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/blk-integrity.h>
45 #include <linux/badblocks.h>
46 #include <linux/sysctl.h>
47 #include <linux/seq_file.h>
48 #include <linux/fs.h>
49 #include <linux/poll.h>
50 #include <linux/ctype.h>
51 #include <linux/string.h>
52 #include <linux/hdreg.h>
53 #include <linux/proc_fs.h>
54 #include <linux/random.h>
55 #include <linux/major.h>
56 #include <linux/module.h>
57 #include <linux/reboot.h>
58 #include <linux/file.h>
59 #include <linux/compat.h>
60 #include <linux/delay.h>
61 #include <linux/raid/md_p.h>
62 #include <linux/raid/md_u.h>
63 #include <linux/raid/detect.h>
64 #include <linux/slab.h>
65 #include <linux/percpu-refcount.h>
66 #include <linux/part_stat.h>
67
68 #include "md.h"
69 #include "md-bitmap.h"
70 #include "md-cluster.h"
71
72 static const char *action_name[NR_SYNC_ACTIONS] = {
73 [ACTION_RESYNC] = "resync",
74 [ACTION_RECOVER] = "recover",
75 [ACTION_CHECK] = "check",
76 [ACTION_REPAIR] = "repair",
77 [ACTION_RESHAPE] = "reshape",
78 [ACTION_FROZEN] = "frozen",
79 [ACTION_IDLE] = "idle",
80 };
81
82 static DEFINE_XARRAY(md_submodule);
83
84 static const struct kobj_type md_ktype;
85
86 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
87 static struct workqueue_struct *md_wq;
88
89 /*
90 * This workqueue is used for sync_work to register new sync_thread, and for
91 * del_work to remove rdev, and for event_work that is only set by dm-raid.
92 *
93 * Noted that sync_work will grab reconfig_mutex, hence never flush this
94 * workqueue whith reconfig_mutex grabbed.
95 */
96 static struct workqueue_struct *md_misc_wq;
97 struct workqueue_struct *md_bitmap_wq;
98
99 static int remove_and_add_spares(struct mddev *mddev,
100 struct md_rdev *this);
101 static void mddev_detach(struct mddev *mddev);
102 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev);
103 static void md_wakeup_thread_directly(struct md_thread __rcu *thread);
104
105 /*
106 * Default number of read corrections we'll attempt on an rdev
107 * before ejecting it from the array. We divide the read error
108 * count by 2 for every hour elapsed between read errors.
109 */
110 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
111 /* Default safemode delay: 200 msec */
112 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
113 /*
114 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
115 * is 1000 KB/sec, so the extra system load does not show up that much.
116 * Increase it if you want to have more _guaranteed_ speed. Note that
117 * the RAID driver will use the maximum available bandwidth if the IO
118 * subsystem is idle. There is also an 'absolute maximum' reconstruction
119 * speed limit - in case reconstruction slows down your system despite
120 * idle IO detection.
121 *
122 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
123 * or /sys/block/mdX/md/sync_speed_{min,max}
124 */
125
126 static int sysctl_speed_limit_min = 1000;
127 static int sysctl_speed_limit_max = 200000;
speed_min(struct mddev * mddev)128 static inline int speed_min(struct mddev *mddev)
129 {
130 return mddev->sync_speed_min ?
131 mddev->sync_speed_min : sysctl_speed_limit_min;
132 }
133
speed_max(struct mddev * mddev)134 static inline int speed_max(struct mddev *mddev)
135 {
136 return mddev->sync_speed_max ?
137 mddev->sync_speed_max : sysctl_speed_limit_max;
138 }
139
rdev_uninit_serial(struct md_rdev * rdev)140 static void rdev_uninit_serial(struct md_rdev *rdev)
141 {
142 if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
143 return;
144
145 kvfree(rdev->serial);
146 rdev->serial = NULL;
147 }
148
rdevs_uninit_serial(struct mddev * mddev)149 static void rdevs_uninit_serial(struct mddev *mddev)
150 {
151 struct md_rdev *rdev;
152
153 rdev_for_each(rdev, mddev)
154 rdev_uninit_serial(rdev);
155 }
156
rdev_init_serial(struct md_rdev * rdev)157 static int rdev_init_serial(struct md_rdev *rdev)
158 {
159 /* serial_nums equals with BARRIER_BUCKETS_NR */
160 int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
161 struct serial_in_rdev *serial = NULL;
162
163 if (test_bit(CollisionCheck, &rdev->flags))
164 return 0;
165
166 serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
167 GFP_KERNEL);
168 if (!serial)
169 return -ENOMEM;
170
171 for (i = 0; i < serial_nums; i++) {
172 struct serial_in_rdev *serial_tmp = &serial[i];
173
174 spin_lock_init(&serial_tmp->serial_lock);
175 serial_tmp->serial_rb = RB_ROOT_CACHED;
176 init_waitqueue_head(&serial_tmp->serial_io_wait);
177 }
178
179 rdev->serial = serial;
180 set_bit(CollisionCheck, &rdev->flags);
181
182 return 0;
183 }
184
rdevs_init_serial(struct mddev * mddev)185 static int rdevs_init_serial(struct mddev *mddev)
186 {
187 struct md_rdev *rdev;
188 int ret = 0;
189
190 rdev_for_each(rdev, mddev) {
191 ret = rdev_init_serial(rdev);
192 if (ret)
193 break;
194 }
195
196 /* Free all resources if pool is not existed */
197 if (ret && !mddev->serial_info_pool)
198 rdevs_uninit_serial(mddev);
199
200 return ret;
201 }
202
203 /*
204 * rdev needs to enable serial stuffs if it meets the conditions:
205 * 1. it is multi-queue device flaged with writemostly.
206 * 2. the write-behind mode is enabled.
207 */
rdev_need_serial(struct md_rdev * rdev)208 static int rdev_need_serial(struct md_rdev *rdev)
209 {
210 return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
211 rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
212 test_bit(WriteMostly, &rdev->flags));
213 }
214
215 /*
216 * Init resource for rdev(s), then create serial_info_pool if:
217 * 1. rdev is the first device which return true from rdev_enable_serial.
218 * 2. rdev is NULL, means we want to enable serialization for all rdevs.
219 */
mddev_create_serial_pool(struct mddev * mddev,struct md_rdev * rdev)220 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev)
221 {
222 int ret = 0;
223
224 if (rdev && !rdev_need_serial(rdev) &&
225 !test_bit(CollisionCheck, &rdev->flags))
226 return;
227
228 if (!rdev)
229 ret = rdevs_init_serial(mddev);
230 else
231 ret = rdev_init_serial(rdev);
232 if (ret)
233 return;
234
235 if (mddev->serial_info_pool == NULL) {
236 /*
237 * already in memalloc noio context by
238 * mddev_suspend()
239 */
240 mddev->serial_info_pool =
241 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
242 sizeof(struct serial_info));
243 if (!mddev->serial_info_pool) {
244 rdevs_uninit_serial(mddev);
245 pr_err("can't alloc memory pool for serialization\n");
246 }
247 }
248 }
249
250 /*
251 * Free resource from rdev(s), and destroy serial_info_pool under conditions:
252 * 1. rdev is the last device flaged with CollisionCheck.
253 * 2. when bitmap is destroyed while policy is not enabled.
254 * 3. for disable policy, the pool is destroyed only when no rdev needs it.
255 */
mddev_destroy_serial_pool(struct mddev * mddev,struct md_rdev * rdev)256 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev)
257 {
258 if (rdev && !test_bit(CollisionCheck, &rdev->flags))
259 return;
260
261 if (mddev->serial_info_pool) {
262 struct md_rdev *temp;
263 int num = 0; /* used to track if other rdevs need the pool */
264
265 rdev_for_each(temp, mddev) {
266 if (!rdev) {
267 if (!mddev->serialize_policy ||
268 !rdev_need_serial(temp))
269 rdev_uninit_serial(temp);
270 else
271 num++;
272 } else if (temp != rdev &&
273 test_bit(CollisionCheck, &temp->flags))
274 num++;
275 }
276
277 if (rdev)
278 rdev_uninit_serial(rdev);
279
280 if (num)
281 pr_info("The mempool could be used by other devices\n");
282 else {
283 mempool_destroy(mddev->serial_info_pool);
284 mddev->serial_info_pool = NULL;
285 }
286 }
287 }
288
289 static struct ctl_table_header *raid_table_header;
290
291 static const struct ctl_table raid_table[] = {
292 {
293 .procname = "speed_limit_min",
294 .data = &sysctl_speed_limit_min,
295 .maxlen = sizeof(int),
296 .mode = S_IRUGO|S_IWUSR,
297 .proc_handler = proc_dointvec,
298 },
299 {
300 .procname = "speed_limit_max",
301 .data = &sysctl_speed_limit_max,
302 .maxlen = sizeof(int),
303 .mode = S_IRUGO|S_IWUSR,
304 .proc_handler = proc_dointvec,
305 },
306 };
307
308 static int start_readonly;
309
310 /*
311 * The original mechanism for creating an md device is to create
312 * a device node in /dev and to open it. This causes races with device-close.
313 * The preferred method is to write to the "new_array" module parameter.
314 * This can avoid races.
315 * Setting create_on_open to false disables the original mechanism
316 * so all the races disappear.
317 */
318 static bool create_on_open = true;
319
320 /*
321 * We have a system wide 'event count' that is incremented
322 * on any 'interesting' event, and readers of /proc/mdstat
323 * can use 'poll' or 'select' to find out when the event
324 * count increases.
325 *
326 * Events are:
327 * start array, stop array, error, add device, remove device,
328 * start build, activate spare
329 */
330 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
331 static atomic_t md_event_count;
md_new_event(void)332 void md_new_event(void)
333 {
334 atomic_inc(&md_event_count);
335 wake_up(&md_event_waiters);
336 }
337 EXPORT_SYMBOL_GPL(md_new_event);
338
339 /*
340 * Enables to iterate over all existing md arrays
341 * all_mddevs_lock protects this list.
342 */
343 static LIST_HEAD(all_mddevs);
344 static DEFINE_SPINLOCK(all_mddevs_lock);
345
is_md_suspended(struct mddev * mddev)346 static bool is_md_suspended(struct mddev *mddev)
347 {
348 return percpu_ref_is_dying(&mddev->active_io);
349 }
350 /* Rather than calling directly into the personality make_request function,
351 * IO requests come here first so that we can check if the device is
352 * being suspended pending a reconfiguration.
353 * We hold a refcount over the call to ->make_request. By the time that
354 * call has finished, the bio has been linked into some internal structure
355 * and so is visible to ->quiesce(), so we don't need the refcount any more.
356 */
is_suspended(struct mddev * mddev,struct bio * bio)357 static bool is_suspended(struct mddev *mddev, struct bio *bio)
358 {
359 if (is_md_suspended(mddev))
360 return true;
361 if (bio_data_dir(bio) != WRITE)
362 return false;
363 if (READ_ONCE(mddev->suspend_lo) >= READ_ONCE(mddev->suspend_hi))
364 return false;
365 if (bio->bi_iter.bi_sector >= READ_ONCE(mddev->suspend_hi))
366 return false;
367 if (bio_end_sector(bio) < READ_ONCE(mddev->suspend_lo))
368 return false;
369 return true;
370 }
371
md_handle_request(struct mddev * mddev,struct bio * bio)372 bool md_handle_request(struct mddev *mddev, struct bio *bio)
373 {
374 check_suspended:
375 if (is_suspended(mddev, bio)) {
376 DEFINE_WAIT(__wait);
377 /* Bail out if REQ_NOWAIT is set for the bio */
378 if (bio->bi_opf & REQ_NOWAIT) {
379 bio_wouldblock_error(bio);
380 return true;
381 }
382 for (;;) {
383 prepare_to_wait(&mddev->sb_wait, &__wait,
384 TASK_UNINTERRUPTIBLE);
385 if (!is_suspended(mddev, bio))
386 break;
387 schedule();
388 }
389 finish_wait(&mddev->sb_wait, &__wait);
390 }
391 if (!percpu_ref_tryget_live(&mddev->active_io))
392 goto check_suspended;
393
394 if (!mddev->pers->make_request(mddev, bio)) {
395 percpu_ref_put(&mddev->active_io);
396 if (!mddev->gendisk && mddev->pers->prepare_suspend)
397 return false;
398 goto check_suspended;
399 }
400
401 percpu_ref_put(&mddev->active_io);
402 return true;
403 }
404 EXPORT_SYMBOL(md_handle_request);
405
md_submit_bio(struct bio * bio)406 static void md_submit_bio(struct bio *bio)
407 {
408 const int rw = bio_data_dir(bio);
409 struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
410
411 if (mddev == NULL || mddev->pers == NULL) {
412 bio_io_error(bio);
413 return;
414 }
415
416 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
417 bio_io_error(bio);
418 return;
419 }
420
421 bio = bio_split_to_limits(bio);
422 if (!bio)
423 return;
424
425 if (mddev->ro == MD_RDONLY && unlikely(rw == WRITE)) {
426 if (bio_sectors(bio) != 0)
427 bio->bi_status = BLK_STS_IOERR;
428 bio_endio(bio);
429 return;
430 }
431
432 /* bio could be mergeable after passing to underlayer */
433 bio->bi_opf &= ~REQ_NOMERGE;
434
435 md_handle_request(mddev, bio);
436 }
437
438 /*
439 * Make sure no new requests are submitted to the device, and any requests that
440 * have been submitted are completely handled.
441 */
mddev_suspend(struct mddev * mddev,bool interruptible)442 int mddev_suspend(struct mddev *mddev, bool interruptible)
443 {
444 int err = 0;
445
446 /*
447 * hold reconfig_mutex to wait for normal io will deadlock, because
448 * other context can't update super_block, and normal io can rely on
449 * updating super_block.
450 */
451 lockdep_assert_not_held(&mddev->reconfig_mutex);
452
453 if (interruptible)
454 err = mutex_lock_interruptible(&mddev->suspend_mutex);
455 else
456 mutex_lock(&mddev->suspend_mutex);
457 if (err)
458 return err;
459
460 if (mddev->suspended) {
461 WRITE_ONCE(mddev->suspended, mddev->suspended + 1);
462 mutex_unlock(&mddev->suspend_mutex);
463 return 0;
464 }
465
466 percpu_ref_kill(&mddev->active_io);
467 if (interruptible)
468 err = wait_event_interruptible(mddev->sb_wait,
469 percpu_ref_is_zero(&mddev->active_io));
470 else
471 wait_event(mddev->sb_wait,
472 percpu_ref_is_zero(&mddev->active_io));
473 if (err) {
474 percpu_ref_resurrect(&mddev->active_io);
475 mutex_unlock(&mddev->suspend_mutex);
476 return err;
477 }
478
479 /*
480 * For raid456, io might be waiting for reshape to make progress,
481 * allow new reshape to start while waiting for io to be done to
482 * prevent deadlock.
483 */
484 WRITE_ONCE(mddev->suspended, mddev->suspended + 1);
485
486 /* restrict memory reclaim I/O during raid array is suspend */
487 mddev->noio_flag = memalloc_noio_save();
488
489 mutex_unlock(&mddev->suspend_mutex);
490 return 0;
491 }
492 EXPORT_SYMBOL_GPL(mddev_suspend);
493
__mddev_resume(struct mddev * mddev,bool recovery_needed)494 static void __mddev_resume(struct mddev *mddev, bool recovery_needed)
495 {
496 lockdep_assert_not_held(&mddev->reconfig_mutex);
497
498 mutex_lock(&mddev->suspend_mutex);
499 WRITE_ONCE(mddev->suspended, mddev->suspended - 1);
500 if (mddev->suspended) {
501 mutex_unlock(&mddev->suspend_mutex);
502 return;
503 }
504
505 /* entred the memalloc scope from mddev_suspend() */
506 memalloc_noio_restore(mddev->noio_flag);
507
508 percpu_ref_resurrect(&mddev->active_io);
509 wake_up(&mddev->sb_wait);
510
511 if (recovery_needed)
512 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
513 md_wakeup_thread(mddev->thread);
514 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
515
516 mutex_unlock(&mddev->suspend_mutex);
517 }
518
mddev_resume(struct mddev * mddev)519 void mddev_resume(struct mddev *mddev)
520 {
521 return __mddev_resume(mddev, true);
522 }
523 EXPORT_SYMBOL_GPL(mddev_resume);
524
525 /* sync bdev before setting device to readonly or stopping raid*/
mddev_set_closing_and_sync_blockdev(struct mddev * mddev,int opener_num)526 static int mddev_set_closing_and_sync_blockdev(struct mddev *mddev, int opener_num)
527 {
528 mutex_lock(&mddev->open_mutex);
529 if (mddev->pers && atomic_read(&mddev->openers) > opener_num) {
530 mutex_unlock(&mddev->open_mutex);
531 return -EBUSY;
532 }
533 if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
534 mutex_unlock(&mddev->open_mutex);
535 return -EBUSY;
536 }
537 mutex_unlock(&mddev->open_mutex);
538
539 sync_blockdev(mddev->gendisk->part0);
540 return 0;
541 }
542
543 /*
544 * The only difference from bio_chain_endio() is that the current
545 * bi_status of bio does not affect the bi_status of parent.
546 */
md_end_flush(struct bio * bio)547 static void md_end_flush(struct bio *bio)
548 {
549 struct bio *parent = bio->bi_private;
550
551 /*
552 * If any flush io error before the power failure,
553 * disk data may be lost.
554 */
555 if (bio->bi_status)
556 pr_err("md: %pg flush io error %d\n", bio->bi_bdev,
557 blk_status_to_errno(bio->bi_status));
558
559 bio_put(bio);
560 bio_endio(parent);
561 }
562
md_flush_request(struct mddev * mddev,struct bio * bio)563 bool md_flush_request(struct mddev *mddev, struct bio *bio)
564 {
565 struct md_rdev *rdev;
566 struct bio *new;
567
568 /*
569 * md_flush_reqeust() should be called under md_handle_request() and
570 * 'active_io' is already grabbed. Hence it's safe to get rdev directly
571 * without rcu protection.
572 */
573 WARN_ON(percpu_ref_is_zero(&mddev->active_io));
574
575 rdev_for_each(rdev, mddev) {
576 if (rdev->raid_disk < 0 || test_bit(Faulty, &rdev->flags))
577 continue;
578
579 new = bio_alloc_bioset(rdev->bdev, 0,
580 REQ_OP_WRITE | REQ_PREFLUSH, GFP_NOIO,
581 &mddev->bio_set);
582 new->bi_private = bio;
583 new->bi_end_io = md_end_flush;
584 bio_inc_remaining(bio);
585 submit_bio(new);
586 }
587
588 if (bio_sectors(bio) == 0) {
589 bio_endio(bio);
590 return true;
591 }
592
593 bio->bi_opf &= ~REQ_PREFLUSH;
594 return false;
595 }
596 EXPORT_SYMBOL(md_flush_request);
597
mddev_get(struct mddev * mddev)598 static inline struct mddev *mddev_get(struct mddev *mddev)
599 {
600 lockdep_assert_held(&all_mddevs_lock);
601
602 if (test_bit(MD_DELETED, &mddev->flags))
603 return NULL;
604 atomic_inc(&mddev->active);
605 return mddev;
606 }
607
608 static void mddev_delayed_delete(struct work_struct *ws);
609
__mddev_put(struct mddev * mddev)610 static void __mddev_put(struct mddev *mddev)
611 {
612 if (mddev->raid_disks || !list_empty(&mddev->disks) ||
613 mddev->ctime || mddev->hold_active)
614 return;
615
616 /* Array is not configured at all, and not held active, so destroy it */
617 set_bit(MD_DELETED, &mddev->flags);
618
619 /*
620 * Call queue_work inside the spinlock so that flush_workqueue() after
621 * mddev_find will succeed in waiting for the work to be done.
622 */
623 queue_work(md_misc_wq, &mddev->del_work);
624 }
625
mddev_put_locked(struct mddev * mddev)626 static void mddev_put_locked(struct mddev *mddev)
627 {
628 if (atomic_dec_and_test(&mddev->active))
629 __mddev_put(mddev);
630 }
631
mddev_put(struct mddev * mddev)632 void mddev_put(struct mddev *mddev)
633 {
634 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
635 return;
636
637 __mddev_put(mddev);
638 spin_unlock(&all_mddevs_lock);
639 }
640
641 static void md_safemode_timeout(struct timer_list *t);
642 static void md_start_sync(struct work_struct *ws);
643
active_io_release(struct percpu_ref * ref)644 static void active_io_release(struct percpu_ref *ref)
645 {
646 struct mddev *mddev = container_of(ref, struct mddev, active_io);
647
648 wake_up(&mddev->sb_wait);
649 }
650
no_op(struct percpu_ref * r)651 static void no_op(struct percpu_ref *r) {}
652
mddev_init(struct mddev * mddev)653 int mddev_init(struct mddev *mddev)
654 {
655
656 if (percpu_ref_init(&mddev->active_io, active_io_release,
657 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
658 return -ENOMEM;
659
660 if (percpu_ref_init(&mddev->writes_pending, no_op,
661 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
662 percpu_ref_exit(&mddev->active_io);
663 return -ENOMEM;
664 }
665
666 /* We want to start with the refcount at zero */
667 percpu_ref_put(&mddev->writes_pending);
668
669 mutex_init(&mddev->open_mutex);
670 mutex_init(&mddev->reconfig_mutex);
671 mutex_init(&mddev->suspend_mutex);
672 mutex_init(&mddev->bitmap_info.mutex);
673 INIT_LIST_HEAD(&mddev->disks);
674 INIT_LIST_HEAD(&mddev->all_mddevs);
675 INIT_LIST_HEAD(&mddev->deleting);
676 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
677 atomic_set(&mddev->active, 1);
678 atomic_set(&mddev->openers, 0);
679 atomic_set(&mddev->sync_seq, 0);
680 spin_lock_init(&mddev->lock);
681 init_waitqueue_head(&mddev->sb_wait);
682 init_waitqueue_head(&mddev->recovery_wait);
683 mddev->reshape_position = MaxSector;
684 mddev->reshape_backwards = 0;
685 mddev->last_sync_action = ACTION_IDLE;
686 mddev->resync_min = 0;
687 mddev->resync_max = MaxSector;
688 mddev->level = LEVEL_NONE;
689 mddev_set_bitmap_ops(mddev);
690
691 INIT_WORK(&mddev->sync_work, md_start_sync);
692 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
693
694 return 0;
695 }
696 EXPORT_SYMBOL_GPL(mddev_init);
697
mddev_destroy(struct mddev * mddev)698 void mddev_destroy(struct mddev *mddev)
699 {
700 percpu_ref_exit(&mddev->active_io);
701 percpu_ref_exit(&mddev->writes_pending);
702 }
703 EXPORT_SYMBOL_GPL(mddev_destroy);
704
mddev_find_locked(dev_t unit)705 static struct mddev *mddev_find_locked(dev_t unit)
706 {
707 struct mddev *mddev;
708
709 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
710 if (mddev->unit == unit)
711 return mddev;
712
713 return NULL;
714 }
715
716 /* find an unused unit number */
mddev_alloc_unit(void)717 static dev_t mddev_alloc_unit(void)
718 {
719 static int next_minor = 512;
720 int start = next_minor;
721 bool is_free = 0;
722 dev_t dev = 0;
723
724 while (!is_free) {
725 dev = MKDEV(MD_MAJOR, next_minor);
726 next_minor++;
727 if (next_minor > MINORMASK)
728 next_minor = 0;
729 if (next_minor == start)
730 return 0; /* Oh dear, all in use. */
731 is_free = !mddev_find_locked(dev);
732 }
733
734 return dev;
735 }
736
mddev_alloc(dev_t unit)737 static struct mddev *mddev_alloc(dev_t unit)
738 {
739 struct mddev *new;
740 int error;
741
742 if (unit && MAJOR(unit) != MD_MAJOR)
743 unit &= ~((1 << MdpMinorShift) - 1);
744
745 new = kzalloc(sizeof(*new), GFP_KERNEL);
746 if (!new)
747 return ERR_PTR(-ENOMEM);
748
749 error = mddev_init(new);
750 if (error)
751 goto out_free_new;
752
753 spin_lock(&all_mddevs_lock);
754 if (unit) {
755 error = -EEXIST;
756 if (mddev_find_locked(unit))
757 goto out_destroy_new;
758 new->unit = unit;
759 if (MAJOR(unit) == MD_MAJOR)
760 new->md_minor = MINOR(unit);
761 else
762 new->md_minor = MINOR(unit) >> MdpMinorShift;
763 new->hold_active = UNTIL_IOCTL;
764 } else {
765 error = -ENODEV;
766 new->unit = mddev_alloc_unit();
767 if (!new->unit)
768 goto out_destroy_new;
769 new->md_minor = MINOR(new->unit);
770 new->hold_active = UNTIL_STOP;
771 }
772
773 list_add(&new->all_mddevs, &all_mddevs);
774 spin_unlock(&all_mddevs_lock);
775 return new;
776
777 out_destroy_new:
778 spin_unlock(&all_mddevs_lock);
779 mddev_destroy(new);
780 out_free_new:
781 kfree(new);
782 return ERR_PTR(error);
783 }
784
mddev_free(struct mddev * mddev)785 static void mddev_free(struct mddev *mddev)
786 {
787 spin_lock(&all_mddevs_lock);
788 list_del(&mddev->all_mddevs);
789 spin_unlock(&all_mddevs_lock);
790
791 mddev_destroy(mddev);
792 kfree(mddev);
793 }
794
795 static const struct attribute_group md_redundancy_group;
796
mddev_unlock(struct mddev * mddev)797 void mddev_unlock(struct mddev *mddev)
798 {
799 struct md_rdev *rdev;
800 struct md_rdev *tmp;
801 LIST_HEAD(delete);
802
803 if (!list_empty(&mddev->deleting))
804 list_splice_init(&mddev->deleting, &delete);
805
806 if (mddev->to_remove) {
807 /* These cannot be removed under reconfig_mutex as
808 * an access to the files will try to take reconfig_mutex
809 * while holding the file unremovable, which leads to
810 * a deadlock.
811 * So hold set sysfs_active while the remove in happeing,
812 * and anything else which might set ->to_remove or my
813 * otherwise change the sysfs namespace will fail with
814 * -EBUSY if sysfs_active is still set.
815 * We set sysfs_active under reconfig_mutex and elsewhere
816 * test it under the same mutex to ensure its correct value
817 * is seen.
818 */
819 const struct attribute_group *to_remove = mddev->to_remove;
820 mddev->to_remove = NULL;
821 mddev->sysfs_active = 1;
822 mutex_unlock(&mddev->reconfig_mutex);
823
824 if (mddev->kobj.sd) {
825 if (to_remove != &md_redundancy_group)
826 sysfs_remove_group(&mddev->kobj, to_remove);
827 if (mddev->pers == NULL ||
828 mddev->pers->sync_request == NULL) {
829 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
830 if (mddev->sysfs_action)
831 sysfs_put(mddev->sysfs_action);
832 if (mddev->sysfs_completed)
833 sysfs_put(mddev->sysfs_completed);
834 if (mddev->sysfs_degraded)
835 sysfs_put(mddev->sysfs_degraded);
836 mddev->sysfs_action = NULL;
837 mddev->sysfs_completed = NULL;
838 mddev->sysfs_degraded = NULL;
839 }
840 }
841 mddev->sysfs_active = 0;
842 } else
843 mutex_unlock(&mddev->reconfig_mutex);
844
845 md_wakeup_thread(mddev->thread);
846 wake_up(&mddev->sb_wait);
847
848 list_for_each_entry_safe(rdev, tmp, &delete, same_set) {
849 list_del_init(&rdev->same_set);
850 kobject_del(&rdev->kobj);
851 export_rdev(rdev, mddev);
852 }
853 }
854 EXPORT_SYMBOL_GPL(mddev_unlock);
855
md_find_rdev_nr_rcu(struct mddev * mddev,int nr)856 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
857 {
858 struct md_rdev *rdev;
859
860 rdev_for_each_rcu(rdev, mddev)
861 if (rdev->desc_nr == nr)
862 return rdev;
863
864 return NULL;
865 }
866 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
867
find_rdev(struct mddev * mddev,dev_t dev)868 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
869 {
870 struct md_rdev *rdev;
871
872 rdev_for_each(rdev, mddev)
873 if (rdev->bdev->bd_dev == dev)
874 return rdev;
875
876 return NULL;
877 }
878
md_find_rdev_rcu(struct mddev * mddev,dev_t dev)879 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
880 {
881 struct md_rdev *rdev;
882
883 rdev_for_each_rcu(rdev, mddev)
884 if (rdev->bdev->bd_dev == dev)
885 return rdev;
886
887 return NULL;
888 }
889 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
890
get_pers(int level,char * clevel)891 static struct md_personality *get_pers(int level, char *clevel)
892 {
893 struct md_personality *ret = NULL;
894 struct md_submodule_head *head;
895 unsigned long i;
896
897 xa_lock(&md_submodule);
898 xa_for_each(&md_submodule, i, head) {
899 if (head->type != MD_PERSONALITY)
900 continue;
901 if ((level != LEVEL_NONE && head->id == level) ||
902 !strcmp(head->name, clevel)) {
903 if (try_module_get(head->owner))
904 ret = (void *)head;
905 break;
906 }
907 }
908 xa_unlock(&md_submodule);
909
910 if (!ret) {
911 if (level != LEVEL_NONE)
912 pr_warn("md: personality for level %d is not loaded!\n",
913 level);
914 else
915 pr_warn("md: personality for level %s is not loaded!\n",
916 clevel);
917 }
918
919 return ret;
920 }
921
put_pers(struct md_personality * pers)922 static void put_pers(struct md_personality *pers)
923 {
924 module_put(pers->head.owner);
925 }
926
927 /* return the offset of the super block in 512byte sectors */
calc_dev_sboffset(struct md_rdev * rdev)928 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
929 {
930 return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
931 }
932
alloc_disk_sb(struct md_rdev * rdev)933 static int alloc_disk_sb(struct md_rdev *rdev)
934 {
935 rdev->sb_page = alloc_page(GFP_KERNEL);
936 if (!rdev->sb_page)
937 return -ENOMEM;
938 return 0;
939 }
940
md_rdev_clear(struct md_rdev * rdev)941 void md_rdev_clear(struct md_rdev *rdev)
942 {
943 if (rdev->sb_page) {
944 put_page(rdev->sb_page);
945 rdev->sb_loaded = 0;
946 rdev->sb_page = NULL;
947 rdev->sb_start = 0;
948 rdev->sectors = 0;
949 }
950 if (rdev->bb_page) {
951 put_page(rdev->bb_page);
952 rdev->bb_page = NULL;
953 }
954 badblocks_exit(&rdev->badblocks);
955 }
956 EXPORT_SYMBOL_GPL(md_rdev_clear);
957
super_written(struct bio * bio)958 static void super_written(struct bio *bio)
959 {
960 struct md_rdev *rdev = bio->bi_private;
961 struct mddev *mddev = rdev->mddev;
962
963 if (bio->bi_status) {
964 pr_err("md: %s gets error=%d\n", __func__,
965 blk_status_to_errno(bio->bi_status));
966 md_error(mddev, rdev);
967 if (!test_bit(Faulty, &rdev->flags)
968 && (bio->bi_opf & MD_FAILFAST)) {
969 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
970 set_bit(LastDev, &rdev->flags);
971 }
972 } else
973 clear_bit(LastDev, &rdev->flags);
974
975 bio_put(bio);
976
977 rdev_dec_pending(rdev, mddev);
978
979 if (atomic_dec_and_test(&mddev->pending_writes))
980 wake_up(&mddev->sb_wait);
981 }
982
md_super_write(struct mddev * mddev,struct md_rdev * rdev,sector_t sector,int size,struct page * page)983 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
984 sector_t sector, int size, struct page *page)
985 {
986 /* write first size bytes of page to sector of rdev
987 * Increment mddev->pending_writes before returning
988 * and decrement it on completion, waking up sb_wait
989 * if zero is reached.
990 * If an error occurred, call md_error
991 */
992 struct bio *bio;
993
994 if (!page)
995 return;
996
997 if (test_bit(Faulty, &rdev->flags))
998 return;
999
1000 bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
1001 1,
1002 REQ_OP_WRITE | REQ_SYNC | REQ_IDLE | REQ_META
1003 | REQ_PREFLUSH | REQ_FUA,
1004 GFP_NOIO, &mddev->sync_set);
1005
1006 atomic_inc(&rdev->nr_pending);
1007
1008 bio->bi_iter.bi_sector = sector;
1009 __bio_add_page(bio, page, size, 0);
1010 bio->bi_private = rdev;
1011 bio->bi_end_io = super_written;
1012
1013 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
1014 test_bit(FailFast, &rdev->flags) &&
1015 !test_bit(LastDev, &rdev->flags))
1016 bio->bi_opf |= MD_FAILFAST;
1017
1018 atomic_inc(&mddev->pending_writes);
1019 submit_bio(bio);
1020 }
1021
md_super_wait(struct mddev * mddev)1022 int md_super_wait(struct mddev *mddev)
1023 {
1024 /* wait for all superblock writes that were scheduled to complete */
1025 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1026 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1027 return -EAGAIN;
1028 return 0;
1029 }
1030
sync_page_io(struct md_rdev * rdev,sector_t sector,int size,struct page * page,blk_opf_t opf,bool metadata_op)1031 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1032 struct page *page, blk_opf_t opf, bool metadata_op)
1033 {
1034 struct bio bio;
1035 struct bio_vec bvec;
1036
1037 if (metadata_op && rdev->meta_bdev)
1038 bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
1039 else
1040 bio_init(&bio, rdev->bdev, &bvec, 1, opf);
1041
1042 if (metadata_op)
1043 bio.bi_iter.bi_sector = sector + rdev->sb_start;
1044 else if (rdev->mddev->reshape_position != MaxSector &&
1045 (rdev->mddev->reshape_backwards ==
1046 (sector >= rdev->mddev->reshape_position)))
1047 bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1048 else
1049 bio.bi_iter.bi_sector = sector + rdev->data_offset;
1050 __bio_add_page(&bio, page, size, 0);
1051
1052 submit_bio_wait(&bio);
1053
1054 return !bio.bi_status;
1055 }
1056 EXPORT_SYMBOL_GPL(sync_page_io);
1057
read_disk_sb(struct md_rdev * rdev,int size)1058 static int read_disk_sb(struct md_rdev *rdev, int size)
1059 {
1060 if (rdev->sb_loaded)
1061 return 0;
1062
1063 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1064 goto fail;
1065 rdev->sb_loaded = 1;
1066 return 0;
1067
1068 fail:
1069 pr_err("md: disabled device %pg, could not read superblock.\n",
1070 rdev->bdev);
1071 return -EINVAL;
1072 }
1073
md_uuid_equal(mdp_super_t * sb1,mdp_super_t * sb2)1074 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1075 {
1076 return sb1->set_uuid0 == sb2->set_uuid0 &&
1077 sb1->set_uuid1 == sb2->set_uuid1 &&
1078 sb1->set_uuid2 == sb2->set_uuid2 &&
1079 sb1->set_uuid3 == sb2->set_uuid3;
1080 }
1081
md_sb_equal(mdp_super_t * sb1,mdp_super_t * sb2)1082 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1083 {
1084 int ret;
1085 mdp_super_t *tmp1, *tmp2;
1086
1087 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1088 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1089
1090 if (!tmp1 || !tmp2) {
1091 ret = 0;
1092 goto abort;
1093 }
1094
1095 *tmp1 = *sb1;
1096 *tmp2 = *sb2;
1097
1098 /*
1099 * nr_disks is not constant
1100 */
1101 tmp1->nr_disks = 0;
1102 tmp2->nr_disks = 0;
1103
1104 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1105 abort:
1106 kfree(tmp1);
1107 kfree(tmp2);
1108 return ret;
1109 }
1110
md_csum_fold(u32 csum)1111 static u32 md_csum_fold(u32 csum)
1112 {
1113 csum = (csum & 0xffff) + (csum >> 16);
1114 return (csum & 0xffff) + (csum >> 16);
1115 }
1116
calc_sb_csum(mdp_super_t * sb)1117 static unsigned int calc_sb_csum(mdp_super_t *sb)
1118 {
1119 u64 newcsum = 0;
1120 u32 *sb32 = (u32*)sb;
1121 int i;
1122 unsigned int disk_csum, csum;
1123
1124 disk_csum = sb->sb_csum;
1125 sb->sb_csum = 0;
1126
1127 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1128 newcsum += sb32[i];
1129 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1130
1131 #ifdef CONFIG_ALPHA
1132 /* This used to use csum_partial, which was wrong for several
1133 * reasons including that different results are returned on
1134 * different architectures. It isn't critical that we get exactly
1135 * the same return value as before (we always csum_fold before
1136 * testing, and that removes any differences). However as we
1137 * know that csum_partial always returned a 16bit value on
1138 * alphas, do a fold to maximise conformity to previous behaviour.
1139 */
1140 sb->sb_csum = md_csum_fold(disk_csum);
1141 #else
1142 sb->sb_csum = disk_csum;
1143 #endif
1144 return csum;
1145 }
1146
1147 /*
1148 * Handle superblock details.
1149 * We want to be able to handle multiple superblock formats
1150 * so we have a common interface to them all, and an array of
1151 * different handlers.
1152 * We rely on user-space to write the initial superblock, and support
1153 * reading and updating of superblocks.
1154 * Interface methods are:
1155 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1156 * loads and validates a superblock on dev.
1157 * if refdev != NULL, compare superblocks on both devices
1158 * Return:
1159 * 0 - dev has a superblock that is compatible with refdev
1160 * 1 - dev has a superblock that is compatible and newer than refdev
1161 * so dev should be used as the refdev in future
1162 * -EINVAL superblock incompatible or invalid
1163 * -othererror e.g. -EIO
1164 *
1165 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1166 * Verify that dev is acceptable into mddev.
1167 * The first time, mddev->raid_disks will be 0, and data from
1168 * dev should be merged in. Subsequent calls check that dev
1169 * is new enough. Return 0 or -EINVAL
1170 *
1171 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1172 * Update the superblock for rdev with data in mddev
1173 * This does not write to disc.
1174 *
1175 */
1176
1177 struct super_type {
1178 char *name;
1179 struct module *owner;
1180 int (*load_super)(struct md_rdev *rdev,
1181 struct md_rdev *refdev,
1182 int minor_version);
1183 int (*validate_super)(struct mddev *mddev,
1184 struct md_rdev *freshest,
1185 struct md_rdev *rdev);
1186 void (*sync_super)(struct mddev *mddev,
1187 struct md_rdev *rdev);
1188 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1189 sector_t num_sectors);
1190 int (*allow_new_offset)(struct md_rdev *rdev,
1191 unsigned long long new_offset);
1192 };
1193
1194 /*
1195 * Check that the given mddev has no bitmap.
1196 *
1197 * This function is called from the run method of all personalities that do not
1198 * support bitmaps. It prints an error message and returns non-zero if mddev
1199 * has a bitmap. Otherwise, it returns 0.
1200 *
1201 */
md_check_no_bitmap(struct mddev * mddev)1202 int md_check_no_bitmap(struct mddev *mddev)
1203 {
1204 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1205 return 0;
1206 pr_warn("%s: bitmaps are not supported for %s\n",
1207 mdname(mddev), mddev->pers->head.name);
1208 return 1;
1209 }
1210 EXPORT_SYMBOL(md_check_no_bitmap);
1211
1212 /*
1213 * load_super for 0.90.0
1214 */
super_90_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1215 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1216 {
1217 mdp_super_t *sb;
1218 int ret;
1219 bool spare_disk = true;
1220
1221 /*
1222 * Calculate the position of the superblock (512byte sectors),
1223 * it's at the end of the disk.
1224 *
1225 * It also happens to be a multiple of 4Kb.
1226 */
1227 rdev->sb_start = calc_dev_sboffset(rdev);
1228
1229 ret = read_disk_sb(rdev, MD_SB_BYTES);
1230 if (ret)
1231 return ret;
1232
1233 ret = -EINVAL;
1234
1235 sb = page_address(rdev->sb_page);
1236
1237 if (sb->md_magic != MD_SB_MAGIC) {
1238 pr_warn("md: invalid raid superblock magic on %pg\n",
1239 rdev->bdev);
1240 goto abort;
1241 }
1242
1243 if (sb->major_version != 0 ||
1244 sb->minor_version < 90 ||
1245 sb->minor_version > 91) {
1246 pr_warn("Bad version number %d.%d on %pg\n",
1247 sb->major_version, sb->minor_version, rdev->bdev);
1248 goto abort;
1249 }
1250
1251 if (sb->raid_disks <= 0)
1252 goto abort;
1253
1254 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1255 pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1256 goto abort;
1257 }
1258
1259 rdev->preferred_minor = sb->md_minor;
1260 rdev->data_offset = 0;
1261 rdev->new_data_offset = 0;
1262 rdev->sb_size = MD_SB_BYTES;
1263 rdev->badblocks.shift = -1;
1264
1265 rdev->desc_nr = sb->this_disk.number;
1266
1267 /* not spare disk */
1268 if (rdev->desc_nr >= 0 && rdev->desc_nr < MD_SB_DISKS &&
1269 sb->disks[rdev->desc_nr].state & ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1270 spare_disk = false;
1271
1272 if (!refdev) {
1273 if (!spare_disk)
1274 ret = 1;
1275 else
1276 ret = 0;
1277 } else {
1278 __u64 ev1, ev2;
1279 mdp_super_t *refsb = page_address(refdev->sb_page);
1280 if (!md_uuid_equal(refsb, sb)) {
1281 pr_warn("md: %pg has different UUID to %pg\n",
1282 rdev->bdev, refdev->bdev);
1283 goto abort;
1284 }
1285 if (!md_sb_equal(refsb, sb)) {
1286 pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1287 rdev->bdev, refdev->bdev);
1288 goto abort;
1289 }
1290 ev1 = md_event(sb);
1291 ev2 = md_event(refsb);
1292
1293 if (!spare_disk && ev1 > ev2)
1294 ret = 1;
1295 else
1296 ret = 0;
1297 }
1298 rdev->sectors = rdev->sb_start;
1299 /* Limit to 4TB as metadata cannot record more than that.
1300 * (not needed for Linear and RAID0 as metadata doesn't
1301 * record this size)
1302 */
1303 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1304 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1305
1306 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1307 /* "this cannot possibly happen" ... */
1308 ret = -EINVAL;
1309
1310 abort:
1311 return ret;
1312 }
1313
md_bitmap_events_cleared(struct mddev * mddev)1314 static u64 md_bitmap_events_cleared(struct mddev *mddev)
1315 {
1316 struct md_bitmap_stats stats;
1317 int err;
1318
1319 err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
1320 if (err)
1321 return 0;
1322
1323 return stats.events_cleared;
1324 }
1325
1326 /*
1327 * validate_super for 0.90.0
1328 * note: we are not using "freshest" for 0.9 superblock
1329 */
super_90_validate(struct mddev * mddev,struct md_rdev * freshest,struct md_rdev * rdev)1330 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1331 {
1332 mdp_disk_t *desc;
1333 mdp_super_t *sb = page_address(rdev->sb_page);
1334 __u64 ev1 = md_event(sb);
1335
1336 rdev->raid_disk = -1;
1337 clear_bit(Faulty, &rdev->flags);
1338 clear_bit(In_sync, &rdev->flags);
1339 clear_bit(Bitmap_sync, &rdev->flags);
1340 clear_bit(WriteMostly, &rdev->flags);
1341
1342 if (mddev->raid_disks == 0) {
1343 mddev->major_version = 0;
1344 mddev->minor_version = sb->minor_version;
1345 mddev->patch_version = sb->patch_version;
1346 mddev->external = 0;
1347 mddev->chunk_sectors = sb->chunk_size >> 9;
1348 mddev->ctime = sb->ctime;
1349 mddev->utime = sb->utime;
1350 mddev->level = sb->level;
1351 mddev->clevel[0] = 0;
1352 mddev->layout = sb->layout;
1353 mddev->raid_disks = sb->raid_disks;
1354 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1355 mddev->events = ev1;
1356 mddev->bitmap_info.offset = 0;
1357 mddev->bitmap_info.space = 0;
1358 /* bitmap can use 60 K after the 4K superblocks */
1359 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1360 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1361 mddev->reshape_backwards = 0;
1362
1363 if (mddev->minor_version >= 91) {
1364 mddev->reshape_position = sb->reshape_position;
1365 mddev->delta_disks = sb->delta_disks;
1366 mddev->new_level = sb->new_level;
1367 mddev->new_layout = sb->new_layout;
1368 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1369 if (mddev->delta_disks < 0)
1370 mddev->reshape_backwards = 1;
1371 } else {
1372 mddev->reshape_position = MaxSector;
1373 mddev->delta_disks = 0;
1374 mddev->new_level = mddev->level;
1375 mddev->new_layout = mddev->layout;
1376 mddev->new_chunk_sectors = mddev->chunk_sectors;
1377 }
1378 if (mddev->level == 0)
1379 mddev->layout = -1;
1380
1381 if (sb->state & (1<<MD_SB_CLEAN))
1382 mddev->recovery_cp = MaxSector;
1383 else {
1384 if (sb->events_hi == sb->cp_events_hi &&
1385 sb->events_lo == sb->cp_events_lo) {
1386 mddev->recovery_cp = sb->recovery_cp;
1387 } else
1388 mddev->recovery_cp = 0;
1389 }
1390
1391 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1392 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1393 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1394 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1395
1396 mddev->max_disks = MD_SB_DISKS;
1397
1398 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1399 mddev->bitmap_info.file == NULL) {
1400 mddev->bitmap_info.offset =
1401 mddev->bitmap_info.default_offset;
1402 mddev->bitmap_info.space =
1403 mddev->bitmap_info.default_space;
1404 }
1405
1406 } else if (mddev->pers == NULL) {
1407 /* Insist on good event counter while assembling, except
1408 * for spares (which don't need an event count) */
1409 ++ev1;
1410 if (sb->disks[rdev->desc_nr].state & (
1411 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1412 if (ev1 < mddev->events)
1413 return -EINVAL;
1414 } else if (mddev->bitmap) {
1415 /* if adding to array with a bitmap, then we can accept an
1416 * older device ... but not too old.
1417 */
1418 if (ev1 < md_bitmap_events_cleared(mddev))
1419 return 0;
1420 if (ev1 < mddev->events)
1421 set_bit(Bitmap_sync, &rdev->flags);
1422 } else {
1423 if (ev1 < mddev->events)
1424 /* just a hot-add of a new device, leave raid_disk at -1 */
1425 return 0;
1426 }
1427
1428 desc = sb->disks + rdev->desc_nr;
1429
1430 if (desc->state & (1<<MD_DISK_FAULTY))
1431 set_bit(Faulty, &rdev->flags);
1432 else if (desc->state & (1<<MD_DISK_SYNC)) {
1433 set_bit(In_sync, &rdev->flags);
1434 rdev->raid_disk = desc->raid_disk;
1435 rdev->saved_raid_disk = desc->raid_disk;
1436 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1437 /* active but not in sync implies recovery up to
1438 * reshape position. We don't know exactly where
1439 * that is, so set to zero for now
1440 */
1441 if (mddev->minor_version >= 91) {
1442 rdev->recovery_offset = 0;
1443 rdev->raid_disk = desc->raid_disk;
1444 }
1445 }
1446 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1447 set_bit(WriteMostly, &rdev->flags);
1448 if (desc->state & (1<<MD_DISK_FAILFAST))
1449 set_bit(FailFast, &rdev->flags);
1450 return 0;
1451 }
1452
1453 /*
1454 * sync_super for 0.90.0
1455 */
super_90_sync(struct mddev * mddev,struct md_rdev * rdev)1456 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1457 {
1458 mdp_super_t *sb;
1459 struct md_rdev *rdev2;
1460 int next_spare = mddev->raid_disks;
1461
1462 /* make rdev->sb match mddev data..
1463 *
1464 * 1/ zero out disks
1465 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1466 * 3/ any empty disks < next_spare become removed
1467 *
1468 * disks[0] gets initialised to REMOVED because
1469 * we cannot be sure from other fields if it has
1470 * been initialised or not.
1471 */
1472 int i;
1473 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1474
1475 rdev->sb_size = MD_SB_BYTES;
1476
1477 sb = page_address(rdev->sb_page);
1478
1479 memset(sb, 0, sizeof(*sb));
1480
1481 sb->md_magic = MD_SB_MAGIC;
1482 sb->major_version = mddev->major_version;
1483 sb->patch_version = mddev->patch_version;
1484 sb->gvalid_words = 0; /* ignored */
1485 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1486 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1487 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1488 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1489
1490 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1491 sb->level = mddev->level;
1492 sb->size = mddev->dev_sectors / 2;
1493 sb->raid_disks = mddev->raid_disks;
1494 sb->md_minor = mddev->md_minor;
1495 sb->not_persistent = 0;
1496 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1497 sb->state = 0;
1498 sb->events_hi = (mddev->events>>32);
1499 sb->events_lo = (u32)mddev->events;
1500
1501 if (mddev->reshape_position == MaxSector)
1502 sb->minor_version = 90;
1503 else {
1504 sb->minor_version = 91;
1505 sb->reshape_position = mddev->reshape_position;
1506 sb->new_level = mddev->new_level;
1507 sb->delta_disks = mddev->delta_disks;
1508 sb->new_layout = mddev->new_layout;
1509 sb->new_chunk = mddev->new_chunk_sectors << 9;
1510 }
1511 mddev->minor_version = sb->minor_version;
1512 if (mddev->in_sync)
1513 {
1514 sb->recovery_cp = mddev->recovery_cp;
1515 sb->cp_events_hi = (mddev->events>>32);
1516 sb->cp_events_lo = (u32)mddev->events;
1517 if (mddev->recovery_cp == MaxSector)
1518 sb->state = (1<< MD_SB_CLEAN);
1519 } else
1520 sb->recovery_cp = 0;
1521
1522 sb->layout = mddev->layout;
1523 sb->chunk_size = mddev->chunk_sectors << 9;
1524
1525 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1526 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1527
1528 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1529 rdev_for_each(rdev2, mddev) {
1530 mdp_disk_t *d;
1531 int desc_nr;
1532 int is_active = test_bit(In_sync, &rdev2->flags);
1533
1534 if (rdev2->raid_disk >= 0 &&
1535 sb->minor_version >= 91)
1536 /* we have nowhere to store the recovery_offset,
1537 * but if it is not below the reshape_position,
1538 * we can piggy-back on that.
1539 */
1540 is_active = 1;
1541 if (rdev2->raid_disk < 0 ||
1542 test_bit(Faulty, &rdev2->flags))
1543 is_active = 0;
1544 if (is_active)
1545 desc_nr = rdev2->raid_disk;
1546 else
1547 desc_nr = next_spare++;
1548 rdev2->desc_nr = desc_nr;
1549 d = &sb->disks[rdev2->desc_nr];
1550 nr_disks++;
1551 d->number = rdev2->desc_nr;
1552 d->major = MAJOR(rdev2->bdev->bd_dev);
1553 d->minor = MINOR(rdev2->bdev->bd_dev);
1554 if (is_active)
1555 d->raid_disk = rdev2->raid_disk;
1556 else
1557 d->raid_disk = rdev2->desc_nr; /* compatibility */
1558 if (test_bit(Faulty, &rdev2->flags))
1559 d->state = (1<<MD_DISK_FAULTY);
1560 else if (is_active) {
1561 d->state = (1<<MD_DISK_ACTIVE);
1562 if (test_bit(In_sync, &rdev2->flags))
1563 d->state |= (1<<MD_DISK_SYNC);
1564 active++;
1565 working++;
1566 } else {
1567 d->state = 0;
1568 spare++;
1569 working++;
1570 }
1571 if (test_bit(WriteMostly, &rdev2->flags))
1572 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1573 if (test_bit(FailFast, &rdev2->flags))
1574 d->state |= (1<<MD_DISK_FAILFAST);
1575 }
1576 /* now set the "removed" and "faulty" bits on any missing devices */
1577 for (i=0 ; i < mddev->raid_disks ; i++) {
1578 mdp_disk_t *d = &sb->disks[i];
1579 if (d->state == 0 && d->number == 0) {
1580 d->number = i;
1581 d->raid_disk = i;
1582 d->state = (1<<MD_DISK_REMOVED);
1583 d->state |= (1<<MD_DISK_FAULTY);
1584 failed++;
1585 }
1586 }
1587 sb->nr_disks = nr_disks;
1588 sb->active_disks = active;
1589 sb->working_disks = working;
1590 sb->failed_disks = failed;
1591 sb->spare_disks = spare;
1592
1593 sb->this_disk = sb->disks[rdev->desc_nr];
1594 sb->sb_csum = calc_sb_csum(sb);
1595 }
1596
1597 /*
1598 * rdev_size_change for 0.90.0
1599 */
1600 static unsigned long long
super_90_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)1601 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1602 {
1603 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1604 return 0; /* component must fit device */
1605 if (rdev->mddev->bitmap_info.offset)
1606 return 0; /* can't move bitmap */
1607 rdev->sb_start = calc_dev_sboffset(rdev);
1608 if (!num_sectors || num_sectors > rdev->sb_start)
1609 num_sectors = rdev->sb_start;
1610 /* Limit to 4TB as metadata cannot record more than that.
1611 * 4TB == 2^32 KB, or 2*2^32 sectors.
1612 */
1613 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1614 num_sectors = (sector_t)(2ULL << 32) - 2;
1615 do {
1616 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1617 rdev->sb_page);
1618 } while (md_super_wait(rdev->mddev) < 0);
1619 return num_sectors;
1620 }
1621
1622 static int
super_90_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)1623 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1624 {
1625 /* non-zero offset changes not possible with v0.90 */
1626 return new_offset == 0;
1627 }
1628
1629 /*
1630 * version 1 superblock
1631 */
1632
calc_sb_1_csum(struct mdp_superblock_1 * sb)1633 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1634 {
1635 __le32 disk_csum;
1636 u32 csum;
1637 unsigned long long newcsum;
1638 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1639 __le32 *isuper = (__le32*)sb;
1640
1641 disk_csum = sb->sb_csum;
1642 sb->sb_csum = 0;
1643 newcsum = 0;
1644 for (; size >= 4; size -= 4)
1645 newcsum += le32_to_cpu(*isuper++);
1646
1647 if (size == 2)
1648 newcsum += le16_to_cpu(*(__le16*) isuper);
1649
1650 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1651 sb->sb_csum = disk_csum;
1652 return cpu_to_le32(csum);
1653 }
1654
super_1_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1655 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1656 {
1657 struct mdp_superblock_1 *sb;
1658 int ret;
1659 sector_t sb_start;
1660 sector_t sectors;
1661 int bmask;
1662 bool spare_disk = true;
1663
1664 /*
1665 * Calculate the position of the superblock in 512byte sectors.
1666 * It is always aligned to a 4K boundary and
1667 * depeding on minor_version, it can be:
1668 * 0: At least 8K, but less than 12K, from end of device
1669 * 1: At start of device
1670 * 2: 4K from start of device.
1671 */
1672 switch(minor_version) {
1673 case 0:
1674 sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1675 sb_start &= ~(sector_t)(4*2-1);
1676 break;
1677 case 1:
1678 sb_start = 0;
1679 break;
1680 case 2:
1681 sb_start = 8;
1682 break;
1683 default:
1684 return -EINVAL;
1685 }
1686 rdev->sb_start = sb_start;
1687
1688 /* superblock is rarely larger than 1K, but it can be larger,
1689 * and it is safe to read 4k, so we do that
1690 */
1691 ret = read_disk_sb(rdev, 4096);
1692 if (ret) return ret;
1693
1694 sb = page_address(rdev->sb_page);
1695
1696 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1697 sb->major_version != cpu_to_le32(1) ||
1698 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1699 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1700 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1701 return -EINVAL;
1702
1703 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1704 pr_warn("md: invalid superblock checksum on %pg\n",
1705 rdev->bdev);
1706 return -EINVAL;
1707 }
1708 if (le64_to_cpu(sb->data_size) < 10) {
1709 pr_warn("md: data_size too small on %pg\n",
1710 rdev->bdev);
1711 return -EINVAL;
1712 }
1713 if (sb->pad0 ||
1714 sb->pad3[0] ||
1715 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1716 /* Some padding is non-zero, might be a new feature */
1717 return -EINVAL;
1718
1719 rdev->preferred_minor = 0xffff;
1720 rdev->data_offset = le64_to_cpu(sb->data_offset);
1721 rdev->new_data_offset = rdev->data_offset;
1722 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1723 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1724 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1725 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1726
1727 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1728 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1729 if (rdev->sb_size & bmask)
1730 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1731
1732 if (minor_version
1733 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1734 return -EINVAL;
1735 if (minor_version
1736 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1737 return -EINVAL;
1738
1739 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1740
1741 if (!rdev->bb_page) {
1742 rdev->bb_page = alloc_page(GFP_KERNEL);
1743 if (!rdev->bb_page)
1744 return -ENOMEM;
1745 }
1746 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1747 rdev->badblocks.count == 0) {
1748 /* need to load the bad block list.
1749 * Currently we limit it to one page.
1750 */
1751 s32 offset;
1752 sector_t bb_sector;
1753 __le64 *bbp;
1754 int i;
1755 int sectors = le16_to_cpu(sb->bblog_size);
1756 if (sectors > (PAGE_SIZE / 512))
1757 return -EINVAL;
1758 offset = le32_to_cpu(sb->bblog_offset);
1759 if (offset == 0)
1760 return -EINVAL;
1761 bb_sector = (long long)offset;
1762 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1763 rdev->bb_page, REQ_OP_READ, true))
1764 return -EIO;
1765 bbp = (__le64 *)page_address(rdev->bb_page);
1766 rdev->badblocks.shift = sb->bblog_shift;
1767 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1768 u64 bb = le64_to_cpu(*bbp);
1769 int count = bb & (0x3ff);
1770 u64 sector = bb >> 10;
1771 sector <<= sb->bblog_shift;
1772 count <<= sb->bblog_shift;
1773 if (bb + 1 == 0)
1774 break;
1775 if (!badblocks_set(&rdev->badblocks, sector, count, 1))
1776 return -EINVAL;
1777 }
1778 } else if (sb->bblog_offset != 0)
1779 rdev->badblocks.shift = 0;
1780
1781 if ((le32_to_cpu(sb->feature_map) &
1782 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1783 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1784 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1785 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1786 }
1787
1788 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1789 sb->level != 0)
1790 return -EINVAL;
1791
1792 /* not spare disk */
1793 if (rdev->desc_nr >= 0 && rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1794 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1795 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1796 spare_disk = false;
1797
1798 if (!refdev) {
1799 if (!spare_disk)
1800 ret = 1;
1801 else
1802 ret = 0;
1803 } else {
1804 __u64 ev1, ev2;
1805 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1806
1807 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1808 sb->level != refsb->level ||
1809 sb->layout != refsb->layout ||
1810 sb->chunksize != refsb->chunksize) {
1811 pr_warn("md: %pg has strangely different superblock to %pg\n",
1812 rdev->bdev,
1813 refdev->bdev);
1814 return -EINVAL;
1815 }
1816 ev1 = le64_to_cpu(sb->events);
1817 ev2 = le64_to_cpu(refsb->events);
1818
1819 if (!spare_disk && ev1 > ev2)
1820 ret = 1;
1821 else
1822 ret = 0;
1823 }
1824 if (minor_version)
1825 sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1826 else
1827 sectors = rdev->sb_start;
1828 if (sectors < le64_to_cpu(sb->data_size))
1829 return -EINVAL;
1830 rdev->sectors = le64_to_cpu(sb->data_size);
1831 return ret;
1832 }
1833
super_1_validate(struct mddev * mddev,struct md_rdev * freshest,struct md_rdev * rdev)1834 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1835 {
1836 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1837 __u64 ev1 = le64_to_cpu(sb->events);
1838 int role;
1839
1840 rdev->raid_disk = -1;
1841 clear_bit(Faulty, &rdev->flags);
1842 clear_bit(In_sync, &rdev->flags);
1843 clear_bit(Bitmap_sync, &rdev->flags);
1844 clear_bit(WriteMostly, &rdev->flags);
1845
1846 if (mddev->raid_disks == 0) {
1847 mddev->major_version = 1;
1848 mddev->patch_version = 0;
1849 mddev->external = 0;
1850 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1851 mddev->ctime = le64_to_cpu(sb->ctime);
1852 mddev->utime = le64_to_cpu(sb->utime);
1853 mddev->level = le32_to_cpu(sb->level);
1854 mddev->clevel[0] = 0;
1855 mddev->layout = le32_to_cpu(sb->layout);
1856 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1857 mddev->dev_sectors = le64_to_cpu(sb->size);
1858 mddev->events = ev1;
1859 mddev->bitmap_info.offset = 0;
1860 mddev->bitmap_info.space = 0;
1861 /* Default location for bitmap is 1K after superblock
1862 * using 3K - total of 4K
1863 */
1864 mddev->bitmap_info.default_offset = 1024 >> 9;
1865 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1866 mddev->reshape_backwards = 0;
1867
1868 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1869 memcpy(mddev->uuid, sb->set_uuid, 16);
1870
1871 mddev->max_disks = (4096-256)/2;
1872
1873 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1874 mddev->bitmap_info.file == NULL) {
1875 mddev->bitmap_info.offset =
1876 (__s32)le32_to_cpu(sb->bitmap_offset);
1877 /* Metadata doesn't record how much space is available.
1878 * For 1.0, we assume we can use up to the superblock
1879 * if before, else to 4K beyond superblock.
1880 * For others, assume no change is possible.
1881 */
1882 if (mddev->minor_version > 0)
1883 mddev->bitmap_info.space = 0;
1884 else if (mddev->bitmap_info.offset > 0)
1885 mddev->bitmap_info.space =
1886 8 - mddev->bitmap_info.offset;
1887 else
1888 mddev->bitmap_info.space =
1889 -mddev->bitmap_info.offset;
1890 }
1891
1892 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1893 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1894 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1895 mddev->new_level = le32_to_cpu(sb->new_level);
1896 mddev->new_layout = le32_to_cpu(sb->new_layout);
1897 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1898 if (mddev->delta_disks < 0 ||
1899 (mddev->delta_disks == 0 &&
1900 (le32_to_cpu(sb->feature_map)
1901 & MD_FEATURE_RESHAPE_BACKWARDS)))
1902 mddev->reshape_backwards = 1;
1903 } else {
1904 mddev->reshape_position = MaxSector;
1905 mddev->delta_disks = 0;
1906 mddev->new_level = mddev->level;
1907 mddev->new_layout = mddev->layout;
1908 mddev->new_chunk_sectors = mddev->chunk_sectors;
1909 }
1910
1911 if (mddev->level == 0 &&
1912 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1913 mddev->layout = -1;
1914
1915 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1916 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1917
1918 if (le32_to_cpu(sb->feature_map) &
1919 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1920 if (le32_to_cpu(sb->feature_map) &
1921 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1922 return -EINVAL;
1923 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1924 (le32_to_cpu(sb->feature_map) &
1925 MD_FEATURE_MULTIPLE_PPLS))
1926 return -EINVAL;
1927 set_bit(MD_HAS_PPL, &mddev->flags);
1928 }
1929 } else if (mddev->pers == NULL) {
1930 /* Insist of good event counter while assembling, except for
1931 * spares (which don't need an event count).
1932 * Similar to mdadm, we allow event counter difference of 1
1933 * from the freshest device.
1934 */
1935 if (rdev->desc_nr >= 0 &&
1936 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1937 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1938 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1939 if (ev1 + 1 < mddev->events)
1940 return -EINVAL;
1941 } else if (mddev->bitmap) {
1942 /* If adding to array with a bitmap, then we can accept an
1943 * older device, but not too old.
1944 */
1945 if (ev1 < md_bitmap_events_cleared(mddev))
1946 return 0;
1947 if (ev1 < mddev->events)
1948 set_bit(Bitmap_sync, &rdev->flags);
1949 } else {
1950 if (ev1 < mddev->events)
1951 /* just a hot-add of a new device, leave raid_disk at -1 */
1952 return 0;
1953 }
1954
1955 if (rdev->desc_nr < 0 ||
1956 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1957 role = MD_DISK_ROLE_SPARE;
1958 rdev->desc_nr = -1;
1959 } else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1960 /*
1961 * If we are assembling, and our event counter is smaller than the
1962 * highest event counter, we cannot trust our superblock about the role.
1963 * It could happen that our rdev was marked as Faulty, and all other
1964 * superblocks were updated with +1 event counter.
1965 * Then, before the next superblock update, which typically happens when
1966 * remove_and_add_spares() removes the device from the array, there was
1967 * a crash or reboot.
1968 * If we allow current rdev without consulting the freshest superblock,
1969 * we could cause data corruption.
1970 * Note that in this case our event counter is smaller by 1 than the
1971 * highest, otherwise, this rdev would not be allowed into array;
1972 * both kernel and mdadm allow event counter difference of 1.
1973 */
1974 struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1975 u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1976
1977 if (rdev->desc_nr >= freshest_max_dev) {
1978 /* this is unexpected, better not proceed */
1979 pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1980 mdname(mddev), rdev->bdev, rdev->desc_nr,
1981 freshest->bdev, freshest_max_dev);
1982 return -EUCLEAN;
1983 }
1984
1985 role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1986 pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1987 mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1988 } else {
1989 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1990 }
1991 switch (role) {
1992 case MD_DISK_ROLE_SPARE: /* spare */
1993 break;
1994 case MD_DISK_ROLE_FAULTY: /* faulty */
1995 set_bit(Faulty, &rdev->flags);
1996 break;
1997 case MD_DISK_ROLE_JOURNAL: /* journal device */
1998 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1999 /* journal device without journal feature */
2000 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
2001 return -EINVAL;
2002 }
2003 set_bit(Journal, &rdev->flags);
2004 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
2005 rdev->raid_disk = 0;
2006 break;
2007 default:
2008 rdev->saved_raid_disk = role;
2009 if ((le32_to_cpu(sb->feature_map) &
2010 MD_FEATURE_RECOVERY_OFFSET)) {
2011 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
2012 if (!(le32_to_cpu(sb->feature_map) &
2013 MD_FEATURE_RECOVERY_BITMAP))
2014 rdev->saved_raid_disk = -1;
2015 } else {
2016 /*
2017 * If the array is FROZEN, then the device can't
2018 * be in_sync with rest of array.
2019 */
2020 if (!test_bit(MD_RECOVERY_FROZEN,
2021 &mddev->recovery))
2022 set_bit(In_sync, &rdev->flags);
2023 }
2024 rdev->raid_disk = role;
2025 break;
2026 }
2027 if (sb->devflags & WriteMostly1)
2028 set_bit(WriteMostly, &rdev->flags);
2029 if (sb->devflags & FailFast1)
2030 set_bit(FailFast, &rdev->flags);
2031 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
2032 set_bit(Replacement, &rdev->flags);
2033
2034 return 0;
2035 }
2036
super_1_sync(struct mddev * mddev,struct md_rdev * rdev)2037 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2038 {
2039 struct mdp_superblock_1 *sb;
2040 struct md_rdev *rdev2;
2041 int max_dev, i;
2042 /* make rdev->sb match mddev and rdev data. */
2043
2044 sb = page_address(rdev->sb_page);
2045
2046 sb->feature_map = 0;
2047 sb->pad0 = 0;
2048 sb->recovery_offset = cpu_to_le64(0);
2049 memset(sb->pad3, 0, sizeof(sb->pad3));
2050
2051 sb->utime = cpu_to_le64((__u64)mddev->utime);
2052 sb->events = cpu_to_le64(mddev->events);
2053 if (mddev->in_sync)
2054 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2055 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2056 sb->resync_offset = cpu_to_le64(MaxSector);
2057 else
2058 sb->resync_offset = cpu_to_le64(0);
2059
2060 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2061
2062 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2063 sb->size = cpu_to_le64(mddev->dev_sectors);
2064 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2065 sb->level = cpu_to_le32(mddev->level);
2066 sb->layout = cpu_to_le32(mddev->layout);
2067 if (test_bit(FailFast, &rdev->flags))
2068 sb->devflags |= FailFast1;
2069 else
2070 sb->devflags &= ~FailFast1;
2071
2072 if (test_bit(WriteMostly, &rdev->flags))
2073 sb->devflags |= WriteMostly1;
2074 else
2075 sb->devflags &= ~WriteMostly1;
2076 sb->data_offset = cpu_to_le64(rdev->data_offset);
2077 sb->data_size = cpu_to_le64(rdev->sectors);
2078
2079 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2080 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2081 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2082 }
2083
2084 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2085 !test_bit(In_sync, &rdev->flags)) {
2086 sb->feature_map |=
2087 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2088 sb->recovery_offset =
2089 cpu_to_le64(rdev->recovery_offset);
2090 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2091 sb->feature_map |=
2092 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2093 }
2094 /* Note: recovery_offset and journal_tail share space */
2095 if (test_bit(Journal, &rdev->flags))
2096 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2097 if (test_bit(Replacement, &rdev->flags))
2098 sb->feature_map |=
2099 cpu_to_le32(MD_FEATURE_REPLACEMENT);
2100
2101 if (mddev->reshape_position != MaxSector) {
2102 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2103 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2104 sb->new_layout = cpu_to_le32(mddev->new_layout);
2105 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2106 sb->new_level = cpu_to_le32(mddev->new_level);
2107 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2108 if (mddev->delta_disks == 0 &&
2109 mddev->reshape_backwards)
2110 sb->feature_map
2111 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2112 if (rdev->new_data_offset != rdev->data_offset) {
2113 sb->feature_map
2114 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2115 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2116 - rdev->data_offset));
2117 }
2118 }
2119
2120 if (mddev_is_clustered(mddev))
2121 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2122
2123 if (rdev->badblocks.count == 0)
2124 /* Nothing to do for bad blocks*/ ;
2125 else if (sb->bblog_offset == 0)
2126 /* Cannot record bad blocks on this device */
2127 md_error(mddev, rdev);
2128 else {
2129 struct badblocks *bb = &rdev->badblocks;
2130 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2131 u64 *p = bb->page;
2132 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2133 if (bb->changed) {
2134 unsigned seq;
2135
2136 retry:
2137 seq = read_seqbegin(&bb->lock);
2138
2139 memset(bbp, 0xff, PAGE_SIZE);
2140
2141 for (i = 0 ; i < bb->count ; i++) {
2142 u64 internal_bb = p[i];
2143 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2144 | BB_LEN(internal_bb));
2145 bbp[i] = cpu_to_le64(store_bb);
2146 }
2147 bb->changed = 0;
2148 if (read_seqretry(&bb->lock, seq))
2149 goto retry;
2150
2151 bb->sector = (rdev->sb_start +
2152 (int)le32_to_cpu(sb->bblog_offset));
2153 bb->size = le16_to_cpu(sb->bblog_size);
2154 }
2155 }
2156
2157 max_dev = 0;
2158 rdev_for_each(rdev2, mddev)
2159 if (rdev2->desc_nr+1 > max_dev)
2160 max_dev = rdev2->desc_nr+1;
2161
2162 if (max_dev > le32_to_cpu(sb->max_dev)) {
2163 int bmask;
2164 sb->max_dev = cpu_to_le32(max_dev);
2165 rdev->sb_size = max_dev * 2 + 256;
2166 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2167 if (rdev->sb_size & bmask)
2168 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2169 } else
2170 max_dev = le32_to_cpu(sb->max_dev);
2171
2172 for (i=0; i<max_dev;i++)
2173 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2174
2175 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2176 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2177
2178 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2179 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2180 sb->feature_map |=
2181 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2182 else
2183 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2184 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2185 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2186 }
2187
2188 rdev_for_each(rdev2, mddev) {
2189 i = rdev2->desc_nr;
2190 if (test_bit(Faulty, &rdev2->flags))
2191 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2192 else if (test_bit(In_sync, &rdev2->flags))
2193 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2194 else if (test_bit(Journal, &rdev2->flags))
2195 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2196 else if (rdev2->raid_disk >= 0)
2197 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2198 else
2199 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2200 }
2201
2202 sb->sb_csum = calc_sb_1_csum(sb);
2203 }
2204
super_1_choose_bm_space(sector_t dev_size)2205 static sector_t super_1_choose_bm_space(sector_t dev_size)
2206 {
2207 sector_t bm_space;
2208
2209 /* if the device is bigger than 8Gig, save 64k for bitmap
2210 * usage, if bigger than 200Gig, save 128k
2211 */
2212 if (dev_size < 64*2)
2213 bm_space = 0;
2214 else if (dev_size - 64*2 >= 200*1024*1024*2)
2215 bm_space = 128*2;
2216 else if (dev_size - 4*2 > 8*1024*1024*2)
2217 bm_space = 64*2;
2218 else
2219 bm_space = 4*2;
2220 return bm_space;
2221 }
2222
2223 static unsigned long long
super_1_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)2224 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2225 {
2226 struct mdp_superblock_1 *sb;
2227 sector_t max_sectors;
2228 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2229 return 0; /* component must fit device */
2230 if (rdev->data_offset != rdev->new_data_offset)
2231 return 0; /* too confusing */
2232 if (rdev->sb_start < rdev->data_offset) {
2233 /* minor versions 1 and 2; superblock before data */
2234 max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2235 if (!num_sectors || num_sectors > max_sectors)
2236 num_sectors = max_sectors;
2237 } else if (rdev->mddev->bitmap_info.offset) {
2238 /* minor version 0 with bitmap we can't move */
2239 return 0;
2240 } else {
2241 /* minor version 0; superblock after data */
2242 sector_t sb_start, bm_space;
2243 sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2244
2245 /* 8K is for superblock */
2246 sb_start = dev_size - 8*2;
2247 sb_start &= ~(sector_t)(4*2 - 1);
2248
2249 bm_space = super_1_choose_bm_space(dev_size);
2250
2251 /* Space that can be used to store date needs to decrease
2252 * superblock bitmap space and bad block space(4K)
2253 */
2254 max_sectors = sb_start - bm_space - 4*2;
2255
2256 if (!num_sectors || num_sectors > max_sectors)
2257 num_sectors = max_sectors;
2258 rdev->sb_start = sb_start;
2259 }
2260 sb = page_address(rdev->sb_page);
2261 sb->data_size = cpu_to_le64(num_sectors);
2262 sb->super_offset = cpu_to_le64(rdev->sb_start);
2263 sb->sb_csum = calc_sb_1_csum(sb);
2264 do {
2265 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2266 rdev->sb_page);
2267 } while (md_super_wait(rdev->mddev) < 0);
2268 return num_sectors;
2269
2270 }
2271
2272 static int
super_1_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)2273 super_1_allow_new_offset(struct md_rdev *rdev,
2274 unsigned long long new_offset)
2275 {
2276 /* All necessary checks on new >= old have been done */
2277 if (new_offset >= rdev->data_offset)
2278 return 1;
2279
2280 /* with 1.0 metadata, there is no metadata to tread on
2281 * so we can always move back */
2282 if (rdev->mddev->minor_version == 0)
2283 return 1;
2284
2285 /* otherwise we must be sure not to step on
2286 * any metadata, so stay:
2287 * 36K beyond start of superblock
2288 * beyond end of badblocks
2289 * beyond write-intent bitmap
2290 */
2291 if (rdev->sb_start + (32+4)*2 > new_offset)
2292 return 0;
2293
2294 if (!rdev->mddev->bitmap_info.file) {
2295 struct mddev *mddev = rdev->mddev;
2296 struct md_bitmap_stats stats;
2297 int err;
2298
2299 err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
2300 if (!err && rdev->sb_start + mddev->bitmap_info.offset +
2301 stats.file_pages * (PAGE_SIZE >> 9) > new_offset)
2302 return 0;
2303 }
2304
2305 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2306 return 0;
2307
2308 return 1;
2309 }
2310
2311 static struct super_type super_types[] = {
2312 [0] = {
2313 .name = "0.90.0",
2314 .owner = THIS_MODULE,
2315 .load_super = super_90_load,
2316 .validate_super = super_90_validate,
2317 .sync_super = super_90_sync,
2318 .rdev_size_change = super_90_rdev_size_change,
2319 .allow_new_offset = super_90_allow_new_offset,
2320 },
2321 [1] = {
2322 .name = "md-1",
2323 .owner = THIS_MODULE,
2324 .load_super = super_1_load,
2325 .validate_super = super_1_validate,
2326 .sync_super = super_1_sync,
2327 .rdev_size_change = super_1_rdev_size_change,
2328 .allow_new_offset = super_1_allow_new_offset,
2329 },
2330 };
2331
sync_super(struct mddev * mddev,struct md_rdev * rdev)2332 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2333 {
2334 if (mddev->sync_super) {
2335 mddev->sync_super(mddev, rdev);
2336 return;
2337 }
2338
2339 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2340
2341 super_types[mddev->major_version].sync_super(mddev, rdev);
2342 }
2343
match_mddev_units(struct mddev * mddev1,struct mddev * mddev2)2344 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2345 {
2346 struct md_rdev *rdev, *rdev2;
2347
2348 rcu_read_lock();
2349 rdev_for_each_rcu(rdev, mddev1) {
2350 if (test_bit(Faulty, &rdev->flags) ||
2351 test_bit(Journal, &rdev->flags) ||
2352 rdev->raid_disk == -1)
2353 continue;
2354 rdev_for_each_rcu(rdev2, mddev2) {
2355 if (test_bit(Faulty, &rdev2->flags) ||
2356 test_bit(Journal, &rdev2->flags) ||
2357 rdev2->raid_disk == -1)
2358 continue;
2359 if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2360 rcu_read_unlock();
2361 return 1;
2362 }
2363 }
2364 }
2365 rcu_read_unlock();
2366 return 0;
2367 }
2368
2369 static LIST_HEAD(pending_raid_disks);
2370
2371 /*
2372 * Try to register data integrity profile for an mddev
2373 *
2374 * This is called when an array is started and after a disk has been kicked
2375 * from the array. It only succeeds if all working and active component devices
2376 * are integrity capable with matching profiles.
2377 */
md_integrity_register(struct mddev * mddev)2378 int md_integrity_register(struct mddev *mddev)
2379 {
2380 if (list_empty(&mddev->disks))
2381 return 0; /* nothing to do */
2382 if (mddev_is_dm(mddev) || !blk_get_integrity(mddev->gendisk))
2383 return 0; /* shouldn't register */
2384
2385 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2386 return 0;
2387 }
2388 EXPORT_SYMBOL(md_integrity_register);
2389
rdev_read_only(struct md_rdev * rdev)2390 static bool rdev_read_only(struct md_rdev *rdev)
2391 {
2392 return bdev_read_only(rdev->bdev) ||
2393 (rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2394 }
2395
bind_rdev_to_array(struct md_rdev * rdev,struct mddev * mddev)2396 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2397 {
2398 char b[BDEVNAME_SIZE];
2399 int err;
2400
2401 /* prevent duplicates */
2402 if (find_rdev(mddev, rdev->bdev->bd_dev))
2403 return -EEXIST;
2404
2405 if (rdev_read_only(rdev) && mddev->pers)
2406 return -EROFS;
2407
2408 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2409 if (!test_bit(Journal, &rdev->flags) &&
2410 rdev->sectors &&
2411 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2412 if (mddev->pers) {
2413 /* Cannot change size, so fail
2414 * If mddev->level <= 0, then we don't care
2415 * about aligning sizes (e.g. linear)
2416 */
2417 if (mddev->level > 0)
2418 return -ENOSPC;
2419 } else
2420 mddev->dev_sectors = rdev->sectors;
2421 }
2422
2423 /* Verify rdev->desc_nr is unique.
2424 * If it is -1, assign a free number, else
2425 * check number is not in use
2426 */
2427 rcu_read_lock();
2428 if (rdev->desc_nr < 0) {
2429 int choice = 0;
2430 if (mddev->pers)
2431 choice = mddev->raid_disks;
2432 while (md_find_rdev_nr_rcu(mddev, choice))
2433 choice++;
2434 rdev->desc_nr = choice;
2435 } else {
2436 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2437 rcu_read_unlock();
2438 return -EBUSY;
2439 }
2440 }
2441 rcu_read_unlock();
2442 if (!test_bit(Journal, &rdev->flags) &&
2443 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2444 pr_warn("md: %s: array is limited to %d devices\n",
2445 mdname(mddev), mddev->max_disks);
2446 return -EBUSY;
2447 }
2448 snprintf(b, sizeof(b), "%pg", rdev->bdev);
2449 strreplace(b, '/', '!');
2450
2451 rdev->mddev = mddev;
2452 pr_debug("md: bind<%s>\n", b);
2453
2454 if (mddev->raid_disks)
2455 mddev_create_serial_pool(mddev, rdev);
2456
2457 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2458 goto fail;
2459
2460 /* failure here is OK */
2461 err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2462 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2463 rdev->sysfs_unack_badblocks =
2464 sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2465 rdev->sysfs_badblocks =
2466 sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2467
2468 list_add_rcu(&rdev->same_set, &mddev->disks);
2469 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2470
2471 /* May as well allow recovery to be retried once */
2472 mddev->recovery_disabled++;
2473
2474 return 0;
2475
2476 fail:
2477 pr_warn("md: failed to register dev-%s for %s\n",
2478 b, mdname(mddev));
2479 mddev_destroy_serial_pool(mddev, rdev);
2480 return err;
2481 }
2482
2483 void md_autodetect_dev(dev_t dev);
2484
2485 /* just for claiming the bdev */
2486 static struct md_rdev claim_rdev;
2487
export_rdev(struct md_rdev * rdev,struct mddev * mddev)2488 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev)
2489 {
2490 pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2491 md_rdev_clear(rdev);
2492 #ifndef MODULE
2493 if (test_bit(AutoDetected, &rdev->flags))
2494 md_autodetect_dev(rdev->bdev->bd_dev);
2495 #endif
2496 fput(rdev->bdev_file);
2497 rdev->bdev = NULL;
2498 kobject_put(&rdev->kobj);
2499 }
2500
md_kick_rdev_from_array(struct md_rdev * rdev)2501 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2502 {
2503 struct mddev *mddev = rdev->mddev;
2504
2505 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2506 list_del_rcu(&rdev->same_set);
2507 pr_debug("md: unbind<%pg>\n", rdev->bdev);
2508 mddev_destroy_serial_pool(rdev->mddev, rdev);
2509 WRITE_ONCE(rdev->mddev, NULL);
2510 sysfs_remove_link(&rdev->kobj, "block");
2511 sysfs_put(rdev->sysfs_state);
2512 sysfs_put(rdev->sysfs_unack_badblocks);
2513 sysfs_put(rdev->sysfs_badblocks);
2514 rdev->sysfs_state = NULL;
2515 rdev->sysfs_unack_badblocks = NULL;
2516 rdev->sysfs_badblocks = NULL;
2517 rdev->badblocks.count = 0;
2518
2519 synchronize_rcu();
2520
2521 /*
2522 * kobject_del() will wait for all in progress writers to be done, where
2523 * reconfig_mutex is held, hence it can't be called under
2524 * reconfig_mutex and it's delayed to mddev_unlock().
2525 */
2526 list_add(&rdev->same_set, &mddev->deleting);
2527 }
2528
export_array(struct mddev * mddev)2529 static void export_array(struct mddev *mddev)
2530 {
2531 struct md_rdev *rdev;
2532
2533 while (!list_empty(&mddev->disks)) {
2534 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2535 same_set);
2536 md_kick_rdev_from_array(rdev);
2537 }
2538 mddev->raid_disks = 0;
2539 mddev->major_version = 0;
2540 }
2541
set_in_sync(struct mddev * mddev)2542 static bool set_in_sync(struct mddev *mddev)
2543 {
2544 lockdep_assert_held(&mddev->lock);
2545 if (!mddev->in_sync) {
2546 mddev->sync_checkers++;
2547 spin_unlock(&mddev->lock);
2548 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2549 spin_lock(&mddev->lock);
2550 if (!mddev->in_sync &&
2551 percpu_ref_is_zero(&mddev->writes_pending)) {
2552 mddev->in_sync = 1;
2553 /*
2554 * Ensure ->in_sync is visible before we clear
2555 * ->sync_checkers.
2556 */
2557 smp_mb();
2558 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2559 sysfs_notify_dirent_safe(mddev->sysfs_state);
2560 }
2561 if (--mddev->sync_checkers == 0)
2562 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2563 }
2564 if (mddev->safemode == 1)
2565 mddev->safemode = 0;
2566 return mddev->in_sync;
2567 }
2568
sync_sbs(struct mddev * mddev,int nospares)2569 static void sync_sbs(struct mddev *mddev, int nospares)
2570 {
2571 /* Update each superblock (in-memory image), but
2572 * if we are allowed to, skip spares which already
2573 * have the right event counter, or have one earlier
2574 * (which would mean they aren't being marked as dirty
2575 * with the rest of the array)
2576 */
2577 struct md_rdev *rdev;
2578 rdev_for_each(rdev, mddev) {
2579 if (rdev->sb_events == mddev->events ||
2580 (nospares &&
2581 rdev->raid_disk < 0 &&
2582 rdev->sb_events+1 == mddev->events)) {
2583 /* Don't update this superblock */
2584 rdev->sb_loaded = 2;
2585 } else {
2586 sync_super(mddev, rdev);
2587 rdev->sb_loaded = 1;
2588 }
2589 }
2590 }
2591
does_sb_need_changing(struct mddev * mddev)2592 static bool does_sb_need_changing(struct mddev *mddev)
2593 {
2594 struct md_rdev *rdev = NULL, *iter;
2595 struct mdp_superblock_1 *sb;
2596 int role;
2597
2598 /* Find a good rdev */
2599 rdev_for_each(iter, mddev)
2600 if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2601 rdev = iter;
2602 break;
2603 }
2604
2605 /* No good device found. */
2606 if (!rdev)
2607 return false;
2608
2609 sb = page_address(rdev->sb_page);
2610 /* Check if a device has become faulty or a spare become active */
2611 rdev_for_each(rdev, mddev) {
2612 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2613 /* Device activated? */
2614 if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2615 !test_bit(Faulty, &rdev->flags))
2616 return true;
2617 /* Device turned faulty? */
2618 if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2619 return true;
2620 }
2621
2622 /* Check if any mddev parameters have changed */
2623 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2624 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2625 (mddev->layout != le32_to_cpu(sb->layout)) ||
2626 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2627 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2628 return true;
2629
2630 return false;
2631 }
2632
md_update_sb(struct mddev * mddev,int force_change)2633 void md_update_sb(struct mddev *mddev, int force_change)
2634 {
2635 struct md_rdev *rdev;
2636 int sync_req;
2637 int nospares = 0;
2638 int any_badblocks_changed = 0;
2639 int ret = -1;
2640
2641 if (!md_is_rdwr(mddev)) {
2642 if (force_change)
2643 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2644 return;
2645 }
2646
2647 repeat:
2648 if (mddev_is_clustered(mddev)) {
2649 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2650 force_change = 1;
2651 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2652 nospares = 1;
2653 ret = mddev->cluster_ops->metadata_update_start(mddev);
2654 /* Has someone else has updated the sb */
2655 if (!does_sb_need_changing(mddev)) {
2656 if (ret == 0)
2657 mddev->cluster_ops->metadata_update_cancel(mddev);
2658 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2659 BIT(MD_SB_CHANGE_DEVS) |
2660 BIT(MD_SB_CHANGE_CLEAN));
2661 return;
2662 }
2663 }
2664
2665 /*
2666 * First make sure individual recovery_offsets are correct
2667 * curr_resync_completed can only be used during recovery.
2668 * During reshape/resync it might use array-addresses rather
2669 * that device addresses.
2670 */
2671 rdev_for_each(rdev, mddev) {
2672 if (rdev->raid_disk >= 0 &&
2673 mddev->delta_disks >= 0 &&
2674 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2675 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2676 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2677 !test_bit(Journal, &rdev->flags) &&
2678 !test_bit(In_sync, &rdev->flags) &&
2679 mddev->curr_resync_completed > rdev->recovery_offset)
2680 rdev->recovery_offset = mddev->curr_resync_completed;
2681
2682 }
2683 if (!mddev->persistent) {
2684 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2685 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2686 if (!mddev->external) {
2687 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2688 rdev_for_each(rdev, mddev) {
2689 if (rdev->badblocks.changed) {
2690 rdev->badblocks.changed = 0;
2691 ack_all_badblocks(&rdev->badblocks);
2692 md_error(mddev, rdev);
2693 }
2694 clear_bit(Blocked, &rdev->flags);
2695 clear_bit(BlockedBadBlocks, &rdev->flags);
2696 wake_up(&rdev->blocked_wait);
2697 }
2698 }
2699 wake_up(&mddev->sb_wait);
2700 return;
2701 }
2702
2703 spin_lock(&mddev->lock);
2704
2705 mddev->utime = ktime_get_real_seconds();
2706
2707 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2708 force_change = 1;
2709 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2710 /* just a clean<-> dirty transition, possibly leave spares alone,
2711 * though if events isn't the right even/odd, we will have to do
2712 * spares after all
2713 */
2714 nospares = 1;
2715 if (force_change)
2716 nospares = 0;
2717 if (mddev->degraded)
2718 /* If the array is degraded, then skipping spares is both
2719 * dangerous and fairly pointless.
2720 * Dangerous because a device that was removed from the array
2721 * might have a event_count that still looks up-to-date,
2722 * so it can be re-added without a resync.
2723 * Pointless because if there are any spares to skip,
2724 * then a recovery will happen and soon that array won't
2725 * be degraded any more and the spare can go back to sleep then.
2726 */
2727 nospares = 0;
2728
2729 sync_req = mddev->in_sync;
2730
2731 /* If this is just a dirty<->clean transition, and the array is clean
2732 * and 'events' is odd, we can roll back to the previous clean state */
2733 if (nospares
2734 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2735 && mddev->can_decrease_events
2736 && mddev->events != 1) {
2737 mddev->events--;
2738 mddev->can_decrease_events = 0;
2739 } else {
2740 /* otherwise we have to go forward and ... */
2741 mddev->events ++;
2742 mddev->can_decrease_events = nospares;
2743 }
2744
2745 /*
2746 * This 64-bit counter should never wrap.
2747 * Either we are in around ~1 trillion A.C., assuming
2748 * 1 reboot per second, or we have a bug...
2749 */
2750 WARN_ON(mddev->events == 0);
2751
2752 rdev_for_each(rdev, mddev) {
2753 if (rdev->badblocks.changed)
2754 any_badblocks_changed++;
2755 if (test_bit(Faulty, &rdev->flags))
2756 set_bit(FaultRecorded, &rdev->flags);
2757 }
2758
2759 sync_sbs(mddev, nospares);
2760 spin_unlock(&mddev->lock);
2761
2762 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2763 mdname(mddev), mddev->in_sync);
2764
2765 mddev_add_trace_msg(mddev, "md md_update_sb");
2766 rewrite:
2767 mddev->bitmap_ops->update_sb(mddev->bitmap);
2768 rdev_for_each(rdev, mddev) {
2769 if (rdev->sb_loaded != 1)
2770 continue; /* no noise on spare devices */
2771
2772 if (!test_bit(Faulty, &rdev->flags)) {
2773 md_super_write(mddev,rdev,
2774 rdev->sb_start, rdev->sb_size,
2775 rdev->sb_page);
2776 pr_debug("md: (write) %pg's sb offset: %llu\n",
2777 rdev->bdev,
2778 (unsigned long long)rdev->sb_start);
2779 rdev->sb_events = mddev->events;
2780 if (rdev->badblocks.size) {
2781 md_super_write(mddev, rdev,
2782 rdev->badblocks.sector,
2783 rdev->badblocks.size << 9,
2784 rdev->bb_page);
2785 rdev->badblocks.size = 0;
2786 }
2787
2788 } else
2789 pr_debug("md: %pg (skipping faulty)\n",
2790 rdev->bdev);
2791 }
2792 if (md_super_wait(mddev) < 0)
2793 goto rewrite;
2794 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2795
2796 if (mddev_is_clustered(mddev) && ret == 0)
2797 mddev->cluster_ops->metadata_update_finish(mddev);
2798
2799 if (mddev->in_sync != sync_req ||
2800 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2801 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2802 /* have to write it out again */
2803 goto repeat;
2804 wake_up(&mddev->sb_wait);
2805 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2806 sysfs_notify_dirent_safe(mddev->sysfs_completed);
2807
2808 rdev_for_each(rdev, mddev) {
2809 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2810 clear_bit(Blocked, &rdev->flags);
2811
2812 if (any_badblocks_changed)
2813 ack_all_badblocks(&rdev->badblocks);
2814 clear_bit(BlockedBadBlocks, &rdev->flags);
2815 wake_up(&rdev->blocked_wait);
2816 }
2817 }
2818 EXPORT_SYMBOL(md_update_sb);
2819
add_bound_rdev(struct md_rdev * rdev)2820 static int add_bound_rdev(struct md_rdev *rdev)
2821 {
2822 struct mddev *mddev = rdev->mddev;
2823 int err = 0;
2824 bool add_journal = test_bit(Journal, &rdev->flags);
2825
2826 if (!mddev->pers->hot_remove_disk || add_journal) {
2827 /* If there is hot_add_disk but no hot_remove_disk
2828 * then added disks for geometry changes,
2829 * and should be added immediately.
2830 */
2831 super_types[mddev->major_version].
2832 validate_super(mddev, NULL/*freshest*/, rdev);
2833 err = mddev->pers->hot_add_disk(mddev, rdev);
2834 if (err) {
2835 md_kick_rdev_from_array(rdev);
2836 return err;
2837 }
2838 }
2839 sysfs_notify_dirent_safe(rdev->sysfs_state);
2840
2841 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2842 if (mddev->degraded)
2843 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2844 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2845 md_new_event();
2846 return 0;
2847 }
2848
2849 /* words written to sysfs files may, or may not, be \n terminated.
2850 * We want to accept with case. For this we use cmd_match.
2851 */
cmd_match(const char * cmd,const char * str)2852 static int cmd_match(const char *cmd, const char *str)
2853 {
2854 /* See if cmd, written into a sysfs file, matches
2855 * str. They must either be the same, or cmd can
2856 * have a trailing newline
2857 */
2858 while (*cmd && *str && *cmd == *str) {
2859 cmd++;
2860 str++;
2861 }
2862 if (*cmd == '\n')
2863 cmd++;
2864 if (*str || *cmd)
2865 return 0;
2866 return 1;
2867 }
2868
2869 struct rdev_sysfs_entry {
2870 struct attribute attr;
2871 ssize_t (*show)(struct md_rdev *, char *);
2872 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2873 };
2874
2875 static ssize_t
state_show(struct md_rdev * rdev,char * page)2876 state_show(struct md_rdev *rdev, char *page)
2877 {
2878 char *sep = ",";
2879 size_t len = 0;
2880 unsigned long flags = READ_ONCE(rdev->flags);
2881
2882 if (test_bit(Faulty, &flags) ||
2883 (!test_bit(ExternalBbl, &flags) &&
2884 rdev->badblocks.unacked_exist))
2885 len += sprintf(page+len, "faulty%s", sep);
2886 if (test_bit(In_sync, &flags))
2887 len += sprintf(page+len, "in_sync%s", sep);
2888 if (test_bit(Journal, &flags))
2889 len += sprintf(page+len, "journal%s", sep);
2890 if (test_bit(WriteMostly, &flags))
2891 len += sprintf(page+len, "write_mostly%s", sep);
2892 if (test_bit(Blocked, &flags) ||
2893 (rdev->badblocks.unacked_exist
2894 && !test_bit(Faulty, &flags)))
2895 len += sprintf(page+len, "blocked%s", sep);
2896 if (!test_bit(Faulty, &flags) &&
2897 !test_bit(Journal, &flags) &&
2898 !test_bit(In_sync, &flags))
2899 len += sprintf(page+len, "spare%s", sep);
2900 if (test_bit(WriteErrorSeen, &flags))
2901 len += sprintf(page+len, "write_error%s", sep);
2902 if (test_bit(WantReplacement, &flags))
2903 len += sprintf(page+len, "want_replacement%s", sep);
2904 if (test_bit(Replacement, &flags))
2905 len += sprintf(page+len, "replacement%s", sep);
2906 if (test_bit(ExternalBbl, &flags))
2907 len += sprintf(page+len, "external_bbl%s", sep);
2908 if (test_bit(FailFast, &flags))
2909 len += sprintf(page+len, "failfast%s", sep);
2910
2911 if (len)
2912 len -= strlen(sep);
2913
2914 return len+sprintf(page+len, "\n");
2915 }
2916
2917 static ssize_t
state_store(struct md_rdev * rdev,const char * buf,size_t len)2918 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2919 {
2920 /* can write
2921 * faulty - simulates an error
2922 * remove - disconnects the device
2923 * writemostly - sets write_mostly
2924 * -writemostly - clears write_mostly
2925 * blocked - sets the Blocked flags
2926 * -blocked - clears the Blocked and possibly simulates an error
2927 * insync - sets Insync providing device isn't active
2928 * -insync - clear Insync for a device with a slot assigned,
2929 * so that it gets rebuilt based on bitmap
2930 * write_error - sets WriteErrorSeen
2931 * -write_error - clears WriteErrorSeen
2932 * {,-}failfast - set/clear FailFast
2933 */
2934
2935 struct mddev *mddev = rdev->mddev;
2936 int err = -EINVAL;
2937 bool need_update_sb = false;
2938
2939 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2940 md_error(rdev->mddev, rdev);
2941
2942 if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2943 err = -EBUSY;
2944 else
2945 err = 0;
2946 } else if (cmd_match(buf, "remove")) {
2947 if (rdev->mddev->pers) {
2948 clear_bit(Blocked, &rdev->flags);
2949 remove_and_add_spares(rdev->mddev, rdev);
2950 }
2951 if (rdev->raid_disk >= 0)
2952 err = -EBUSY;
2953 else {
2954 err = 0;
2955 if (mddev_is_clustered(mddev))
2956 err = mddev->cluster_ops->remove_disk(mddev, rdev);
2957
2958 if (err == 0) {
2959 md_kick_rdev_from_array(rdev);
2960 if (mddev->pers)
2961 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2962 md_new_event();
2963 }
2964 }
2965 } else if (cmd_match(buf, "writemostly")) {
2966 set_bit(WriteMostly, &rdev->flags);
2967 mddev_create_serial_pool(rdev->mddev, rdev);
2968 need_update_sb = true;
2969 err = 0;
2970 } else if (cmd_match(buf, "-writemostly")) {
2971 mddev_destroy_serial_pool(rdev->mddev, rdev);
2972 clear_bit(WriteMostly, &rdev->flags);
2973 need_update_sb = true;
2974 err = 0;
2975 } else if (cmd_match(buf, "blocked")) {
2976 set_bit(Blocked, &rdev->flags);
2977 err = 0;
2978 } else if (cmd_match(buf, "-blocked")) {
2979 if (!test_bit(Faulty, &rdev->flags) &&
2980 !test_bit(ExternalBbl, &rdev->flags) &&
2981 rdev->badblocks.unacked_exist) {
2982 /* metadata handler doesn't understand badblocks,
2983 * so we need to fail the device
2984 */
2985 md_error(rdev->mddev, rdev);
2986 }
2987 clear_bit(Blocked, &rdev->flags);
2988 clear_bit(BlockedBadBlocks, &rdev->flags);
2989 wake_up(&rdev->blocked_wait);
2990 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2991
2992 err = 0;
2993 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2994 set_bit(In_sync, &rdev->flags);
2995 err = 0;
2996 } else if (cmd_match(buf, "failfast")) {
2997 set_bit(FailFast, &rdev->flags);
2998 need_update_sb = true;
2999 err = 0;
3000 } else if (cmd_match(buf, "-failfast")) {
3001 clear_bit(FailFast, &rdev->flags);
3002 need_update_sb = true;
3003 err = 0;
3004 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
3005 !test_bit(Journal, &rdev->flags)) {
3006 if (rdev->mddev->pers == NULL) {
3007 clear_bit(In_sync, &rdev->flags);
3008 rdev->saved_raid_disk = rdev->raid_disk;
3009 rdev->raid_disk = -1;
3010 err = 0;
3011 }
3012 } else if (cmd_match(buf, "write_error")) {
3013 set_bit(WriteErrorSeen, &rdev->flags);
3014 err = 0;
3015 } else if (cmd_match(buf, "-write_error")) {
3016 clear_bit(WriteErrorSeen, &rdev->flags);
3017 err = 0;
3018 } else if (cmd_match(buf, "want_replacement")) {
3019 /* Any non-spare device that is not a replacement can
3020 * become want_replacement at any time, but we then need to
3021 * check if recovery is needed.
3022 */
3023 if (rdev->raid_disk >= 0 &&
3024 !test_bit(Journal, &rdev->flags) &&
3025 !test_bit(Replacement, &rdev->flags))
3026 set_bit(WantReplacement, &rdev->flags);
3027 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3028 err = 0;
3029 } else if (cmd_match(buf, "-want_replacement")) {
3030 /* Clearing 'want_replacement' is always allowed.
3031 * Once replacements starts it is too late though.
3032 */
3033 err = 0;
3034 clear_bit(WantReplacement, &rdev->flags);
3035 } else if (cmd_match(buf, "replacement")) {
3036 /* Can only set a device as a replacement when array has not
3037 * yet been started. Once running, replacement is automatic
3038 * from spares, or by assigning 'slot'.
3039 */
3040 if (rdev->mddev->pers)
3041 err = -EBUSY;
3042 else {
3043 set_bit(Replacement, &rdev->flags);
3044 err = 0;
3045 }
3046 } else if (cmd_match(buf, "-replacement")) {
3047 /* Similarly, can only clear Replacement before start */
3048 if (rdev->mddev->pers)
3049 err = -EBUSY;
3050 else {
3051 clear_bit(Replacement, &rdev->flags);
3052 err = 0;
3053 }
3054 } else if (cmd_match(buf, "re-add")) {
3055 if (!rdev->mddev->pers)
3056 err = -EINVAL;
3057 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3058 rdev->saved_raid_disk >= 0) {
3059 /* clear_bit is performed _after_ all the devices
3060 * have their local Faulty bit cleared. If any writes
3061 * happen in the meantime in the local node, they
3062 * will land in the local bitmap, which will be synced
3063 * by this node eventually
3064 */
3065 if (!mddev_is_clustered(rdev->mddev) ||
3066 (err = mddev->cluster_ops->gather_bitmaps(rdev)) == 0) {
3067 clear_bit(Faulty, &rdev->flags);
3068 err = add_bound_rdev(rdev);
3069 }
3070 } else
3071 err = -EBUSY;
3072 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3073 set_bit(ExternalBbl, &rdev->flags);
3074 rdev->badblocks.shift = 0;
3075 err = 0;
3076 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3077 clear_bit(ExternalBbl, &rdev->flags);
3078 err = 0;
3079 }
3080 if (need_update_sb)
3081 md_update_sb(mddev, 1);
3082 if (!err)
3083 sysfs_notify_dirent_safe(rdev->sysfs_state);
3084 return err ? err : len;
3085 }
3086 static struct rdev_sysfs_entry rdev_state =
3087 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3088
3089 static ssize_t
errors_show(struct md_rdev * rdev,char * page)3090 errors_show(struct md_rdev *rdev, char *page)
3091 {
3092 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3093 }
3094
3095 static ssize_t
errors_store(struct md_rdev * rdev,const char * buf,size_t len)3096 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3097 {
3098 unsigned int n;
3099 int rv;
3100
3101 rv = kstrtouint(buf, 10, &n);
3102 if (rv < 0)
3103 return rv;
3104 atomic_set(&rdev->corrected_errors, n);
3105 return len;
3106 }
3107 static struct rdev_sysfs_entry rdev_errors =
3108 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3109
3110 static ssize_t
slot_show(struct md_rdev * rdev,char * page)3111 slot_show(struct md_rdev *rdev, char *page)
3112 {
3113 if (test_bit(Journal, &rdev->flags))
3114 return sprintf(page, "journal\n");
3115 else if (rdev->raid_disk < 0)
3116 return sprintf(page, "none\n");
3117 else
3118 return sprintf(page, "%d\n", rdev->raid_disk);
3119 }
3120
3121 static ssize_t
slot_store(struct md_rdev * rdev,const char * buf,size_t len)3122 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3123 {
3124 int slot;
3125 int err;
3126
3127 if (test_bit(Journal, &rdev->flags))
3128 return -EBUSY;
3129 if (strncmp(buf, "none", 4)==0)
3130 slot = -1;
3131 else {
3132 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3133 if (err < 0)
3134 return err;
3135 if (slot < 0)
3136 /* overflow */
3137 return -ENOSPC;
3138 }
3139 if (rdev->mddev->pers && slot == -1) {
3140 /* Setting 'slot' on an active array requires also
3141 * updating the 'rd%d' link, and communicating
3142 * with the personality with ->hot_*_disk.
3143 * For now we only support removing
3144 * failed/spare devices. This normally happens automatically,
3145 * but not when the metadata is externally managed.
3146 */
3147 if (rdev->raid_disk == -1)
3148 return -EEXIST;
3149 /* personality does all needed checks */
3150 if (rdev->mddev->pers->hot_remove_disk == NULL)
3151 return -EINVAL;
3152 clear_bit(Blocked, &rdev->flags);
3153 remove_and_add_spares(rdev->mddev, rdev);
3154 if (rdev->raid_disk >= 0)
3155 return -EBUSY;
3156 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3157 } else if (rdev->mddev->pers) {
3158 /* Activating a spare .. or possibly reactivating
3159 * if we ever get bitmaps working here.
3160 */
3161 int err;
3162
3163 if (rdev->raid_disk != -1)
3164 return -EBUSY;
3165
3166 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3167 return -EBUSY;
3168
3169 if (rdev->mddev->pers->hot_add_disk == NULL)
3170 return -EINVAL;
3171
3172 if (slot >= rdev->mddev->raid_disks &&
3173 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3174 return -ENOSPC;
3175
3176 rdev->raid_disk = slot;
3177 if (test_bit(In_sync, &rdev->flags))
3178 rdev->saved_raid_disk = slot;
3179 else
3180 rdev->saved_raid_disk = -1;
3181 clear_bit(In_sync, &rdev->flags);
3182 clear_bit(Bitmap_sync, &rdev->flags);
3183 err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3184 if (err) {
3185 rdev->raid_disk = -1;
3186 return err;
3187 } else
3188 sysfs_notify_dirent_safe(rdev->sysfs_state);
3189 /* failure here is OK */;
3190 sysfs_link_rdev(rdev->mddev, rdev);
3191 /* don't wakeup anyone, leave that to userspace. */
3192 } else {
3193 if (slot >= rdev->mddev->raid_disks &&
3194 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3195 return -ENOSPC;
3196 rdev->raid_disk = slot;
3197 /* assume it is working */
3198 clear_bit(Faulty, &rdev->flags);
3199 clear_bit(WriteMostly, &rdev->flags);
3200 set_bit(In_sync, &rdev->flags);
3201 sysfs_notify_dirent_safe(rdev->sysfs_state);
3202 }
3203 return len;
3204 }
3205
3206 static struct rdev_sysfs_entry rdev_slot =
3207 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3208
3209 static ssize_t
offset_show(struct md_rdev * rdev,char * page)3210 offset_show(struct md_rdev *rdev, char *page)
3211 {
3212 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3213 }
3214
3215 static ssize_t
offset_store(struct md_rdev * rdev,const char * buf,size_t len)3216 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3217 {
3218 unsigned long long offset;
3219 if (kstrtoull(buf, 10, &offset) < 0)
3220 return -EINVAL;
3221 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3222 return -EBUSY;
3223 if (rdev->sectors && rdev->mddev->external)
3224 /* Must set offset before size, so overlap checks
3225 * can be sane */
3226 return -EBUSY;
3227 rdev->data_offset = offset;
3228 rdev->new_data_offset = offset;
3229 return len;
3230 }
3231
3232 static struct rdev_sysfs_entry rdev_offset =
3233 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3234
new_offset_show(struct md_rdev * rdev,char * page)3235 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3236 {
3237 return sprintf(page, "%llu\n",
3238 (unsigned long long)rdev->new_data_offset);
3239 }
3240
new_offset_store(struct md_rdev * rdev,const char * buf,size_t len)3241 static ssize_t new_offset_store(struct md_rdev *rdev,
3242 const char *buf, size_t len)
3243 {
3244 unsigned long long new_offset;
3245 struct mddev *mddev = rdev->mddev;
3246
3247 if (kstrtoull(buf, 10, &new_offset) < 0)
3248 return -EINVAL;
3249
3250 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3251 return -EBUSY;
3252 if (new_offset == rdev->data_offset)
3253 /* reset is always permitted */
3254 ;
3255 else if (new_offset > rdev->data_offset) {
3256 /* must not push array size beyond rdev_sectors */
3257 if (new_offset - rdev->data_offset
3258 + mddev->dev_sectors > rdev->sectors)
3259 return -E2BIG;
3260 }
3261 /* Metadata worries about other space details. */
3262
3263 /* decreasing the offset is inconsistent with a backwards
3264 * reshape.
3265 */
3266 if (new_offset < rdev->data_offset &&
3267 mddev->reshape_backwards)
3268 return -EINVAL;
3269 /* Increasing offset is inconsistent with forwards
3270 * reshape. reshape_direction should be set to
3271 * 'backwards' first.
3272 */
3273 if (new_offset > rdev->data_offset &&
3274 !mddev->reshape_backwards)
3275 return -EINVAL;
3276
3277 if (mddev->pers && mddev->persistent &&
3278 !super_types[mddev->major_version]
3279 .allow_new_offset(rdev, new_offset))
3280 return -E2BIG;
3281 rdev->new_data_offset = new_offset;
3282 if (new_offset > rdev->data_offset)
3283 mddev->reshape_backwards = 1;
3284 else if (new_offset < rdev->data_offset)
3285 mddev->reshape_backwards = 0;
3286
3287 return len;
3288 }
3289 static struct rdev_sysfs_entry rdev_new_offset =
3290 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3291
3292 static ssize_t
rdev_size_show(struct md_rdev * rdev,char * page)3293 rdev_size_show(struct md_rdev *rdev, char *page)
3294 {
3295 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3296 }
3297
md_rdevs_overlap(struct md_rdev * a,struct md_rdev * b)3298 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3299 {
3300 /* check if two start/length pairs overlap */
3301 if (a->data_offset + a->sectors <= b->data_offset)
3302 return false;
3303 if (b->data_offset + b->sectors <= a->data_offset)
3304 return false;
3305 return true;
3306 }
3307
md_rdev_overlaps(struct md_rdev * rdev)3308 static bool md_rdev_overlaps(struct md_rdev *rdev)
3309 {
3310 struct mddev *mddev;
3311 struct md_rdev *rdev2;
3312
3313 spin_lock(&all_mddevs_lock);
3314 list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3315 if (test_bit(MD_DELETED, &mddev->flags))
3316 continue;
3317 rdev_for_each(rdev2, mddev) {
3318 if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3319 md_rdevs_overlap(rdev, rdev2)) {
3320 spin_unlock(&all_mddevs_lock);
3321 return true;
3322 }
3323 }
3324 }
3325 spin_unlock(&all_mddevs_lock);
3326 return false;
3327 }
3328
strict_blocks_to_sectors(const char * buf,sector_t * sectors)3329 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3330 {
3331 unsigned long long blocks;
3332 sector_t new;
3333
3334 if (kstrtoull(buf, 10, &blocks) < 0)
3335 return -EINVAL;
3336
3337 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3338 return -EINVAL; /* sector conversion overflow */
3339
3340 new = blocks * 2;
3341 if (new != blocks * 2)
3342 return -EINVAL; /* unsigned long long to sector_t overflow */
3343
3344 *sectors = new;
3345 return 0;
3346 }
3347
3348 static ssize_t
rdev_size_store(struct md_rdev * rdev,const char * buf,size_t len)3349 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3350 {
3351 struct mddev *my_mddev = rdev->mddev;
3352 sector_t oldsectors = rdev->sectors;
3353 sector_t sectors;
3354
3355 if (test_bit(Journal, &rdev->flags))
3356 return -EBUSY;
3357 if (strict_blocks_to_sectors(buf, §ors) < 0)
3358 return -EINVAL;
3359 if (rdev->data_offset != rdev->new_data_offset)
3360 return -EINVAL; /* too confusing */
3361 if (my_mddev->pers && rdev->raid_disk >= 0) {
3362 if (my_mddev->persistent) {
3363 sectors = super_types[my_mddev->major_version].
3364 rdev_size_change(rdev, sectors);
3365 if (!sectors)
3366 return -EBUSY;
3367 } else if (!sectors)
3368 sectors = bdev_nr_sectors(rdev->bdev) -
3369 rdev->data_offset;
3370 if (!my_mddev->pers->resize)
3371 /* Cannot change size for RAID0 or Linear etc */
3372 return -EINVAL;
3373 }
3374 if (sectors < my_mddev->dev_sectors)
3375 return -EINVAL; /* component must fit device */
3376
3377 rdev->sectors = sectors;
3378
3379 /*
3380 * Check that all other rdevs with the same bdev do not overlap. This
3381 * check does not provide a hard guarantee, it just helps avoid
3382 * dangerous mistakes.
3383 */
3384 if (sectors > oldsectors && my_mddev->external &&
3385 md_rdev_overlaps(rdev)) {
3386 /*
3387 * Someone else could have slipped in a size change here, but
3388 * doing so is just silly. We put oldsectors back because we
3389 * know it is safe, and trust userspace not to race with itself.
3390 */
3391 rdev->sectors = oldsectors;
3392 return -EBUSY;
3393 }
3394 return len;
3395 }
3396
3397 static struct rdev_sysfs_entry rdev_size =
3398 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3399
recovery_start_show(struct md_rdev * rdev,char * page)3400 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3401 {
3402 unsigned long long recovery_start = rdev->recovery_offset;
3403
3404 if (test_bit(In_sync, &rdev->flags) ||
3405 recovery_start == MaxSector)
3406 return sprintf(page, "none\n");
3407
3408 return sprintf(page, "%llu\n", recovery_start);
3409 }
3410
recovery_start_store(struct md_rdev * rdev,const char * buf,size_t len)3411 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3412 {
3413 unsigned long long recovery_start;
3414
3415 if (cmd_match(buf, "none"))
3416 recovery_start = MaxSector;
3417 else if (kstrtoull(buf, 10, &recovery_start))
3418 return -EINVAL;
3419
3420 if (rdev->mddev->pers &&
3421 rdev->raid_disk >= 0)
3422 return -EBUSY;
3423
3424 rdev->recovery_offset = recovery_start;
3425 if (recovery_start == MaxSector)
3426 set_bit(In_sync, &rdev->flags);
3427 else
3428 clear_bit(In_sync, &rdev->flags);
3429 return len;
3430 }
3431
3432 static struct rdev_sysfs_entry rdev_recovery_start =
3433 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3434
3435 /* sysfs access to bad-blocks list.
3436 * We present two files.
3437 * 'bad-blocks' lists sector numbers and lengths of ranges that
3438 * are recorded as bad. The list is truncated to fit within
3439 * the one-page limit of sysfs.
3440 * Writing "sector length" to this file adds an acknowledged
3441 * bad block list.
3442 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3443 * been acknowledged. Writing to this file adds bad blocks
3444 * without acknowledging them. This is largely for testing.
3445 */
bb_show(struct md_rdev * rdev,char * page)3446 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3447 {
3448 return badblocks_show(&rdev->badblocks, page, 0);
3449 }
bb_store(struct md_rdev * rdev,const char * page,size_t len)3450 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3451 {
3452 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3453 /* Maybe that ack was all we needed */
3454 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3455 wake_up(&rdev->blocked_wait);
3456 return rv;
3457 }
3458 static struct rdev_sysfs_entry rdev_bad_blocks =
3459 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3460
ubb_show(struct md_rdev * rdev,char * page)3461 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3462 {
3463 return badblocks_show(&rdev->badblocks, page, 1);
3464 }
ubb_store(struct md_rdev * rdev,const char * page,size_t len)3465 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3466 {
3467 return badblocks_store(&rdev->badblocks, page, len, 1);
3468 }
3469 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3470 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3471
3472 static ssize_t
ppl_sector_show(struct md_rdev * rdev,char * page)3473 ppl_sector_show(struct md_rdev *rdev, char *page)
3474 {
3475 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3476 }
3477
3478 static ssize_t
ppl_sector_store(struct md_rdev * rdev,const char * buf,size_t len)3479 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3480 {
3481 unsigned long long sector;
3482
3483 if (kstrtoull(buf, 10, §or) < 0)
3484 return -EINVAL;
3485 if (sector != (sector_t)sector)
3486 return -EINVAL;
3487
3488 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3489 rdev->raid_disk >= 0)
3490 return -EBUSY;
3491
3492 if (rdev->mddev->persistent) {
3493 if (rdev->mddev->major_version == 0)
3494 return -EINVAL;
3495 if ((sector > rdev->sb_start &&
3496 sector - rdev->sb_start > S16_MAX) ||
3497 (sector < rdev->sb_start &&
3498 rdev->sb_start - sector > -S16_MIN))
3499 return -EINVAL;
3500 rdev->ppl.offset = sector - rdev->sb_start;
3501 } else if (!rdev->mddev->external) {
3502 return -EBUSY;
3503 }
3504 rdev->ppl.sector = sector;
3505 return len;
3506 }
3507
3508 static struct rdev_sysfs_entry rdev_ppl_sector =
3509 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3510
3511 static ssize_t
ppl_size_show(struct md_rdev * rdev,char * page)3512 ppl_size_show(struct md_rdev *rdev, char *page)
3513 {
3514 return sprintf(page, "%u\n", rdev->ppl.size);
3515 }
3516
3517 static ssize_t
ppl_size_store(struct md_rdev * rdev,const char * buf,size_t len)3518 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3519 {
3520 unsigned int size;
3521
3522 if (kstrtouint(buf, 10, &size) < 0)
3523 return -EINVAL;
3524
3525 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3526 rdev->raid_disk >= 0)
3527 return -EBUSY;
3528
3529 if (rdev->mddev->persistent) {
3530 if (rdev->mddev->major_version == 0)
3531 return -EINVAL;
3532 if (size > U16_MAX)
3533 return -EINVAL;
3534 } else if (!rdev->mddev->external) {
3535 return -EBUSY;
3536 }
3537 rdev->ppl.size = size;
3538 return len;
3539 }
3540
3541 static struct rdev_sysfs_entry rdev_ppl_size =
3542 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3543
3544 static struct attribute *rdev_default_attrs[] = {
3545 &rdev_state.attr,
3546 &rdev_errors.attr,
3547 &rdev_slot.attr,
3548 &rdev_offset.attr,
3549 &rdev_new_offset.attr,
3550 &rdev_size.attr,
3551 &rdev_recovery_start.attr,
3552 &rdev_bad_blocks.attr,
3553 &rdev_unack_bad_blocks.attr,
3554 &rdev_ppl_sector.attr,
3555 &rdev_ppl_size.attr,
3556 NULL,
3557 };
3558 ATTRIBUTE_GROUPS(rdev_default);
3559 static ssize_t
rdev_attr_show(struct kobject * kobj,struct attribute * attr,char * page)3560 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3561 {
3562 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3563 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3564
3565 if (!entry->show)
3566 return -EIO;
3567 if (!rdev->mddev)
3568 return -ENODEV;
3569 return entry->show(rdev, page);
3570 }
3571
3572 static ssize_t
rdev_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)3573 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3574 const char *page, size_t length)
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 struct kernfs_node *kn = NULL;
3579 bool suspend = false;
3580 ssize_t rv;
3581 struct mddev *mddev = READ_ONCE(rdev->mddev);
3582
3583 if (!entry->store)
3584 return -EIO;
3585 if (!capable(CAP_SYS_ADMIN))
3586 return -EACCES;
3587 if (!mddev)
3588 return -ENODEV;
3589
3590 if (entry->store == state_store) {
3591 if (cmd_match(page, "remove"))
3592 kn = sysfs_break_active_protection(kobj, attr);
3593 if (cmd_match(page, "remove") || cmd_match(page, "re-add") ||
3594 cmd_match(page, "writemostly") ||
3595 cmd_match(page, "-writemostly"))
3596 suspend = true;
3597 }
3598
3599 rv = suspend ? mddev_suspend_and_lock(mddev) : mddev_lock(mddev);
3600 if (!rv) {
3601 if (rdev->mddev == NULL)
3602 rv = -ENODEV;
3603 else
3604 rv = entry->store(rdev, page, length);
3605 suspend ? mddev_unlock_and_resume(mddev) : mddev_unlock(mddev);
3606 }
3607
3608 if (kn)
3609 sysfs_unbreak_active_protection(kn);
3610
3611 return rv;
3612 }
3613
rdev_free(struct kobject * ko)3614 static void rdev_free(struct kobject *ko)
3615 {
3616 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3617 kfree(rdev);
3618 }
3619 static const struct sysfs_ops rdev_sysfs_ops = {
3620 .show = rdev_attr_show,
3621 .store = rdev_attr_store,
3622 };
3623 static const struct kobj_type rdev_ktype = {
3624 .release = rdev_free,
3625 .sysfs_ops = &rdev_sysfs_ops,
3626 .default_groups = rdev_default_groups,
3627 };
3628
md_rdev_init(struct md_rdev * rdev)3629 int md_rdev_init(struct md_rdev *rdev)
3630 {
3631 rdev->desc_nr = -1;
3632 rdev->saved_raid_disk = -1;
3633 rdev->raid_disk = -1;
3634 rdev->flags = 0;
3635 rdev->data_offset = 0;
3636 rdev->new_data_offset = 0;
3637 rdev->sb_events = 0;
3638 rdev->last_read_error = 0;
3639 rdev->sb_loaded = 0;
3640 rdev->bb_page = NULL;
3641 atomic_set(&rdev->nr_pending, 0);
3642 atomic_set(&rdev->read_errors, 0);
3643 atomic_set(&rdev->corrected_errors, 0);
3644
3645 INIT_LIST_HEAD(&rdev->same_set);
3646 init_waitqueue_head(&rdev->blocked_wait);
3647
3648 /* Add space to store bad block list.
3649 * This reserves the space even on arrays where it cannot
3650 * be used - I wonder if that matters
3651 */
3652 return badblocks_init(&rdev->badblocks, 0);
3653 }
3654 EXPORT_SYMBOL_GPL(md_rdev_init);
3655
3656 /*
3657 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3658 *
3659 * mark the device faulty if:
3660 *
3661 * - the device is nonexistent (zero size)
3662 * - the device has no valid superblock
3663 *
3664 * a faulty rdev _never_ has rdev->sb set.
3665 */
md_import_device(dev_t newdev,int super_format,int super_minor)3666 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3667 {
3668 struct md_rdev *rdev;
3669 sector_t size;
3670 int err;
3671
3672 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3673 if (!rdev)
3674 return ERR_PTR(-ENOMEM);
3675
3676 err = md_rdev_init(rdev);
3677 if (err)
3678 goto out_free_rdev;
3679 err = alloc_disk_sb(rdev);
3680 if (err)
3681 goto out_clear_rdev;
3682
3683 rdev->bdev_file = bdev_file_open_by_dev(newdev,
3684 BLK_OPEN_READ | BLK_OPEN_WRITE,
3685 super_format == -2 ? &claim_rdev : rdev, NULL);
3686 if (IS_ERR(rdev->bdev_file)) {
3687 pr_warn("md: could not open device unknown-block(%u,%u).\n",
3688 MAJOR(newdev), MINOR(newdev));
3689 err = PTR_ERR(rdev->bdev_file);
3690 goto out_clear_rdev;
3691 }
3692 rdev->bdev = file_bdev(rdev->bdev_file);
3693
3694 kobject_init(&rdev->kobj, &rdev_ktype);
3695
3696 size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3697 if (!size) {
3698 pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3699 rdev->bdev);
3700 err = -EINVAL;
3701 goto out_blkdev_put;
3702 }
3703
3704 if (super_format >= 0) {
3705 err = super_types[super_format].
3706 load_super(rdev, NULL, super_minor);
3707 if (err == -EINVAL) {
3708 pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3709 rdev->bdev,
3710 super_format, super_minor);
3711 goto out_blkdev_put;
3712 }
3713 if (err < 0) {
3714 pr_warn("md: could not read %pg's sb, not importing!\n",
3715 rdev->bdev);
3716 goto out_blkdev_put;
3717 }
3718 }
3719
3720 return rdev;
3721
3722 out_blkdev_put:
3723 fput(rdev->bdev_file);
3724 out_clear_rdev:
3725 md_rdev_clear(rdev);
3726 out_free_rdev:
3727 kfree(rdev);
3728 return ERR_PTR(err);
3729 }
3730
3731 /*
3732 * Check a full RAID array for plausibility
3733 */
3734
analyze_sbs(struct mddev * mddev)3735 static int analyze_sbs(struct mddev *mddev)
3736 {
3737 int i;
3738 struct md_rdev *rdev, *freshest, *tmp;
3739
3740 freshest = NULL;
3741 rdev_for_each_safe(rdev, tmp, mddev)
3742 switch (super_types[mddev->major_version].
3743 load_super(rdev, freshest, mddev->minor_version)) {
3744 case 1:
3745 freshest = rdev;
3746 break;
3747 case 0:
3748 break;
3749 default:
3750 pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3751 rdev->bdev);
3752 md_kick_rdev_from_array(rdev);
3753 }
3754
3755 /* Cannot find a valid fresh disk */
3756 if (!freshest) {
3757 pr_warn("md: cannot find a valid disk\n");
3758 return -EINVAL;
3759 }
3760
3761 super_types[mddev->major_version].
3762 validate_super(mddev, NULL/*freshest*/, freshest);
3763
3764 i = 0;
3765 rdev_for_each_safe(rdev, tmp, mddev) {
3766 if (mddev->max_disks &&
3767 (rdev->desc_nr >= mddev->max_disks ||
3768 i > mddev->max_disks)) {
3769 pr_warn("md: %s: %pg: only %d devices permitted\n",
3770 mdname(mddev), rdev->bdev,
3771 mddev->max_disks);
3772 md_kick_rdev_from_array(rdev);
3773 continue;
3774 }
3775 if (rdev != freshest) {
3776 if (super_types[mddev->major_version].
3777 validate_super(mddev, freshest, rdev)) {
3778 pr_warn("md: kicking non-fresh %pg from array!\n",
3779 rdev->bdev);
3780 md_kick_rdev_from_array(rdev);
3781 continue;
3782 }
3783 }
3784 if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3785 !test_bit(Journal, &rdev->flags)) {
3786 rdev->raid_disk = -1;
3787 clear_bit(In_sync, &rdev->flags);
3788 }
3789 }
3790
3791 return 0;
3792 }
3793
3794 /* Read a fixed-point number.
3795 * Numbers in sysfs attributes should be in "standard" units where
3796 * possible, so time should be in seconds.
3797 * However we internally use a a much smaller unit such as
3798 * milliseconds or jiffies.
3799 * This function takes a decimal number with a possible fractional
3800 * component, and produces an integer which is the result of
3801 * multiplying that number by 10^'scale'.
3802 * all without any floating-point arithmetic.
3803 */
strict_strtoul_scaled(const char * cp,unsigned long * res,int scale)3804 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3805 {
3806 unsigned long result = 0;
3807 long decimals = -1;
3808 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3809 if (*cp == '.')
3810 decimals = 0;
3811 else if (decimals < scale) {
3812 unsigned int value;
3813 value = *cp - '0';
3814 result = result * 10 + value;
3815 if (decimals >= 0)
3816 decimals++;
3817 }
3818 cp++;
3819 }
3820 if (*cp == '\n')
3821 cp++;
3822 if (*cp)
3823 return -EINVAL;
3824 if (decimals < 0)
3825 decimals = 0;
3826 *res = result * int_pow(10, scale - decimals);
3827 return 0;
3828 }
3829
3830 static ssize_t
safe_delay_show(struct mddev * mddev,char * page)3831 safe_delay_show(struct mddev *mddev, char *page)
3832 {
3833 unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3834
3835 return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3836 }
3837 static ssize_t
safe_delay_store(struct mddev * mddev,const char * cbuf,size_t len)3838 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3839 {
3840 unsigned long msec;
3841
3842 if (mddev_is_clustered(mddev)) {
3843 pr_warn("md: Safemode is disabled for clustered mode\n");
3844 return -EINVAL;
3845 }
3846
3847 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3848 return -EINVAL;
3849 if (msec == 0)
3850 mddev->safemode_delay = 0;
3851 else {
3852 unsigned long old_delay = mddev->safemode_delay;
3853 unsigned long new_delay = (msec*HZ)/1000;
3854
3855 if (new_delay == 0)
3856 new_delay = 1;
3857 mddev->safemode_delay = new_delay;
3858 if (new_delay < old_delay || old_delay == 0)
3859 mod_timer(&mddev->safemode_timer, jiffies+1);
3860 }
3861 return len;
3862 }
3863 static struct md_sysfs_entry md_safe_delay =
3864 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3865
3866 static ssize_t
level_show(struct mddev * mddev,char * page)3867 level_show(struct mddev *mddev, char *page)
3868 {
3869 struct md_personality *p;
3870 int ret;
3871 spin_lock(&mddev->lock);
3872 p = mddev->pers;
3873 if (p)
3874 ret = sprintf(page, "%s\n", p->head.name);
3875 else if (mddev->clevel[0])
3876 ret = sprintf(page, "%s\n", mddev->clevel);
3877 else if (mddev->level != LEVEL_NONE)
3878 ret = sprintf(page, "%d\n", mddev->level);
3879 else
3880 ret = 0;
3881 spin_unlock(&mddev->lock);
3882 return ret;
3883 }
3884
3885 static ssize_t
level_store(struct mddev * mddev,const char * buf,size_t len)3886 level_store(struct mddev *mddev, const char *buf, size_t len)
3887 {
3888 char clevel[16];
3889 ssize_t rv;
3890 size_t slen = len;
3891 struct md_personality *pers, *oldpers;
3892 long level;
3893 void *priv, *oldpriv;
3894 struct md_rdev *rdev;
3895
3896 if (slen == 0 || slen >= sizeof(clevel))
3897 return -EINVAL;
3898
3899 rv = mddev_suspend_and_lock(mddev);
3900 if (rv)
3901 return rv;
3902
3903 if (mddev->pers == NULL) {
3904 memcpy(mddev->clevel, buf, slen);
3905 if (mddev->clevel[slen-1] == '\n')
3906 slen--;
3907 mddev->clevel[slen] = 0;
3908 mddev->level = LEVEL_NONE;
3909 rv = len;
3910 goto out_unlock;
3911 }
3912 rv = -EROFS;
3913 if (!md_is_rdwr(mddev))
3914 goto out_unlock;
3915
3916 /* request to change the personality. Need to ensure:
3917 * - array is not engaged in resync/recovery/reshape
3918 * - old personality can be suspended
3919 * - new personality will access other array.
3920 */
3921
3922 rv = -EBUSY;
3923 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3924 mddev->reshape_position != MaxSector ||
3925 mddev->sysfs_active)
3926 goto out_unlock;
3927
3928 rv = -EINVAL;
3929 if (!mddev->pers->quiesce) {
3930 pr_warn("md: %s: %s does not support online personality change\n",
3931 mdname(mddev), mddev->pers->head.name);
3932 goto out_unlock;
3933 }
3934
3935 /* Now find the new personality */
3936 memcpy(clevel, buf, slen);
3937 if (clevel[slen-1] == '\n')
3938 slen--;
3939 clevel[slen] = 0;
3940 if (kstrtol(clevel, 10, &level))
3941 level = LEVEL_NONE;
3942
3943 if (request_module("md-%s", clevel) != 0)
3944 request_module("md-level-%s", clevel);
3945 pers = get_pers(level, clevel);
3946 if (!pers) {
3947 rv = -EINVAL;
3948 goto out_unlock;
3949 }
3950
3951 if (pers == mddev->pers) {
3952 /* Nothing to do! */
3953 put_pers(pers);
3954 rv = len;
3955 goto out_unlock;
3956 }
3957 if (!pers->takeover) {
3958 put_pers(pers);
3959 pr_warn("md: %s: %s does not support personality takeover\n",
3960 mdname(mddev), clevel);
3961 rv = -EINVAL;
3962 goto out_unlock;
3963 }
3964
3965 rdev_for_each(rdev, mddev)
3966 rdev->new_raid_disk = rdev->raid_disk;
3967
3968 /* ->takeover must set new_* and/or delta_disks
3969 * if it succeeds, and may set them when it fails.
3970 */
3971 priv = pers->takeover(mddev);
3972 if (IS_ERR(priv)) {
3973 mddev->new_level = mddev->level;
3974 mddev->new_layout = mddev->layout;
3975 mddev->new_chunk_sectors = mddev->chunk_sectors;
3976 mddev->raid_disks -= mddev->delta_disks;
3977 mddev->delta_disks = 0;
3978 mddev->reshape_backwards = 0;
3979 put_pers(pers);
3980 pr_warn("md: %s: %s would not accept array\n",
3981 mdname(mddev), clevel);
3982 rv = PTR_ERR(priv);
3983 goto out_unlock;
3984 }
3985
3986 /* Looks like we have a winner */
3987 mddev_detach(mddev);
3988
3989 spin_lock(&mddev->lock);
3990 oldpers = mddev->pers;
3991 oldpriv = mddev->private;
3992 mddev->pers = pers;
3993 mddev->private = priv;
3994 strscpy(mddev->clevel, pers->head.name, sizeof(mddev->clevel));
3995 mddev->level = mddev->new_level;
3996 mddev->layout = mddev->new_layout;
3997 mddev->chunk_sectors = mddev->new_chunk_sectors;
3998 mddev->delta_disks = 0;
3999 mddev->reshape_backwards = 0;
4000 mddev->degraded = 0;
4001 spin_unlock(&mddev->lock);
4002
4003 if (oldpers->sync_request == NULL &&
4004 mddev->external) {
4005 /* We are converting from a no-redundancy array
4006 * to a redundancy array and metadata is managed
4007 * externally so we need to be sure that writes
4008 * won't block due to a need to transition
4009 * clean->dirty
4010 * until external management is started.
4011 */
4012 mddev->in_sync = 0;
4013 mddev->safemode_delay = 0;
4014 mddev->safemode = 0;
4015 }
4016
4017 oldpers->free(mddev, oldpriv);
4018
4019 if (oldpers->sync_request == NULL &&
4020 pers->sync_request != NULL) {
4021 /* need to add the md_redundancy_group */
4022 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4023 pr_warn("md: cannot register extra attributes for %s\n",
4024 mdname(mddev));
4025 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4026 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4027 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4028 }
4029 if (oldpers->sync_request != NULL &&
4030 pers->sync_request == NULL) {
4031 /* need to remove the md_redundancy_group */
4032 if (mddev->to_remove == NULL)
4033 mddev->to_remove = &md_redundancy_group;
4034 }
4035
4036 put_pers(oldpers);
4037
4038 rdev_for_each(rdev, mddev) {
4039 if (rdev->raid_disk < 0)
4040 continue;
4041 if (rdev->new_raid_disk >= mddev->raid_disks)
4042 rdev->new_raid_disk = -1;
4043 if (rdev->new_raid_disk == rdev->raid_disk)
4044 continue;
4045 sysfs_unlink_rdev(mddev, rdev);
4046 }
4047 rdev_for_each(rdev, mddev) {
4048 if (rdev->raid_disk < 0)
4049 continue;
4050 if (rdev->new_raid_disk == rdev->raid_disk)
4051 continue;
4052 rdev->raid_disk = rdev->new_raid_disk;
4053 if (rdev->raid_disk < 0)
4054 clear_bit(In_sync, &rdev->flags);
4055 else {
4056 if (sysfs_link_rdev(mddev, rdev))
4057 pr_warn("md: cannot register rd%d for %s after level change\n",
4058 rdev->raid_disk, mdname(mddev));
4059 }
4060 }
4061
4062 if (pers->sync_request == NULL) {
4063 /* this is now an array without redundancy, so
4064 * it must always be in_sync
4065 */
4066 mddev->in_sync = 1;
4067 del_timer_sync(&mddev->safemode_timer);
4068 }
4069 pers->run(mddev);
4070 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4071 if (!mddev->thread)
4072 md_update_sb(mddev, 1);
4073 sysfs_notify_dirent_safe(mddev->sysfs_level);
4074 md_new_event();
4075 rv = len;
4076 out_unlock:
4077 mddev_unlock_and_resume(mddev);
4078 return rv;
4079 }
4080
4081 static struct md_sysfs_entry md_level =
4082 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4083
4084 static ssize_t
new_level_show(struct mddev * mddev,char * page)4085 new_level_show(struct mddev *mddev, char *page)
4086 {
4087 return sprintf(page, "%d\n", mddev->new_level);
4088 }
4089
4090 static ssize_t
new_level_store(struct mddev * mddev,const char * buf,size_t len)4091 new_level_store(struct mddev *mddev, const char *buf, size_t len)
4092 {
4093 unsigned int n;
4094 int err;
4095
4096 err = kstrtouint(buf, 10, &n);
4097 if (err < 0)
4098 return err;
4099 err = mddev_lock(mddev);
4100 if (err)
4101 return err;
4102
4103 mddev->new_level = n;
4104 md_update_sb(mddev, 1);
4105
4106 mddev_unlock(mddev);
4107 return len;
4108 }
4109 static struct md_sysfs_entry md_new_level =
4110 __ATTR(new_level, 0664, new_level_show, new_level_store);
4111
4112 static ssize_t
layout_show(struct mddev * mddev,char * page)4113 layout_show(struct mddev *mddev, char *page)
4114 {
4115 /* just a number, not meaningful for all levels */
4116 if (mddev->reshape_position != MaxSector &&
4117 mddev->layout != mddev->new_layout)
4118 return sprintf(page, "%d (%d)\n",
4119 mddev->new_layout, mddev->layout);
4120 return sprintf(page, "%d\n", mddev->layout);
4121 }
4122
4123 static ssize_t
layout_store(struct mddev * mddev,const char * buf,size_t len)4124 layout_store(struct mddev *mddev, const char *buf, size_t len)
4125 {
4126 unsigned int n;
4127 int err;
4128
4129 err = kstrtouint(buf, 10, &n);
4130 if (err < 0)
4131 return err;
4132 err = mddev_lock(mddev);
4133 if (err)
4134 return err;
4135
4136 if (mddev->pers) {
4137 if (mddev->pers->check_reshape == NULL)
4138 err = -EBUSY;
4139 else if (!md_is_rdwr(mddev))
4140 err = -EROFS;
4141 else {
4142 mddev->new_layout = n;
4143 err = mddev->pers->check_reshape(mddev);
4144 if (err)
4145 mddev->new_layout = mddev->layout;
4146 }
4147 } else {
4148 mddev->new_layout = n;
4149 if (mddev->reshape_position == MaxSector)
4150 mddev->layout = n;
4151 }
4152 mddev_unlock(mddev);
4153 return err ?: len;
4154 }
4155 static struct md_sysfs_entry md_layout =
4156 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4157
4158 static ssize_t
raid_disks_show(struct mddev * mddev,char * page)4159 raid_disks_show(struct mddev *mddev, char *page)
4160 {
4161 if (mddev->raid_disks == 0)
4162 return 0;
4163 if (mddev->reshape_position != MaxSector &&
4164 mddev->delta_disks != 0)
4165 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4166 mddev->raid_disks - mddev->delta_disks);
4167 return sprintf(page, "%d\n", mddev->raid_disks);
4168 }
4169
4170 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4171
4172 static ssize_t
raid_disks_store(struct mddev * mddev,const char * buf,size_t len)4173 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4174 {
4175 unsigned int n;
4176 int err;
4177
4178 err = kstrtouint(buf, 10, &n);
4179 if (err < 0)
4180 return err;
4181
4182 err = mddev_lock(mddev);
4183 if (err)
4184 return err;
4185 if (mddev->pers)
4186 err = update_raid_disks(mddev, n);
4187 else if (mddev->reshape_position != MaxSector) {
4188 struct md_rdev *rdev;
4189 int olddisks = mddev->raid_disks - mddev->delta_disks;
4190
4191 err = -EINVAL;
4192 rdev_for_each(rdev, mddev) {
4193 if (olddisks < n &&
4194 rdev->data_offset < rdev->new_data_offset)
4195 goto out_unlock;
4196 if (olddisks > n &&
4197 rdev->data_offset > rdev->new_data_offset)
4198 goto out_unlock;
4199 }
4200 err = 0;
4201 mddev->delta_disks = n - olddisks;
4202 mddev->raid_disks = n;
4203 mddev->reshape_backwards = (mddev->delta_disks < 0);
4204 } else
4205 mddev->raid_disks = n;
4206 out_unlock:
4207 mddev_unlock(mddev);
4208 return err ? err : len;
4209 }
4210 static struct md_sysfs_entry md_raid_disks =
4211 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4212
4213 static ssize_t
uuid_show(struct mddev * mddev,char * page)4214 uuid_show(struct mddev *mddev, char *page)
4215 {
4216 return sprintf(page, "%pU\n", mddev->uuid);
4217 }
4218 static struct md_sysfs_entry md_uuid =
4219 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4220
4221 static ssize_t
chunk_size_show(struct mddev * mddev,char * page)4222 chunk_size_show(struct mddev *mddev, char *page)
4223 {
4224 if (mddev->reshape_position != MaxSector &&
4225 mddev->chunk_sectors != mddev->new_chunk_sectors)
4226 return sprintf(page, "%d (%d)\n",
4227 mddev->new_chunk_sectors << 9,
4228 mddev->chunk_sectors << 9);
4229 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4230 }
4231
4232 static ssize_t
chunk_size_store(struct mddev * mddev,const char * buf,size_t len)4233 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4234 {
4235 unsigned long n;
4236 int err;
4237
4238 err = kstrtoul(buf, 10, &n);
4239 if (err < 0)
4240 return err;
4241
4242 err = mddev_lock(mddev);
4243 if (err)
4244 return err;
4245 if (mddev->pers) {
4246 if (mddev->pers->check_reshape == NULL)
4247 err = -EBUSY;
4248 else if (!md_is_rdwr(mddev))
4249 err = -EROFS;
4250 else {
4251 mddev->new_chunk_sectors = n >> 9;
4252 err = mddev->pers->check_reshape(mddev);
4253 if (err)
4254 mddev->new_chunk_sectors = mddev->chunk_sectors;
4255 }
4256 } else {
4257 mddev->new_chunk_sectors = n >> 9;
4258 if (mddev->reshape_position == MaxSector)
4259 mddev->chunk_sectors = n >> 9;
4260 }
4261 mddev_unlock(mddev);
4262 return err ?: len;
4263 }
4264 static struct md_sysfs_entry md_chunk_size =
4265 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4266
4267 static ssize_t
resync_start_show(struct mddev * mddev,char * page)4268 resync_start_show(struct mddev *mddev, char *page)
4269 {
4270 if (mddev->recovery_cp == MaxSector)
4271 return sprintf(page, "none\n");
4272 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4273 }
4274
4275 static ssize_t
resync_start_store(struct mddev * mddev,const char * buf,size_t len)4276 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4277 {
4278 unsigned long long n;
4279 int err;
4280
4281 if (cmd_match(buf, "none"))
4282 n = MaxSector;
4283 else {
4284 err = kstrtoull(buf, 10, &n);
4285 if (err < 0)
4286 return err;
4287 if (n != (sector_t)n)
4288 return -EINVAL;
4289 }
4290
4291 err = mddev_lock(mddev);
4292 if (err)
4293 return err;
4294 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4295 err = -EBUSY;
4296
4297 if (!err) {
4298 mddev->recovery_cp = n;
4299 if (mddev->pers)
4300 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4301 }
4302 mddev_unlock(mddev);
4303 return err ?: len;
4304 }
4305 static struct md_sysfs_entry md_resync_start =
4306 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4307 resync_start_show, resync_start_store);
4308
4309 /*
4310 * The array state can be:
4311 *
4312 * clear
4313 * No devices, no size, no level
4314 * Equivalent to STOP_ARRAY ioctl
4315 * inactive
4316 * May have some settings, but array is not active
4317 * all IO results in error
4318 * When written, doesn't tear down array, but just stops it
4319 * suspended (not supported yet)
4320 * All IO requests will block. The array can be reconfigured.
4321 * Writing this, if accepted, will block until array is quiescent
4322 * readonly
4323 * no resync can happen. no superblocks get written.
4324 * write requests fail
4325 * read-auto
4326 * like readonly, but behaves like 'clean' on a write request.
4327 *
4328 * clean - no pending writes, but otherwise active.
4329 * When written to inactive array, starts without resync
4330 * If a write request arrives then
4331 * if metadata is known, mark 'dirty' and switch to 'active'.
4332 * if not known, block and switch to write-pending
4333 * If written to an active array that has pending writes, then fails.
4334 * active
4335 * fully active: IO and resync can be happening.
4336 * When written to inactive array, starts with resync
4337 *
4338 * write-pending
4339 * clean, but writes are blocked waiting for 'active' to be written.
4340 *
4341 * active-idle
4342 * like active, but no writes have been seen for a while (100msec).
4343 *
4344 * broken
4345 * Array is failed. It's useful because mounted-arrays aren't stopped
4346 * when array is failed, so this state will at least alert the user that
4347 * something is wrong.
4348 */
4349 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4350 write_pending, active_idle, broken, bad_word};
4351 static char *array_states[] = {
4352 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4353 "write-pending", "active-idle", "broken", NULL };
4354
match_word(const char * word,char ** list)4355 static int match_word(const char *word, char **list)
4356 {
4357 int n;
4358 for (n=0; list[n]; n++)
4359 if (cmd_match(word, list[n]))
4360 break;
4361 return n;
4362 }
4363
4364 static ssize_t
array_state_show(struct mddev * mddev,char * page)4365 array_state_show(struct mddev *mddev, char *page)
4366 {
4367 enum array_state st = inactive;
4368
4369 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4370 switch(mddev->ro) {
4371 case MD_RDONLY:
4372 st = readonly;
4373 break;
4374 case MD_AUTO_READ:
4375 st = read_auto;
4376 break;
4377 case MD_RDWR:
4378 spin_lock(&mddev->lock);
4379 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4380 st = write_pending;
4381 else if (mddev->in_sync)
4382 st = clean;
4383 else if (mddev->safemode)
4384 st = active_idle;
4385 else
4386 st = active;
4387 spin_unlock(&mddev->lock);
4388 }
4389
4390 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4391 st = broken;
4392 } else {
4393 if (list_empty(&mddev->disks) &&
4394 mddev->raid_disks == 0 &&
4395 mddev->dev_sectors == 0)
4396 st = clear;
4397 else
4398 st = inactive;
4399 }
4400 return sprintf(page, "%s\n", array_states[st]);
4401 }
4402
4403 static int do_md_stop(struct mddev *mddev, int ro);
4404 static int md_set_readonly(struct mddev *mddev);
4405 static int restart_array(struct mddev *mddev);
4406
4407 static ssize_t
array_state_store(struct mddev * mddev,const char * buf,size_t len)4408 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4409 {
4410 int err = 0;
4411 enum array_state st = match_word(buf, array_states);
4412
4413 /* No lock dependent actions */
4414 switch (st) {
4415 case suspended: /* not supported yet */
4416 case write_pending: /* cannot be set */
4417 case active_idle: /* cannot be set */
4418 case broken: /* cannot be set */
4419 case bad_word:
4420 return -EINVAL;
4421 case clear:
4422 case readonly:
4423 case inactive:
4424 case read_auto:
4425 if (!mddev->pers || !md_is_rdwr(mddev))
4426 break;
4427 /* write sysfs will not open mddev and opener should be 0 */
4428 err = mddev_set_closing_and_sync_blockdev(mddev, 0);
4429 if (err)
4430 return err;
4431 break;
4432 default:
4433 break;
4434 }
4435
4436 if (mddev->pers && (st == active || st == clean) &&
4437 mddev->ro != MD_RDONLY) {
4438 /* don't take reconfig_mutex when toggling between
4439 * clean and active
4440 */
4441 spin_lock(&mddev->lock);
4442 if (st == active) {
4443 restart_array(mddev);
4444 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4445 md_wakeup_thread(mddev->thread);
4446 wake_up(&mddev->sb_wait);
4447 } else /* st == clean */ {
4448 restart_array(mddev);
4449 if (!set_in_sync(mddev))
4450 err = -EBUSY;
4451 }
4452 if (!err)
4453 sysfs_notify_dirent_safe(mddev->sysfs_state);
4454 spin_unlock(&mddev->lock);
4455 return err ?: len;
4456 }
4457 err = mddev_lock(mddev);
4458 if (err)
4459 return err;
4460
4461 switch (st) {
4462 case inactive:
4463 /* stop an active array, return 0 otherwise */
4464 if (mddev->pers)
4465 err = do_md_stop(mddev, 2);
4466 break;
4467 case clear:
4468 err = do_md_stop(mddev, 0);
4469 break;
4470 case readonly:
4471 if (mddev->pers)
4472 err = md_set_readonly(mddev);
4473 else {
4474 mddev->ro = MD_RDONLY;
4475 set_disk_ro(mddev->gendisk, 1);
4476 err = do_md_run(mddev);
4477 }
4478 break;
4479 case read_auto:
4480 if (mddev->pers) {
4481 if (md_is_rdwr(mddev))
4482 err = md_set_readonly(mddev);
4483 else if (mddev->ro == MD_RDONLY)
4484 err = restart_array(mddev);
4485 if (err == 0) {
4486 mddev->ro = MD_AUTO_READ;
4487 set_disk_ro(mddev->gendisk, 0);
4488 }
4489 } else {
4490 mddev->ro = MD_AUTO_READ;
4491 err = do_md_run(mddev);
4492 }
4493 break;
4494 case clean:
4495 if (mddev->pers) {
4496 err = restart_array(mddev);
4497 if (err)
4498 break;
4499 spin_lock(&mddev->lock);
4500 if (!set_in_sync(mddev))
4501 err = -EBUSY;
4502 spin_unlock(&mddev->lock);
4503 } else
4504 err = -EINVAL;
4505 break;
4506 case active:
4507 if (mddev->pers) {
4508 err = restart_array(mddev);
4509 if (err)
4510 break;
4511 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4512 wake_up(&mddev->sb_wait);
4513 err = 0;
4514 } else {
4515 mddev->ro = MD_RDWR;
4516 set_disk_ro(mddev->gendisk, 0);
4517 err = do_md_run(mddev);
4518 }
4519 break;
4520 default:
4521 err = -EINVAL;
4522 break;
4523 }
4524
4525 if (!err) {
4526 if (mddev->hold_active == UNTIL_IOCTL)
4527 mddev->hold_active = 0;
4528 sysfs_notify_dirent_safe(mddev->sysfs_state);
4529 }
4530 mddev_unlock(mddev);
4531
4532 if (st == readonly || st == read_auto || st == inactive ||
4533 (err && st == clear))
4534 clear_bit(MD_CLOSING, &mddev->flags);
4535
4536 return err ?: len;
4537 }
4538 static struct md_sysfs_entry md_array_state =
4539 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4540
4541 static ssize_t
max_corrected_read_errors_show(struct mddev * mddev,char * page)4542 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4543 return sprintf(page, "%d\n",
4544 atomic_read(&mddev->max_corr_read_errors));
4545 }
4546
4547 static ssize_t
max_corrected_read_errors_store(struct mddev * mddev,const char * buf,size_t len)4548 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4549 {
4550 unsigned int n;
4551 int rv;
4552
4553 rv = kstrtouint(buf, 10, &n);
4554 if (rv < 0)
4555 return rv;
4556 if (n > INT_MAX)
4557 return -EINVAL;
4558 atomic_set(&mddev->max_corr_read_errors, n);
4559 return len;
4560 }
4561
4562 static struct md_sysfs_entry max_corr_read_errors =
4563 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4564 max_corrected_read_errors_store);
4565
4566 static ssize_t
null_show(struct mddev * mddev,char * page)4567 null_show(struct mddev *mddev, char *page)
4568 {
4569 return -EINVAL;
4570 }
4571
4572 static ssize_t
new_dev_store(struct mddev * mddev,const char * buf,size_t len)4573 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4574 {
4575 /* buf must be %d:%d\n? giving major and minor numbers */
4576 /* The new device is added to the array.
4577 * If the array has a persistent superblock, we read the
4578 * superblock to initialise info and check validity.
4579 * Otherwise, only checking done is that in bind_rdev_to_array,
4580 * which mainly checks size.
4581 */
4582 char *e;
4583 int major = simple_strtoul(buf, &e, 10);
4584 int minor;
4585 dev_t dev;
4586 struct md_rdev *rdev;
4587 int err;
4588
4589 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4590 return -EINVAL;
4591 minor = simple_strtoul(e+1, &e, 10);
4592 if (*e && *e != '\n')
4593 return -EINVAL;
4594 dev = MKDEV(major, minor);
4595 if (major != MAJOR(dev) ||
4596 minor != MINOR(dev))
4597 return -EOVERFLOW;
4598
4599 err = mddev_suspend_and_lock(mddev);
4600 if (err)
4601 return err;
4602 if (mddev->persistent) {
4603 rdev = md_import_device(dev, mddev->major_version,
4604 mddev->minor_version);
4605 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4606 struct md_rdev *rdev0
4607 = list_entry(mddev->disks.next,
4608 struct md_rdev, same_set);
4609 err = super_types[mddev->major_version]
4610 .load_super(rdev, rdev0, mddev->minor_version);
4611 if (err < 0)
4612 goto out;
4613 }
4614 } else if (mddev->external)
4615 rdev = md_import_device(dev, -2, -1);
4616 else
4617 rdev = md_import_device(dev, -1, -1);
4618
4619 if (IS_ERR(rdev)) {
4620 mddev_unlock_and_resume(mddev);
4621 return PTR_ERR(rdev);
4622 }
4623 err = bind_rdev_to_array(rdev, mddev);
4624 out:
4625 if (err)
4626 export_rdev(rdev, mddev);
4627 mddev_unlock_and_resume(mddev);
4628 if (!err)
4629 md_new_event();
4630 return err ? err : len;
4631 }
4632
4633 static struct md_sysfs_entry md_new_device =
4634 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4635
4636 static ssize_t
bitmap_store(struct mddev * mddev,const char * buf,size_t len)4637 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4638 {
4639 char *end;
4640 unsigned long chunk, end_chunk;
4641 int err;
4642
4643 err = mddev_lock(mddev);
4644 if (err)
4645 return err;
4646 if (!mddev->bitmap)
4647 goto out;
4648 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4649 while (*buf) {
4650 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4651 if (buf == end)
4652 break;
4653
4654 if (*end == '-') { /* range */
4655 buf = end + 1;
4656 end_chunk = simple_strtoul(buf, &end, 0);
4657 if (buf == end)
4658 break;
4659 }
4660
4661 if (*end && !isspace(*end))
4662 break;
4663
4664 mddev->bitmap_ops->dirty_bits(mddev, chunk, end_chunk);
4665 buf = skip_spaces(end);
4666 }
4667 mddev->bitmap_ops->unplug(mddev, true); /* flush the bits to disk */
4668 out:
4669 mddev_unlock(mddev);
4670 return len;
4671 }
4672
4673 static struct md_sysfs_entry md_bitmap =
4674 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4675
4676 static ssize_t
size_show(struct mddev * mddev,char * page)4677 size_show(struct mddev *mddev, char *page)
4678 {
4679 return sprintf(page, "%llu\n",
4680 (unsigned long long)mddev->dev_sectors / 2);
4681 }
4682
4683 static int update_size(struct mddev *mddev, sector_t num_sectors);
4684
4685 static ssize_t
size_store(struct mddev * mddev,const char * buf,size_t len)4686 size_store(struct mddev *mddev, const char *buf, size_t len)
4687 {
4688 /* If array is inactive, we can reduce the component size, but
4689 * not increase it (except from 0).
4690 * If array is active, we can try an on-line resize
4691 */
4692 sector_t sectors;
4693 int err = strict_blocks_to_sectors(buf, §ors);
4694
4695 if (err < 0)
4696 return err;
4697 err = mddev_lock(mddev);
4698 if (err)
4699 return err;
4700 if (mddev->pers) {
4701 err = update_size(mddev, sectors);
4702 if (err == 0)
4703 md_update_sb(mddev, 1);
4704 } else {
4705 if (mddev->dev_sectors == 0 ||
4706 mddev->dev_sectors > sectors)
4707 mddev->dev_sectors = sectors;
4708 else
4709 err = -ENOSPC;
4710 }
4711 mddev_unlock(mddev);
4712 return err ? err : len;
4713 }
4714
4715 static struct md_sysfs_entry md_size =
4716 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4717
4718 /* Metadata version.
4719 * This is one of
4720 * 'none' for arrays with no metadata (good luck...)
4721 * 'external' for arrays with externally managed metadata,
4722 * or N.M for internally known formats
4723 */
4724 static ssize_t
metadata_show(struct mddev * mddev,char * page)4725 metadata_show(struct mddev *mddev, char *page)
4726 {
4727 if (mddev->persistent)
4728 return sprintf(page, "%d.%d\n",
4729 mddev->major_version, mddev->minor_version);
4730 else if (mddev->external)
4731 return sprintf(page, "external:%s\n", mddev->metadata_type);
4732 else
4733 return sprintf(page, "none\n");
4734 }
4735
4736 static ssize_t
metadata_store(struct mddev * mddev,const char * buf,size_t len)4737 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4738 {
4739 int major, minor;
4740 char *e;
4741 int err;
4742 /* Changing the details of 'external' metadata is
4743 * always permitted. Otherwise there must be
4744 * no devices attached to the array.
4745 */
4746
4747 err = mddev_lock(mddev);
4748 if (err)
4749 return err;
4750 err = -EBUSY;
4751 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4752 ;
4753 else if (!list_empty(&mddev->disks))
4754 goto out_unlock;
4755
4756 err = 0;
4757 if (cmd_match(buf, "none")) {
4758 mddev->persistent = 0;
4759 mddev->external = 0;
4760 mddev->major_version = 0;
4761 mddev->minor_version = 90;
4762 goto out_unlock;
4763 }
4764 if (strncmp(buf, "external:", 9) == 0) {
4765 size_t namelen = len-9;
4766 if (namelen >= sizeof(mddev->metadata_type))
4767 namelen = sizeof(mddev->metadata_type)-1;
4768 memcpy(mddev->metadata_type, buf+9, namelen);
4769 mddev->metadata_type[namelen] = 0;
4770 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4771 mddev->metadata_type[--namelen] = 0;
4772 mddev->persistent = 0;
4773 mddev->external = 1;
4774 mddev->major_version = 0;
4775 mddev->minor_version = 90;
4776 goto out_unlock;
4777 }
4778 major = simple_strtoul(buf, &e, 10);
4779 err = -EINVAL;
4780 if (e==buf || *e != '.')
4781 goto out_unlock;
4782 buf = e+1;
4783 minor = simple_strtoul(buf, &e, 10);
4784 if (e==buf || (*e && *e != '\n') )
4785 goto out_unlock;
4786 err = -ENOENT;
4787 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4788 goto out_unlock;
4789 mddev->major_version = major;
4790 mddev->minor_version = minor;
4791 mddev->persistent = 1;
4792 mddev->external = 0;
4793 err = 0;
4794 out_unlock:
4795 mddev_unlock(mddev);
4796 return err ?: len;
4797 }
4798
4799 static struct md_sysfs_entry md_metadata =
4800 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4801
md_sync_action(struct mddev * mddev)4802 enum sync_action md_sync_action(struct mddev *mddev)
4803 {
4804 unsigned long recovery = mddev->recovery;
4805
4806 /*
4807 * frozen has the highest priority, means running sync_thread will be
4808 * stopped immediately, and no new sync_thread can start.
4809 */
4810 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4811 return ACTION_FROZEN;
4812
4813 /*
4814 * read-only array can't register sync_thread, and it can only
4815 * add/remove spares.
4816 */
4817 if (!md_is_rdwr(mddev))
4818 return ACTION_IDLE;
4819
4820 /*
4821 * idle means no sync_thread is running, and no new sync_thread is
4822 * requested.
4823 */
4824 if (!test_bit(MD_RECOVERY_RUNNING, &recovery) &&
4825 !test_bit(MD_RECOVERY_NEEDED, &recovery))
4826 return ACTION_IDLE;
4827
4828 if (test_bit(MD_RECOVERY_RESHAPE, &recovery) ||
4829 mddev->reshape_position != MaxSector)
4830 return ACTION_RESHAPE;
4831
4832 if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4833 return ACTION_RECOVER;
4834
4835 if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4836 /*
4837 * MD_RECOVERY_CHECK must be paired with
4838 * MD_RECOVERY_REQUESTED.
4839 */
4840 if (test_bit(MD_RECOVERY_CHECK, &recovery))
4841 return ACTION_CHECK;
4842 if (test_bit(MD_RECOVERY_REQUESTED, &recovery))
4843 return ACTION_REPAIR;
4844 return ACTION_RESYNC;
4845 }
4846
4847 /*
4848 * MD_RECOVERY_NEEDED or MD_RECOVERY_RUNNING is set, however, no
4849 * sync_action is specified.
4850 */
4851 return ACTION_IDLE;
4852 }
4853
md_sync_action_by_name(const char * page)4854 enum sync_action md_sync_action_by_name(const char *page)
4855 {
4856 enum sync_action action;
4857
4858 for (action = 0; action < NR_SYNC_ACTIONS; ++action) {
4859 if (cmd_match(page, action_name[action]))
4860 return action;
4861 }
4862
4863 return NR_SYNC_ACTIONS;
4864 }
4865
md_sync_action_name(enum sync_action action)4866 const char *md_sync_action_name(enum sync_action action)
4867 {
4868 return action_name[action];
4869 }
4870
4871 static ssize_t
action_show(struct mddev * mddev,char * page)4872 action_show(struct mddev *mddev, char *page)
4873 {
4874 enum sync_action action = md_sync_action(mddev);
4875
4876 return sprintf(page, "%s\n", md_sync_action_name(action));
4877 }
4878
4879 /**
4880 * stop_sync_thread() - wait for sync_thread to stop if it's running.
4881 * @mddev: the array.
4882 * @locked: if set, reconfig_mutex will still be held after this function
4883 * return; if not set, reconfig_mutex will be released after this
4884 * function return.
4885 */
stop_sync_thread(struct mddev * mddev,bool locked)4886 static void stop_sync_thread(struct mddev *mddev, bool locked)
4887 {
4888 int sync_seq = atomic_read(&mddev->sync_seq);
4889
4890 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4891 if (!locked)
4892 mddev_unlock(mddev);
4893 return;
4894 }
4895
4896 mddev_unlock(mddev);
4897
4898 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4899 /*
4900 * Thread might be blocked waiting for metadata update which will now
4901 * never happen
4902 */
4903 md_wakeup_thread_directly(mddev->sync_thread);
4904 if (work_pending(&mddev->sync_work))
4905 flush_work(&mddev->sync_work);
4906
4907 wait_event(resync_wait,
4908 !test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4909 (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery) &&
4910 sync_seq != atomic_read(&mddev->sync_seq)));
4911
4912 if (locked)
4913 mddev_lock_nointr(mddev);
4914 }
4915
md_idle_sync_thread(struct mddev * mddev)4916 void md_idle_sync_thread(struct mddev *mddev)
4917 {
4918 lockdep_assert_held(&mddev->reconfig_mutex);
4919
4920 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4921 stop_sync_thread(mddev, true);
4922 }
4923 EXPORT_SYMBOL_GPL(md_idle_sync_thread);
4924
md_frozen_sync_thread(struct mddev * mddev)4925 void md_frozen_sync_thread(struct mddev *mddev)
4926 {
4927 lockdep_assert_held(&mddev->reconfig_mutex);
4928
4929 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4930 stop_sync_thread(mddev, true);
4931 }
4932 EXPORT_SYMBOL_GPL(md_frozen_sync_thread);
4933
md_unfrozen_sync_thread(struct mddev * mddev)4934 void md_unfrozen_sync_thread(struct mddev *mddev)
4935 {
4936 lockdep_assert_held(&mddev->reconfig_mutex);
4937
4938 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4939 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4940 md_wakeup_thread(mddev->thread);
4941 sysfs_notify_dirent_safe(mddev->sysfs_action);
4942 }
4943 EXPORT_SYMBOL_GPL(md_unfrozen_sync_thread);
4944
mddev_start_reshape(struct mddev * mddev)4945 static int mddev_start_reshape(struct mddev *mddev)
4946 {
4947 int ret;
4948
4949 if (mddev->pers->start_reshape == NULL)
4950 return -EINVAL;
4951
4952 if (mddev->reshape_position == MaxSector ||
4953 mddev->pers->check_reshape == NULL ||
4954 mddev->pers->check_reshape(mddev)) {
4955 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4956 ret = mddev->pers->start_reshape(mddev);
4957 if (ret)
4958 return ret;
4959 } else {
4960 /*
4961 * If reshape is still in progress, and md_check_recovery() can
4962 * continue to reshape, don't restart reshape because data can
4963 * be corrupted for raid456.
4964 */
4965 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4966 }
4967
4968 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4969 return 0;
4970 }
4971
4972 static ssize_t
action_store(struct mddev * mddev,const char * page,size_t len)4973 action_store(struct mddev *mddev, const char *page, size_t len)
4974 {
4975 int ret;
4976 enum sync_action action;
4977
4978 if (!mddev->pers || !mddev->pers->sync_request)
4979 return -EINVAL;
4980
4981 retry:
4982 if (work_busy(&mddev->sync_work))
4983 flush_work(&mddev->sync_work);
4984
4985 ret = mddev_lock(mddev);
4986 if (ret)
4987 return ret;
4988
4989 if (work_busy(&mddev->sync_work)) {
4990 mddev_unlock(mddev);
4991 goto retry;
4992 }
4993
4994 action = md_sync_action_by_name(page);
4995
4996 /* TODO: mdadm rely on "idle" to start sync_thread. */
4997 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4998 switch (action) {
4999 case ACTION_FROZEN:
5000 md_frozen_sync_thread(mddev);
5001 ret = len;
5002 goto out;
5003 case ACTION_IDLE:
5004 md_idle_sync_thread(mddev);
5005 break;
5006 case ACTION_RESHAPE:
5007 case ACTION_RECOVER:
5008 case ACTION_CHECK:
5009 case ACTION_REPAIR:
5010 case ACTION_RESYNC:
5011 ret = -EBUSY;
5012 goto out;
5013 default:
5014 ret = -EINVAL;
5015 goto out;
5016 }
5017 } else {
5018 switch (action) {
5019 case ACTION_FROZEN:
5020 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5021 ret = len;
5022 goto out;
5023 case ACTION_RESHAPE:
5024 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5025 ret = mddev_start_reshape(mddev);
5026 if (ret)
5027 goto out;
5028 break;
5029 case ACTION_RECOVER:
5030 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5031 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5032 break;
5033 case ACTION_CHECK:
5034 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5035 fallthrough;
5036 case ACTION_REPAIR:
5037 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
5038 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5039 fallthrough;
5040 case ACTION_RESYNC:
5041 case ACTION_IDLE:
5042 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5043 break;
5044 default:
5045 ret = -EINVAL;
5046 goto out;
5047 }
5048 }
5049
5050 if (mddev->ro == MD_AUTO_READ) {
5051 /* A write to sync_action is enough to justify
5052 * canceling read-auto mode
5053 */
5054 mddev->ro = MD_RDWR;
5055 md_wakeup_thread(mddev->sync_thread);
5056 }
5057
5058 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5059 md_wakeup_thread(mddev->thread);
5060 sysfs_notify_dirent_safe(mddev->sysfs_action);
5061 ret = len;
5062
5063 out:
5064 mddev_unlock(mddev);
5065 return ret;
5066 }
5067
5068 static struct md_sysfs_entry md_scan_mode =
5069 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
5070
5071 static ssize_t
last_sync_action_show(struct mddev * mddev,char * page)5072 last_sync_action_show(struct mddev *mddev, char *page)
5073 {
5074 return sprintf(page, "%s\n",
5075 md_sync_action_name(mddev->last_sync_action));
5076 }
5077
5078 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
5079
5080 static ssize_t
mismatch_cnt_show(struct mddev * mddev,char * page)5081 mismatch_cnt_show(struct mddev *mddev, char *page)
5082 {
5083 return sprintf(page, "%llu\n",
5084 (unsigned long long)
5085 atomic64_read(&mddev->resync_mismatches));
5086 }
5087
5088 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
5089
5090 static ssize_t
sync_min_show(struct mddev * mddev,char * page)5091 sync_min_show(struct mddev *mddev, char *page)
5092 {
5093 return sprintf(page, "%d (%s)\n", speed_min(mddev),
5094 mddev->sync_speed_min ? "local": "system");
5095 }
5096
5097 static ssize_t
sync_min_store(struct mddev * mddev,const char * buf,size_t len)5098 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
5099 {
5100 unsigned int min;
5101 int rv;
5102
5103 if (strncmp(buf, "system", 6)==0) {
5104 min = 0;
5105 } else {
5106 rv = kstrtouint(buf, 10, &min);
5107 if (rv < 0)
5108 return rv;
5109 if (min == 0)
5110 return -EINVAL;
5111 }
5112 mddev->sync_speed_min = min;
5113 return len;
5114 }
5115
5116 static struct md_sysfs_entry md_sync_min =
5117 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
5118
5119 static ssize_t
sync_max_show(struct mddev * mddev,char * page)5120 sync_max_show(struct mddev *mddev, char *page)
5121 {
5122 return sprintf(page, "%d (%s)\n", speed_max(mddev),
5123 mddev->sync_speed_max ? "local": "system");
5124 }
5125
5126 static ssize_t
sync_max_store(struct mddev * mddev,const char * buf,size_t len)5127 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
5128 {
5129 unsigned int max;
5130 int rv;
5131
5132 if (strncmp(buf, "system", 6)==0) {
5133 max = 0;
5134 } else {
5135 rv = kstrtouint(buf, 10, &max);
5136 if (rv < 0)
5137 return rv;
5138 if (max == 0)
5139 return -EINVAL;
5140 }
5141 mddev->sync_speed_max = max;
5142 return len;
5143 }
5144
5145 static struct md_sysfs_entry md_sync_max =
5146 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
5147
5148 static ssize_t
degraded_show(struct mddev * mddev,char * page)5149 degraded_show(struct mddev *mddev, char *page)
5150 {
5151 return sprintf(page, "%d\n", mddev->degraded);
5152 }
5153 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5154
5155 static ssize_t
sync_force_parallel_show(struct mddev * mddev,char * page)5156 sync_force_parallel_show(struct mddev *mddev, char *page)
5157 {
5158 return sprintf(page, "%d\n", mddev->parallel_resync);
5159 }
5160
5161 static ssize_t
sync_force_parallel_store(struct mddev * mddev,const char * buf,size_t len)5162 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5163 {
5164 long n;
5165
5166 if (kstrtol(buf, 10, &n))
5167 return -EINVAL;
5168
5169 if (n != 0 && n != 1)
5170 return -EINVAL;
5171
5172 mddev->parallel_resync = n;
5173
5174 if (mddev->sync_thread)
5175 wake_up(&resync_wait);
5176
5177 return len;
5178 }
5179
5180 /* force parallel resync, even with shared block devices */
5181 static struct md_sysfs_entry md_sync_force_parallel =
5182 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5183 sync_force_parallel_show, sync_force_parallel_store);
5184
5185 static ssize_t
sync_speed_show(struct mddev * mddev,char * page)5186 sync_speed_show(struct mddev *mddev, char *page)
5187 {
5188 unsigned long resync, dt, db;
5189 if (mddev->curr_resync == MD_RESYNC_NONE)
5190 return sprintf(page, "none\n");
5191 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5192 dt = (jiffies - mddev->resync_mark) / HZ;
5193 if (!dt) dt++;
5194 db = resync - mddev->resync_mark_cnt;
5195 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5196 }
5197
5198 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5199
5200 static ssize_t
sync_completed_show(struct mddev * mddev,char * page)5201 sync_completed_show(struct mddev *mddev, char *page)
5202 {
5203 unsigned long long max_sectors, resync;
5204
5205 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5206 return sprintf(page, "none\n");
5207
5208 if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5209 mddev->curr_resync == MD_RESYNC_DELAYED)
5210 return sprintf(page, "delayed\n");
5211
5212 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5213 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5214 max_sectors = mddev->resync_max_sectors;
5215 else
5216 max_sectors = mddev->dev_sectors;
5217
5218 resync = mddev->curr_resync_completed;
5219 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5220 }
5221
5222 static struct md_sysfs_entry md_sync_completed =
5223 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5224
5225 static ssize_t
min_sync_show(struct mddev * mddev,char * page)5226 min_sync_show(struct mddev *mddev, char *page)
5227 {
5228 return sprintf(page, "%llu\n",
5229 (unsigned long long)mddev->resync_min);
5230 }
5231 static ssize_t
min_sync_store(struct mddev * mddev,const char * buf,size_t len)5232 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5233 {
5234 unsigned long long min;
5235 int err;
5236
5237 if (kstrtoull(buf, 10, &min))
5238 return -EINVAL;
5239
5240 spin_lock(&mddev->lock);
5241 err = -EINVAL;
5242 if (min > mddev->resync_max)
5243 goto out_unlock;
5244
5245 err = -EBUSY;
5246 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5247 goto out_unlock;
5248
5249 /* Round down to multiple of 4K for safety */
5250 mddev->resync_min = round_down(min, 8);
5251 err = 0;
5252
5253 out_unlock:
5254 spin_unlock(&mddev->lock);
5255 return err ?: len;
5256 }
5257
5258 static struct md_sysfs_entry md_min_sync =
5259 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5260
5261 static ssize_t
max_sync_show(struct mddev * mddev,char * page)5262 max_sync_show(struct mddev *mddev, char *page)
5263 {
5264 if (mddev->resync_max == MaxSector)
5265 return sprintf(page, "max\n");
5266 else
5267 return sprintf(page, "%llu\n",
5268 (unsigned long long)mddev->resync_max);
5269 }
5270 static ssize_t
max_sync_store(struct mddev * mddev,const char * buf,size_t len)5271 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5272 {
5273 int err;
5274 spin_lock(&mddev->lock);
5275 if (strncmp(buf, "max", 3) == 0)
5276 mddev->resync_max = MaxSector;
5277 else {
5278 unsigned long long max;
5279 int chunk;
5280
5281 err = -EINVAL;
5282 if (kstrtoull(buf, 10, &max))
5283 goto out_unlock;
5284 if (max < mddev->resync_min)
5285 goto out_unlock;
5286
5287 err = -EBUSY;
5288 if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5289 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5290 goto out_unlock;
5291
5292 /* Must be a multiple of chunk_size */
5293 chunk = mddev->chunk_sectors;
5294 if (chunk) {
5295 sector_t temp = max;
5296
5297 err = -EINVAL;
5298 if (sector_div(temp, chunk))
5299 goto out_unlock;
5300 }
5301 mddev->resync_max = max;
5302 }
5303 wake_up(&mddev->recovery_wait);
5304 err = 0;
5305 out_unlock:
5306 spin_unlock(&mddev->lock);
5307 return err ?: len;
5308 }
5309
5310 static struct md_sysfs_entry md_max_sync =
5311 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5312
5313 static ssize_t
suspend_lo_show(struct mddev * mddev,char * page)5314 suspend_lo_show(struct mddev *mddev, char *page)
5315 {
5316 return sprintf(page, "%llu\n",
5317 (unsigned long long)READ_ONCE(mddev->suspend_lo));
5318 }
5319
5320 static ssize_t
suspend_lo_store(struct mddev * mddev,const char * buf,size_t len)5321 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5322 {
5323 unsigned long long new;
5324 int err;
5325
5326 err = kstrtoull(buf, 10, &new);
5327 if (err < 0)
5328 return err;
5329 if (new != (sector_t)new)
5330 return -EINVAL;
5331
5332 err = mddev_suspend(mddev, true);
5333 if (err)
5334 return err;
5335
5336 WRITE_ONCE(mddev->suspend_lo, new);
5337 mddev_resume(mddev);
5338
5339 return len;
5340 }
5341 static struct md_sysfs_entry md_suspend_lo =
5342 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5343
5344 static ssize_t
suspend_hi_show(struct mddev * mddev,char * page)5345 suspend_hi_show(struct mddev *mddev, char *page)
5346 {
5347 return sprintf(page, "%llu\n",
5348 (unsigned long long)READ_ONCE(mddev->suspend_hi));
5349 }
5350
5351 static ssize_t
suspend_hi_store(struct mddev * mddev,const char * buf,size_t len)5352 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5353 {
5354 unsigned long long new;
5355 int err;
5356
5357 err = kstrtoull(buf, 10, &new);
5358 if (err < 0)
5359 return err;
5360 if (new != (sector_t)new)
5361 return -EINVAL;
5362
5363 err = mddev_suspend(mddev, true);
5364 if (err)
5365 return err;
5366
5367 WRITE_ONCE(mddev->suspend_hi, new);
5368 mddev_resume(mddev);
5369
5370 return len;
5371 }
5372 static struct md_sysfs_entry md_suspend_hi =
5373 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5374
5375 static ssize_t
reshape_position_show(struct mddev * mddev,char * page)5376 reshape_position_show(struct mddev *mddev, char *page)
5377 {
5378 if (mddev->reshape_position != MaxSector)
5379 return sprintf(page, "%llu\n",
5380 (unsigned long long)mddev->reshape_position);
5381 strcpy(page, "none\n");
5382 return 5;
5383 }
5384
5385 static ssize_t
reshape_position_store(struct mddev * mddev,const char * buf,size_t len)5386 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5387 {
5388 struct md_rdev *rdev;
5389 unsigned long long new;
5390 int err;
5391
5392 err = kstrtoull(buf, 10, &new);
5393 if (err < 0)
5394 return err;
5395 if (new != (sector_t)new)
5396 return -EINVAL;
5397 err = mddev_lock(mddev);
5398 if (err)
5399 return err;
5400 err = -EBUSY;
5401 if (mddev->pers)
5402 goto unlock;
5403 mddev->reshape_position = new;
5404 mddev->delta_disks = 0;
5405 mddev->reshape_backwards = 0;
5406 mddev->new_level = mddev->level;
5407 mddev->new_layout = mddev->layout;
5408 mddev->new_chunk_sectors = mddev->chunk_sectors;
5409 rdev_for_each(rdev, mddev)
5410 rdev->new_data_offset = rdev->data_offset;
5411 err = 0;
5412 unlock:
5413 mddev_unlock(mddev);
5414 return err ?: len;
5415 }
5416
5417 static struct md_sysfs_entry md_reshape_position =
5418 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5419 reshape_position_store);
5420
5421 static ssize_t
reshape_direction_show(struct mddev * mddev,char * page)5422 reshape_direction_show(struct mddev *mddev, char *page)
5423 {
5424 return sprintf(page, "%s\n",
5425 mddev->reshape_backwards ? "backwards" : "forwards");
5426 }
5427
5428 static ssize_t
reshape_direction_store(struct mddev * mddev,const char * buf,size_t len)5429 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5430 {
5431 int backwards = 0;
5432 int err;
5433
5434 if (cmd_match(buf, "forwards"))
5435 backwards = 0;
5436 else if (cmd_match(buf, "backwards"))
5437 backwards = 1;
5438 else
5439 return -EINVAL;
5440 if (mddev->reshape_backwards == backwards)
5441 return len;
5442
5443 err = mddev_lock(mddev);
5444 if (err)
5445 return err;
5446 /* check if we are allowed to change */
5447 if (mddev->delta_disks)
5448 err = -EBUSY;
5449 else if (mddev->persistent &&
5450 mddev->major_version == 0)
5451 err = -EINVAL;
5452 else
5453 mddev->reshape_backwards = backwards;
5454 mddev_unlock(mddev);
5455 return err ?: len;
5456 }
5457
5458 static struct md_sysfs_entry md_reshape_direction =
5459 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5460 reshape_direction_store);
5461
5462 static ssize_t
array_size_show(struct mddev * mddev,char * page)5463 array_size_show(struct mddev *mddev, char *page)
5464 {
5465 if (mddev->external_size)
5466 return sprintf(page, "%llu\n",
5467 (unsigned long long)mddev->array_sectors/2);
5468 else
5469 return sprintf(page, "default\n");
5470 }
5471
5472 static ssize_t
array_size_store(struct mddev * mddev,const char * buf,size_t len)5473 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5474 {
5475 sector_t sectors;
5476 int err;
5477
5478 err = mddev_lock(mddev);
5479 if (err)
5480 return err;
5481
5482 /* cluster raid doesn't support change array_sectors */
5483 if (mddev_is_clustered(mddev)) {
5484 mddev_unlock(mddev);
5485 return -EINVAL;
5486 }
5487
5488 if (strncmp(buf, "default", 7) == 0) {
5489 if (mddev->pers)
5490 sectors = mddev->pers->size(mddev, 0, 0);
5491 else
5492 sectors = mddev->array_sectors;
5493
5494 mddev->external_size = 0;
5495 } else {
5496 if (strict_blocks_to_sectors(buf, §ors) < 0)
5497 err = -EINVAL;
5498 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5499 err = -E2BIG;
5500 else
5501 mddev->external_size = 1;
5502 }
5503
5504 if (!err) {
5505 mddev->array_sectors = sectors;
5506 if (mddev->pers)
5507 set_capacity_and_notify(mddev->gendisk,
5508 mddev->array_sectors);
5509 }
5510 mddev_unlock(mddev);
5511 return err ?: len;
5512 }
5513
5514 static struct md_sysfs_entry md_array_size =
5515 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5516 array_size_store);
5517
5518 static ssize_t
consistency_policy_show(struct mddev * mddev,char * page)5519 consistency_policy_show(struct mddev *mddev, char *page)
5520 {
5521 int ret;
5522
5523 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5524 ret = sprintf(page, "journal\n");
5525 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5526 ret = sprintf(page, "ppl\n");
5527 } else if (mddev->bitmap) {
5528 ret = sprintf(page, "bitmap\n");
5529 } else if (mddev->pers) {
5530 if (mddev->pers->sync_request)
5531 ret = sprintf(page, "resync\n");
5532 else
5533 ret = sprintf(page, "none\n");
5534 } else {
5535 ret = sprintf(page, "unknown\n");
5536 }
5537
5538 return ret;
5539 }
5540
5541 static ssize_t
consistency_policy_store(struct mddev * mddev,const char * buf,size_t len)5542 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5543 {
5544 int err = 0;
5545
5546 if (mddev->pers) {
5547 if (mddev->pers->change_consistency_policy)
5548 err = mddev->pers->change_consistency_policy(mddev, buf);
5549 else
5550 err = -EBUSY;
5551 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5552 set_bit(MD_HAS_PPL, &mddev->flags);
5553 } else {
5554 err = -EINVAL;
5555 }
5556
5557 return err ? err : len;
5558 }
5559
5560 static struct md_sysfs_entry md_consistency_policy =
5561 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5562 consistency_policy_store);
5563
fail_last_dev_show(struct mddev * mddev,char * page)5564 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5565 {
5566 return sprintf(page, "%d\n", mddev->fail_last_dev);
5567 }
5568
5569 /*
5570 * Setting fail_last_dev to true to allow last device to be forcibly removed
5571 * from RAID1/RAID10.
5572 */
5573 static ssize_t
fail_last_dev_store(struct mddev * mddev,const char * buf,size_t len)5574 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5575 {
5576 int ret;
5577 bool value;
5578
5579 ret = kstrtobool(buf, &value);
5580 if (ret)
5581 return ret;
5582
5583 if (value != mddev->fail_last_dev)
5584 mddev->fail_last_dev = value;
5585
5586 return len;
5587 }
5588 static struct md_sysfs_entry md_fail_last_dev =
5589 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5590 fail_last_dev_store);
5591
serialize_policy_show(struct mddev * mddev,char * page)5592 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5593 {
5594 if (mddev->pers == NULL || (mddev->pers->head.id != ID_RAID1))
5595 return sprintf(page, "n/a\n");
5596 else
5597 return sprintf(page, "%d\n", mddev->serialize_policy);
5598 }
5599
5600 /*
5601 * Setting serialize_policy to true to enforce write IO is not reordered
5602 * for raid1.
5603 */
5604 static ssize_t
serialize_policy_store(struct mddev * mddev,const char * buf,size_t len)5605 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5606 {
5607 int err;
5608 bool value;
5609
5610 err = kstrtobool(buf, &value);
5611 if (err)
5612 return err;
5613
5614 if (value == mddev->serialize_policy)
5615 return len;
5616
5617 err = mddev_suspend_and_lock(mddev);
5618 if (err)
5619 return err;
5620 if (mddev->pers == NULL || (mddev->pers->head.id != ID_RAID1)) {
5621 pr_err("md: serialize_policy is only effective for raid1\n");
5622 err = -EINVAL;
5623 goto unlock;
5624 }
5625
5626 if (value)
5627 mddev_create_serial_pool(mddev, NULL);
5628 else
5629 mddev_destroy_serial_pool(mddev, NULL);
5630 mddev->serialize_policy = value;
5631 unlock:
5632 mddev_unlock_and_resume(mddev);
5633 return err ?: len;
5634 }
5635
5636 static struct md_sysfs_entry md_serialize_policy =
5637 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5638 serialize_policy_store);
5639
5640
5641 static struct attribute *md_default_attrs[] = {
5642 &md_level.attr,
5643 &md_new_level.attr,
5644 &md_layout.attr,
5645 &md_raid_disks.attr,
5646 &md_uuid.attr,
5647 &md_chunk_size.attr,
5648 &md_size.attr,
5649 &md_resync_start.attr,
5650 &md_metadata.attr,
5651 &md_new_device.attr,
5652 &md_safe_delay.attr,
5653 &md_array_state.attr,
5654 &md_reshape_position.attr,
5655 &md_reshape_direction.attr,
5656 &md_array_size.attr,
5657 &max_corr_read_errors.attr,
5658 &md_consistency_policy.attr,
5659 &md_fail_last_dev.attr,
5660 &md_serialize_policy.attr,
5661 NULL,
5662 };
5663
5664 static const struct attribute_group md_default_group = {
5665 .attrs = md_default_attrs,
5666 };
5667
5668 static struct attribute *md_redundancy_attrs[] = {
5669 &md_scan_mode.attr,
5670 &md_last_scan_mode.attr,
5671 &md_mismatches.attr,
5672 &md_sync_min.attr,
5673 &md_sync_max.attr,
5674 &md_sync_speed.attr,
5675 &md_sync_force_parallel.attr,
5676 &md_sync_completed.attr,
5677 &md_min_sync.attr,
5678 &md_max_sync.attr,
5679 &md_suspend_lo.attr,
5680 &md_suspend_hi.attr,
5681 &md_bitmap.attr,
5682 &md_degraded.attr,
5683 NULL,
5684 };
5685 static const struct attribute_group md_redundancy_group = {
5686 .name = NULL,
5687 .attrs = md_redundancy_attrs,
5688 };
5689
5690 static const struct attribute_group *md_attr_groups[] = {
5691 &md_default_group,
5692 &md_bitmap_group,
5693 NULL,
5694 };
5695
5696 static ssize_t
md_attr_show(struct kobject * kobj,struct attribute * attr,char * page)5697 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5698 {
5699 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5700 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5701 ssize_t rv;
5702
5703 if (!entry->show)
5704 return -EIO;
5705 spin_lock(&all_mddevs_lock);
5706 if (!mddev_get(mddev)) {
5707 spin_unlock(&all_mddevs_lock);
5708 return -EBUSY;
5709 }
5710 spin_unlock(&all_mddevs_lock);
5711
5712 rv = entry->show(mddev, page);
5713 mddev_put(mddev);
5714 return rv;
5715 }
5716
5717 static ssize_t
md_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)5718 md_attr_store(struct kobject *kobj, struct attribute *attr,
5719 const char *page, size_t length)
5720 {
5721 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5722 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5723 ssize_t rv;
5724
5725 if (!entry->store)
5726 return -EIO;
5727 if (!capable(CAP_SYS_ADMIN))
5728 return -EACCES;
5729 spin_lock(&all_mddevs_lock);
5730 if (!mddev_get(mddev)) {
5731 spin_unlock(&all_mddevs_lock);
5732 return -EBUSY;
5733 }
5734 spin_unlock(&all_mddevs_lock);
5735 rv = entry->store(mddev, page, length);
5736 mddev_put(mddev);
5737 return rv;
5738 }
5739
md_kobj_release(struct kobject * ko)5740 static void md_kobj_release(struct kobject *ko)
5741 {
5742 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5743
5744 if (mddev->sysfs_state)
5745 sysfs_put(mddev->sysfs_state);
5746 if (mddev->sysfs_level)
5747 sysfs_put(mddev->sysfs_level);
5748
5749 del_gendisk(mddev->gendisk);
5750 put_disk(mddev->gendisk);
5751 }
5752
5753 static const struct sysfs_ops md_sysfs_ops = {
5754 .show = md_attr_show,
5755 .store = md_attr_store,
5756 };
5757 static const struct kobj_type md_ktype = {
5758 .release = md_kobj_release,
5759 .sysfs_ops = &md_sysfs_ops,
5760 .default_groups = md_attr_groups,
5761 };
5762
5763 int mdp_major = 0;
5764
5765 /* stack the limit for all rdevs into lim */
mddev_stack_rdev_limits(struct mddev * mddev,struct queue_limits * lim,unsigned int flags)5766 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
5767 unsigned int flags)
5768 {
5769 struct md_rdev *rdev;
5770
5771 rdev_for_each(rdev, mddev) {
5772 queue_limits_stack_bdev(lim, rdev->bdev, rdev->data_offset,
5773 mddev->gendisk->disk_name);
5774 if ((flags & MDDEV_STACK_INTEGRITY) &&
5775 !queue_limits_stack_integrity_bdev(lim, rdev->bdev))
5776 return -EINVAL;
5777 }
5778
5779 return 0;
5780 }
5781 EXPORT_SYMBOL_GPL(mddev_stack_rdev_limits);
5782
5783 /* apply the extra stacking limits from a new rdev into mddev */
mddev_stack_new_rdev(struct mddev * mddev,struct md_rdev * rdev)5784 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev)
5785 {
5786 struct queue_limits lim;
5787
5788 if (mddev_is_dm(mddev))
5789 return 0;
5790
5791 lim = queue_limits_start_update(mddev->gendisk->queue);
5792 queue_limits_stack_bdev(&lim, rdev->bdev, rdev->data_offset,
5793 mddev->gendisk->disk_name);
5794
5795 if (!queue_limits_stack_integrity_bdev(&lim, rdev->bdev)) {
5796 pr_err("%s: incompatible integrity profile for %pg\n",
5797 mdname(mddev), rdev->bdev);
5798 queue_limits_cancel_update(mddev->gendisk->queue);
5799 return -ENXIO;
5800 }
5801
5802 return queue_limits_commit_update(mddev->gendisk->queue, &lim);
5803 }
5804 EXPORT_SYMBOL_GPL(mddev_stack_new_rdev);
5805
5806 /* update the optimal I/O size after a reshape */
mddev_update_io_opt(struct mddev * mddev,unsigned int nr_stripes)5807 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes)
5808 {
5809 struct queue_limits lim;
5810
5811 if (mddev_is_dm(mddev))
5812 return;
5813
5814 /* don't bother updating io_opt if we can't suspend the array */
5815 if (mddev_suspend(mddev, false) < 0)
5816 return;
5817 lim = queue_limits_start_update(mddev->gendisk->queue);
5818 lim.io_opt = lim.io_min * nr_stripes;
5819 queue_limits_commit_update(mddev->gendisk->queue, &lim);
5820 mddev_resume(mddev);
5821 }
5822 EXPORT_SYMBOL_GPL(mddev_update_io_opt);
5823
mddev_delayed_delete(struct work_struct * ws)5824 static void mddev_delayed_delete(struct work_struct *ws)
5825 {
5826 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5827
5828 kobject_put(&mddev->kobj);
5829 }
5830
md_init_stacking_limits(struct queue_limits * lim)5831 void md_init_stacking_limits(struct queue_limits *lim)
5832 {
5833 blk_set_stacking_limits(lim);
5834 lim->features = BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA |
5835 BLK_FEAT_IO_STAT | BLK_FEAT_NOWAIT;
5836 }
5837 EXPORT_SYMBOL_GPL(md_init_stacking_limits);
5838
md_alloc(dev_t dev,char * name)5839 struct mddev *md_alloc(dev_t dev, char *name)
5840 {
5841 /*
5842 * If dev is zero, name is the name of a device to allocate with
5843 * an arbitrary minor number. It will be "md_???"
5844 * If dev is non-zero it must be a device number with a MAJOR of
5845 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5846 * the device is being created by opening a node in /dev.
5847 * If "name" is not NULL, the device is being created by
5848 * writing to /sys/module/md_mod/parameters/new_array.
5849 */
5850 static DEFINE_MUTEX(disks_mutex);
5851 struct mddev *mddev;
5852 struct gendisk *disk;
5853 int partitioned;
5854 int shift;
5855 int unit;
5856 int error;
5857
5858 /*
5859 * Wait for any previous instance of this device to be completely
5860 * removed (mddev_delayed_delete).
5861 */
5862 flush_workqueue(md_misc_wq);
5863
5864 mutex_lock(&disks_mutex);
5865 mddev = mddev_alloc(dev);
5866 if (IS_ERR(mddev)) {
5867 error = PTR_ERR(mddev);
5868 goto out_unlock;
5869 }
5870
5871 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5872 shift = partitioned ? MdpMinorShift : 0;
5873 unit = MINOR(mddev->unit) >> shift;
5874
5875 if (name && !dev) {
5876 /* Need to ensure that 'name' is not a duplicate.
5877 */
5878 struct mddev *mddev2;
5879 spin_lock(&all_mddevs_lock);
5880
5881 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5882 if (mddev2->gendisk &&
5883 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5884 spin_unlock(&all_mddevs_lock);
5885 error = -EEXIST;
5886 goto out_free_mddev;
5887 }
5888 spin_unlock(&all_mddevs_lock);
5889 }
5890 if (name && dev)
5891 /*
5892 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5893 */
5894 mddev->hold_active = UNTIL_STOP;
5895
5896 disk = blk_alloc_disk(NULL, NUMA_NO_NODE);
5897 if (IS_ERR(disk)) {
5898 error = PTR_ERR(disk);
5899 goto out_free_mddev;
5900 }
5901
5902 disk->major = MAJOR(mddev->unit);
5903 disk->first_minor = unit << shift;
5904 disk->minors = 1 << shift;
5905 if (name)
5906 strcpy(disk->disk_name, name);
5907 else if (partitioned)
5908 sprintf(disk->disk_name, "md_d%d", unit);
5909 else
5910 sprintf(disk->disk_name, "md%d", unit);
5911 disk->fops = &md_fops;
5912 disk->private_data = mddev;
5913
5914 disk->events |= DISK_EVENT_MEDIA_CHANGE;
5915 mddev->gendisk = disk;
5916 error = add_disk(disk);
5917 if (error)
5918 goto out_put_disk;
5919
5920 kobject_init(&mddev->kobj, &md_ktype);
5921 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5922 if (error) {
5923 /*
5924 * The disk is already live at this point. Clear the hold flag
5925 * and let mddev_put take care of the deletion, as it isn't any
5926 * different from a normal close on last release now.
5927 */
5928 mddev->hold_active = 0;
5929 mutex_unlock(&disks_mutex);
5930 mddev_put(mddev);
5931 return ERR_PTR(error);
5932 }
5933
5934 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5935 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5936 mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5937 mutex_unlock(&disks_mutex);
5938 return mddev;
5939
5940 out_put_disk:
5941 put_disk(disk);
5942 out_free_mddev:
5943 mddev_free(mddev);
5944 out_unlock:
5945 mutex_unlock(&disks_mutex);
5946 return ERR_PTR(error);
5947 }
5948
md_alloc_and_put(dev_t dev,char * name)5949 static int md_alloc_and_put(dev_t dev, char *name)
5950 {
5951 struct mddev *mddev = md_alloc(dev, name);
5952
5953 if (IS_ERR(mddev))
5954 return PTR_ERR(mddev);
5955 mddev_put(mddev);
5956 return 0;
5957 }
5958
md_probe(dev_t dev)5959 static void md_probe(dev_t dev)
5960 {
5961 if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5962 return;
5963 if (create_on_open)
5964 md_alloc_and_put(dev, NULL);
5965 }
5966
add_named_array(const char * val,const struct kernel_param * kp)5967 static int add_named_array(const char *val, const struct kernel_param *kp)
5968 {
5969 /*
5970 * val must be "md_*" or "mdNNN".
5971 * For "md_*" we allocate an array with a large free minor number, and
5972 * set the name to val. val must not already be an active name.
5973 * For "mdNNN" we allocate an array with the minor number NNN
5974 * which must not already be in use.
5975 */
5976 int len = strlen(val);
5977 char buf[DISK_NAME_LEN];
5978 unsigned long devnum;
5979
5980 while (len && val[len-1] == '\n')
5981 len--;
5982 if (len >= DISK_NAME_LEN)
5983 return -E2BIG;
5984 strscpy(buf, val, len+1);
5985 if (strncmp(buf, "md_", 3) == 0)
5986 return md_alloc_and_put(0, buf);
5987 if (strncmp(buf, "md", 2) == 0 &&
5988 isdigit(buf[2]) &&
5989 kstrtoul(buf+2, 10, &devnum) == 0 &&
5990 devnum <= MINORMASK)
5991 return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5992
5993 return -EINVAL;
5994 }
5995
md_safemode_timeout(struct timer_list * t)5996 static void md_safemode_timeout(struct timer_list *t)
5997 {
5998 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5999
6000 mddev->safemode = 1;
6001 if (mddev->external)
6002 sysfs_notify_dirent_safe(mddev->sysfs_state);
6003
6004 md_wakeup_thread(mddev->thread);
6005 }
6006
6007 static int start_dirty_degraded;
6008
md_run(struct mddev * mddev)6009 int md_run(struct mddev *mddev)
6010 {
6011 int err;
6012 struct md_rdev *rdev;
6013 struct md_personality *pers;
6014 bool nowait = true;
6015
6016 if (list_empty(&mddev->disks))
6017 /* cannot run an array with no devices.. */
6018 return -EINVAL;
6019
6020 if (mddev->pers)
6021 return -EBUSY;
6022 /* Cannot run until previous stop completes properly */
6023 if (mddev->sysfs_active)
6024 return -EBUSY;
6025
6026 /*
6027 * Analyze all RAID superblock(s)
6028 */
6029 if (!mddev->raid_disks) {
6030 if (!mddev->persistent)
6031 return -EINVAL;
6032 err = analyze_sbs(mddev);
6033 if (err)
6034 return -EINVAL;
6035 }
6036
6037 if (mddev->level != LEVEL_NONE)
6038 request_module("md-level-%d", mddev->level);
6039 else if (mddev->clevel[0])
6040 request_module("md-%s", mddev->clevel);
6041
6042 /*
6043 * Drop all container device buffers, from now on
6044 * the only valid external interface is through the md
6045 * device.
6046 */
6047 mddev->has_superblocks = false;
6048 rdev_for_each(rdev, mddev) {
6049 if (test_bit(Faulty, &rdev->flags))
6050 continue;
6051 sync_blockdev(rdev->bdev);
6052 invalidate_bdev(rdev->bdev);
6053 if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
6054 mddev->ro = MD_RDONLY;
6055 if (!mddev_is_dm(mddev))
6056 set_disk_ro(mddev->gendisk, 1);
6057 }
6058
6059 if (rdev->sb_page)
6060 mddev->has_superblocks = true;
6061
6062 /* perform some consistency tests on the device.
6063 * We don't want the data to overlap the metadata,
6064 * Internal Bitmap issues have been handled elsewhere.
6065 */
6066 if (rdev->meta_bdev) {
6067 /* Nothing to check */;
6068 } else if (rdev->data_offset < rdev->sb_start) {
6069 if (mddev->dev_sectors &&
6070 rdev->data_offset + mddev->dev_sectors
6071 > rdev->sb_start) {
6072 pr_warn("md: %s: data overlaps metadata\n",
6073 mdname(mddev));
6074 return -EINVAL;
6075 }
6076 } else {
6077 if (rdev->sb_start + rdev->sb_size/512
6078 > rdev->data_offset) {
6079 pr_warn("md: %s: metadata overlaps data\n",
6080 mdname(mddev));
6081 return -EINVAL;
6082 }
6083 }
6084 sysfs_notify_dirent_safe(rdev->sysfs_state);
6085 nowait = nowait && bdev_nowait(rdev->bdev);
6086 }
6087
6088 if (!bioset_initialized(&mddev->bio_set)) {
6089 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
6090 if (err)
6091 return err;
6092 }
6093 if (!bioset_initialized(&mddev->sync_set)) {
6094 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
6095 if (err)
6096 goto exit_bio_set;
6097 }
6098
6099 if (!bioset_initialized(&mddev->io_clone_set)) {
6100 err = bioset_init(&mddev->io_clone_set, BIO_POOL_SIZE,
6101 offsetof(struct md_io_clone, bio_clone), 0);
6102 if (err)
6103 goto exit_sync_set;
6104 }
6105
6106 pers = get_pers(mddev->level, mddev->clevel);
6107 if (!pers) {
6108 err = -EINVAL;
6109 goto abort;
6110 }
6111 if (mddev->level != pers->head.id) {
6112 mddev->level = pers->head.id;
6113 mddev->new_level = pers->head.id;
6114 }
6115 strscpy(mddev->clevel, pers->head.name, sizeof(mddev->clevel));
6116
6117 if (mddev->reshape_position != MaxSector &&
6118 pers->start_reshape == NULL) {
6119 /* This personality cannot handle reshaping... */
6120 put_pers(pers);
6121 err = -EINVAL;
6122 goto abort;
6123 }
6124
6125 if (pers->sync_request) {
6126 /* Warn if this is a potentially silly
6127 * configuration.
6128 */
6129 struct md_rdev *rdev2;
6130 int warned = 0;
6131
6132 rdev_for_each(rdev, mddev)
6133 rdev_for_each(rdev2, mddev) {
6134 if (rdev < rdev2 &&
6135 rdev->bdev->bd_disk ==
6136 rdev2->bdev->bd_disk) {
6137 pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
6138 mdname(mddev),
6139 rdev->bdev,
6140 rdev2->bdev);
6141 warned = 1;
6142 }
6143 }
6144
6145 if (warned)
6146 pr_warn("True protection against single-disk failure might be compromised.\n");
6147 }
6148
6149 /* dm-raid expect sync_thread to be frozen until resume */
6150 if (mddev->gendisk)
6151 mddev->recovery = 0;
6152
6153 /* may be over-ridden by personality */
6154 mddev->resync_max_sectors = mddev->dev_sectors;
6155
6156 mddev->ok_start_degraded = start_dirty_degraded;
6157
6158 if (start_readonly && md_is_rdwr(mddev))
6159 mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
6160
6161 err = pers->run(mddev);
6162 if (err)
6163 pr_warn("md: pers->run() failed ...\n");
6164 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
6165 WARN_ONCE(!mddev->external_size,
6166 "%s: default size too small, but 'external_size' not in effect?\n",
6167 __func__);
6168 pr_warn("md: invalid array_size %llu > default size %llu\n",
6169 (unsigned long long)mddev->array_sectors / 2,
6170 (unsigned long long)pers->size(mddev, 0, 0) / 2);
6171 err = -EINVAL;
6172 }
6173 if (err == 0 && pers->sync_request &&
6174 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
6175 err = mddev->bitmap_ops->create(mddev, -1);
6176 if (err)
6177 pr_warn("%s: failed to create bitmap (%d)\n",
6178 mdname(mddev), err);
6179 }
6180 if (err)
6181 goto bitmap_abort;
6182
6183 if (mddev->bitmap_info.max_write_behind > 0) {
6184 bool create_pool = false;
6185
6186 rdev_for_each(rdev, mddev) {
6187 if (test_bit(WriteMostly, &rdev->flags) &&
6188 rdev_init_serial(rdev))
6189 create_pool = true;
6190 }
6191 if (create_pool && mddev->serial_info_pool == NULL) {
6192 mddev->serial_info_pool =
6193 mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6194 sizeof(struct serial_info));
6195 if (!mddev->serial_info_pool) {
6196 err = -ENOMEM;
6197 goto bitmap_abort;
6198 }
6199 }
6200 }
6201
6202 if (pers->sync_request) {
6203 if (mddev->kobj.sd &&
6204 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6205 pr_warn("md: cannot register extra attributes for %s\n",
6206 mdname(mddev));
6207 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6208 mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6209 mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6210 } else if (mddev->ro == MD_AUTO_READ)
6211 mddev->ro = MD_RDWR;
6212
6213 atomic_set(&mddev->max_corr_read_errors,
6214 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6215 mddev->safemode = 0;
6216 if (mddev_is_clustered(mddev))
6217 mddev->safemode_delay = 0;
6218 else
6219 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6220 mddev->in_sync = 1;
6221 smp_wmb();
6222 spin_lock(&mddev->lock);
6223 mddev->pers = pers;
6224 spin_unlock(&mddev->lock);
6225 rdev_for_each(rdev, mddev)
6226 if (rdev->raid_disk >= 0)
6227 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6228
6229 if (mddev->degraded && md_is_rdwr(mddev))
6230 /* This ensures that recovering status is reported immediately
6231 * via sysfs - until a lack of spares is confirmed.
6232 */
6233 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6234 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6235
6236 if (mddev->sb_flags)
6237 md_update_sb(mddev, 0);
6238
6239 md_new_event();
6240 return 0;
6241
6242 bitmap_abort:
6243 mddev_detach(mddev);
6244 if (mddev->private)
6245 pers->free(mddev, mddev->private);
6246 mddev->private = NULL;
6247 put_pers(pers);
6248 mddev->bitmap_ops->destroy(mddev);
6249 abort:
6250 bioset_exit(&mddev->io_clone_set);
6251 exit_sync_set:
6252 bioset_exit(&mddev->sync_set);
6253 exit_bio_set:
6254 bioset_exit(&mddev->bio_set);
6255 return err;
6256 }
6257 EXPORT_SYMBOL_GPL(md_run);
6258
do_md_run(struct mddev * mddev)6259 int do_md_run(struct mddev *mddev)
6260 {
6261 int err;
6262
6263 set_bit(MD_NOT_READY, &mddev->flags);
6264 err = md_run(mddev);
6265 if (err)
6266 goto out;
6267
6268 err = mddev->bitmap_ops->load(mddev);
6269 if (err) {
6270 mddev->bitmap_ops->destroy(mddev);
6271 goto out;
6272 }
6273
6274 if (mddev_is_clustered(mddev))
6275 md_allow_write(mddev);
6276
6277 /* run start up tasks that require md_thread */
6278 md_start(mddev);
6279
6280 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6281
6282 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6283 clear_bit(MD_NOT_READY, &mddev->flags);
6284 mddev->changed = 1;
6285 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6286 sysfs_notify_dirent_safe(mddev->sysfs_state);
6287 sysfs_notify_dirent_safe(mddev->sysfs_action);
6288 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6289 out:
6290 clear_bit(MD_NOT_READY, &mddev->flags);
6291 return err;
6292 }
6293
md_start(struct mddev * mddev)6294 int md_start(struct mddev *mddev)
6295 {
6296 int ret = 0;
6297
6298 if (mddev->pers->start) {
6299 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6300 ret = mddev->pers->start(mddev);
6301 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6302 md_wakeup_thread(mddev->sync_thread);
6303 }
6304 return ret;
6305 }
6306 EXPORT_SYMBOL_GPL(md_start);
6307
restart_array(struct mddev * mddev)6308 static int restart_array(struct mddev *mddev)
6309 {
6310 struct gendisk *disk = mddev->gendisk;
6311 struct md_rdev *rdev;
6312 bool has_journal = false;
6313 bool has_readonly = false;
6314
6315 /* Complain if it has no devices */
6316 if (list_empty(&mddev->disks))
6317 return -ENXIO;
6318 if (!mddev->pers)
6319 return -EINVAL;
6320 if (md_is_rdwr(mddev))
6321 return -EBUSY;
6322
6323 rcu_read_lock();
6324 rdev_for_each_rcu(rdev, mddev) {
6325 if (test_bit(Journal, &rdev->flags) &&
6326 !test_bit(Faulty, &rdev->flags))
6327 has_journal = true;
6328 if (rdev_read_only(rdev))
6329 has_readonly = true;
6330 }
6331 rcu_read_unlock();
6332 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6333 /* Don't restart rw with journal missing/faulty */
6334 return -EINVAL;
6335 if (has_readonly)
6336 return -EROFS;
6337
6338 mddev->safemode = 0;
6339 mddev->ro = MD_RDWR;
6340 set_disk_ro(disk, 0);
6341 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6342 /* Kick recovery or resync if necessary */
6343 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6344 md_wakeup_thread(mddev->sync_thread);
6345 sysfs_notify_dirent_safe(mddev->sysfs_state);
6346 return 0;
6347 }
6348
md_clean(struct mddev * mddev)6349 static void md_clean(struct mddev *mddev)
6350 {
6351 mddev->array_sectors = 0;
6352 mddev->external_size = 0;
6353 mddev->dev_sectors = 0;
6354 mddev->raid_disks = 0;
6355 mddev->recovery_cp = 0;
6356 mddev->resync_min = 0;
6357 mddev->resync_max = MaxSector;
6358 mddev->reshape_position = MaxSector;
6359 /* we still need mddev->external in export_rdev, do not clear it yet */
6360 mddev->persistent = 0;
6361 mddev->level = LEVEL_NONE;
6362 mddev->clevel[0] = 0;
6363 /*
6364 * Don't clear MD_CLOSING, or mddev can be opened again.
6365 * 'hold_active != 0' means mddev is still in the creation
6366 * process and will be used later.
6367 */
6368 if (mddev->hold_active)
6369 mddev->flags = 0;
6370 else
6371 mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6372 mddev->sb_flags = 0;
6373 mddev->ro = MD_RDWR;
6374 mddev->metadata_type[0] = 0;
6375 mddev->chunk_sectors = 0;
6376 mddev->ctime = mddev->utime = 0;
6377 mddev->layout = 0;
6378 mddev->max_disks = 0;
6379 mddev->events = 0;
6380 mddev->can_decrease_events = 0;
6381 mddev->delta_disks = 0;
6382 mddev->reshape_backwards = 0;
6383 mddev->new_level = LEVEL_NONE;
6384 mddev->new_layout = 0;
6385 mddev->new_chunk_sectors = 0;
6386 mddev->curr_resync = MD_RESYNC_NONE;
6387 atomic64_set(&mddev->resync_mismatches, 0);
6388 mddev->suspend_lo = mddev->suspend_hi = 0;
6389 mddev->sync_speed_min = mddev->sync_speed_max = 0;
6390 mddev->recovery = 0;
6391 mddev->in_sync = 0;
6392 mddev->changed = 0;
6393 mddev->degraded = 0;
6394 mddev->safemode = 0;
6395 mddev->private = NULL;
6396 mddev->cluster_info = NULL;
6397 mddev->bitmap_info.offset = 0;
6398 mddev->bitmap_info.default_offset = 0;
6399 mddev->bitmap_info.default_space = 0;
6400 mddev->bitmap_info.chunksize = 0;
6401 mddev->bitmap_info.daemon_sleep = 0;
6402 mddev->bitmap_info.max_write_behind = 0;
6403 mddev->bitmap_info.nodes = 0;
6404 }
6405
__md_stop_writes(struct mddev * mddev)6406 static void __md_stop_writes(struct mddev *mddev)
6407 {
6408 del_timer_sync(&mddev->safemode_timer);
6409
6410 if (mddev->pers && mddev->pers->quiesce) {
6411 mddev->pers->quiesce(mddev, 1);
6412 mddev->pers->quiesce(mddev, 0);
6413 }
6414
6415 mddev->bitmap_ops->flush(mddev);
6416
6417 if (md_is_rdwr(mddev) &&
6418 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6419 mddev->sb_flags)) {
6420 /* mark array as shutdown cleanly */
6421 if (!mddev_is_clustered(mddev))
6422 mddev->in_sync = 1;
6423 md_update_sb(mddev, 1);
6424 }
6425 /* disable policy to guarantee rdevs free resources for serialization */
6426 mddev->serialize_policy = 0;
6427 mddev_destroy_serial_pool(mddev, NULL);
6428 }
6429
md_stop_writes(struct mddev * mddev)6430 void md_stop_writes(struct mddev *mddev)
6431 {
6432 mddev_lock_nointr(mddev);
6433 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6434 stop_sync_thread(mddev, true);
6435 __md_stop_writes(mddev);
6436 mddev_unlock(mddev);
6437 }
6438 EXPORT_SYMBOL_GPL(md_stop_writes);
6439
mddev_detach(struct mddev * mddev)6440 static void mddev_detach(struct mddev *mddev)
6441 {
6442 mddev->bitmap_ops->wait_behind_writes(mddev);
6443 if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6444 mddev->pers->quiesce(mddev, 1);
6445 mddev->pers->quiesce(mddev, 0);
6446 }
6447 md_unregister_thread(mddev, &mddev->thread);
6448
6449 /* the unplug fn references 'conf' */
6450 if (!mddev_is_dm(mddev))
6451 blk_sync_queue(mddev->gendisk->queue);
6452 }
6453
__md_stop(struct mddev * mddev)6454 static void __md_stop(struct mddev *mddev)
6455 {
6456 struct md_personality *pers = mddev->pers;
6457
6458 mddev->bitmap_ops->destroy(mddev);
6459 mddev_detach(mddev);
6460 spin_lock(&mddev->lock);
6461 mddev->pers = NULL;
6462 spin_unlock(&mddev->lock);
6463 if (mddev->private)
6464 pers->free(mddev, mddev->private);
6465 mddev->private = NULL;
6466 if (pers->sync_request && mddev->to_remove == NULL)
6467 mddev->to_remove = &md_redundancy_group;
6468 put_pers(pers);
6469 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6470
6471 bioset_exit(&mddev->bio_set);
6472 bioset_exit(&mddev->sync_set);
6473 bioset_exit(&mddev->io_clone_set);
6474 }
6475
md_stop(struct mddev * mddev)6476 void md_stop(struct mddev *mddev)
6477 {
6478 lockdep_assert_held(&mddev->reconfig_mutex);
6479
6480 /* stop the array and free an attached data structures.
6481 * This is called from dm-raid
6482 */
6483 __md_stop_writes(mddev);
6484 __md_stop(mddev);
6485 }
6486
6487 EXPORT_SYMBOL_GPL(md_stop);
6488
6489 /* ensure 'mddev->pers' exist before calling md_set_readonly() */
md_set_readonly(struct mddev * mddev)6490 static int md_set_readonly(struct mddev *mddev)
6491 {
6492 int err = 0;
6493 int did_freeze = 0;
6494
6495 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6496 return -EBUSY;
6497
6498 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6499 did_freeze = 1;
6500 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6501 }
6502
6503 stop_sync_thread(mddev, false);
6504 wait_event(mddev->sb_wait,
6505 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6506 mddev_lock_nointr(mddev);
6507
6508 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6509 pr_warn("md: %s still in use.\n",mdname(mddev));
6510 err = -EBUSY;
6511 goto out;
6512 }
6513
6514 __md_stop_writes(mddev);
6515
6516 if (mddev->ro == MD_RDONLY) {
6517 err = -ENXIO;
6518 goto out;
6519 }
6520
6521 mddev->ro = MD_RDONLY;
6522 set_disk_ro(mddev->gendisk, 1);
6523
6524 out:
6525 if (!err || did_freeze) {
6526 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6527 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6528 sysfs_notify_dirent_safe(mddev->sysfs_state);
6529 }
6530
6531 return err;
6532 }
6533
6534 /* mode:
6535 * 0 - completely stop and dis-assemble array
6536 * 2 - stop but do not disassemble array
6537 */
do_md_stop(struct mddev * mddev,int mode)6538 static int do_md_stop(struct mddev *mddev, int mode)
6539 {
6540 struct gendisk *disk = mddev->gendisk;
6541 struct md_rdev *rdev;
6542 int did_freeze = 0;
6543
6544 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6545 did_freeze = 1;
6546 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6547 }
6548
6549 stop_sync_thread(mddev, true);
6550
6551 if (mddev->sysfs_active ||
6552 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6553 pr_warn("md: %s still in use.\n",mdname(mddev));
6554 if (did_freeze) {
6555 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6556 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6557 }
6558 return -EBUSY;
6559 }
6560 if (mddev->pers) {
6561 if (!md_is_rdwr(mddev))
6562 set_disk_ro(disk, 0);
6563
6564 __md_stop_writes(mddev);
6565 __md_stop(mddev);
6566
6567 /* tell userspace to handle 'inactive' */
6568 sysfs_notify_dirent_safe(mddev->sysfs_state);
6569
6570 rdev_for_each(rdev, mddev)
6571 if (rdev->raid_disk >= 0)
6572 sysfs_unlink_rdev(mddev, rdev);
6573
6574 set_capacity_and_notify(disk, 0);
6575 mddev->changed = 1;
6576
6577 if (!md_is_rdwr(mddev))
6578 mddev->ro = MD_RDWR;
6579 }
6580 /*
6581 * Free resources if final stop
6582 */
6583 if (mode == 0) {
6584 pr_info("md: %s stopped.\n", mdname(mddev));
6585
6586 if (mddev->bitmap_info.file) {
6587 struct file *f = mddev->bitmap_info.file;
6588 spin_lock(&mddev->lock);
6589 mddev->bitmap_info.file = NULL;
6590 spin_unlock(&mddev->lock);
6591 fput(f);
6592 }
6593 mddev->bitmap_info.offset = 0;
6594
6595 export_array(mddev);
6596
6597 md_clean(mddev);
6598 if (mddev->hold_active == UNTIL_STOP)
6599 mddev->hold_active = 0;
6600 }
6601 md_new_event();
6602 sysfs_notify_dirent_safe(mddev->sysfs_state);
6603 return 0;
6604 }
6605
6606 #ifndef MODULE
autorun_array(struct mddev * mddev)6607 static void autorun_array(struct mddev *mddev)
6608 {
6609 struct md_rdev *rdev;
6610 int err;
6611
6612 if (list_empty(&mddev->disks))
6613 return;
6614
6615 pr_info("md: running: ");
6616
6617 rdev_for_each(rdev, mddev) {
6618 pr_cont("<%pg>", rdev->bdev);
6619 }
6620 pr_cont("\n");
6621
6622 err = do_md_run(mddev);
6623 if (err) {
6624 pr_warn("md: do_md_run() returned %d\n", err);
6625 do_md_stop(mddev, 0);
6626 }
6627 }
6628
6629 /*
6630 * lets try to run arrays based on all disks that have arrived
6631 * until now. (those are in pending_raid_disks)
6632 *
6633 * the method: pick the first pending disk, collect all disks with
6634 * the same UUID, remove all from the pending list and put them into
6635 * the 'same_array' list. Then order this list based on superblock
6636 * update time (freshest comes first), kick out 'old' disks and
6637 * compare superblocks. If everything's fine then run it.
6638 *
6639 * If "unit" is allocated, then bump its reference count
6640 */
autorun_devices(int part)6641 static void autorun_devices(int part)
6642 {
6643 struct md_rdev *rdev0, *rdev, *tmp;
6644 struct mddev *mddev;
6645
6646 pr_info("md: autorun ...\n");
6647 while (!list_empty(&pending_raid_disks)) {
6648 int unit;
6649 dev_t dev;
6650 LIST_HEAD(candidates);
6651 rdev0 = list_entry(pending_raid_disks.next,
6652 struct md_rdev, same_set);
6653
6654 pr_debug("md: considering %pg ...\n", rdev0->bdev);
6655 INIT_LIST_HEAD(&candidates);
6656 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6657 if (super_90_load(rdev, rdev0, 0) >= 0) {
6658 pr_debug("md: adding %pg ...\n",
6659 rdev->bdev);
6660 list_move(&rdev->same_set, &candidates);
6661 }
6662 /*
6663 * now we have a set of devices, with all of them having
6664 * mostly sane superblocks. It's time to allocate the
6665 * mddev.
6666 */
6667 if (part) {
6668 dev = MKDEV(mdp_major,
6669 rdev0->preferred_minor << MdpMinorShift);
6670 unit = MINOR(dev) >> MdpMinorShift;
6671 } else {
6672 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6673 unit = MINOR(dev);
6674 }
6675 if (rdev0->preferred_minor != unit) {
6676 pr_warn("md: unit number in %pg is bad: %d\n",
6677 rdev0->bdev, rdev0->preferred_minor);
6678 break;
6679 }
6680
6681 mddev = md_alloc(dev, NULL);
6682 if (IS_ERR(mddev))
6683 break;
6684
6685 if (mddev_suspend_and_lock(mddev))
6686 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6687 else if (mddev->raid_disks || mddev->major_version
6688 || !list_empty(&mddev->disks)) {
6689 pr_warn("md: %s already running, cannot run %pg\n",
6690 mdname(mddev), rdev0->bdev);
6691 mddev_unlock_and_resume(mddev);
6692 } else {
6693 pr_debug("md: created %s\n", mdname(mddev));
6694 mddev->persistent = 1;
6695 rdev_for_each_list(rdev, tmp, &candidates) {
6696 list_del_init(&rdev->same_set);
6697 if (bind_rdev_to_array(rdev, mddev))
6698 export_rdev(rdev, mddev);
6699 }
6700 autorun_array(mddev);
6701 mddev_unlock_and_resume(mddev);
6702 }
6703 /* on success, candidates will be empty, on error
6704 * it won't...
6705 */
6706 rdev_for_each_list(rdev, tmp, &candidates) {
6707 list_del_init(&rdev->same_set);
6708 export_rdev(rdev, mddev);
6709 }
6710 mddev_put(mddev);
6711 }
6712 pr_info("md: ... autorun DONE.\n");
6713 }
6714 #endif /* !MODULE */
6715
get_version(void __user * arg)6716 static int get_version(void __user *arg)
6717 {
6718 mdu_version_t ver;
6719
6720 ver.major = MD_MAJOR_VERSION;
6721 ver.minor = MD_MINOR_VERSION;
6722 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6723
6724 if (copy_to_user(arg, &ver, sizeof(ver)))
6725 return -EFAULT;
6726
6727 return 0;
6728 }
6729
get_array_info(struct mddev * mddev,void __user * arg)6730 static int get_array_info(struct mddev *mddev, void __user *arg)
6731 {
6732 mdu_array_info_t info;
6733 int nr,working,insync,failed,spare;
6734 struct md_rdev *rdev;
6735
6736 nr = working = insync = failed = spare = 0;
6737 rcu_read_lock();
6738 rdev_for_each_rcu(rdev, mddev) {
6739 nr++;
6740 if (test_bit(Faulty, &rdev->flags))
6741 failed++;
6742 else {
6743 working++;
6744 if (test_bit(In_sync, &rdev->flags))
6745 insync++;
6746 else if (test_bit(Journal, &rdev->flags))
6747 /* TODO: add journal count to md_u.h */
6748 ;
6749 else
6750 spare++;
6751 }
6752 }
6753 rcu_read_unlock();
6754
6755 info.major_version = mddev->major_version;
6756 info.minor_version = mddev->minor_version;
6757 info.patch_version = MD_PATCHLEVEL_VERSION;
6758 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6759 info.level = mddev->level;
6760 info.size = mddev->dev_sectors / 2;
6761 if (info.size != mddev->dev_sectors / 2) /* overflow */
6762 info.size = -1;
6763 info.nr_disks = nr;
6764 info.raid_disks = mddev->raid_disks;
6765 info.md_minor = mddev->md_minor;
6766 info.not_persistent= !mddev->persistent;
6767
6768 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6769 info.state = 0;
6770 if (mddev->in_sync)
6771 info.state = (1<<MD_SB_CLEAN);
6772 if (mddev->bitmap && mddev->bitmap_info.offset)
6773 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6774 if (mddev_is_clustered(mddev))
6775 info.state |= (1<<MD_SB_CLUSTERED);
6776 info.active_disks = insync;
6777 info.working_disks = working;
6778 info.failed_disks = failed;
6779 info.spare_disks = spare;
6780
6781 info.layout = mddev->layout;
6782 info.chunk_size = mddev->chunk_sectors << 9;
6783
6784 if (copy_to_user(arg, &info, sizeof(info)))
6785 return -EFAULT;
6786
6787 return 0;
6788 }
6789
get_bitmap_file(struct mddev * mddev,void __user * arg)6790 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6791 {
6792 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6793 char *ptr;
6794 int err;
6795
6796 file = kzalloc(sizeof(*file), GFP_NOIO);
6797 if (!file)
6798 return -ENOMEM;
6799
6800 err = 0;
6801 spin_lock(&mddev->lock);
6802 /* bitmap enabled */
6803 if (mddev->bitmap_info.file) {
6804 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6805 sizeof(file->pathname));
6806 if (IS_ERR(ptr))
6807 err = PTR_ERR(ptr);
6808 else
6809 memmove(file->pathname, ptr,
6810 sizeof(file->pathname)-(ptr-file->pathname));
6811 }
6812 spin_unlock(&mddev->lock);
6813
6814 if (err == 0 &&
6815 copy_to_user(arg, file, sizeof(*file)))
6816 err = -EFAULT;
6817
6818 kfree(file);
6819 return err;
6820 }
6821
get_disk_info(struct mddev * mddev,void __user * arg)6822 static int get_disk_info(struct mddev *mddev, void __user * arg)
6823 {
6824 mdu_disk_info_t info;
6825 struct md_rdev *rdev;
6826
6827 if (copy_from_user(&info, arg, sizeof(info)))
6828 return -EFAULT;
6829
6830 rcu_read_lock();
6831 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6832 if (rdev) {
6833 info.major = MAJOR(rdev->bdev->bd_dev);
6834 info.minor = MINOR(rdev->bdev->bd_dev);
6835 info.raid_disk = rdev->raid_disk;
6836 info.state = 0;
6837 if (test_bit(Faulty, &rdev->flags))
6838 info.state |= (1<<MD_DISK_FAULTY);
6839 else if (test_bit(In_sync, &rdev->flags)) {
6840 info.state |= (1<<MD_DISK_ACTIVE);
6841 info.state |= (1<<MD_DISK_SYNC);
6842 }
6843 if (test_bit(Journal, &rdev->flags))
6844 info.state |= (1<<MD_DISK_JOURNAL);
6845 if (test_bit(WriteMostly, &rdev->flags))
6846 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6847 if (test_bit(FailFast, &rdev->flags))
6848 info.state |= (1<<MD_DISK_FAILFAST);
6849 } else {
6850 info.major = info.minor = 0;
6851 info.raid_disk = -1;
6852 info.state = (1<<MD_DISK_REMOVED);
6853 }
6854 rcu_read_unlock();
6855
6856 if (copy_to_user(arg, &info, sizeof(info)))
6857 return -EFAULT;
6858
6859 return 0;
6860 }
6861
md_add_new_disk(struct mddev * mddev,struct mdu_disk_info_s * info)6862 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6863 {
6864 struct md_rdev *rdev;
6865 dev_t dev = MKDEV(info->major,info->minor);
6866
6867 if (mddev_is_clustered(mddev) &&
6868 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6869 pr_warn("%s: Cannot add to clustered mddev.\n",
6870 mdname(mddev));
6871 return -EINVAL;
6872 }
6873
6874 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6875 return -EOVERFLOW;
6876
6877 if (!mddev->raid_disks) {
6878 int err;
6879 /* expecting a device which has a superblock */
6880 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6881 if (IS_ERR(rdev)) {
6882 pr_warn("md: md_import_device returned %ld\n",
6883 PTR_ERR(rdev));
6884 return PTR_ERR(rdev);
6885 }
6886 if (!list_empty(&mddev->disks)) {
6887 struct md_rdev *rdev0
6888 = list_entry(mddev->disks.next,
6889 struct md_rdev, same_set);
6890 err = super_types[mddev->major_version]
6891 .load_super(rdev, rdev0, mddev->minor_version);
6892 if (err < 0) {
6893 pr_warn("md: %pg has different UUID to %pg\n",
6894 rdev->bdev,
6895 rdev0->bdev);
6896 export_rdev(rdev, mddev);
6897 return -EINVAL;
6898 }
6899 }
6900 err = bind_rdev_to_array(rdev, mddev);
6901 if (err)
6902 export_rdev(rdev, mddev);
6903 return err;
6904 }
6905
6906 /*
6907 * md_add_new_disk can be used once the array is assembled
6908 * to add "hot spares". They must already have a superblock
6909 * written
6910 */
6911 if (mddev->pers) {
6912 int err;
6913 if (!mddev->pers->hot_add_disk) {
6914 pr_warn("%s: personality does not support diskops!\n",
6915 mdname(mddev));
6916 return -EINVAL;
6917 }
6918 if (mddev->persistent)
6919 rdev = md_import_device(dev, mddev->major_version,
6920 mddev->minor_version);
6921 else
6922 rdev = md_import_device(dev, -1, -1);
6923 if (IS_ERR(rdev)) {
6924 pr_warn("md: md_import_device returned %ld\n",
6925 PTR_ERR(rdev));
6926 return PTR_ERR(rdev);
6927 }
6928 /* set saved_raid_disk if appropriate */
6929 if (!mddev->persistent) {
6930 if (info->state & (1<<MD_DISK_SYNC) &&
6931 info->raid_disk < mddev->raid_disks) {
6932 rdev->raid_disk = info->raid_disk;
6933 clear_bit(Bitmap_sync, &rdev->flags);
6934 } else
6935 rdev->raid_disk = -1;
6936 rdev->saved_raid_disk = rdev->raid_disk;
6937 } else
6938 super_types[mddev->major_version].
6939 validate_super(mddev, NULL/*freshest*/, rdev);
6940 if ((info->state & (1<<MD_DISK_SYNC)) &&
6941 rdev->raid_disk != info->raid_disk) {
6942 /* This was a hot-add request, but events doesn't
6943 * match, so reject it.
6944 */
6945 export_rdev(rdev, mddev);
6946 return -EINVAL;
6947 }
6948
6949 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6950 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6951 set_bit(WriteMostly, &rdev->flags);
6952 else
6953 clear_bit(WriteMostly, &rdev->flags);
6954 if (info->state & (1<<MD_DISK_FAILFAST))
6955 set_bit(FailFast, &rdev->flags);
6956 else
6957 clear_bit(FailFast, &rdev->flags);
6958
6959 if (info->state & (1<<MD_DISK_JOURNAL)) {
6960 struct md_rdev *rdev2;
6961 bool has_journal = false;
6962
6963 /* make sure no existing journal disk */
6964 rdev_for_each(rdev2, mddev) {
6965 if (test_bit(Journal, &rdev2->flags)) {
6966 has_journal = true;
6967 break;
6968 }
6969 }
6970 if (has_journal || mddev->bitmap) {
6971 export_rdev(rdev, mddev);
6972 return -EBUSY;
6973 }
6974 set_bit(Journal, &rdev->flags);
6975 }
6976 /*
6977 * check whether the device shows up in other nodes
6978 */
6979 if (mddev_is_clustered(mddev)) {
6980 if (info->state & (1 << MD_DISK_CANDIDATE))
6981 set_bit(Candidate, &rdev->flags);
6982 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6983 /* --add initiated by this node */
6984 err = mddev->cluster_ops->add_new_disk(mddev, rdev);
6985 if (err) {
6986 export_rdev(rdev, mddev);
6987 return err;
6988 }
6989 }
6990 }
6991
6992 rdev->raid_disk = -1;
6993 err = bind_rdev_to_array(rdev, mddev);
6994
6995 if (err)
6996 export_rdev(rdev, mddev);
6997
6998 if (mddev_is_clustered(mddev)) {
6999 if (info->state & (1 << MD_DISK_CANDIDATE)) {
7000 if (!err) {
7001 err = mddev->cluster_ops->new_disk_ack(
7002 mddev, err == 0);
7003 if (err)
7004 md_kick_rdev_from_array(rdev);
7005 }
7006 } else {
7007 if (err)
7008 mddev->cluster_ops->add_new_disk_cancel(mddev);
7009 else
7010 err = add_bound_rdev(rdev);
7011 }
7012
7013 } else if (!err)
7014 err = add_bound_rdev(rdev);
7015
7016 return err;
7017 }
7018
7019 /* otherwise, md_add_new_disk is only allowed
7020 * for major_version==0 superblocks
7021 */
7022 if (mddev->major_version != 0) {
7023 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
7024 return -EINVAL;
7025 }
7026
7027 if (!(info->state & (1<<MD_DISK_FAULTY))) {
7028 int err;
7029 rdev = md_import_device(dev, -1, 0);
7030 if (IS_ERR(rdev)) {
7031 pr_warn("md: error, md_import_device() returned %ld\n",
7032 PTR_ERR(rdev));
7033 return PTR_ERR(rdev);
7034 }
7035 rdev->desc_nr = info->number;
7036 if (info->raid_disk < mddev->raid_disks)
7037 rdev->raid_disk = info->raid_disk;
7038 else
7039 rdev->raid_disk = -1;
7040
7041 if (rdev->raid_disk < mddev->raid_disks)
7042 if (info->state & (1<<MD_DISK_SYNC))
7043 set_bit(In_sync, &rdev->flags);
7044
7045 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
7046 set_bit(WriteMostly, &rdev->flags);
7047 if (info->state & (1<<MD_DISK_FAILFAST))
7048 set_bit(FailFast, &rdev->flags);
7049
7050 if (!mddev->persistent) {
7051 pr_debug("md: nonpersistent superblock ...\n");
7052 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7053 } else
7054 rdev->sb_start = calc_dev_sboffset(rdev);
7055 rdev->sectors = rdev->sb_start;
7056
7057 err = bind_rdev_to_array(rdev, mddev);
7058 if (err) {
7059 export_rdev(rdev, mddev);
7060 return err;
7061 }
7062 }
7063
7064 return 0;
7065 }
7066
hot_remove_disk(struct mddev * mddev,dev_t dev)7067 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
7068 {
7069 struct md_rdev *rdev;
7070
7071 if (!mddev->pers)
7072 return -ENODEV;
7073
7074 rdev = find_rdev(mddev, dev);
7075 if (!rdev)
7076 return -ENXIO;
7077
7078 if (rdev->raid_disk < 0)
7079 goto kick_rdev;
7080
7081 clear_bit(Blocked, &rdev->flags);
7082 remove_and_add_spares(mddev, rdev);
7083
7084 if (rdev->raid_disk >= 0)
7085 goto busy;
7086
7087 kick_rdev:
7088 if (mddev_is_clustered(mddev) &&
7089 mddev->cluster_ops->remove_disk(mddev, rdev))
7090 goto busy;
7091
7092 md_kick_rdev_from_array(rdev);
7093 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7094 if (!mddev->thread)
7095 md_update_sb(mddev, 1);
7096 md_new_event();
7097
7098 return 0;
7099 busy:
7100 pr_debug("md: cannot remove active disk %pg from %s ...\n",
7101 rdev->bdev, mdname(mddev));
7102 return -EBUSY;
7103 }
7104
hot_add_disk(struct mddev * mddev,dev_t dev)7105 static int hot_add_disk(struct mddev *mddev, dev_t dev)
7106 {
7107 int err;
7108 struct md_rdev *rdev;
7109
7110 if (!mddev->pers)
7111 return -ENODEV;
7112
7113 if (mddev->major_version != 0) {
7114 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
7115 mdname(mddev));
7116 return -EINVAL;
7117 }
7118 if (!mddev->pers->hot_add_disk) {
7119 pr_warn("%s: personality does not support diskops!\n",
7120 mdname(mddev));
7121 return -EINVAL;
7122 }
7123
7124 rdev = md_import_device(dev, -1, 0);
7125 if (IS_ERR(rdev)) {
7126 pr_warn("md: error, md_import_device() returned %ld\n",
7127 PTR_ERR(rdev));
7128 return -EINVAL;
7129 }
7130
7131 if (mddev->persistent)
7132 rdev->sb_start = calc_dev_sboffset(rdev);
7133 else
7134 rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7135
7136 rdev->sectors = rdev->sb_start;
7137
7138 if (test_bit(Faulty, &rdev->flags)) {
7139 pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
7140 rdev->bdev, mdname(mddev));
7141 err = -EINVAL;
7142 goto abort_export;
7143 }
7144
7145 clear_bit(In_sync, &rdev->flags);
7146 rdev->desc_nr = -1;
7147 rdev->saved_raid_disk = -1;
7148 err = bind_rdev_to_array(rdev, mddev);
7149 if (err)
7150 goto abort_export;
7151
7152 /*
7153 * The rest should better be atomic, we can have disk failures
7154 * noticed in interrupt contexts ...
7155 */
7156
7157 rdev->raid_disk = -1;
7158
7159 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7160 if (!mddev->thread)
7161 md_update_sb(mddev, 1);
7162 /*
7163 * Kick recovery, maybe this spare has to be added to the
7164 * array immediately.
7165 */
7166 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7167 md_new_event();
7168 return 0;
7169
7170 abort_export:
7171 export_rdev(rdev, mddev);
7172 return err;
7173 }
7174
set_bitmap_file(struct mddev * mddev,int fd)7175 static int set_bitmap_file(struct mddev *mddev, int fd)
7176 {
7177 int err = 0;
7178
7179 if (mddev->pers) {
7180 if (!mddev->pers->quiesce || !mddev->thread)
7181 return -EBUSY;
7182 if (mddev->recovery || mddev->sync_thread)
7183 return -EBUSY;
7184 /* we should be able to change the bitmap.. */
7185 }
7186
7187 if (fd >= 0) {
7188 struct inode *inode;
7189 struct file *f;
7190
7191 if (mddev->bitmap || mddev->bitmap_info.file)
7192 return -EEXIST; /* cannot add when bitmap is present */
7193
7194 if (!IS_ENABLED(CONFIG_MD_BITMAP_FILE)) {
7195 pr_warn("%s: bitmap files not supported by this kernel\n",
7196 mdname(mddev));
7197 return -EINVAL;
7198 }
7199 pr_warn("%s: using deprecated bitmap file support\n",
7200 mdname(mddev));
7201
7202 f = fget(fd);
7203
7204 if (f == NULL) {
7205 pr_warn("%s: error: failed to get bitmap file\n",
7206 mdname(mddev));
7207 return -EBADF;
7208 }
7209
7210 inode = f->f_mapping->host;
7211 if (!S_ISREG(inode->i_mode)) {
7212 pr_warn("%s: error: bitmap file must be a regular file\n",
7213 mdname(mddev));
7214 err = -EBADF;
7215 } else if (!(f->f_mode & FMODE_WRITE)) {
7216 pr_warn("%s: error: bitmap file must open for write\n",
7217 mdname(mddev));
7218 err = -EBADF;
7219 } else if (atomic_read(&inode->i_writecount) != 1) {
7220 pr_warn("%s: error: bitmap file is already in use\n",
7221 mdname(mddev));
7222 err = -EBUSY;
7223 }
7224 if (err) {
7225 fput(f);
7226 return err;
7227 }
7228 mddev->bitmap_info.file = f;
7229 mddev->bitmap_info.offset = 0; /* file overrides offset */
7230 } else if (mddev->bitmap == NULL)
7231 return -ENOENT; /* cannot remove what isn't there */
7232 err = 0;
7233 if (mddev->pers) {
7234 if (fd >= 0) {
7235 err = mddev->bitmap_ops->create(mddev, -1);
7236 if (!err)
7237 err = mddev->bitmap_ops->load(mddev);
7238
7239 if (err) {
7240 mddev->bitmap_ops->destroy(mddev);
7241 fd = -1;
7242 }
7243 } else if (fd < 0) {
7244 mddev->bitmap_ops->destroy(mddev);
7245 }
7246 }
7247
7248 if (fd < 0) {
7249 struct file *f = mddev->bitmap_info.file;
7250 if (f) {
7251 spin_lock(&mddev->lock);
7252 mddev->bitmap_info.file = NULL;
7253 spin_unlock(&mddev->lock);
7254 fput(f);
7255 }
7256 }
7257
7258 return err;
7259 }
7260
7261 /*
7262 * md_set_array_info is used two different ways
7263 * The original usage is when creating a new array.
7264 * In this usage, raid_disks is > 0 and it together with
7265 * level, size, not_persistent,layout,chunksize determine the
7266 * shape of the array.
7267 * This will always create an array with a type-0.90.0 superblock.
7268 * The newer usage is when assembling an array.
7269 * In this case raid_disks will be 0, and the major_version field is
7270 * use to determine which style super-blocks are to be found on the devices.
7271 * The minor and patch _version numbers are also kept incase the
7272 * super_block handler wishes to interpret them.
7273 */
md_set_array_info(struct mddev * mddev,struct mdu_array_info_s * info)7274 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7275 {
7276 if (info->raid_disks == 0) {
7277 /* just setting version number for superblock loading */
7278 if (info->major_version < 0 ||
7279 info->major_version >= ARRAY_SIZE(super_types) ||
7280 super_types[info->major_version].name == NULL) {
7281 /* maybe try to auto-load a module? */
7282 pr_warn("md: superblock version %d not known\n",
7283 info->major_version);
7284 return -EINVAL;
7285 }
7286 mddev->major_version = info->major_version;
7287 mddev->minor_version = info->minor_version;
7288 mddev->patch_version = info->patch_version;
7289 mddev->persistent = !info->not_persistent;
7290 /* ensure mddev_put doesn't delete this now that there
7291 * is some minimal configuration.
7292 */
7293 mddev->ctime = ktime_get_real_seconds();
7294 return 0;
7295 }
7296 mddev->major_version = MD_MAJOR_VERSION;
7297 mddev->minor_version = MD_MINOR_VERSION;
7298 mddev->patch_version = MD_PATCHLEVEL_VERSION;
7299 mddev->ctime = ktime_get_real_seconds();
7300
7301 mddev->level = info->level;
7302 mddev->clevel[0] = 0;
7303 mddev->dev_sectors = 2 * (sector_t)info->size;
7304 mddev->raid_disks = info->raid_disks;
7305 /* don't set md_minor, it is determined by which /dev/md* was
7306 * openned
7307 */
7308 if (info->state & (1<<MD_SB_CLEAN))
7309 mddev->recovery_cp = MaxSector;
7310 else
7311 mddev->recovery_cp = 0;
7312 mddev->persistent = ! info->not_persistent;
7313 mddev->external = 0;
7314
7315 mddev->layout = info->layout;
7316 if (mddev->level == 0)
7317 /* Cannot trust RAID0 layout info here */
7318 mddev->layout = -1;
7319 mddev->chunk_sectors = info->chunk_size >> 9;
7320
7321 if (mddev->persistent) {
7322 mddev->max_disks = MD_SB_DISKS;
7323 mddev->flags = 0;
7324 mddev->sb_flags = 0;
7325 }
7326 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7327
7328 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7329 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7330 mddev->bitmap_info.offset = 0;
7331
7332 mddev->reshape_position = MaxSector;
7333
7334 /*
7335 * Generate a 128 bit UUID
7336 */
7337 get_random_bytes(mddev->uuid, 16);
7338
7339 mddev->new_level = mddev->level;
7340 mddev->new_chunk_sectors = mddev->chunk_sectors;
7341 mddev->new_layout = mddev->layout;
7342 mddev->delta_disks = 0;
7343 mddev->reshape_backwards = 0;
7344
7345 return 0;
7346 }
7347
md_set_array_sectors(struct mddev * mddev,sector_t array_sectors)7348 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7349 {
7350 lockdep_assert_held(&mddev->reconfig_mutex);
7351
7352 if (mddev->external_size)
7353 return;
7354
7355 mddev->array_sectors = array_sectors;
7356 }
7357 EXPORT_SYMBOL(md_set_array_sectors);
7358
update_size(struct mddev * mddev,sector_t num_sectors)7359 static int update_size(struct mddev *mddev, sector_t num_sectors)
7360 {
7361 struct md_rdev *rdev;
7362 int rv;
7363 int fit = (num_sectors == 0);
7364 sector_t old_dev_sectors = mddev->dev_sectors;
7365
7366 if (mddev->pers->resize == NULL)
7367 return -EINVAL;
7368 /* The "num_sectors" is the number of sectors of each device that
7369 * is used. This can only make sense for arrays with redundancy.
7370 * linear and raid0 always use whatever space is available. We can only
7371 * consider changing this number if no resync or reconstruction is
7372 * happening, and if the new size is acceptable. It must fit before the
7373 * sb_start or, if that is <data_offset, it must fit before the size
7374 * of each device. If num_sectors is zero, we find the largest size
7375 * that fits.
7376 */
7377 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
7378 return -EBUSY;
7379 if (!md_is_rdwr(mddev))
7380 return -EROFS;
7381
7382 rdev_for_each(rdev, mddev) {
7383 sector_t avail = rdev->sectors;
7384
7385 if (fit && (num_sectors == 0 || num_sectors > avail))
7386 num_sectors = avail;
7387 if (avail < num_sectors)
7388 return -ENOSPC;
7389 }
7390 rv = mddev->pers->resize(mddev, num_sectors);
7391 if (!rv) {
7392 if (mddev_is_clustered(mddev))
7393 mddev->cluster_ops->update_size(mddev, old_dev_sectors);
7394 else if (!mddev_is_dm(mddev))
7395 set_capacity_and_notify(mddev->gendisk,
7396 mddev->array_sectors);
7397 }
7398 return rv;
7399 }
7400
update_raid_disks(struct mddev * mddev,int raid_disks)7401 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7402 {
7403 int rv;
7404 struct md_rdev *rdev;
7405 /* change the number of raid disks */
7406 if (mddev->pers->check_reshape == NULL)
7407 return -EINVAL;
7408 if (!md_is_rdwr(mddev))
7409 return -EROFS;
7410 if (raid_disks <= 0 ||
7411 (mddev->max_disks && raid_disks >= mddev->max_disks))
7412 return -EINVAL;
7413 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7414 test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7415 mddev->reshape_position != MaxSector)
7416 return -EBUSY;
7417
7418 rdev_for_each(rdev, mddev) {
7419 if (mddev->raid_disks < raid_disks &&
7420 rdev->data_offset < rdev->new_data_offset)
7421 return -EINVAL;
7422 if (mddev->raid_disks > raid_disks &&
7423 rdev->data_offset > rdev->new_data_offset)
7424 return -EINVAL;
7425 }
7426
7427 mddev->delta_disks = raid_disks - mddev->raid_disks;
7428 if (mddev->delta_disks < 0)
7429 mddev->reshape_backwards = 1;
7430 else if (mddev->delta_disks > 0)
7431 mddev->reshape_backwards = 0;
7432
7433 rv = mddev->pers->check_reshape(mddev);
7434 if (rv < 0) {
7435 mddev->delta_disks = 0;
7436 mddev->reshape_backwards = 0;
7437 }
7438 return rv;
7439 }
7440
get_cluster_ops(struct mddev * mddev)7441 static int get_cluster_ops(struct mddev *mddev)
7442 {
7443 xa_lock(&md_submodule);
7444 mddev->cluster_ops = xa_load(&md_submodule, ID_CLUSTER);
7445 if (mddev->cluster_ops &&
7446 !try_module_get(mddev->cluster_ops->head.owner))
7447 mddev->cluster_ops = NULL;
7448 xa_unlock(&md_submodule);
7449
7450 return mddev->cluster_ops == NULL ? -ENOENT : 0;
7451 }
7452
put_cluster_ops(struct mddev * mddev)7453 static void put_cluster_ops(struct mddev *mddev)
7454 {
7455 if (!mddev->cluster_ops)
7456 return;
7457
7458 mddev->cluster_ops->leave(mddev);
7459 module_put(mddev->cluster_ops->head.owner);
7460 mddev->cluster_ops = NULL;
7461 }
7462
7463 /*
7464 * update_array_info is used to change the configuration of an
7465 * on-line array.
7466 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7467 * fields in the info are checked against the array.
7468 * Any differences that cannot be handled will cause an error.
7469 * Normally, only one change can be managed at a time.
7470 */
update_array_info(struct mddev * mddev,mdu_array_info_t * info)7471 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7472 {
7473 int rv = 0;
7474 int cnt = 0;
7475 int state = 0;
7476
7477 /* calculate expected state,ignoring low bits */
7478 if (mddev->bitmap && mddev->bitmap_info.offset)
7479 state |= (1 << MD_SB_BITMAP_PRESENT);
7480
7481 if (mddev->major_version != info->major_version ||
7482 mddev->minor_version != info->minor_version ||
7483 /* mddev->patch_version != info->patch_version || */
7484 mddev->ctime != info->ctime ||
7485 mddev->level != info->level ||
7486 /* mddev->layout != info->layout || */
7487 mddev->persistent != !info->not_persistent ||
7488 mddev->chunk_sectors != info->chunk_size >> 9 ||
7489 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7490 ((state^info->state) & 0xfffffe00)
7491 )
7492 return -EINVAL;
7493 /* Check there is only one change */
7494 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7495 cnt++;
7496 if (mddev->raid_disks != info->raid_disks)
7497 cnt++;
7498 if (mddev->layout != info->layout)
7499 cnt++;
7500 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7501 cnt++;
7502 if (cnt == 0)
7503 return 0;
7504 if (cnt > 1)
7505 return -EINVAL;
7506
7507 if (mddev->layout != info->layout) {
7508 /* Change layout
7509 * we don't need to do anything at the md level, the
7510 * personality will take care of it all.
7511 */
7512 if (mddev->pers->check_reshape == NULL)
7513 return -EINVAL;
7514 else {
7515 mddev->new_layout = info->layout;
7516 rv = mddev->pers->check_reshape(mddev);
7517 if (rv)
7518 mddev->new_layout = mddev->layout;
7519 return rv;
7520 }
7521 }
7522 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7523 rv = update_size(mddev, (sector_t)info->size * 2);
7524
7525 if (mddev->raid_disks != info->raid_disks)
7526 rv = update_raid_disks(mddev, info->raid_disks);
7527
7528 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7529 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7530 rv = -EINVAL;
7531 goto err;
7532 }
7533 if (mddev->recovery || mddev->sync_thread) {
7534 rv = -EBUSY;
7535 goto err;
7536 }
7537 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7538 /* add the bitmap */
7539 if (mddev->bitmap) {
7540 rv = -EEXIST;
7541 goto err;
7542 }
7543 if (mddev->bitmap_info.default_offset == 0) {
7544 rv = -EINVAL;
7545 goto err;
7546 }
7547 mddev->bitmap_info.offset =
7548 mddev->bitmap_info.default_offset;
7549 mddev->bitmap_info.space =
7550 mddev->bitmap_info.default_space;
7551 rv = mddev->bitmap_ops->create(mddev, -1);
7552 if (!rv)
7553 rv = mddev->bitmap_ops->load(mddev);
7554
7555 if (rv)
7556 mddev->bitmap_ops->destroy(mddev);
7557 } else {
7558 struct md_bitmap_stats stats;
7559
7560 rv = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
7561 if (rv)
7562 goto err;
7563
7564 if (stats.file) {
7565 rv = -EINVAL;
7566 goto err;
7567 }
7568
7569 if (mddev->bitmap_info.nodes) {
7570 /* hold PW on all the bitmap lock */
7571 if (mddev->cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7572 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7573 rv = -EPERM;
7574 mddev->cluster_ops->unlock_all_bitmaps(mddev);
7575 goto err;
7576 }
7577
7578 mddev->bitmap_info.nodes = 0;
7579 put_cluster_ops(mddev);
7580 mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7581 }
7582 mddev->bitmap_ops->destroy(mddev);
7583 mddev->bitmap_info.offset = 0;
7584 }
7585 }
7586 md_update_sb(mddev, 1);
7587 return rv;
7588 err:
7589 return rv;
7590 }
7591
set_disk_faulty(struct mddev * mddev,dev_t dev)7592 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7593 {
7594 struct md_rdev *rdev;
7595 int err = 0;
7596
7597 if (mddev->pers == NULL)
7598 return -ENODEV;
7599
7600 rcu_read_lock();
7601 rdev = md_find_rdev_rcu(mddev, dev);
7602 if (!rdev)
7603 err = -ENODEV;
7604 else {
7605 md_error(mddev, rdev);
7606 if (test_bit(MD_BROKEN, &mddev->flags))
7607 err = -EBUSY;
7608 }
7609 rcu_read_unlock();
7610 return err;
7611 }
7612
7613 /*
7614 * We have a problem here : there is no easy way to give a CHS
7615 * virtual geometry. We currently pretend that we have a 2 heads
7616 * 4 sectors (with a BIG number of cylinders...). This drives
7617 * dosfs just mad... ;-)
7618 */
md_getgeo(struct block_device * bdev,struct hd_geometry * geo)7619 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7620 {
7621 struct mddev *mddev = bdev->bd_disk->private_data;
7622
7623 geo->heads = 2;
7624 geo->sectors = 4;
7625 geo->cylinders = mddev->array_sectors / 8;
7626 return 0;
7627 }
7628
md_ioctl_valid(unsigned int cmd)7629 static inline int md_ioctl_valid(unsigned int cmd)
7630 {
7631 switch (cmd) {
7632 case GET_ARRAY_INFO:
7633 case GET_DISK_INFO:
7634 case RAID_VERSION:
7635 return 0;
7636 case ADD_NEW_DISK:
7637 case GET_BITMAP_FILE:
7638 case HOT_ADD_DISK:
7639 case HOT_REMOVE_DISK:
7640 case RESTART_ARRAY_RW:
7641 case RUN_ARRAY:
7642 case SET_ARRAY_INFO:
7643 case SET_BITMAP_FILE:
7644 case SET_DISK_FAULTY:
7645 case STOP_ARRAY:
7646 case STOP_ARRAY_RO:
7647 case CLUSTERED_DISK_NACK:
7648 if (!capable(CAP_SYS_ADMIN))
7649 return -EACCES;
7650 return 0;
7651 default:
7652 return -ENOTTY;
7653 }
7654 }
7655
md_ioctl_need_suspend(unsigned int cmd)7656 static bool md_ioctl_need_suspend(unsigned int cmd)
7657 {
7658 switch (cmd) {
7659 case ADD_NEW_DISK:
7660 case HOT_ADD_DISK:
7661 case HOT_REMOVE_DISK:
7662 case SET_BITMAP_FILE:
7663 case SET_ARRAY_INFO:
7664 return true;
7665 default:
7666 return false;
7667 }
7668 }
7669
__md_set_array_info(struct mddev * mddev,void __user * argp)7670 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7671 {
7672 mdu_array_info_t info;
7673 int err;
7674
7675 if (!argp)
7676 memset(&info, 0, sizeof(info));
7677 else if (copy_from_user(&info, argp, sizeof(info)))
7678 return -EFAULT;
7679
7680 if (mddev->pers) {
7681 err = update_array_info(mddev, &info);
7682 if (err)
7683 pr_warn("md: couldn't update array info. %d\n", err);
7684 return err;
7685 }
7686
7687 if (!list_empty(&mddev->disks)) {
7688 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7689 return -EBUSY;
7690 }
7691
7692 if (mddev->raid_disks) {
7693 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7694 return -EBUSY;
7695 }
7696
7697 err = md_set_array_info(mddev, &info);
7698 if (err)
7699 pr_warn("md: couldn't set array info. %d\n", err);
7700
7701 return err;
7702 }
7703
md_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)7704 static int md_ioctl(struct block_device *bdev, blk_mode_t mode,
7705 unsigned int cmd, unsigned long arg)
7706 {
7707 int err = 0;
7708 void __user *argp = (void __user *)arg;
7709 struct mddev *mddev = NULL;
7710
7711 err = md_ioctl_valid(cmd);
7712 if (err)
7713 return err;
7714
7715 /*
7716 * Commands dealing with the RAID driver but not any
7717 * particular array:
7718 */
7719 if (cmd == RAID_VERSION)
7720 return get_version(argp);
7721
7722 /*
7723 * Commands creating/starting a new array:
7724 */
7725
7726 mddev = bdev->bd_disk->private_data;
7727
7728 /* Some actions do not requires the mutex */
7729 switch (cmd) {
7730 case GET_ARRAY_INFO:
7731 if (!mddev->raid_disks && !mddev->external)
7732 return -ENODEV;
7733 return get_array_info(mddev, argp);
7734
7735 case GET_DISK_INFO:
7736 if (!mddev->raid_disks && !mddev->external)
7737 return -ENODEV;
7738 return get_disk_info(mddev, argp);
7739
7740 case SET_DISK_FAULTY:
7741 return set_disk_faulty(mddev, new_decode_dev(arg));
7742
7743 case GET_BITMAP_FILE:
7744 return get_bitmap_file(mddev, argp);
7745 }
7746
7747 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7748 /* Need to flush page cache, and ensure no-one else opens
7749 * and writes
7750 */
7751 err = mddev_set_closing_and_sync_blockdev(mddev, 1);
7752 if (err)
7753 return err;
7754 }
7755
7756 if (!md_is_rdwr(mddev))
7757 flush_work(&mddev->sync_work);
7758
7759 err = md_ioctl_need_suspend(cmd) ? mddev_suspend_and_lock(mddev) :
7760 mddev_lock(mddev);
7761 if (err) {
7762 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7763 err, cmd);
7764 goto out;
7765 }
7766
7767 if (cmd == SET_ARRAY_INFO) {
7768 err = __md_set_array_info(mddev, argp);
7769 goto unlock;
7770 }
7771
7772 /*
7773 * Commands querying/configuring an existing array:
7774 */
7775 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7776 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7777 if ((!mddev->raid_disks && !mddev->external)
7778 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7779 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7780 && cmd != GET_BITMAP_FILE) {
7781 err = -ENODEV;
7782 goto unlock;
7783 }
7784
7785 /*
7786 * Commands even a read-only array can execute:
7787 */
7788 switch (cmd) {
7789 case RESTART_ARRAY_RW:
7790 err = restart_array(mddev);
7791 goto unlock;
7792
7793 case STOP_ARRAY:
7794 err = do_md_stop(mddev, 0);
7795 goto unlock;
7796
7797 case STOP_ARRAY_RO:
7798 if (mddev->pers)
7799 err = md_set_readonly(mddev);
7800 goto unlock;
7801
7802 case HOT_REMOVE_DISK:
7803 err = hot_remove_disk(mddev, new_decode_dev(arg));
7804 goto unlock;
7805
7806 case ADD_NEW_DISK:
7807 /* We can support ADD_NEW_DISK on read-only arrays
7808 * only if we are re-adding a preexisting device.
7809 * So require mddev->pers and MD_DISK_SYNC.
7810 */
7811 if (mddev->pers) {
7812 mdu_disk_info_t info;
7813 if (copy_from_user(&info, argp, sizeof(info)))
7814 err = -EFAULT;
7815 else if (!(info.state & (1<<MD_DISK_SYNC)))
7816 /* Need to clear read-only for this */
7817 break;
7818 else
7819 err = md_add_new_disk(mddev, &info);
7820 goto unlock;
7821 }
7822 break;
7823 }
7824
7825 /*
7826 * The remaining ioctls are changing the state of the
7827 * superblock, so we do not allow them on read-only arrays.
7828 */
7829 if (!md_is_rdwr(mddev) && mddev->pers) {
7830 if (mddev->ro != MD_AUTO_READ) {
7831 err = -EROFS;
7832 goto unlock;
7833 }
7834 mddev->ro = MD_RDWR;
7835 sysfs_notify_dirent_safe(mddev->sysfs_state);
7836 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7837 /* mddev_unlock will wake thread */
7838 /* If a device failed while we were read-only, we
7839 * need to make sure the metadata is updated now.
7840 */
7841 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7842 mddev_unlock(mddev);
7843 wait_event(mddev->sb_wait,
7844 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7845 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7846 mddev_lock_nointr(mddev);
7847 }
7848 }
7849
7850 switch (cmd) {
7851 case ADD_NEW_DISK:
7852 {
7853 mdu_disk_info_t info;
7854 if (copy_from_user(&info, argp, sizeof(info)))
7855 err = -EFAULT;
7856 else
7857 err = md_add_new_disk(mddev, &info);
7858 goto unlock;
7859 }
7860
7861 case CLUSTERED_DISK_NACK:
7862 if (mddev_is_clustered(mddev))
7863 mddev->cluster_ops->new_disk_ack(mddev, false);
7864 else
7865 err = -EINVAL;
7866 goto unlock;
7867
7868 case HOT_ADD_DISK:
7869 err = hot_add_disk(mddev, new_decode_dev(arg));
7870 goto unlock;
7871
7872 case RUN_ARRAY:
7873 err = do_md_run(mddev);
7874 goto unlock;
7875
7876 case SET_BITMAP_FILE:
7877 err = set_bitmap_file(mddev, (int)arg);
7878 goto unlock;
7879
7880 default:
7881 err = -EINVAL;
7882 goto unlock;
7883 }
7884
7885 unlock:
7886 if (mddev->hold_active == UNTIL_IOCTL &&
7887 err != -EINVAL)
7888 mddev->hold_active = 0;
7889
7890 md_ioctl_need_suspend(cmd) ? mddev_unlock_and_resume(mddev) :
7891 mddev_unlock(mddev);
7892
7893 out:
7894 if (cmd == STOP_ARRAY_RO || (err && cmd == STOP_ARRAY))
7895 clear_bit(MD_CLOSING, &mddev->flags);
7896 return err;
7897 }
7898 #ifdef CONFIG_COMPAT
md_compat_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)7899 static int md_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
7900 unsigned int cmd, unsigned long arg)
7901 {
7902 switch (cmd) {
7903 case HOT_REMOVE_DISK:
7904 case HOT_ADD_DISK:
7905 case SET_DISK_FAULTY:
7906 case SET_BITMAP_FILE:
7907 /* These take in integer arg, do not convert */
7908 break;
7909 default:
7910 arg = (unsigned long)compat_ptr(arg);
7911 break;
7912 }
7913
7914 return md_ioctl(bdev, mode, cmd, arg);
7915 }
7916 #endif /* CONFIG_COMPAT */
7917
md_set_read_only(struct block_device * bdev,bool ro)7918 static int md_set_read_only(struct block_device *bdev, bool ro)
7919 {
7920 struct mddev *mddev = bdev->bd_disk->private_data;
7921 int err;
7922
7923 err = mddev_lock(mddev);
7924 if (err)
7925 return err;
7926
7927 if (!mddev->raid_disks && !mddev->external) {
7928 err = -ENODEV;
7929 goto out_unlock;
7930 }
7931
7932 /*
7933 * Transitioning to read-auto need only happen for arrays that call
7934 * md_write_start and which are not ready for writes yet.
7935 */
7936 if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7937 err = restart_array(mddev);
7938 if (err)
7939 goto out_unlock;
7940 mddev->ro = MD_AUTO_READ;
7941 }
7942
7943 out_unlock:
7944 mddev_unlock(mddev);
7945 return err;
7946 }
7947
md_open(struct gendisk * disk,blk_mode_t mode)7948 static int md_open(struct gendisk *disk, blk_mode_t mode)
7949 {
7950 struct mddev *mddev;
7951 int err;
7952
7953 spin_lock(&all_mddevs_lock);
7954 mddev = mddev_get(disk->private_data);
7955 spin_unlock(&all_mddevs_lock);
7956 if (!mddev)
7957 return -ENODEV;
7958
7959 err = mutex_lock_interruptible(&mddev->open_mutex);
7960 if (err)
7961 goto out;
7962
7963 err = -ENODEV;
7964 if (test_bit(MD_CLOSING, &mddev->flags))
7965 goto out_unlock;
7966
7967 atomic_inc(&mddev->openers);
7968 mutex_unlock(&mddev->open_mutex);
7969
7970 disk_check_media_change(disk);
7971 return 0;
7972
7973 out_unlock:
7974 mutex_unlock(&mddev->open_mutex);
7975 out:
7976 mddev_put(mddev);
7977 return err;
7978 }
7979
md_release(struct gendisk * disk)7980 static void md_release(struct gendisk *disk)
7981 {
7982 struct mddev *mddev = disk->private_data;
7983
7984 BUG_ON(!mddev);
7985 atomic_dec(&mddev->openers);
7986 mddev_put(mddev);
7987 }
7988
md_check_events(struct gendisk * disk,unsigned int clearing)7989 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7990 {
7991 struct mddev *mddev = disk->private_data;
7992 unsigned int ret = 0;
7993
7994 if (mddev->changed)
7995 ret = DISK_EVENT_MEDIA_CHANGE;
7996 mddev->changed = 0;
7997 return ret;
7998 }
7999
md_free_disk(struct gendisk * disk)8000 static void md_free_disk(struct gendisk *disk)
8001 {
8002 struct mddev *mddev = disk->private_data;
8003
8004 mddev_free(mddev);
8005 }
8006
8007 const struct block_device_operations md_fops =
8008 {
8009 .owner = THIS_MODULE,
8010 .submit_bio = md_submit_bio,
8011 .open = md_open,
8012 .release = md_release,
8013 .ioctl = md_ioctl,
8014 #ifdef CONFIG_COMPAT
8015 .compat_ioctl = md_compat_ioctl,
8016 #endif
8017 .getgeo = md_getgeo,
8018 .check_events = md_check_events,
8019 .set_read_only = md_set_read_only,
8020 .free_disk = md_free_disk,
8021 };
8022
md_thread(void * arg)8023 static int md_thread(void *arg)
8024 {
8025 struct md_thread *thread = arg;
8026
8027 /*
8028 * md_thread is a 'system-thread', it's priority should be very
8029 * high. We avoid resource deadlocks individually in each
8030 * raid personality. (RAID5 does preallocation) We also use RR and
8031 * the very same RT priority as kswapd, thus we will never get
8032 * into a priority inversion deadlock.
8033 *
8034 * we definitely have to have equal or higher priority than
8035 * bdflush, otherwise bdflush will deadlock if there are too
8036 * many dirty RAID5 blocks.
8037 */
8038
8039 allow_signal(SIGKILL);
8040 while (!kthread_should_stop()) {
8041
8042 /* We need to wait INTERRUPTIBLE so that
8043 * we don't add to the load-average.
8044 * That means we need to be sure no signals are
8045 * pending
8046 */
8047 if (signal_pending(current))
8048 flush_signals(current);
8049
8050 wait_event_interruptible_timeout
8051 (thread->wqueue,
8052 test_bit(THREAD_WAKEUP, &thread->flags)
8053 || kthread_should_stop() || kthread_should_park(),
8054 thread->timeout);
8055
8056 clear_bit(THREAD_WAKEUP, &thread->flags);
8057 if (kthread_should_park())
8058 kthread_parkme();
8059 if (!kthread_should_stop())
8060 thread->run(thread);
8061 }
8062
8063 return 0;
8064 }
8065
md_wakeup_thread_directly(struct md_thread __rcu * thread)8066 static void md_wakeup_thread_directly(struct md_thread __rcu *thread)
8067 {
8068 struct md_thread *t;
8069
8070 rcu_read_lock();
8071 t = rcu_dereference(thread);
8072 if (t)
8073 wake_up_process(t->tsk);
8074 rcu_read_unlock();
8075 }
8076
md_wakeup_thread(struct md_thread __rcu * thread)8077 void md_wakeup_thread(struct md_thread __rcu *thread)
8078 {
8079 struct md_thread *t;
8080
8081 rcu_read_lock();
8082 t = rcu_dereference(thread);
8083 if (t) {
8084 pr_debug("md: waking up MD thread %s.\n", t->tsk->comm);
8085 set_bit(THREAD_WAKEUP, &t->flags);
8086 if (wq_has_sleeper(&t->wqueue))
8087 wake_up(&t->wqueue);
8088 }
8089 rcu_read_unlock();
8090 }
8091 EXPORT_SYMBOL(md_wakeup_thread);
8092
md_register_thread(void (* run)(struct md_thread *),struct mddev * mddev,const char * name)8093 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
8094 struct mddev *mddev, const char *name)
8095 {
8096 struct md_thread *thread;
8097
8098 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
8099 if (!thread)
8100 return NULL;
8101
8102 init_waitqueue_head(&thread->wqueue);
8103
8104 thread->run = run;
8105 thread->mddev = mddev;
8106 thread->timeout = MAX_SCHEDULE_TIMEOUT;
8107 thread->tsk = kthread_run(md_thread, thread,
8108 "%s_%s",
8109 mdname(thread->mddev),
8110 name);
8111 if (IS_ERR(thread->tsk)) {
8112 kfree(thread);
8113 return NULL;
8114 }
8115 return thread;
8116 }
8117 EXPORT_SYMBOL(md_register_thread);
8118
md_unregister_thread(struct mddev * mddev,struct md_thread __rcu ** threadp)8119 void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp)
8120 {
8121 struct md_thread *thread = rcu_dereference_protected(*threadp,
8122 lockdep_is_held(&mddev->reconfig_mutex));
8123
8124 if (!thread)
8125 return;
8126
8127 rcu_assign_pointer(*threadp, NULL);
8128 synchronize_rcu();
8129
8130 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
8131 kthread_stop(thread->tsk);
8132 kfree(thread);
8133 }
8134 EXPORT_SYMBOL(md_unregister_thread);
8135
md_error(struct mddev * mddev,struct md_rdev * rdev)8136 void md_error(struct mddev *mddev, struct md_rdev *rdev)
8137 {
8138 if (!rdev || test_bit(Faulty, &rdev->flags))
8139 return;
8140
8141 if (!mddev->pers || !mddev->pers->error_handler)
8142 return;
8143 mddev->pers->error_handler(mddev, rdev);
8144
8145 if (mddev->pers->head.id == ID_RAID0 ||
8146 mddev->pers->head.id == ID_LINEAR)
8147 return;
8148
8149 if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
8150 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8151 sysfs_notify_dirent_safe(rdev->sysfs_state);
8152 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8153 if (!test_bit(MD_BROKEN, &mddev->flags)) {
8154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8155 md_wakeup_thread(mddev->thread);
8156 }
8157 if (mddev->event_work.func)
8158 queue_work(md_misc_wq, &mddev->event_work);
8159 md_new_event();
8160 }
8161 EXPORT_SYMBOL(md_error);
8162
8163 /* seq_file implementation /proc/mdstat */
8164
status_unused(struct seq_file * seq)8165 static void status_unused(struct seq_file *seq)
8166 {
8167 int i = 0;
8168 struct md_rdev *rdev;
8169
8170 seq_printf(seq, "unused devices: ");
8171
8172 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8173 i++;
8174 seq_printf(seq, "%pg ", rdev->bdev);
8175 }
8176 if (!i)
8177 seq_printf(seq, "<none>");
8178
8179 seq_printf(seq, "\n");
8180 }
8181
status_personalities(struct seq_file * seq)8182 static void status_personalities(struct seq_file *seq)
8183 {
8184 struct md_submodule_head *head;
8185 unsigned long i;
8186
8187 seq_puts(seq, "Personalities : ");
8188
8189 xa_lock(&md_submodule);
8190 xa_for_each(&md_submodule, i, head)
8191 if (head->type == MD_PERSONALITY)
8192 seq_printf(seq, "[%s] ", head->name);
8193 xa_unlock(&md_submodule);
8194
8195 seq_puts(seq, "\n");
8196 }
8197
status_resync(struct seq_file * seq,struct mddev * mddev)8198 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8199 {
8200 sector_t max_sectors, resync, res;
8201 unsigned long dt, db = 0;
8202 sector_t rt, curr_mark_cnt, resync_mark_cnt;
8203 int scale, recovery_active;
8204 unsigned int per_milli;
8205
8206 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8207 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8208 max_sectors = mddev->resync_max_sectors;
8209 else
8210 max_sectors = mddev->dev_sectors;
8211
8212 resync = mddev->curr_resync;
8213 if (resync < MD_RESYNC_ACTIVE) {
8214 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8215 /* Still cleaning up */
8216 resync = max_sectors;
8217 } else if (resync > max_sectors) {
8218 resync = max_sectors;
8219 } else {
8220 res = atomic_read(&mddev->recovery_active);
8221 /*
8222 * Resync has started, but the subtraction has overflowed or
8223 * yielded one of the special values. Force it to active to
8224 * ensure the status reports an active resync.
8225 */
8226 if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8227 resync = MD_RESYNC_ACTIVE;
8228 else
8229 resync -= res;
8230 }
8231
8232 if (resync == MD_RESYNC_NONE) {
8233 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8234 struct md_rdev *rdev;
8235
8236 rdev_for_each(rdev, mddev)
8237 if (rdev->raid_disk >= 0 &&
8238 !test_bit(Faulty, &rdev->flags) &&
8239 rdev->recovery_offset != MaxSector &&
8240 rdev->recovery_offset) {
8241 seq_printf(seq, "\trecover=REMOTE");
8242 return 1;
8243 }
8244 if (mddev->reshape_position != MaxSector)
8245 seq_printf(seq, "\treshape=REMOTE");
8246 else
8247 seq_printf(seq, "\tresync=REMOTE");
8248 return 1;
8249 }
8250 if (mddev->recovery_cp < MaxSector) {
8251 seq_printf(seq, "\tresync=PENDING");
8252 return 1;
8253 }
8254 return 0;
8255 }
8256 if (resync < MD_RESYNC_ACTIVE) {
8257 seq_printf(seq, "\tresync=DELAYED");
8258 return 1;
8259 }
8260
8261 WARN_ON(max_sectors == 0);
8262 /* Pick 'scale' such that (resync>>scale)*1000 will fit
8263 * in a sector_t, and (max_sectors>>scale) will fit in a
8264 * u32, as those are the requirements for sector_div.
8265 * Thus 'scale' must be at least 10
8266 */
8267 scale = 10;
8268 if (sizeof(sector_t) > sizeof(unsigned long)) {
8269 while ( max_sectors/2 > (1ULL<<(scale+32)))
8270 scale++;
8271 }
8272 res = (resync>>scale)*1000;
8273 sector_div(res, (u32)((max_sectors>>scale)+1));
8274
8275 per_milli = res;
8276 {
8277 int i, x = per_milli/50, y = 20-x;
8278 seq_printf(seq, "[");
8279 for (i = 0; i < x; i++)
8280 seq_printf(seq, "=");
8281 seq_printf(seq, ">");
8282 for (i = 0; i < y; i++)
8283 seq_printf(seq, ".");
8284 seq_printf(seq, "] ");
8285 }
8286 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8287 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8288 "reshape" :
8289 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8290 "check" :
8291 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8292 "resync" : "recovery"))),
8293 per_milli/10, per_milli % 10,
8294 (unsigned long long) resync/2,
8295 (unsigned long long) max_sectors/2);
8296
8297 /*
8298 * dt: time from mark until now
8299 * db: blocks written from mark until now
8300 * rt: remaining time
8301 *
8302 * rt is a sector_t, which is always 64bit now. We are keeping
8303 * the original algorithm, but it is not really necessary.
8304 *
8305 * Original algorithm:
8306 * So we divide before multiply in case it is 32bit and close
8307 * to the limit.
8308 * We scale the divisor (db) by 32 to avoid losing precision
8309 * near the end of resync when the number of remaining sectors
8310 * is close to 'db'.
8311 * We then divide rt by 32 after multiplying by db to compensate.
8312 * The '+1' avoids division by zero if db is very small.
8313 */
8314 dt = ((jiffies - mddev->resync_mark) / HZ);
8315 if (!dt) dt++;
8316
8317 curr_mark_cnt = mddev->curr_mark_cnt;
8318 recovery_active = atomic_read(&mddev->recovery_active);
8319 resync_mark_cnt = mddev->resync_mark_cnt;
8320
8321 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8322 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8323
8324 rt = max_sectors - resync; /* number of remaining sectors */
8325 rt = div64_u64(rt, db/32+1);
8326 rt *= dt;
8327 rt >>= 5;
8328
8329 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8330 ((unsigned long)rt % 60)/6);
8331
8332 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8333 return 1;
8334 }
8335
md_seq_start(struct seq_file * seq,loff_t * pos)8336 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8337 __acquires(&all_mddevs_lock)
8338 {
8339 seq->poll_event = atomic_read(&md_event_count);
8340 spin_lock(&all_mddevs_lock);
8341
8342 return seq_list_start_head(&all_mddevs, *pos);
8343 }
8344
md_seq_next(struct seq_file * seq,void * v,loff_t * pos)8345 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8346 {
8347 return seq_list_next(v, &all_mddevs, pos);
8348 }
8349
md_seq_stop(struct seq_file * seq,void * v)8350 static void md_seq_stop(struct seq_file *seq, void *v)
8351 __releases(&all_mddevs_lock)
8352 {
8353 spin_unlock(&all_mddevs_lock);
8354 }
8355
md_bitmap_status(struct seq_file * seq,struct mddev * mddev)8356 static void md_bitmap_status(struct seq_file *seq, struct mddev *mddev)
8357 {
8358 struct md_bitmap_stats stats;
8359 unsigned long used_pages;
8360 unsigned long chunk_kb;
8361 int err;
8362
8363 err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
8364 if (err)
8365 return;
8366
8367 chunk_kb = mddev->bitmap_info.chunksize >> 10;
8368 used_pages = stats.pages - stats.missing_pages;
8369
8370 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], %lu%s chunk",
8371 used_pages, stats.pages, used_pages << (PAGE_SHIFT - 10),
8372 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
8373 chunk_kb ? "KB" : "B");
8374
8375 if (stats.file) {
8376 seq_puts(seq, ", file: ");
8377 seq_file_path(seq, stats.file, " \t\n");
8378 }
8379
8380 seq_putc(seq, '\n');
8381 }
8382
md_seq_show(struct seq_file * seq,void * v)8383 static int md_seq_show(struct seq_file *seq, void *v)
8384 {
8385 struct mddev *mddev;
8386 sector_t sectors;
8387 struct md_rdev *rdev;
8388
8389 if (v == &all_mddevs) {
8390 status_personalities(seq);
8391 if (list_empty(&all_mddevs))
8392 status_unused(seq);
8393 return 0;
8394 }
8395
8396 mddev = list_entry(v, struct mddev, all_mddevs);
8397 if (!mddev_get(mddev))
8398 return 0;
8399
8400 spin_unlock(&all_mddevs_lock);
8401
8402 /* prevent bitmap to be freed after checking */
8403 mutex_lock(&mddev->bitmap_info.mutex);
8404
8405 spin_lock(&mddev->lock);
8406 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8407 seq_printf(seq, "%s : ", mdname(mddev));
8408 if (mddev->pers) {
8409 if (test_bit(MD_BROKEN, &mddev->flags))
8410 seq_printf(seq, "broken");
8411 else
8412 seq_printf(seq, "active");
8413 if (mddev->ro == MD_RDONLY)
8414 seq_printf(seq, " (read-only)");
8415 if (mddev->ro == MD_AUTO_READ)
8416 seq_printf(seq, " (auto-read-only)");
8417 seq_printf(seq, " %s", mddev->pers->head.name);
8418 } else {
8419 seq_printf(seq, "inactive");
8420 }
8421
8422 sectors = 0;
8423 rcu_read_lock();
8424 rdev_for_each_rcu(rdev, mddev) {
8425 seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8426
8427 if (test_bit(WriteMostly, &rdev->flags))
8428 seq_printf(seq, "(W)");
8429 if (test_bit(Journal, &rdev->flags))
8430 seq_printf(seq, "(J)");
8431 if (test_bit(Faulty, &rdev->flags)) {
8432 seq_printf(seq, "(F)");
8433 continue;
8434 }
8435 if (rdev->raid_disk < 0)
8436 seq_printf(seq, "(S)"); /* spare */
8437 if (test_bit(Replacement, &rdev->flags))
8438 seq_printf(seq, "(R)");
8439 sectors += rdev->sectors;
8440 }
8441 rcu_read_unlock();
8442
8443 if (!list_empty(&mddev->disks)) {
8444 if (mddev->pers)
8445 seq_printf(seq, "\n %llu blocks",
8446 (unsigned long long)
8447 mddev->array_sectors / 2);
8448 else
8449 seq_printf(seq, "\n %llu blocks",
8450 (unsigned long long)sectors / 2);
8451 }
8452 if (mddev->persistent) {
8453 if (mddev->major_version != 0 ||
8454 mddev->minor_version != 90) {
8455 seq_printf(seq," super %d.%d",
8456 mddev->major_version,
8457 mddev->minor_version);
8458 }
8459 } else if (mddev->external)
8460 seq_printf(seq, " super external:%s",
8461 mddev->metadata_type);
8462 else
8463 seq_printf(seq, " super non-persistent");
8464
8465 if (mddev->pers) {
8466 mddev->pers->status(seq, mddev);
8467 seq_printf(seq, "\n ");
8468 if (mddev->pers->sync_request) {
8469 if (status_resync(seq, mddev))
8470 seq_printf(seq, "\n ");
8471 }
8472 } else
8473 seq_printf(seq, "\n ");
8474
8475 md_bitmap_status(seq, mddev);
8476
8477 seq_printf(seq, "\n");
8478 }
8479 spin_unlock(&mddev->lock);
8480 mutex_unlock(&mddev->bitmap_info.mutex);
8481 spin_lock(&all_mddevs_lock);
8482
8483 if (mddev == list_last_entry(&all_mddevs, struct mddev, all_mddevs))
8484 status_unused(seq);
8485
8486 mddev_put_locked(mddev);
8487 return 0;
8488 }
8489
8490 static const struct seq_operations md_seq_ops = {
8491 .start = md_seq_start,
8492 .next = md_seq_next,
8493 .stop = md_seq_stop,
8494 .show = md_seq_show,
8495 };
8496
md_seq_open(struct inode * inode,struct file * file)8497 static int md_seq_open(struct inode *inode, struct file *file)
8498 {
8499 struct seq_file *seq;
8500 int error;
8501
8502 error = seq_open(file, &md_seq_ops);
8503 if (error)
8504 return error;
8505
8506 seq = file->private_data;
8507 seq->poll_event = atomic_read(&md_event_count);
8508 return error;
8509 }
8510
8511 static int md_unloading;
mdstat_poll(struct file * filp,poll_table * wait)8512 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8513 {
8514 struct seq_file *seq = filp->private_data;
8515 __poll_t mask;
8516
8517 if (md_unloading)
8518 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8519 poll_wait(filp, &md_event_waiters, wait);
8520
8521 /* always allow read */
8522 mask = EPOLLIN | EPOLLRDNORM;
8523
8524 if (seq->poll_event != atomic_read(&md_event_count))
8525 mask |= EPOLLERR | EPOLLPRI;
8526 return mask;
8527 }
8528
8529 static const struct proc_ops mdstat_proc_ops = {
8530 .proc_open = md_seq_open,
8531 .proc_read = seq_read,
8532 .proc_lseek = seq_lseek,
8533 .proc_release = seq_release,
8534 .proc_poll = mdstat_poll,
8535 };
8536
register_md_submodule(struct md_submodule_head * msh)8537 int register_md_submodule(struct md_submodule_head *msh)
8538 {
8539 return xa_insert(&md_submodule, msh->id, msh, GFP_KERNEL);
8540 }
8541 EXPORT_SYMBOL_GPL(register_md_submodule);
8542
unregister_md_submodule(struct md_submodule_head * msh)8543 void unregister_md_submodule(struct md_submodule_head *msh)
8544 {
8545 xa_erase(&md_submodule, msh->id);
8546 }
8547 EXPORT_SYMBOL_GPL(unregister_md_submodule);
8548
md_setup_cluster(struct mddev * mddev,int nodes)8549 int md_setup_cluster(struct mddev *mddev, int nodes)
8550 {
8551 int ret = get_cluster_ops(mddev);
8552
8553 if (ret) {
8554 request_module("md-cluster");
8555 ret = get_cluster_ops(mddev);
8556 }
8557
8558 /* ensure module won't be unloaded */
8559 if (ret) {
8560 pr_warn("can't find md-cluster module or get its reference.\n");
8561 return ret;
8562 }
8563
8564 ret = mddev->cluster_ops->join(mddev, nodes);
8565 if (!ret)
8566 mddev->safemode_delay = 0;
8567 return ret;
8568 }
8569
md_cluster_stop(struct mddev * mddev)8570 void md_cluster_stop(struct mddev *mddev)
8571 {
8572 put_cluster_ops(mddev);
8573 }
8574
is_mddev_idle(struct mddev * mddev,int init)8575 static int is_mddev_idle(struct mddev *mddev, int init)
8576 {
8577 struct md_rdev *rdev;
8578 int idle;
8579 int curr_events;
8580
8581 idle = 1;
8582 rcu_read_lock();
8583 rdev_for_each_rcu(rdev, mddev) {
8584 struct gendisk *disk = rdev->bdev->bd_disk;
8585
8586 if (!init && !blk_queue_io_stat(disk->queue))
8587 continue;
8588
8589 curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8590 atomic_read(&disk->sync_io);
8591 /* sync IO will cause sync_io to increase before the disk_stats
8592 * as sync_io is counted when a request starts, and
8593 * disk_stats is counted when it completes.
8594 * So resync activity will cause curr_events to be smaller than
8595 * when there was no such activity.
8596 * non-sync IO will cause disk_stat to increase without
8597 * increasing sync_io so curr_events will (eventually)
8598 * be larger than it was before. Once it becomes
8599 * substantially larger, the test below will cause
8600 * the array to appear non-idle, and resync will slow
8601 * down.
8602 * If there is a lot of outstanding resync activity when
8603 * we set last_event to curr_events, then all that activity
8604 * completing might cause the array to appear non-idle
8605 * and resync will be slowed down even though there might
8606 * not have been non-resync activity. This will only
8607 * happen once though. 'last_events' will soon reflect
8608 * the state where there is little or no outstanding
8609 * resync requests, and further resync activity will
8610 * always make curr_events less than last_events.
8611 *
8612 */
8613 if (init || curr_events - rdev->last_events > 64) {
8614 rdev->last_events = curr_events;
8615 idle = 0;
8616 }
8617 }
8618 rcu_read_unlock();
8619 return idle;
8620 }
8621
md_done_sync(struct mddev * mddev,int blocks,int ok)8622 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8623 {
8624 /* another "blocks" (512byte) blocks have been synced */
8625 atomic_sub(blocks, &mddev->recovery_active);
8626 wake_up(&mddev->recovery_wait);
8627 if (!ok) {
8628 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8629 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8630 md_wakeup_thread(mddev->thread);
8631 // stop recovery, signal do_sync ....
8632 }
8633 }
8634 EXPORT_SYMBOL(md_done_sync);
8635
8636 /* md_write_start(mddev, bi)
8637 * If we need to update some array metadata (e.g. 'active' flag
8638 * in superblock) before writing, schedule a superblock update
8639 * and wait for it to complete.
8640 * A return value of 'false' means that the write wasn't recorded
8641 * and cannot proceed as the array is being suspend.
8642 */
md_write_start(struct mddev * mddev,struct bio * bi)8643 void md_write_start(struct mddev *mddev, struct bio *bi)
8644 {
8645 int did_change = 0;
8646
8647 if (bio_data_dir(bi) != WRITE)
8648 return;
8649
8650 BUG_ON(mddev->ro == MD_RDONLY);
8651 if (mddev->ro == MD_AUTO_READ) {
8652 /* need to switch to read/write */
8653 mddev->ro = MD_RDWR;
8654 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8655 md_wakeup_thread(mddev->thread);
8656 md_wakeup_thread(mddev->sync_thread);
8657 did_change = 1;
8658 }
8659 rcu_read_lock();
8660 percpu_ref_get(&mddev->writes_pending);
8661 smp_mb(); /* Match smp_mb in set_in_sync() */
8662 if (mddev->safemode == 1)
8663 mddev->safemode = 0;
8664 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8665 if (mddev->in_sync || mddev->sync_checkers) {
8666 spin_lock(&mddev->lock);
8667 if (mddev->in_sync) {
8668 mddev->in_sync = 0;
8669 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8670 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8671 md_wakeup_thread(mddev->thread);
8672 did_change = 1;
8673 }
8674 spin_unlock(&mddev->lock);
8675 }
8676 rcu_read_unlock();
8677 if (did_change)
8678 sysfs_notify_dirent_safe(mddev->sysfs_state);
8679 if (!mddev->has_superblocks)
8680 return;
8681 wait_event(mddev->sb_wait,
8682 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8683 }
8684 EXPORT_SYMBOL(md_write_start);
8685
8686 /* md_write_inc can only be called when md_write_start() has
8687 * already been called at least once of the current request.
8688 * It increments the counter and is useful when a single request
8689 * is split into several parts. Each part causes an increment and
8690 * so needs a matching md_write_end().
8691 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8692 * a spinlocked region.
8693 */
md_write_inc(struct mddev * mddev,struct bio * bi)8694 void md_write_inc(struct mddev *mddev, struct bio *bi)
8695 {
8696 if (bio_data_dir(bi) != WRITE)
8697 return;
8698 WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8699 percpu_ref_get(&mddev->writes_pending);
8700 }
8701 EXPORT_SYMBOL(md_write_inc);
8702
md_write_end(struct mddev * mddev)8703 void md_write_end(struct mddev *mddev)
8704 {
8705 percpu_ref_put(&mddev->writes_pending);
8706
8707 if (mddev->safemode == 2)
8708 md_wakeup_thread(mddev->thread);
8709 else if (mddev->safemode_delay)
8710 /* The roundup() ensures this only performs locking once
8711 * every ->safemode_delay jiffies
8712 */
8713 mod_timer(&mddev->safemode_timer,
8714 roundup(jiffies, mddev->safemode_delay) +
8715 mddev->safemode_delay);
8716 }
8717
8718 EXPORT_SYMBOL(md_write_end);
8719
8720 /* This is used by raid0 and raid10 */
md_submit_discard_bio(struct mddev * mddev,struct md_rdev * rdev,struct bio * bio,sector_t start,sector_t size)8721 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8722 struct bio *bio, sector_t start, sector_t size)
8723 {
8724 struct bio *discard_bio = NULL;
8725
8726 if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8727 &discard_bio) || !discard_bio)
8728 return;
8729
8730 bio_chain(discard_bio, bio);
8731 bio_clone_blkg_association(discard_bio, bio);
8732 mddev_trace_remap(mddev, discard_bio, bio->bi_iter.bi_sector);
8733 submit_bio_noacct(discard_bio);
8734 }
8735 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8736
md_bitmap_start(struct mddev * mddev,struct md_io_clone * md_io_clone)8737 static void md_bitmap_start(struct mddev *mddev,
8738 struct md_io_clone *md_io_clone)
8739 {
8740 if (mddev->pers->bitmap_sector)
8741 mddev->pers->bitmap_sector(mddev, &md_io_clone->offset,
8742 &md_io_clone->sectors);
8743
8744 mddev->bitmap_ops->startwrite(mddev, md_io_clone->offset,
8745 md_io_clone->sectors);
8746 }
8747
md_bitmap_end(struct mddev * mddev,struct md_io_clone * md_io_clone)8748 static void md_bitmap_end(struct mddev *mddev, struct md_io_clone *md_io_clone)
8749 {
8750 mddev->bitmap_ops->endwrite(mddev, md_io_clone->offset,
8751 md_io_clone->sectors);
8752 }
8753
md_end_clone_io(struct bio * bio)8754 static void md_end_clone_io(struct bio *bio)
8755 {
8756 struct md_io_clone *md_io_clone = bio->bi_private;
8757 struct bio *orig_bio = md_io_clone->orig_bio;
8758 struct mddev *mddev = md_io_clone->mddev;
8759
8760 if (bio_data_dir(orig_bio) == WRITE && mddev->bitmap)
8761 md_bitmap_end(mddev, md_io_clone);
8762
8763 if (bio->bi_status && !orig_bio->bi_status)
8764 orig_bio->bi_status = bio->bi_status;
8765
8766 if (md_io_clone->start_time)
8767 bio_end_io_acct(orig_bio, md_io_clone->start_time);
8768
8769 bio_put(bio);
8770 bio_endio(orig_bio);
8771 percpu_ref_put(&mddev->active_io);
8772 }
8773
md_clone_bio(struct mddev * mddev,struct bio ** bio)8774 static void md_clone_bio(struct mddev *mddev, struct bio **bio)
8775 {
8776 struct block_device *bdev = (*bio)->bi_bdev;
8777 struct md_io_clone *md_io_clone;
8778 struct bio *clone =
8779 bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_clone_set);
8780
8781 md_io_clone = container_of(clone, struct md_io_clone, bio_clone);
8782 md_io_clone->orig_bio = *bio;
8783 md_io_clone->mddev = mddev;
8784 if (blk_queue_io_stat(bdev->bd_disk->queue))
8785 md_io_clone->start_time = bio_start_io_acct(*bio);
8786
8787 if (bio_data_dir(*bio) == WRITE && mddev->bitmap) {
8788 md_io_clone->offset = (*bio)->bi_iter.bi_sector;
8789 md_io_clone->sectors = bio_sectors(*bio);
8790 md_bitmap_start(mddev, md_io_clone);
8791 }
8792
8793 clone->bi_end_io = md_end_clone_io;
8794 clone->bi_private = md_io_clone;
8795 *bio = clone;
8796 }
8797
md_account_bio(struct mddev * mddev,struct bio ** bio)8798 void md_account_bio(struct mddev *mddev, struct bio **bio)
8799 {
8800 percpu_ref_get(&mddev->active_io);
8801 md_clone_bio(mddev, bio);
8802 }
8803 EXPORT_SYMBOL_GPL(md_account_bio);
8804
md_free_cloned_bio(struct bio * bio)8805 void md_free_cloned_bio(struct bio *bio)
8806 {
8807 struct md_io_clone *md_io_clone = bio->bi_private;
8808 struct bio *orig_bio = md_io_clone->orig_bio;
8809 struct mddev *mddev = md_io_clone->mddev;
8810
8811 if (bio_data_dir(orig_bio) == WRITE && mddev->bitmap)
8812 md_bitmap_end(mddev, md_io_clone);
8813
8814 if (bio->bi_status && !orig_bio->bi_status)
8815 orig_bio->bi_status = bio->bi_status;
8816
8817 if (md_io_clone->start_time)
8818 bio_end_io_acct(orig_bio, md_io_clone->start_time);
8819
8820 bio_put(bio);
8821 percpu_ref_put(&mddev->active_io);
8822 }
8823 EXPORT_SYMBOL_GPL(md_free_cloned_bio);
8824
8825 /* md_allow_write(mddev)
8826 * Calling this ensures that the array is marked 'active' so that writes
8827 * may proceed without blocking. It is important to call this before
8828 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8829 * Must be called with mddev_lock held.
8830 */
md_allow_write(struct mddev * mddev)8831 void md_allow_write(struct mddev *mddev)
8832 {
8833 if (!mddev->pers)
8834 return;
8835 if (!md_is_rdwr(mddev))
8836 return;
8837 if (!mddev->pers->sync_request)
8838 return;
8839
8840 spin_lock(&mddev->lock);
8841 if (mddev->in_sync) {
8842 mddev->in_sync = 0;
8843 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8844 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8845 if (mddev->safemode_delay &&
8846 mddev->safemode == 0)
8847 mddev->safemode = 1;
8848 spin_unlock(&mddev->lock);
8849 md_update_sb(mddev, 0);
8850 sysfs_notify_dirent_safe(mddev->sysfs_state);
8851 /* wait for the dirty state to be recorded in the metadata */
8852 wait_event(mddev->sb_wait,
8853 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8854 } else
8855 spin_unlock(&mddev->lock);
8856 }
8857 EXPORT_SYMBOL_GPL(md_allow_write);
8858
md_sync_max_sectors(struct mddev * mddev,enum sync_action action)8859 static sector_t md_sync_max_sectors(struct mddev *mddev,
8860 enum sync_action action)
8861 {
8862 switch (action) {
8863 case ACTION_RESYNC:
8864 case ACTION_CHECK:
8865 case ACTION_REPAIR:
8866 atomic64_set(&mddev->resync_mismatches, 0);
8867 fallthrough;
8868 case ACTION_RESHAPE:
8869 return mddev->resync_max_sectors;
8870 case ACTION_RECOVER:
8871 return mddev->dev_sectors;
8872 default:
8873 return 0;
8874 }
8875 }
8876
md_sync_position(struct mddev * mddev,enum sync_action action)8877 static sector_t md_sync_position(struct mddev *mddev, enum sync_action action)
8878 {
8879 sector_t start = 0;
8880 struct md_rdev *rdev;
8881
8882 switch (action) {
8883 case ACTION_CHECK:
8884 case ACTION_REPAIR:
8885 return mddev->resync_min;
8886 case ACTION_RESYNC:
8887 if (!mddev->bitmap)
8888 return mddev->recovery_cp;
8889 return 0;
8890 case ACTION_RESHAPE:
8891 /*
8892 * If the original node aborts reshaping then we continue the
8893 * reshaping, so set again to avoid restart reshape from the
8894 * first beginning
8895 */
8896 if (mddev_is_clustered(mddev) &&
8897 mddev->reshape_position != MaxSector)
8898 return mddev->reshape_position;
8899 return 0;
8900 case ACTION_RECOVER:
8901 start = MaxSector;
8902 rcu_read_lock();
8903 rdev_for_each_rcu(rdev, mddev)
8904 if (rdev->raid_disk >= 0 &&
8905 !test_bit(Journal, &rdev->flags) &&
8906 !test_bit(Faulty, &rdev->flags) &&
8907 !test_bit(In_sync, &rdev->flags) &&
8908 rdev->recovery_offset < start)
8909 start = rdev->recovery_offset;
8910 rcu_read_unlock();
8911
8912 /* If there is a bitmap, we need to make sure all
8913 * writes that started before we added a spare
8914 * complete before we start doing a recovery.
8915 * Otherwise the write might complete and (via
8916 * bitmap_endwrite) set a bit in the bitmap after the
8917 * recovery has checked that bit and skipped that
8918 * region.
8919 */
8920 if (mddev->bitmap) {
8921 mddev->pers->quiesce(mddev, 1);
8922 mddev->pers->quiesce(mddev, 0);
8923 }
8924 return start;
8925 default:
8926 return MaxSector;
8927 }
8928 }
8929
8930 #define SYNC_MARKS 10
8931 #define SYNC_MARK_STEP (3*HZ)
8932 #define UPDATE_FREQUENCY (5*60*HZ)
md_do_sync(struct md_thread * thread)8933 void md_do_sync(struct md_thread *thread)
8934 {
8935 struct mddev *mddev = thread->mddev;
8936 struct mddev *mddev2;
8937 unsigned int currspeed = 0, window;
8938 sector_t max_sectors,j, io_sectors, recovery_done;
8939 unsigned long mark[SYNC_MARKS];
8940 unsigned long update_time;
8941 sector_t mark_cnt[SYNC_MARKS];
8942 int last_mark,m;
8943 sector_t last_check;
8944 int skipped = 0;
8945 struct md_rdev *rdev;
8946 enum sync_action action;
8947 const char *desc;
8948 struct blk_plug plug;
8949 int ret;
8950
8951 /* just incase thread restarts... */
8952 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8953 return;
8954
8955 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8956 goto skip;
8957
8958 if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
8959 !md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8960 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8961 goto skip;
8962 }
8963
8964 if (mddev_is_clustered(mddev)) {
8965 ret = mddev->cluster_ops->resync_start(mddev);
8966 if (ret)
8967 goto skip;
8968
8969 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8970 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8971 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8972 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8973 && ((unsigned long long)mddev->curr_resync_completed
8974 < (unsigned long long)mddev->resync_max_sectors))
8975 goto skip;
8976 }
8977
8978 action = md_sync_action(mddev);
8979 desc = md_sync_action_name(action);
8980 mddev->last_sync_action = action;
8981
8982 /*
8983 * Before starting a resync we must have set curr_resync to
8984 * 2, and then checked that every "conflicting" array has curr_resync
8985 * less than ours. When we find one that is the same or higher
8986 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8987 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8988 * This will mean we have to start checking from the beginning again.
8989 *
8990 */
8991 if (mddev_is_clustered(mddev))
8992 mddev->cluster_ops->resync_start_notify(mddev);
8993 do {
8994 int mddev2_minor = -1;
8995 mddev->curr_resync = MD_RESYNC_DELAYED;
8996
8997 try_again:
8998 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8999 goto skip;
9000 spin_lock(&all_mddevs_lock);
9001 list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
9002 if (test_bit(MD_DELETED, &mddev2->flags))
9003 continue;
9004 if (mddev2 == mddev)
9005 continue;
9006 if (!mddev->parallel_resync
9007 && mddev2->curr_resync
9008 && match_mddev_units(mddev, mddev2)) {
9009 DEFINE_WAIT(wq);
9010 if (mddev < mddev2 &&
9011 mddev->curr_resync == MD_RESYNC_DELAYED) {
9012 /* arbitrarily yield */
9013 mddev->curr_resync = MD_RESYNC_YIELDED;
9014 wake_up(&resync_wait);
9015 }
9016 if (mddev > mddev2 &&
9017 mddev->curr_resync == MD_RESYNC_YIELDED)
9018 /* no need to wait here, we can wait the next
9019 * time 'round when curr_resync == 2
9020 */
9021 continue;
9022 /* We need to wait 'interruptible' so as not to
9023 * contribute to the load average, and not to
9024 * be caught by 'softlockup'
9025 */
9026 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
9027 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9028 mddev2->curr_resync >= mddev->curr_resync) {
9029 if (mddev2_minor != mddev2->md_minor) {
9030 mddev2_minor = mddev2->md_minor;
9031 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
9032 desc, mdname(mddev),
9033 mdname(mddev2));
9034 }
9035 spin_unlock(&all_mddevs_lock);
9036
9037 if (signal_pending(current))
9038 flush_signals(current);
9039 schedule();
9040 finish_wait(&resync_wait, &wq);
9041 goto try_again;
9042 }
9043 finish_wait(&resync_wait, &wq);
9044 }
9045 }
9046 spin_unlock(&all_mddevs_lock);
9047 } while (mddev->curr_resync < MD_RESYNC_DELAYED);
9048
9049 max_sectors = md_sync_max_sectors(mddev, action);
9050 j = md_sync_position(mddev, action);
9051
9052 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
9053 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
9054 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
9055 speed_max(mddev), desc);
9056
9057 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
9058
9059 io_sectors = 0;
9060 for (m = 0; m < SYNC_MARKS; m++) {
9061 mark[m] = jiffies;
9062 mark_cnt[m] = io_sectors;
9063 }
9064 last_mark = 0;
9065 mddev->resync_mark = mark[last_mark];
9066 mddev->resync_mark_cnt = mark_cnt[last_mark];
9067
9068 /*
9069 * Tune reconstruction:
9070 */
9071 window = 32 * (PAGE_SIZE / 512);
9072 pr_debug("md: using %dk window, over a total of %lluk.\n",
9073 window/2, (unsigned long long)max_sectors/2);
9074
9075 atomic_set(&mddev->recovery_active, 0);
9076 last_check = 0;
9077
9078 if (j >= MD_RESYNC_ACTIVE) {
9079 pr_debug("md: resuming %s of %s from checkpoint.\n",
9080 desc, mdname(mddev));
9081 mddev->curr_resync = j;
9082 } else
9083 mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
9084 mddev->curr_resync_completed = j;
9085 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9086 md_new_event();
9087 update_time = jiffies;
9088
9089 blk_start_plug(&plug);
9090 while (j < max_sectors) {
9091 sector_t sectors;
9092
9093 skipped = 0;
9094
9095 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9096 ((mddev->curr_resync > mddev->curr_resync_completed &&
9097 (mddev->curr_resync - mddev->curr_resync_completed)
9098 > (max_sectors >> 4)) ||
9099 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
9100 (j - mddev->curr_resync_completed)*2
9101 >= mddev->resync_max - mddev->curr_resync_completed ||
9102 mddev->curr_resync_completed > mddev->resync_max
9103 )) {
9104 /* time to update curr_resync_completed */
9105 wait_event(mddev->recovery_wait,
9106 atomic_read(&mddev->recovery_active) == 0);
9107 mddev->curr_resync_completed = j;
9108 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
9109 j > mddev->recovery_cp)
9110 mddev->recovery_cp = j;
9111 update_time = jiffies;
9112 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
9113 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9114 }
9115
9116 while (j >= mddev->resync_max &&
9117 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9118 /* As this condition is controlled by user-space,
9119 * we can block indefinitely, so use '_interruptible'
9120 * to avoid triggering warnings.
9121 */
9122 flush_signals(current); /* just in case */
9123 wait_event_interruptible(mddev->recovery_wait,
9124 mddev->resync_max > j
9125 || test_bit(MD_RECOVERY_INTR,
9126 &mddev->recovery));
9127 }
9128
9129 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9130 break;
9131
9132 sectors = mddev->pers->sync_request(mddev, j, max_sectors,
9133 &skipped);
9134 if (sectors == 0) {
9135 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9136 break;
9137 }
9138
9139 if (!skipped) { /* actual IO requested */
9140 io_sectors += sectors;
9141 atomic_add(sectors, &mddev->recovery_active);
9142 }
9143
9144 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9145 break;
9146
9147 j += sectors;
9148 if (j > max_sectors)
9149 /* when skipping, extra large numbers can be returned. */
9150 j = max_sectors;
9151 if (j >= MD_RESYNC_ACTIVE)
9152 mddev->curr_resync = j;
9153 mddev->curr_mark_cnt = io_sectors;
9154 if (last_check == 0)
9155 /* this is the earliest that rebuild will be
9156 * visible in /proc/mdstat
9157 */
9158 md_new_event();
9159
9160 if (last_check + window > io_sectors || j == max_sectors)
9161 continue;
9162
9163 last_check = io_sectors;
9164 repeat:
9165 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
9166 /* step marks */
9167 int next = (last_mark+1) % SYNC_MARKS;
9168
9169 mddev->resync_mark = mark[next];
9170 mddev->resync_mark_cnt = mark_cnt[next];
9171 mark[next] = jiffies;
9172 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
9173 last_mark = next;
9174 }
9175
9176 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9177 break;
9178
9179 /*
9180 * this loop exits only if either when we are slower than
9181 * the 'hard' speed limit, or the system was IO-idle for
9182 * a jiffy.
9183 * the system might be non-idle CPU-wise, but we only care
9184 * about not overloading the IO subsystem. (things like an
9185 * e2fsck being done on the RAID array should execute fast)
9186 */
9187 cond_resched();
9188
9189 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9190 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9191 /((jiffies-mddev->resync_mark)/HZ +1) +1;
9192
9193 if (currspeed > speed_min(mddev)) {
9194 if (currspeed > speed_max(mddev)) {
9195 msleep(500);
9196 goto repeat;
9197 }
9198 if (!is_mddev_idle(mddev, 0)) {
9199 /*
9200 * Give other IO more of a chance.
9201 * The faster the devices, the less we wait.
9202 */
9203 wait_event(mddev->recovery_wait,
9204 !atomic_read(&mddev->recovery_active));
9205 }
9206 }
9207 }
9208 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9209 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9210 ? "interrupted" : "done");
9211 /*
9212 * this also signals 'finished resyncing' to md_stop
9213 */
9214 blk_finish_plug(&plug);
9215 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9216
9217 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9218 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9219 mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9220 mddev->curr_resync_completed = mddev->curr_resync;
9221 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9222 }
9223 mddev->pers->sync_request(mddev, max_sectors, max_sectors, &skipped);
9224
9225 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9226 mddev->curr_resync > MD_RESYNC_ACTIVE) {
9227 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9228 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9229 if (mddev->curr_resync >= mddev->recovery_cp) {
9230 pr_debug("md: checkpointing %s of %s.\n",
9231 desc, mdname(mddev));
9232 if (test_bit(MD_RECOVERY_ERROR,
9233 &mddev->recovery))
9234 mddev->recovery_cp =
9235 mddev->curr_resync_completed;
9236 else
9237 mddev->recovery_cp =
9238 mddev->curr_resync;
9239 }
9240 } else
9241 mddev->recovery_cp = MaxSector;
9242 } else {
9243 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9244 mddev->curr_resync = MaxSector;
9245 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9246 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9247 rcu_read_lock();
9248 rdev_for_each_rcu(rdev, mddev)
9249 if (rdev->raid_disk >= 0 &&
9250 mddev->delta_disks >= 0 &&
9251 !test_bit(Journal, &rdev->flags) &&
9252 !test_bit(Faulty, &rdev->flags) &&
9253 !test_bit(In_sync, &rdev->flags) &&
9254 rdev->recovery_offset < mddev->curr_resync)
9255 rdev->recovery_offset = mddev->curr_resync;
9256 rcu_read_unlock();
9257 }
9258 }
9259 }
9260 skip:
9261 /* set CHANGE_PENDING here since maybe another update is needed,
9262 * so other nodes are informed. It should be harmless for normal
9263 * raid */
9264 set_mask_bits(&mddev->sb_flags, 0,
9265 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9266
9267 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9268 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9269 mddev->delta_disks > 0 &&
9270 mddev->pers->finish_reshape &&
9271 mddev->pers->size &&
9272 !mddev_is_dm(mddev)) {
9273 mddev_lock_nointr(mddev);
9274 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9275 mddev_unlock(mddev);
9276 if (!mddev_is_clustered(mddev))
9277 set_capacity_and_notify(mddev->gendisk,
9278 mddev->array_sectors);
9279 }
9280
9281 spin_lock(&mddev->lock);
9282 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9283 /* We completed so min/max setting can be forgotten if used. */
9284 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9285 mddev->resync_min = 0;
9286 mddev->resync_max = MaxSector;
9287 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9288 mddev->resync_min = mddev->curr_resync_completed;
9289 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9290 mddev->curr_resync = MD_RESYNC_NONE;
9291 spin_unlock(&mddev->lock);
9292
9293 wake_up(&resync_wait);
9294 md_wakeup_thread(mddev->thread);
9295 return;
9296 }
9297 EXPORT_SYMBOL_GPL(md_do_sync);
9298
rdev_removeable(struct md_rdev * rdev)9299 static bool rdev_removeable(struct md_rdev *rdev)
9300 {
9301 /* rdev is not used. */
9302 if (rdev->raid_disk < 0)
9303 return false;
9304
9305 /* There are still inflight io, don't remove this rdev. */
9306 if (atomic_read(&rdev->nr_pending))
9307 return false;
9308
9309 /*
9310 * An error occurred but has not yet been acknowledged by the metadata
9311 * handler, don't remove this rdev.
9312 */
9313 if (test_bit(Blocked, &rdev->flags))
9314 return false;
9315
9316 /* Fautly rdev is not used, it's safe to remove it. */
9317 if (test_bit(Faulty, &rdev->flags))
9318 return true;
9319
9320 /* Journal disk can only be removed if it's faulty. */
9321 if (test_bit(Journal, &rdev->flags))
9322 return false;
9323
9324 /*
9325 * 'In_sync' is cleared while 'raid_disk' is valid, which means
9326 * replacement has just become active from pers->spare_active(), and
9327 * then pers->hot_remove_disk() will replace this rdev with replacement.
9328 */
9329 if (!test_bit(In_sync, &rdev->flags))
9330 return true;
9331
9332 return false;
9333 }
9334
rdev_is_spare(struct md_rdev * rdev)9335 static bool rdev_is_spare(struct md_rdev *rdev)
9336 {
9337 return !test_bit(Candidate, &rdev->flags) && rdev->raid_disk >= 0 &&
9338 !test_bit(In_sync, &rdev->flags) &&
9339 !test_bit(Journal, &rdev->flags) &&
9340 !test_bit(Faulty, &rdev->flags);
9341 }
9342
rdev_addable(struct md_rdev * rdev)9343 static bool rdev_addable(struct md_rdev *rdev)
9344 {
9345 /* rdev is already used, don't add it again. */
9346 if (test_bit(Candidate, &rdev->flags) || rdev->raid_disk >= 0 ||
9347 test_bit(Faulty, &rdev->flags))
9348 return false;
9349
9350 /* Allow to add journal disk. */
9351 if (test_bit(Journal, &rdev->flags))
9352 return true;
9353
9354 /* Allow to add if array is read-write. */
9355 if (md_is_rdwr(rdev->mddev))
9356 return true;
9357
9358 /*
9359 * For read-only array, only allow to readd a rdev. And if bitmap is
9360 * used, don't allow to readd a rdev that is too old.
9361 */
9362 if (rdev->saved_raid_disk >= 0 && !test_bit(Bitmap_sync, &rdev->flags))
9363 return true;
9364
9365 return false;
9366 }
9367
md_spares_need_change(struct mddev * mddev)9368 static bool md_spares_need_change(struct mddev *mddev)
9369 {
9370 struct md_rdev *rdev;
9371
9372 rcu_read_lock();
9373 rdev_for_each_rcu(rdev, mddev) {
9374 if (rdev_removeable(rdev) || rdev_addable(rdev)) {
9375 rcu_read_unlock();
9376 return true;
9377 }
9378 }
9379 rcu_read_unlock();
9380 return false;
9381 }
9382
remove_and_add_spares(struct mddev * mddev,struct md_rdev * this)9383 static int remove_and_add_spares(struct mddev *mddev,
9384 struct md_rdev *this)
9385 {
9386 struct md_rdev *rdev;
9387 int spares = 0;
9388 int removed = 0;
9389
9390 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9391 /* Mustn't remove devices when resync thread is running */
9392 return 0;
9393
9394 rdev_for_each(rdev, mddev) {
9395 if ((this == NULL || rdev == this) && rdev_removeable(rdev) &&
9396 !mddev->pers->hot_remove_disk(mddev, rdev)) {
9397 sysfs_unlink_rdev(mddev, rdev);
9398 rdev->saved_raid_disk = rdev->raid_disk;
9399 rdev->raid_disk = -1;
9400 removed++;
9401 }
9402 }
9403
9404 if (removed && mddev->kobj.sd)
9405 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9406
9407 if (this && removed)
9408 goto no_add;
9409
9410 rdev_for_each(rdev, mddev) {
9411 if (this && this != rdev)
9412 continue;
9413 if (rdev_is_spare(rdev))
9414 spares++;
9415 if (!rdev_addable(rdev))
9416 continue;
9417 if (!test_bit(Journal, &rdev->flags))
9418 rdev->recovery_offset = 0;
9419 if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9420 /* failure here is OK */
9421 sysfs_link_rdev(mddev, rdev);
9422 if (!test_bit(Journal, &rdev->flags))
9423 spares++;
9424 md_new_event();
9425 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9426 }
9427 }
9428 no_add:
9429 if (removed)
9430 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9431 return spares;
9432 }
9433
md_choose_sync_action(struct mddev * mddev,int * spares)9434 static bool md_choose_sync_action(struct mddev *mddev, int *spares)
9435 {
9436 /* Check if reshape is in progress first. */
9437 if (mddev->reshape_position != MaxSector) {
9438 if (mddev->pers->check_reshape == NULL ||
9439 mddev->pers->check_reshape(mddev) != 0)
9440 return false;
9441
9442 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9443 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9444 return true;
9445 }
9446
9447 /* Check if resync is in progress. */
9448 if (mddev->recovery_cp < MaxSector) {
9449 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9450 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9451 return true;
9452 }
9453
9454 /*
9455 * Remove any failed drives, then add spares if possible. Spares are
9456 * also removed and re-added, to allow the personality to fail the
9457 * re-add.
9458 */
9459 *spares = remove_and_add_spares(mddev, NULL);
9460 if (*spares) {
9461 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9462 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9463 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9464
9465 /* Start new recovery. */
9466 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9467 return true;
9468 }
9469
9470 /* Delay to choose resync/check/repair in md_do_sync(). */
9471 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9472 return true;
9473
9474 /* Nothing to be done */
9475 return false;
9476 }
9477
md_start_sync(struct work_struct * ws)9478 static void md_start_sync(struct work_struct *ws)
9479 {
9480 struct mddev *mddev = container_of(ws, struct mddev, sync_work);
9481 int spares = 0;
9482 bool suspend = false;
9483 char *name;
9484
9485 /*
9486 * If reshape is still in progress, spares won't be added or removed
9487 * from conf until reshape is done.
9488 */
9489 if (mddev->reshape_position == MaxSector &&
9490 md_spares_need_change(mddev)) {
9491 suspend = true;
9492 mddev_suspend(mddev, false);
9493 }
9494
9495 mddev_lock_nointr(mddev);
9496 if (!md_is_rdwr(mddev)) {
9497 /*
9498 * On a read-only array we can:
9499 * - remove failed devices
9500 * - add already-in_sync devices if the array itself is in-sync.
9501 * As we only add devices that are already in-sync, we can
9502 * activate the spares immediately.
9503 */
9504 remove_and_add_spares(mddev, NULL);
9505 goto not_running;
9506 }
9507
9508 if (!md_choose_sync_action(mddev, &spares))
9509 goto not_running;
9510
9511 if (!mddev->pers->sync_request)
9512 goto not_running;
9513
9514 /*
9515 * We are adding a device or devices to an array which has the bitmap
9516 * stored on all devices. So make sure all bitmap pages get written.
9517 */
9518 if (spares)
9519 mddev->bitmap_ops->write_all(mddev);
9520
9521 name = test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ?
9522 "reshape" : "resync";
9523 rcu_assign_pointer(mddev->sync_thread,
9524 md_register_thread(md_do_sync, mddev, name));
9525 if (!mddev->sync_thread) {
9526 pr_warn("%s: could not start resync thread...\n",
9527 mdname(mddev));
9528 /* leave the spares where they are, it shouldn't hurt */
9529 goto not_running;
9530 }
9531
9532 mddev_unlock(mddev);
9533 /*
9534 * md_start_sync was triggered by MD_RECOVERY_NEEDED, so we should
9535 * not set it again. Otherwise, we may cause issue like this one:
9536 * https://bugzilla.kernel.org/show_bug.cgi?id=218200
9537 * Therefore, use __mddev_resume(mddev, false).
9538 */
9539 if (suspend)
9540 __mddev_resume(mddev, false);
9541 md_wakeup_thread(mddev->sync_thread);
9542 sysfs_notify_dirent_safe(mddev->sysfs_action);
9543 md_new_event();
9544 return;
9545
9546 not_running:
9547 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9548 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9549 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9550 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9551 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9552 mddev_unlock(mddev);
9553 /*
9554 * md_start_sync was triggered by MD_RECOVERY_NEEDED, so we should
9555 * not set it again. Otherwise, we may cause issue like this one:
9556 * https://bugzilla.kernel.org/show_bug.cgi?id=218200
9557 * Therefore, use __mddev_resume(mddev, false).
9558 */
9559 if (suspend)
9560 __mddev_resume(mddev, false);
9561
9562 wake_up(&resync_wait);
9563 if (test_and_clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
9564 mddev->sysfs_action)
9565 sysfs_notify_dirent_safe(mddev->sysfs_action);
9566 }
9567
unregister_sync_thread(struct mddev * mddev)9568 static void unregister_sync_thread(struct mddev *mddev)
9569 {
9570 if (!test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9571 /* resync/recovery still happening */
9572 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9573 return;
9574 }
9575
9576 if (WARN_ON_ONCE(!mddev->sync_thread))
9577 return;
9578
9579 md_reap_sync_thread(mddev);
9580 }
9581
9582 /*
9583 * This routine is regularly called by all per-raid-array threads to
9584 * deal with generic issues like resync and super-block update.
9585 * Raid personalities that don't have a thread (linear/raid0) do not
9586 * need this as they never do any recovery or update the superblock.
9587 *
9588 * It does not do any resync itself, but rather "forks" off other threads
9589 * to do that as needed.
9590 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9591 * "->recovery" and create a thread at ->sync_thread.
9592 * When the thread finishes it sets MD_RECOVERY_DONE
9593 * and wakeups up this thread which will reap the thread and finish up.
9594 * This thread also removes any faulty devices (with nr_pending == 0).
9595 *
9596 * The overall approach is:
9597 * 1/ if the superblock needs updating, update it.
9598 * 2/ If a recovery thread is running, don't do anything else.
9599 * 3/ If recovery has finished, clean up, possibly marking spares active.
9600 * 4/ If there are any faulty devices, remove them.
9601 * 5/ If array is degraded, try to add spares devices
9602 * 6/ If array has spares or is not in-sync, start a resync thread.
9603 */
md_check_recovery(struct mddev * mddev)9604 void md_check_recovery(struct mddev *mddev)
9605 {
9606 if (mddev->bitmap)
9607 mddev->bitmap_ops->daemon_work(mddev);
9608
9609 if (signal_pending(current)) {
9610 if (mddev->pers->sync_request && !mddev->external) {
9611 pr_debug("md: %s in immediate safe mode\n",
9612 mdname(mddev));
9613 mddev->safemode = 2;
9614 }
9615 flush_signals(current);
9616 }
9617
9618 if (!md_is_rdwr(mddev) &&
9619 !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) &&
9620 !test_bit(MD_RECOVERY_DONE, &mddev->recovery))
9621 return;
9622 if ( ! (
9623 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9624 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9625 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9626 (mddev->external == 0 && mddev->safemode == 1) ||
9627 (mddev->safemode == 2
9628 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9629 ))
9630 return;
9631
9632 if (mddev_trylock(mddev)) {
9633 bool try_set_sync = mddev->safemode != 0;
9634
9635 if (!mddev->external && mddev->safemode == 1)
9636 mddev->safemode = 0;
9637
9638 if (!md_is_rdwr(mddev)) {
9639 struct md_rdev *rdev;
9640
9641 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9642 unregister_sync_thread(mddev);
9643 goto unlock;
9644 }
9645
9646 if (!mddev->external && mddev->in_sync)
9647 /*
9648 * 'Blocked' flag not needed as failed devices
9649 * will be recorded if array switched to read/write.
9650 * Leaving it set will prevent the device
9651 * from being removed.
9652 */
9653 rdev_for_each(rdev, mddev)
9654 clear_bit(Blocked, &rdev->flags);
9655
9656 /*
9657 * There is no thread, but we need to call
9658 * ->spare_active and clear saved_raid_disk
9659 */
9660 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9661 md_reap_sync_thread(mddev);
9662
9663 /*
9664 * Let md_start_sync() to remove and add rdevs to the
9665 * array.
9666 */
9667 if (md_spares_need_change(mddev)) {
9668 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9669 queue_work(md_misc_wq, &mddev->sync_work);
9670 }
9671
9672 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9673 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9674 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9675
9676 goto unlock;
9677 }
9678
9679 if (mddev_is_clustered(mddev)) {
9680 struct md_rdev *rdev, *tmp;
9681 /* kick the device if another node issued a
9682 * remove disk.
9683 */
9684 rdev_for_each_safe(rdev, tmp, mddev) {
9685 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9686 rdev->raid_disk < 0)
9687 md_kick_rdev_from_array(rdev);
9688 }
9689 }
9690
9691 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9692 spin_lock(&mddev->lock);
9693 set_in_sync(mddev);
9694 spin_unlock(&mddev->lock);
9695 }
9696
9697 if (mddev->sb_flags)
9698 md_update_sb(mddev, 0);
9699
9700 /*
9701 * Never start a new sync thread if MD_RECOVERY_RUNNING is
9702 * still set.
9703 */
9704 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9705 unregister_sync_thread(mddev);
9706 goto unlock;
9707 }
9708
9709 /* Set RUNNING before clearing NEEDED to avoid
9710 * any transients in the value of "sync_action".
9711 */
9712 mddev->curr_resync_completed = 0;
9713 spin_lock(&mddev->lock);
9714 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9715 spin_unlock(&mddev->lock);
9716 /* Clear some bits that don't mean anything, but
9717 * might be left set
9718 */
9719 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9720 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9721
9722 if (test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) &&
9723 !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
9724 queue_work(md_misc_wq, &mddev->sync_work);
9725 } else {
9726 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9727 wake_up(&resync_wait);
9728 }
9729
9730 unlock:
9731 wake_up(&mddev->sb_wait);
9732 mddev_unlock(mddev);
9733 }
9734 }
9735 EXPORT_SYMBOL(md_check_recovery);
9736
md_reap_sync_thread(struct mddev * mddev)9737 void md_reap_sync_thread(struct mddev *mddev)
9738 {
9739 struct md_rdev *rdev;
9740 sector_t old_dev_sectors = mddev->dev_sectors;
9741 bool is_reshaped = false;
9742
9743 /* resync has finished, collect result */
9744 md_unregister_thread(mddev, &mddev->sync_thread);
9745 atomic_inc(&mddev->sync_seq);
9746
9747 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9748 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9749 mddev->degraded != mddev->raid_disks) {
9750 /* success...*/
9751 /* activate any spares */
9752 if (mddev->pers->spare_active(mddev)) {
9753 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9754 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9755 }
9756 }
9757 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9758 mddev->pers->finish_reshape) {
9759 mddev->pers->finish_reshape(mddev);
9760 if (mddev_is_clustered(mddev))
9761 is_reshaped = true;
9762 }
9763
9764 /* If array is no-longer degraded, then any saved_raid_disk
9765 * information must be scrapped.
9766 */
9767 if (!mddev->degraded)
9768 rdev_for_each(rdev, mddev)
9769 rdev->saved_raid_disk = -1;
9770
9771 md_update_sb(mddev, 1);
9772 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9773 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9774 * clustered raid */
9775 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9776 mddev->cluster_ops->resync_finish(mddev);
9777 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9778 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9779 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9780 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9781 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9782 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9783 /*
9784 * We call mddev->cluster_ops->update_size here because sync_size could
9785 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9786 * so it is time to update size across cluster.
9787 */
9788 if (mddev_is_clustered(mddev) && is_reshaped
9789 && !test_bit(MD_CLOSING, &mddev->flags))
9790 mddev->cluster_ops->update_size(mddev, old_dev_sectors);
9791 /* flag recovery needed just to double check */
9792 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9793 sysfs_notify_dirent_safe(mddev->sysfs_completed);
9794 sysfs_notify_dirent_safe(mddev->sysfs_action);
9795 md_new_event();
9796 if (mddev->event_work.func)
9797 queue_work(md_misc_wq, &mddev->event_work);
9798 wake_up(&resync_wait);
9799 }
9800 EXPORT_SYMBOL(md_reap_sync_thread);
9801
md_wait_for_blocked_rdev(struct md_rdev * rdev,struct mddev * mddev)9802 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9803 {
9804 sysfs_notify_dirent_safe(rdev->sysfs_state);
9805 wait_event_timeout(rdev->blocked_wait, !rdev_blocked(rdev),
9806 msecs_to_jiffies(5000));
9807 rdev_dec_pending(rdev, mddev);
9808 }
9809 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9810
md_finish_reshape(struct mddev * mddev)9811 void md_finish_reshape(struct mddev *mddev)
9812 {
9813 /* called be personality module when reshape completes. */
9814 struct md_rdev *rdev;
9815
9816 rdev_for_each(rdev, mddev) {
9817 if (rdev->data_offset > rdev->new_data_offset)
9818 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9819 else
9820 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9821 rdev->data_offset = rdev->new_data_offset;
9822 }
9823 }
9824 EXPORT_SYMBOL(md_finish_reshape);
9825
9826 /* Bad block management */
9827
9828 /* Returns true on success, false on failure */
rdev_set_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9829 bool rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9830 int is_new)
9831 {
9832 struct mddev *mddev = rdev->mddev;
9833
9834 /*
9835 * Recording new badblocks for faulty rdev will force unnecessary
9836 * super block updating. This is fragile for external management because
9837 * userspace daemon may trying to remove this device and deadlock may
9838 * occur. This will be probably solved in the mdadm, but it is safer to
9839 * avoid it.
9840 */
9841 if (test_bit(Faulty, &rdev->flags))
9842 return true;
9843
9844 if (is_new)
9845 s += rdev->new_data_offset;
9846 else
9847 s += rdev->data_offset;
9848
9849 if (!badblocks_set(&rdev->badblocks, s, sectors, 0))
9850 return false;
9851
9852 /* Make sure they get written out promptly */
9853 if (test_bit(ExternalBbl, &rdev->flags))
9854 sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9855 sysfs_notify_dirent_safe(rdev->sysfs_state);
9856 set_mask_bits(&mddev->sb_flags, 0,
9857 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9858 md_wakeup_thread(rdev->mddev->thread);
9859 return true;
9860 }
9861 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9862
rdev_clear_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9863 void rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9864 int is_new)
9865 {
9866 if (is_new)
9867 s += rdev->new_data_offset;
9868 else
9869 s += rdev->data_offset;
9870
9871 if (!badblocks_clear(&rdev->badblocks, s, sectors))
9872 return;
9873
9874 if (test_bit(ExternalBbl, &rdev->flags))
9875 sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9876 }
9877 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9878
md_notify_reboot(struct notifier_block * this,unsigned long code,void * x)9879 static int md_notify_reboot(struct notifier_block *this,
9880 unsigned long code, void *x)
9881 {
9882 struct mddev *mddev;
9883 int need_delay = 0;
9884
9885 spin_lock(&all_mddevs_lock);
9886 list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
9887 if (!mddev_get(mddev))
9888 continue;
9889 spin_unlock(&all_mddevs_lock);
9890 if (mddev_trylock(mddev)) {
9891 if (mddev->pers)
9892 __md_stop_writes(mddev);
9893 if (mddev->persistent)
9894 mddev->safemode = 2;
9895 mddev_unlock(mddev);
9896 }
9897 need_delay = 1;
9898 spin_lock(&all_mddevs_lock);
9899 mddev_put_locked(mddev);
9900 }
9901 spin_unlock(&all_mddevs_lock);
9902
9903 /*
9904 * certain more exotic SCSI devices are known to be
9905 * volatile wrt too early system reboots. While the
9906 * right place to handle this issue is the given
9907 * driver, we do want to have a safe RAID driver ...
9908 */
9909 if (need_delay)
9910 msleep(1000);
9911
9912 return NOTIFY_DONE;
9913 }
9914
9915 static struct notifier_block md_notifier = {
9916 .notifier_call = md_notify_reboot,
9917 .next = NULL,
9918 .priority = INT_MAX, /* before any real devices */
9919 };
9920
md_geninit(void)9921 static void md_geninit(void)
9922 {
9923 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9924
9925 proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9926 }
9927
md_init(void)9928 static int __init md_init(void)
9929 {
9930 int ret = -ENOMEM;
9931
9932 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9933 if (!md_wq)
9934 goto err_wq;
9935
9936 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9937 if (!md_misc_wq)
9938 goto err_misc_wq;
9939
9940 md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
9941 0);
9942 if (!md_bitmap_wq)
9943 goto err_bitmap_wq;
9944
9945 ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9946 if (ret < 0)
9947 goto err_md;
9948
9949 ret = __register_blkdev(0, "mdp", md_probe);
9950 if (ret < 0)
9951 goto err_mdp;
9952 mdp_major = ret;
9953
9954 register_reboot_notifier(&md_notifier);
9955 raid_table_header = register_sysctl("dev/raid", raid_table);
9956
9957 md_geninit();
9958 return 0;
9959
9960 err_mdp:
9961 unregister_blkdev(MD_MAJOR, "md");
9962 err_md:
9963 destroy_workqueue(md_bitmap_wq);
9964 err_bitmap_wq:
9965 destroy_workqueue(md_misc_wq);
9966 err_misc_wq:
9967 destroy_workqueue(md_wq);
9968 err_wq:
9969 return ret;
9970 }
9971
check_sb_changes(struct mddev * mddev,struct md_rdev * rdev)9972 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9973 {
9974 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9975 struct md_rdev *rdev2, *tmp;
9976 int role, ret;
9977
9978 /*
9979 * If size is changed in another node then we need to
9980 * do resize as well.
9981 */
9982 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9983 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9984 if (ret)
9985 pr_info("md-cluster: resize failed\n");
9986 else
9987 mddev->bitmap_ops->update_sb(mddev->bitmap);
9988 }
9989
9990 /* Check for change of roles in the active devices */
9991 rdev_for_each_safe(rdev2, tmp, mddev) {
9992 if (test_bit(Faulty, &rdev2->flags))
9993 continue;
9994
9995 /* Check if the roles changed */
9996 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9997
9998 if (test_bit(Candidate, &rdev2->flags)) {
9999 if (role == MD_DISK_ROLE_FAULTY) {
10000 pr_info("md: Removing Candidate device %pg because add failed\n",
10001 rdev2->bdev);
10002 md_kick_rdev_from_array(rdev2);
10003 continue;
10004 }
10005 else
10006 clear_bit(Candidate, &rdev2->flags);
10007 }
10008
10009 if (role != rdev2->raid_disk) {
10010 /*
10011 * got activated except reshape is happening.
10012 */
10013 if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
10014 !(le32_to_cpu(sb->feature_map) &
10015 MD_FEATURE_RESHAPE_ACTIVE) &&
10016 !mddev->cluster_ops->resync_status_get(mddev)) {
10017 /*
10018 * -1 to make raid1_add_disk() set conf->fullsync
10019 * to 1. This could avoid skipping sync when the
10020 * remote node is down during resyncing.
10021 */
10022 if ((le32_to_cpu(sb->feature_map)
10023 & MD_FEATURE_RECOVERY_OFFSET))
10024 rdev2->saved_raid_disk = -1;
10025 else
10026 rdev2->saved_raid_disk = role;
10027 ret = remove_and_add_spares(mddev, rdev2);
10028 pr_info("Activated spare: %pg\n",
10029 rdev2->bdev);
10030 /* wakeup mddev->thread here, so array could
10031 * perform resync with the new activated disk */
10032 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
10033 md_wakeup_thread(mddev->thread);
10034 }
10035 /* device faulty
10036 * We just want to do the minimum to mark the disk
10037 * as faulty. The recovery is performed by the
10038 * one who initiated the error.
10039 */
10040 if (role == MD_DISK_ROLE_FAULTY ||
10041 role == MD_DISK_ROLE_JOURNAL) {
10042 md_error(mddev, rdev2);
10043 clear_bit(Blocked, &rdev2->flags);
10044 }
10045 }
10046 }
10047
10048 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
10049 ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
10050 if (ret)
10051 pr_warn("md: updating array disks failed. %d\n", ret);
10052 }
10053
10054 /*
10055 * Since mddev->delta_disks has already updated in update_raid_disks,
10056 * so it is time to check reshape.
10057 */
10058 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
10059 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
10060 /*
10061 * reshape is happening in the remote node, we need to
10062 * update reshape_position and call start_reshape.
10063 */
10064 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
10065 if (mddev->pers->update_reshape_pos)
10066 mddev->pers->update_reshape_pos(mddev);
10067 if (mddev->pers->start_reshape)
10068 mddev->pers->start_reshape(mddev);
10069 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
10070 mddev->reshape_position != MaxSector &&
10071 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
10072 /* reshape is just done in another node. */
10073 mddev->reshape_position = MaxSector;
10074 if (mddev->pers->update_reshape_pos)
10075 mddev->pers->update_reshape_pos(mddev);
10076 }
10077
10078 /* Finally set the event to be up to date */
10079 mddev->events = le64_to_cpu(sb->events);
10080 }
10081
read_rdev(struct mddev * mddev,struct md_rdev * rdev)10082 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
10083 {
10084 int err;
10085 struct page *swapout = rdev->sb_page;
10086 struct mdp_superblock_1 *sb;
10087
10088 /* Store the sb page of the rdev in the swapout temporary
10089 * variable in case we err in the future
10090 */
10091 rdev->sb_page = NULL;
10092 err = alloc_disk_sb(rdev);
10093 if (err == 0) {
10094 ClearPageUptodate(rdev->sb_page);
10095 rdev->sb_loaded = 0;
10096 err = super_types[mddev->major_version].
10097 load_super(rdev, NULL, mddev->minor_version);
10098 }
10099 if (err < 0) {
10100 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
10101 __func__, __LINE__, rdev->desc_nr, err);
10102 if (rdev->sb_page)
10103 put_page(rdev->sb_page);
10104 rdev->sb_page = swapout;
10105 rdev->sb_loaded = 1;
10106 return err;
10107 }
10108
10109 sb = page_address(rdev->sb_page);
10110 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
10111 * is not set
10112 */
10113
10114 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
10115 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
10116
10117 /* The other node finished recovery, call spare_active to set
10118 * device In_sync and mddev->degraded
10119 */
10120 if (rdev->recovery_offset == MaxSector &&
10121 !test_bit(In_sync, &rdev->flags) &&
10122 mddev->pers->spare_active(mddev))
10123 sysfs_notify_dirent_safe(mddev->sysfs_degraded);
10124
10125 put_page(swapout);
10126 return 0;
10127 }
10128
md_reload_sb(struct mddev * mddev,int nr)10129 void md_reload_sb(struct mddev *mddev, int nr)
10130 {
10131 struct md_rdev *rdev = NULL, *iter;
10132 int err;
10133
10134 /* Find the rdev */
10135 rdev_for_each_rcu(iter, mddev) {
10136 if (iter->desc_nr == nr) {
10137 rdev = iter;
10138 break;
10139 }
10140 }
10141
10142 if (!rdev) {
10143 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
10144 return;
10145 }
10146
10147 err = read_rdev(mddev, rdev);
10148 if (err < 0)
10149 return;
10150
10151 check_sb_changes(mddev, rdev);
10152
10153 /* Read all rdev's to update recovery_offset */
10154 rdev_for_each_rcu(rdev, mddev) {
10155 if (!test_bit(Faulty, &rdev->flags))
10156 read_rdev(mddev, rdev);
10157 }
10158 }
10159 EXPORT_SYMBOL(md_reload_sb);
10160
10161 #ifndef MODULE
10162
10163 /*
10164 * Searches all registered partitions for autorun RAID arrays
10165 * at boot time.
10166 */
10167
10168 static DEFINE_MUTEX(detected_devices_mutex);
10169 static LIST_HEAD(all_detected_devices);
10170 struct detected_devices_node {
10171 struct list_head list;
10172 dev_t dev;
10173 };
10174
md_autodetect_dev(dev_t dev)10175 void md_autodetect_dev(dev_t dev)
10176 {
10177 struct detected_devices_node *node_detected_dev;
10178
10179 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
10180 if (node_detected_dev) {
10181 node_detected_dev->dev = dev;
10182 mutex_lock(&detected_devices_mutex);
10183 list_add_tail(&node_detected_dev->list, &all_detected_devices);
10184 mutex_unlock(&detected_devices_mutex);
10185 }
10186 }
10187
md_autostart_arrays(int part)10188 void md_autostart_arrays(int part)
10189 {
10190 struct md_rdev *rdev;
10191 struct detected_devices_node *node_detected_dev;
10192 dev_t dev;
10193 int i_scanned, i_passed;
10194
10195 i_scanned = 0;
10196 i_passed = 0;
10197
10198 pr_info("md: Autodetecting RAID arrays.\n");
10199
10200 mutex_lock(&detected_devices_mutex);
10201 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
10202 i_scanned++;
10203 node_detected_dev = list_entry(all_detected_devices.next,
10204 struct detected_devices_node, list);
10205 list_del(&node_detected_dev->list);
10206 dev = node_detected_dev->dev;
10207 kfree(node_detected_dev);
10208 mutex_unlock(&detected_devices_mutex);
10209 rdev = md_import_device(dev,0, 90);
10210 mutex_lock(&detected_devices_mutex);
10211 if (IS_ERR(rdev))
10212 continue;
10213
10214 if (test_bit(Faulty, &rdev->flags))
10215 continue;
10216
10217 set_bit(AutoDetected, &rdev->flags);
10218 list_add(&rdev->same_set, &pending_raid_disks);
10219 i_passed++;
10220 }
10221 mutex_unlock(&detected_devices_mutex);
10222
10223 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
10224
10225 autorun_devices(part);
10226 }
10227
10228 #endif /* !MODULE */
10229
md_exit(void)10230 static __exit void md_exit(void)
10231 {
10232 struct mddev *mddev;
10233 int delay = 1;
10234
10235 unregister_blkdev(MD_MAJOR,"md");
10236 unregister_blkdev(mdp_major, "mdp");
10237 unregister_reboot_notifier(&md_notifier);
10238 unregister_sysctl_table(raid_table_header);
10239
10240 /* We cannot unload the modules while some process is
10241 * waiting for us in select() or poll() - wake them up
10242 */
10243 md_unloading = 1;
10244 while (waitqueue_active(&md_event_waiters)) {
10245 /* not safe to leave yet */
10246 wake_up(&md_event_waiters);
10247 msleep(delay);
10248 delay += delay;
10249 }
10250 remove_proc_entry("mdstat", NULL);
10251
10252 spin_lock(&all_mddevs_lock);
10253 list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
10254 if (!mddev_get(mddev))
10255 continue;
10256 spin_unlock(&all_mddevs_lock);
10257 export_array(mddev);
10258 mddev->ctime = 0;
10259 mddev->hold_active = 0;
10260 /*
10261 * As the mddev is now fully clear, mddev_put will schedule
10262 * the mddev for destruction by a workqueue, and the
10263 * destroy_workqueue() below will wait for that to complete.
10264 */
10265 spin_lock(&all_mddevs_lock);
10266 mddev_put_locked(mddev);
10267 }
10268 spin_unlock(&all_mddevs_lock);
10269
10270 destroy_workqueue(md_misc_wq);
10271 destroy_workqueue(md_bitmap_wq);
10272 destroy_workqueue(md_wq);
10273 }
10274
10275 subsys_initcall(md_init);
module_exit(md_exit)10276 module_exit(md_exit)
10277
10278 static int get_ro(char *buffer, const struct kernel_param *kp)
10279 {
10280 return sprintf(buffer, "%d\n", start_readonly);
10281 }
set_ro(const char * val,const struct kernel_param * kp)10282 static int set_ro(const char *val, const struct kernel_param *kp)
10283 {
10284 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
10285 }
10286
10287 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
10288 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10289 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10290 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10291
10292 MODULE_LICENSE("GPL");
10293 MODULE_DESCRIPTION("MD RAID framework");
10294 MODULE_ALIAS("md");
10295 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
10296