULL) {
3019 clear_bit(In_sync, &rdev->flags);
3020 rdev->saved_raid_disk = rdev->raid_disk;
3021 rdev->raid_disk = -1;
3022 err = 0;
3023 }
3024 } else if (cmd_match(buf, "write_error")) {
3025 set_bit(WriteErrorSeen, &rdev->flags);
3026 err = 0;
3027 } else if (cmd_match(buf, "-write_error")) {
3028 clear_bit(WriteErrorSeen, &rdev->flags);
3029 err = 0;
3030 } else if (cmd_match(buf, "want_replacement")) {
3031 /* Any non-spare device that is not a replacement can
3032 * become want_replacement at any time, but we then need to
3033 * check if recovery is needed.
3034 */
3035 if (rdev->raid_disk >= 0 &&
3036 !test_bit(Journal, &rdev->flags) &&
3037 !test_bit(Replacement, &rdev->flags))
3038 set_bit(WantReplacement, &rdev->flags);
3039 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3040 md_wakeup_thread(rdev->mddev->thread);
3041 err = 0;
3042 } else if (cmd_match(buf, "-want_replacement")) {
3043 /* Clearing 'want_replacement' is always allowed.
3044 * Once replacements starts it is too late though.
3045 */
3046 err = 0;
3047 clear_bit(WantReplacement, &rdev->flags);
3048 } else if (cmd_match(buf, "replacement")) {
3049 /* Can only set a device as a replacement when array has not
3050 * yet been started. Once running, replacement is automatic
3051 * from spares, or by assigning 'slot'.
3052 */
3053 if (rdev->mddev->pers)
3054 err = -EBUSY;
3055 else {
3056 set_bit(Replacement, &rdev->flags);
3057 err = 0;
3058 }
3059 } else if (cmd_match(buf, "-replacement")) {
3060 /* Similarly, can only clear Replacement before start */
3061 if (rdev->mddev->pers)
3062 err = -EBUSY;
3063 else {
3064 clear_bit(Replacement, &rdev->flags);
3065 err = 0;
3066 }
3067 } else if (cmd_match(buf, "re-add")) {
3068 if (!rdev->mddev->pers)
3069 err = -EINVAL;
3070 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3071 rdev->saved_raid_disk >= 0) {
3072 /* clear_bit is performed _after_ all the devices
3073 * have their local Faulty bit cleared. If any writes
3074 * happen in the meantime in the local node, they
3075 * will land in the local bitmap, which will be synced
3076 * by this node eventually
3077 */
3078 if (!mddev_is_clustered(rdev->mddev) ||
3079 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3080 clear_bit(Faulty, &rdev->flags);
3081 err = add_bound_rdev(rdev);
3082 }
3083 } else
3084 err = -EBUSY;
3085 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3086 set_bit(ExternalBbl, &rdev->flags);
3087 rdev->badblocks.shift = 0;
3088 err = 0;
3089 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3090 clear_bit(ExternalBbl, &rdev->flags);
3091 err = 0;
3092 }
3093 if (!err)
3094 sysfs_notify_dirent_safe(rdev->sysfs_state);
3095 return err ? err : len;
3096}
3097static struct rdev_sysfs_entry rdev_state =
3098__ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3099
3100static ssize_t
3101errors_show(struct md_rdev *rdev, char *page)
3102{
3103 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3104}
3105
3106static ssize_t
3107errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3108{
3109 unsigned int n;
3110 int rv;
3111
3112 rv = kstrtouint(buf, 10, &n);
3113 if (rv < 0)
3114 return rv;
3115 atomic_set(&rdev->corrected_errors, n);
3116 return len;
3117}
3118static struct rdev_sysfs_entry rdev_errors =
3119__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3120
3121static ssize_t
3122slot_show(struct md_rdev *rdev, char *page)
3123{
3124 if (test_bit(Journal, &rdev->flags))
3125 return sprintf(page, "journal\n");
3126 else if (rdev->raid_disk < 0)
3127 return sprintf(page, "none\n");
3128 else
3129 return sprintf(page, "%d\n", rdev->raid_disk);
3130}
3131
3132static ssize_t
3133slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3134{
3135 int slot;
3136 int err;
3137
3138 if (test_bit(Journal, &rdev->flags))
3139 return -EBUSY;
3140 if (strncmp(buf, "none", 4)==0)
3141 slot = -1;
3142 else {
3143 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3144 if (err < 0)
3145 return err;
3146 }
3147 if (rdev->mddev->pers && slot == -1) {
3148 /* Setting 'slot' on an active array requires also
3149 * updating the 'rd%d' link, and communicating
3150 * with the personality with ->hot_*_disk.
3151 * For now we only support removing
3152 * failed/spare devices. This normally happens automatically,
3153 * but not when the metadata is externally managed.
3154 */
3155 if (rdev->raid_disk == -1)
3156 return -EEXIST;
3157 /* personality does all needed checks */
3158 if (rdev->mddev->pers->hot_remove_disk == NULL)
3159 return -EINVAL;
3160 clear_bit(Blocked, &rdev->flags);
3161 remove_and_add_spares(rdev->mddev, rdev);
3162 if (rdev->raid_disk >= 0)
3163 return -EBUSY;
3164 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3165 md_wakeup_thread(rdev->mddev->thread);
3166 } else if (rdev->mddev->pers) {
3167 /* Activating a spare .. or possibly reactivating
3168 * if we ever get bitmaps working here.
3169 */
3170 int err;
3171
3172 if (rdev->raid_disk != -1)
3173 return -EBUSY;
3174
3175 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3176 return -EBUSY;
3177
3178 if (rdev->mddev->pers->hot_add_disk == NULL)
3179 return -EINVAL;
3180
3181 if (slot >= rdev->mddev->raid_disks &&
3182 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3183 return -ENOSPC;
3184
3185 rdev->raid_disk = slot;
3186 if (test_bit(In_sync, &rdev->flags))
3187 rdev->saved_raid_disk = slot;
3188 else
3189 rdev->saved_raid_disk = -1;
3190 clear_bit(In_sync, &rdev->flags);
3191 clear_bit(Bitmap_sync, &rdev->flags);
3192 err = rdev->mddev->pers->
3193 hot_add_disk(rdev->mddev, rdev);
3194 if (err) {
3195 rdev->raid_disk = -1;
3196 return err;
3197 } else
3198 sysfs_notify_dirent_safe(rdev->sysfs_state);
3199 if (sysfs_link_rdev(rdev->mddev, rdev))
3200 /* failure here is OK */;
3201 /* don't wakeup anyone, leave that to userspace. */
3202 } else {
3203 if (slot >= rdev->mddev->raid_disks &&
3204 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3205 return -ENOSPC;
3206 rdev->raid_disk = slot;
3207 /* assume it is working */
3208 clear_bit(Faulty, &rdev->flags);
3209 clear_bit(WriteMostly, &rdev->flags);
3210 set_bit(In_sync, &rdev->flags);
3211 sysfs_notify_dirent_safe(rdev->sysfs_state);
3212 }
3213 return len;
3214}
3215
3216static struct rdev_sysfs_entry rdev_slot =
3217__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3218
3219static ssize_t
3220offset_show(struct md_rdev *rdev, char *page)
3221{
3222 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3223}
3224
3225static ssize_t
3226offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3227{
3228 unsigned long long offset;
3229 if (kstrtoull(buf, 10, &offset) < 0)
3230 return -EINVAL;
3231 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3232 return -EBUSY;
3233 if (rdev->sectors && rdev->mddev->external)
3234 /* Must set offset before size, so overlap checks
3235 * can be sane */
3236 return -EBUSY;
3237 rdev->data_offset = offset;
3238 rdev->new_data_offset = offset;
3239 return len;
3240}
3241
3242static struct rdev_sysfs_entry rdev_offset =
3243__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3244
3245static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3246{
3247 return sprintf(page, "%llu\n",
3248 (unsigned long long)rdev->new_data_offset);
3249}
3250
3251static ssize_t new_offset_store(struct md_rdev *rdev,
3252 const char *buf, size_t len)
3253{
3254 unsigned long long new_offset;
3255 struct mddev *mddev = rdev->mddev;
3256
3257 if (kstrtoull(buf, 10, &new_offset) < 0)
3258 return -EINVAL;
3259
3260 if (mddev->sync_thread ||
3261 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3262 return -EBUSY;
3263 if (new_offset == rdev->data_offset)
3264 /* reset is always permitted */
3265 ;
3266 else if (new_offset > rdev->data_offset) {
3267 /* must not push array size beyond rdev_sectors */
3268 if (new_offset - rdev->data_offset
3269 + mddev->dev_sectors > rdev->sectors)
3270 return -E2BIG;
3271 }
3272 /* Metadata worries about other space details. */
3273
3274 /* decreasing the offset is inconsistent with a backwards
3275 * reshape.
3276 */
3277 if (new_offset < rdev->data_offset &&
3278 mddev->reshape_backwards)
3279 return -EINVAL;
3280 /* Increasing offset is inconsistent with forwards
3281 * reshape. reshape_direction should be set to
3282 * 'backwards' first.
3283 */
3284 if (new_offset > rdev->data_offset &&
3285 !mddev->reshape_backwards)
3286 return -EINVAL;
3287
3288 if (mddev->pers && mddev->persistent &&
3289 !super_types[mddev->major_version]
3290 .allow_new_offset(rdev, new_offset))
3291 return -E2BIG;
3292 rdev->new_data_offset = new_offset;
3293 if (new_offset > rdev->data_offset)
3294 mddev->reshape_backwards = 1;
3295 else if (new_offset < rdev->data_offset)
3296 mddev->reshape_backwards = 0;
3297
3298 return len;
3299}
3300static struct rdev_sysfs_entry rdev_new_offset =
3301__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3302
3303static ssize_t
3304rdev_size_show(struct md_rdev *rdev, char *page)
3305{
3306 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3307}
3308
3309static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3310{
3311 /* check if two start/length pairs overlap */
3312 if (s1+l1 <= s2)
3313 return 0;
3314 if (s2+l2 <= s1)
3315 return 0;
3316 return 1;
3317}
3318
3319static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3320{
3321 unsigned long long blocks;
3322 sector_t new;
3323
3324 if (kstrtoull(buf, 10, &blocks) < 0)
3325 return -EINVAL;
3326
3327 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3328 return -EINVAL; /* sector conversion overflow */
3329
3330 new = blocks * 2;
3331 if (new != blocks * 2)
3332 return -EINVAL; /* unsigned long long to sector_t overflow */
3333
3334 *sectors = new;
3335 return 0;
3336}
3337
3338static ssize_t
3339rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3340{
3341 struct mddev *my_mddev = rdev->mddev;
3342 sector_t oldsectors = rdev->sectors;
3343 sector_t sectors;
3344
3345 if (test_bit(Journal, &rdev->flags))
3346 return -EBUSY;
3347 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3348 return -EINVAL;
3349 if (rdev->data_offset != rdev->new_data_offset)
3350 return -EINVAL; /* too confusing */
3351 if (my_mddev->pers && rdev->raid_disk >= 0) {
3352 if (my_mddev->persistent) {
3353 sectors = super_types[my_mddev->major_version].
3354 rdev_size_change(rdev, sectors);
3355 if (!sectors)
3356 return -EBUSY;
3357 } else if (!sectors)
3358 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3359 rdev->data_offset;
3360 if (!my_mddev->pers->resize)
3361 /* Cannot change size for RAID0 or Linear etc */
3362 return -EINVAL;
3363 }
3364 if (sectors < my_mddev->dev_sectors)
3365 return -EINVAL; /* component must fit device */
3366
3367 rdev->sectors = sectors;
3368 if (sectors > oldsectors && my_mddev->external) {
3369 /* Need to check that all other rdevs with the same
3370 * ->bdev do not overlap. 'rcu' is sufficient to walk
3371 * the rdev lists safely.
3372 * This check does not provide a hard guarantee, it
3373 * just helps avoid dangerous mistakes.
3374 */
3375 struct mddev *mddev;
3376 int overlap = 0;
3377 struct list_head *tmp;
3378
3379 rcu_read_lock();
3380 for_each_mddev(mddev, tmp) {
3381 struct md_rdev *rdev2;
3382
3383 rdev_for_each(rdev2, mddev)
3384 if (rdev->bdev == rdev2->bdev &&
3385 rdev != rdev2 &&
3386 overlaps(rdev->data_offset, rdev->sectors,
3387 rdev2->data_offset,
3388 rdev2->sectors)) {
3389 overlap = 1;
3390 break;
3391 }
3392 if (overlap) {
3393 mddev_put(mddev);
3394 break;
3395 }
3396 }
3397 rcu_read_unlock();
3398 if (overlap) {
3399 /* Someone else could have slipped in a size
3400 * change here, but doing so is just silly.
3401 * We put oldsectors back because we *know* it is
3402 * safe, and trust userspace not to race with
3403 * itself
3404 */
3405 rdev->sectors = oldsectors;
3406 return -EBUSY;
3407 }
3408 }
3409 return len;
3410}
3411
3412static struct rdev_sysfs_entry rdev_size =
3413__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3414
3415static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3416{
3417 unsigned long long recovery_start = rdev->recovery_offset;
3418
3419 if (test_bit(In_sync, &rdev->flags) ||
3420 recovery_start == MaxSector)
3421 return sprintf(page, "none\n");
3422
3423 return sprintf(page, "%llu\n", recovery_start);
3424}
3425
3426static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3427{
3428 unsigned long long recovery_start;
3429
3430 if (cmd_match(buf, "none"))
3431 recovery_start = MaxSector;
3432 else if (kstrtoull(buf, 10, &recovery_start))
3433 return -EINVAL;
3434
3435 if (rdev->mddev->pers &&
3436 rdev->raid_disk >= 0)
3437 return -EBUSY;
3438
3439 rdev->recovery_offset = recovery_start;
3440 if (recovery_start == MaxSector)
3441 set_bit(In_sync, &rdev->flags);
3442 else
3443 clear_bit(In_sync, &rdev->flags);
3444 return len;
3445}
3446
3447static struct rdev_sysfs_entry rdev_recovery_start =
3448__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3449
3450/* sysfs access to bad-blocks list.
3451 * We present two files.
3452 * 'bad-blocks' lists sector numbers and lengths of ranges that
3453 * are recorded as bad. The list is truncated to fit within
3454 * the one-page limit of sysfs.
3455 * Writing "sector length" to this file adds an acknowledged
3456 * bad block list.
3457 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3458 * been acknowledged. Writing to this file adds bad blocks
3459 * without acknowledging them. This is largely for testing.
3460 */
3461static ssize_t bb_show(struct md_rdev *rdev, char *page)
3462{
3463 return badblocks_show(&rdev->badblocks, page, 0);
3464}
3465static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3466{
3467 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3468 /* Maybe that ack was all we needed */
3469 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3470 wake_up(&rdev->blocked_wait);
3471 return rv;
3472}
3473static struct rdev_sysfs_entry rdev_bad_blocks =
3474__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3475
3476static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3477{
3478 return badblocks_show(&rdev->badblocks, page, 1);
3479}
3480static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3481{
3482 return badblocks_store(&rdev->badblocks, page, len, 1);
3483}
3484static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3485__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3486
3487static ssize_t
3488ppl_sector_show(struct md_rdev *rdev, char *page)
3489{
3490 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3491}
3492
3493static ssize_t
3494ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3495{
3496 unsigned long long sector;
3497
3498 if (kstrtoull(buf, 10, &sector) < 0)
3499 return -EINVAL;
3500 if (sector != (sector_t)sector)
3501 return -EINVAL;
3502
3503 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3504 rdev->raid_disk >= 0)
3505 return -EBUSY;
3506
3507 if (rdev->mddev->persistent) {
3508 if (rdev->mddev->major_version == 0)
3509 return -EINVAL;
3510 if ((sector > rdev->sb_start &&
3511 sector - rdev->sb_start > S16_MAX) ||
3512 (sector < rdev->sb_start &&
3513 rdev->sb_start - sector > -S16_MIN))
3514 return -EINVAL;
3515 rdev->ppl.offset = sector - rdev->sb_start;
3516 } else if (!rdev->mddev->external) {
3517 return -EBUSY;
3518 }
3519 rdev->ppl.sector = sector;
3520 return len;
3521}
3522
3523static struct rdev_sysfs_entry rdev_ppl_sector =
3524__ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3525
3526static ssize_t
3527ppl_size_show(struct md_rdev *rdev, char *page)
3528{
3529 return sprintf(page, "%u\n", rdev->ppl.size);
3530}
3531
3532static ssize_t
3533ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3534{
3535 unsigned int size;
3536
3537 if (kstrtouint(buf, 10, &size) < 0)
3538 return -EINVAL;
3539
3540 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3541 rdev->raid_disk >= 0)
3542 return -EBUSY;
3543
3544 if (rdev->mddev->persistent) {
3545 if (rdev->mddev->major_version == 0)
3546 return -EINVAL;
3547 if (size > U16_MAX)
3548 return -EINVAL;
3549 } else if (!rdev->mddev->external) {
3550 return -EBUSY;
3551 }
3552 rdev->ppl.size = size;
3553 return len;
3554}
3555
3556static struct rdev_sysfs_entry rdev_ppl_size =
3557__ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3558
3559static struct attribute *rdev_default_attrs[] = {
3560 &rdev_state.attr,
3561 &rdev_errors.attr,
3562 &rdev_slot.attr,
3563 &rdev_offset.attr,
3564 &rdev_new_offset.attr,
3565 &rdev_size.attr,
3566 &rdev_recovery_start.attr,
3567 &rdev_bad_blocks.attr,
3568 &rdev_unack_bad_blocks.attr,
3569 &rdev_ppl_sector.attr,
3570 &rdev_ppl_size.attr,
3571 NULL,
3572};
3573static ssize_t
3574rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3575{
3576 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3577 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3578
3579 if (!entry->show)
3580 return -EIO;
3581 if (!rdev->mddev)
3582 return -ENODEV;
3583 return entry->show(rdev, page);
3584}
3585
3586static ssize_t
3587rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3588 const char *page, size_t length)
3589{
3590 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3591 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3592 ssize_t rv;
3593 struct mddev *mddev = rdev->mddev;
3594
3595 if (!entry->store)
3596 return -EIO;
3597 if (!capable(CAP_SYS_ADMIN))
3598 return -EACCES;
3599 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3600 if (!rv) {
3601 if (rdev->mddev == NULL)
3602 rv = -ENODEV;
3603 else
3604 rv = entry->store(rdev, page, length);
3605 mddev_unlock(mddev);
3606 }
3607 return rv;
3608}
3609
3610static void rdev_free(struct kobject *ko)
3611{
3612 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3613 kfree(rdev);
3614}
3615static const struct sysfs_ops rdev_sysfs_ops = {
3616 .show = rdev_attr_show,
3617 .store = rdev_attr_store,
3618};
3619static struct kobj_type rdev_ktype = {
3620 .release = rdev_free,
3621 .sysfs_ops = &rdev_sysfs_ops,
3622 .default_attrs = rdev_default_attrs,
3623};
3624
3625int md_rdev_init(struct md_rdev *rdev)
3626{
3627 rdev->desc_nr = -1;
3628 rdev->saved_raid_disk = -1;
3629 rdev->raid_disk = -1;
3630 rdev->flags = 0;
3631 rdev->data_offset = 0;
3632 rdev->new_data_offset = 0;
3633 rdev->sb_events = 0;
3634 rdev->last_read_error = 0;
3635 rdev->sb_loaded = 0;
3636 rdev->bb_page = NULL;
3637 atomic_set(&rdev->nr_pending, 0);
3638 atomic_set(&rdev->read_errors, 0);
3639 atomic_set(&rdev->corrected_errors, 0);
3640
3641 INIT_LIST_HEAD(&rdev->same_set);
3642 init_waitqueue_head(&rdev->blocked_wait);
3643
3644 /* Add space to store bad block list.
3645 * This reserves the space even on arrays where it cannot
3646 * be used - I wonder if that matters
3647 */
3648 return badblocks_init(&rdev->badblocks, 0);
3649}
3650EXPORT_SYMBOL_GPL(md_rdev_init);
3651/*
3652 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3653 *
3654 * mark the device faulty if:
3655 *
3656 * - the device is nonexistent (zero size)
3657 * - the device has no valid superblock
3658 *
3659 * a faulty rdev _never_ has rdev->sb set.
3660 */
3661static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3662{
3663 char b[BDEVNAME_SIZE];
3664 int err;
3665 struct md_rdev *rdev;
3666 sector_t size;
3667
3668 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3669 if (!rdev)
3670 return ERR_PTR(-ENOMEM);
3671
3672 err = md_rdev_init(rdev);
3673 if (err)
3674 goto abort_free;
3675 err = alloc_disk_sb(rdev);
3676 if (err)
3677 goto abort_free;
3678
3679 err = lock_rdev(rdev, newdev, super_format == -2);
3680 if (err)
3681 goto abort_free;
3682
3683 kobject_init(&rdev->kobj, &rdev_ktype);
3684
3685 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3686 if (!size) {
3687 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3688 bdevname(rdev->bdev,b));
3689 err = -EINVAL;
3690 goto abort_free;
3691 }
3692
3693 if (super_format >= 0) {
3694 err = super_types[super_format].
3695 load_super(rdev, NULL, super_minor);
3696 if (err == -EINVAL) {
3697 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3698 bdevname(rdev->bdev,b),
3699 super_format, super_minor);
3700 goto abort_free;
3701 }
3702 if (err < 0) {
3703 pr_warn("md: could not read %s's sb, not importing!\n",
3704 bdevname(rdev->bdev,b));
3705 goto abort_free;
3706 }
3707 }
3708
3709 return rdev;
3710
3711abort_free:
3712 if (rdev->bdev)
3713 unlock_rdev(rdev);
3714 md_rdev_clear(rdev);
3715 kfree(rdev);
3716 return ERR_PTR(err);
3717}
3718
3719/*
3720 * Check a full RAID array for plausibility
3721 */
3722
3723static int analyze_sbs(struct mddev *mddev)
3724{
3725 int i;
3726 struct md_rdev *rdev, *freshest, *tmp;
3727 char b[BDEVNAME_SIZE];
3728
3729 freshest = NULL;
3730 rdev_for_each_safe(rdev, tmp, mddev)
3731 switch (super_types[mddev->major_version].
3732 load_super(rdev, freshest, mddev->minor_version)) {
3733 case 1:
3734 freshest = rdev;
3735 break;
3736 case 0:
3737 break;
3738 default:
3739 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3740 bdevname(rdev->bdev,b));
3741 md_kick_rdev_from_array(rdev);
3742 }
3743
3744 /* Cannot find a valid fresh disk */
3745 if (!freshest) {
3746 pr_warn("md: cannot find a valid disk\n");
3747 return -EINVAL;
3748 }
3749
3750 super_types[mddev->major_version].
3751 validate_super(mddev, freshest);
3752
3753 i = 0;
3754 rdev_for_each_safe(rdev, tmp, mddev) {
3755 if (mddev->max_disks &&
3756 (rdev->desc_nr >= mddev->max_disks ||
3757 i > mddev->max_disks)) {
3758 pr_warn("md: %s: %s: only %d devices permitted\n",
3759 mdname(mddev), bdevname(rdev->bdev, b),
3760 mddev->max_disks);
3761 md_kick_rdev_from_array(rdev);
3762 continue;
3763 }
3764 if (rdev != freshest) {
3765 if (super_types[mddev->major_version].
3766 validate_super(mddev, rdev)) {
3767 pr_warn("md: kicking non-fresh %s from array!\n",
3768 bdevname(rdev->bdev,b));
3769 md_kick_rdev_from_array(rdev);
3770 continue;
3771 }
3772 }
3773 if (mddev->level == LEVEL_MULTIPATH) {
3774 rdev->desc_nr = i++;
3775 rdev->raid_disk = rdev->desc_nr;
3776 set_bit(In_sync, &rdev->flags);
3777 } else if (rdev->raid_disk >=
3778 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3779 !test_bit(Journal, &rdev->flags)) {
3780 rdev->raid_disk = -1;
3781 clear_bit(In_sync, &rdev->flags);
3782 }
3783 }
3784
3785 return 0;
3786}
3787
3788/* Read a fixed-point number.
3789 * Numbers in sysfs attributes should be in "standard" units where
3790 * possible, so time should be in seconds.
3791 * However we internally use a a much smaller unit such as
3792 * milliseconds or jiffies.
3793 * This function takes a decimal number with a possible fractional
3794 * component, and produces an integer which is the result of
3795 * multiplying that number by 10^'scale'.
3796 * all without any floating-point arithmetic.
3797 */
3798int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3799{
3800 unsigned long result = 0;
3801 long decimals = -1;
3802 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3803 if (*cp == '.')
3804 decimals = 0;
3805 else if (decimals < scale) {
3806 unsigned int value;
3807 value = *cp - '0';
3808 result = result * 10 + value;
3809 if (decimals >= 0)
3810 decimals++;
3811 }
3812 cp++;
3813 }
3814 if (*cp == '\n')
3815 cp++;
3816 if (*cp)
3817 return -EINVAL;
3818 if (decimals < 0)
3819 decimals = 0;
3820 *res = result * int_pow(10, scale - decimals);
3821 return 0;
3822}
3823
3824static ssize_t
3825safe_delay_show(struct mddev *mddev, char *page)
3826{
3827 int msec = (mddev->safemode_delay*1000)/HZ;
3828 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3829}
3830static ssize_t
3831safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3832{
3833 unsigned long msec;
3834
3835 if (mddev_is_clustered(mddev)) {
3836 pr_warn("md: Safemode is disabled for clustered mode\n");
3837 return -EINVAL;
3838 }
3839
3840 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3841 return -EINVAL;
3842 if (msec == 0)
3843 mddev->safemode_delay = 0;
3844 else {
3845 unsigned long old_delay = mddev->safemode_delay;
3846 unsigned long new_delay = (msec*HZ)/1000;
3847
3848 if (new_delay == 0)
3849 new_delay = 1;
3850 mddev->safemode_delay = new_delay;
3851 if (new_delay < old_delay || old_delay == 0)
3852 mod_timer(&mddev->safemode_timer, jiffies+1);
3853 }
3854 return len;
3855}
3856static struct md_sysfs_entry md_safe_delay =
3857__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3858
3859static ssize_t
3860level_show(struct mddev *mddev, char *page)
3861{
3862 struct md_personality *p;
3863 int ret;
3864 spin_lock(&mddev->lock);
3865 p = mddev->pers;
3866 if (p)
3867 ret = sprintf(page, "%s\n", p->name);
3868 else if (mddev->clevel[0])
3869 ret = sprintf(page, "%s\n", mddev->clevel);
3870 else if (mddev->level != LEVEL_NONE)
3871 ret = sprintf(page, "%d\n", mddev->level);
3872 else
3873 ret = 0;
3874 spin_unlock(&mddev->lock);
3875 return ret;
3876}
3877
3878static ssize_t
3879level_store(struct mddev *mddev, const char *buf, size_t len)
3880{
3881 char clevel[16];
3882 ssize_t rv;
3883 size_t slen = len;
3884 struct md_personality *pers, *oldpers;
3885 long level;
3886 void *priv, *oldpriv;
3887 struct md_rdev *rdev;
3888
3889 if (slen == 0 || slen >= sizeof(clevel))
3890 return -EINVAL;
3891
3892 rv = mddev_lock(mddev);
3893 if (rv)
3894 return rv;
3895
3896 if (mddev->pers == NULL) {
3897 strncpy(mddev->clevel, buf, slen);
3898 if (mddev->clevel[slen-1] == '\n')
3899 slen--;
3900 mddev->clevel[slen] = 0;
3901 mddev->level = LEVEL_NONE;
3902 rv = len;
3903 goto out_unlock;
3904 }
3905 rv = -EROFS;
3906 if (mddev->ro)
3907 goto out_unlock;
3908
3909 /* request to change the personality. Need to ensure:
3910 * - array is not engaged in resync/recovery/reshape
3911 * - old personality can be suspended
3912 * - new personality will access other array.
3913 */
3914
3915 rv = -EBUSY;
3916 if (mddev->sync_thread ||
3917 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3918 mddev->reshape_position != MaxSector ||
3919 mddev->sysfs_active)
3920 goto out_unlock;
3921
3922 rv = -EINVAL;
3923 if (!mddev->pers->quiesce) {
3924 pr_warn("md: %s: %s does not support online personality change\n",
3925 mdname(mddev), mddev->pers->name);
3926 goto out_unlock;
3927 }
3928
3929 /* Now find the new personality */
3930 strncpy(clevel, buf, slen);
3931 if (clevel[slen-1] == '\n')
3932 slen--;
3933 clevel[slen] = 0;
3934 if (kstrtol(clevel, 10, &level))
3935 level = LEVEL_NONE;
3936
3937 if (request_module("md-%s", clevel) != 0)
3938 request_module("md-level-%s", clevel);
3939 spin_lock(&pers_lock);
3940 pers = find_pers(level, clevel);
3941 if (!pers || !try_module_get(pers->owner)) {
3942 spin_unlock(&pers_lock);
3943 pr_warn("md: personality %s not loaded\n", clevel);
3944 rv = -EINVAL;
3945 goto out_unlock;
3946 }
3947 spin_unlock(&pers_lock);
3948
3949 if (pers == mddev->pers) {
3950 /* Nothing to do! */
3951 module_put(pers->owner);
3952 rv = len;
3953 goto out_unlock;
3954 }
3955 if (!pers->takeover) {
3956 module_put(pers->owner);
3957 pr_warn("md: %s: %s does not support personality takeover\n",
3958 mdname(mddev), clevel);
3959 rv = -EINVAL;
3960 goto out_unlock;
3961 }
3962
3963 rdev_for_each(rdev, mddev)
3964 rdev->new_raid_disk = rdev->raid_disk;
3965
3966 /* ->takeover must set new_* and/or delta_disks
3967 * if it succeeds, and may set them when it fails.
3968 */
3969 priv = pers->takeover(mddev);
3970 if (IS_ERR(priv)) {
3971 mddev->new_level = mddev->level;
3972 mddev->new_layout = mddev->layout;
3973 mddev->new_chunk_sectors = mddev->chunk_sectors;
3974 mddev->raid_disks -= mddev->delta_disks;
3975 mddev->delta_disks = 0;
3976 mddev->reshape_backwards = 0;
3977 module_put(pers->owner);
3978 pr_warn("md: %s: %s would not accept array\n",
3979 mdname(mddev), clevel);
3980 rv = PTR_ERR(priv);
3981 goto out_unlock;
3982 }
3983
3984 /* Looks like we have a winner */
3985 mddev_suspend(mddev);
3986 mddev_detach(mddev);
3987
3988 spin_lock(&mddev->lock);
3989 oldpers = mddev->pers;
3990 oldpriv = mddev->private;
3991 mddev->pers = pers;
3992 mddev->private = priv;
3993 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3994 mddev->level = mddev->new_level;
3995 mddev->layout = mddev->new_layout;
3996 mddev->chunk_sectors = mddev->new_chunk_sectors;
3997 mddev->delta_disks = 0;
3998 mddev->reshape_backwards = 0;
3999 mddev->degraded = 0;
4000 spin_unlock(&mddev->lock);
4001
4002 if (oldpers->sync_request == NULL &&
4003 mddev->external) {
4004 /* We are converting from a no-redundancy array
4005 * to a redundancy array and metadata is managed
4006 * externally so we need to be sure that writes
4007 * won't block due to a need to transition
4008 * clean->dirty
4009 * until external management is started.
4010 */
4011 mddev->in_sync = 0;
4012 mddev->safemode_delay = 0;
4013 mddev->safemode = 0;
4014 }
4015
4016 oldpers->free(mddev, oldpriv);
4017
4018 if (oldpers->sync_request == NULL &&
4019 pers->sync_request != NULL) {
4020 /* need to add the md_redundancy_group */
4021 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4022 pr_warn("md: cannot register extra attributes for %s\n",
4023 mdname(mddev));
4024 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4025 }
4026 if (oldpers->sync_request != NULL &&
4027 pers->sync_request == NULL) {
4028 /* need to remove the md_redundancy_group */
4029 if (mddev->to_remove == NULL)
4030 mddev->to_remove = &md_redundancy_group;
4031 }
4032
4033 module_put(oldpers->owner);
4034
4035 rdev_for_each(rdev, mddev) {
4036 if (rdev->raid_disk < 0)
4037 continue;
4038 if (rdev->new_raid_disk >= mddev->raid_disks)
4039 rdev->new_raid_disk = -1;
4040 if (rdev->new_raid_disk == rdev->raid_disk)
4041 continue;
4042 sysfs_unlink_rdev(mddev, rdev);
4043 }
4044 rdev_for_each(rdev, mddev) {
4045 if (rdev->raid_disk < 0)
4046 continue;
4047 if (rdev->new_raid_disk == rdev->raid_disk)
4048 continue;
4049 rdev->raid_disk = rdev->new_raid_disk;
4050 if (rdev->raid_disk < 0)
4051 clear_bit(In_sync, &rdev->flags);
4052 else {
4053 if (sysfs_link_rdev(mddev, rdev))
4054 pr_warn("md: cannot register rd%d for %s after level change\n",
4055 rdev->raid_disk, mdname(mddev));
4056 }
4057 }
4058
4059 if (pers->sync_request == NULL) {
4060 /* this is now an array without redundancy, so
4061 * it must always be in_sync
4062 */
4063 mddev->in_sync = 1;
4064 del_timer_sync(&mddev->safemode_timer);
4065 }
4066 blk_set_stacking_limits(&mddev->queue->limits);
4067 pers->run(mddev);
4068 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4069 mddev_resume(mddev);
4070 if (!mddev->thread)
4071 md_update_sb(mddev, 1);
4072 sysfs_notify(&mddev->kobj, NULL, "level");
4073 md_new_event(mddev);
4074 rv = len;
4075out_unlock:
4076 mddev_unlock(mddev);
4077 return rv;
4078}
4079
4080static struct md_sysfs_entry md_level =
4081__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4082
4083static ssize_t
4084layout_show(struct mddev *mddev, char *page)
4085{
4086 /* just a number, not meaningful for all levels */
4087 if (mddev->reshape_position != MaxSector &&
4088 mddev->layout != mddev->new_layout)
4089 return sprintf(page, "%d (%d)\n",
4090 mddev->new_layout, mddev->layout);
4091 return sprintf(page, "%d\n", mddev->layout);
4092}
4093
4094static ssize_t
4095layout_store(struct mddev *mddev, const char *buf, size_t len)
4096{
4097 unsigned int n;
4098 int err;
4099
4100 err = kstrtouint(buf, 10, &n);
4101 if (err < 0)
4102 return err;
4103 err = mddev_lock(mddev);
4104 if (err)
4105 return err;
4106
4107 if (mddev->pers) {
4108 if (mddev->pers->check_reshape == NULL)
4109 err = -EBUSY;
4110 else if (mddev->ro)
4111 err = -EROFS;
4112 else {
4113 mddev->new_layout = n;
4114 err = mddev->pers->check_reshape(mddev);
4115 if (err)
4116 mddev->new_layout = mddev->layout;
4117 }
4118 } else {
4119 mddev->new_layout = n;
4120 if (mddev->reshape_position == MaxSector)
4121 mddev->layout = n;
4122 }
4123 mddev_unlock(mddev);
4124 return err ?: len;
4125}
4126static struct md_sysfs_entry md_layout =
4127__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4128
4129static ssize_t
4130raid_disks_show(struct mddev *mddev, char *page)
4131{
4132 if (mddev->raid_disks == 0)
4133 return 0;
4134 if (mddev->reshape_position != MaxSector &&
4135 mddev->delta_disks != 0)
4136 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4137 mddev->raid_disks - mddev->delta_disks);
4138 return sprintf(page, "%d\n", mddev->raid_disks);
4139}
4140
4141static int update_raid_disks(struct mddev *mddev, int raid_disks);
4142
4143static ssize_t
4144raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4145{
4146 unsigned int n;
4147 int err;
4148
4149 err = kstrtouint(buf, 10, &n);
4150 if (err < 0)
4151 return err;
4152
4153 err = mddev_lock(mddev);
4154 if (err)
4155 return err;
4156 if (mddev->pers)
4157 err = update_raid_disks(mddev, n);
4158 else if (mddev->reshape_position != MaxSector) {
4159 struct md_rdev *rdev;
4160 int olddisks = mddev->raid_disks - mddev->delta_disks;
4161
4162 err = -EINVAL;
4163 rdev_for_each(rdev, mddev) {
4164 if (olddisks < n &&
4165 rdev->data_offset < rdev->new_data_offset)
4166 goto out_unlock;
4167 if (olddisks > n &&
4168 rdev->data_offset > rdev->new_data_offset)
4169 goto out_unlock;
4170 }
4171 err = 0;
4172 mddev->delta_disks = n - olddisks;
4173 mddev->raid_disks = n;
4174 mddev->reshape_backwards = (mddev->delta_disks < 0);
4175 } else
4176 mddev->raid_disks = n;
4177out_unlock:
4178 mddev_unlock(mddev);
4179 return err ? err : len;
4180}
4181static struct md_sysfs_entry md_raid_disks =
4182__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4183
4184static ssize_t
4185chunk_size_show(struct mddev *mddev, char *page)
4186{
4187 if (mddev->reshape_position != MaxSector &&
4188 mddev->chunk_sectors != mddev->new_chunk_sectors)
4189 return sprintf(page, "%d (%d)\n",
4190 mddev->new_chunk_sectors << 9,
4191 mddev->chunk_sectors << 9);
4192 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4193}
4194
4195static ssize_t
4196chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4197{
4198 unsigned long n;
4199 int err;
4200
4201 err = kstrtoul(buf, 10, &n);
4202 if (err < 0)
4203 return err;
4204
4205 err = mddev_lock(mddev);
4206 if (err)
4207 return err;
4208 if (mddev->pers) {
4209 if (mddev->pers->check_reshape == NULL)
4210 err = -EBUSY;
4211 else if (mddev->ro)
4212 err = -EROFS;
4213 else {
4214 mddev->new_chunk_sectors = n >> 9;
4215 err = mddev->pers->check_reshape(mddev);
4216 if (err)
4217 mddev->new_chunk_sectors = mddev->chunk_sectors;
4218 }
4219 } else {
4220 mddev->new_chunk_sectors = n >> 9;
4221 if (mddev->reshape_position == MaxSector)
4222 mddev->chunk_sectors = n >> 9;
4223 }
4224 mddev_unlock(mddev);
4225 return err ?: len;
4226}
4227static struct md_sysfs_entry md_chunk_size =
4228__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4229
4230static ssize_t
4231resync_start_show(struct mddev *mddev, char *page)
4232{
4233 if (mddev->recovery_cp == MaxSector)
4234 return sprintf(page, "none\n");
4235 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4236}
4237
4238static ssize_t
4239resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4240{
4241 unsigned long long n;
4242 int err;
4243
4244 if (cmd_match(buf, "none"))
4245 n = MaxSector;
4246 else {
4247 err = kstrtoull(buf, 10, &n);
4248 if (err < 0)
4249 return err;
4250 if (n != (sector_t)n)
4251 return -EINVAL;
4252 }
4253
4254 err = mddev_lock(mddev);
4255 if (err)
4256 return err;
4257 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4258 err = -EBUSY;
4259
4260 if (!err) {
4261 mddev->recovery_cp = n;
4262 if (mddev->pers)
4263 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4264 }
4265 mddev_unlock(mddev);
4266 return err ?: len;
4267}
4268static struct md_sysfs_entry md_resync_start =
4269__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4270 resync_start_show, resync_start_store);
4271
4272/*
4273 * The array state can be:
4274 *
4275 * clear
4276 * No devices, no size, no level
4277 * Equivalent to STOP_ARRAY ioctl
4278 * inactive
4279 * May have some settings, but array is not active
4280 * all IO results in error
4281 * When written, doesn't tear down array, but just stops it
4282 * suspended (not supported yet)
4283 * All IO requests will block. The array can be reconfigured.
4284 * Writing this, if accepted, will block until array is quiescent
4285 * readonly
4286 * no resync can happen. no superblocks get written.
4287 * write requests fail
4288 * read-auto
4289 * like readonly, but behaves like 'clean' on a write request.
4290 *
4291 * clean - no pending writes, but otherwise active.
4292 * When written to inactive array, starts without resync
4293 * If a write request arrives then
4294 * if metadata is known, mark 'dirty' and switch to 'active'.
4295 * if not known, block and switch to write-pending
4296 * If written to an active array that has pending writes, then fails.
4297 * active
4298 * fully active: IO and resync can be happening.
4299 * When written to inactive array, starts with resync
4300 *
4301 * write-pending
4302 * clean, but writes are blocked waiting for 'active' to be written.
4303 *
4304 * active-idle
4305 * like active, but no writes have been seen for a while (100msec).
4306 *
4307 * broken
4308 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4309 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4310 * when a member is gone, so this state will at least alert the
4311 * user that something is wrong.
4312 */
4313enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4314 write_pending, active_idle, broken, bad_word};
4315static char *array_states[] = {
4316 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4317 "write-pending", "active-idle", "broken", NULL };
4318
4319static int match_word(const char *word, char **list)
4320{
4321 int n;
4322 for (n=0; list[n]; n++)
4323 if (cmd_match(word, list[n]))
4324 break;
4325 return n;
4326}
4327
4328static ssize_t
4329array_state_show(struct mddev *mddev, char *page)
4330{
4331 enum array_state st = inactive;
4332
4333 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4334 switch(mddev->ro) {
4335 case 1:
4336 st = readonly;
4337 break;
4338 case 2:
4339 st = read_auto;
4340 break;
4341 case 0:
4342 spin_lock(&mddev->lock);
4343 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4344 st = write_pending;
4345 else if (mddev->in_sync)
4346 st = clean;
4347 else if (mddev->safemode)
4348 st = active_idle;
4349 else
4350 st = active;
4351 spin_unlock(&mddev->lock);
4352 }
4353
4354 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4355 st = broken;
4356 } else {
4357 if (list_empty(&mddev->disks) &&
4358 mddev->raid_disks == 0 &&
4359 mddev->dev_sectors == 0)
4360 st = clear;
4361 else
4362 st = inactive;
4363 }
4364 return sprintf(page, "%s\n", array_states[st]);
4365}
4366
4367static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4368static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4369static int do_md_run(struct mddev *mddev);
4370static int restart_array(struct mddev *mddev);
4371
4372static ssize_t
4373array_state_store(struct mddev *mddev, const char *buf, size_t len)
4374{
4375 int err = 0;
4376 enum array_state st = match_word(buf, array_states);
4377
4378 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4379 /* don't take reconfig_mutex when toggling between
4380 * clean and active
4381 */
4382 spin_lock(&mddev->lock);
4383 if (st == active) {
4384 restart_array(mddev);
4385 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4386 md_wakeup_thread(mddev->thread);
4387 wake_up(&mddev->sb_wait);
4388 } else /* st == clean */ {
4389 restart_array(mddev);
4390 if (!set_in_sync(mddev))
4391 err = -EBUSY;
4392 }
4393 if (!err)
4394 sysfs_notify_dirent_safe(mddev->sysfs_state);
4395 spin_unlock(&mddev->lock);
4396 return err ?: len;
4397 }
4398 err = mddev_lock(mddev);
4399 if (err)
4400 return err;
4401 err = -EINVAL;
4402 switch(st) {
4403 case bad_word:
4404 break;
4405 case clear:
4406 /* stopping an active array */
4407 err = do_md_stop(mddev, 0, NULL);
4408 break;
4409 case inactive:
4410 /* stopping an active array */
4411 if (mddev->pers)
4412 err = do_md_stop(mddev, 2, NULL);
4413 else
4414 err = 0; /* already inactive */
4415 break;
4416 case suspended:
4417 break; /* not supported yet */
4418 case readonly:
4419 if (mddev->pers)
4420 err = md_set_readonly(mddev, NULL);
4421 else {
4422 mddev->ro = 1;
4423 set_disk_ro(mddev->gendisk, 1);
4424 err = do_md_run(mddev);
4425 }
4426 break;
4427 case read_auto:
4428 if (mddev->pers) {
4429 if (mddev->ro == 0)
4430 err = md_set_readonly(mddev, NULL);
4431 else if (mddev->ro == 1)
4432 err = restart_array(mddev);
4433 if (err == 0) {
4434 mddev->ro = 2;
4435 set_disk_ro(mddev->gendisk, 0);
4436 }
4437 } else {
4438 mddev->ro = 2;
4439 err = do_md_run(mddev);
4440 }
4441 break;
4442 case clean:
4443 if (mddev->pers) {
4444 err = restart_array(mddev);
4445 if (err)
4446 break;
4447 spin_lock(&mddev->lock);
4448 if (!set_in_sync(mddev))
4449 err = -EBUSY;
4450 spin_unlock(&mddev->lock);
4451 } else
4452 err = -EINVAL;
4453 break;
4454 case active:
4455 if (mddev->pers) {
4456 err = restart_array(mddev);
4457 if (err)
4458 break;
4459 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4460 wake_up(&mddev->sb_wait);
4461 err = 0;
4462 } else {
4463 mddev->ro = 0;
4464 set_disk_ro(mddev->gendisk, 0);
4465 err = do_md_run(mddev);
4466 }
4467 break;
4468 case write_pending:
4469 case active_idle:
4470 case broken:
4471 /* these cannot be set */
4472 break;
4473 }
4474
4475 if (!err) {
4476 if (mddev->hold_active == UNTIL_IOCTL)
4477 mddev->hold_active = 0;
4478 sysfs_notify_dirent_safe(mddev->sysfs_state);
4479 }
4480 mddev_unlock(mddev);
4481 return err ?: len;
4482}
4483static struct md_sysfs_entry md_array_state =
4484__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4485
4486static ssize_t
4487max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4488 return sprintf(page, "%d\n",
4489 atomic_read(&mddev->max_corr_read_errors));
4490}
4491
4492static ssize_t
4493max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4494{
4495 unsigned int n;
4496 int rv;
4497
4498 rv = kstrtouint(buf, 10, &n);
4499 if (rv < 0)
4500 return rv;
4501 atomic_set(&mddev->max_corr_read_errors, n);
4502 return len;
4503}
4504
4505static struct md_sysfs_entry max_corr_read_errors =
4506__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4507 max_corrected_read_errors_store);
4508
4509static ssize_t
4510null_show(struct mddev *mddev, char *page)
4511{
4512 return -EINVAL;
4513}
4514
4515static ssize_t
4516new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4517{
4518 /* buf must be %d:%d\n? giving major and minor numbers */
4519 /* The new device is added to the array.
4520 * If the array has a persistent superblock, we read the
4521 * superblock to initialise info and check validity.
4522 * Otherwise, only checking done is that in bind_rdev_to_array,
4523 * which mainly checks size.
4524 */
4525 char *e;
4526 int major = simple_strtoul(buf, &e, 10);
4527 int minor;
4528 dev_t dev;
4529 struct md_rdev *rdev;
4530 int err;
4531
4532 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4533 return -EINVAL;
4534 minor = simple_strtoul(e+1, &e, 10);
4535 if (*e && *e != '\n')
4536 return -EINVAL;
4537 dev = MKDEV(major, minor);
4538 if (major != MAJOR(dev) ||
4539 minor != MINOR(dev))
4540 return -EOVERFLOW;
4541
4542 flush_workqueue(md_misc_wq);
4543
4544 err = mddev_lock(mddev);
4545 if (err)
4546 return err;
4547 if (mddev->persistent) {
4548 rdev = md_import_device(dev, mddev->major_version,
4549 mddev->minor_version);
4550 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4551 struct md_rdev *rdev0
4552 = list_entry(mddev->disks.next,
4553 struct md_rdev, same_set);
4554 err = super_types[mddev->major_version]
4555 .load_super(rdev, rdev0, mddev->minor_version);
4556 if (err < 0)
4557 goto out;
4558 }
4559 } else if (mddev->external)
4560 rdev = md_import_device(dev, -2, -1);
4561 else
4562 rdev = md_import_device(dev, -1, -1);
4563
4564 if (IS_ERR(rdev)) {
4565 mddev_unlock(mddev);
4566 return PTR_ERR(rdev);
4567 }
4568 err = bind_rdev_to_array(rdev, mddev);
4569 out:
4570 if (err)
4571 export_rdev(rdev);
4572 mddev_unlock(mddev);
4573 if (!err)
4574 md_new_event(mddev);
4575 return err ? err : len;
4576}
4577
4578static struct md_sysfs_entry md_new_device =
4579__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4580
4581static ssize_t
4582bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4583{
4584 char *end;
4585 unsigned long chunk, end_chunk;
4586 int err;
4587
4588 err = mddev_lock(mddev);
4589 if (err)
4590 return err;
4591 if (!mddev->bitmap)
4592 goto out;
4593 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4594 while (*buf) {
4595 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4596 if (buf == end) break;
4597 if (*end == '-') { /* range */
4598 buf = end + 1;
4599 end_chunk = simple_strtoul(buf, &end, 0);
4600 if (buf == end) break;
4601 }
4602 if (*end && !isspace(*end)) break;
4603 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4604 buf = skip_spaces(end);
4605 }
4606 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4607out:
4608 mddev_unlock(mddev);
4609 return len;
4610}
4611
4612static struct md_sysfs_entry md_bitmap =
4613__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4614
4615static ssize_t
4616size_show(struct mddev *mddev, char *page)
4617{
4618 return sprintf(page, "%llu\n",
4619 (unsigned long long)mddev->dev_sectors / 2);
4620}
4621
4622static int update_size(struct mddev *mddev, sector_t num_sectors);
4623
4624static ssize_t
4625size_store(struct mddev *mddev, const char *buf, size_t len)
4626{
4627 /* If array is inactive, we can reduce the component size, but
4628 * not increase it (except from 0).
4629 * If array is active, we can try an on-line resize
4630 */
4631 sector_t sectors;
4632 int err = strict_blocks_to_sectors(buf, &sectors);
4633
4634 if (err < 0)
4635 return err;
4636 err = mddev_lock(mddev);
4637 if (err)
4638 return err;
4639 if (mddev->pers) {
4640 err = update_size(mddev, sectors);
4641 if (err == 0)
4642 md_update_sb(mddev, 1);
4643 } else {
4644 if (mddev->dev_sectors == 0 ||
4645 mddev->dev_sectors > sectors)
4646 mddev->dev_sectors = sectors;
4647 else
4648 err = -ENOSPC;
4649 }
4650 mddev_unlock(mddev);
4651 return err ? err : len;
4652}
4653
4654static struct md_sysfs_entry md_size =
4655__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4656
4657/* Metadata version.
4658 * This is one of
4659 * 'none' for arrays with no metadata (good luck...)
4660 * 'external' for arrays with externally managed metadata,
4661 * or N.M for internally known formats
4662 */
4663static ssize_t
4664metadata_show(struct mddev *mddev, char *page)
4665{
4666 if (mddev->persistent)
4667 return sprintf(page, "%d.%d\n",
4668 mddev->major_version, mddev->minor_version);
4669 else if (mddev->external)
4670 return sprintf(page, "external:%s\n", mddev->metadata_type);
4671 else
4672 return sprintf(page, "none\n");
4673}
4674
4675static ssize_t
4676metadata_store(struct mddev *mddev, const char *buf, size_t len)
4677{
4678 int major, minor;
4679 char *e;
4680 int err;
4681 /* Changing the details of 'external' metadata is
4682 * always permitted. Otherwise there must be
4683 * no devices attached to the array.
4684 */
4685
4686 err = mddev_lock(mddev);
4687 if (err)
4688 return err;
4689 err = -EBUSY;
4690 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4691 ;
4692 else if (!list_empty(&mddev->disks))
4693 goto out_unlock;
4694
4695 err = 0;
4696 if (cmd_match(buf, "none")) {
4697 mddev->persistent = 0;
4698 mddev->external = 0;
4699 mddev->major_version = 0;
4700 mddev->minor_version = 90;
4701 goto out_unlock;
4702 }
4703 if (strncmp(buf, "external:", 9) == 0) {
4704 size_t namelen = len-9;
4705 if (namelen >= sizeof(mddev->metadata_type))
4706 namelen = sizeof(mddev->metadata_type)-1;
4707 strncpy(mddev->metadata_type, buf+9, namelen);
4708 mddev->metadata_type[namelen] = 0;
4709 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4710 mddev->metadata_type[--namelen] = 0;
4711 mddev->persistent = 0;
4712 mddev->external = 1;
4713 mddev->major_version = 0;
4714 mddev->minor_version = 90;
4715 goto out_unlock;
4716 }
4717 major = simple_strtoul(buf, &e, 10);
4718 err = -EINVAL;
4719 if (e==buf || *e != '.')
4720 goto out_unlock;
4721 buf = e+1;
4722 minor = simple_strtoul(buf, &e, 10);
4723 if (e==buf || (*e && *e != '\n') )
4724 goto out_unlock;
4725 err = -ENOENT;
4726 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4727 goto out_unlock;
4728 mddev->major_version = major;
4729 mddev->minor_version = minor;
4730 mddev->persistent = 1;
4731 mddev->external = 0;
4732 err = 0;
4733out_unlock:
4734 mddev_unlock(mddev);
4735 return err ?: len;
4736}
4737
4738static struct md_sysfs_entry md_metadata =
4739__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4740
4741static ssize_t
4742action_show(struct mddev *mddev, char *page)
4743{
4744 char *type = "idle";
4745 unsigned long recovery = mddev->recovery;
4746 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4747 type = "frozen";
4748 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4749 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4750 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4751 type = "reshape";
4752 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4753 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4754 type = "resync";
4755 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4756 type = "check";
4757 else
4758 type = "repair";
4759 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4760 type = "recover";
4761 else if (mddev->reshape_position != MaxSector)
4762 type = "reshape";
4763 }
4764 return sprintf(page, "%s\n", type);
4765}
4766
4767static ssize_t
4768action_store(struct mddev *mddev, const char *page, size_t len)
4769{
4770 if (!mddev->pers || !mddev->pers->sync_request)
4771 return -EINVAL;
4772
4773
4774 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4775 if (cmd_match(page, "frozen"))
4776 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4777 else
4778 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4779 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4780 mddev_lock(mddev) == 0) {
4781 flush_workqueue(md_misc_wq);
4782 if (mddev->sync_thread) {
4783 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4784 md_reap_sync_thread(mddev);
4785 }
4786 mddev_unlock(mddev);
4787 }
4788 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4789 return -EBUSY;
4790 else if (cmd_match(page, "resync"))
4791 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4792 else if (cmd_match(page, "recover")) {
4793 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4794 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4795 } else if (cmd_match(page, "reshape")) {
4796 int err;
4797 if (mddev->pers->start_reshape == NULL)
4798 return -EINVAL;
4799 err = mddev_lock(mddev);
4800 if (!err) {
4801 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4802 err = -EBUSY;
4803 else {
4804 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4805 err = mddev->pers->start_reshape(mddev);
4806 }
4807 mddev_unlock(mddev);
4808 }
4809 if (err)
4810 return err;
4811 sysfs_notify(&mddev->kobj, NULL, "degraded");
4812 } else {
4813 if (cmd_match(page, "check"))
4814 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4815 else if (!cmd_match(page, "repair"))
4816 return -EINVAL;
4817 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4818 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4819 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4820 }
4821 if (mddev->ro == 2) {
4822 /* A write to sync_action is enough to justify
4823 * canceling read-auto mode
4824 */
4825 mddev->ro = 0;
4826 md_wakeup_thread(mddev->sync_thread);
4827 }
4828 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4829 md_wakeup_thread(mddev->thread);
4830 sysfs_notify_dirent_safe(mddev->sysfs_action);
4831 return len;
4832}
4833
4834static struct md_sysfs_entry md_scan_mode =
4835__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4836
4837static ssize_t
4838last_sync_action_show(struct mddev *mddev, char *page)
4839{
4840 return sprintf(page, "%s\n", mddev->last_sync_action);
4841}
4842
4843static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4844
4845static ssize_t
4846mismatch_cnt_show(struct mddev *mddev, char *page)
4847{
4848 return sprintf(page, "%llu\n",
4849 (unsigned long long)
4850 atomic64_read(&mddev->resync_mismatches));
4851}
4852
4853static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4854
4855static ssize_t
4856sync_min_show(struct mddev *mddev, char *page)
4857{
4858 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4859 mddev->sync_speed_min ? "local": "system");
4860}
4861
4862static ssize_t
4863sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4864{
4865 unsigned int min;
4866 int rv;
4867
4868 if (strncmp(buf, "system", 6)==0) {
4869 min = 0;
4870 } else {
4871 rv = kstrtouint(buf, 10, &min);
4872 if (rv < 0)
4873 return rv;
4874 if (min == 0)
4875 return -EINVAL;
4876 }
4877 mddev->sync_speed_min = min;
4878 return len;
4879}
4880
4881static struct md_sysfs_entry md_sync_min =
4882__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4883
4884static ssize_t
4885sync_max_show(struct mddev *mddev, char *page)
4886{
4887 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4888 mddev->sync_speed_max ? "local": "system");
4889}
4890
4891static ssize_t
4892sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4893{
4894 unsigned int max;
4895 int rv;
4896
4897 if (strncmp(buf, "system", 6)==0) {
4898 max = 0;
4899 } else {
4900 rv = kstrtouint(buf, 10, &max);
4901 if (rv < 0)
4902 return rv;
4903 if (max == 0)
4904 return -EINVAL;
4905 }
4906 mddev->sync_speed_max = max;
4907 return len;
4908}
4909
4910static struct md_sysfs_entry md_sync_max =
4911__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4912
4913static ssize_t
4914degraded_show(struct mddev *mddev, char *page)
4915{
4916 return sprintf(page, "%d\n", mddev->degraded);
4917}
4918static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4919
4920static ssize_t
4921sync_force_parallel_show(struct mddev *mddev, char *page)
4922{
4923 return sprintf(page, "%d\n", mddev->parallel_resync);
4924}
4925
4926static ssize_t
4927sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4928{
4929 long n;
4930
4931 if (kstrtol(buf, 10, &n))
4932 return -EINVAL;
4933
4934 if (n != 0 && n != 1)
4935 return -EINVAL;
4936
4937 mddev->parallel_resync = n;
4938
4939 if (mddev->sync_thread)
4940 wake_up(&resync_wait);
4941
4942 return len;
4943}
4944
4945/* force parallel resync, even with shared block devices */
4946static struct md_sysfs_entry md_sync_force_parallel =
4947__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4948 sync_force_parallel_show, sync_force_parallel_store);
4949
4950static ssize_t
4951sync_speed_show(struct mddev *mddev, char *page)
4952{
4953 unsigned long resync, dt, db;
4954 if (mddev->curr_resync == 0)
4955 return sprintf(page, "none\n");
4956 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4957 dt = (jiffies - mddev->resync_mark) / HZ;
4958 if (!dt) dt++;
4959 db = resync - mddev->resync_mark_cnt;
4960 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4961}
4962
4963static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4964
4965static ssize_t
4966sync_completed_show(struct mddev *mddev, char *page)
4967{
4968 unsigned long long max_sectors, resync;
4969
4970 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4971 return sprintf(page, "none\n");
4972
4973 if (mddev->curr_resync == 1 ||
4974 mddev->curr_resync == 2)
4975 return sprintf(page, "delayed\n");
4976
4977 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4978 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4979 max_sectors = mddev->resync_max_sectors;
4980 else
4981 max_sectors = mddev->dev_sectors;
4982
4983 resync = mddev->curr_resync_completed;
4984 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4985}
4986
4987static struct md_sysfs_entry md_sync_completed =
4988 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4989
4990static ssize_t
4991min_sync_show(struct mddev *mddev, char *page)
4992{
4993 return sprintf(page, "%llu\n",
4994 (unsigned long long)mddev->resync_min);
4995}
4996static ssize_t
4997min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4998{
4999 unsigned long long min;
5000 int err;
5001
5002 if (kstrtoull(buf, 10, &min))
5003 return -EINVAL;
5004
5005 spin_lock(&mddev->lock);
5006 err = -EINVAL;
5007 if (min > mddev->resync_max)
5008 goto out_unlock;
5009
5010 err = -EBUSY;
5011 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5012 goto out_unlock;
5013
5014 /* Round down to multiple of 4K for safety */
5015 mddev->resync_min = round_down(min, 8);
5016 err = 0;
5017
5018out_unlock:
5019 spin_unlock(&mddev->lock);
5020 return err ?: len;
5021}
5022
5023static struct md_sysfs_entry md_min_sync =
5024__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5025
5026static ssize_t
5027max_sync_show(struct mddev *mddev, char *page)
5028{
5029 if (mddev->resync_max == MaxSector)
5030 return sprintf(page, "max\n");
5031 else
5032 return sprintf(page, "%llu\n",
5033 (unsigned long long)mddev->resync_max);
5034}
5035static ssize_t
5036max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5037{
5038 int err;
5039 spin_lock(&mddev->lock);
5040 if (strncmp(buf, "max", 3) == 0)
5041 mddev->resync_max = MaxSector;
5042 else {
5043 unsigned long long max;
5044 int chunk;
5045
5046 err = -EINVAL;
5047 if (kstrtoull(buf, 10, &max))
5048 goto out_unlock;
5049 if (max < mddev->resync_min)
5050 goto out_unlock;
5051
5052 err = -EBUSY;
5053 if (max < mddev->resync_max &&
5054 mddev->ro == 0 &&
5055 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5056 goto out_unlock;
5057
5058 /* Must be a multiple of chunk_size */
5059 chunk = mddev->chunk_sectors;
5060 if (chunk) {
5061 sector_t temp = max;
5062
5063 err = -EINVAL;
5064 if (sector_div(temp, chunk))
5065 goto out_unlock;
5066 }
5067 mddev->resync_max = max;
5068 }
5069 wake_up(&mddev->recovery_wait);
5070 err = 0;
5071out_unlock:
5072 spin_unlock(&mddev->lock);
5073 return err ?: len;
5074}
5075
5076static struct md_sysfs_entry md_max_sync =
5077__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5078
5079static ssize_t
5080suspend_lo_show(struct mddev *mddev, char *page)
5081{
5082 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
5083}
5084
5085static ssize_t
5086suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5087{
5088 unsigned long long new;
5089 int err;
5090
5091 err = kstrtoull(buf, 10, &new);
5092 if (err < 0)
5093 return err;
5094 if (new != (sector_t)new)
5095 return -EINVAL;
5096
5097 err = mddev_lock(mddev);
5098 if (err)
5099 return err;
5100 err = -EINVAL;
5101 if (mddev->pers == NULL ||
5102 mddev->pers->quiesce == NULL)
5103 goto unlock;
5104 mddev_suspend(mddev);
5105 mddev->suspend_lo = new;
5106 mddev_resume(mddev);
5107
5108 err = 0;
5109unlock:
5110 mddev_unlock(mddev);
5111 return err ?: len;
5112}
5113static struct md_sysfs_entry md_suspend_lo =
5114__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5115
5116static ssize_t
5117suspend_hi_show(struct mddev *mddev, char *page)
5118{
5119 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5120}
5121
5122static ssize_t
5123suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5124{
5125 unsigned long long new;
5126 int err;
5127
5128 err = kstrtoull(buf, 10, &new);
5129 if (err < 0)
5130 return err;
5131 if (new != (sector_t)new)
5132 return -EINVAL;
5133
5134 err = mddev_lock(mddev);
5135 if (err)
5136 return err;
5137 err = -EINVAL;
5138 if (mddev->pers == NULL)
5139 goto unlock;
5140
5141 mddev_suspend(mddev);
5142 mddev->suspend_hi = new;
5143 mddev_resume(mddev);
5144
5145 err = 0;
5146unlock:
5147 mddev_unlock(mddev);
5148 return err ?: len;
5149}
5150static struct md_sysfs_entry md_suspend_hi =
5151__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5152
5153static ssize_t
5154reshape_position_show(struct mddev *mddev, char *page)
5155{
5156 if (mddev->reshape_position != MaxSector)
5157 return sprintf(page, "%llu\n",
5158 (unsigned long long)mddev->reshape_position);
5159 strcpy(page, "none\n");
5160 return 5;
5161}
5162
5163static ssize_t
5164reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5165{
5166 struct md_rdev *rdev;
5167 unsigned long long new;
5168 int err;
5169
5170 err = kstrtoull(buf, 10, &new);
5171 if (err < 0)
5172 return err;
5173 if (new != (sector_t)new)
5174 return -EINVAL;
5175 err = mddev_lock(mddev);
5176 if (err)
5177 return err;
5178 err = -EBUSY;
5179 if (mddev->pers)
5180 goto unlock;
5181 mddev->reshape_position = new;
5182 mddev->delta_disks = 0;
5183 mddev->reshape_backwards = 0;
5184 mddev->new_level = mddev->level;
5185 mddev->new_layout = mddev->layout;
5186 mddev->new_chunk_sectors = mddev->chunk_sectors;
5187 rdev_for_each(rdev, mddev)
5188 rdev->new_data_offset = rdev->data_offset;
5189 err = 0;
5190unlock:
5191 mddev_unlock(mddev);
5192 return err ?: len;
5193}
5194
5195static struct md_sysfs_entry md_reshape_position =
5196__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5197 reshape_position_store);
5198
5199static ssize_t
5200reshape_direction_show(struct mddev *mddev, char *page)
5201{
5202 return sprintf(page, "%s\n",
5203 mddev->reshape_backwards ? "backwards" : "forwards");
5204}
5205
5206static ssize_t
5207reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5208{
5209 int backwards = 0;
5210 int err;
5211
5212 if (cmd_match(buf, "forwards"))
5213 backwards = 0;
5214 else if (cmd_match(buf, "backwards"))
5215 backwards = 1;
5216 else
5217 return -EINVAL;
5218 if (mddev->reshape_backwards == backwards)
5219 return len;
5220
5221 err = mddev_lock(mddev);
5222 if (err)
5223 return err;
5224 /* check if we are allowed to change */
5225 if (mddev->delta_disks)
5226 err = -EBUSY;
5227 else if (mddev->persistent &&
5228 mddev->major_version == 0)
5229 err = -EINVAL;
5230 else
5231 mddev->reshape_backwards = backwards;
5232 mddev_unlock(mddev);
5233 return err ?: len;
5234}
5235
5236static struct md_sysfs_entry md_reshape_direction =
5237__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5238 reshape_direction_store);
5239
5240static ssize_t
5241array_size_show(struct mddev *mddev, char *page)
5242{
5243 if (mddev->external_size)
5244 return sprintf(page, "%llu\n",
5245 (unsigned long long)mddev->array_sectors/2);
5246 else
5247 return sprintf(page, "default\n");
5248}
5249
5250static ssize_t
5251array_size_store(struct mddev *mddev, const char *buf, size_t len)
5252{
5253 sector_t sectors;
5254 int err;
5255
5256 err = mddev_lock(mddev);
5257 if (err)
5258 return err;
5259
5260 /* cluster raid doesn't support change array_sectors */
5261 if (mddev_is_clustered(mddev)) {
5262 mddev_unlock(mddev);
5263 return -EINVAL;
5264 }
5265
5266 if (strncmp(buf, "default", 7) == 0) {
5267 if (mddev->pers)
5268 sectors = mddev->pers->size(mddev, 0, 0);
5269 else
5270 sectors = mddev->array_sectors;
5271
5272 mddev->external_size = 0;
5273 } else {
5274 if (strict_blocks_to_sectors(buf, &sectors) < 0)
5275 err = -EINVAL;
5276 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5277 err = -E2BIG;
5278 else
5279 mddev->external_size = 1;
5280 }
5281
5282 if (!err) {
5283 mddev->array_sectors = sectors;
5284 if (mddev->pers) {
5285 set_capacity(mddev->gendisk, mddev->array_sectors);
5286 revalidate_disk(mddev->gendisk);
5287 }
5288 }
5289 mddev_unlock(mddev);
5290 return err ?: len;
5291}
5292
5293static struct md_sysfs_entry md_array_size =
5294__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5295 array_size_store);
5296
5297static ssize_t
5298consistency_policy_show(struct mddev *mddev, char *page)
5299{
5300 int ret;
5301
5302 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5303 ret = sprintf(page, "journal\n");
5304 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5305 ret = sprintf(page, "ppl\n");
5306 } else if (mddev->bitmap) {
5307 ret = sprintf(page, "bitmap\n");
5308 } else if (mddev->pers) {
5309 if (mddev->pers->sync_request)
5310 ret = sprintf(page, "resync\n");
5311 else
5312 ret = sprintf(page, "none\n");
5313 } else {
5314 ret = sprintf(page, "unknown\n");
5315 }
5316
5317 return ret;
5318}
5319
5320static ssize_t
5321consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5322{
5323 int err = 0;
5324
5325 if (mddev->pers) {
5326 if (mddev->pers->change_consistency_policy)
5327 err = mddev->pers->change_consistency_policy(mddev, buf);
5328 else
5329 err = -EBUSY;
5330 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5331 set_bit(MD_HAS_PPL, &mddev->flags);
5332 } else {
5333 err = -EINVAL;
5334 }
5335
5336 return err ? err : len;
5337}
5338
5339static struct md_sysfs_entry md_consistency_policy =
5340__ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5341 consistency_policy_store);
5342
5343static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5344{
5345 return sprintf(page, "%d\n", mddev->fail_last_dev);
5346}
5347
5348/*
5349 * Setting fail_last_dev to true to allow last device to be forcibly removed
5350 * from RAID1/RAID10.
5351 */
5352static ssize_t
5353fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5354{
5355 int ret;
5356 bool value;
5357
5358 ret = kstrtobool(buf, &value);
5359 if (ret)
5360 return ret;
5361
5362 if (value != mddev->fail_last_dev)
5363 mddev->fail_last_dev = value;
5364
5365 return len;
5366}
5367static struct md_sysfs_entry md_fail_last_dev =
5368__ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5369 fail_last_dev_store);
5370
5371static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5372{
5373 if (mddev->pers == NULL || (mddev->pers->level != 1))
5374 return sprintf(page, "n/a\n");
5375 else
5376 return sprintf(page, "%d\n", mddev->serialize_policy);
5377}
5378
5379/*
5380 * Setting serialize_policy to true to enforce write IO is not reordered
5381 * for raid1.
5382 */
5383static ssize_t
5384serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5385{
5386 int err;
5387 bool value;
5388
5389 err = kstrtobool(buf, &value);
5390 if (err)
5391 return err;
5392
5393 if (value == mddev->serialize_policy)
5394 return len;
5395
5396 err = mddev_lock(mddev);
5397 if (err)
5398 return err;
5399 if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5400 pr_err("md: serialize_policy is only effective for raid1\n");
5401 err = -EINVAL;
5402 goto unlock;
5403 }
5404
5405 mddev_suspend(mddev);
5406 if (value)
5407 mddev_create_serial_pool(mddev, NULL, true);
5408 else
5409 mddev_destroy_serial_pool(mddev, NULL, true);
5410 mddev->serialize_policy = value;
5411 mddev_resume(mddev);
5412unlock:
5413 mddev_unlock(mddev);
5414 return err ?: len;
5415}
5416
5417static struct md_sysfs_entry md_serialize_policy =
5418__ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5419 serialize_policy_store);
5420
5421
5422static struct attribute *md_default_attrs[] = {
5423 &md_level.attr,
5424 &md_layout.attr,
5425 &md_raid_disks.attr,
5426 &md_chunk_size.attr,
5427 &md_size.attr,
5428 &md_resync_start.attr,
5429 &md_metadata.attr,
5430 &md_new_device.attr,
5431 &md_safe_delay.attr,
5432 &md_array_state.attr,
5433 &md_reshape_position.attr,
5434 &md_reshape_direction.attr,
5435 &md_array_size.attr,
5436 &max_corr_read_errors.attr,
5437 &md_consistency_policy.attr,
5438 &md_fail_last_dev.attr,
5439 &md_serialize_policy.attr,
5440 NULL,
5441};
5442
5443static struct attribute *md_redundancy_attrs[] = {
5444 &md_scan_mode.attr,
5445 &md_last_scan_mode.attr,
5446 &md_mismatches.attr,
5447 &md_sync_min.attr,
5448 &md_sync_max.attr,
5449 &md_sync_speed.attr,
5450 &md_sync_force_parallel.attr,
5451 &md_sync_completed.attr,
5452 &md_min_sync.attr,
5453 &md_max_sync.attr,
5454 &md_suspend_lo.attr,
5455 &md_suspend_hi.attr,
5456 &md_bitmap.attr,
5457 &md_degraded.attr,
5458 NULL,
5459};
5460static struct attribute_group md_redundancy_group = {
5461 .name = NULL,
5462 .attrs = md_redundancy_attrs,
5463};
5464
5465static ssize_t
5466md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5467{
5468 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5469 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5470 ssize_t rv;
5471
5472 if (!entry->show)
5473 return -EIO;
5474 spin_lock(&all_mddevs_lock);
5475 if (list_empty(&mddev->all_mddevs)) {
5476 spin_unlock(&all_mddevs_lock);
5477 return -EBUSY;
5478 }
5479 mddev_get(mddev);
5480 spin_unlock(&all_mddevs_lock);
5481
5482 rv = entry->show(mddev, page);
5483 mddev_put(mddev);
5484 return rv;
5485}
5486
5487static ssize_t
5488md_attr_store(struct kobject *kobj, struct attribute *attr,
5489 const char *page, size_t length)
5490{
5491 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5492 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5493 ssize_t rv;
5494
5495 if (!entry->store)
5496 return -EIO;
5497 if (!capable(CAP_SYS_ADMIN))
5498 return -EACCES;
5499 spin_lock(&all_mddevs_lock);
5500 if (list_empty(&mddev->all_mddevs)) {
5501 spin_unlock(&all_mddevs_lock);
5502 return -EBUSY;
5503 }
5504 mddev_get(mddev);
5505 spin_unlock(&all_mddevs_lock);
5506 rv = entry->store(mddev, page, length);
5507 mddev_put(mddev);
5508 return rv;
5509}
5510
5511static void md_free(struct kobject *ko)
5512{
5513 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5514
5515 if (mddev->sysfs_state)
5516 sysfs_put(mddev->sysfs_state);
5517
5518 if (mddev->gendisk)
5519 del_gendisk(mddev->gendisk);
5520 if (mddev->queue)
5521 blk_cleanup_queue(mddev->queue);
5522 if (mddev->gendisk)
5523 put_disk(mddev->gendisk);
5524 percpu_ref_exit(&mddev->writes_pending);
5525
5526 bioset_exit(&mddev->bio_set);
5527 bioset_exit(&mddev->sync_set);
5528 kfree(mddev);
5529}
5530
5531static const struct sysfs_ops md_sysfs_ops = {
5532 .show = md_attr_show,
5533 .store = md_attr_store,
5534};
5535static struct kobj_type md_ktype = {
5536 .release = md_free,
5537 .sysfs_ops = &md_sysfs_ops,
5538 .default_attrs = md_default_attrs,
5539};
5540
5541int mdp_major = 0;
5542
5543static void mddev_delayed_delete(struct work_struct *ws)
5544{
5545 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5546
5547 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5548 kobject_del(&mddev->kobj);
5549 kobject_put(&mddev->kobj);
5550}
5551
5552static void no_op(struct percpu_ref *r) {}
5553
5554int mddev_init_writes_pending(struct mddev *mddev)
5555{
5556 if (mddev->writes_pending.percpu_count_ptr)
5557 return 0;
5558 if (percpu_ref_init(&mddev->writes_pending, no_op,
5559 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5560 return -ENOMEM;
5561 /* We want to start with the refcount at zero */
5562 percpu_ref_put(&mddev->writes_pending);
5563 return 0;
5564}
5565EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5566
5567static int md_alloc(dev_t dev, char *name)
5568{
5569 /*
5570 * If dev is zero, name is the name of a device to allocate with
5571 * an arbitrary minor number. It will be "md_???"
5572 * If dev is non-zero it must be a device number with a MAJOR of
5573 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5574 * the device is being created by opening a node in /dev.
5575 * If "name" is not NULL, the device is being created by
5576 * writing to /sys/module/md_mod/parameters/new_array.
5577 */
5578 static DEFINE_MUTEX(disks_mutex);
5579 struct mddev *mddev = mddev_find(dev);
5580 struct gendisk *disk;
5581 int partitioned;
5582 int shift;
5583 int unit;
5584 int error;
5585
5586 if (!mddev)
5587 return -ENODEV;
5588
5589 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5590 shift = partitioned ? MdpMinorShift : 0;
5591 unit = MINOR(mddev->unit) >> shift;
5592
5593 /* wait for any previous instance of this device to be
5594 * completely removed (mddev_delayed_delete).
5595 */
5596 flush_workqueue(md_misc_wq);
5597
5598 mutex_lock(&disks_mutex);
5599 error = -EEXIST;
5600 if (mddev->gendisk)
5601 goto abort;
5602
5603 if (name && !dev) {
5604 /* Need to ensure that 'name' is not a duplicate.
5605 */
5606 struct mddev *mddev2;
5607 spin_lock(&all_mddevs_lock);
5608
5609 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5610 if (mddev2->gendisk &&
5611 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5612 spin_unlock(&all_mddevs_lock);
5613 goto abort;
5614 }
5615 spin_unlock(&all_mddevs_lock);
5616 }
5617 if (name && dev)
5618 /*
5619 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5620 */
5621 mddev->hold_active = UNTIL_STOP;
5622
5623 error = -ENOMEM;
5624 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5625 if (!mddev->queue)
5626 goto abort;
5627 mddev->queue->queuedata = mddev;
5628
5629 blk_queue_make_request(mddev->queue, md_make_request);
5630 blk_set_stacking_limits(&mddev->queue->limits);
5631
5632 disk = alloc_disk(1 << shift);
5633 if (!disk) {
5634 blk_cleanup_queue(mddev->queue);
5635 mddev->queue = NULL;
5636 goto abort;
5637 }
5638 disk->major = MAJOR(mddev->unit);
5639 disk->first_minor = unit << shift;
5640 if (name)
5641 strcpy(disk->disk_name, name);
5642 else if (partitioned)
5643 sprintf(disk->disk_name, "md_d%d", unit);
5644 else
5645 sprintf(disk->disk_name, "md%d", unit);
5646 disk->fops = &md_fops;
5647 disk->private_data = mddev;
5648 disk->queue = mddev->queue;
5649 blk_queue_write_cache(mddev->queue, true, true);
5650 /* Allow extended partitions. This makes the
5651 * 'mdp' device redundant, but we can't really
5652 * remove it now.
5653 */
5654 disk->flags |= GENHD_FL_EXT_DEVT;
5655 mddev->gendisk = disk;
5656 /* As soon as we call add_disk(), another thread could get
5657 * through to md_open, so make sure it doesn't get too far
5658 */
5659 mutex_lock(&mddev->open_mutex);
5660 add_disk(disk);
5661
5662 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5663 if (error) {
5664 /* This isn't possible, but as kobject_init_and_add is marked
5665 * __must_check, we must do something with the result
5666 */
5667 pr_debug("md: cannot register %s/md - name in use\n",
5668 disk->disk_name);
5669 error = 0;
5670 }
5671 if (mddev->kobj.sd &&
5672 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5673 pr_debug("pointless warning\n");
5674 mutex_unlock(&mddev->open_mutex);
5675 abort:
5676 mutex_unlock(&disks_mutex);
5677 if (!error && mddev->kobj.sd) {
5678 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5679 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5680 }
5681 mddev_put(mddev);
5682 return error;
5683}
5684
5685static struct kobject *md_probe(dev_t dev, int *part, void *data)
5686{
5687 if (create_on_open)
5688 md_alloc(dev, NULL);
5689 return NULL;
5690}
5691
5692static int add_named_array(const char *val, const struct kernel_param *kp)
5693{
5694 /*
5695 * val must be "md_*" or "mdNNN".
5696 * For "md_*" we allocate an array with a large free minor number, and
5697 * set the name to val. val must not already be an active name.
5698 * For "mdNNN" we allocate an array with the minor number NNN
5699 * which must not already be in use.
5700 */
5701 int len = strlen(val);
5702 char buf[DISK_NAME_LEN];
5703 unsigned long devnum;
5704
5705 while (len && val[len-1] == '\n')
5706 len--;
5707 if (len >= DISK_NAME_LEN)
5708 return -E2BIG;
5709 strlcpy(buf, val, len+1);
5710 if (strncmp(buf, "md_", 3) == 0)
5711 return md_alloc(0, buf);
5712 if (strncmp(buf, "md", 2) == 0 &&
5713 isdigit(buf[2]) &&
5714 kstrtoul(buf+2, 10, &devnum) == 0 &&
5715 devnum <= MINORMASK)
5716 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5717
5718 return -EINVAL;
5719}
5720
5721static void md_safemode_timeout(struct timer_list *t)
5722{
5723 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5724
5725 mddev->safemode = 1;
5726 if (mddev->external)
5727 sysfs_notify_dirent_safe(mddev->sysfs_state);
5728
5729 md_wakeup_thread(mddev->thread);
5730}
5731
5732static int start_dirty_degraded;
5733
5734int md_run(struct mddev *mddev)
5735{
5736 int err;
5737 struct md_rdev *rdev;
5738 struct md_personality *pers;
5739
5740 if (list_empty(&mddev->disks))
5741 /* cannot run an array with no devices.. */
5742 return -EINVAL;
5743
5744 if (mddev->pers)
5745 return -EBUSY;
5746 /* Cannot run until previous stop completes properly */
5747 if (mddev->sysfs_active)
5748 return -EBUSY;
5749
5750 /*
5751 * Analyze all RAID superblock(s)
5752 */
5753 if (!mddev->raid_disks) {
5754 if (!mddev->persistent)
5755 return -EINVAL;
5756 err = analyze_sbs(mddev);
5757 if (err)
5758 return -EINVAL;
5759 }
5760
5761 if (mddev->level != LEVEL_NONE)
5762 request_module("md-level-%d", mddev->level);
5763 else if (mddev->clevel[0])
5764 request_module("md-%s", mddev->clevel);
5765
5766 /*
5767 * Drop all container device buffers, from now on
5768 * the only valid external interface is through the md
5769 * device.
5770 */
5771 mddev->has_superblocks = false;
5772 rdev_for_each(rdev, mddev) {
5773 if (test_bit(Faulty, &rdev->flags))
5774 continue;
5775 sync_blockdev(rdev->bdev);
5776 invalidate_bdev(rdev->bdev);
5777 if (mddev->ro != 1 &&
5778 (bdev_read_only(rdev->bdev) ||
5779 bdev_read_only(rdev->meta_bdev))) {
5780 mddev->ro = 1;
5781 if (mddev->gendisk)
5782 set_disk_ro(mddev->gendisk, 1);
5783 }
5784
5785 if (rdev->sb_page)
5786 mddev->has_superblocks = true;
5787
5788 /* perform some consistency tests on the device.
5789 * We don't want the data to overlap the metadata,
5790 * Internal Bitmap issues have been handled elsewhere.
5791 */
5792 if (rdev->meta_bdev) {
5793 /* Nothing to check */;
5794 } else if (rdev->data_offset < rdev->sb_start) {
5795 if (mddev->dev_sectors &&
5796 rdev->data_offset + mddev->dev_sectors
5797 > rdev->sb_start) {
5798 pr_warn("md: %s: data overlaps metadata\n",
5799 mdname(mddev));
5800 return -EINVAL;
5801 }
5802 } else {
5803 if (rdev->sb_start + rdev->sb_size/512
5804 > rdev->data_offset) {
5805 pr_warn("md: %s: metadata overlaps data\n",
5806 mdname(mddev));
5807 return -EINVAL;
5808 }
5809 }
5810 sysfs_notify_dirent_safe(rdev->sysfs_state);
5811 }
5812
5813 if (!bioset_initialized(&mddev->bio_set)) {
5814 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5815 if (err)
5816 return err;
5817 }
5818 if (!bioset_initialized(&mddev->sync_set)) {
5819 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5820 if (err)
5821 return err;
5822 }
5823
5824 spin_lock(&pers_lock);
5825 pers = find_pers(mddev->level, mddev->clevel);
5826 if (!pers || !try_module_get(pers->owner)) {
5827 spin_unlock(&pers_lock);
5828 if (mddev->level != LEVEL_NONE)
5829 pr_warn("md: personality for level %d is not loaded!\n",
5830 mddev->level);
5831 else
5832 pr_warn("md: personality for level %s is not loaded!\n",
5833 mddev->clevel);
5834 err = -EINVAL;
5835 goto abort;
5836 }
5837 spin_unlock(&pers_lock);
5838 if (mddev->level != pers->level) {
5839 mddev->level = pers->level;
5840 mddev->new_level = pers->level;
5841 }
5842 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5843
5844 if (mddev->reshape_position != MaxSector &&
5845 pers->start_reshape == NULL) {
5846 /* This personality cannot handle reshaping... */
5847 module_put(pers->owner);
5848 err = -EINVAL;
5849 goto abort;
5850 }
5851
5852 if (pers->sync_request) {
5853 /* Warn if this is a potentially silly
5854 * configuration.
5855 */
5856 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5857 struct md_rdev *rdev2;
5858 int warned = 0;
5859
5860 rdev_for_each(rdev, mddev)
5861 rdev_for_each(rdev2, mddev) {
5862 if (rdev < rdev2 &&
5863 rdev->bdev->bd_contains ==
5864 rdev2->bdev->bd_contains) {
5865 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5866 mdname(mddev),
5867 bdevname(rdev->bdev,b),
5868 bdevname(rdev2->bdev,b2));
5869 warned = 1;
5870 }
5871 }
5872
5873 if (warned)
5874 pr_warn("True protection against single-disk failure might be compromised.\n");
5875 }
5876
5877 mddev->recovery = 0;
5878 /* may be over-ridden by personality */
5879 mddev->resync_max_sectors = mddev->dev_sectors;
5880
5881 mddev->ok_start_degraded = start_dirty_degraded;
5882
5883 if (start_readonly && mddev->ro == 0)
5884 mddev->ro = 2; /* read-only, but switch on first write */
5885
5886 err = pers->run(mddev);
5887 if (err)
5888 pr_warn("md: pers->run() failed ...\n");
5889 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5890 WARN_ONCE(!mddev->external_size,
5891 "%s: default size too small, but 'external_size' not in effect?\n",
5892 __func__);
5893 pr_warn("md: invalid array_size %llu > default size %llu\n",
5894 (unsigned long long)mddev->array_sectors / 2,
5895 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5896 err = -EINVAL;
5897 }
5898 if (err == 0 && pers->sync_request &&
5899 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5900 struct bitmap *bitmap;
5901
5902 bitmap = md_bitmap_create(mddev, -1);
5903 if (IS_ERR(bitmap)) {
5904 err = PTR_ERR(bitmap);
5905 pr_warn("%s: failed to create bitmap (%d)\n",
5906 mdname(mddev), err);
5907 } else
5908 mddev->bitmap = bitmap;
5909
5910 }
5911 if (err)
5912 goto bitmap_abort;
5913
5914 if (mddev->bitmap_info.max_write_behind > 0) {
5915 bool create_pool = false;
5916
5917 rdev_for_each(rdev, mddev) {
5918 if (test_bit(WriteMostly, &rdev->flags) &&
5919 rdev_init_serial(rdev))
5920 create_pool = true;
5921 }
5922 if (create_pool && mddev->serial_info_pool == NULL) {
5923 mddev->serial_info_pool =
5924 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
5925 sizeof(struct serial_info));
5926 if (!mddev->serial_info_pool) {
5927 err = -ENOMEM;
5928 goto bitmap_abort;
5929 }
5930 }
5931 }
5932
5933 if (mddev->queue) {
5934 bool nonrot = true;
5935
5936 rdev_for_each(rdev, mddev) {
5937 if (rdev->raid_disk >= 0 &&
5938 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5939 nonrot = false;
5940 break;
5941 }
5942 }
5943 if (mddev->degraded)
5944 nonrot = false;
5945 if (nonrot)
5946 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5947 else
5948 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5949 mddev->queue->backing_dev_info->congested_data = mddev;
5950 mddev->queue->backing_dev_info->congested_fn = md_congested;
5951 }
5952 if (pers->sync_request) {
5953 if (mddev->kobj.sd &&
5954 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5955 pr_warn("md: cannot register extra attributes for %s\n",
5956 mdname(mddev));
5957 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5958 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5959 mddev->ro = 0;
5960
5961 atomic_set(&mddev->max_corr_read_errors,
5962 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5963 mddev->safemode = 0;
5964 if (mddev_is_clustered(mddev))
5965 mddev->safemode_delay = 0;
5966 else
5967 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5968 mddev->in_sync = 1;
5969 smp_wmb();
5970 spin_lock(&mddev->lock);
5971 mddev->pers = pers;
5972 spin_unlock(&mddev->lock);
5973 rdev_for_each(rdev, mddev)
5974 if (rdev->raid_disk >= 0)
5975 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
5976
5977 if (mddev->degraded && !mddev->ro)
5978 /* This ensures that recovering status is reported immediately
5979 * via sysfs - until a lack of spares is confirmed.
5980 */
5981 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5982 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5983
5984 if (mddev->sb_flags)
5985 md_update_sb(mddev, 0);
5986
5987 md_new_event(mddev);
5988 return 0;
5989
5990bitmap_abort:
5991 mddev_detach(mddev);
5992 if (mddev->private)
5993 pers->free(mddev, mddev->private);
5994 mddev->private = NULL;
5995 module_put(pers->owner);
5996 md_bitmap_destroy(mddev);
5997abort:
5998 bioset_exit(&mddev->bio_set);
5999 bioset_exit(&mddev->sync_set);
6000 return err;
6001}
6002EXPORT_SYMBOL_GPL(md_run);
6003
6004static int do_md_run(struct mddev *mddev)
6005{
6006 int err;
6007
6008 set_bit(MD_NOT_READY, &mddev->flags);
6009 err = md_run(mddev);
6010 if (err)
6011 goto out;
6012 err = md_bitmap_load(mddev);
6013 if (err) {
6014 md_bitmap_destroy(mddev);
6015 goto out;
6016 }
6017
6018 if (mddev_is_clustered(mddev))
6019 md_allow_write(mddev);
6020
6021 /* run start up tasks that require md_thread */
6022 md_start(mddev);
6023
6024 md_wakeup_thread(mddev->thread);
6025 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6026
6027 set_capacity(mddev->gendisk, mddev->array_sectors);
6028 revalidate_disk(mddev->gendisk);
6029 clear_bit(MD_NOT_READY, &mddev->flags);
6030 mddev->changed = 1;
6031 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6032 sysfs_notify_dirent_safe(mddev->sysfs_state);
6033 sysfs_notify_dirent_safe(mddev->sysfs_action);
6034 sysfs_notify(&mddev->kobj, NULL, "degraded");
6035out:
6036 clear_bit(MD_NOT_READY, &mddev->flags);
6037 return err;
6038}
6039
6040int md_start(struct mddev *mddev)
6041{
6042 int ret = 0;
6043
6044 if (mddev->pers->start) {
6045 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6046 md_wakeup_thread(mddev->thread);
6047 ret = mddev->pers->start(mddev);
6048 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6049 md_wakeup_thread(mddev->sync_thread);
6050 }
6051 return ret;
6052}
6053EXPORT_SYMBOL_GPL(md_start);
6054
6055static int restart_array(struct mddev *mddev)
6056{
6057 struct gendisk *disk = mddev->gendisk;
6058 struct md_rdev *rdev;
6059 bool has_journal = false;
6060 bool has_readonly = false;
6061
6062 /* Complain if it has no devices */
6063 if (list_empty(&mddev->disks))
6064 return -ENXIO;
6065 if (!mddev->pers)
6066 return -EINVAL;
6067 if (!mddev->ro)
6068 return -EBUSY;
6069
6070 rcu_read_lock();
6071 rdev_for_each_rcu(rdev, mddev) {
6072 if (test_bit(Journal, &rdev->flags) &&
6073 !test_bit(Faulty, &rdev->flags))
6074 has_journal = true;
6075 if (bdev_read_only(rdev->bdev))
6076 has_readonly = true;
6077 }
6078 rcu_read_unlock();
6079 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6080 /* Don't restart rw with journal missing/faulty */
6081 return -EINVAL;
6082 if (has_readonly)
6083 return -EROFS;
6084
6085 mddev->safemode = 0;
6086 mddev->ro = 0;
6087 set_disk_ro(disk, 0);
6088 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6089 /* Kick recovery or resync if necessary */
6090 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6091 md_wakeup_thread(mddev->thread);
6092 md_wakeup_thread(mddev->sync_thread);
6093 sysfs_notify_dirent_safe(mddev->sysfs_state);
6094 return 0;
6095}
6096
6097static void md_clean(struct mddev *mddev)
6098{
6099 mddev->array_sectors = 0;
6100 mddev->external_size = 0;
6101 mddev->dev_sectors = 0;
6102 mddev->raid_disks = 0;
6103 mddev->recovery_cp = 0;
6104 mddev->resync_min = 0;
6105 mddev->resync_max = MaxSector;
6106 mddev->reshape_position = MaxSector;
6107 mddev->external = 0;
6108 mddev->persistent = 0;
6109 mddev->level = LEVEL_NONE;
6110 mddev->clevel[0] = 0;
6111 mddev->flags = 0;
6112 mddev->sb_flags = 0;
6113 mddev->ro = 0;
6114 mddev->metadata_type[0] = 0;
6115 mddev->chunk_sectors = 0;
6116 mddev->ctime = mddev->utime = 0;
6117 mddev->layout = 0;
6118 mddev->max_disks = 0;
6119 mddev->events = 0;
6120 mddev->can_decrease_events = 0;
6121 mddev->delta_disks = 0;
6122 mddev->reshape_backwards = 0;
6123 mddev->new_level = LEVEL_NONE;
6124 mddev->new_layout = 0;
6125 mddev->new_chunk_sectors = 0;
6126 mddev->curr_resync = 0;
6127 atomic64_set(&mddev->resync_mismatches, 0);
6128 mddev->suspend_lo = mddev->suspend_hi = 0;
6129 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6130 mddev->recovery = 0;
6131 mddev->in_sync = 0;
6132 mddev->changed = 0;
6133 mddev->degraded = 0;
6134 mddev->safemode = 0;
6135 mddev->private = NULL;
6136 mddev->cluster_info = NULL;
6137 mddev->bitmap_info.offset = 0;
6138 mddev->bitmap_info.default_offset = 0;
6139 mddev->bitmap_info.default_space = 0;
6140 mddev->bitmap_info.chunksize = 0;
6141 mddev->bitmap_info.daemon_sleep = 0;
6142 mddev->bitmap_info.max_write_behind = 0;
6143 mddev->bitmap_info.nodes = 0;
6144}
6145
6146static void __md_stop_writes(struct mddev *mddev)
6147{
6148 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6149 flush_workqueue(md_misc_wq);
6150 if (mddev->sync_thread) {
6151 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6152 md_reap_sync_thread(mddev);
6153 }
6154
6155 del_timer_sync(&mddev->safemode_timer);
6156
6157 if (mddev->pers && mddev->pers->quiesce) {
6158 mddev->pers->quiesce(mddev, 1);
6159 mddev->pers->quiesce(mddev, 0);
6160 }
6161 md_bitmap_flush(mddev);
6162
6163 if (mddev->ro == 0 &&
6164 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6165 mddev->sb_flags)) {
6166 /* mark array as shutdown cleanly */
6167 if (!mddev_is_clustered(mddev))
6168 mddev->in_sync = 1;
6169 md_update_sb(mddev, 1);
6170 }
6171 /* disable policy to guarantee rdevs free resources for serialization */
6172 mddev->serialize_policy = 0;
6173 mddev_destroy_serial_pool(mddev, NULL, true);
6174}
6175
6176void md_stop_writes(struct mddev *mddev)
6177{
6178 mddev_lock_nointr(mddev);
6179 __md_stop_writes(mddev);
6180 mddev_unlock(mddev);
6181}
6182EXPORT_SYMBOL_GPL(md_stop_writes);
6183
6184static void mddev_detach(struct mddev *mddev)
6185{
6186 md_bitmap_wait_behind_writes(mddev);
6187 if (mddev->pers && mddev->pers->quiesce) {
6188 mddev->pers->quiesce(mddev, 1);
6189 mddev->pers->quiesce(mddev, 0);
6190 }
6191 md_unregister_thread(&mddev->thread);
6192 if (mddev->queue)
6193 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6194}
6195
6196static void __md_stop(struct mddev *mddev)
6197{
6198 struct md_personality *pers = mddev->pers;
6199 md_bitmap_destroy(mddev);
6200 mddev_detach(mddev);
6201 /* Ensure ->event_work is done */
6202 flush_workqueue(md_misc_wq);
6203 spin_lock(&mddev->lock);
6204 mddev->pers = NULL;
6205 spin_unlock(&mddev->lock);
6206 pers->free(mddev, mddev->private);
6207 mddev->private = NULL;
6208 if (pers->sync_request && mddev->to_remove == NULL)
6209 mddev->to_remove = &md_redundancy_group;
6210 module_put(pers->owner);
6211 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6212}
6213
6214void md_stop(struct mddev *mddev)
6215{
6216 /* stop the array and free an attached data structures.
6217 * This is called from dm-raid
6218 */
6219 __md_stop(mddev);
6220 bioset_exit(&mddev->bio_set);
6221 bioset_exit(&mddev->sync_set);
6222}
6223
6224EXPORT_SYMBOL_GPL(md_stop);
6225
6226static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6227{
6228 int err = 0;
6229 int did_freeze = 0;
6230
6231 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6232 did_freeze = 1;
6233 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6234 md_wakeup_thread(mddev->thread);
6235 }
6236 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6237 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6238 if (mddev->sync_thread)
6239 /* Thread might be blocked waiting for metadata update
6240 * which will now never happen */
6241 wake_up_process(mddev->sync_thread->tsk);
6242
6243 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6244 return -EBUSY;
6245 mddev_unlock(mddev);
6246 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6247 &mddev->recovery));
6248 wait_event(mddev->sb_wait,
6249 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6250 mddev_lock_nointr(mddev);
6251
6252 mutex_lock(&mddev->open_mutex);
6253 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6254 mddev->sync_thread ||
6255 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6256 pr_warn("md: %s still in use.\n",mdname(mddev));
6257 if (did_freeze) {
6258 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6259 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6260 md_wakeup_thread(mddev->thread);
6261 }
6262 err = -EBUSY;
6263 goto out;
6264 }
6265 if (mddev->pers) {
6266 __md_stop_writes(mddev);
6267
6268 err = -ENXIO;
6269 if (mddev->ro==1)
6270 goto out;
6271 mddev->ro = 1;
6272 set_disk_ro(mddev->gendisk, 1);
6273 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6274 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6275 md_wakeup_thread(mddev->thread);
6276 sysfs_notify_dirent_safe(mddev->sysfs_state);
6277 err = 0;
6278 }
6279out:
6280 mutex_unlock(&mddev->open_mutex);
6281 return err;
6282}
6283
6284/* mode:
6285 * 0 - completely stop and dis-assemble array
6286 * 2 - stop but do not disassemble array
6287 */
6288static int do_md_stop(struct mddev *mddev, int mode,
6289 struct block_device *bdev)
6290{
6291 struct gendisk *disk = mddev->gendisk;
6292 struct md_rdev *rdev;
6293 int did_freeze = 0;
6294
6295 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6296 did_freeze = 1;
6297 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6298 md_wakeup_thread(mddev->thread);
6299 }
6300 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6301 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6302 if (mddev->sync_thread)
6303 /* Thread might be blocked waiting for metadata update
6304 * which will now never happen */
6305 wake_up_process(mddev->sync_thread->tsk);
6306
6307 mddev_unlock(mddev);
6308 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6309 !test_bit(MD_RECOVERY_RUNNING,
6310 &mddev->recovery)));
6311 mddev_lock_nointr(mddev);
6312
6313 mutex_lock(&mddev->open_mutex);
6314 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6315 mddev->sysfs_active ||
6316 mddev->sync_thread ||
6317 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6318 pr_warn("md: %s still in use.\n",mdname(mddev));
6319 mutex_unlock(&mddev->open_mutex);
6320 if (did_freeze) {
6321 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6322 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6323 md_wakeup_thread(mddev->thread);
6324 }
6325 return -EBUSY;
6326 }
6327 if (mddev->pers) {
6328 if (mddev->ro)
6329 set_disk_ro(disk, 0);
6330
6331 __md_stop_writes(mddev);
6332 __md_stop(mddev);
6333 mddev->queue->backing_dev_info->congested_fn = NULL;
6334
6335 /* tell userspace to handle 'inactive' */
6336 sysfs_notify_dirent_safe(mddev->sysfs_state);
6337
6338 rdev_for_each(rdev, mddev)
6339 if (rdev->raid_disk >= 0)
6340 sysfs_unlink_rdev(mddev, rdev);
6341
6342 set_capacity(disk, 0);
6343 mutex_unlock(&mddev->open_mutex);
6344 mddev->changed = 1;
6345 revalidate_disk(disk);
6346
6347 if (mddev->ro)
6348 mddev->ro = 0;
6349 } else
6350 mutex_unlock(&mddev->open_mutex);
6351 /*
6352 * Free resources if final stop
6353 */
6354 if (mode == 0) {
6355 pr_info("md: %s stopped.\n", mdname(mddev));
6356
6357 if (mddev->bitmap_info.file) {
6358 struct file *f = mddev->bitmap_info.file;
6359 spin_lock(&mddev->lock);
6360 mddev->bitmap_info.file = NULL;
6361 spin_unlock(&mddev->lock);
6362 fput(f);
6363 }
6364 mddev->bitmap_info.offset = 0;
6365
6366 export_array(mddev);
6367
6368 md_clean(mddev);
6369 if (mddev->hold_active == UNTIL_STOP)
6370 mddev->hold_active = 0;
6371 }
6372 md_new_event(mddev);
6373 sysfs_notify_dirent_safe(mddev->sysfs_state);
6374 return 0;
6375}
6376
6377#ifndef MODULE
6378static void autorun_array(struct mddev *mddev)
6379{
6380 struct md_rdev *rdev;
6381 int err;
6382
6383 if (list_empty(&mddev->disks))
6384 return;
6385
6386 pr_info("md: running: ");
6387
6388 rdev_for_each(rdev, mddev) {
6389 char b[BDEVNAME_SIZE];
6390 pr_cont("<%s>", bdevname(rdev->bdev,b));
6391 }
6392 pr_cont("\n");
6393
6394 err = do_md_run(mddev);
6395 if (err) {
6396 pr_warn("md: do_md_run() returned %d\n", err);
6397 do_md_stop(mddev, 0, NULL);
6398 }
6399}
6400
6401/*
6402 * lets try to run arrays based on all disks that have arrived
6403 * until now. (those are in pending_raid_disks)
6404 *
6405 * the method: pick the first pending disk, collect all disks with
6406 * the same UUID, remove all from the pending list and put them into
6407 * the 'same_array' list. Then order this list based on superblock
6408 * update time (freshest comes first), kick out 'old' disks and
6409 * compare superblocks. If everything's fine then run it.
6410 *
6411 * If "unit" is allocated, then bump its reference count
6412 */
6413static void autorun_devices(int part)
6414{
6415 struct md_rdev *rdev0, *rdev, *tmp;
6416 struct mddev *mddev;
6417 char b[BDEVNAME_SIZE];
6418
6419 pr_info("md: autorun ...\n");
6420 while (!list_empty(&pending_raid_disks)) {
6421 int unit;
6422 dev_t dev;
6423 LIST_HEAD(candidates);
6424 rdev0 = list_entry(pending_raid_disks.next,
6425 struct md_rdev, same_set);
6426
6427 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6428 INIT_LIST_HEAD(&candidates);
6429 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6430 if (super_90_load(rdev, rdev0, 0) >= 0) {
6431 pr_debug("md: adding %s ...\n",
6432 bdevname(rdev->bdev,b));
6433 list_move(&rdev->same_set, &candidates);
6434 }
6435 /*
6436 * now we have a set of devices, with all of them having
6437 * mostly sane superblocks. It's time to allocate the
6438 * mddev.
6439 */
6440 if (part) {
6441 dev = MKDEV(mdp_major,
6442 rdev0->preferred_minor << MdpMinorShift);
6443 unit = MINOR(dev) >> MdpMinorShift;
6444 } else {
6445 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6446 unit = MINOR(dev);
6447 }
6448 if (rdev0->preferred_minor != unit) {
6449 pr_warn("md: unit number in %s is bad: %d\n",
6450 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6451 break;
6452 }
6453
6454 md_probe(dev, NULL, NULL);
6455 mddev = mddev_find(dev);
6456 if (!mddev || !mddev->gendisk) {
6457 if (mddev)
6458 mddev_put(mddev);
6459 break;
6460 }
6461 if (mddev_lock(mddev))
6462 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6463 else if (mddev->raid_disks || mddev->major_version
6464 || !list_empty(&mddev->disks)) {
6465 pr_warn("md: %s already running, cannot run %s\n",
6466 mdname(mddev), bdevname(rdev0->bdev,b));
6467 mddev_unlock(mddev);
6468 } else {
6469 pr_debug("md: created %s\n", mdname(mddev));
6470 mddev->persistent = 1;
6471 rdev_for_each_list(rdev, tmp, &candidates) {
6472 list_del_init(&rdev->same_set);
6473 if (bind_rdev_to_array(rdev, mddev))
6474 export_rdev(rdev);
6475 }
6476 autorun_array(mddev);
6477 mddev_unlock(mddev);
6478 }
6479 /* on success, candidates will be empty, on error
6480 * it won't...
6481 */
6482 rdev_for_each_list(rdev, tmp, &candidates) {
6483 list_del_init(&rdev->same_set);
6484 export_rdev(rdev);
6485 }
6486 mddev_put(mddev);
6487 }
6488 pr_info("md: ... autorun DONE.\n");
6489}
6490#endif /* !MODULE */
6491
6492static int get_version(void __user *arg)
6493{
6494 mdu_version_t ver;
6495
6496 ver.major = MD_MAJOR_VERSION;
6497 ver.minor = MD_MINOR_VERSION;
6498 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6499
6500 if (copy_to_user(arg, &ver, sizeof(ver)))
6501 return -EFAULT;
6502
6503 return 0;
6504}
6505
6506static int get_array_info(struct mddev *mddev, void __user *arg)
6507{
6508 mdu_array_info_t info;
6509 int nr,working,insync,failed,spare;
6510 struct md_rdev *rdev;
6511
6512 nr = working = insync = failed = spare = 0;
6513 rcu_read_lock();
6514 rdev_for_each_rcu(rdev, mddev) {
6515 nr++;
6516 if (test_bit(Faulty, &rdev->flags))
6517 failed++;
6518 else {
6519 working++;
6520 if (test_bit(In_sync, &rdev->flags))
6521 insync++;
6522 else if (test_bit(Journal, &rdev->flags))
6523 /* TODO: add journal count to md_u.h */
6524 ;
6525 else
6526 spare++;
6527 }
6528 }
6529 rcu_read_unlock();
6530
6531 info.major_version = mddev->major_version;
6532 info.minor_version = mddev->minor_version;
6533 info.patch_version = MD_PATCHLEVEL_VERSION;
6534 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6535 info.level = mddev->level;
6536 info.size = mddev->dev_sectors / 2;
6537 if (info.size != mddev->dev_sectors / 2) /* overflow */
6538 info.size = -1;
6539 info.nr_disks = nr;
6540 info.raid_disks = mddev->raid_disks;
6541 info.md_minor = mddev->md_minor;
6542 info.not_persistent= !mddev->persistent;
6543
6544 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6545 info.state = 0;
6546 if (mddev->in_sync)
6547 info.state = (1<<MD_SB_CLEAN);
6548 if (mddev->bitmap && mddev->bitmap_info.offset)
6549 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6550 if (mddev_is_clustered(mddev))
6551 info.state |= (1<<MD_SB_CLUSTERED);
6552 info.active_disks = insync;
6553 info.working_disks = working;
6554 info.failed_disks = failed;
6555 info.spare_disks = spare;
6556
6557 info.layout = mddev->layout;
6558 info.chunk_size = mddev->chunk_sectors << 9;
6559
6560 if (copy_to_user(arg, &info, sizeof(info)))
6561 return -EFAULT;
6562
6563 return 0;
6564}
6565
6566static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6567{
6568 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6569 char *ptr;
6570 int err;
6571
6572 file = kzalloc(sizeof(*file), GFP_NOIO);
6573 if (!file)
6574 return -ENOMEM;
6575
6576 err = 0;
6577 spin_lock(&mddev->lock);
6578 /* bitmap enabled */
6579 if (mddev->bitmap_info.file) {
6580 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6581 sizeof(file->pathname));
6582 if (IS_ERR(ptr))
6583 err = PTR_ERR(ptr);
6584 else
6585 memmove(file->pathname, ptr,
6586 sizeof(file->pathname)-(ptr-file->pathname));
6587 }
6588 spin_unlock(&mddev->lock);
6589
6590 if (err == 0 &&
6591 copy_to_user(arg, file, sizeof(*file)))
6592 err = -EFAULT;
6593
6594 kfree(file);
6595 return err;
6596}
6597
6598static int get_disk_info(struct mddev *mddev, void __user * arg)
6599{
6600 mdu_disk_info_t info;
6601 struct md_rdev *rdev;
6602
6603 if (copy_from_user(&info, arg, sizeof(info)))
6604 return -EFAULT;
6605
6606 rcu_read_lock();
6607 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6608 if (rdev) {
6609 info.major = MAJOR(rdev->bdev->bd_dev);
6610 info.minor = MINOR(rdev->bdev->bd_dev);
6611 info.raid_disk = rdev->raid_disk;
6612 info.state = 0;
6613 if (test_bit(Faulty, &rdev->flags))
6614 info.state |= (1<<MD_DISK_FAULTY);
6615 else if (test_bit(In_sync, &rdev->flags)) {
6616 info.state |= (1<<MD_DISK_ACTIVE);
6617 info.state |= (1<<MD_DISK_SYNC);
6618 }
6619 if (test_bit(Journal, &rdev->flags))
6620 info.state |= (1<<MD_DISK_JOURNAL);
6621 if (test_bit(WriteMostly, &rdev->flags))
6622 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6623 if (test_bit(FailFast, &rdev->flags))
6624 info.state |= (1<<MD_DISK_FAILFAST);
6625 } else {
6626 info.major = info.minor = 0;
6627 info.raid_disk = -1;
6628 info.state = (1<<MD_DISK_REMOVED);
6629 }
6630 rcu_read_unlock();
6631
6632 if (copy_to_user(arg, &info, sizeof(info)))
6633 return -EFAULT;
6634
6635 return 0;
6636}
6637
6638static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6639{
6640 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6641 struct md_rdev *rdev;
6642 dev_t dev = MKDEV(info->major,info->minor);
6643
6644 if (mddev_is_clustered(mddev) &&
6645 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6646 pr_warn("%s: Cannot add to clustered mddev.\n",
6647 mdname(mddev));
6648 return -EINVAL;
6649 }
6650
6651 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6652 return -EOVERFLOW;
6653
6654 if (!mddev->raid_disks) {
6655 int err;
6656 /* expecting a device which has a superblock */
6657 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6658 if (IS_ERR(rdev)) {
6659 pr_warn("md: md_import_device returned %ld\n",
6660 PTR_ERR(rdev));
6661 return PTR_ERR(rdev);
6662 }
6663 if (!list_empty(&mddev->disks)) {
6664 struct md_rdev *rdev0
6665 = list_entry(mddev->disks.next,
6666 struct md_rdev, same_set);
6667 err = super_types[mddev->major_version]
6668 .load_super(rdev, rdev0, mddev->minor_version);
6669 if (err < 0) {
6670 pr_warn("md: %s has different UUID to %s\n",
6671 bdevname(rdev->bdev,b),
6672 bdevname(rdev0->bdev,b2));
6673 export_rdev(rdev);
6674 return -EINVAL;
6675 }
6676 }
6677 err = bind_rdev_to_array(rdev, mddev);
6678 if (err)
6679 export_rdev(rdev);
6680 return err;
6681 }
6682
6683 /*
6684 * add_new_disk can be used once the array is assembled
6685 * to add "hot spares". They must already have a superblock
6686 * written
6687 */
6688 if (mddev->pers) {
6689 int err;
6690 if (!mddev->pers->hot_add_disk) {
6691 pr_warn("%s: personality does not support diskops!\n",
6692 mdname(mddev));
6693 return -EINVAL;
6694 }
6695 if (mddev->persistent)
6696 rdev = md_import_device(dev, mddev->major_version,
6697 mddev->minor_version);
6698 else
6699 rdev = md_import_device(dev, -1, -1);
6700 if (IS_ERR(rdev)) {
6701 pr_warn("md: md_import_device returned %ld\n",
6702 PTR_ERR(rdev));
6703 return PTR_ERR(rdev);
6704 }
6705 /* set saved_raid_disk if appropriate */
6706 if (!mddev->persistent) {
6707 if (info->state & (1<<MD_DISK_SYNC) &&
6708 info->raid_disk < mddev->raid_disks) {
6709 rdev->raid_disk = info->raid_disk;
6710 set_bit(In_sync, &rdev->flags);
6711 clear_bit(Bitmap_sync, &rdev->flags);
6712 } else
6713 rdev->raid_disk = -1;
6714 rdev->saved_raid_disk = rdev->raid_disk;
6715 } else
6716 super_types[mddev->major_version].
6717 validate_super(mddev, rdev);
6718 if ((info->state & (1<<MD_DISK_SYNC)) &&
6719 rdev->raid_disk != info->raid_disk) {
6720 /* This was a hot-add request, but events doesn't
6721 * match, so reject it.
6722 */
6723 export_rdev(rdev);
6724 return -EINVAL;
6725 }
6726
6727 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6728 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6729 set_bit(WriteMostly, &rdev->flags);
6730 else
6731 clear_bit(WriteMostly, &rdev->flags);
6732 if (info->state & (1<<MD_DISK_FAILFAST))
6733 set_bit(FailFast, &rdev->flags);
6734 else
6735 clear_bit(FailFast, &rdev->flags);
6736
6737 if (info->state & (1<<MD_DISK_JOURNAL)) {
6738 struct md_rdev *rdev2;
6739 bool has_journal = false;
6740
6741 /* make sure no existing journal disk */
6742 rdev_for_each(rdev2, mddev) {
6743 if (test_bit(Journal, &rdev2->flags)) {
6744 has_journal = true;
6745 break;
6746 }
6747 }
6748 if (has_journal || mddev->bitmap) {
6749 export_rdev(rdev);
6750 return -EBUSY;
6751 }
6752 set_bit(Journal, &rdev->flags);
6753 }
6754 /*
6755 * check whether the device shows up in other nodes
6756 */
6757 if (mddev_is_clustered(mddev)) {
6758 if (info->state & (1 << MD_DISK_CANDIDATE))
6759 set_bit(Candidate, &rdev->flags);
6760 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6761 /* --add initiated by this node */
6762 err = md_cluster_ops->add_new_disk(mddev, rdev);
6763 if (err) {
6764 export_rdev(rdev);
6765 return err;
6766 }
6767 }
6768 }
6769
6770 rdev->raid_disk = -1;
6771 err = bind_rdev_to_array(rdev, mddev);
6772
6773 if (err)
6774 export_rdev(rdev);
6775
6776 if (mddev_is_clustered(mddev)) {
6777 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6778 if (!err) {
6779 err = md_cluster_ops->new_disk_ack(mddev,
6780 err == 0);
6781 if (err)
6782 md_kick_rdev_from_array(rdev);
6783 }
6784 } else {
6785 if (err)
6786 md_cluster_ops->add_new_disk_cancel(mddev);
6787 else
6788 err = add_bound_rdev(rdev);
6789 }
6790
6791 } else if (!err)
6792 err = add_bound_rdev(rdev);
6793
6794 return err;
6795 }
6796
6797 /* otherwise, add_new_disk is only allowed
6798 * for major_version==0 superblocks
6799 */
6800 if (mddev->major_version != 0) {
6801 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6802 return -EINVAL;
6803 }
6804
6805 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6806 int err;
6807 rdev = md_import_device(dev, -1, 0);
6808 if (IS_ERR(rdev)) {
6809 pr_warn("md: error, md_import_device() returned %ld\n",
6810 PTR_ERR(rdev));
6811 return PTR_ERR(rdev);
6812 }
6813 rdev->desc_nr = info->number;
6814 if (info->raid_disk < mddev->raid_disks)
6815 rdev->raid_disk = info->raid_disk;
6816 else
6817 rdev->raid_disk = -1;
6818
6819 if (rdev->raid_disk < mddev->raid_disks)
6820 if (info->state & (1<<MD_DISK_SYNC))
6821 set_bit(In_sync, &rdev->flags);
6822
6823 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6824 set_bit(WriteMostly, &rdev->flags);
6825 if (info->state & (1<<MD_DISK_FAILFAST))
6826 set_bit(FailFast, &rdev->flags);
6827
6828 if (!mddev->persistent) {
6829 pr_debug("md: nonpersistent superblock ...\n");
6830 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6831 } else
6832 rdev->sb_start = calc_dev_sboffset(rdev);
6833 rdev->sectors = rdev->sb_start;
6834
6835 err = bind_rdev_to_array(rdev, mddev);
6836 if (err) {
6837 export_rdev(rdev);
6838 return err;
6839 }
6840 }
6841
6842 return 0;
6843}
6844
6845static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6846{
6847 char b[BDEVNAME_SIZE];
6848 struct md_rdev *rdev;
6849
6850 if (!mddev->pers)
6851 return -ENODEV;
6852
6853 rdev = find_rdev(mddev, dev);
6854 if (!rdev)
6855 return -ENXIO;
6856
6857 if (rdev->raid_disk < 0)
6858 goto kick_rdev;
6859
6860 clear_bit(Blocked, &rdev->flags);
6861 remove_and_add_spares(mddev, rdev);
6862
6863 if (rdev->raid_disk >= 0)
6864 goto busy;
6865
6866kick_rdev:
6867 if (mddev_is_clustered(mddev))
6868 md_cluster_ops->remove_disk(mddev, rdev);
6869
6870 md_kick_rdev_from_array(rdev);
6871 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6872 if (mddev->thread)
6873 md_wakeup_thread(mddev->thread);
6874 else
6875 md_update_sb(mddev, 1);
6876 md_new_event(mddev);
6877
6878 return 0;
6879busy:
6880 pr_debug("md: cannot remove active disk %s from %s ...\n",
6881 bdevname(rdev->bdev,b), mdname(mddev));
6882 return -EBUSY;
6883}
6884
6885static int hot_add_disk(struct mddev *mddev, dev_t dev)
6886{
6887 char b[BDEVNAME_SIZE];
6888 int err;
6889 struct md_rdev *rdev;
6890
6891 if (!mddev->pers)
6892 return -ENODEV;
6893
6894 if (mddev->major_version != 0) {
6895 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6896 mdname(mddev));
6897 return -EINVAL;
6898 }
6899 if (!mddev->pers->hot_add_disk) {
6900 pr_warn("%s: personality does not support diskops!\n",
6901 mdname(mddev));
6902 return -EINVAL;
6903 }
6904
6905 rdev = md_import_device(dev, -1, 0);
6906 if (IS_ERR(rdev)) {
6907 pr_warn("md: error, md_import_device() returned %ld\n",
6908 PTR_ERR(rdev));
6909 return -EINVAL;
6910 }
6911
6912 if (mddev->persistent)
6913 rdev->sb_start = calc_dev_sboffset(rdev);
6914 else
6915 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6916
6917 rdev->sectors = rdev->sb_start;
6918
6919 if (test_bit(Faulty, &rdev->flags)) {
6920 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6921 bdevname(rdev->bdev,b), mdname(mddev));
6922 err = -EINVAL;
6923 goto abort_export;
6924 }
6925
6926 clear_bit(In_sync, &rdev->flags);
6927 rdev->desc_nr = -1;
6928 rdev->saved_raid_disk = -1;
6929 err = bind_rdev_to_array(rdev, mddev);
6930 if (err)
6931 goto abort_export;
6932
6933 /*
6934 * The rest should better be atomic, we can have disk failures
6935 * noticed in interrupt contexts ...
6936 */
6937
6938 rdev->raid_disk = -1;
6939
6940 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6941 if (!mddev->thread)
6942 md_update_sb(mddev, 1);
6943 /*
6944 * Kick recovery, maybe this spare has to be added to the
6945 * array immediately.
6946 */
6947 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6948 md_wakeup_thread(mddev->thread);
6949 md_new_event(mddev);
6950 return 0;
6951
6952abort_export:
6953 export_rdev(rdev);
6954 return err;
6955}
6956
6957static int set_bitmap_file(struct mddev *mddev, int fd)
6958{
6959 int err = 0;
6960
6961 if (mddev->pers) {
6962 if (!mddev->pers->quiesce || !mddev->thread)
6963 return -EBUSY;
6964 if (mddev->recovery || mddev->sync_thread)
6965 return -EBUSY;
6966 /* we should be able to change the bitmap.. */
6967 }
6968
6969 if (fd >= 0) {
6970 struct inode *inode;
6971 struct file *f;
6972
6973 if (mddev->bitmap || mddev->bitmap_info.file)
6974 return -EEXIST; /* cannot add when bitmap is present */
6975 f = fget(fd);
6976
6977 if (f == NULL) {
6978 pr_warn("%s: error: failed to get bitmap file\n",
6979 mdname(mddev));
6980 return -EBADF;
6981 }
6982
6983 inode = f->f_mapping->host;
6984 if (!S_ISREG(inode->i_mode)) {
6985 pr_warn("%s: error: bitmap file must be a regular file\n",
6986 mdname(mddev));
6987 err = -EBADF;
6988 } else if (!(f->f_mode & FMODE_WRITE)) {
6989 pr_warn("%s: error: bitmap file must open for write\n",
6990 mdname(mddev));
6991 err = -EBADF;
6992 } else if (atomic_read(&inode->i_writecount) != 1) {
6993 pr_warn("%s: error: bitmap file is already in use\n",
6994 mdname(mddev));
6995 err = -EBUSY;
6996 }
6997 if (err) {
6998 fput(f);
6999 return err;
7000 }
7001 mddev->bitmap_info.file = f;
7002 mddev->bitmap_info.offset = 0; /* file overrides offset */
7003 } else if (mddev->bitmap == NULL)
7004 return -ENOENT; /* cannot remove what isn't there */
7005 err = 0;
7006 if (mddev->pers) {
7007 if (fd >= 0) {
7008 struct bitmap *bitmap;
7009
7010 bitmap = md_bitmap_create(mddev, -1);
7011 mddev_suspend(mddev);
7012 if (!IS_ERR(bitmap)) {
7013 mddev->bitmap = bitmap;
7014 err = md_bitmap_load(mddev);
7015 } else
7016 err = PTR_ERR(bitmap);
7017 if (err) {
7018 md_bitmap_destroy(mddev);
7019 fd = -1;
7020 }
7021 mddev_resume(mddev);
7022 } else if (fd < 0) {
7023 mddev_suspend(mddev);
7024 md_bitmap_destroy(mddev);
7025 mddev_resume(mddev);
7026 }
7027 }
7028 if (fd < 0) {
7029 struct file *f = mddev->bitmap_info.file;
7030 if (f) {
7031 spin_lock(&mddev->lock);
7032 mddev->bitmap_info.file = NULL;
7033 spin_unlock(&mddev->lock);
7034 fput(f);
7035 }
7036 }
7037
7038 return err;
7039}
7040
7041/*
7042 * set_array_info is used two different ways
7043 * The original usage is when creating a new array.
7044 * In this usage, raid_disks is > 0 and it together with
7045 * level, size, not_persistent,layout,chunksize determine the
7046 * shape of the array.
7047 * This will always create an array with a type-0.90.0 superblock.
7048 * The newer usage is when assembling an array.
7049 * In this case raid_disks will be 0, and the major_version field is
7050 * use to determine which style super-blocks are to be found on the devices.
7051 * The minor and patch _version numbers are also kept incase the
7052 * super_block handler wishes to interpret them.
7053 */
7054static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
7055{
7056
7057 if (info->raid_disks == 0) {
7058 /* just setting version number for superblock loading */
7059 if (info->major_version < 0 ||
7060 info->major_version >= ARRAY_SIZE(super_types) ||
7061 super_types[info->major_version].name == NULL) {
7062 /* maybe try to auto-load a module? */
7063 pr_warn("md: superblock version %d not known\n",
7064 info->major_version);
7065 return -EINVAL;
7066 }
7067 mddev->major_version = info->major_version;
7068 mddev->minor_version = info->minor_version;
7069 mddev->patch_version = info->patch_version;
7070 mddev->persistent = !info->not_persistent;
7071 /* ensure mddev_put doesn't delete this now that there
7072 * is some minimal configuration.
7073 */
7074 mddev->ctime = ktime_get_real_seconds();
7075 return 0;
7076 }
7077 mddev->major_version = MD_MAJOR_VERSION;
7078 mddev->minor_version = MD_MINOR_VERSION;
7079 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7080 mddev->ctime = ktime_get_real_seconds();
7081
7082 mddev->level = info->level;
7083 mddev->clevel[0] = 0;
7084 mddev->dev_sectors = 2 * (sector_t)info->size;
7085 mddev->raid_disks = info->raid_disks;
7086 /* don't set md_minor, it is determined by which /dev/md* was
7087 * openned
7088 */
7089 if (info->state & (1<<MD_SB_CLEAN))
7090 mddev->recovery_cp = MaxSector;
7091 else
7092 mddev->recovery_cp = 0;
7093 mddev->persistent = ! info->not_persistent;
7094 mddev->external = 0;
7095
7096 mddev->layout = info->layout;
7097 if (mddev->level == 0)
7098 /* Cannot trust RAID0 layout info here */
7099 mddev->layout = -1;
7100 mddev->chunk_sectors = info->chunk_size >> 9;
7101
7102 if (mddev->persistent) {
7103 mddev->max_disks = MD_SB_DISKS;
7104 mddev->flags = 0;
7105 mddev->sb_flags = 0;
7106 }
7107 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7108
7109 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7110 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7111 mddev->bitmap_info.offset = 0;
7112
7113 mddev->reshape_position = MaxSector;
7114
7115 /*
7116 * Generate a 128 bit UUID
7117 */
7118 get_random_bytes(mddev->uuid, 16);
7119
7120 mddev->new_level = mddev->level;
7121 mddev->new_chunk_sectors = mddev->chunk_sectors;
7122 mddev->new_layout = mddev->layout;
7123 mddev->delta_disks = 0;
7124 mddev->reshape_backwards = 0;
7125
7126 return 0;
7127}
7128
7129void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7130{
7131 lockdep_assert_held(&mddev->reconfig_mutex);
7132
7133 if (mddev->external_size)
7134 return;
7135
7136 mddev->array_sectors = array_sectors;
7137}
7138EXPORT_SYMBOL(md_set_array_sectors);
7139
7140static int update_size(struct mddev *mddev, sector_t num_sectors)
7141{
7142 struct md_rdev *rdev;
7143 int rv;
7144 int fit = (num_sectors == 0);
7145 sector_t old_dev_sectors = mddev->dev_sectors;
7146
7147 if (mddev->pers->resize == NULL)
7148 return -EINVAL;
7149 /* The "num_sectors" is the number of sectors of each device that
7150 * is used. This can only make sense for arrays with redundancy.
7151 * linear and raid0 always use whatever space is available. We can only
7152 * consider changing this number if no resync or reconstruction is
7153 * happening, and if the new size is acceptable. It must fit before the
7154 * sb_start or, if that is <data_offset, it must fit before the size
7155 * of each device. If num_sectors is zero, we find the largest size
7156 * that fits.
7157 */
7158 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7159 mddev->sync_thread)
7160 return -EBUSY;
7161 if (mddev->ro)
7162 return -EROFS;
7163
7164 rdev_for_each(rdev, mddev) {
7165 sector_t avail = rdev->sectors;
7166
7167 if (fit && (num_sectors == 0 || num_sectors > avail))
7168 num_sectors = avail;
7169 if (avail < num_sectors)
7170 return -ENOSPC;
7171 }
7172 rv = mddev->pers->resize(mddev, num_sectors);
7173 if (!rv) {
7174 if (mddev_is_clustered(mddev))
7175 md_cluster_ops->update_size(mddev, old_dev_sectors);
7176 else if (mddev->queue) {
7177 set_capacity(mddev->gendisk, mddev->array_sectors);
7178 revalidate_disk(mddev->gendisk);
7179 }
7180 }
7181 return rv;
7182}
7183
7184static int update_raid_disks(struct mddev *mddev, int raid_disks)
7185{
7186 int rv;
7187 struct md_rdev *rdev;
7188 /* change the number of raid disks */
7189 if (mddev->pers->check_reshape == NULL)
7190 return -EINVAL;
7191 if (mddev->ro)
7192 return -EROFS;
7193 if (raid_disks <= 0 ||
7194 (mddev->max_disks && raid_disks >= mddev->max_disks))
7195 return -EINVAL;
7196 if (mddev->sync_thread ||
7197 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7198 mddev->reshape_position != MaxSector)
7199 return -EBUSY;
7200
7201 rdev_for_each(rdev, mddev) {
7202 if (mddev->raid_disks < raid_disks &&
7203 rdev->data_offset < rdev->new_data_offset)
7204 return -EINVAL;
7205 if (mddev->raid_disks > raid_disks &&
7206 rdev->data_offset > rdev->new_data_offset)
7207 return -EINVAL;
7208 }
7209
7210 mddev->delta_disks = raid_disks - mddev->raid_disks;
7211 if (mddev->delta_disks < 0)
7212 mddev->reshape_backwards = 1;
7213 else if (mddev->delta_disks > 0)
7214 mddev->reshape_backwards = 0;
7215
7216 rv = mddev->pers->check_reshape(mddev);
7217 if (rv < 0) {
7218 mddev->delta_disks = 0;
7219 mddev->reshape_backwards = 0;
7220 }
7221 return rv;
7222}
7223
7224/*
7225 * update_array_info is used to change the configuration of an
7226 * on-line array.
7227 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7228 * fields in the info are checked against the array.
7229 * Any differences that cannot be handled will cause an error.
7230 * Normally, only one change can be managed at a time.
7231 */
7232static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7233{
7234 int rv = 0;
7235 int cnt = 0;
7236 int state = 0;
7237
7238 /* calculate expected state,ignoring low bits */
7239 if (mddev->bitmap && mddev->bitmap_info.offset)
7240 state |= (1 << MD_SB_BITMAP_PRESENT);
7241
7242 if (mddev->major_version != info->major_version ||
7243 mddev->minor_version != info->minor_version ||
7244/* mddev->patch_version != info->patch_version || */
7245 mddev->ctime != info->ctime ||
7246 mddev->level != info->level ||
7247/* mddev->layout != info->layout || */
7248 mddev->persistent != !info->not_persistent ||
7249 mddev->chunk_sectors != info->chunk_size >> 9 ||
7250 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7251 ((state^info->state) & 0xfffffe00)
7252 )
7253 return -EINVAL;
7254 /* Check there is only one change */
7255 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7256 cnt++;
7257 if (mddev->raid_disks != info->raid_disks)
7258 cnt++;
7259 if (mddev->layout != info->layout)
7260 cnt++;
7261 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7262 cnt++;
7263 if (cnt == 0)
7264 return 0;
7265 if (cnt > 1)
7266 return -EINVAL;
7267
7268 if (mddev->layout != info->layout) {
7269 /* Change layout
7270 * we don't need to do anything at the md level, the
7271 * personality will take care of it all.
7272 */
7273 if (mddev->pers->check_reshape == NULL)
7274 return -EINVAL;
7275 else {
7276 mddev->new_layout = info->layout;
7277 rv = mddev->pers->check_reshape(mddev);
7278 if (rv)
7279 mddev->new_layout = mddev->layout;
7280 return rv;
7281 }
7282 }
7283 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7284 rv = update_size(mddev, (sector_t)info->size * 2);
7285
7286 if (mddev->raid_disks != info->raid_disks)
7287 rv = update_raid_disks(mddev, info->raid_disks);
7288
7289 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7290 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7291 rv = -EINVAL;
7292 goto err;
7293 }
7294 if (mddev->recovery || mddev->sync_thread) {
7295 rv = -EBUSY;
7296 goto err;
7297 }
7298 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7299 struct bitmap *bitmap;
7300 /* add the bitmap */
7301 if (mddev->bitmap) {
7302 rv = -EEXIST;
7303 goto err;
7304 }
7305 if (mddev->bitmap_info.default_offset == 0) {
7306 rv = -EINVAL;
7307 goto err;
7308 }
7309 mddev->bitmap_info.offset =
7310 mddev->bitmap_info.default_offset;
7311 mddev->bitmap_info.space =
7312 mddev->bitmap_info.default_space;
7313 bitmap = md_bitmap_create(mddev, -1);
7314 mddev_suspend(mddev);
7315 if (!IS_ERR(bitmap)) {
7316 mddev->bitmap = bitmap;
7317 rv = md_bitmap_load(mddev);
7318 } else
7319 rv = PTR_ERR(bitmap);
7320 if (rv)
7321 md_bitmap_destroy(mddev);
7322 mddev_resume(mddev);
7323 } else {
7324 /* remove the bitmap */
7325 if (!mddev->bitmap) {
7326 rv = -ENOENT;
7327 goto err;
7328 }
7329 if (mddev->bitmap->storage.file) {
7330 rv = -EINVAL;
7331 goto err;
7332 }
7333 if (mddev->bitmap_info.nodes) {
7334 /* hold PW on all the bitmap lock */
7335 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7336 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7337 rv = -EPERM;
7338 md_cluster_ops->unlock_all_bitmaps(mddev);
7339 goto err;
7340 }
7341
7342 mddev->bitmap_info.nodes = 0;
7343 md_cluster_ops->leave(mddev);
7344 }
7345 mddev_suspend(mddev);
7346 md_bitmap_destroy(mddev);
7347 mddev_resume(mddev);
7348 mddev->bitmap_info.offset = 0;
7349 }
7350 }
7351 md_update_sb(mddev, 1);
7352 return rv;
7353err:
7354 return rv;
7355}
7356
7357static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7358{
7359 struct md_rdev *rdev;
7360 int err = 0;
7361
7362 if (mddev->pers == NULL)
7363 return -ENODEV;
7364
7365 rcu_read_lock();
7366 rdev = md_find_rdev_rcu(mddev, dev);
7367 if (!rdev)
7368 err = -ENODEV;
7369 else {
7370 md_error(mddev, rdev);
7371 if (!test_bit(Faulty, &rdev->flags))
7372 err = -EBUSY;
7373 }
7374 rcu_read_unlock();
7375 return err;
7376}
7377
7378/*
7379 * We have a problem here : there is no easy way to give a CHS
7380 * virtual geometry. We currently pretend that we have a 2 heads
7381 * 4 sectors (with a BIG number of cylinders...). This drives
7382 * dosfs just mad... ;-)
7383 */
7384static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7385{
7386 struct mddev *mddev = bdev->bd_disk->private_data;
7387
7388 geo->heads = 2;
7389 geo->sectors = 4;
7390 geo->cylinders = mddev->array_sectors / 8;
7391 return 0;
7392}
7393
7394static inline bool md_ioctl_valid(unsigned int cmd)
7395{
7396 switch (cmd) {
7397 case ADD_NEW_DISK:
7398 case BLKROSET:
7399 case GET_ARRAY_INFO:
7400 case GET_BITMAP_FILE:
7401 case GET_DISK_INFO:
7402 case HOT_ADD_DISK:
7403 case HOT_REMOVE_DISK:
7404 case RAID_AUTORUN:
7405 case RAID_VERSION:
7406 case RESTART_ARRAY_RW:
7407 case RUN_ARRAY:
7408 case SET_ARRAY_INFO:
7409 case SET_BITMAP_FILE:
7410 case SET_DISK_FAULTY:
7411 case STOP_ARRAY:
7412 case STOP_ARRAY_RO:
7413 case CLUSTERED_DISK_NACK:
7414 return true;
7415 default:
7416 return false;
7417 }
7418}
7419
7420static int md_ioctl(struct block_device *bdev, fmode_t mode,
7421 unsigned int cmd, unsigned long arg)
7422{
7423 int err = 0;
7424 void __user *argp = (void __user *)arg;
7425 struct mddev *mddev = NULL;
7426 int ro;
7427 bool did_set_md_closing = false;
7428
7429 if (!md_ioctl_valid(cmd))
7430 return -ENOTTY;
7431
7432 switch (cmd) {
7433 case RAID_VERSION:
7434 case GET_ARRAY_INFO:
7435 case GET_DISK_INFO:
7436 break;
7437 default:
7438 if (!capable(CAP_SYS_ADMIN))
7439 return -EACCES;
7440 }
7441
7442 /*
7443 * Commands dealing with the RAID driver but not any
7444 * particular array:
7445 */
7446 switch (cmd) {
7447 case RAID_VERSION:
7448 err = get_version(argp);
7449 goto out;
7450
7451#ifndef MODULE
7452 case RAID_AUTORUN:
7453 err = 0;
7454 autostart_arrays(arg);
7455 goto out;
7456#endif
7457 default:;
7458 }
7459
7460 /*
7461 * Commands creating/starting a new array:
7462 */
7463
7464 mddev = bdev->bd_disk->private_data;
7465
7466 if (!mddev) {
7467 BUG();
7468 goto out;
7469 }
7470
7471 /* Some actions do not requires the mutex */
7472 switch (cmd) {
7473 case GET_ARRAY_INFO:
7474 if (!mddev->raid_disks && !mddev->external)
7475 err = -ENODEV;
7476 else
7477 err = get_array_info(mddev, argp);
7478 goto out;
7479
7480 case GET_DISK_INFO:
7481 if (!mddev->raid_disks && !mddev->external)
7482 err = -ENODEV;
7483 else
7484 err = get_disk_info(mddev, argp);
7485 goto out;
7486
7487 case SET_DISK_FAULTY:
7488 err = set_disk_faulty(mddev, new_decode_dev(arg));
7489 goto out;
7490
7491 case GET_BITMAP_FILE:
7492 err = get_bitmap_file(mddev, argp);
7493 goto out;
7494
7495 }
7496
7497 if (cmd == ADD_NEW_DISK)
7498 /* need to ensure md_delayed_delete() has completed */
7499 flush_workqueue(md_misc_wq);
7500
7501 if (cmd == HOT_REMOVE_DISK)
7502 /* need to ensure recovery thread has run */
7503 wait_event_interruptible_timeout(mddev->sb_wait,
7504 !test_bit(MD_RECOVERY_NEEDED,
7505 &mddev->recovery),
7506 msecs_to_jiffies(5000));
7507 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7508 /* Need to flush page cache, and ensure no-one else opens
7509 * and writes
7510 */
7511 mutex_lock(&mddev->open_mutex);
7512 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7513 mutex_unlock(&mddev->open_mutex);
7514 err = -EBUSY;
7515 goto out;
7516 }
7517 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7518 set_bit(MD_CLOSING, &mddev->flags);
7519 did_set_md_closing = true;
7520 mutex_unlock(&mddev->open_mutex);
7521 sync_blockdev(bdev);
7522 }
7523 err = mddev_lock(mddev);
7524 if (err) {
7525 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7526 err, cmd);
7527 goto out;
7528 }
7529
7530 if (cmd == SET_ARRAY_INFO) {
7531 mdu_array_info_t info;
7532 if (!arg)
7533 memset(&info, 0, sizeof(info));
7534 else if (copy_from_user(&info, argp, sizeof(info))) {
7535 err = -EFAULT;
7536 goto unlock;
7537 }
7538 if (mddev->pers) {
7539 err = update_array_info(mddev, &info);
7540 if (err) {
7541 pr_warn("md: couldn't update array info. %d\n", err);
7542 goto unlock;
7543 }
7544 goto unlock;
7545 }
7546 if (!list_empty(&mddev->disks)) {
7547 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7548 err = -EBUSY;
7549 goto unlock;
7550 }
7551 if (mddev->raid_disks) {
7552 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7553 err = -EBUSY;
7554 goto unlock;
7555 }
7556 err = set_array_info(mddev, &info);
7557 if (err) {
7558 pr_warn("md: couldn't set array info. %d\n", err);
7559 goto unlock;
7560 }
7561 goto unlock;
7562 }
7563
7564 /*
7565 * Commands querying/configuring an existing array:
7566 */
7567 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7568 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7569 if ((!mddev->raid_disks && !mddev->external)
7570 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7571 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7572 && cmd != GET_BITMAP_FILE) {
7573 err = -ENODEV;
7574 goto unlock;
7575 }
7576
7577 /*
7578 * Commands even a read-only array can execute:
7579 */
7580 switch (cmd) {
7581 case RESTART_ARRAY_RW:
7582 err = restart_array(mddev);
7583 goto unlock;
7584
7585 case STOP_ARRAY:
7586 err = do_md_stop(mddev, 0, bdev);
7587 goto unlock;
7588
7589 case STOP_ARRAY_RO:
7590 err = md_set_readonly(mddev, bdev);
7591 goto unlock;
7592
7593 case HOT_REMOVE_DISK:
7594 err = hot_remove_disk(mddev, new_decode_dev(arg));
7595 goto unlock;
7596
7597 case ADD_NEW_DISK:
7598 /* We can support ADD_NEW_DISK on read-only arrays
7599 * only if we are re-adding a preexisting device.
7600 * So require mddev->pers and MD_DISK_SYNC.
7601 */
7602 if (mddev->pers) {
7603 mdu_disk_info_t info;
7604 if (copy_from_user(&info, argp, sizeof(info)))
7605 err = -EFAULT;
7606 else if (!(info.state & (1<<MD_DISK_SYNC)))
7607 /* Need to clear read-only for this */
7608 break;
7609 else
7610 err = add_new_disk(mddev, &info);
7611 goto unlock;
7612 }
7613 break;
7614
7615 case BLKROSET:
7616 if (get_user(ro, (int __user *)(arg))) {
7617 err = -EFAULT;
7618 goto unlock;
7619 }
7620 err = -EINVAL;
7621
7622 /* if the bdev is going readonly the value of mddev->ro
7623 * does not matter, no writes are coming
7624 */
7625 if (ro)
7626 goto unlock;
7627
7628 /* are we are already prepared for writes? */
7629 if (mddev->ro != 1)
7630 goto unlock;
7631
7632 /* transitioning to readauto need only happen for
7633 * arrays that call md_write_start
7634 */
7635 if (mddev->pers) {
7636 err = restart_array(mddev);
7637 if (err == 0) {
7638 mddev->ro = 2;
7639 set_disk_ro(mddev->gendisk, 0);
7640 }
7641 }
7642 goto unlock;
7643 }
7644
7645 /*
7646 * The remaining ioctls are changing the state of the
7647 * superblock, so we do not allow them on read-only arrays.
7648 */
7649 if (mddev->ro && mddev->pers) {
7650 if (mddev->ro == 2) {
7651 mddev->ro = 0;
7652 sysfs_notify_dirent_safe(mddev->sysfs_state);
7653 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7654 /* mddev_unlock will wake thread */
7655 /* If a device failed while we were read-only, we
7656 * need to make sure the metadata is updated now.
7657 */
7658 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7659 mddev_unlock(mddev);
7660 wait_event(mddev->sb_wait,
7661 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7662 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7663 mddev_lock_nointr(mddev);
7664 }
7665 } else {
7666 err = -EROFS;
7667 goto unlock;
7668 }
7669 }
7670
7671 switch (cmd) {
7672 case ADD_NEW_DISK:
7673 {
7674 mdu_disk_info_t info;
7675 if (copy_from_user(&info, argp, sizeof(info)))
7676 err = -EFAULT;
7677 else
7678 err = add_new_disk(mddev, &info);
7679 goto unlock;
7680 }
7681
7682 case CLUSTERED_DISK_NACK:
7683 if (mddev_is_clustered(mddev))
7684 md_cluster_ops->new_disk_ack(mddev, false);
7685 else
7686 err = -EINVAL;
7687 goto unlock;
7688
7689 case HOT_ADD_DISK:
7690 err = hot_add_disk(mddev, new_decode_dev(arg));
7691 goto unlock;
7692
7693 case RUN_ARRAY:
7694 err = do_md_run(mddev);
7695 goto unlock;
7696
7697 case SET_BITMAP_FILE:
7698 err = set_bitmap_file(mddev, (int)arg);
7699 goto unlock;
7700
7701 default:
7702 err = -EINVAL;
7703 goto unlock;
7704 }
7705
7706unlock:
7707 if (mddev->hold_active == UNTIL_IOCTL &&
7708 err != -EINVAL)
7709 mddev->hold_active = 0;
7710 mddev_unlock(mddev);
7711out:
7712 if(did_set_md_closing)
7713 clear_bit(MD_CLOSING, &mddev->flags);
7714 return err;
7715}
7716#ifdef CONFIG_COMPAT
7717static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7718 unsigned int cmd, unsigned long arg)
7719{
7720 switch (cmd) {
7721 case HOT_REMOVE_DISK:
7722 case HOT_ADD_DISK:
7723 case SET_DISK_FAULTY:
7724 case SET_BITMAP_FILE:
7725 /* These take in integer arg, do not convert */
7726 break;
7727 default:
7728 arg = (unsigned long)compat_ptr(arg);
7729 break;
7730 }
7731
7732 return md_ioctl(bdev, mode, cmd, arg);
7733}
7734#endif /* CONFIG_COMPAT */
7735
7736static int md_open(struct block_device *bdev, fmode_t mode)
7737{
7738 /*
7739 * Succeed if we can lock the mddev, which confirms that
7740 * it isn't being stopped right now.
7741 */
7742 struct mddev *mddev = mddev_find(bdev->bd_dev);
7743 int err;
7744
7745 if (!mddev)
7746 return -ENODEV;
7747
7748 if (mddev->gendisk != bdev->bd_disk) {
7749 /* we are racing with mddev_put which is discarding this
7750 * bd_disk.
7751 */
7752 mddev_put(mddev);
7753 /* Wait until bdev->bd_disk is definitely gone */
7754 flush_workqueue(md_misc_wq);
7755 /* Then retry the open from the top */
7756 return -ERESTARTSYS;
7757 }
7758 BUG_ON(mddev != bdev->bd_disk->private_data);
7759
7760 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7761 goto out;
7762
7763 if (test_bit(MD_CLOSING, &mddev->flags)) {
7764 mutex_unlock(&mddev->open_mutex);
7765 err = -ENODEV;
7766 goto out;
7767 }
7768
7769 err = 0;
7770 atomic_inc(&mddev->openers);
7771 mutex_unlock(&mddev->open_mutex);
7772
7773 check_disk_change(bdev);
7774 out:
7775 if (err)
7776 mddev_put(mddev);
7777 return err;
7778}
7779
7780static void md_release(struct gendisk *disk, fmode_t mode)
7781{
7782 struct mddev *mddev = disk->private_data;
7783
7784 BUG_ON(!mddev);
7785 atomic_dec(&mddev->openers);
7786 mddev_put(mddev);
7787}
7788
7789static int md_media_changed(struct gendisk *disk)
7790{
7791 struct mddev *mddev = disk->private_data;
7792
7793 return mddev->changed;
7794}
7795
7796static int md_revalidate(struct gendisk *disk)
7797{
7798 struct mddev *mddev = disk->private_data;
7799
7800 mddev->changed = 0;
7801 return 0;
7802}
7803static const struct block_device_operations md_fops =
7804{
7805 .owner = THIS_MODULE,
7806 .open = md_open,
7807 .release = md_release,
7808 .ioctl = md_ioctl,
7809#ifdef CONFIG_COMPAT
7810 .compat_ioctl = md_compat_ioctl,
7811#endif
7812 .getgeo = md_getgeo,
7813 .media_changed = md_media_changed,
7814 .revalidate_disk= md_revalidate,
7815};
7816
7817static int md_thread(void *arg)
7818{
7819 struct md_thread *thread = arg;
7820
7821 /*
7822 * md_thread is a 'system-thread', it's priority should be very
7823 * high. We avoid resource deadlocks individually in each
7824 * raid personality. (RAID5 does preallocation) We also use RR and
7825 * the very same RT priority as kswapd, thus we will never get
7826 * into a priority inversion deadlock.
7827 *
7828 * we definitely have to have equal or higher priority than
7829 * bdflush, otherwise bdflush will deadlock if there are too
7830 * many dirty RAID5 blocks.
7831 */
7832
7833 allow_signal(SIGKILL);
7834 while (!kthread_should_stop()) {
7835
7836 /* We need to wait INTERRUPTIBLE so that
7837 * we don't add to the load-average.
7838 * That means we need to be sure no signals are
7839 * pending
7840 */
7841 if (signal_pending(current))
7842 flush_signals(current);
7843
7844 wait_event_interruptible_timeout
7845 (thread->wqueue,
7846 test_bit(THREAD_WAKEUP, &thread->flags)
7847 || kthread_should_stop() || kthread_should_park(),
7848 thread->timeout);
7849
7850 clear_bit(THREAD_WAKEUP, &thread->flags);
7851 if (kthread_should_park())
7852 kthread_parkme();
7853 if (!kthread_should_stop())
7854 thread->run(thread);
7855 }
7856
7857 return 0;
7858}
7859
7860void md_wakeup_thread(struct md_thread *thread)
7861{
7862 if (thread) {
7863 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7864 set_bit(THREAD_WAKEUP, &thread->flags);
7865 wake_up(&thread->wqueue);
7866 }
7867}
7868EXPORT_SYMBOL(md_wakeup_thread);
7869
7870struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7871 struct mddev *mddev, const char *name)
7872{
7873 struct md_thread *thread;
7874
7875 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7876 if (!thread)
7877 return NULL;
7878
7879 init_waitqueue_head(&thread->wqueue);
7880
7881 thread->run = run;
7882 thread->mddev = mddev;
7883 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7884 thread->tsk = kthread_run(md_thread, thread,
7885 "%s_%s",
7886 mdname(thread->mddev),
7887 name);
7888 if (IS_ERR(thread->tsk)) {
7889 kfree(thread);
7890 return NULL;
7891 }
7892 return thread;
7893}
7894EXPORT_SYMBOL(md_register_thread);
7895
7896void md_unregister_thread(struct md_thread **threadp)
7897{
7898 struct md_thread *thread = *threadp;
7899 if (!thread)
7900 return;
7901 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7902 /* Locking ensures that mddev_unlock does not wake_up a
7903 * non-existent thread
7904 */
7905 spin_lock(&pers_lock);
7906 *threadp = NULL;
7907 spin_unlock(&pers_lock);
7908
7909 kthread_stop(thread->tsk);
7910 kfree(thread);
7911}
7912EXPORT_SYMBOL(md_unregister_thread);
7913
7914void md_error(struct mddev *mddev, struct md_rdev *rdev)
7915{
7916 if (!rdev || test_bit(Faulty, &rdev->flags))
7917 return;
7918
7919 if (!mddev->pers || !mddev->pers->error_handler)
7920 return;
7921 mddev->pers->error_handler(mddev,rdev);
7922 if (mddev->degraded)
7923 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7924 sysfs_notify_dirent_safe(rdev->sysfs_state);
7925 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7926 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7927 md_wakeup_thread(mddev->thread);
7928 if (mddev->event_work.func)
7929 queue_work(md_misc_wq, &mddev->event_work);
7930 md_new_event(mddev);
7931}
7932EXPORT_SYMBOL(md_error);
7933
7934/* seq_file implementation /proc/mdstat */
7935
7936static void status_unused(struct seq_file *seq)
7937{
7938 int i = 0;
7939 struct md_rdev *rdev;
7940
7941 seq_printf(seq, "unused devices: ");
7942
7943 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7944 char b[BDEVNAME_SIZE];
7945 i++;
7946 seq_printf(seq, "%s ",
7947 bdevname(rdev->bdev,b));
7948 }
7949 if (!i)
7950 seq_printf(seq, "<none>");
7951
7952 seq_printf(seq, "\n");
7953}
7954
7955static int status_resync(struct seq_file *seq, struct mddev *mddev)
7956{
7957 sector_t max_sectors, resync, res;
7958 unsigned long dt, db = 0;
7959 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7960 int scale, recovery_active;
7961 unsigned int per_milli;
7962
7963 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7964 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7965 max_sectors = mddev->resync_max_sectors;
7966 else
7967 max_sectors = mddev->dev_sectors;
7968
7969 resync = mddev->curr_resync;
7970 if (resync <= 3) {
7971 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7972 /* Still cleaning up */
7973 resync = max_sectors;
7974 } else if (resync > max_sectors)
7975 resync = max_sectors;
7976 else
7977 resync -= atomic_read(&mddev->recovery_active);
7978
7979 if (resync == 0) {
7980 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
7981 struct md_rdev *rdev;
7982
7983 rdev_for_each(rdev, mddev)
7984 if (rdev->raid_disk >= 0 &&
7985 !test_bit(Faulty, &rdev->flags) &&
7986 rdev->recovery_offset != MaxSector &&
7987 rdev->recovery_offset) {
7988 seq_printf(seq, "\trecover=REMOTE");
7989 return 1;
7990 }
7991 if (mddev->reshape_position != MaxSector)
7992 seq_printf(seq, "\treshape=REMOTE");
7993 else
7994 seq_printf(seq, "\tresync=REMOTE");
7995 return 1;
7996 }
7997 if (mddev->recovery_cp < MaxSector) {
7998 seq_printf(seq, "\tresync=PENDING");
7999 return 1;
8000 }
8001 return 0;
8002 }
8003 if (resync < 3) {
8004 seq_printf(seq, "\tresync=DELAYED");
8005 return 1;
8006 }
8007
8008 WARN_ON(max_sectors == 0);
8009 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8010 * in a sector_t, and (max_sectors>>scale) will fit in a
8011 * u32, as those are the requirements for sector_div.
8012 * Thus 'scale' must be at least 10
8013 */
8014 scale = 10;
8015 if (sizeof(sector_t) > sizeof(unsigned long)) {
8016 while ( max_sectors/2 > (1ULL<<(scale+32)))
8017 scale++;
8018 }
8019 res = (resync>>scale)*1000;
8020 sector_div(res, (u32)((max_sectors>>scale)+1));
8021
8022 per_milli = res;
8023 {
8024 int i, x = per_milli/50, y = 20-x;
8025 seq_printf(seq, "[");
8026 for (i = 0; i < x; i++)
8027 seq_printf(seq, "=");
8028 seq_printf(seq, ">");
8029 for (i = 0; i < y; i++)
8030 seq_printf(seq, ".");
8031 seq_printf(seq, "] ");
8032 }
8033 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8034 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8035 "reshape" :
8036 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8037 "check" :
8038 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8039 "resync" : "recovery"))),
8040 per_milli/10, per_milli % 10,
8041 (unsigned long long) resync/2,
8042 (unsigned long long) max_sectors/2);
8043
8044 /*
8045 * dt: time from mark until now
8046 * db: blocks written from mark until now
8047 * rt: remaining time
8048 *
8049 * rt is a sector_t, which is always 64bit now. We are keeping
8050 * the original algorithm, but it is not really necessary.
8051 *
8052 * Original algorithm:
8053 * So we divide before multiply in case it is 32bit and close
8054 * to the limit.
8055 * We scale the divisor (db) by 32 to avoid losing precision
8056 * near the end of resync when the number of remaining sectors
8057 * is close to 'db'.
8058 * We then divide rt by 32 after multiplying by db to compensate.
8059 * The '+1' avoids division by zero if db is very small.
8060 */
8061 dt = ((jiffies - mddev->resync_mark) / HZ);
8062 if (!dt) dt++;
8063
8064 curr_mark_cnt = mddev->curr_mark_cnt;
8065 recovery_active = atomic_read(&mddev->recovery_active);
8066 resync_mark_cnt = mddev->resync_mark_cnt;
8067
8068 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8069 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8070
8071 rt = max_sectors - resync; /* number of remaining sectors */
8072 rt = div64_u64(rt, db/32+1);
8073 rt *= dt;
8074 rt >>= 5;
8075
8076 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8077 ((unsigned long)rt % 60)/6);
8078
8079 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8080 return 1;
8081}
8082
8083static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8084{
8085 struct list_head *tmp;
8086 loff_t l = *pos;
8087 struct mddev *mddev;
8088
8089 if (l >= 0x10000)
8090 return NULL;
8091 if (!l--)
8092 /* header */
8093 return (void*)1;
8094
8095 spin_lock(&all_mddevs_lock);
8096 list_for_each(tmp,&all_mddevs)
8097 if (!l--) {
8098 mddev = list_entry(tmp, struct mddev, all_mddevs);
8099 mddev_get(mddev);
8100 spin_unlock(&all_mddevs_lock);
8101 return mddev;
8102 }
8103 spin_unlock(&all_mddevs_lock);
8104 if (!l--)
8105 return (void*)2;/* tail */
8106 return NULL;
8107}
8108
8109static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8110{
8111 struct list_head *tmp;
8112 struct mddev *next_mddev, *mddev = v;
8113
8114 ++*pos;
8115 if (v == (void*)2)
8116 return NULL;
8117
8118 spin_lock(&all_mddevs_lock);
8119 if (v == (void*)1)
8120 tmp = all_mddevs.next;
8121 else
8122 tmp = mddev->all_mddevs.next;
8123 if (tmp != &all_mddevs)
8124 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
8125 else {
8126 next_mddev = (void*)2;
8127 *pos = 0x10000;
8128 }
8129 spin_unlock(&all_mddevs_lock);
8130
8131 if (v != (void*)1)
8132 mddev_put(mddev);
8133 return next_mddev;
8134
8135}
8136
8137static void md_seq_stop(struct seq_file *seq, void *v)
8138{
8139 struct mddev *mddev = v;
8140
8141 if (mddev && v != (void*)1 && v != (void*)2)
8142 mddev_put(mddev);
8143}
8144
8145static int md_seq_show(struct seq_file *seq, void *v)
8146{
8147 struct mddev *mddev = v;
8148 sector_t sectors;
8149 struct md_rdev *rdev;
8150
8151 if (v == (void*)1) {
8152 struct md_personality *pers;
8153 seq_printf(seq, "Personalities : ");
8154 spin_lock(&pers_lock);
8155 list_for_each_entry(pers, &pers_list, list)
8156 seq_printf(seq, "[%s] ", pers->name);
8157
8158 spin_unlock(&pers_lock);
8159 seq_printf(seq, "\n");
8160 seq->poll_event = atomic_read(&md_event_count);
8161 return 0;
8162 }
8163 if (v == (void*)2) {
8164 status_unused(seq);
8165 return 0;
8166 }
8167
8168 spin_lock(&mddev->lock);
8169 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8170 seq_printf(seq, "%s : %sactive", mdname(mddev),
8171 mddev->pers ? "" : "in");
8172 if (mddev->pers) {
8173 if (mddev->ro==1)
8174 seq_printf(seq, " (read-only)");
8175 if (mddev->ro==2)
8176 seq_printf(seq, " (auto-read-only)");
8177 seq_printf(seq, " %s", mddev->pers->name);
8178 }
8179
8180 sectors = 0;
8181 rcu_read_lock();
8182 rdev_for_each_rcu(rdev, mddev) {
8183 char b[BDEVNAME_SIZE];
8184 seq_printf(seq, " %s[%d]",
8185 bdevname(rdev->bdev,b), rdev->desc_nr);
8186 if (test_bit(WriteMostly, &rdev->flags))
8187 seq_printf(seq, "(W)");
8188 if (test_bit(Journal, &rdev->flags))
8189 seq_printf(seq, "(J)");
8190 if (test_bit(Faulty, &rdev->flags)) {
8191 seq_printf(seq, "(F)");
8192 continue;
8193 }
8194 if (rdev->raid_disk < 0)
8195 seq_printf(seq, "(S)"); /* spare */
8196 if (test_bit(Replacement, &rdev->flags))
8197 seq_printf(seq, "(R)");
8198 sectors += rdev->sectors;
8199 }
8200 rcu_read_unlock();
8201
8202 if (!list_empty(&mddev->disks)) {
8203 if (mddev->pers)
8204 seq_printf(seq, "\n %llu blocks",
8205 (unsigned long long)
8206 mddev->array_sectors / 2);
8207 else
8208 seq_printf(seq, "\n %llu blocks",
8209 (unsigned long long)sectors / 2);
8210 }
8211 if (mddev->persistent) {
8212 if (mddev->major_version != 0 ||
8213 mddev->minor_version != 90) {
8214 seq_printf(seq," super %d.%d",
8215 mddev->major_version,
8216 mddev->minor_version);
8217 }
8218 } else if (mddev->external)
8219 seq_printf(seq, " super external:%s",
8220 mddev->metadata_type);
8221 else
8222 seq_printf(seq, " super non-persistent");
8223
8224 if (mddev->pers) {
8225 mddev->pers->status(seq, mddev);
8226 seq_printf(seq, "\n ");
8227 if (mddev->pers->sync_request) {
8228 if (status_resync(seq, mddev))
8229 seq_printf(seq, "\n ");
8230 }
8231 } else
8232 seq_printf(seq, "\n ");
8233
8234 md_bitmap_status(seq, mddev->bitmap);
8235
8236 seq_printf(seq, "\n");
8237 }
8238 spin_unlock(&mddev->lock);
8239
8240 return 0;
8241}
8242
8243static const struct seq_operations md_seq_ops = {
8244 .start = md_seq_start,
8245 .next = md_seq_next,
8246 .stop = md_seq_stop,
8247 .show = md_seq_show,
8248};
8249
8250static int md_seq_open(struct inode *inode, struct file *file)
8251{
8252 struct seq_file *seq;
8253 int error;
8254
8255 error = seq_open(file, &md_seq_ops);
8256 if (error)
8257 return error;
8258
8259 seq = file->private_data;
8260 seq->poll_event = atomic_read(&md_event_count);
8261 return error;
8262}
8263
8264static int md_unloading;
8265static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8266{
8267 struct seq_file *seq = filp->private_data;
8268 __poll_t mask;
8269
8270 if (md_unloading)
8271 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8272 poll_wait(filp, &md_event_waiters, wait);
8273
8274 /* always allow read */
8275 mask = EPOLLIN | EPOLLRDNORM;
8276
8277 if (seq->poll_event != atomic_read(&md_event_count))
8278 mask |= EPOLLERR | EPOLLPRI;
8279 return mask;
8280}
8281
8282static const struct proc_ops mdstat_proc_ops = {
8283 .proc_open = md_seq_open,
8284 .proc_read = seq_read,
8285 .proc_lseek = seq_lseek,
8286 .proc_release = seq_release,
8287 .proc_poll = mdstat_poll,
8288};
8289
8290int register_md_personality(struct md_personality *p)
8291{
8292 pr_debug("md: %s personality registered for level %d\n",
8293 p->name, p->level);
8294 spin_lock(&pers_lock);
8295 list_add_tail(&p->list, &pers_list);
8296 spin_unlock(&pers_lock);
8297 return 0;
8298}
8299EXPORT_SYMBOL(register_md_personality);
8300
8301int unregister_md_personality(struct md_personality *p)
8302{
8303 pr_debug("md: %s personality unregistered\n", p->name);
8304 spin_lock(&pers_lock);
8305 list_del_init(&p->list);
8306 spin_unlock(&pers_lock);
8307 return 0;
8308}
8309EXPORT_SYMBOL(unregister_md_personality);
8310
8311int register_md_cluster_operations(struct md_cluster_operations *ops,
8312 struct module *module)
8313{
8314 int ret = 0;
8315 spin_lock(&pers_lock);
8316 if (md_cluster_ops != NULL)
8317 ret = -EALREADY;
8318 else {
8319 md_cluster_ops = ops;
8320 md_cluster_mod = module;
8321 }
8322 spin_unlock(&pers_lock);
8323 return ret;
8324}
8325EXPORT_SYMBOL(register_md_cluster_operations);
8326
8327int unregister_md_cluster_operations(void)
8328{
8329 spin_lock(&pers_lock);
8330 md_cluster_ops = NULL;
8331 spin_unlock(&pers_lock);
8332 return 0;
8333}
8334EXPORT_SYMBOL(unregister_md_cluster_operations);
8335
8336int md_setup_cluster(struct mddev *mddev, int nodes)
8337{
8338 if (!md_cluster_ops)
8339 request_module("md-cluster");
8340 spin_lock(&pers_lock);
8341 /* ensure module won't be unloaded */
8342 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8343 pr_warn("can't find md-cluster module or get it's reference.\n");
8344 spin_unlock(&pers_lock);
8345 return -ENOENT;
8346 }
8347 spin_unlock(&pers_lock);
8348
8349 return md_cluster_ops->join(mddev, nodes);
8350}
8351
8352void md_cluster_stop(struct mddev *mddev)
8353{
8354 if (!md_cluster_ops)
8355 return;
8356 md_cluster_ops->leave(mddev);
8357 module_put(md_cluster_mod);
8358}
8359
8360static int is_mddev_idle(struct mddev *mddev, int init)
8361{
8362 struct md_rdev *rdev;
8363 int idle;
8364 int curr_events;
8365
8366 idle = 1;
8367 rcu_read_lock();
8368 rdev_for_each_rcu(rdev, mddev) {
8369 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
8370 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8371 atomic_read(&disk->sync_io);
8372 /* sync IO will cause sync_io to increase before the disk_stats
8373 * as sync_io is counted when a request starts, and
8374 * disk_stats is counted when it completes.
8375 * So resync activity will cause curr_events to be smaller than
8376 * when there was no such activity.
8377 * non-sync IO will cause disk_stat to increase without
8378 * increasing sync_io so curr_events will (eventually)
8379 * be larger than it was before. Once it becomes
8380 * substantially larger, the test below will cause
8381 * the array to appear non-idle, and resync will slow
8382 * down.
8383 * If there is a lot of outstanding resync activity when
8384 * we set last_event to curr_events, then all that activity
8385 * completing might cause the array to appear non-idle
8386 * and resync will be slowed down even though there might
8387 * not have been non-resync activity. This will only
8388 * happen once though. 'last_events' will soon reflect
8389 * the state where there is little or no outstanding
8390 * resync requests, and further resync activity will
8391 * always make curr_events less than last_events.
8392 *
8393 */
8394 if (init || curr_events - rdev->last_events > 64) {
8395 rdev->last_events = curr_events;
8396 idle = 0;
8397 }
8398 }
8399 rcu_read_unlock();
8400 return idle;
8401}
8402
8403void md_done_sync(struct mddev *mddev, int blocks, int ok)
8404{
8405 /* another "blocks" (512byte) blocks have been synced */
8406 atomic_sub(blocks, &mddev->recovery_active);
8407 wake_up(&mddev->recovery_wait);
8408 if (!ok) {
8409 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8410 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8411 md_wakeup_thread(mddev->thread);
8412 // stop recovery, signal do_sync ....
8413 }
8414}
8415EXPORT_SYMBOL(md_done_sync);
8416
8417/* md_write_start(mddev, bi)
8418 * If we need to update some array metadata (e.g. 'active' flag
8419 * in superblock) before writing, schedule a superblock update
8420 * and wait for it to complete.
8421 * A return value of 'false' means that the write wasn't recorded
8422 * and cannot proceed as the array is being suspend.
8423 */
8424bool md_write_start(struct mddev *mddev, struct bio *bi)
8425{
8426 int did_change = 0;
8427
8428 if (bio_data_dir(bi) != WRITE)
8429 return true;
8430
8431 BUG_ON(mddev->ro == 1);
8432 if (mddev->ro == 2) {
8433 /* need to switch to read/write */
8434 mddev->ro = 0;
8435 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8436 md_wakeup_thread(mddev->thread);
8437 md_wakeup_thread(mddev->sync_thread);
8438 did_change = 1;
8439 }
8440 rcu_read_lock();
8441 percpu_ref_get(&mddev->writes_pending);
8442 smp_mb(); /* Match smp_mb in set_in_sync() */
8443 if (mddev->safemode == 1)
8444 mddev->safemode = 0;
8445 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8446 if (mddev->in_sync || mddev->sync_checkers) {
8447 spin_lock(&mddev->lock);
8448 if (mddev->in_sync) {
8449 mddev->in_sync = 0;
8450 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8451 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8452 md_wakeup_thread(mddev->thread);
8453 did_change = 1;
8454 }
8455 spin_unlock(&mddev->lock);
8456 }
8457 rcu_read_unlock();
8458 if (did_change)
8459 sysfs_notify_dirent_safe(mddev->sysfs_state);
8460 if (!mddev->has_superblocks)
8461 return true;
8462 wait_event(mddev->sb_wait,
8463 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8464 mddev->suspended);
8465 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8466 percpu_ref_put(&mddev->writes_pending);
8467 return false;
8468 }
8469 return true;
8470}
8471EXPORT_SYMBOL(md_write_start);
8472
8473/* md_write_inc can only be called when md_write_start() has
8474 * already been called at least once of the current request.
8475 * It increments the counter and is useful when a single request
8476 * is split into several parts. Each part causes an increment and
8477 * so needs a matching md_write_end().
8478 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8479 * a spinlocked region.
8480 */
8481void md_write_inc(struct mddev *mddev, struct bio *bi)
8482{
8483 if (bio_data_dir(bi) != WRITE)
8484 return;
8485 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8486 percpu_ref_get(&mddev->writes_pending);
8487}
8488EXPORT_SYMBOL(md_write_inc);
8489
8490void md_write_end(struct mddev *mddev)
8491{
8492 percpu_ref_put(&mddev->writes_pending);
8493
8494 if (mddev->safemode == 2)
8495 md_wakeup_thread(mddev->thread);
8496 else if (mddev->safemode_delay)
8497 /* The roundup() ensures this only performs locking once
8498 * every ->safemode_delay jiffies
8499 */
8500 mod_timer(&mddev->safemode_timer,
8501 roundup(jiffies, mddev->safemode_delay) +
8502 mddev->safemode_delay);
8503}
8504
8505EXPORT_SYMBOL(md_write_end);
8506
8507/* md_allow_write(mddev)
8508 * Calling this ensures that the array is marked 'active' so that writes
8509 * may proceed without blocking. It is important to call this before
8510 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8511 * Must be called with mddev_lock held.
8512 */
8513void md_allow_write(struct mddev *mddev)
8514{
8515 if (!mddev->pers)
8516 return;
8517 if (mddev->ro)
8518 return;
8519 if (!mddev->pers->sync_request)
8520 return;
8521
8522 spin_lock(&mddev->lock);
8523 if (mddev->in_sync) {
8524 mddev->in_sync = 0;
8525 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8526 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8527 if (mddev->safemode_delay &&
8528 mddev->safemode == 0)
8529 mddev->safemode = 1;
8530 spin_unlock(&mddev->lock);
8531 md_update_sb(mddev, 0);
8532 sysfs_notify_dirent_safe(mddev->sysfs_state);
8533 /* wait for the dirty state to be recorded in the metadata */
8534 wait_event(mddev->sb_wait,
8535 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8536 } else
8537 spin_unlock(&mddev->lock);
8538}
8539EXPORT_SYMBOL_GPL(md_allow_write);
8540
8541#define SYNC_MARKS 10
8542#define SYNC_MARK_STEP (3*HZ)
8543#define UPDATE_FREQUENCY (5*60*HZ)
8544void md_do_sync(struct md_thread *thread)
8545{
8546 struct mddev *mddev = thread->mddev;
8547 struct mddev *mddev2;
8548 unsigned int currspeed = 0, window;
8549 sector_t max_sectors,j, io_sectors, recovery_done;
8550 unsigned long mark[SYNC_MARKS];
8551 unsigned long update_time;
8552 sector_t mark_cnt[SYNC_MARKS];
8553 int last_mark,m;
8554 struct list_head *tmp;
8555 sector_t last_check;
8556 int skipped = 0;
8557 struct md_rdev *rdev;
8558 char *desc, *action = NULL;
8559 struct blk_plug plug;
8560 int ret;
8561
8562 /* just incase thread restarts... */
8563 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8564 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8565 return;
8566 if (mddev->ro) {/* never try to sync a read-only array */
8567 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8568 return;
8569 }
8570
8571 if (mddev_is_clustered(mddev)) {
8572 ret = md_cluster_ops->resync_start(mddev);
8573 if (ret)
8574 goto skip;
8575
8576 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8577 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8578 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8579 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8580 && ((unsigned long long)mddev->curr_resync_completed
8581 < (unsigned long long)mddev->resync_max_sectors))
8582 goto skip;
8583 }
8584
8585 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8586 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8587 desc = "data-check";
8588 action = "check";
8589 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8590 desc = "requested-resync";
8591 action = "repair";
8592 } else
8593 desc = "resync";
8594 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8595 desc = "reshape";
8596 else
8597 desc = "recovery";
8598
8599 mddev->last_sync_action = action ?: desc;
8600
8601 /* we overload curr_resync somewhat here.
8602 * 0 == not engaged in resync at all
8603 * 2 == checking that there is no conflict with another sync
8604 * 1 == like 2, but have yielded to allow conflicting resync to
8605 * commence
8606 * other == active in resync - this many blocks
8607 *
8608 * Before starting a resync we must have set curr_resync to
8609 * 2, and then checked that every "conflicting" array has curr_resync
8610 * less than ours. When we find one that is the same or higher
8611 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8612 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8613 * This will mean we have to start checking from the beginning again.
8614 *
8615 */
8616
8617 do {
8618 int mddev2_minor = -1;
8619 mddev->curr_resync = 2;
8620
8621 try_again:
8622 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8623 goto skip;
8624 for_each_mddev(mddev2, tmp) {
8625 if (mddev2 == mddev)
8626 continue;
8627 if (!mddev->parallel_resync
8628 && mddev2->curr_resync
8629 && match_mddev_units(mddev, mddev2)) {
8630 DEFINE_WAIT(wq);
8631 if (mddev < mddev2 && mddev->curr_resync == 2) {
8632 /* arbitrarily yield */
8633 mddev->curr_resync = 1;
8634 wake_up(&resync_wait);
8635 }
8636 if (mddev > mddev2 && mddev->curr_resync == 1)
8637 /* no need to wait here, we can wait the next
8638 * time 'round when curr_resync == 2
8639 */
8640 continue;
8641 /* We need to wait 'interruptible' so as not to
8642 * contribute to the load average, and not to
8643 * be caught by 'softlockup'
8644 */
8645 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8646 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8647 mddev2->curr_resync >= mddev->curr_resync) {
8648 if (mddev2_minor != mddev2->md_minor) {
8649 mddev2_minor = mddev2->md_minor;
8650 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8651 desc, mdname(mddev),
8652 mdname(mddev2));
8653 }
8654 mddev_put(mddev2);
8655 if (signal_pending(current))
8656 flush_signals(current);
8657 schedule();
8658 finish_wait(&resync_wait, &wq);
8659 goto try_again;
8660 }
8661 finish_wait(&resync_wait, &wq);
8662 }
8663 }
8664 } while (mddev->curr_resync < 2);
8665
8666 j = 0;
8667 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8668 /* resync follows the size requested by the personality,
8669 * which defaults to physical size, but can be virtual size
8670 */
8671 max_sectors = mddev->resync_max_sectors;
8672 atomic64_set(&mddev->resync_mismatches, 0);
8673 /* we don't use the checkpoint if there's a bitmap */
8674 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8675 j = mddev->resync_min;
8676 else if (!mddev->bitmap)
8677 j = mddev->recovery_cp;
8678
8679 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8680 max_sectors = mddev->resync_max_sectors;
8681 /*
8682 * If the original node aborts reshaping then we continue the
8683 * reshaping, so set j again to avoid restart reshape from the
8684 * first beginning
8685 */
8686 if (mddev_is_clustered(mddev) &&
8687 mddev->reshape_position != MaxSector)
8688 j = mddev->reshape_position;
8689 } else {
8690 /* recovery follows the physical size of devices */
8691 max_sectors = mddev->dev_sectors;
8692 j = MaxSector;
8693 rcu_read_lock();
8694 rdev_for_each_rcu(rdev, mddev)
8695 if (rdev->raid_disk >= 0 &&
8696 !test_bit(Journal, &rdev->flags) &&
8697 !test_bit(Faulty, &rdev->flags) &&
8698 !test_bit(In_sync, &rdev->flags) &&
8699 rdev->recovery_offset < j)
8700 j = rdev->recovery_offset;
8701 rcu_read_unlock();
8702
8703 /* If there is a bitmap, we need to make sure all
8704 * writes that started before we added a spare
8705 * complete before we start doing a recovery.
8706 * Otherwise the write might complete and (via
8707 * bitmap_endwrite) set a bit in the bitmap after the
8708 * recovery has checked that bit and skipped that
8709 * region.
8710 */
8711 if (mddev->bitmap) {
8712 mddev->pers->quiesce(mddev, 1);
8713 mddev->pers->quiesce(mddev, 0);
8714 }
8715 }
8716
8717 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8718 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8719 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8720 speed_max(mddev), desc);
8721
8722 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8723
8724 io_sectors = 0;
8725 for (m = 0; m < SYNC_MARKS; m++) {
8726 mark[m] = jiffies;
8727 mark_cnt[m] = io_sectors;
8728 }
8729 last_mark = 0;
8730 mddev->resync_mark = mark[last_mark];
8731 mddev->resync_mark_cnt = mark_cnt[last_mark];
8732
8733 /*
8734 * Tune reconstruction:
8735 */
8736 window = 32 * (PAGE_SIZE / 512);
8737 pr_debug("md: using %dk window, over a total of %lluk.\n",
8738 window/2, (unsigned long long)max_sectors/2);
8739
8740 atomic_set(&mddev->recovery_active, 0);
8741 last_check = 0;
8742
8743 if (j>2) {
8744 pr_debug("md: resuming %s of %s from checkpoint.\n",
8745 desc, mdname(mddev));
8746 mddev->curr_resync = j;
8747 } else
8748 mddev->curr_resync = 3; /* no longer delayed */
8749 mddev->curr_resync_completed = j;
8750 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8751 md_new_event(mddev);
8752 update_time = jiffies;
8753
8754 blk_start_plug(&plug);
8755 while (j < max_sectors) {
8756 sector_t sectors;
8757
8758 skipped = 0;
8759
8760 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8761 ((mddev->curr_resync > mddev->curr_resync_completed &&
8762 (mddev->curr_resync - mddev->curr_resync_completed)
8763 > (max_sectors >> 4)) ||
8764 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8765 (j - mddev->curr_resync_completed)*2
8766 >= mddev->resync_max - mddev->curr_resync_completed ||
8767 mddev->curr_resync_completed > mddev->resync_max
8768 )) {
8769 /* time to update curr_resync_completed */
8770 wait_event(mddev->recovery_wait,
8771 atomic_read(&mddev->recovery_active) == 0);
8772 mddev->curr_resync_completed = j;
8773 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8774 j > mddev->recovery_cp)
8775 mddev->recovery_cp = j;
8776 update_time = jiffies;
8777 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8778 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8779 }
8780
8781 while (j >= mddev->resync_max &&
8782 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8783 /* As this condition is controlled by user-space,
8784 * we can block indefinitely, so use '_interruptible'
8785 * to avoid triggering warnings.
8786 */
8787 flush_signals(current); /* just in case */
8788 wait_event_interruptible(mddev->recovery_wait,
8789 mddev->resync_max > j
8790 || test_bit(MD_RECOVERY_INTR,
8791 &mddev->recovery));
8792 }
8793
8794 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8795 break;
8796
8797 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8798 if (sectors == 0) {
8799 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8800 break;
8801 }
8802
8803 if (!skipped) { /* actual IO requested */
8804 io_sectors += sectors;
8805 atomic_add(sectors, &mddev->recovery_active);
8806 }
8807
8808 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8809 break;
8810
8811 j += sectors;
8812 if (j > max_sectors)
8813 /* when skipping, extra large numbers can be returned. */
8814 j = max_sectors;
8815 if (j > 2)
8816 mddev->curr_resync = j;
8817 mddev->curr_mark_cnt = io_sectors;
8818 if (last_check == 0)
8819 /* this is the earliest that rebuild will be
8820 * visible in /proc/mdstat
8821 */
8822 md_new_event(mddev);
8823
8824 if (last_check + window > io_sectors || j == max_sectors)
8825 continue;
8826
8827 last_check = io_sectors;
8828 repeat:
8829 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8830 /* step marks */
8831 int next = (last_mark+1) % SYNC_MARKS;
8832
8833 mddev->resync_mark = mark[next];
8834 mddev->resync_mark_cnt = mark_cnt[next];
8835 mark[next] = jiffies;
8836 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8837 last_mark = next;
8838 }
8839
8840 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8841 break;
8842
8843 /*
8844 * this loop exits only if either when we are slower than
8845 * the 'hard' speed limit, or the system was IO-idle for
8846 * a jiffy.
8847 * the system might be non-idle CPU-wise, but we only care
8848 * about not overloading the IO subsystem. (things like an
8849 * e2fsck being done on the RAID array should execute fast)
8850 */
8851 cond_resched();
8852
8853 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8854 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8855 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8856
8857 if (currspeed > speed_min(mddev)) {
8858 if (currspeed > speed_max(mddev)) {
8859 msleep(500);
8860 goto repeat;
8861 }
8862 if (!is_mddev_idle(mddev, 0)) {
8863 /*
8864 * Give other IO more of a chance.
8865 * The faster the devices, the less we wait.
8866 */
8867 wait_event(mddev->recovery_wait,
8868 !atomic_read(&mddev->recovery_active));
8869 }
8870 }
8871 }
8872 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8873 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8874 ? "interrupted" : "done");
8875 /*
8876 * this also signals 'finished resyncing' to md_stop
8877 */
8878 blk_finish_plug(&plug);
8879 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8880
8881 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8882 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8883 mddev->curr_resync > 3) {
8884 mddev->curr_resync_completed = mddev->curr_resync;
8885 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8886 }
8887 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8888
8889 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8890 mddev->curr_resync > 3) {
8891 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8892 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8893 if (mddev->curr_resync >= mddev->recovery_cp) {
8894 pr_debug("md: checkpointing %s of %s.\n",
8895 desc, mdname(mddev));
8896 if (test_bit(MD_RECOVERY_ERROR,
8897 &mddev->recovery))
8898 mddev->recovery_cp =
8899 mddev->curr_resync_completed;
8900 else
8901 mddev->recovery_cp =
8902 mddev->curr_resync;
8903 }
8904 } else
8905 mddev->recovery_cp = MaxSector;
8906 } else {
8907 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8908 mddev->curr_resync = MaxSector;
8909 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8910 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8911 rcu_read_lock();
8912 rdev_for_each_rcu(rdev, mddev)
8913 if (rdev->raid_disk >= 0 &&
8914 mddev->delta_disks >= 0 &&
8915 !test_bit(Journal, &rdev->flags) &&
8916 !test_bit(Faulty, &rdev->flags) &&
8917 !test_bit(In_sync, &rdev->flags) &&
8918 rdev->recovery_offset < mddev->curr_resync)
8919 rdev->recovery_offset = mddev->curr_resync;
8920 rcu_read_unlock();
8921 }
8922 }
8923 }
8924 skip:
8925 /* set CHANGE_PENDING here since maybe another update is needed,
8926 * so other nodes are informed. It should be harmless for normal
8927 * raid */
8928 set_mask_bits(&mddev->sb_flags, 0,
8929 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8930
8931 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8932 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8933 mddev->delta_disks > 0 &&
8934 mddev->pers->finish_reshape &&
8935 mddev->pers->size &&
8936 mddev->queue) {
8937 mddev_lock_nointr(mddev);
8938 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8939 mddev_unlock(mddev);
8940 if (!mddev_is_clustered(mddev)) {
8941 set_capacity(mddev->gendisk, mddev->array_sectors);
8942 revalidate_disk(mddev->gendisk);
8943 }
8944 }
8945
8946 spin_lock(&mddev->lock);
8947 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8948 /* We completed so min/max setting can be forgotten if used. */
8949 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8950 mddev->resync_min = 0;
8951 mddev->resync_max = MaxSector;
8952 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8953 mddev->resync_min = mddev->curr_resync_completed;
8954 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8955 mddev->curr_resync = 0;
8956 spin_unlock(&mddev->lock);
8957
8958 wake_up(&resync_wait);
8959 md_wakeup_thread(mddev->thread);
8960 return;
8961}
8962EXPORT_SYMBOL_GPL(md_do_sync);
8963
8964static int remove_and_add_spares(struct mddev *mddev,
8965 struct md_rdev *this)
8966{
8967 struct md_rdev *rdev;
8968 int spares = 0;
8969 int removed = 0;
8970 bool remove_some = false;
8971
8972 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8973 /* Mustn't remove devices when resync thread is running */
8974 return 0;
8975
8976 rdev_for_each(rdev, mddev) {
8977 if ((this == NULL || rdev == this) &&
8978 rdev->raid_disk >= 0 &&
8979 !test_bit(Blocked, &rdev->flags) &&
8980 test_bit(Faulty, &rdev->flags) &&
8981 atomic_read(&rdev->nr_pending)==0) {
8982 /* Faulty non-Blocked devices with nr_pending == 0
8983 * never get nr_pending incremented,
8984 * never get Faulty cleared, and never get Blocked set.
8985 * So we can synchronize_rcu now rather than once per device
8986 */
8987 remove_some = true;
8988 set_bit(RemoveSynchronized, &rdev->flags);
8989 }
8990 }
8991
8992 if (remove_some)
8993 synchronize_rcu();
8994 rdev_for_each(rdev, mddev) {
8995 if ((this == NULL || rdev == this) &&
8996 rdev->raid_disk >= 0 &&
8997 !test_bit(Blocked, &rdev->flags) &&
8998 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8999 (!test_bit(In_sync, &rdev->flags) &&
9000 !test_bit(Journal, &rdev->flags))) &&
9001 atomic_read(&rdev->nr_pending)==0)) {
9002 if (mddev->pers->hot_remove_disk(
9003 mddev, rdev) == 0) {
9004 sysfs_unlink_rdev(mddev, rdev);
9005 rdev->saved_raid_disk = rdev->raid_disk;
9006 rdev->raid_disk = -1;
9007 removed++;
9008 }
9009 }
9010 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
9011 clear_bit(RemoveSynchronized, &rdev->flags);
9012 }
9013
9014 if (removed && mddev->kobj.sd)
9015 sysfs_notify(&mddev->kobj, NULL, "degraded");
9016
9017 if (this && removed)
9018 goto no_add;
9019
9020 rdev_for_each(rdev, mddev) {
9021 if (this && this != rdev)
9022 continue;
9023 if (test_bit(Candidate, &rdev->flags))
9024 continue;
9025 if (rdev->raid_disk >= 0 &&
9026 !test_bit(In_sync, &rdev->flags) &&
9027 !test_bit(Journal, &rdev->flags) &&
9028 !test_bit(Faulty, &rdev->flags))
9029 spares++;
9030 if (rdev->raid_disk >= 0)
9031 continue;
9032 if (test_bit(Faulty, &rdev->flags))
9033 continue;
9034 if (!test_bit(Journal, &rdev->flags)) {
9035 if (mddev->ro &&
9036 ! (rdev->saved_raid_disk >= 0 &&
9037 !test_bit(Bitmap_sync, &rdev->flags)))
9038 continue;
9039
9040 rdev->recovery_offset = 0;
9041 }
9042 if (mddev->pers->
9043 hot_add_disk(mddev, rdev) == 0) {
9044 if (sysfs_link_rdev(mddev, rdev))
9045 /* failure here is OK */;
9046 if (!test_bit(Journal, &rdev->flags))
9047 spares++;
9048 md_new_event(mddev);
9049 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9050 }
9051 }
9052no_add:
9053 if (removed)
9054 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9055 return spares;
9056}
9057
9058static void md_start_sync(struct work_struct *ws)
9059{
9060 struct mddev *mddev = container_of(ws, struct mddev, del_work);
9061
9062 mddev->sync_thread = md_register_thread(md_do_sync,
9063 mddev,
9064 "resync");
9065 if (!mddev->sync_thread) {
9066 pr_warn("%s: could not start resync thread...\n",
9067 mdname(mddev));
9068 /* leave the spares where they are, it shouldn't hurt */
9069 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9070 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9071 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9072 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9073 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9074 wake_up(&resync_wait);
9075 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9076 &mddev->recovery))
9077 if (mddev->sysfs_action)
9078 sysfs_notify_dirent_safe(mddev->sysfs_action);
9079 } else
9080 md_wakeup_thread(mddev->sync_thread);
9081 sysfs_notify_dirent_safe(mddev->sysfs_action);
9082 md_new_event(mddev);
9083}
9084
9085/*
9086 * This routine is regularly called by all per-raid-array threads to
9087 * deal with generic issues like resync and super-block update.
9088 * Raid personalities that don't have a thread (linear/raid0) do not
9089 * need this as they never do any recovery or update the superblock.
9090 *
9091 * It does not do any resync itself, but rather "forks" off other threads
9092 * to do that as needed.
9093 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9094 * "->recovery" and create a thread at ->sync_thread.
9095 * When the thread finishes it sets MD_RECOVERY_DONE
9096 * and wakeups up this thread which will reap the thread and finish up.
9097 * This thread also removes any faulty devices (with nr_pending == 0).
9098 *
9099 * The overall approach is:
9100 * 1/ if the superblock needs updating, update it.
9101 * 2/ If a recovery thread is running, don't do anything else.
9102 * 3/ If recovery has finished, clean up, possibly marking spares active.
9103 * 4/ If there are any faulty devices, remove them.
9104 * 5/ If array is degraded, try to add spares devices
9105 * 6/ If array has spares or is not in-sync, start a resync thread.
9106 */
9107void md_check_recovery(struct mddev *mddev)
9108{
9109 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
9110 /* Write superblock - thread that called mddev_suspend()
9111 * holds reconfig_mutex for us.
9112 */
9113 set_bit(MD_UPDATING_SB, &mddev->flags);
9114 smp_mb__after_atomic();
9115 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
9116 md_update_sb(mddev, 0);
9117 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
9118 wake_up(&mddev->sb_wait);
9119 }
9120
9121 if (mddev->suspended)
9122 return;
9123
9124 if (mddev->bitmap)
9125 md_bitmap_daemon_work(mddev);
9126
9127 if (signal_pending(current)) {
9128 if (mddev->pers->sync_request && !mddev->external) {
9129 pr_debug("md: %s in immediate safe mode\n",
9130 mdname(mddev));
9131 mddev->safemode = 2;
9132 }
9133 flush_signals(current);
9134 }
9135
9136 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
9137 return;
9138 if ( ! (
9139 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9140 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9141 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9142 (mddev->external == 0 && mddev->safemode == 1) ||
9143 (mddev->safemode == 2
9144 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9145 ))
9146 return;
9147
9148 if (mddev_trylock(mddev)) {
9149 int spares = 0;
9150 bool try_set_sync = mddev->safemode != 0;
9151
9152 if (!mddev->external && mddev->safemode == 1)
9153 mddev->safemode = 0;
9154
9155 if (mddev->ro) {
9156 struct md_rdev *rdev;
9157 if (!mddev->external && mddev->in_sync)
9158 /* 'Blocked' flag not needed as failed devices
9159 * will be recorded if array switched to read/write.
9160 * Leaving it set will prevent the device
9161 * from being removed.
9162 */
9163 rdev_for_each(rdev, mddev)
9164 clear_bit(Blocked, &rdev->flags);
9165 /* On a read-only array we can:
9166 * - remove failed devices
9167 * - add already-in_sync devices if the array itself
9168 * is in-sync.
9169 * As we only add devices that are already in-sync,
9170 * we can activate the spares immediately.
9171 */
9172 remove_and_add_spares(mddev, NULL);
9173 /* There is no thread, but we need to call
9174 * ->spare_active and clear saved_raid_disk
9175 */
9176 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9177 md_reap_sync_thread(mddev);
9178 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9179 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9180 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9181 goto unlock;
9182 }
9183
9184 if (mddev_is_clustered(mddev)) {
9185 struct md_rdev *rdev;
9186 /* kick the device if another node issued a
9187 * remove disk.
9188 */
9189 rdev_for_each(rdev, mddev) {
9190 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9191 rdev->raid_disk < 0)
9192 md_kick_rdev_from_array(rdev);
9193 }
9194 }
9195
9196 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9197 spin_lock(&mddev->lock);
9198 set_in_sync(mddev);
9199 spin_unlock(&mddev->lock);
9200 }
9201
9202 if (mddev->sb_flags)
9203 md_update_sb(mddev, 0);
9204
9205 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9206 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9207 /* resync/recovery still happening */
9208 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9209 goto unlock;
9210 }
9211 if (mddev->sync_thread) {
9212 md_reap_sync_thread(mddev);
9213 goto unlock;
9214 }
9215 /* Set RUNNING before clearing NEEDED to avoid
9216 * any transients in the value of "sync_action".
9217 */
9218 mddev->curr_resync_completed = 0;
9219 spin_lock(&mddev->lock);
9220 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9221 spin_unlock(&mddev->lock);
9222 /* Clear some bits that don't mean anything, but
9223 * might be left set
9224 */
9225 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9226 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9227
9228 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9229 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9230 goto not_running;
9231 /* no recovery is running.
9232 * remove any failed drives, then
9233 * add spares if possible.
9234 * Spares are also removed and re-added, to allow
9235 * the personality to fail the re-add.
9236 */
9237
9238 if (mddev->reshape_position != MaxSector) {
9239 if (mddev->pers->check_reshape == NULL ||
9240 mddev->pers->check_reshape(mddev) != 0)
9241 /* Cannot proceed */
9242 goto not_running;
9243 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9244 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9245 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9246 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9247 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9248 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9249 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9250 } else if (mddev->recovery_cp < MaxSector) {
9251 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9252 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9253 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9254 /* nothing to be done ... */
9255 goto not_running;
9256
9257 if (mddev->pers->sync_request) {
9258 if (spares) {
9259 /* We are adding a device or devices to an array
9260 * which has the bitmap stored on all devices.
9261 * So make sure all bitmap pages get written
9262 */
9263 md_bitmap_write_all(mddev->bitmap);
9264 }
9265 INIT_WORK(&mddev->del_work, md_start_sync);
9266 queue_work(md_misc_wq, &mddev->del_work);
9267 goto unlock;
9268 }
9269 not_running:
9270 if (!mddev->sync_thread) {
9271 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9272 wake_up(&resync_wait);
9273 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9274 &mddev->recovery))
9275 if (mddev->sysfs_action)
9276 sysfs_notify_dirent_safe(mddev->sysfs_action);
9277 }
9278 unlock:
9279 wake_up(&mddev->sb_wait);
9280 mddev_unlock(mddev);
9281 }
9282}
9283EXPORT_SYMBOL(md_check_recovery);
9284
9285void md_reap_sync_thread(struct mddev *mddev)
9286{
9287 struct md_rdev *rdev;
9288 sector_t old_dev_sectors = mddev->dev_sectors;
9289 bool is_reshaped = false;
9290
9291 /* resync has finished, collect result */
9292 md_unregister_thread(&mddev->sync_thread);
9293 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9294 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9295 mddev->degraded != mddev->raid_disks) {
9296 /* success...*/
9297 /* activate any spares */
9298 if (mddev->pers->spare_active(mddev)) {
9299 sysfs_notify(&mddev->kobj, NULL,
9300 "degraded");
9301 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9302 }
9303 }
9304 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9305 mddev->pers->finish_reshape) {
9306 mddev->pers->finish_reshape(mddev);
9307 if (mddev_is_clustered(mddev))
9308 is_reshaped = true;
9309 }
9310
9311 /* If array is no-longer degraded, then any saved_raid_disk
9312 * information must be scrapped.
9313 */
9314 if (!mddev->degraded)
9315 rdev_for_each(rdev, mddev)
9316 rdev->saved_raid_disk = -1;
9317
9318 md_update_sb(mddev, 1);
9319 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9320 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9321 * clustered raid */
9322 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9323 md_cluster_ops->resync_finish(mddev);
9324 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9325 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9326 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9327 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9328 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9329 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9330 /*
9331 * We call md_cluster_ops->update_size here because sync_size could
9332 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9333 * so it is time to update size across cluster.
9334 */
9335 if (mddev_is_clustered(mddev) && is_reshaped
9336 && !test_bit(MD_CLOSING, &mddev->flags))
9337 md_cluster_ops->update_size(mddev, old_dev_sectors);
9338 wake_up(&resync_wait);
9339 /* flag recovery needed just to double check */
9340 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9341 sysfs_notify_dirent_safe(mddev->sysfs_action);
9342 md_new_event(mddev);
9343 if (mddev->event_work.func)
9344 queue_work(md_misc_wq, &mddev->event_work);
9345}
9346EXPORT_SYMBOL(md_reap_sync_thread);
9347
9348void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9349{
9350 sysfs_notify_dirent_safe(rdev->sysfs_state);
9351 wait_event_timeout(rdev->blocked_wait,
9352 !test_bit(Blocked, &rdev->flags) &&
9353 !test_bit(BlockedBadBlocks, &rdev->flags),
9354 msecs_to_jiffies(5000));
9355 rdev_dec_pending(rdev, mddev);
9356}
9357EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9358
9359void md_finish_reshape(struct mddev *mddev)
9360{
9361 /* called be personality module when reshape completes. */
9362 struct md_rdev *rdev;
9363
9364 rdev_for_each(rdev, mddev) {
9365 if (rdev->data_offset > rdev->new_data_offset)
9366 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9367 else
9368 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9369 rdev->data_offset = rdev->new_data_offset;
9370 }
9371}
9372EXPORT_SYMBOL(md_finish_reshape);
9373
9374/* Bad block management */
9375
9376/* Returns 1 on success, 0 on failure */
9377int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9378 int is_new)
9379{
9380 struct mddev *mddev = rdev->mddev;
9381 int rv;
9382 if (is_new)
9383 s += rdev->new_data_offset;
9384 else
9385 s += rdev->data_offset;
9386 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9387 if (rv == 0) {
9388 /* Make sure they get written out promptly */
9389 if (test_bit(ExternalBbl, &rdev->flags))
9390 sysfs_notify(&rdev->kobj, NULL,
9391 "unacknowledged_bad_blocks");
9392 sysfs_notify_dirent_safe(rdev->sysfs_state);
9393 set_mask_bits(&mddev->sb_flags, 0,
9394 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9395 md_wakeup_thread(rdev->mddev->thread);
9396 return 1;
9397 } else
9398 return 0;
9399}
9400EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9401
9402int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9403 int is_new)
9404{
9405 int rv;
9406 if (is_new)
9407 s += rdev->new_data_offset;
9408 else
9409 s += rdev->data_offset;
9410 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9411 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9412 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9413 return rv;
9414}
9415EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9416
9417static int md_notify_reboot(struct notifier_block *this,
9418 unsigned long code, void *x)
9419{
9420 struct list_head *tmp;
9421 struct mddev *mddev;
9422 int need_delay = 0;
9423
9424 for_each_mddev(mddev, tmp) {
9425 if (mddev_trylock(mddev)) {
9426 if (mddev->pers)
9427 __md_stop_writes(mddev);
9428 if (mddev->persistent)
9429 mddev->safemode = 2;
9430 mddev_unlock(mddev);
9431 }
9432 need_delay = 1;
9433 }
9434 /*
9435 * certain more exotic SCSI devices are known to be
9436 * volatile wrt too early system reboots. While the
9437 * right place to handle this issue is the given
9438 * driver, we do want to have a safe RAID driver ...
9439 */
9440 if (need_delay)
9441 mdelay(1000*1);
9442
9443 return NOTIFY_DONE;
9444}
9445
9446static struct notifier_block md_notifier = {
9447 .notifier_call = md_notify_reboot,
9448 .next = NULL,
9449 .priority = INT_MAX, /* before any real devices */
9450};
9451
9452static void md_geninit(void)
9453{
9454 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9455
9456 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9457}
9458
9459static int __init md_init(void)
9460{
9461 int ret = -ENOMEM;
9462
9463 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9464 if (!md_wq)
9465 goto err_wq;
9466
9467 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9468 if (!md_misc_wq)
9469 goto err_misc_wq;
9470
9471 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9472 goto err_md;
9473
9474 if ((ret = register_blkdev(0, "mdp")) < 0)
9475 goto err_mdp;
9476 mdp_major = ret;
9477
9478 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9479 md_probe, NULL, NULL);
9480 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9481 md_probe, NULL, NULL);
9482
9483 register_reboot_notifier(&md_notifier);
9484 raid_table_header = register_sysctl_table(raid_root_table);
9485
9486 md_geninit();
9487 return 0;
9488
9489err_mdp:
9490 unregister_blkdev(MD_MAJOR, "md");
9491err_md:
9492 destroy_workqueue(md_misc_wq);
9493err_misc_wq:
9494 destroy_workqueue(md_wq);
9495err_wq:
9496 return ret;
9497}
9498
9499static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9500{
9501 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9502 struct md_rdev *rdev2;
9503 int role, ret;
9504 char b[BDEVNAME_SIZE];
9505
9506 /*
9507 * If size is changed in another node then we need to
9508 * do resize as well.
9509 */
9510 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9511 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9512 if (ret)
9513 pr_info("md-cluster: resize failed\n");
9514 else
9515 md_bitmap_update_sb(mddev->bitmap);
9516 }
9517
9518 /* Check for change of roles in the active devices */
9519 rdev_for_each(rdev2, mddev) {
9520 if (test_bit(Faulty, &rdev2->flags))
9521 continue;
9522
9523 /* Check if the roles changed */
9524 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9525
9526 if (test_bit(Candidate, &rdev2->flags)) {
9527 if (role == 0xfffe) {
9528 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9529 md_kick_rdev_from_array(rdev2);
9530 continue;
9531 }
9532 else
9533 clear_bit(Candidate, &rdev2->flags);
9534 }
9535
9536 if (role != rdev2->raid_disk) {
9537 /*
9538 * got activated except reshape is happening.
9539 */
9540 if (rdev2->raid_disk == -1 && role != 0xffff &&
9541 !(le32_to_cpu(sb->feature_map) &
9542 MD_FEATURE_RESHAPE_ACTIVE)) {
9543 rdev2->saved_raid_disk = role;
9544 ret = remove_and_add_spares(mddev, rdev2);
9545 pr_info("Activated spare: %s\n",
9546 bdevname(rdev2->bdev,b));
9547 /* wakeup mddev->thread here, so array could
9548 * perform resync with the new activated disk */
9549 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9550 md_wakeup_thread(mddev->thread);
9551 }
9552 /* device faulty
9553 * We just want to do the minimum to mark the disk
9554 * as faulty. The recovery is performed by the
9555 * one who initiated the error.
9556 */
9557 if ((role == 0xfffe) || (role == 0xfffd)) {
9558 md_error(mddev, rdev2);
9559 clear_bit(Blocked, &rdev2->flags);
9560 }
9561 }
9562 }
9563
9564 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9565 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9566
9567 /*
9568 * Since mddev->delta_disks has already updated in update_raid_disks,
9569 * so it is time to check reshape.
9570 */
9571 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9572 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9573 /*
9574 * reshape is happening in the remote node, we need to
9575 * update reshape_position and call start_reshape.
9576 */
9577 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9578 if (mddev->pers->update_reshape_pos)
9579 mddev->pers->update_reshape_pos(mddev);
9580 if (mddev->pers->start_reshape)
9581 mddev->pers->start_reshape(mddev);
9582 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9583 mddev->reshape_position != MaxSector &&
9584 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9585 /* reshape is just done in another node. */
9586 mddev->reshape_position = MaxSector;
9587 if (mddev->pers->update_reshape_pos)
9588 mddev->pers->update_reshape_pos(mddev);
9589 }
9590
9591 /* Finally set the event to be up to date */
9592 mddev->events = le64_to_cpu(sb->events);
9593}
9594
9595static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9596{
9597 int err;
9598 struct page *swapout = rdev->sb_page;
9599 struct mdp_superblock_1 *sb;
9600
9601 /* Store the sb page of the rdev in the swapout temporary
9602 * variable in case we err in the future
9603 */
9604 rdev->sb_page = NULL;
9605 err = alloc_disk_sb(rdev);
9606 if (err == 0) {
9607 ClearPageUptodate(rdev->sb_page);
9608 rdev->sb_loaded = 0;
9609 err = super_types[mddev->major_version].
9610 load_super(rdev, NULL, mddev->minor_version);
9611 }
9612 if (err < 0) {
9613 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9614 __func__, __LINE__, rdev->desc_nr, err);
9615 if (rdev->sb_page)
9616 put_page(rdev->sb_page);
9617 rdev->sb_page = swapout;
9618 rdev->sb_loaded = 1;
9619 return err;
9620 }
9621
9622 sb = page_address(rdev->sb_page);
9623 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9624 * is not set
9625 */
9626
9627 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9628 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9629
9630 /* The other node finished recovery, call spare_active to set
9631 * device In_sync and mddev->degraded
9632 */
9633 if (rdev->recovery_offset == MaxSector &&
9634 !test_bit(In_sync, &rdev->flags) &&
9635 mddev->pers->spare_active(mddev))
9636 sysfs_notify(&mddev->kobj, NULL, "degraded");
9637
9638 put_page(swapout);
9639 return 0;
9640}
9641
9642void md_reload_sb(struct mddev *mddev, int nr)
9643{
9644 struct md_rdev *rdev;
9645 int err;
9646
9647 /* Find the rdev */
9648 rdev_for_each_rcu(rdev, mddev) {
9649 if (rdev->desc_nr == nr)
9650 break;
9651 }
9652
9653 if (!rdev || rdev->desc_nr != nr) {
9654 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9655 return;
9656 }
9657
9658 err = read_rdev(mddev, rdev);
9659 if (err < 0)
9660 return;
9661
9662 check_sb_changes(mddev, rdev);
9663
9664 /* Read all rdev's to update recovery_offset */
9665 rdev_for_each_rcu(rdev, mddev) {
9666 if (!test_bit(Faulty, &rdev->flags))
9667 read_rdev(mddev, rdev);
9668 }
9669}
9670EXPORT_SYMBOL(md_reload_sb);
9671
9672#ifndef MODULE
9673
9674/*
9675 * Searches all registered partitions for autorun RAID arrays
9676 * at boot time.
9677 */
9678
9679static DEFINE_MUTEX(detected_devices_mutex);
9680static LIST_HEAD(all_detected_devices);
9681struct detected_devices_node {
9682 struct list_head list;
9683 dev_t dev;
9684};
9685
9686void md_autodetect_dev(dev_t dev)
9687{
9688 struct detected_devices_node *node_detected_dev;
9689
9690 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9691 if (node_detected_dev) {
9692 node_detected_dev->dev = dev;
9693 mutex_lock(&detected_devices_mutex);
9694 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9695 mutex_unlock(&detected_devices_mutex);
9696 }
9697}
9698
9699static void autostart_arrays(int part)
9700{
9701 struct md_rdev *rdev;
9702 struct detected_devices_node *node_detected_dev;
9703 dev_t dev;
9704 int i_scanned, i_passed;
9705
9706 i_scanned = 0;
9707 i_passed = 0;
9708
9709 pr_info("md: Autodetecting RAID arrays.\n");
9710
9711 mutex_lock(&detected_devices_mutex);
9712 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9713 i_scanned++;
9714 node_detected_dev = list_entry(all_detected_devices.next,
9715 struct detected_devices_node, list);
9716 list_del(&node_detected_dev->list);
9717 dev = node_detected_dev->dev;
9718 kfree(node_detected_dev);
9719 mutex_unlock(&detected_devices_mutex);
9720 rdev = md_import_device(dev,0, 90);
9721 mutex_lock(&detected_devices_mutex);
9722 if (IS_ERR(rdev))
9723 continue;
9724
9725 if (test_bit(Faulty, &rdev->flags))
9726 continue;
9727
9728 set_bit(AutoDetected, &rdev->flags);
9729 list_add(&rdev->same_set, &pending_raid_disks);
9730 i_passed++;
9731 }
9732 mutex_unlock(&detected_devices_mutex);
9733
9734 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9735
9736 autorun_devices(part);
9737}
9738
9739#endif /* !MODULE */
9740
9741static __exit void md_exit(void)
9742{
9743 struct mddev *mddev;
9744 struct list_head *tmp;
9745 int delay = 1;
9746
9747 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9748 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9749
9750 unregister_blkdev(MD_MAJOR,"md");
9751 unregister_blkdev(mdp_major, "mdp");
9752 unregister_reboot_notifier(&md_notifier);
9753 unregister_sysctl_table(raid_table_header);
9754
9755 /* We cannot unload the modules while some process is
9756 * waiting for us in select() or poll() - wake them up
9757 */
9758 md_unloading = 1;
9759 while (waitqueue_active(&md_event_waiters)) {
9760 /* not safe to leave yet */
9761 wake_up(&md_event_waiters);
9762 msleep(delay);
9763 delay += delay;
9764 }
9765 remove_proc_entry("mdstat", NULL);
9766
9767 for_each_mddev(mddev, tmp) {
9768 export_array(mddev);
9769 mddev->ctime = 0;
9770 mddev->hold_active = 0;
9771 /*
9772 * for_each_mddev() will call mddev_put() at the end of each
9773 * iteration. As the mddev is now fully clear, this will
9774 * schedule the mddev for destruction by a workqueue, and the
9775 * destroy_workqueue() below will wait for that to complete.
9776 */
9777 }
9778 destroy_workqueue(md_misc_wq);
9779 destroy_workqueue(md_wq);
9780}
9781
9782subsys_initcall(md_init);
9783module_exit(md_exit)
9784
9785static int get_ro(char *buffer, const struct kernel_param *kp)
9786{
9787 return sprintf(buffer, "%d", start_readonly);
9788}
9789static int set_ro(const char *val, const struct kernel_param *kp)
9790{
9791 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9792}
9793
9794module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9795module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9796module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9797module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9798
9799MODULE_LICENSE("GPL");
9800MODULE_DESCRIPTION("MD RAID framework");
9801MODULE_ALIAS("md");
9802MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Ubuntu Jammy Linux kernel mirror