xref: /linux/drivers/md/md.c (revision 3fd6c59042dbba50391e30862beac979491145fe)
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 /* pers_list is a list of registered personalities protected by pers_lock. */
83 static LIST_HEAD(pers_list);
84 static DEFINE_SPINLOCK(pers_lock);
85 
86 static const struct kobj_type md_ktype;
87 
88 const struct md_cluster_operations *md_cluster_ops;
89 EXPORT_SYMBOL(md_cluster_ops);
90 static struct module *md_cluster_mod;
91 
92 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
93 static struct workqueue_struct *md_wq;
94 
95 /*
96  * This workqueue is used for sync_work to register new sync_thread, and for
97  * del_work to remove rdev, and for event_work that is only set by dm-raid.
98  *
99  * Noted that sync_work will grab reconfig_mutex, hence never flush this
100  * workqueue whith reconfig_mutex grabbed.
101  */
102 static struct workqueue_struct *md_misc_wq;
103 struct workqueue_struct *md_bitmap_wq;
104 
105 static int remove_and_add_spares(struct mddev *mddev,
106 				 struct md_rdev *this);
107 static void mddev_detach(struct mddev *mddev);
108 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev);
109 static void md_wakeup_thread_directly(struct md_thread __rcu *thread);
110 
111 /*
112  * Default number of read corrections we'll attempt on an rdev
113  * before ejecting it from the array. We divide the read error
114  * count by 2 for every hour elapsed between read errors.
115  */
116 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
117 /* Default safemode delay: 200 msec */
118 #define DEFAULT_SAFEMODE_DELAY ((200 * HZ)/1000 +1)
119 /*
120  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
121  * is 1000 KB/sec, so the extra system load does not show up that much.
122  * Increase it if you want to have more _guaranteed_ speed. Note that
123  * the RAID driver will use the maximum available bandwidth if the IO
124  * subsystem is idle. There is also an 'absolute maximum' reconstruction
125  * speed limit - in case reconstruction slows down your system despite
126  * idle IO detection.
127  *
128  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
129  * or /sys/block/mdX/md/sync_speed_{min,max}
130  */
131 
132 static int sysctl_speed_limit_min = 1000;
133 static int sysctl_speed_limit_max = 200000;
speed_min(struct mddev * mddev)134 static inline int speed_min(struct mddev *mddev)
135 {
136 	return mddev->sync_speed_min ?
137 		mddev->sync_speed_min : sysctl_speed_limit_min;
138 }
139 
speed_max(struct mddev * mddev)140 static inline int speed_max(struct mddev *mddev)
141 {
142 	return mddev->sync_speed_max ?
143 		mddev->sync_speed_max : sysctl_speed_limit_max;
144 }
145 
rdev_uninit_serial(struct md_rdev * rdev)146 static void rdev_uninit_serial(struct md_rdev *rdev)
147 {
148 	if (!test_and_clear_bit(CollisionCheck, &rdev->flags))
149 		return;
150 
151 	kvfree(rdev->serial);
152 	rdev->serial = NULL;
153 }
154 
rdevs_uninit_serial(struct mddev * mddev)155 static void rdevs_uninit_serial(struct mddev *mddev)
156 {
157 	struct md_rdev *rdev;
158 
159 	rdev_for_each(rdev, mddev)
160 		rdev_uninit_serial(rdev);
161 }
162 
rdev_init_serial(struct md_rdev * rdev)163 static int rdev_init_serial(struct md_rdev *rdev)
164 {
165 	/* serial_nums equals with BARRIER_BUCKETS_NR */
166 	int i, serial_nums = 1 << ((PAGE_SHIFT - ilog2(sizeof(atomic_t))));
167 	struct serial_in_rdev *serial = NULL;
168 
169 	if (test_bit(CollisionCheck, &rdev->flags))
170 		return 0;
171 
172 	serial = kvmalloc(sizeof(struct serial_in_rdev) * serial_nums,
173 			  GFP_KERNEL);
174 	if (!serial)
175 		return -ENOMEM;
176 
177 	for (i = 0; i < serial_nums; i++) {
178 		struct serial_in_rdev *serial_tmp = &serial[i];
179 
180 		spin_lock_init(&serial_tmp->serial_lock);
181 		serial_tmp->serial_rb = RB_ROOT_CACHED;
182 		init_waitqueue_head(&serial_tmp->serial_io_wait);
183 	}
184 
185 	rdev->serial = serial;
186 	set_bit(CollisionCheck, &rdev->flags);
187 
188 	return 0;
189 }
190 
rdevs_init_serial(struct mddev * mddev)191 static int rdevs_init_serial(struct mddev *mddev)
192 {
193 	struct md_rdev *rdev;
194 	int ret = 0;
195 
196 	rdev_for_each(rdev, mddev) {
197 		ret = rdev_init_serial(rdev);
198 		if (ret)
199 			break;
200 	}
201 
202 	/* Free all resources if pool is not existed */
203 	if (ret && !mddev->serial_info_pool)
204 		rdevs_uninit_serial(mddev);
205 
206 	return ret;
207 }
208 
209 /*
210  * rdev needs to enable serial stuffs if it meets the conditions:
211  * 1. it is multi-queue device flaged with writemostly.
212  * 2. the write-behind mode is enabled.
213  */
rdev_need_serial(struct md_rdev * rdev)214 static int rdev_need_serial(struct md_rdev *rdev)
215 {
216 	return (rdev && rdev->mddev->bitmap_info.max_write_behind > 0 &&
217 		rdev->bdev->bd_disk->queue->nr_hw_queues != 1 &&
218 		test_bit(WriteMostly, &rdev->flags));
219 }
220 
221 /*
222  * Init resource for rdev(s), then create serial_info_pool if:
223  * 1. rdev is the first device which return true from rdev_enable_serial.
224  * 2. rdev is NULL, means we want to enable serialization for all rdevs.
225  */
mddev_create_serial_pool(struct mddev * mddev,struct md_rdev * rdev)226 void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev)
227 {
228 	int ret = 0;
229 
230 	if (rdev && !rdev_need_serial(rdev) &&
231 	    !test_bit(CollisionCheck, &rdev->flags))
232 		return;
233 
234 	if (!rdev)
235 		ret = rdevs_init_serial(mddev);
236 	else
237 		ret = rdev_init_serial(rdev);
238 	if (ret)
239 		return;
240 
241 	if (mddev->serial_info_pool == NULL) {
242 		/*
243 		 * already in memalloc noio context by
244 		 * mddev_suspend()
245 		 */
246 		mddev->serial_info_pool =
247 			mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
248 						sizeof(struct serial_info));
249 		if (!mddev->serial_info_pool) {
250 			rdevs_uninit_serial(mddev);
251 			pr_err("can't alloc memory pool for serialization\n");
252 		}
253 	}
254 }
255 
256 /*
257  * Free resource from rdev(s), and destroy serial_info_pool under conditions:
258  * 1. rdev is the last device flaged with CollisionCheck.
259  * 2. when bitmap is destroyed while policy is not enabled.
260  * 3. for disable policy, the pool is destroyed only when no rdev needs it.
261  */
mddev_destroy_serial_pool(struct mddev * mddev,struct md_rdev * rdev)262 void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev)
263 {
264 	if (rdev && !test_bit(CollisionCheck, &rdev->flags))
265 		return;
266 
267 	if (mddev->serial_info_pool) {
268 		struct md_rdev *temp;
269 		int num = 0; /* used to track if other rdevs need the pool */
270 
271 		rdev_for_each(temp, mddev) {
272 			if (!rdev) {
273 				if (!mddev->serialize_policy ||
274 				    !rdev_need_serial(temp))
275 					rdev_uninit_serial(temp);
276 				else
277 					num++;
278 			} else if (temp != rdev &&
279 				   test_bit(CollisionCheck, &temp->flags))
280 				num++;
281 		}
282 
283 		if (rdev)
284 			rdev_uninit_serial(rdev);
285 
286 		if (num)
287 			pr_info("The mempool could be used by other devices\n");
288 		else {
289 			mempool_destroy(mddev->serial_info_pool);
290 			mddev->serial_info_pool = NULL;
291 		}
292 	}
293 }
294 
295 static struct ctl_table_header *raid_table_header;
296 
297 static struct ctl_table raid_table[] = {
298 	{
299 		.procname	= "speed_limit_min",
300 		.data		= &sysctl_speed_limit_min,
301 		.maxlen		= sizeof(int),
302 		.mode		= S_IRUGO|S_IWUSR,
303 		.proc_handler	= proc_dointvec,
304 	},
305 	{
306 		.procname	= "speed_limit_max",
307 		.data		= &sysctl_speed_limit_max,
308 		.maxlen		= sizeof(int),
309 		.mode		= S_IRUGO|S_IWUSR,
310 		.proc_handler	= proc_dointvec,
311 	},
312 };
313 
314 static int start_readonly;
315 
316 /*
317  * The original mechanism for creating an md device is to create
318  * a device node in /dev and to open it.  This causes races with device-close.
319  * The preferred method is to write to the "new_array" module parameter.
320  * This can avoid races.
321  * Setting create_on_open to false disables the original mechanism
322  * so all the races disappear.
323  */
324 static bool create_on_open = true;
325 
326 /*
327  * We have a system wide 'event count' that is incremented
328  * on any 'interesting' event, and readers of /proc/mdstat
329  * can use 'poll' or 'select' to find out when the event
330  * count increases.
331  *
332  * Events are:
333  *  start array, stop array, error, add device, remove device,
334  *  start build, activate spare
335  */
336 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
337 static atomic_t md_event_count;
md_new_event(void)338 void md_new_event(void)
339 {
340 	atomic_inc(&md_event_count);
341 	wake_up(&md_event_waiters);
342 }
343 EXPORT_SYMBOL_GPL(md_new_event);
344 
345 /*
346  * Enables to iterate over all existing md arrays
347  * all_mddevs_lock protects this list.
348  */
349 static LIST_HEAD(all_mddevs);
350 static DEFINE_SPINLOCK(all_mddevs_lock);
351 
is_md_suspended(struct mddev * mddev)352 static bool is_md_suspended(struct mddev *mddev)
353 {
354 	return percpu_ref_is_dying(&mddev->active_io);
355 }
356 /* Rather than calling directly into the personality make_request function,
357  * IO requests come here first so that we can check if the device is
358  * being suspended pending a reconfiguration.
359  * We hold a refcount over the call to ->make_request.  By the time that
360  * call has finished, the bio has been linked into some internal structure
361  * and so is visible to ->quiesce(), so we don't need the refcount any more.
362  */
is_suspended(struct mddev * mddev,struct bio * bio)363 static bool is_suspended(struct mddev *mddev, struct bio *bio)
364 {
365 	if (is_md_suspended(mddev))
366 		return true;
367 	if (bio_data_dir(bio) != WRITE)
368 		return false;
369 	if (READ_ONCE(mddev->suspend_lo) >= READ_ONCE(mddev->suspend_hi))
370 		return false;
371 	if (bio->bi_iter.bi_sector >= READ_ONCE(mddev->suspend_hi))
372 		return false;
373 	if (bio_end_sector(bio) < READ_ONCE(mddev->suspend_lo))
374 		return false;
375 	return true;
376 }
377 
md_handle_request(struct mddev * mddev,struct bio * bio)378 bool md_handle_request(struct mddev *mddev, struct bio *bio)
379 {
380 check_suspended:
381 	if (is_suspended(mddev, bio)) {
382 		DEFINE_WAIT(__wait);
383 		/* Bail out if REQ_NOWAIT is set for the bio */
384 		if (bio->bi_opf & REQ_NOWAIT) {
385 			bio_wouldblock_error(bio);
386 			return true;
387 		}
388 		for (;;) {
389 			prepare_to_wait(&mddev->sb_wait, &__wait,
390 					TASK_UNINTERRUPTIBLE);
391 			if (!is_suspended(mddev, bio))
392 				break;
393 			schedule();
394 		}
395 		finish_wait(&mddev->sb_wait, &__wait);
396 	}
397 	if (!percpu_ref_tryget_live(&mddev->active_io))
398 		goto check_suspended;
399 
400 	if (!mddev->pers->make_request(mddev, bio)) {
401 		percpu_ref_put(&mddev->active_io);
402 		if (!mddev->gendisk && mddev->pers->prepare_suspend)
403 			return false;
404 		goto check_suspended;
405 	}
406 
407 	percpu_ref_put(&mddev->active_io);
408 	return true;
409 }
410 EXPORT_SYMBOL(md_handle_request);
411 
md_submit_bio(struct bio * bio)412 static void md_submit_bio(struct bio *bio)
413 {
414 	const int rw = bio_data_dir(bio);
415 	struct mddev *mddev = bio->bi_bdev->bd_disk->private_data;
416 
417 	if (mddev == NULL || mddev->pers == NULL) {
418 		bio_io_error(bio);
419 		return;
420 	}
421 
422 	if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
423 		bio_io_error(bio);
424 		return;
425 	}
426 
427 	bio = bio_split_to_limits(bio);
428 	if (!bio)
429 		return;
430 
431 	if (mddev->ro == MD_RDONLY && unlikely(rw == WRITE)) {
432 		if (bio_sectors(bio) != 0)
433 			bio->bi_status = BLK_STS_IOERR;
434 		bio_endio(bio);
435 		return;
436 	}
437 
438 	/* bio could be mergeable after passing to underlayer */
439 	bio->bi_opf &= ~REQ_NOMERGE;
440 
441 	md_handle_request(mddev, bio);
442 }
443 
444 /*
445  * Make sure no new requests are submitted to the device, and any requests that
446  * have been submitted are completely handled.
447  */
mddev_suspend(struct mddev * mddev,bool interruptible)448 int mddev_suspend(struct mddev *mddev, bool interruptible)
449 {
450 	int err = 0;
451 
452 	/*
453 	 * hold reconfig_mutex to wait for normal io will deadlock, because
454 	 * other context can't update super_block, and normal io can rely on
455 	 * updating super_block.
456 	 */
457 	lockdep_assert_not_held(&mddev->reconfig_mutex);
458 
459 	if (interruptible)
460 		err = mutex_lock_interruptible(&mddev->suspend_mutex);
461 	else
462 		mutex_lock(&mddev->suspend_mutex);
463 	if (err)
464 		return err;
465 
466 	if (mddev->suspended) {
467 		WRITE_ONCE(mddev->suspended, mddev->suspended + 1);
468 		mutex_unlock(&mddev->suspend_mutex);
469 		return 0;
470 	}
471 
472 	percpu_ref_kill(&mddev->active_io);
473 	if (interruptible)
474 		err = wait_event_interruptible(mddev->sb_wait,
475 				percpu_ref_is_zero(&mddev->active_io));
476 	else
477 		wait_event(mddev->sb_wait,
478 				percpu_ref_is_zero(&mddev->active_io));
479 	if (err) {
480 		percpu_ref_resurrect(&mddev->active_io);
481 		mutex_unlock(&mddev->suspend_mutex);
482 		return err;
483 	}
484 
485 	/*
486 	 * For raid456, io might be waiting for reshape to make progress,
487 	 * allow new reshape to start while waiting for io to be done to
488 	 * prevent deadlock.
489 	 */
490 	WRITE_ONCE(mddev->suspended, mddev->suspended + 1);
491 
492 	/* restrict memory reclaim I/O during raid array is suspend */
493 	mddev->noio_flag = memalloc_noio_save();
494 
495 	mutex_unlock(&mddev->suspend_mutex);
496 	return 0;
497 }
498 EXPORT_SYMBOL_GPL(mddev_suspend);
499 
__mddev_resume(struct mddev * mddev,bool recovery_needed)500 static void __mddev_resume(struct mddev *mddev, bool recovery_needed)
501 {
502 	lockdep_assert_not_held(&mddev->reconfig_mutex);
503 
504 	mutex_lock(&mddev->suspend_mutex);
505 	WRITE_ONCE(mddev->suspended, mddev->suspended - 1);
506 	if (mddev->suspended) {
507 		mutex_unlock(&mddev->suspend_mutex);
508 		return;
509 	}
510 
511 	/* entred the memalloc scope from mddev_suspend() */
512 	memalloc_noio_restore(mddev->noio_flag);
513 
514 	percpu_ref_resurrect(&mddev->active_io);
515 	wake_up(&mddev->sb_wait);
516 
517 	if (recovery_needed)
518 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
519 	md_wakeup_thread(mddev->thread);
520 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
521 
522 	mutex_unlock(&mddev->suspend_mutex);
523 }
524 
mddev_resume(struct mddev * mddev)525 void mddev_resume(struct mddev *mddev)
526 {
527 	return __mddev_resume(mddev, true);
528 }
529 EXPORT_SYMBOL_GPL(mddev_resume);
530 
531 /* sync bdev before setting device to readonly or stopping raid*/
mddev_set_closing_and_sync_blockdev(struct mddev * mddev,int opener_num)532 static int mddev_set_closing_and_sync_blockdev(struct mddev *mddev, int opener_num)
533 {
534 	mutex_lock(&mddev->open_mutex);
535 	if (mddev->pers && atomic_read(&mddev->openers) > opener_num) {
536 		mutex_unlock(&mddev->open_mutex);
537 		return -EBUSY;
538 	}
539 	if (test_and_set_bit(MD_CLOSING, &mddev->flags)) {
540 		mutex_unlock(&mddev->open_mutex);
541 		return -EBUSY;
542 	}
543 	mutex_unlock(&mddev->open_mutex);
544 
545 	sync_blockdev(mddev->gendisk->part0);
546 	return 0;
547 }
548 
549 /*
550  * The only difference from bio_chain_endio() is that the current
551  * bi_status of bio does not affect the bi_status of parent.
552  */
md_end_flush(struct bio * bio)553 static void md_end_flush(struct bio *bio)
554 {
555 	struct bio *parent = bio->bi_private;
556 
557 	/*
558 	 * If any flush io error before the power failure,
559 	 * disk data may be lost.
560 	 */
561 	if (bio->bi_status)
562 		pr_err("md: %pg flush io error %d\n", bio->bi_bdev,
563 			blk_status_to_errno(bio->bi_status));
564 
565 	bio_put(bio);
566 	bio_endio(parent);
567 }
568 
md_flush_request(struct mddev * mddev,struct bio * bio)569 bool md_flush_request(struct mddev *mddev, struct bio *bio)
570 {
571 	struct md_rdev *rdev;
572 	struct bio *new;
573 
574 	/*
575 	 * md_flush_reqeust() should be called under md_handle_request() and
576 	 * 'active_io' is already grabbed. Hence it's safe to get rdev directly
577 	 * without rcu protection.
578 	 */
579 	WARN_ON(percpu_ref_is_zero(&mddev->active_io));
580 
581 	rdev_for_each(rdev, mddev) {
582 		if (rdev->raid_disk < 0 || test_bit(Faulty, &rdev->flags))
583 			continue;
584 
585 		new = bio_alloc_bioset(rdev->bdev, 0,
586 				       REQ_OP_WRITE | REQ_PREFLUSH, GFP_NOIO,
587 				       &mddev->bio_set);
588 		new->bi_private = bio;
589 		new->bi_end_io = md_end_flush;
590 		bio_inc_remaining(bio);
591 		submit_bio(new);
592 	}
593 
594 	if (bio_sectors(bio) == 0) {
595 		bio_endio(bio);
596 		return true;
597 	}
598 
599 	bio->bi_opf &= ~REQ_PREFLUSH;
600 	return false;
601 }
602 EXPORT_SYMBOL(md_flush_request);
603 
mddev_get(struct mddev * mddev)604 static inline struct mddev *mddev_get(struct mddev *mddev)
605 {
606 	lockdep_assert_held(&all_mddevs_lock);
607 
608 	if (test_bit(MD_DELETED, &mddev->flags))
609 		return NULL;
610 	atomic_inc(&mddev->active);
611 	return mddev;
612 }
613 
614 static void mddev_delayed_delete(struct work_struct *ws);
615 
__mddev_put(struct mddev * mddev)616 static void __mddev_put(struct mddev *mddev)
617 {
618 	if (mddev->raid_disks || !list_empty(&mddev->disks) ||
619 	    mddev->ctime || mddev->hold_active)
620 		return;
621 
622 	/* Array is not configured at all, and not held active, so destroy it */
623 	set_bit(MD_DELETED, &mddev->flags);
624 
625 	/*
626 	 * Call queue_work inside the spinlock so that flush_workqueue() after
627 	 * mddev_find will succeed in waiting for the work to be done.
628 	 */
629 	queue_work(md_misc_wq, &mddev->del_work);
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 
find_pers(int level,char * clevel)891 static struct md_personality *find_pers(int level, char *clevel)
892 {
893 	struct md_personality *pers;
894 	list_for_each_entry(pers, &pers_list, list) {
895 		if (level != LEVEL_NONE && pers->level == level)
896 			return pers;
897 		if (strcmp(pers->name, clevel)==0)
898 			return pers;
899 	}
900 	return NULL;
901 }
902 
903 /* return the offset of the super block in 512byte sectors */
calc_dev_sboffset(struct md_rdev * rdev)904 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
905 {
906 	return MD_NEW_SIZE_SECTORS(bdev_nr_sectors(rdev->bdev));
907 }
908 
alloc_disk_sb(struct md_rdev * rdev)909 static int alloc_disk_sb(struct md_rdev *rdev)
910 {
911 	rdev->sb_page = alloc_page(GFP_KERNEL);
912 	if (!rdev->sb_page)
913 		return -ENOMEM;
914 	return 0;
915 }
916 
md_rdev_clear(struct md_rdev * rdev)917 void md_rdev_clear(struct md_rdev *rdev)
918 {
919 	if (rdev->sb_page) {
920 		put_page(rdev->sb_page);
921 		rdev->sb_loaded = 0;
922 		rdev->sb_page = NULL;
923 		rdev->sb_start = 0;
924 		rdev->sectors = 0;
925 	}
926 	if (rdev->bb_page) {
927 		put_page(rdev->bb_page);
928 		rdev->bb_page = NULL;
929 	}
930 	badblocks_exit(&rdev->badblocks);
931 }
932 EXPORT_SYMBOL_GPL(md_rdev_clear);
933 
super_written(struct bio * bio)934 static void super_written(struct bio *bio)
935 {
936 	struct md_rdev *rdev = bio->bi_private;
937 	struct mddev *mddev = rdev->mddev;
938 
939 	if (bio->bi_status) {
940 		pr_err("md: %s gets error=%d\n", __func__,
941 		       blk_status_to_errno(bio->bi_status));
942 		md_error(mddev, rdev);
943 		if (!test_bit(Faulty, &rdev->flags)
944 		    && (bio->bi_opf & MD_FAILFAST)) {
945 			set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
946 			set_bit(LastDev, &rdev->flags);
947 		}
948 	} else
949 		clear_bit(LastDev, &rdev->flags);
950 
951 	bio_put(bio);
952 
953 	rdev_dec_pending(rdev, mddev);
954 
955 	if (atomic_dec_and_test(&mddev->pending_writes))
956 		wake_up(&mddev->sb_wait);
957 }
958 
md_super_write(struct mddev * mddev,struct md_rdev * rdev,sector_t sector,int size,struct page * page)959 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
960 		   sector_t sector, int size, struct page *page)
961 {
962 	/* write first size bytes of page to sector of rdev
963 	 * Increment mddev->pending_writes before returning
964 	 * and decrement it on completion, waking up sb_wait
965 	 * if zero is reached.
966 	 * If an error occurred, call md_error
967 	 */
968 	struct bio *bio;
969 
970 	if (!page)
971 		return;
972 
973 	if (test_bit(Faulty, &rdev->flags))
974 		return;
975 
976 	bio = bio_alloc_bioset(rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev,
977 			      1,
978 			      REQ_OP_WRITE | REQ_SYNC | REQ_IDLE | REQ_META
979 				  | REQ_PREFLUSH | REQ_FUA,
980 			      GFP_NOIO, &mddev->sync_set);
981 
982 	atomic_inc(&rdev->nr_pending);
983 
984 	bio->bi_iter.bi_sector = sector;
985 	__bio_add_page(bio, page, size, 0);
986 	bio->bi_private = rdev;
987 	bio->bi_end_io = super_written;
988 
989 	if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
990 	    test_bit(FailFast, &rdev->flags) &&
991 	    !test_bit(LastDev, &rdev->flags))
992 		bio->bi_opf |= MD_FAILFAST;
993 
994 	atomic_inc(&mddev->pending_writes);
995 	submit_bio(bio);
996 }
997 
md_super_wait(struct mddev * mddev)998 int md_super_wait(struct mddev *mddev)
999 {
1000 	/* wait for all superblock writes that were scheduled to complete */
1001 	wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1002 	if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
1003 		return -EAGAIN;
1004 	return 0;
1005 }
1006 
sync_page_io(struct md_rdev * rdev,sector_t sector,int size,struct page * page,blk_opf_t opf,bool metadata_op)1007 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
1008 		 struct page *page, blk_opf_t opf, bool metadata_op)
1009 {
1010 	struct bio bio;
1011 	struct bio_vec bvec;
1012 
1013 	if (metadata_op && rdev->meta_bdev)
1014 		bio_init(&bio, rdev->meta_bdev, &bvec, 1, opf);
1015 	else
1016 		bio_init(&bio, rdev->bdev, &bvec, 1, opf);
1017 
1018 	if (metadata_op)
1019 		bio.bi_iter.bi_sector = sector + rdev->sb_start;
1020 	else if (rdev->mddev->reshape_position != MaxSector &&
1021 		 (rdev->mddev->reshape_backwards ==
1022 		  (sector >= rdev->mddev->reshape_position)))
1023 		bio.bi_iter.bi_sector = sector + rdev->new_data_offset;
1024 	else
1025 		bio.bi_iter.bi_sector = sector + rdev->data_offset;
1026 	__bio_add_page(&bio, page, size, 0);
1027 
1028 	submit_bio_wait(&bio);
1029 
1030 	return !bio.bi_status;
1031 }
1032 EXPORT_SYMBOL_GPL(sync_page_io);
1033 
read_disk_sb(struct md_rdev * rdev,int size)1034 static int read_disk_sb(struct md_rdev *rdev, int size)
1035 {
1036 	if (rdev->sb_loaded)
1037 		return 0;
1038 
1039 	if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, true))
1040 		goto fail;
1041 	rdev->sb_loaded = 1;
1042 	return 0;
1043 
1044 fail:
1045 	pr_err("md: disabled device %pg, could not read superblock.\n",
1046 	       rdev->bdev);
1047 	return -EINVAL;
1048 }
1049 
md_uuid_equal(mdp_super_t * sb1,mdp_super_t * sb2)1050 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1051 {
1052 	return	sb1->set_uuid0 == sb2->set_uuid0 &&
1053 		sb1->set_uuid1 == sb2->set_uuid1 &&
1054 		sb1->set_uuid2 == sb2->set_uuid2 &&
1055 		sb1->set_uuid3 == sb2->set_uuid3;
1056 }
1057 
md_sb_equal(mdp_super_t * sb1,mdp_super_t * sb2)1058 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1059 {
1060 	int ret;
1061 	mdp_super_t *tmp1, *tmp2;
1062 
1063 	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
1064 	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
1065 
1066 	if (!tmp1 || !tmp2) {
1067 		ret = 0;
1068 		goto abort;
1069 	}
1070 
1071 	*tmp1 = *sb1;
1072 	*tmp2 = *sb2;
1073 
1074 	/*
1075 	 * nr_disks is not constant
1076 	 */
1077 	tmp1->nr_disks = 0;
1078 	tmp2->nr_disks = 0;
1079 
1080 	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1081 abort:
1082 	kfree(tmp1);
1083 	kfree(tmp2);
1084 	return ret;
1085 }
1086 
md_csum_fold(u32 csum)1087 static u32 md_csum_fold(u32 csum)
1088 {
1089 	csum = (csum & 0xffff) + (csum >> 16);
1090 	return (csum & 0xffff) + (csum >> 16);
1091 }
1092 
calc_sb_csum(mdp_super_t * sb)1093 static unsigned int calc_sb_csum(mdp_super_t *sb)
1094 {
1095 	u64 newcsum = 0;
1096 	u32 *sb32 = (u32*)sb;
1097 	int i;
1098 	unsigned int disk_csum, csum;
1099 
1100 	disk_csum = sb->sb_csum;
1101 	sb->sb_csum = 0;
1102 
1103 	for (i = 0; i < MD_SB_BYTES/4 ; i++)
1104 		newcsum += sb32[i];
1105 	csum = (newcsum & 0xffffffff) + (newcsum>>32);
1106 
1107 #ifdef CONFIG_ALPHA
1108 	/* This used to use csum_partial, which was wrong for several
1109 	 * reasons including that different results are returned on
1110 	 * different architectures.  It isn't critical that we get exactly
1111 	 * the same return value as before (we always csum_fold before
1112 	 * testing, and that removes any differences).  However as we
1113 	 * know that csum_partial always returned a 16bit value on
1114 	 * alphas, do a fold to maximise conformity to previous behaviour.
1115 	 */
1116 	sb->sb_csum = md_csum_fold(disk_csum);
1117 #else
1118 	sb->sb_csum = disk_csum;
1119 #endif
1120 	return csum;
1121 }
1122 
1123 /*
1124  * Handle superblock details.
1125  * We want to be able to handle multiple superblock formats
1126  * so we have a common interface to them all, and an array of
1127  * different handlers.
1128  * We rely on user-space to write the initial superblock, and support
1129  * reading and updating of superblocks.
1130  * Interface methods are:
1131  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1132  *      loads and validates a superblock on dev.
1133  *      if refdev != NULL, compare superblocks on both devices
1134  *    Return:
1135  *      0 - dev has a superblock that is compatible with refdev
1136  *      1 - dev has a superblock that is compatible and newer than refdev
1137  *          so dev should be used as the refdev in future
1138  *     -EINVAL superblock incompatible or invalid
1139  *     -othererror e.g. -EIO
1140  *
1141  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1142  *      Verify that dev is acceptable into mddev.
1143  *       The first time, mddev->raid_disks will be 0, and data from
1144  *       dev should be merged in.  Subsequent calls check that dev
1145  *       is new enough.  Return 0 or -EINVAL
1146  *
1147  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1148  *     Update the superblock for rdev with data in mddev
1149  *     This does not write to disc.
1150  *
1151  */
1152 
1153 struct super_type  {
1154 	char		    *name;
1155 	struct module	    *owner;
1156 	int		    (*load_super)(struct md_rdev *rdev,
1157 					  struct md_rdev *refdev,
1158 					  int minor_version);
1159 	int		    (*validate_super)(struct mddev *mddev,
1160 					      struct md_rdev *freshest,
1161 					      struct md_rdev *rdev);
1162 	void		    (*sync_super)(struct mddev *mddev,
1163 					  struct md_rdev *rdev);
1164 	unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1165 						sector_t num_sectors);
1166 	int		    (*allow_new_offset)(struct md_rdev *rdev,
1167 						unsigned long long new_offset);
1168 };
1169 
1170 /*
1171  * Check that the given mddev has no bitmap.
1172  *
1173  * This function is called from the run method of all personalities that do not
1174  * support bitmaps. It prints an error message and returns non-zero if mddev
1175  * has a bitmap. Otherwise, it returns 0.
1176  *
1177  */
md_check_no_bitmap(struct mddev * mddev)1178 int md_check_no_bitmap(struct mddev *mddev)
1179 {
1180 	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1181 		return 0;
1182 	pr_warn("%s: bitmaps are not supported for %s\n",
1183 		mdname(mddev), mddev->pers->name);
1184 	return 1;
1185 }
1186 EXPORT_SYMBOL(md_check_no_bitmap);
1187 
1188 /*
1189  * load_super for 0.90.0
1190  */
super_90_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1191 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1192 {
1193 	mdp_super_t *sb;
1194 	int ret;
1195 	bool spare_disk = true;
1196 
1197 	/*
1198 	 * Calculate the position of the superblock (512byte sectors),
1199 	 * it's at the end of the disk.
1200 	 *
1201 	 * It also happens to be a multiple of 4Kb.
1202 	 */
1203 	rdev->sb_start = calc_dev_sboffset(rdev);
1204 
1205 	ret = read_disk_sb(rdev, MD_SB_BYTES);
1206 	if (ret)
1207 		return ret;
1208 
1209 	ret = -EINVAL;
1210 
1211 	sb = page_address(rdev->sb_page);
1212 
1213 	if (sb->md_magic != MD_SB_MAGIC) {
1214 		pr_warn("md: invalid raid superblock magic on %pg\n",
1215 			rdev->bdev);
1216 		goto abort;
1217 	}
1218 
1219 	if (sb->major_version != 0 ||
1220 	    sb->minor_version < 90 ||
1221 	    sb->minor_version > 91) {
1222 		pr_warn("Bad version number %d.%d on %pg\n",
1223 			sb->major_version, sb->minor_version, rdev->bdev);
1224 		goto abort;
1225 	}
1226 
1227 	if (sb->raid_disks <= 0)
1228 		goto abort;
1229 
1230 	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1231 		pr_warn("md: invalid superblock checksum on %pg\n", rdev->bdev);
1232 		goto abort;
1233 	}
1234 
1235 	rdev->preferred_minor = sb->md_minor;
1236 	rdev->data_offset = 0;
1237 	rdev->new_data_offset = 0;
1238 	rdev->sb_size = MD_SB_BYTES;
1239 	rdev->badblocks.shift = -1;
1240 
1241 	rdev->desc_nr = sb->this_disk.number;
1242 
1243 	/* not spare disk */
1244 	if (rdev->desc_nr >= 0 && rdev->desc_nr < MD_SB_DISKS &&
1245 	    sb->disks[rdev->desc_nr].state & ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1246 		spare_disk = false;
1247 
1248 	if (!refdev) {
1249 		if (!spare_disk)
1250 			ret = 1;
1251 		else
1252 			ret = 0;
1253 	} else {
1254 		__u64 ev1, ev2;
1255 		mdp_super_t *refsb = page_address(refdev->sb_page);
1256 		if (!md_uuid_equal(refsb, sb)) {
1257 			pr_warn("md: %pg has different UUID to %pg\n",
1258 				rdev->bdev, refdev->bdev);
1259 			goto abort;
1260 		}
1261 		if (!md_sb_equal(refsb, sb)) {
1262 			pr_warn("md: %pg has same UUID but different superblock to %pg\n",
1263 				rdev->bdev, refdev->bdev);
1264 			goto abort;
1265 		}
1266 		ev1 = md_event(sb);
1267 		ev2 = md_event(refsb);
1268 
1269 		if (!spare_disk && ev1 > ev2)
1270 			ret = 1;
1271 		else
1272 			ret = 0;
1273 	}
1274 	rdev->sectors = rdev->sb_start;
1275 	/* Limit to 4TB as metadata cannot record more than that.
1276 	 * (not needed for Linear and RAID0 as metadata doesn't
1277 	 * record this size)
1278 	 */
1279 	if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1280 		rdev->sectors = (sector_t)(2ULL << 32) - 2;
1281 
1282 	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1283 		/* "this cannot possibly happen" ... */
1284 		ret = -EINVAL;
1285 
1286  abort:
1287 	return ret;
1288 }
1289 
md_bitmap_events_cleared(struct mddev * mddev)1290 static u64 md_bitmap_events_cleared(struct mddev *mddev)
1291 {
1292 	struct md_bitmap_stats stats;
1293 	int err;
1294 
1295 	err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
1296 	if (err)
1297 		return 0;
1298 
1299 	return stats.events_cleared;
1300 }
1301 
1302 /*
1303  * validate_super for 0.90.0
1304  * note: we are not using "freshest" for 0.9 superblock
1305  */
super_90_validate(struct mddev * mddev,struct md_rdev * freshest,struct md_rdev * rdev)1306 static int super_90_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1307 {
1308 	mdp_disk_t *desc;
1309 	mdp_super_t *sb = page_address(rdev->sb_page);
1310 	__u64 ev1 = md_event(sb);
1311 
1312 	rdev->raid_disk = -1;
1313 	clear_bit(Faulty, &rdev->flags);
1314 	clear_bit(In_sync, &rdev->flags);
1315 	clear_bit(Bitmap_sync, &rdev->flags);
1316 	clear_bit(WriteMostly, &rdev->flags);
1317 
1318 	if (mddev->raid_disks == 0) {
1319 		mddev->major_version = 0;
1320 		mddev->minor_version = sb->minor_version;
1321 		mddev->patch_version = sb->patch_version;
1322 		mddev->external = 0;
1323 		mddev->chunk_sectors = sb->chunk_size >> 9;
1324 		mddev->ctime = sb->ctime;
1325 		mddev->utime = sb->utime;
1326 		mddev->level = sb->level;
1327 		mddev->clevel[0] = 0;
1328 		mddev->layout = sb->layout;
1329 		mddev->raid_disks = sb->raid_disks;
1330 		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1331 		mddev->events = ev1;
1332 		mddev->bitmap_info.offset = 0;
1333 		mddev->bitmap_info.space = 0;
1334 		/* bitmap can use 60 K after the 4K superblocks */
1335 		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1336 		mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1337 		mddev->reshape_backwards = 0;
1338 
1339 		if (mddev->minor_version >= 91) {
1340 			mddev->reshape_position = sb->reshape_position;
1341 			mddev->delta_disks = sb->delta_disks;
1342 			mddev->new_level = sb->new_level;
1343 			mddev->new_layout = sb->new_layout;
1344 			mddev->new_chunk_sectors = sb->new_chunk >> 9;
1345 			if (mddev->delta_disks < 0)
1346 				mddev->reshape_backwards = 1;
1347 		} else {
1348 			mddev->reshape_position = MaxSector;
1349 			mddev->delta_disks = 0;
1350 			mddev->new_level = mddev->level;
1351 			mddev->new_layout = mddev->layout;
1352 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1353 		}
1354 		if (mddev->level == 0)
1355 			mddev->layout = -1;
1356 
1357 		if (sb->state & (1<<MD_SB_CLEAN))
1358 			mddev->recovery_cp = MaxSector;
1359 		else {
1360 			if (sb->events_hi == sb->cp_events_hi &&
1361 				sb->events_lo == sb->cp_events_lo) {
1362 				mddev->recovery_cp = sb->recovery_cp;
1363 			} else
1364 				mddev->recovery_cp = 0;
1365 		}
1366 
1367 		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1368 		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1369 		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1370 		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1371 
1372 		mddev->max_disks = MD_SB_DISKS;
1373 
1374 		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1375 		    mddev->bitmap_info.file == NULL) {
1376 			mddev->bitmap_info.offset =
1377 				mddev->bitmap_info.default_offset;
1378 			mddev->bitmap_info.space =
1379 				mddev->bitmap_info.default_space;
1380 		}
1381 
1382 	} else if (mddev->pers == NULL) {
1383 		/* Insist on good event counter while assembling, except
1384 		 * for spares (which don't need an event count) */
1385 		++ev1;
1386 		if (sb->disks[rdev->desc_nr].state & (
1387 			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1388 			if (ev1 < mddev->events)
1389 				return -EINVAL;
1390 	} else if (mddev->bitmap) {
1391 		/* if adding to array with a bitmap, then we can accept an
1392 		 * older device ... but not too old.
1393 		 */
1394 		if (ev1 < md_bitmap_events_cleared(mddev))
1395 			return 0;
1396 		if (ev1 < mddev->events)
1397 			set_bit(Bitmap_sync, &rdev->flags);
1398 	} else {
1399 		if (ev1 < mddev->events)
1400 			/* just a hot-add of a new device, leave raid_disk at -1 */
1401 			return 0;
1402 	}
1403 
1404 	desc = sb->disks + rdev->desc_nr;
1405 
1406 	if (desc->state & (1<<MD_DISK_FAULTY))
1407 		set_bit(Faulty, &rdev->flags);
1408 	else if (desc->state & (1<<MD_DISK_SYNC)) {
1409 		set_bit(In_sync, &rdev->flags);
1410 		rdev->raid_disk = desc->raid_disk;
1411 		rdev->saved_raid_disk = desc->raid_disk;
1412 	} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1413 		/* active but not in sync implies recovery up to
1414 		 * reshape position.  We don't know exactly where
1415 		 * that is, so set to zero for now
1416 		 */
1417 		if (mddev->minor_version >= 91) {
1418 			rdev->recovery_offset = 0;
1419 			rdev->raid_disk = desc->raid_disk;
1420 		}
1421 	}
1422 	if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1423 		set_bit(WriteMostly, &rdev->flags);
1424 	if (desc->state & (1<<MD_DISK_FAILFAST))
1425 		set_bit(FailFast, &rdev->flags);
1426 	return 0;
1427 }
1428 
1429 /*
1430  * sync_super for 0.90.0
1431  */
super_90_sync(struct mddev * mddev,struct md_rdev * rdev)1432 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1433 {
1434 	mdp_super_t *sb;
1435 	struct md_rdev *rdev2;
1436 	int next_spare = mddev->raid_disks;
1437 
1438 	/* make rdev->sb match mddev data..
1439 	 *
1440 	 * 1/ zero out disks
1441 	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1442 	 * 3/ any empty disks < next_spare become removed
1443 	 *
1444 	 * disks[0] gets initialised to REMOVED because
1445 	 * we cannot be sure from other fields if it has
1446 	 * been initialised or not.
1447 	 */
1448 	int i;
1449 	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1450 
1451 	rdev->sb_size = MD_SB_BYTES;
1452 
1453 	sb = page_address(rdev->sb_page);
1454 
1455 	memset(sb, 0, sizeof(*sb));
1456 
1457 	sb->md_magic = MD_SB_MAGIC;
1458 	sb->major_version = mddev->major_version;
1459 	sb->patch_version = mddev->patch_version;
1460 	sb->gvalid_words  = 0; /* ignored */
1461 	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1462 	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1463 	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1464 	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1465 
1466 	sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1467 	sb->level = mddev->level;
1468 	sb->size = mddev->dev_sectors / 2;
1469 	sb->raid_disks = mddev->raid_disks;
1470 	sb->md_minor = mddev->md_minor;
1471 	sb->not_persistent = 0;
1472 	sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1473 	sb->state = 0;
1474 	sb->events_hi = (mddev->events>>32);
1475 	sb->events_lo = (u32)mddev->events;
1476 
1477 	if (mddev->reshape_position == MaxSector)
1478 		sb->minor_version = 90;
1479 	else {
1480 		sb->minor_version = 91;
1481 		sb->reshape_position = mddev->reshape_position;
1482 		sb->new_level = mddev->new_level;
1483 		sb->delta_disks = mddev->delta_disks;
1484 		sb->new_layout = mddev->new_layout;
1485 		sb->new_chunk = mddev->new_chunk_sectors << 9;
1486 	}
1487 	mddev->minor_version = sb->minor_version;
1488 	if (mddev->in_sync)
1489 	{
1490 		sb->recovery_cp = mddev->recovery_cp;
1491 		sb->cp_events_hi = (mddev->events>>32);
1492 		sb->cp_events_lo = (u32)mddev->events;
1493 		if (mddev->recovery_cp == MaxSector)
1494 			sb->state = (1<< MD_SB_CLEAN);
1495 	} else
1496 		sb->recovery_cp = 0;
1497 
1498 	sb->layout = mddev->layout;
1499 	sb->chunk_size = mddev->chunk_sectors << 9;
1500 
1501 	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1502 		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1503 
1504 	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1505 	rdev_for_each(rdev2, mddev) {
1506 		mdp_disk_t *d;
1507 		int desc_nr;
1508 		int is_active = test_bit(In_sync, &rdev2->flags);
1509 
1510 		if (rdev2->raid_disk >= 0 &&
1511 		    sb->minor_version >= 91)
1512 			/* we have nowhere to store the recovery_offset,
1513 			 * but if it is not below the reshape_position,
1514 			 * we can piggy-back on that.
1515 			 */
1516 			is_active = 1;
1517 		if (rdev2->raid_disk < 0 ||
1518 		    test_bit(Faulty, &rdev2->flags))
1519 			is_active = 0;
1520 		if (is_active)
1521 			desc_nr = rdev2->raid_disk;
1522 		else
1523 			desc_nr = next_spare++;
1524 		rdev2->desc_nr = desc_nr;
1525 		d = &sb->disks[rdev2->desc_nr];
1526 		nr_disks++;
1527 		d->number = rdev2->desc_nr;
1528 		d->major = MAJOR(rdev2->bdev->bd_dev);
1529 		d->minor = MINOR(rdev2->bdev->bd_dev);
1530 		if (is_active)
1531 			d->raid_disk = rdev2->raid_disk;
1532 		else
1533 			d->raid_disk = rdev2->desc_nr; /* compatibility */
1534 		if (test_bit(Faulty, &rdev2->flags))
1535 			d->state = (1<<MD_DISK_FAULTY);
1536 		else if (is_active) {
1537 			d->state = (1<<MD_DISK_ACTIVE);
1538 			if (test_bit(In_sync, &rdev2->flags))
1539 				d->state |= (1<<MD_DISK_SYNC);
1540 			active++;
1541 			working++;
1542 		} else {
1543 			d->state = 0;
1544 			spare++;
1545 			working++;
1546 		}
1547 		if (test_bit(WriteMostly, &rdev2->flags))
1548 			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1549 		if (test_bit(FailFast, &rdev2->flags))
1550 			d->state |= (1<<MD_DISK_FAILFAST);
1551 	}
1552 	/* now set the "removed" and "faulty" bits on any missing devices */
1553 	for (i=0 ; i < mddev->raid_disks ; i++) {
1554 		mdp_disk_t *d = &sb->disks[i];
1555 		if (d->state == 0 && d->number == 0) {
1556 			d->number = i;
1557 			d->raid_disk = i;
1558 			d->state = (1<<MD_DISK_REMOVED);
1559 			d->state |= (1<<MD_DISK_FAULTY);
1560 			failed++;
1561 		}
1562 	}
1563 	sb->nr_disks = nr_disks;
1564 	sb->active_disks = active;
1565 	sb->working_disks = working;
1566 	sb->failed_disks = failed;
1567 	sb->spare_disks = spare;
1568 
1569 	sb->this_disk = sb->disks[rdev->desc_nr];
1570 	sb->sb_csum = calc_sb_csum(sb);
1571 }
1572 
1573 /*
1574  * rdev_size_change for 0.90.0
1575  */
1576 static unsigned long long
super_90_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)1577 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1578 {
1579 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1580 		return 0; /* component must fit device */
1581 	if (rdev->mddev->bitmap_info.offset)
1582 		return 0; /* can't move bitmap */
1583 	rdev->sb_start = calc_dev_sboffset(rdev);
1584 	if (!num_sectors || num_sectors > rdev->sb_start)
1585 		num_sectors = rdev->sb_start;
1586 	/* Limit to 4TB as metadata cannot record more than that.
1587 	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1588 	 */
1589 	if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1590 		num_sectors = (sector_t)(2ULL << 32) - 2;
1591 	do {
1592 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1593 		       rdev->sb_page);
1594 	} while (md_super_wait(rdev->mddev) < 0);
1595 	return num_sectors;
1596 }
1597 
1598 static int
super_90_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)1599 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1600 {
1601 	/* non-zero offset changes not possible with v0.90 */
1602 	return new_offset == 0;
1603 }
1604 
1605 /*
1606  * version 1 superblock
1607  */
1608 
calc_sb_1_csum(struct mdp_superblock_1 * sb)1609 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1610 {
1611 	__le32 disk_csum;
1612 	u32 csum;
1613 	unsigned long long newcsum;
1614 	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1615 	__le32 *isuper = (__le32*)sb;
1616 
1617 	disk_csum = sb->sb_csum;
1618 	sb->sb_csum = 0;
1619 	newcsum = 0;
1620 	for (; size >= 4; size -= 4)
1621 		newcsum += le32_to_cpu(*isuper++);
1622 
1623 	if (size == 2)
1624 		newcsum += le16_to_cpu(*(__le16*) isuper);
1625 
1626 	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1627 	sb->sb_csum = disk_csum;
1628 	return cpu_to_le32(csum);
1629 }
1630 
super_1_load(struct md_rdev * rdev,struct md_rdev * refdev,int minor_version)1631 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1632 {
1633 	struct mdp_superblock_1 *sb;
1634 	int ret;
1635 	sector_t sb_start;
1636 	sector_t sectors;
1637 	int bmask;
1638 	bool spare_disk = true;
1639 
1640 	/*
1641 	 * Calculate the position of the superblock in 512byte sectors.
1642 	 * It is always aligned to a 4K boundary and
1643 	 * depeding on minor_version, it can be:
1644 	 * 0: At least 8K, but less than 12K, from end of device
1645 	 * 1: At start of device
1646 	 * 2: 4K from start of device.
1647 	 */
1648 	switch(minor_version) {
1649 	case 0:
1650 		sb_start = bdev_nr_sectors(rdev->bdev) - 8 * 2;
1651 		sb_start &= ~(sector_t)(4*2-1);
1652 		break;
1653 	case 1:
1654 		sb_start = 0;
1655 		break;
1656 	case 2:
1657 		sb_start = 8;
1658 		break;
1659 	default:
1660 		return -EINVAL;
1661 	}
1662 	rdev->sb_start = sb_start;
1663 
1664 	/* superblock is rarely larger than 1K, but it can be larger,
1665 	 * and it is safe to read 4k, so we do that
1666 	 */
1667 	ret = read_disk_sb(rdev, 4096);
1668 	if (ret) return ret;
1669 
1670 	sb = page_address(rdev->sb_page);
1671 
1672 	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1673 	    sb->major_version != cpu_to_le32(1) ||
1674 	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1675 	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1676 	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1677 		return -EINVAL;
1678 
1679 	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1680 		pr_warn("md: invalid superblock checksum on %pg\n",
1681 			rdev->bdev);
1682 		return -EINVAL;
1683 	}
1684 	if (le64_to_cpu(sb->data_size) < 10) {
1685 		pr_warn("md: data_size too small on %pg\n",
1686 			rdev->bdev);
1687 		return -EINVAL;
1688 	}
1689 	if (sb->pad0 ||
1690 	    sb->pad3[0] ||
1691 	    memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1692 		/* Some padding is non-zero, might be a new feature */
1693 		return -EINVAL;
1694 
1695 	rdev->preferred_minor = 0xffff;
1696 	rdev->data_offset = le64_to_cpu(sb->data_offset);
1697 	rdev->new_data_offset = rdev->data_offset;
1698 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1699 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1700 		rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1701 	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1702 
1703 	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1704 	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1705 	if (rdev->sb_size & bmask)
1706 		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1707 
1708 	if (minor_version
1709 	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1710 		return -EINVAL;
1711 	if (minor_version
1712 	    && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1713 		return -EINVAL;
1714 
1715 	rdev->desc_nr = le32_to_cpu(sb->dev_number);
1716 
1717 	if (!rdev->bb_page) {
1718 		rdev->bb_page = alloc_page(GFP_KERNEL);
1719 		if (!rdev->bb_page)
1720 			return -ENOMEM;
1721 	}
1722 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1723 	    rdev->badblocks.count == 0) {
1724 		/* need to load the bad block list.
1725 		 * Currently we limit it to one page.
1726 		 */
1727 		s32 offset;
1728 		sector_t bb_sector;
1729 		__le64 *bbp;
1730 		int i;
1731 		int sectors = le16_to_cpu(sb->bblog_size);
1732 		if (sectors > (PAGE_SIZE / 512))
1733 			return -EINVAL;
1734 		offset = le32_to_cpu(sb->bblog_offset);
1735 		if (offset == 0)
1736 			return -EINVAL;
1737 		bb_sector = (long long)offset;
1738 		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1739 				  rdev->bb_page, REQ_OP_READ, true))
1740 			return -EIO;
1741 		bbp = (__le64 *)page_address(rdev->bb_page);
1742 		rdev->badblocks.shift = sb->bblog_shift;
1743 		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1744 			u64 bb = le64_to_cpu(*bbp);
1745 			int count = bb & (0x3ff);
1746 			u64 sector = bb >> 10;
1747 			sector <<= sb->bblog_shift;
1748 			count <<= sb->bblog_shift;
1749 			if (bb + 1 == 0)
1750 				break;
1751 			if (badblocks_set(&rdev->badblocks, sector, count, 1))
1752 				return -EINVAL;
1753 		}
1754 	} else if (sb->bblog_offset != 0)
1755 		rdev->badblocks.shift = 0;
1756 
1757 	if ((le32_to_cpu(sb->feature_map) &
1758 	    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1759 		rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1760 		rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1761 		rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1762 	}
1763 
1764 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1765 	    sb->level != 0)
1766 		return -EINVAL;
1767 
1768 	/* not spare disk */
1769 	if (rdev->desc_nr >= 0 && rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1770 	    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1771 	     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1772 		spare_disk = false;
1773 
1774 	if (!refdev) {
1775 		if (!spare_disk)
1776 			ret = 1;
1777 		else
1778 			ret = 0;
1779 	} else {
1780 		__u64 ev1, ev2;
1781 		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1782 
1783 		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1784 		    sb->level != refsb->level ||
1785 		    sb->layout != refsb->layout ||
1786 		    sb->chunksize != refsb->chunksize) {
1787 			pr_warn("md: %pg has strangely different superblock to %pg\n",
1788 				rdev->bdev,
1789 				refdev->bdev);
1790 			return -EINVAL;
1791 		}
1792 		ev1 = le64_to_cpu(sb->events);
1793 		ev2 = le64_to_cpu(refsb->events);
1794 
1795 		if (!spare_disk && ev1 > ev2)
1796 			ret = 1;
1797 		else
1798 			ret = 0;
1799 	}
1800 	if (minor_version)
1801 		sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
1802 	else
1803 		sectors = rdev->sb_start;
1804 	if (sectors < le64_to_cpu(sb->data_size))
1805 		return -EINVAL;
1806 	rdev->sectors = le64_to_cpu(sb->data_size);
1807 	return ret;
1808 }
1809 
super_1_validate(struct mddev * mddev,struct md_rdev * freshest,struct md_rdev * rdev)1810 static int super_1_validate(struct mddev *mddev, struct md_rdev *freshest, struct md_rdev *rdev)
1811 {
1812 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1813 	__u64 ev1 = le64_to_cpu(sb->events);
1814 	int role;
1815 
1816 	rdev->raid_disk = -1;
1817 	clear_bit(Faulty, &rdev->flags);
1818 	clear_bit(In_sync, &rdev->flags);
1819 	clear_bit(Bitmap_sync, &rdev->flags);
1820 	clear_bit(WriteMostly, &rdev->flags);
1821 
1822 	if (mddev->raid_disks == 0) {
1823 		mddev->major_version = 1;
1824 		mddev->patch_version = 0;
1825 		mddev->external = 0;
1826 		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1827 		mddev->ctime = le64_to_cpu(sb->ctime);
1828 		mddev->utime = le64_to_cpu(sb->utime);
1829 		mddev->level = le32_to_cpu(sb->level);
1830 		mddev->clevel[0] = 0;
1831 		mddev->layout = le32_to_cpu(sb->layout);
1832 		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1833 		mddev->dev_sectors = le64_to_cpu(sb->size);
1834 		mddev->events = ev1;
1835 		mddev->bitmap_info.offset = 0;
1836 		mddev->bitmap_info.space = 0;
1837 		/* Default location for bitmap is 1K after superblock
1838 		 * using 3K - total of 4K
1839 		 */
1840 		mddev->bitmap_info.default_offset = 1024 >> 9;
1841 		mddev->bitmap_info.default_space = (4096-1024) >> 9;
1842 		mddev->reshape_backwards = 0;
1843 
1844 		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1845 		memcpy(mddev->uuid, sb->set_uuid, 16);
1846 
1847 		mddev->max_disks =  (4096-256)/2;
1848 
1849 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1850 		    mddev->bitmap_info.file == NULL) {
1851 			mddev->bitmap_info.offset =
1852 				(__s32)le32_to_cpu(sb->bitmap_offset);
1853 			/* Metadata doesn't record how much space is available.
1854 			 * For 1.0, we assume we can use up to the superblock
1855 			 * if before, else to 4K beyond superblock.
1856 			 * For others, assume no change is possible.
1857 			 */
1858 			if (mddev->minor_version > 0)
1859 				mddev->bitmap_info.space = 0;
1860 			else if (mddev->bitmap_info.offset > 0)
1861 				mddev->bitmap_info.space =
1862 					8 - mddev->bitmap_info.offset;
1863 			else
1864 				mddev->bitmap_info.space =
1865 					-mddev->bitmap_info.offset;
1866 		}
1867 
1868 		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1869 			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1870 			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1871 			mddev->new_level = le32_to_cpu(sb->new_level);
1872 			mddev->new_layout = le32_to_cpu(sb->new_layout);
1873 			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1874 			if (mddev->delta_disks < 0 ||
1875 			    (mddev->delta_disks == 0 &&
1876 			     (le32_to_cpu(sb->feature_map)
1877 			      & MD_FEATURE_RESHAPE_BACKWARDS)))
1878 				mddev->reshape_backwards = 1;
1879 		} else {
1880 			mddev->reshape_position = MaxSector;
1881 			mddev->delta_disks = 0;
1882 			mddev->new_level = mddev->level;
1883 			mddev->new_layout = mddev->layout;
1884 			mddev->new_chunk_sectors = mddev->chunk_sectors;
1885 		}
1886 
1887 		if (mddev->level == 0 &&
1888 		    !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1889 			mddev->layout = -1;
1890 
1891 		if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1892 			set_bit(MD_HAS_JOURNAL, &mddev->flags);
1893 
1894 		if (le32_to_cpu(sb->feature_map) &
1895 		    (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1896 			if (le32_to_cpu(sb->feature_map) &
1897 			    (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1898 				return -EINVAL;
1899 			if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1900 			    (le32_to_cpu(sb->feature_map) &
1901 					    MD_FEATURE_MULTIPLE_PPLS))
1902 				return -EINVAL;
1903 			set_bit(MD_HAS_PPL, &mddev->flags);
1904 		}
1905 	} else if (mddev->pers == NULL) {
1906 		/* Insist of good event counter while assembling, except for
1907 		 * spares (which don't need an event count).
1908 		 * Similar to mdadm, we allow event counter difference of 1
1909 		 * from the freshest device.
1910 		 */
1911 		if (rdev->desc_nr >= 0 &&
1912 		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1913 		    (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1914 		     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1915 			if (ev1 + 1 < mddev->events)
1916 				return -EINVAL;
1917 	} else if (mddev->bitmap) {
1918 		/* If adding to array with a bitmap, then we can accept an
1919 		 * older device, but not too old.
1920 		 */
1921 		if (ev1 < md_bitmap_events_cleared(mddev))
1922 			return 0;
1923 		if (ev1 < mddev->events)
1924 			set_bit(Bitmap_sync, &rdev->flags);
1925 	} else {
1926 		if (ev1 < mddev->events)
1927 			/* just a hot-add of a new device, leave raid_disk at -1 */
1928 			return 0;
1929 	}
1930 
1931 	if (rdev->desc_nr < 0 ||
1932 	    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1933 		role = MD_DISK_ROLE_SPARE;
1934 		rdev->desc_nr = -1;
1935 	} else if (mddev->pers == NULL && freshest && ev1 < mddev->events) {
1936 		/*
1937 		 * If we are assembling, and our event counter is smaller than the
1938 		 * highest event counter, we cannot trust our superblock about the role.
1939 		 * It could happen that our rdev was marked as Faulty, and all other
1940 		 * superblocks were updated with +1 event counter.
1941 		 * Then, before the next superblock update, which typically happens when
1942 		 * remove_and_add_spares() removes the device from the array, there was
1943 		 * a crash or reboot.
1944 		 * If we allow current rdev without consulting the freshest superblock,
1945 		 * we could cause data corruption.
1946 		 * Note that in this case our event counter is smaller by 1 than the
1947 		 * highest, otherwise, this rdev would not be allowed into array;
1948 		 * both kernel and mdadm allow event counter difference of 1.
1949 		 */
1950 		struct mdp_superblock_1 *freshest_sb = page_address(freshest->sb_page);
1951 		u32 freshest_max_dev = le32_to_cpu(freshest_sb->max_dev);
1952 
1953 		if (rdev->desc_nr >= freshest_max_dev) {
1954 			/* this is unexpected, better not proceed */
1955 			pr_warn("md: %s: rdev[%pg]: desc_nr(%d) >= freshest(%pg)->sb->max_dev(%u)\n",
1956 				mdname(mddev), rdev->bdev, rdev->desc_nr,
1957 				freshest->bdev, freshest_max_dev);
1958 			return -EUCLEAN;
1959 		}
1960 
1961 		role = le16_to_cpu(freshest_sb->dev_roles[rdev->desc_nr]);
1962 		pr_debug("md: %s: rdev[%pg]: role=%d(0x%x) according to freshest %pg\n",
1963 			 mdname(mddev), rdev->bdev, role, role, freshest->bdev);
1964 	} else {
1965 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1966 	}
1967 	switch (role) {
1968 	case MD_DISK_ROLE_SPARE: /* spare */
1969 		break;
1970 	case MD_DISK_ROLE_FAULTY: /* faulty */
1971 		set_bit(Faulty, &rdev->flags);
1972 		break;
1973 	case MD_DISK_ROLE_JOURNAL: /* journal device */
1974 		if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1975 			/* journal device without journal feature */
1976 			pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1977 			return -EINVAL;
1978 		}
1979 		set_bit(Journal, &rdev->flags);
1980 		rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1981 		rdev->raid_disk = 0;
1982 		break;
1983 	default:
1984 		rdev->saved_raid_disk = role;
1985 		if ((le32_to_cpu(sb->feature_map) &
1986 		     MD_FEATURE_RECOVERY_OFFSET)) {
1987 			rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1988 			if (!(le32_to_cpu(sb->feature_map) &
1989 			      MD_FEATURE_RECOVERY_BITMAP))
1990 				rdev->saved_raid_disk = -1;
1991 		} else {
1992 			/*
1993 			 * If the array is FROZEN, then the device can't
1994 			 * be in_sync with rest of array.
1995 			 */
1996 			if (!test_bit(MD_RECOVERY_FROZEN,
1997 				      &mddev->recovery))
1998 				set_bit(In_sync, &rdev->flags);
1999 		}
2000 		rdev->raid_disk = role;
2001 		break;
2002 	}
2003 	if (sb->devflags & WriteMostly1)
2004 		set_bit(WriteMostly, &rdev->flags);
2005 	if (sb->devflags & FailFast1)
2006 		set_bit(FailFast, &rdev->flags);
2007 	if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
2008 		set_bit(Replacement, &rdev->flags);
2009 
2010 	return 0;
2011 }
2012 
super_1_sync(struct mddev * mddev,struct md_rdev * rdev)2013 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
2014 {
2015 	struct mdp_superblock_1 *sb;
2016 	struct md_rdev *rdev2;
2017 	int max_dev, i;
2018 	/* make rdev->sb match mddev and rdev data. */
2019 
2020 	sb = page_address(rdev->sb_page);
2021 
2022 	sb->feature_map = 0;
2023 	sb->pad0 = 0;
2024 	sb->recovery_offset = cpu_to_le64(0);
2025 	memset(sb->pad3, 0, sizeof(sb->pad3));
2026 
2027 	sb->utime = cpu_to_le64((__u64)mddev->utime);
2028 	sb->events = cpu_to_le64(mddev->events);
2029 	if (mddev->in_sync)
2030 		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
2031 	else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
2032 		sb->resync_offset = cpu_to_le64(MaxSector);
2033 	else
2034 		sb->resync_offset = cpu_to_le64(0);
2035 
2036 	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
2037 
2038 	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
2039 	sb->size = cpu_to_le64(mddev->dev_sectors);
2040 	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
2041 	sb->level = cpu_to_le32(mddev->level);
2042 	sb->layout = cpu_to_le32(mddev->layout);
2043 	if (test_bit(FailFast, &rdev->flags))
2044 		sb->devflags |= FailFast1;
2045 	else
2046 		sb->devflags &= ~FailFast1;
2047 
2048 	if (test_bit(WriteMostly, &rdev->flags))
2049 		sb->devflags |= WriteMostly1;
2050 	else
2051 		sb->devflags &= ~WriteMostly1;
2052 	sb->data_offset = cpu_to_le64(rdev->data_offset);
2053 	sb->data_size = cpu_to_le64(rdev->sectors);
2054 
2055 	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
2056 		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
2057 		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
2058 	}
2059 
2060 	if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
2061 	    !test_bit(In_sync, &rdev->flags)) {
2062 		sb->feature_map |=
2063 			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
2064 		sb->recovery_offset =
2065 			cpu_to_le64(rdev->recovery_offset);
2066 		if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
2067 			sb->feature_map |=
2068 				cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
2069 	}
2070 	/* Note: recovery_offset and journal_tail share space  */
2071 	if (test_bit(Journal, &rdev->flags))
2072 		sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2073 	if (test_bit(Replacement, &rdev->flags))
2074 		sb->feature_map |=
2075 			cpu_to_le32(MD_FEATURE_REPLACEMENT);
2076 
2077 	if (mddev->reshape_position != MaxSector) {
2078 		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
2079 		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
2080 		sb->new_layout = cpu_to_le32(mddev->new_layout);
2081 		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
2082 		sb->new_level = cpu_to_le32(mddev->new_level);
2083 		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2084 		if (mddev->delta_disks == 0 &&
2085 		    mddev->reshape_backwards)
2086 			sb->feature_map
2087 				|= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
2088 		if (rdev->new_data_offset != rdev->data_offset) {
2089 			sb->feature_map
2090 				|= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
2091 			sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
2092 							     - rdev->data_offset));
2093 		}
2094 	}
2095 
2096 	if (mddev_is_clustered(mddev))
2097 		sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
2098 
2099 	if (rdev->badblocks.count == 0)
2100 		/* Nothing to do for bad blocks*/ ;
2101 	else if (sb->bblog_offset == 0)
2102 		/* Cannot record bad blocks on this device */
2103 		md_error(mddev, rdev);
2104 	else {
2105 		struct badblocks *bb = &rdev->badblocks;
2106 		__le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2107 		u64 *p = bb->page;
2108 		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2109 		if (bb->changed) {
2110 			unsigned seq;
2111 
2112 retry:
2113 			seq = read_seqbegin(&bb->lock);
2114 
2115 			memset(bbp, 0xff, PAGE_SIZE);
2116 
2117 			for (i = 0 ; i < bb->count ; i++) {
2118 				u64 internal_bb = p[i];
2119 				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2120 						| BB_LEN(internal_bb));
2121 				bbp[i] = cpu_to_le64(store_bb);
2122 			}
2123 			bb->changed = 0;
2124 			if (read_seqretry(&bb->lock, seq))
2125 				goto retry;
2126 
2127 			bb->sector = (rdev->sb_start +
2128 				      (int)le32_to_cpu(sb->bblog_offset));
2129 			bb->size = le16_to_cpu(sb->bblog_size);
2130 		}
2131 	}
2132 
2133 	max_dev = 0;
2134 	rdev_for_each(rdev2, mddev)
2135 		if (rdev2->desc_nr+1 > max_dev)
2136 			max_dev = rdev2->desc_nr+1;
2137 
2138 	if (max_dev > le32_to_cpu(sb->max_dev)) {
2139 		int bmask;
2140 		sb->max_dev = cpu_to_le32(max_dev);
2141 		rdev->sb_size = max_dev * 2 + 256;
2142 		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2143 		if (rdev->sb_size & bmask)
2144 			rdev->sb_size = (rdev->sb_size | bmask) + 1;
2145 	} else
2146 		max_dev = le32_to_cpu(sb->max_dev);
2147 
2148 	for (i=0; i<max_dev;i++)
2149 		sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2150 
2151 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2152 		sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2153 
2154 	if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2155 		if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2156 			sb->feature_map |=
2157 			    cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2158 		else
2159 			sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2160 		sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2161 		sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2162 	}
2163 
2164 	rdev_for_each(rdev2, mddev) {
2165 		i = rdev2->desc_nr;
2166 		if (test_bit(Faulty, &rdev2->flags))
2167 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2168 		else if (test_bit(In_sync, &rdev2->flags))
2169 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2170 		else if (test_bit(Journal, &rdev2->flags))
2171 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2172 		else if (rdev2->raid_disk >= 0)
2173 			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2174 		else
2175 			sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2176 	}
2177 
2178 	sb->sb_csum = calc_sb_1_csum(sb);
2179 }
2180 
super_1_choose_bm_space(sector_t dev_size)2181 static sector_t super_1_choose_bm_space(sector_t dev_size)
2182 {
2183 	sector_t bm_space;
2184 
2185 	/* if the device is bigger than 8Gig, save 64k for bitmap
2186 	 * usage, if bigger than 200Gig, save 128k
2187 	 */
2188 	if (dev_size < 64*2)
2189 		bm_space = 0;
2190 	else if (dev_size - 64*2 >= 200*1024*1024*2)
2191 		bm_space = 128*2;
2192 	else if (dev_size - 4*2 > 8*1024*1024*2)
2193 		bm_space = 64*2;
2194 	else
2195 		bm_space = 4*2;
2196 	return bm_space;
2197 }
2198 
2199 static unsigned long long
super_1_rdev_size_change(struct md_rdev * rdev,sector_t num_sectors)2200 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2201 {
2202 	struct mdp_superblock_1 *sb;
2203 	sector_t max_sectors;
2204 	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2205 		return 0; /* component must fit device */
2206 	if (rdev->data_offset != rdev->new_data_offset)
2207 		return 0; /* too confusing */
2208 	if (rdev->sb_start < rdev->data_offset) {
2209 		/* minor versions 1 and 2; superblock before data */
2210 		max_sectors = bdev_nr_sectors(rdev->bdev) - rdev->data_offset;
2211 		if (!num_sectors || num_sectors > max_sectors)
2212 			num_sectors = max_sectors;
2213 	} else if (rdev->mddev->bitmap_info.offset) {
2214 		/* minor version 0 with bitmap we can't move */
2215 		return 0;
2216 	} else {
2217 		/* minor version 0; superblock after data */
2218 		sector_t sb_start, bm_space;
2219 		sector_t dev_size = bdev_nr_sectors(rdev->bdev);
2220 
2221 		/* 8K is for superblock */
2222 		sb_start = dev_size - 8*2;
2223 		sb_start &= ~(sector_t)(4*2 - 1);
2224 
2225 		bm_space = super_1_choose_bm_space(dev_size);
2226 
2227 		/* Space that can be used to store date needs to decrease
2228 		 * superblock bitmap space and bad block space(4K)
2229 		 */
2230 		max_sectors = sb_start - bm_space - 4*2;
2231 
2232 		if (!num_sectors || num_sectors > max_sectors)
2233 			num_sectors = max_sectors;
2234 		rdev->sb_start = sb_start;
2235 	}
2236 	sb = page_address(rdev->sb_page);
2237 	sb->data_size = cpu_to_le64(num_sectors);
2238 	sb->super_offset = cpu_to_le64(rdev->sb_start);
2239 	sb->sb_csum = calc_sb_1_csum(sb);
2240 	do {
2241 		md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2242 			       rdev->sb_page);
2243 	} while (md_super_wait(rdev->mddev) < 0);
2244 	return num_sectors;
2245 
2246 }
2247 
2248 static int
super_1_allow_new_offset(struct md_rdev * rdev,unsigned long long new_offset)2249 super_1_allow_new_offset(struct md_rdev *rdev,
2250 			 unsigned long long new_offset)
2251 {
2252 	/* All necessary checks on new >= old have been done */
2253 	if (new_offset >= rdev->data_offset)
2254 		return 1;
2255 
2256 	/* with 1.0 metadata, there is no metadata to tread on
2257 	 * so we can always move back */
2258 	if (rdev->mddev->minor_version == 0)
2259 		return 1;
2260 
2261 	/* otherwise we must be sure not to step on
2262 	 * any metadata, so stay:
2263 	 * 36K beyond start of superblock
2264 	 * beyond end of badblocks
2265 	 * beyond write-intent bitmap
2266 	 */
2267 	if (rdev->sb_start + (32+4)*2 > new_offset)
2268 		return 0;
2269 
2270 	if (!rdev->mddev->bitmap_info.file) {
2271 		struct mddev *mddev = rdev->mddev;
2272 		struct md_bitmap_stats stats;
2273 		int err;
2274 
2275 		err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
2276 		if (!err && rdev->sb_start + mddev->bitmap_info.offset +
2277 		    stats.file_pages * (PAGE_SIZE >> 9) > new_offset)
2278 			return 0;
2279 	}
2280 
2281 	if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2282 		return 0;
2283 
2284 	return 1;
2285 }
2286 
2287 static struct super_type super_types[] = {
2288 	[0] = {
2289 		.name	= "0.90.0",
2290 		.owner	= THIS_MODULE,
2291 		.load_super	    = super_90_load,
2292 		.validate_super	    = super_90_validate,
2293 		.sync_super	    = super_90_sync,
2294 		.rdev_size_change   = super_90_rdev_size_change,
2295 		.allow_new_offset   = super_90_allow_new_offset,
2296 	},
2297 	[1] = {
2298 		.name	= "md-1",
2299 		.owner	= THIS_MODULE,
2300 		.load_super	    = super_1_load,
2301 		.validate_super	    = super_1_validate,
2302 		.sync_super	    = super_1_sync,
2303 		.rdev_size_change   = super_1_rdev_size_change,
2304 		.allow_new_offset   = super_1_allow_new_offset,
2305 	},
2306 };
2307 
sync_super(struct mddev * mddev,struct md_rdev * rdev)2308 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2309 {
2310 	if (mddev->sync_super) {
2311 		mddev->sync_super(mddev, rdev);
2312 		return;
2313 	}
2314 
2315 	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2316 
2317 	super_types[mddev->major_version].sync_super(mddev, rdev);
2318 }
2319 
match_mddev_units(struct mddev * mddev1,struct mddev * mddev2)2320 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2321 {
2322 	struct md_rdev *rdev, *rdev2;
2323 
2324 	rcu_read_lock();
2325 	rdev_for_each_rcu(rdev, mddev1) {
2326 		if (test_bit(Faulty, &rdev->flags) ||
2327 		    test_bit(Journal, &rdev->flags) ||
2328 		    rdev->raid_disk == -1)
2329 			continue;
2330 		rdev_for_each_rcu(rdev2, mddev2) {
2331 			if (test_bit(Faulty, &rdev2->flags) ||
2332 			    test_bit(Journal, &rdev2->flags) ||
2333 			    rdev2->raid_disk == -1)
2334 				continue;
2335 			if (rdev->bdev->bd_disk == rdev2->bdev->bd_disk) {
2336 				rcu_read_unlock();
2337 				return 1;
2338 			}
2339 		}
2340 	}
2341 	rcu_read_unlock();
2342 	return 0;
2343 }
2344 
2345 static LIST_HEAD(pending_raid_disks);
2346 
2347 /*
2348  * Try to register data integrity profile for an mddev
2349  *
2350  * This is called when an array is started and after a disk has been kicked
2351  * from the array. It only succeeds if all working and active component devices
2352  * are integrity capable with matching profiles.
2353  */
md_integrity_register(struct mddev * mddev)2354 int md_integrity_register(struct mddev *mddev)
2355 {
2356 	if (list_empty(&mddev->disks))
2357 		return 0; /* nothing to do */
2358 	if (mddev_is_dm(mddev) || !blk_get_integrity(mddev->gendisk))
2359 		return 0; /* shouldn't register */
2360 
2361 	pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2362 	if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE) ||
2363 	    (mddev->level != 1 && mddev->level != 10 &&
2364 	     bioset_integrity_create(&mddev->io_clone_set, BIO_POOL_SIZE))) {
2365 		/*
2366 		 * No need to handle the failure of bioset_integrity_create,
2367 		 * because the function is called by md_run() -> pers->run(),
2368 		 * md_run calls bioset_exit -> bioset_integrity_free in case
2369 		 * of failure case.
2370 		 */
2371 		pr_err("md: failed to create integrity pool for %s\n",
2372 		       mdname(mddev));
2373 		return -EINVAL;
2374 	}
2375 	return 0;
2376 }
2377 EXPORT_SYMBOL(md_integrity_register);
2378 
rdev_read_only(struct md_rdev * rdev)2379 static bool rdev_read_only(struct md_rdev *rdev)
2380 {
2381 	return bdev_read_only(rdev->bdev) ||
2382 		(rdev->meta_bdev && bdev_read_only(rdev->meta_bdev));
2383 }
2384 
bind_rdev_to_array(struct md_rdev * rdev,struct mddev * mddev)2385 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2386 {
2387 	char b[BDEVNAME_SIZE];
2388 	int err;
2389 
2390 	/* prevent duplicates */
2391 	if (find_rdev(mddev, rdev->bdev->bd_dev))
2392 		return -EEXIST;
2393 
2394 	if (rdev_read_only(rdev) && mddev->pers)
2395 		return -EROFS;
2396 
2397 	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2398 	if (!test_bit(Journal, &rdev->flags) &&
2399 	    rdev->sectors &&
2400 	    (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2401 		if (mddev->pers) {
2402 			/* Cannot change size, so fail
2403 			 * If mddev->level <= 0, then we don't care
2404 			 * about aligning sizes (e.g. linear)
2405 			 */
2406 			if (mddev->level > 0)
2407 				return -ENOSPC;
2408 		} else
2409 			mddev->dev_sectors = rdev->sectors;
2410 	}
2411 
2412 	/* Verify rdev->desc_nr is unique.
2413 	 * If it is -1, assign a free number, else
2414 	 * check number is not in use
2415 	 */
2416 	rcu_read_lock();
2417 	if (rdev->desc_nr < 0) {
2418 		int choice = 0;
2419 		if (mddev->pers)
2420 			choice = mddev->raid_disks;
2421 		while (md_find_rdev_nr_rcu(mddev, choice))
2422 			choice++;
2423 		rdev->desc_nr = choice;
2424 	} else {
2425 		if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2426 			rcu_read_unlock();
2427 			return -EBUSY;
2428 		}
2429 	}
2430 	rcu_read_unlock();
2431 	if (!test_bit(Journal, &rdev->flags) &&
2432 	    mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2433 		pr_warn("md: %s: array is limited to %d devices\n",
2434 			mdname(mddev), mddev->max_disks);
2435 		return -EBUSY;
2436 	}
2437 	snprintf(b, sizeof(b), "%pg", rdev->bdev);
2438 	strreplace(b, '/', '!');
2439 
2440 	rdev->mddev = mddev;
2441 	pr_debug("md: bind<%s>\n", b);
2442 
2443 	if (mddev->raid_disks)
2444 		mddev_create_serial_pool(mddev, rdev);
2445 
2446 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2447 		goto fail;
2448 
2449 	/* failure here is OK */
2450 	err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
2451 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2452 	rdev->sysfs_unack_badblocks =
2453 		sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
2454 	rdev->sysfs_badblocks =
2455 		sysfs_get_dirent_safe(rdev->kobj.sd, "bad_blocks");
2456 
2457 	list_add_rcu(&rdev->same_set, &mddev->disks);
2458 	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2459 
2460 	/* May as well allow recovery to be retried once */
2461 	mddev->recovery_disabled++;
2462 
2463 	return 0;
2464 
2465  fail:
2466 	pr_warn("md: failed to register dev-%s for %s\n",
2467 		b, mdname(mddev));
2468 	mddev_destroy_serial_pool(mddev, rdev);
2469 	return err;
2470 }
2471 
2472 void md_autodetect_dev(dev_t dev);
2473 
2474 /* just for claiming the bdev */
2475 static struct md_rdev claim_rdev;
2476 
export_rdev(struct md_rdev * rdev,struct mddev * mddev)2477 static void export_rdev(struct md_rdev *rdev, struct mddev *mddev)
2478 {
2479 	pr_debug("md: export_rdev(%pg)\n", rdev->bdev);
2480 	md_rdev_clear(rdev);
2481 #ifndef MODULE
2482 	if (test_bit(AutoDetected, &rdev->flags))
2483 		md_autodetect_dev(rdev->bdev->bd_dev);
2484 #endif
2485 	fput(rdev->bdev_file);
2486 	rdev->bdev = NULL;
2487 	kobject_put(&rdev->kobj);
2488 }
2489 
md_kick_rdev_from_array(struct md_rdev * rdev)2490 static void md_kick_rdev_from_array(struct md_rdev *rdev)
2491 {
2492 	struct mddev *mddev = rdev->mddev;
2493 
2494 	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2495 	list_del_rcu(&rdev->same_set);
2496 	pr_debug("md: unbind<%pg>\n", rdev->bdev);
2497 	mddev_destroy_serial_pool(rdev->mddev, rdev);
2498 	WRITE_ONCE(rdev->mddev, NULL);
2499 	sysfs_remove_link(&rdev->kobj, "block");
2500 	sysfs_put(rdev->sysfs_state);
2501 	sysfs_put(rdev->sysfs_unack_badblocks);
2502 	sysfs_put(rdev->sysfs_badblocks);
2503 	rdev->sysfs_state = NULL;
2504 	rdev->sysfs_unack_badblocks = NULL;
2505 	rdev->sysfs_badblocks = NULL;
2506 	rdev->badblocks.count = 0;
2507 
2508 	synchronize_rcu();
2509 
2510 	/*
2511 	 * kobject_del() will wait for all in progress writers to be done, where
2512 	 * reconfig_mutex is held, hence it can't be called under
2513 	 * reconfig_mutex and it's delayed to mddev_unlock().
2514 	 */
2515 	list_add(&rdev->same_set, &mddev->deleting);
2516 }
2517 
export_array(struct mddev * mddev)2518 static void export_array(struct mddev *mddev)
2519 {
2520 	struct md_rdev *rdev;
2521 
2522 	while (!list_empty(&mddev->disks)) {
2523 		rdev = list_first_entry(&mddev->disks, struct md_rdev,
2524 					same_set);
2525 		md_kick_rdev_from_array(rdev);
2526 	}
2527 	mddev->raid_disks = 0;
2528 	mddev->major_version = 0;
2529 }
2530 
set_in_sync(struct mddev * mddev)2531 static bool set_in_sync(struct mddev *mddev)
2532 {
2533 	lockdep_assert_held(&mddev->lock);
2534 	if (!mddev->in_sync) {
2535 		mddev->sync_checkers++;
2536 		spin_unlock(&mddev->lock);
2537 		percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2538 		spin_lock(&mddev->lock);
2539 		if (!mddev->in_sync &&
2540 		    percpu_ref_is_zero(&mddev->writes_pending)) {
2541 			mddev->in_sync = 1;
2542 			/*
2543 			 * Ensure ->in_sync is visible before we clear
2544 			 * ->sync_checkers.
2545 			 */
2546 			smp_mb();
2547 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2548 			sysfs_notify_dirent_safe(mddev->sysfs_state);
2549 		}
2550 		if (--mddev->sync_checkers == 0)
2551 			percpu_ref_switch_to_percpu(&mddev->writes_pending);
2552 	}
2553 	if (mddev->safemode == 1)
2554 		mddev->safemode = 0;
2555 	return mddev->in_sync;
2556 }
2557 
sync_sbs(struct mddev * mddev,int nospares)2558 static void sync_sbs(struct mddev *mddev, int nospares)
2559 {
2560 	/* Update each superblock (in-memory image), but
2561 	 * if we are allowed to, skip spares which already
2562 	 * have the right event counter, or have one earlier
2563 	 * (which would mean they aren't being marked as dirty
2564 	 * with the rest of the array)
2565 	 */
2566 	struct md_rdev *rdev;
2567 	rdev_for_each(rdev, mddev) {
2568 		if (rdev->sb_events == mddev->events ||
2569 		    (nospares &&
2570 		     rdev->raid_disk < 0 &&
2571 		     rdev->sb_events+1 == mddev->events)) {
2572 			/* Don't update this superblock */
2573 			rdev->sb_loaded = 2;
2574 		} else {
2575 			sync_super(mddev, rdev);
2576 			rdev->sb_loaded = 1;
2577 		}
2578 	}
2579 }
2580 
does_sb_need_changing(struct mddev * mddev)2581 static bool does_sb_need_changing(struct mddev *mddev)
2582 {
2583 	struct md_rdev *rdev = NULL, *iter;
2584 	struct mdp_superblock_1 *sb;
2585 	int role;
2586 
2587 	/* Find a good rdev */
2588 	rdev_for_each(iter, mddev)
2589 		if ((iter->raid_disk >= 0) && !test_bit(Faulty, &iter->flags)) {
2590 			rdev = iter;
2591 			break;
2592 		}
2593 
2594 	/* No good device found. */
2595 	if (!rdev)
2596 		return false;
2597 
2598 	sb = page_address(rdev->sb_page);
2599 	/* Check if a device has become faulty or a spare become active */
2600 	rdev_for_each(rdev, mddev) {
2601 		role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2602 		/* Device activated? */
2603 		if (role == MD_DISK_ROLE_SPARE && rdev->raid_disk >= 0 &&
2604 		    !test_bit(Faulty, &rdev->flags))
2605 			return true;
2606 		/* Device turned faulty? */
2607 		if (test_bit(Faulty, &rdev->flags) && (role < MD_DISK_ROLE_MAX))
2608 			return true;
2609 	}
2610 
2611 	/* Check if any mddev parameters have changed */
2612 	if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2613 	    (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2614 	    (mddev->layout != le32_to_cpu(sb->layout)) ||
2615 	    (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2616 	    (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2617 		return true;
2618 
2619 	return false;
2620 }
2621 
md_update_sb(struct mddev * mddev,int force_change)2622 void md_update_sb(struct mddev *mddev, int force_change)
2623 {
2624 	struct md_rdev *rdev;
2625 	int sync_req;
2626 	int nospares = 0;
2627 	int any_badblocks_changed = 0;
2628 	int ret = -1;
2629 
2630 	if (!md_is_rdwr(mddev)) {
2631 		if (force_change)
2632 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2633 		return;
2634 	}
2635 
2636 repeat:
2637 	if (mddev_is_clustered(mddev)) {
2638 		if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2639 			force_change = 1;
2640 		if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2641 			nospares = 1;
2642 		ret = md_cluster_ops->metadata_update_start(mddev);
2643 		/* Has someone else has updated the sb */
2644 		if (!does_sb_need_changing(mddev)) {
2645 			if (ret == 0)
2646 				md_cluster_ops->metadata_update_cancel(mddev);
2647 			bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2648 							 BIT(MD_SB_CHANGE_DEVS) |
2649 							 BIT(MD_SB_CHANGE_CLEAN));
2650 			return;
2651 		}
2652 	}
2653 
2654 	/*
2655 	 * First make sure individual recovery_offsets are correct
2656 	 * curr_resync_completed can only be used during recovery.
2657 	 * During reshape/resync it might use array-addresses rather
2658 	 * that device addresses.
2659 	 */
2660 	rdev_for_each(rdev, mddev) {
2661 		if (rdev->raid_disk >= 0 &&
2662 		    mddev->delta_disks >= 0 &&
2663 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2664 		    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2665 		    !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2666 		    !test_bit(Journal, &rdev->flags) &&
2667 		    !test_bit(In_sync, &rdev->flags) &&
2668 		    mddev->curr_resync_completed > rdev->recovery_offset)
2669 				rdev->recovery_offset = mddev->curr_resync_completed;
2670 
2671 	}
2672 	if (!mddev->persistent) {
2673 		clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2674 		clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2675 		if (!mddev->external) {
2676 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2677 			rdev_for_each(rdev, mddev) {
2678 				if (rdev->badblocks.changed) {
2679 					rdev->badblocks.changed = 0;
2680 					ack_all_badblocks(&rdev->badblocks);
2681 					md_error(mddev, rdev);
2682 				}
2683 				clear_bit(Blocked, &rdev->flags);
2684 				clear_bit(BlockedBadBlocks, &rdev->flags);
2685 				wake_up(&rdev->blocked_wait);
2686 			}
2687 		}
2688 		wake_up(&mddev->sb_wait);
2689 		return;
2690 	}
2691 
2692 	spin_lock(&mddev->lock);
2693 
2694 	mddev->utime = ktime_get_real_seconds();
2695 
2696 	if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2697 		force_change = 1;
2698 	if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2699 		/* just a clean<-> dirty transition, possibly leave spares alone,
2700 		 * though if events isn't the right even/odd, we will have to do
2701 		 * spares after all
2702 		 */
2703 		nospares = 1;
2704 	if (force_change)
2705 		nospares = 0;
2706 	if (mddev->degraded)
2707 		/* If the array is degraded, then skipping spares is both
2708 		 * dangerous and fairly pointless.
2709 		 * Dangerous because a device that was removed from the array
2710 		 * might have a event_count that still looks up-to-date,
2711 		 * so it can be re-added without a resync.
2712 		 * Pointless because if there are any spares to skip,
2713 		 * then a recovery will happen and soon that array won't
2714 		 * be degraded any more and the spare can go back to sleep then.
2715 		 */
2716 		nospares = 0;
2717 
2718 	sync_req = mddev->in_sync;
2719 
2720 	/* If this is just a dirty<->clean transition, and the array is clean
2721 	 * and 'events' is odd, we can roll back to the previous clean state */
2722 	if (nospares
2723 	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2724 	    && mddev->can_decrease_events
2725 	    && mddev->events != 1) {
2726 		mddev->events--;
2727 		mddev->can_decrease_events = 0;
2728 	} else {
2729 		/* otherwise we have to go forward and ... */
2730 		mddev->events ++;
2731 		mddev->can_decrease_events = nospares;
2732 	}
2733 
2734 	/*
2735 	 * This 64-bit counter should never wrap.
2736 	 * Either we are in around ~1 trillion A.C., assuming
2737 	 * 1 reboot per second, or we have a bug...
2738 	 */
2739 	WARN_ON(mddev->events == 0);
2740 
2741 	rdev_for_each(rdev, mddev) {
2742 		if (rdev->badblocks.changed)
2743 			any_badblocks_changed++;
2744 		if (test_bit(Faulty, &rdev->flags))
2745 			set_bit(FaultRecorded, &rdev->flags);
2746 	}
2747 
2748 	sync_sbs(mddev, nospares);
2749 	spin_unlock(&mddev->lock);
2750 
2751 	pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2752 		 mdname(mddev), mddev->in_sync);
2753 
2754 	mddev_add_trace_msg(mddev, "md md_update_sb");
2755 rewrite:
2756 	mddev->bitmap_ops->update_sb(mddev->bitmap);
2757 	rdev_for_each(rdev, mddev) {
2758 		if (rdev->sb_loaded != 1)
2759 			continue; /* no noise on spare devices */
2760 
2761 		if (!test_bit(Faulty, &rdev->flags)) {
2762 			md_super_write(mddev,rdev,
2763 				       rdev->sb_start, rdev->sb_size,
2764 				       rdev->sb_page);
2765 			pr_debug("md: (write) %pg's sb offset: %llu\n",
2766 				 rdev->bdev,
2767 				 (unsigned long long)rdev->sb_start);
2768 			rdev->sb_events = mddev->events;
2769 			if (rdev->badblocks.size) {
2770 				md_super_write(mddev, rdev,
2771 					       rdev->badblocks.sector,
2772 					       rdev->badblocks.size << 9,
2773 					       rdev->bb_page);
2774 				rdev->badblocks.size = 0;
2775 			}
2776 
2777 		} else
2778 			pr_debug("md: %pg (skipping faulty)\n",
2779 				 rdev->bdev);
2780 	}
2781 	if (md_super_wait(mddev) < 0)
2782 		goto rewrite;
2783 	/* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2784 
2785 	if (mddev_is_clustered(mddev) && ret == 0)
2786 		md_cluster_ops->metadata_update_finish(mddev);
2787 
2788 	if (mddev->in_sync != sync_req ||
2789 	    !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2790 			       BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2791 		/* have to write it out again */
2792 		goto repeat;
2793 	wake_up(&mddev->sb_wait);
2794 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2795 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
2796 
2797 	rdev_for_each(rdev, mddev) {
2798 		if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2799 			clear_bit(Blocked, &rdev->flags);
2800 
2801 		if (any_badblocks_changed)
2802 			ack_all_badblocks(&rdev->badblocks);
2803 		clear_bit(BlockedBadBlocks, &rdev->flags);
2804 		wake_up(&rdev->blocked_wait);
2805 	}
2806 }
2807 EXPORT_SYMBOL(md_update_sb);
2808 
add_bound_rdev(struct md_rdev * rdev)2809 static int add_bound_rdev(struct md_rdev *rdev)
2810 {
2811 	struct mddev *mddev = rdev->mddev;
2812 	int err = 0;
2813 	bool add_journal = test_bit(Journal, &rdev->flags);
2814 
2815 	if (!mddev->pers->hot_remove_disk || add_journal) {
2816 		/* If there is hot_add_disk but no hot_remove_disk
2817 		 * then added disks for geometry changes,
2818 		 * and should be added immediately.
2819 		 */
2820 		super_types[mddev->major_version].
2821 			validate_super(mddev, NULL/*freshest*/, rdev);
2822 		err = mddev->pers->hot_add_disk(mddev, rdev);
2823 		if (err) {
2824 			md_kick_rdev_from_array(rdev);
2825 			return err;
2826 		}
2827 	}
2828 	sysfs_notify_dirent_safe(rdev->sysfs_state);
2829 
2830 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2831 	if (mddev->degraded)
2832 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2833 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2834 	md_new_event();
2835 	return 0;
2836 }
2837 
2838 /* words written to sysfs files may, or may not, be \n terminated.
2839  * We want to accept with case. For this we use cmd_match.
2840  */
cmd_match(const char * cmd,const char * str)2841 static int cmd_match(const char *cmd, const char *str)
2842 {
2843 	/* See if cmd, written into a sysfs file, matches
2844 	 * str.  They must either be the same, or cmd can
2845 	 * have a trailing newline
2846 	 */
2847 	while (*cmd && *str && *cmd == *str) {
2848 		cmd++;
2849 		str++;
2850 	}
2851 	if (*cmd == '\n')
2852 		cmd++;
2853 	if (*str || *cmd)
2854 		return 0;
2855 	return 1;
2856 }
2857 
2858 struct rdev_sysfs_entry {
2859 	struct attribute attr;
2860 	ssize_t (*show)(struct md_rdev *, char *);
2861 	ssize_t (*store)(struct md_rdev *, const char *, size_t);
2862 };
2863 
2864 static ssize_t
state_show(struct md_rdev * rdev,char * page)2865 state_show(struct md_rdev *rdev, char *page)
2866 {
2867 	char *sep = ",";
2868 	size_t len = 0;
2869 	unsigned long flags = READ_ONCE(rdev->flags);
2870 
2871 	if (test_bit(Faulty, &flags) ||
2872 	    (!test_bit(ExternalBbl, &flags) &&
2873 	    rdev->badblocks.unacked_exist))
2874 		len += sprintf(page+len, "faulty%s", sep);
2875 	if (test_bit(In_sync, &flags))
2876 		len += sprintf(page+len, "in_sync%s", sep);
2877 	if (test_bit(Journal, &flags))
2878 		len += sprintf(page+len, "journal%s", sep);
2879 	if (test_bit(WriteMostly, &flags))
2880 		len += sprintf(page+len, "write_mostly%s", sep);
2881 	if (test_bit(Blocked, &flags) ||
2882 	    (rdev->badblocks.unacked_exist
2883 	     && !test_bit(Faulty, &flags)))
2884 		len += sprintf(page+len, "blocked%s", sep);
2885 	if (!test_bit(Faulty, &flags) &&
2886 	    !test_bit(Journal, &flags) &&
2887 	    !test_bit(In_sync, &flags))
2888 		len += sprintf(page+len, "spare%s", sep);
2889 	if (test_bit(WriteErrorSeen, &flags))
2890 		len += sprintf(page+len, "write_error%s", sep);
2891 	if (test_bit(WantReplacement, &flags))
2892 		len += sprintf(page+len, "want_replacement%s", sep);
2893 	if (test_bit(Replacement, &flags))
2894 		len += sprintf(page+len, "replacement%s", sep);
2895 	if (test_bit(ExternalBbl, &flags))
2896 		len += sprintf(page+len, "external_bbl%s", sep);
2897 	if (test_bit(FailFast, &flags))
2898 		len += sprintf(page+len, "failfast%s", sep);
2899 
2900 	if (len)
2901 		len -= strlen(sep);
2902 
2903 	return len+sprintf(page+len, "\n");
2904 }
2905 
2906 static ssize_t
state_store(struct md_rdev * rdev,const char * buf,size_t len)2907 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2908 {
2909 	/* can write
2910 	 *  faulty  - simulates an error
2911 	 *  remove  - disconnects the device
2912 	 *  writemostly - sets write_mostly
2913 	 *  -writemostly - clears write_mostly
2914 	 *  blocked - sets the Blocked flags
2915 	 *  -blocked - clears the Blocked and possibly simulates an error
2916 	 *  insync - sets Insync providing device isn't active
2917 	 *  -insync - clear Insync for a device with a slot assigned,
2918 	 *            so that it gets rebuilt based on bitmap
2919 	 *  write_error - sets WriteErrorSeen
2920 	 *  -write_error - clears WriteErrorSeen
2921 	 *  {,-}failfast - set/clear FailFast
2922 	 */
2923 
2924 	struct mddev *mddev = rdev->mddev;
2925 	int err = -EINVAL;
2926 	bool need_update_sb = false;
2927 
2928 	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2929 		md_error(rdev->mddev, rdev);
2930 
2931 		if (test_bit(MD_BROKEN, &rdev->mddev->flags))
2932 			err = -EBUSY;
2933 		else
2934 			err = 0;
2935 	} else if (cmd_match(buf, "remove")) {
2936 		if (rdev->mddev->pers) {
2937 			clear_bit(Blocked, &rdev->flags);
2938 			remove_and_add_spares(rdev->mddev, rdev);
2939 		}
2940 		if (rdev->raid_disk >= 0)
2941 			err = -EBUSY;
2942 		else {
2943 			err = 0;
2944 			if (mddev_is_clustered(mddev))
2945 				err = md_cluster_ops->remove_disk(mddev, rdev);
2946 
2947 			if (err == 0) {
2948 				md_kick_rdev_from_array(rdev);
2949 				if (mddev->pers)
2950 					set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2951 				md_new_event();
2952 			}
2953 		}
2954 	} else if (cmd_match(buf, "writemostly")) {
2955 		set_bit(WriteMostly, &rdev->flags);
2956 		mddev_create_serial_pool(rdev->mddev, rdev);
2957 		need_update_sb = true;
2958 		err = 0;
2959 	} else if (cmd_match(buf, "-writemostly")) {
2960 		mddev_destroy_serial_pool(rdev->mddev, rdev);
2961 		clear_bit(WriteMostly, &rdev->flags);
2962 		need_update_sb = true;
2963 		err = 0;
2964 	} else if (cmd_match(buf, "blocked")) {
2965 		set_bit(Blocked, &rdev->flags);
2966 		err = 0;
2967 	} else if (cmd_match(buf, "-blocked")) {
2968 		if (!test_bit(Faulty, &rdev->flags) &&
2969 		    !test_bit(ExternalBbl, &rdev->flags) &&
2970 		    rdev->badblocks.unacked_exist) {
2971 			/* metadata handler doesn't understand badblocks,
2972 			 * so we need to fail the device
2973 			 */
2974 			md_error(rdev->mddev, rdev);
2975 		}
2976 		clear_bit(Blocked, &rdev->flags);
2977 		clear_bit(BlockedBadBlocks, &rdev->flags);
2978 		wake_up(&rdev->blocked_wait);
2979 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2980 
2981 		err = 0;
2982 	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2983 		set_bit(In_sync, &rdev->flags);
2984 		err = 0;
2985 	} else if (cmd_match(buf, "failfast")) {
2986 		set_bit(FailFast, &rdev->flags);
2987 		need_update_sb = true;
2988 		err = 0;
2989 	} else if (cmd_match(buf, "-failfast")) {
2990 		clear_bit(FailFast, &rdev->flags);
2991 		need_update_sb = true;
2992 		err = 0;
2993 	} else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2994 		   !test_bit(Journal, &rdev->flags)) {
2995 		if (rdev->mddev->pers == NULL) {
2996 			clear_bit(In_sync, &rdev->flags);
2997 			rdev->saved_raid_disk = rdev->raid_disk;
2998 			rdev->raid_disk = -1;
2999 			err = 0;
3000 		}
3001 	} else if (cmd_match(buf, "write_error")) {
3002 		set_bit(WriteErrorSeen, &rdev->flags);
3003 		err = 0;
3004 	} else if (cmd_match(buf, "-write_error")) {
3005 		clear_bit(WriteErrorSeen, &rdev->flags);
3006 		err = 0;
3007 	} else if (cmd_match(buf, "want_replacement")) {
3008 		/* Any non-spare device that is not a replacement can
3009 		 * become want_replacement at any time, but we then need to
3010 		 * check if recovery is needed.
3011 		 */
3012 		if (rdev->raid_disk >= 0 &&
3013 		    !test_bit(Journal, &rdev->flags) &&
3014 		    !test_bit(Replacement, &rdev->flags))
3015 			set_bit(WantReplacement, &rdev->flags);
3016 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3017 		err = 0;
3018 	} else if (cmd_match(buf, "-want_replacement")) {
3019 		/* Clearing 'want_replacement' is always allowed.
3020 		 * Once replacements starts it is too late though.
3021 		 */
3022 		err = 0;
3023 		clear_bit(WantReplacement, &rdev->flags);
3024 	} else if (cmd_match(buf, "replacement")) {
3025 		/* Can only set a device as a replacement when array has not
3026 		 * yet been started.  Once running, replacement is automatic
3027 		 * from spares, or by assigning 'slot'.
3028 		 */
3029 		if (rdev->mddev->pers)
3030 			err = -EBUSY;
3031 		else {
3032 			set_bit(Replacement, &rdev->flags);
3033 			err = 0;
3034 		}
3035 	} else if (cmd_match(buf, "-replacement")) {
3036 		/* Similarly, can only clear Replacement before start */
3037 		if (rdev->mddev->pers)
3038 			err = -EBUSY;
3039 		else {
3040 			clear_bit(Replacement, &rdev->flags);
3041 			err = 0;
3042 		}
3043 	} else if (cmd_match(buf, "re-add")) {
3044 		if (!rdev->mddev->pers)
3045 			err = -EINVAL;
3046 		else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
3047 				rdev->saved_raid_disk >= 0) {
3048 			/* clear_bit is performed _after_ all the devices
3049 			 * have their local Faulty bit cleared. If any writes
3050 			 * happen in the meantime in the local node, they
3051 			 * will land in the local bitmap, which will be synced
3052 			 * by this node eventually
3053 			 */
3054 			if (!mddev_is_clustered(rdev->mddev) ||
3055 			    (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
3056 				clear_bit(Faulty, &rdev->flags);
3057 				err = add_bound_rdev(rdev);
3058 			}
3059 		} else
3060 			err = -EBUSY;
3061 	} else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
3062 		set_bit(ExternalBbl, &rdev->flags);
3063 		rdev->badblocks.shift = 0;
3064 		err = 0;
3065 	} else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
3066 		clear_bit(ExternalBbl, &rdev->flags);
3067 		err = 0;
3068 	}
3069 	if (need_update_sb)
3070 		md_update_sb(mddev, 1);
3071 	if (!err)
3072 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3073 	return err ? err : len;
3074 }
3075 static struct rdev_sysfs_entry rdev_state =
3076 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3077 
3078 static ssize_t
errors_show(struct md_rdev * rdev,char * page)3079 errors_show(struct md_rdev *rdev, char *page)
3080 {
3081 	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3082 }
3083 
3084 static ssize_t
errors_store(struct md_rdev * rdev,const char * buf,size_t len)3085 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3086 {
3087 	unsigned int n;
3088 	int rv;
3089 
3090 	rv = kstrtouint(buf, 10, &n);
3091 	if (rv < 0)
3092 		return rv;
3093 	atomic_set(&rdev->corrected_errors, n);
3094 	return len;
3095 }
3096 static struct rdev_sysfs_entry rdev_errors =
3097 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3098 
3099 static ssize_t
slot_show(struct md_rdev * rdev,char * page)3100 slot_show(struct md_rdev *rdev, char *page)
3101 {
3102 	if (test_bit(Journal, &rdev->flags))
3103 		return sprintf(page, "journal\n");
3104 	else if (rdev->raid_disk < 0)
3105 		return sprintf(page, "none\n");
3106 	else
3107 		return sprintf(page, "%d\n", rdev->raid_disk);
3108 }
3109 
3110 static ssize_t
slot_store(struct md_rdev * rdev,const char * buf,size_t len)3111 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3112 {
3113 	int slot;
3114 	int err;
3115 
3116 	if (test_bit(Journal, &rdev->flags))
3117 		return -EBUSY;
3118 	if (strncmp(buf, "none", 4)==0)
3119 		slot = -1;
3120 	else {
3121 		err = kstrtouint(buf, 10, (unsigned int *)&slot);
3122 		if (err < 0)
3123 			return err;
3124 		if (slot < 0)
3125 			/* overflow */
3126 			return -ENOSPC;
3127 	}
3128 	if (rdev->mddev->pers && slot == -1) {
3129 		/* Setting 'slot' on an active array requires also
3130 		 * updating the 'rd%d' link, and communicating
3131 		 * with the personality with ->hot_*_disk.
3132 		 * For now we only support removing
3133 		 * failed/spare devices.  This normally happens automatically,
3134 		 * but not when the metadata is externally managed.
3135 		 */
3136 		if (rdev->raid_disk == -1)
3137 			return -EEXIST;
3138 		/* personality does all needed checks */
3139 		if (rdev->mddev->pers->hot_remove_disk == NULL)
3140 			return -EINVAL;
3141 		clear_bit(Blocked, &rdev->flags);
3142 		remove_and_add_spares(rdev->mddev, rdev);
3143 		if (rdev->raid_disk >= 0)
3144 			return -EBUSY;
3145 		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3146 	} else if (rdev->mddev->pers) {
3147 		/* Activating a spare .. or possibly reactivating
3148 		 * if we ever get bitmaps working here.
3149 		 */
3150 		int err;
3151 
3152 		if (rdev->raid_disk != -1)
3153 			return -EBUSY;
3154 
3155 		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3156 			return -EBUSY;
3157 
3158 		if (rdev->mddev->pers->hot_add_disk == NULL)
3159 			return -EINVAL;
3160 
3161 		if (slot >= rdev->mddev->raid_disks &&
3162 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3163 			return -ENOSPC;
3164 
3165 		rdev->raid_disk = slot;
3166 		if (test_bit(In_sync, &rdev->flags))
3167 			rdev->saved_raid_disk = slot;
3168 		else
3169 			rdev->saved_raid_disk = -1;
3170 		clear_bit(In_sync, &rdev->flags);
3171 		clear_bit(Bitmap_sync, &rdev->flags);
3172 		err = rdev->mddev->pers->hot_add_disk(rdev->mddev, rdev);
3173 		if (err) {
3174 			rdev->raid_disk = -1;
3175 			return err;
3176 		} else
3177 			sysfs_notify_dirent_safe(rdev->sysfs_state);
3178 		/* failure here is OK */;
3179 		sysfs_link_rdev(rdev->mddev, rdev);
3180 		/* don't wakeup anyone, leave that to userspace. */
3181 	} else {
3182 		if (slot >= rdev->mddev->raid_disks &&
3183 		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3184 			return -ENOSPC;
3185 		rdev->raid_disk = slot;
3186 		/* assume it is working */
3187 		clear_bit(Faulty, &rdev->flags);
3188 		clear_bit(WriteMostly, &rdev->flags);
3189 		set_bit(In_sync, &rdev->flags);
3190 		sysfs_notify_dirent_safe(rdev->sysfs_state);
3191 	}
3192 	return len;
3193 }
3194 
3195 static struct rdev_sysfs_entry rdev_slot =
3196 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3197 
3198 static ssize_t
offset_show(struct md_rdev * rdev,char * page)3199 offset_show(struct md_rdev *rdev, char *page)
3200 {
3201 	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3202 }
3203 
3204 static ssize_t
offset_store(struct md_rdev * rdev,const char * buf,size_t len)3205 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3206 {
3207 	unsigned long long offset;
3208 	if (kstrtoull(buf, 10, &offset) < 0)
3209 		return -EINVAL;
3210 	if (rdev->mddev->pers && rdev->raid_disk >= 0)
3211 		return -EBUSY;
3212 	if (rdev->sectors && rdev->mddev->external)
3213 		/* Must set offset before size, so overlap checks
3214 		 * can be sane */
3215 		return -EBUSY;
3216 	rdev->data_offset = offset;
3217 	rdev->new_data_offset = offset;
3218 	return len;
3219 }
3220 
3221 static struct rdev_sysfs_entry rdev_offset =
3222 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3223 
new_offset_show(struct md_rdev * rdev,char * page)3224 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3225 {
3226 	return sprintf(page, "%llu\n",
3227 		       (unsigned long long)rdev->new_data_offset);
3228 }
3229 
new_offset_store(struct md_rdev * rdev,const char * buf,size_t len)3230 static ssize_t new_offset_store(struct md_rdev *rdev,
3231 				const char *buf, size_t len)
3232 {
3233 	unsigned long long new_offset;
3234 	struct mddev *mddev = rdev->mddev;
3235 
3236 	if (kstrtoull(buf, 10, &new_offset) < 0)
3237 		return -EINVAL;
3238 
3239 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3240 		return -EBUSY;
3241 	if (new_offset == rdev->data_offset)
3242 		/* reset is always permitted */
3243 		;
3244 	else if (new_offset > rdev->data_offset) {
3245 		/* must not push array size beyond rdev_sectors */
3246 		if (new_offset - rdev->data_offset
3247 		    + mddev->dev_sectors > rdev->sectors)
3248 				return -E2BIG;
3249 	}
3250 	/* Metadata worries about other space details. */
3251 
3252 	/* decreasing the offset is inconsistent with a backwards
3253 	 * reshape.
3254 	 */
3255 	if (new_offset < rdev->data_offset &&
3256 	    mddev->reshape_backwards)
3257 		return -EINVAL;
3258 	/* Increasing offset is inconsistent with forwards
3259 	 * reshape.  reshape_direction should be set to
3260 	 * 'backwards' first.
3261 	 */
3262 	if (new_offset > rdev->data_offset &&
3263 	    !mddev->reshape_backwards)
3264 		return -EINVAL;
3265 
3266 	if (mddev->pers && mddev->persistent &&
3267 	    !super_types[mddev->major_version]
3268 	    .allow_new_offset(rdev, new_offset))
3269 		return -E2BIG;
3270 	rdev->new_data_offset = new_offset;
3271 	if (new_offset > rdev->data_offset)
3272 		mddev->reshape_backwards = 1;
3273 	else if (new_offset < rdev->data_offset)
3274 		mddev->reshape_backwards = 0;
3275 
3276 	return len;
3277 }
3278 static struct rdev_sysfs_entry rdev_new_offset =
3279 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3280 
3281 static ssize_t
rdev_size_show(struct md_rdev * rdev,char * page)3282 rdev_size_show(struct md_rdev *rdev, char *page)
3283 {
3284 	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3285 }
3286 
md_rdevs_overlap(struct md_rdev * a,struct md_rdev * b)3287 static int md_rdevs_overlap(struct md_rdev *a, struct md_rdev *b)
3288 {
3289 	/* check if two start/length pairs overlap */
3290 	if (a->data_offset + a->sectors <= b->data_offset)
3291 		return false;
3292 	if (b->data_offset + b->sectors <= a->data_offset)
3293 		return false;
3294 	return true;
3295 }
3296 
md_rdev_overlaps(struct md_rdev * rdev)3297 static bool md_rdev_overlaps(struct md_rdev *rdev)
3298 {
3299 	struct mddev *mddev;
3300 	struct md_rdev *rdev2;
3301 
3302 	spin_lock(&all_mddevs_lock);
3303 	list_for_each_entry(mddev, &all_mddevs, all_mddevs) {
3304 		if (test_bit(MD_DELETED, &mddev->flags))
3305 			continue;
3306 		rdev_for_each(rdev2, mddev) {
3307 			if (rdev != rdev2 && rdev->bdev == rdev2->bdev &&
3308 			    md_rdevs_overlap(rdev, rdev2)) {
3309 				spin_unlock(&all_mddevs_lock);
3310 				return true;
3311 			}
3312 		}
3313 	}
3314 	spin_unlock(&all_mddevs_lock);
3315 	return false;
3316 }
3317 
strict_blocks_to_sectors(const char * buf,sector_t * sectors)3318 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3319 {
3320 	unsigned long long blocks;
3321 	sector_t new;
3322 
3323 	if (kstrtoull(buf, 10, &blocks) < 0)
3324 		return -EINVAL;
3325 
3326 	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3327 		return -EINVAL; /* sector conversion overflow */
3328 
3329 	new = blocks * 2;
3330 	if (new != blocks * 2)
3331 		return -EINVAL; /* unsigned long long to sector_t overflow */
3332 
3333 	*sectors = new;
3334 	return 0;
3335 }
3336 
3337 static ssize_t
rdev_size_store(struct md_rdev * rdev,const char * buf,size_t len)3338 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3339 {
3340 	struct mddev *my_mddev = rdev->mddev;
3341 	sector_t oldsectors = rdev->sectors;
3342 	sector_t sectors;
3343 
3344 	if (test_bit(Journal, &rdev->flags))
3345 		return -EBUSY;
3346 	if (strict_blocks_to_sectors(buf, &sectors) < 0)
3347 		return -EINVAL;
3348 	if (rdev->data_offset != rdev->new_data_offset)
3349 		return -EINVAL; /* too confusing */
3350 	if (my_mddev->pers && rdev->raid_disk >= 0) {
3351 		if (my_mddev->persistent) {
3352 			sectors = super_types[my_mddev->major_version].
3353 				rdev_size_change(rdev, sectors);
3354 			if (!sectors)
3355 				return -EBUSY;
3356 		} else if (!sectors)
3357 			sectors = bdev_nr_sectors(rdev->bdev) -
3358 				rdev->data_offset;
3359 		if (!my_mddev->pers->resize)
3360 			/* Cannot change size for RAID0 or Linear etc */
3361 			return -EINVAL;
3362 	}
3363 	if (sectors < my_mddev->dev_sectors)
3364 		return -EINVAL; /* component must fit device */
3365 
3366 	rdev->sectors = sectors;
3367 
3368 	/*
3369 	 * Check that all other rdevs with the same bdev do not overlap.  This
3370 	 * check does not provide a hard guarantee, it just helps avoid
3371 	 * dangerous mistakes.
3372 	 */
3373 	if (sectors > oldsectors && my_mddev->external &&
3374 	    md_rdev_overlaps(rdev)) {
3375 		/*
3376 		 * Someone else could have slipped in a size change here, but
3377 		 * doing so is just silly.  We put oldsectors back because we
3378 		 * know it is safe, and trust userspace not to race with itself.
3379 		 */
3380 		rdev->sectors = oldsectors;
3381 		return -EBUSY;
3382 	}
3383 	return len;
3384 }
3385 
3386 static struct rdev_sysfs_entry rdev_size =
3387 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3388 
recovery_start_show(struct md_rdev * rdev,char * page)3389 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3390 {
3391 	unsigned long long recovery_start = rdev->recovery_offset;
3392 
3393 	if (test_bit(In_sync, &rdev->flags) ||
3394 	    recovery_start == MaxSector)
3395 		return sprintf(page, "none\n");
3396 
3397 	return sprintf(page, "%llu\n", recovery_start);
3398 }
3399 
recovery_start_store(struct md_rdev * rdev,const char * buf,size_t len)3400 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3401 {
3402 	unsigned long long recovery_start;
3403 
3404 	if (cmd_match(buf, "none"))
3405 		recovery_start = MaxSector;
3406 	else if (kstrtoull(buf, 10, &recovery_start))
3407 		return -EINVAL;
3408 
3409 	if (rdev->mddev->pers &&
3410 	    rdev->raid_disk >= 0)
3411 		return -EBUSY;
3412 
3413 	rdev->recovery_offset = recovery_start;
3414 	if (recovery_start == MaxSector)
3415 		set_bit(In_sync, &rdev->flags);
3416 	else
3417 		clear_bit(In_sync, &rdev->flags);
3418 	return len;
3419 }
3420 
3421 static struct rdev_sysfs_entry rdev_recovery_start =
3422 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3423 
3424 /* sysfs access to bad-blocks list.
3425  * We present two files.
3426  * 'bad-blocks' lists sector numbers and lengths of ranges that
3427  *    are recorded as bad.  The list is truncated to fit within
3428  *    the one-page limit of sysfs.
3429  *    Writing "sector length" to this file adds an acknowledged
3430  *    bad block list.
3431  * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3432  *    been acknowledged.  Writing to this file adds bad blocks
3433  *    without acknowledging them.  This is largely for testing.
3434  */
bb_show(struct md_rdev * rdev,char * page)3435 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3436 {
3437 	return badblocks_show(&rdev->badblocks, page, 0);
3438 }
bb_store(struct md_rdev * rdev,const char * page,size_t len)3439 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3440 {
3441 	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3442 	/* Maybe that ack was all we needed */
3443 	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3444 		wake_up(&rdev->blocked_wait);
3445 	return rv;
3446 }
3447 static struct rdev_sysfs_entry rdev_bad_blocks =
3448 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3449 
ubb_show(struct md_rdev * rdev,char * page)3450 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3451 {
3452 	return badblocks_show(&rdev->badblocks, page, 1);
3453 }
ubb_store(struct md_rdev * rdev,const char * page,size_t len)3454 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3455 {
3456 	return badblocks_store(&rdev->badblocks, page, len, 1);
3457 }
3458 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3459 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3460 
3461 static ssize_t
ppl_sector_show(struct md_rdev * rdev,char * page)3462 ppl_sector_show(struct md_rdev *rdev, char *page)
3463 {
3464 	return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3465 }
3466 
3467 static ssize_t
ppl_sector_store(struct md_rdev * rdev,const char * buf,size_t len)3468 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3469 {
3470 	unsigned long long sector;
3471 
3472 	if (kstrtoull(buf, 10, &sector) < 0)
3473 		return -EINVAL;
3474 	if (sector != (sector_t)sector)
3475 		return -EINVAL;
3476 
3477 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3478 	    rdev->raid_disk >= 0)
3479 		return -EBUSY;
3480 
3481 	if (rdev->mddev->persistent) {
3482 		if (rdev->mddev->major_version == 0)
3483 			return -EINVAL;
3484 		if ((sector > rdev->sb_start &&
3485 		     sector - rdev->sb_start > S16_MAX) ||
3486 		    (sector < rdev->sb_start &&
3487 		     rdev->sb_start - sector > -S16_MIN))
3488 			return -EINVAL;
3489 		rdev->ppl.offset = sector - rdev->sb_start;
3490 	} else if (!rdev->mddev->external) {
3491 		return -EBUSY;
3492 	}
3493 	rdev->ppl.sector = sector;
3494 	return len;
3495 }
3496 
3497 static struct rdev_sysfs_entry rdev_ppl_sector =
3498 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3499 
3500 static ssize_t
ppl_size_show(struct md_rdev * rdev,char * page)3501 ppl_size_show(struct md_rdev *rdev, char *page)
3502 {
3503 	return sprintf(page, "%u\n", rdev->ppl.size);
3504 }
3505 
3506 static ssize_t
ppl_size_store(struct md_rdev * rdev,const char * buf,size_t len)3507 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3508 {
3509 	unsigned int size;
3510 
3511 	if (kstrtouint(buf, 10, &size) < 0)
3512 		return -EINVAL;
3513 
3514 	if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3515 	    rdev->raid_disk >= 0)
3516 		return -EBUSY;
3517 
3518 	if (rdev->mddev->persistent) {
3519 		if (rdev->mddev->major_version == 0)
3520 			return -EINVAL;
3521 		if (size > U16_MAX)
3522 			return -EINVAL;
3523 	} else if (!rdev->mddev->external) {
3524 		return -EBUSY;
3525 	}
3526 	rdev->ppl.size = size;
3527 	return len;
3528 }
3529 
3530 static struct rdev_sysfs_entry rdev_ppl_size =
3531 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3532 
3533 static struct attribute *rdev_default_attrs[] = {
3534 	&rdev_state.attr,
3535 	&rdev_errors.attr,
3536 	&rdev_slot.attr,
3537 	&rdev_offset.attr,
3538 	&rdev_new_offset.attr,
3539 	&rdev_size.attr,
3540 	&rdev_recovery_start.attr,
3541 	&rdev_bad_blocks.attr,
3542 	&rdev_unack_bad_blocks.attr,
3543 	&rdev_ppl_sector.attr,
3544 	&rdev_ppl_size.attr,
3545 	NULL,
3546 };
3547 ATTRIBUTE_GROUPS(rdev_default);
3548 static ssize_t
rdev_attr_show(struct kobject * kobj,struct attribute * attr,char * page)3549 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3550 {
3551 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3552 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3553 
3554 	if (!entry->show)
3555 		return -EIO;
3556 	if (!rdev->mddev)
3557 		return -ENODEV;
3558 	return entry->show(rdev, page);
3559 }
3560 
3561 static ssize_t
rdev_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)3562 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3563 	      const char *page, size_t length)
3564 {
3565 	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3566 	struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3567 	struct kernfs_node *kn = NULL;
3568 	bool suspend = false;
3569 	ssize_t rv;
3570 	struct mddev *mddev = READ_ONCE(rdev->mddev);
3571 
3572 	if (!entry->store)
3573 		return -EIO;
3574 	if (!capable(CAP_SYS_ADMIN))
3575 		return -EACCES;
3576 	if (!mddev)
3577 		return -ENODEV;
3578 
3579 	if (entry->store == state_store) {
3580 		if (cmd_match(page, "remove"))
3581 			kn = sysfs_break_active_protection(kobj, attr);
3582 		if (cmd_match(page, "remove") || cmd_match(page, "re-add") ||
3583 		    cmd_match(page, "writemostly") ||
3584 		    cmd_match(page, "-writemostly"))
3585 			suspend = true;
3586 	}
3587 
3588 	rv = suspend ? mddev_suspend_and_lock(mddev) : mddev_lock(mddev);
3589 	if (!rv) {
3590 		if (rdev->mddev == NULL)
3591 			rv = -ENODEV;
3592 		else
3593 			rv = entry->store(rdev, page, length);
3594 		suspend ? mddev_unlock_and_resume(mddev) : mddev_unlock(mddev);
3595 	}
3596 
3597 	if (kn)
3598 		sysfs_unbreak_active_protection(kn);
3599 
3600 	return rv;
3601 }
3602 
rdev_free(struct kobject * ko)3603 static void rdev_free(struct kobject *ko)
3604 {
3605 	struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3606 	kfree(rdev);
3607 }
3608 static const struct sysfs_ops rdev_sysfs_ops = {
3609 	.show		= rdev_attr_show,
3610 	.store		= rdev_attr_store,
3611 };
3612 static const struct kobj_type rdev_ktype = {
3613 	.release	= rdev_free,
3614 	.sysfs_ops	= &rdev_sysfs_ops,
3615 	.default_groups	= rdev_default_groups,
3616 };
3617 
md_rdev_init(struct md_rdev * rdev)3618 int md_rdev_init(struct md_rdev *rdev)
3619 {
3620 	rdev->desc_nr = -1;
3621 	rdev->saved_raid_disk = -1;
3622 	rdev->raid_disk = -1;
3623 	rdev->flags = 0;
3624 	rdev->data_offset = 0;
3625 	rdev->new_data_offset = 0;
3626 	rdev->sb_events = 0;
3627 	rdev->last_read_error = 0;
3628 	rdev->sb_loaded = 0;
3629 	rdev->bb_page = NULL;
3630 	atomic_set(&rdev->nr_pending, 0);
3631 	atomic_set(&rdev->read_errors, 0);
3632 	atomic_set(&rdev->corrected_errors, 0);
3633 
3634 	INIT_LIST_HEAD(&rdev->same_set);
3635 	init_waitqueue_head(&rdev->blocked_wait);
3636 
3637 	/* Add space to store bad block list.
3638 	 * This reserves the space even on arrays where it cannot
3639 	 * be used - I wonder if that matters
3640 	 */
3641 	return badblocks_init(&rdev->badblocks, 0);
3642 }
3643 EXPORT_SYMBOL_GPL(md_rdev_init);
3644 
3645 /*
3646  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3647  *
3648  * mark the device faulty if:
3649  *
3650  *   - the device is nonexistent (zero size)
3651  *   - the device has no valid superblock
3652  *
3653  * a faulty rdev _never_ has rdev->sb set.
3654  */
md_import_device(dev_t newdev,int super_format,int super_minor)3655 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3656 {
3657 	struct md_rdev *rdev;
3658 	sector_t size;
3659 	int err;
3660 
3661 	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3662 	if (!rdev)
3663 		return ERR_PTR(-ENOMEM);
3664 
3665 	err = md_rdev_init(rdev);
3666 	if (err)
3667 		goto out_free_rdev;
3668 	err = alloc_disk_sb(rdev);
3669 	if (err)
3670 		goto out_clear_rdev;
3671 
3672 	rdev->bdev_file = bdev_file_open_by_dev(newdev,
3673 			BLK_OPEN_READ | BLK_OPEN_WRITE,
3674 			super_format == -2 ? &claim_rdev : rdev, NULL);
3675 	if (IS_ERR(rdev->bdev_file)) {
3676 		pr_warn("md: could not open device unknown-block(%u,%u).\n",
3677 			MAJOR(newdev), MINOR(newdev));
3678 		err = PTR_ERR(rdev->bdev_file);
3679 		goto out_clear_rdev;
3680 	}
3681 	rdev->bdev = file_bdev(rdev->bdev_file);
3682 
3683 	kobject_init(&rdev->kobj, &rdev_ktype);
3684 
3685 	size = bdev_nr_bytes(rdev->bdev) >> BLOCK_SIZE_BITS;
3686 	if (!size) {
3687 		pr_warn("md: %pg has zero or unknown size, marking faulty!\n",
3688 			rdev->bdev);
3689 		err = -EINVAL;
3690 		goto out_blkdev_put;
3691 	}
3692 
3693 	if (super_format >= 0) {
3694 		err = super_types[super_format].
3695 			load_super(rdev, NULL, super_minor);
3696 		if (err == -EINVAL) {
3697 			pr_warn("md: %pg does not have a valid v%d.%d superblock, not importing!\n",
3698 				rdev->bdev,
3699 				super_format, super_minor);
3700 			goto out_blkdev_put;
3701 		}
3702 		if (err < 0) {
3703 			pr_warn("md: could not read %pg's sb, not importing!\n",
3704 				rdev->bdev);
3705 			goto out_blkdev_put;
3706 		}
3707 	}
3708 
3709 	return rdev;
3710 
3711 out_blkdev_put:
3712 	fput(rdev->bdev_file);
3713 out_clear_rdev:
3714 	md_rdev_clear(rdev);
3715 out_free_rdev:
3716 	kfree(rdev);
3717 	return ERR_PTR(err);
3718 }
3719 
3720 /*
3721  * Check a full RAID array for plausibility
3722  */
3723 
analyze_sbs(struct mddev * mddev)3724 static int analyze_sbs(struct mddev *mddev)
3725 {
3726 	int i;
3727 	struct md_rdev *rdev, *freshest, *tmp;
3728 
3729 	freshest = NULL;
3730 	rdev_for_each_safe(rdev, tmp, mddev)
3731 		switch (super_types[mddev->major_version].
3732 			load_super(rdev, freshest, mddev->minor_version)) {
3733 		case 1:
3734 			freshest = rdev;
3735 			break;
3736 		case 0:
3737 			break;
3738 		default:
3739 			pr_warn("md: fatal superblock inconsistency in %pg -- removing from array\n",
3740 				rdev->bdev);
3741 			md_kick_rdev_from_array(rdev);
3742 		}
3743 
3744 	/* Cannot find a valid fresh disk */
3745 	if (!freshest) {
3746 		pr_warn("md: cannot find a valid disk\n");
3747 		return -EINVAL;
3748 	}
3749 
3750 	super_types[mddev->major_version].
3751 		validate_super(mddev, NULL/*freshest*/, freshest);
3752 
3753 	i = 0;
3754 	rdev_for_each_safe(rdev, tmp, mddev) {
3755 		if (mddev->max_disks &&
3756 		    (rdev->desc_nr >= mddev->max_disks ||
3757 		     i > mddev->max_disks)) {
3758 			pr_warn("md: %s: %pg: only %d devices permitted\n",
3759 				mdname(mddev), rdev->bdev,
3760 				mddev->max_disks);
3761 			md_kick_rdev_from_array(rdev);
3762 			continue;
3763 		}
3764 		if (rdev != freshest) {
3765 			if (super_types[mddev->major_version].
3766 			    validate_super(mddev, freshest, rdev)) {
3767 				pr_warn("md: kicking non-fresh %pg from array!\n",
3768 					rdev->bdev);
3769 				md_kick_rdev_from_array(rdev);
3770 				continue;
3771 			}
3772 		}
3773 		if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3774 		    !test_bit(Journal, &rdev->flags)) {
3775 			rdev->raid_disk = -1;
3776 			clear_bit(In_sync, &rdev->flags);
3777 		}
3778 	}
3779 
3780 	return 0;
3781 }
3782 
3783 /* Read a fixed-point number.
3784  * Numbers in sysfs attributes should be in "standard" units where
3785  * possible, so time should be in seconds.
3786  * However we internally use a a much smaller unit such as
3787  * milliseconds or jiffies.
3788  * This function takes a decimal number with a possible fractional
3789  * component, and produces an integer which is the result of
3790  * multiplying that number by 10^'scale'.
3791  * all without any floating-point arithmetic.
3792  */
strict_strtoul_scaled(const char * cp,unsigned long * res,int scale)3793 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3794 {
3795 	unsigned long result = 0;
3796 	long decimals = -1;
3797 	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3798 		if (*cp == '.')
3799 			decimals = 0;
3800 		else if (decimals < scale) {
3801 			unsigned int value;
3802 			value = *cp - '0';
3803 			result = result * 10 + value;
3804 			if (decimals >= 0)
3805 				decimals++;
3806 		}
3807 		cp++;
3808 	}
3809 	if (*cp == '\n')
3810 		cp++;
3811 	if (*cp)
3812 		return -EINVAL;
3813 	if (decimals < 0)
3814 		decimals = 0;
3815 	*res = result * int_pow(10, scale - decimals);
3816 	return 0;
3817 }
3818 
3819 static ssize_t
safe_delay_show(struct mddev * mddev,char * page)3820 safe_delay_show(struct mddev *mddev, char *page)
3821 {
3822 	unsigned int msec = ((unsigned long)mddev->safemode_delay*1000)/HZ;
3823 
3824 	return sprintf(page, "%u.%03u\n", msec/1000, msec%1000);
3825 }
3826 static ssize_t
safe_delay_store(struct mddev * mddev,const char * cbuf,size_t len)3827 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3828 {
3829 	unsigned long msec;
3830 
3831 	if (mddev_is_clustered(mddev)) {
3832 		pr_warn("md: Safemode is disabled for clustered mode\n");
3833 		return -EINVAL;
3834 	}
3835 
3836 	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0 || msec > UINT_MAX / HZ)
3837 		return -EINVAL;
3838 	if (msec == 0)
3839 		mddev->safemode_delay = 0;
3840 	else {
3841 		unsigned long old_delay = mddev->safemode_delay;
3842 		unsigned long new_delay = (msec*HZ)/1000;
3843 
3844 		if (new_delay == 0)
3845 			new_delay = 1;
3846 		mddev->safemode_delay = new_delay;
3847 		if (new_delay < old_delay || old_delay == 0)
3848 			mod_timer(&mddev->safemode_timer, jiffies+1);
3849 	}
3850 	return len;
3851 }
3852 static struct md_sysfs_entry md_safe_delay =
3853 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3854 
3855 static ssize_t
level_show(struct mddev * mddev,char * page)3856 level_show(struct mddev *mddev, char *page)
3857 {
3858 	struct md_personality *p;
3859 	int ret;
3860 	spin_lock(&mddev->lock);
3861 	p = mddev->pers;
3862 	if (p)
3863 		ret = sprintf(page, "%s\n", p->name);
3864 	else if (mddev->clevel[0])
3865 		ret = sprintf(page, "%s\n", mddev->clevel);
3866 	else if (mddev->level != LEVEL_NONE)
3867 		ret = sprintf(page, "%d\n", mddev->level);
3868 	else
3869 		ret = 0;
3870 	spin_unlock(&mddev->lock);
3871 	return ret;
3872 }
3873 
3874 static ssize_t
level_store(struct mddev * mddev,const char * buf,size_t len)3875 level_store(struct mddev *mddev, const char *buf, size_t len)
3876 {
3877 	char clevel[16];
3878 	ssize_t rv;
3879 	size_t slen = len;
3880 	struct md_personality *pers, *oldpers;
3881 	long level;
3882 	void *priv, *oldpriv;
3883 	struct md_rdev *rdev;
3884 
3885 	if (slen == 0 || slen >= sizeof(clevel))
3886 		return -EINVAL;
3887 
3888 	rv = mddev_suspend_and_lock(mddev);
3889 	if (rv)
3890 		return rv;
3891 
3892 	if (mddev->pers == NULL) {
3893 		memcpy(mddev->clevel, buf, slen);
3894 		if (mddev->clevel[slen-1] == '\n')
3895 			slen--;
3896 		mddev->clevel[slen] = 0;
3897 		mddev->level = LEVEL_NONE;
3898 		rv = len;
3899 		goto out_unlock;
3900 	}
3901 	rv = -EROFS;
3902 	if (!md_is_rdwr(mddev))
3903 		goto out_unlock;
3904 
3905 	/* request to change the personality.  Need to ensure:
3906 	 *  - array is not engaged in resync/recovery/reshape
3907 	 *  - old personality can be suspended
3908 	 *  - new personality will access other array.
3909 	 */
3910 
3911 	rv = -EBUSY;
3912 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3913 	    mddev->reshape_position != MaxSector ||
3914 	    mddev->sysfs_active)
3915 		goto out_unlock;
3916 
3917 	rv = -EINVAL;
3918 	if (!mddev->pers->quiesce) {
3919 		pr_warn("md: %s: %s does not support online personality change\n",
3920 			mdname(mddev), mddev->pers->name);
3921 		goto out_unlock;
3922 	}
3923 
3924 	/* Now find the new personality */
3925 	memcpy(clevel, buf, slen);
3926 	if (clevel[slen-1] == '\n')
3927 		slen--;
3928 	clevel[slen] = 0;
3929 	if (kstrtol(clevel, 10, &level))
3930 		level = LEVEL_NONE;
3931 
3932 	if (request_module("md-%s", clevel) != 0)
3933 		request_module("md-level-%s", clevel);
3934 	spin_lock(&pers_lock);
3935 	pers = find_pers(level, clevel);
3936 	if (!pers || !try_module_get(pers->owner)) {
3937 		spin_unlock(&pers_lock);
3938 		pr_warn("md: personality %s not loaded\n", clevel);
3939 		rv = -EINVAL;
3940 		goto out_unlock;
3941 	}
3942 	spin_unlock(&pers_lock);
3943 
3944 	if (pers == mddev->pers) {
3945 		/* Nothing to do! */
3946 		module_put(pers->owner);
3947 		rv = len;
3948 		goto out_unlock;
3949 	}
3950 	if (!pers->takeover) {
3951 		module_put(pers->owner);
3952 		pr_warn("md: %s: %s does not support personality takeover\n",
3953 			mdname(mddev), clevel);
3954 		rv = -EINVAL;
3955 		goto out_unlock;
3956 	}
3957 
3958 	rdev_for_each(rdev, mddev)
3959 		rdev->new_raid_disk = rdev->raid_disk;
3960 
3961 	/* ->takeover must set new_* and/or delta_disks
3962 	 * if it succeeds, and may set them when it fails.
3963 	 */
3964 	priv = pers->takeover(mddev);
3965 	if (IS_ERR(priv)) {
3966 		mddev->new_level = mddev->level;
3967 		mddev->new_layout = mddev->layout;
3968 		mddev->new_chunk_sectors = mddev->chunk_sectors;
3969 		mddev->raid_disks -= mddev->delta_disks;
3970 		mddev->delta_disks = 0;
3971 		mddev->reshape_backwards = 0;
3972 		module_put(pers->owner);
3973 		pr_warn("md: %s: %s would not accept array\n",
3974 			mdname(mddev), clevel);
3975 		rv = PTR_ERR(priv);
3976 		goto out_unlock;
3977 	}
3978 
3979 	/* Looks like we have a winner */
3980 	mddev_detach(mddev);
3981 
3982 	spin_lock(&mddev->lock);
3983 	oldpers = mddev->pers;
3984 	oldpriv = mddev->private;
3985 	mddev->pers = pers;
3986 	mddev->private = priv;
3987 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3988 	mddev->level = mddev->new_level;
3989 	mddev->layout = mddev->new_layout;
3990 	mddev->chunk_sectors = mddev->new_chunk_sectors;
3991 	mddev->delta_disks = 0;
3992 	mddev->reshape_backwards = 0;
3993 	mddev->degraded = 0;
3994 	spin_unlock(&mddev->lock);
3995 
3996 	if (oldpers->sync_request == NULL &&
3997 	    mddev->external) {
3998 		/* We are converting from a no-redundancy array
3999 		 * to a redundancy array and metadata is managed
4000 		 * externally so we need to be sure that writes
4001 		 * won't block due to a need to transition
4002 		 *      clean->dirty
4003 		 * until external management is started.
4004 		 */
4005 		mddev->in_sync = 0;
4006 		mddev->safemode_delay = 0;
4007 		mddev->safemode = 0;
4008 	}
4009 
4010 	oldpers->free(mddev, oldpriv);
4011 
4012 	if (oldpers->sync_request == NULL &&
4013 	    pers->sync_request != NULL) {
4014 		/* need to add the md_redundancy_group */
4015 		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4016 			pr_warn("md: cannot register extra attributes for %s\n",
4017 				mdname(mddev));
4018 		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4019 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
4020 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
4021 	}
4022 	if (oldpers->sync_request != NULL &&
4023 	    pers->sync_request == NULL) {
4024 		/* need to remove the md_redundancy_group */
4025 		if (mddev->to_remove == NULL)
4026 			mddev->to_remove = &md_redundancy_group;
4027 	}
4028 
4029 	module_put(oldpers->owner);
4030 
4031 	rdev_for_each(rdev, mddev) {
4032 		if (rdev->raid_disk < 0)
4033 			continue;
4034 		if (rdev->new_raid_disk >= mddev->raid_disks)
4035 			rdev->new_raid_disk = -1;
4036 		if (rdev->new_raid_disk == rdev->raid_disk)
4037 			continue;
4038 		sysfs_unlink_rdev(mddev, rdev);
4039 	}
4040 	rdev_for_each(rdev, mddev) {
4041 		if (rdev->raid_disk < 0)
4042 			continue;
4043 		if (rdev->new_raid_disk == rdev->raid_disk)
4044 			continue;
4045 		rdev->raid_disk = rdev->new_raid_disk;
4046 		if (rdev->raid_disk < 0)
4047 			clear_bit(In_sync, &rdev->flags);
4048 		else {
4049 			if (sysfs_link_rdev(mddev, rdev))
4050 				pr_warn("md: cannot register rd%d for %s after level change\n",
4051 					rdev->raid_disk, mdname(mddev));
4052 		}
4053 	}
4054 
4055 	if (pers->sync_request == NULL) {
4056 		/* this is now an array without redundancy, so
4057 		 * it must always be in_sync
4058 		 */
4059 		mddev->in_sync = 1;
4060 		del_timer_sync(&mddev->safemode_timer);
4061 	}
4062 	pers->run(mddev);
4063 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
4064 	if (!mddev->thread)
4065 		md_update_sb(mddev, 1);
4066 	sysfs_notify_dirent_safe(mddev->sysfs_level);
4067 	md_new_event();
4068 	rv = len;
4069 out_unlock:
4070 	mddev_unlock_and_resume(mddev);
4071 	return rv;
4072 }
4073 
4074 static struct md_sysfs_entry md_level =
4075 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
4076 
4077 static ssize_t
new_level_show(struct mddev * mddev,char * page)4078 new_level_show(struct mddev *mddev, char *page)
4079 {
4080 	return sprintf(page, "%d\n", mddev->new_level);
4081 }
4082 
4083 static ssize_t
new_level_store(struct mddev * mddev,const char * buf,size_t len)4084 new_level_store(struct mddev *mddev, const char *buf, size_t len)
4085 {
4086 	unsigned int n;
4087 	int err;
4088 
4089 	err = kstrtouint(buf, 10, &n);
4090 	if (err < 0)
4091 		return err;
4092 	err = mddev_lock(mddev);
4093 	if (err)
4094 		return err;
4095 
4096 	mddev->new_level = n;
4097 	md_update_sb(mddev, 1);
4098 
4099 	mddev_unlock(mddev);
4100 	return len;
4101 }
4102 static struct md_sysfs_entry md_new_level =
4103 __ATTR(new_level, 0664, new_level_show, new_level_store);
4104 
4105 static ssize_t
layout_show(struct mddev * mddev,char * page)4106 layout_show(struct mddev *mddev, char *page)
4107 {
4108 	/* just a number, not meaningful for all levels */
4109 	if (mddev->reshape_position != MaxSector &&
4110 	    mddev->layout != mddev->new_layout)
4111 		return sprintf(page, "%d (%d)\n",
4112 			       mddev->new_layout, mddev->layout);
4113 	return sprintf(page, "%d\n", mddev->layout);
4114 }
4115 
4116 static ssize_t
layout_store(struct mddev * mddev,const char * buf,size_t len)4117 layout_store(struct mddev *mddev, const char *buf, size_t len)
4118 {
4119 	unsigned int n;
4120 	int err;
4121 
4122 	err = kstrtouint(buf, 10, &n);
4123 	if (err < 0)
4124 		return err;
4125 	err = mddev_lock(mddev);
4126 	if (err)
4127 		return err;
4128 
4129 	if (mddev->pers) {
4130 		if (mddev->pers->check_reshape == NULL)
4131 			err = -EBUSY;
4132 		else if (!md_is_rdwr(mddev))
4133 			err = -EROFS;
4134 		else {
4135 			mddev->new_layout = n;
4136 			err = mddev->pers->check_reshape(mddev);
4137 			if (err)
4138 				mddev->new_layout = mddev->layout;
4139 		}
4140 	} else {
4141 		mddev->new_layout = n;
4142 		if (mddev->reshape_position == MaxSector)
4143 			mddev->layout = n;
4144 	}
4145 	mddev_unlock(mddev);
4146 	return err ?: len;
4147 }
4148 static struct md_sysfs_entry md_layout =
4149 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4150 
4151 static ssize_t
raid_disks_show(struct mddev * mddev,char * page)4152 raid_disks_show(struct mddev *mddev, char *page)
4153 {
4154 	if (mddev->raid_disks == 0)
4155 		return 0;
4156 	if (mddev->reshape_position != MaxSector &&
4157 	    mddev->delta_disks != 0)
4158 		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4159 			       mddev->raid_disks - mddev->delta_disks);
4160 	return sprintf(page, "%d\n", mddev->raid_disks);
4161 }
4162 
4163 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4164 
4165 static ssize_t
raid_disks_store(struct mddev * mddev,const char * buf,size_t len)4166 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4167 {
4168 	unsigned int n;
4169 	int err;
4170 
4171 	err = kstrtouint(buf, 10, &n);
4172 	if (err < 0)
4173 		return err;
4174 
4175 	err = mddev_lock(mddev);
4176 	if (err)
4177 		return err;
4178 	if (mddev->pers)
4179 		err = update_raid_disks(mddev, n);
4180 	else if (mddev->reshape_position != MaxSector) {
4181 		struct md_rdev *rdev;
4182 		int olddisks = mddev->raid_disks - mddev->delta_disks;
4183 
4184 		err = -EINVAL;
4185 		rdev_for_each(rdev, mddev) {
4186 			if (olddisks < n &&
4187 			    rdev->data_offset < rdev->new_data_offset)
4188 				goto out_unlock;
4189 			if (olddisks > n &&
4190 			    rdev->data_offset > rdev->new_data_offset)
4191 				goto out_unlock;
4192 		}
4193 		err = 0;
4194 		mddev->delta_disks = n - olddisks;
4195 		mddev->raid_disks = n;
4196 		mddev->reshape_backwards = (mddev->delta_disks < 0);
4197 	} else
4198 		mddev->raid_disks = n;
4199 out_unlock:
4200 	mddev_unlock(mddev);
4201 	return err ? err : len;
4202 }
4203 static struct md_sysfs_entry md_raid_disks =
4204 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4205 
4206 static ssize_t
uuid_show(struct mddev * mddev,char * page)4207 uuid_show(struct mddev *mddev, char *page)
4208 {
4209 	return sprintf(page, "%pU\n", mddev->uuid);
4210 }
4211 static struct md_sysfs_entry md_uuid =
4212 __ATTR(uuid, S_IRUGO, uuid_show, NULL);
4213 
4214 static ssize_t
chunk_size_show(struct mddev * mddev,char * page)4215 chunk_size_show(struct mddev *mddev, char *page)
4216 {
4217 	if (mddev->reshape_position != MaxSector &&
4218 	    mddev->chunk_sectors != mddev->new_chunk_sectors)
4219 		return sprintf(page, "%d (%d)\n",
4220 			       mddev->new_chunk_sectors << 9,
4221 			       mddev->chunk_sectors << 9);
4222 	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4223 }
4224 
4225 static ssize_t
chunk_size_store(struct mddev * mddev,const char * buf,size_t len)4226 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4227 {
4228 	unsigned long n;
4229 	int err;
4230 
4231 	err = kstrtoul(buf, 10, &n);
4232 	if (err < 0)
4233 		return err;
4234 
4235 	err = mddev_lock(mddev);
4236 	if (err)
4237 		return err;
4238 	if (mddev->pers) {
4239 		if (mddev->pers->check_reshape == NULL)
4240 			err = -EBUSY;
4241 		else if (!md_is_rdwr(mddev))
4242 			err = -EROFS;
4243 		else {
4244 			mddev->new_chunk_sectors = n >> 9;
4245 			err = mddev->pers->check_reshape(mddev);
4246 			if (err)
4247 				mddev->new_chunk_sectors = mddev->chunk_sectors;
4248 		}
4249 	} else {
4250 		mddev->new_chunk_sectors = n >> 9;
4251 		if (mddev->reshape_position == MaxSector)
4252 			mddev->chunk_sectors = n >> 9;
4253 	}
4254 	mddev_unlock(mddev);
4255 	return err ?: len;
4256 }
4257 static struct md_sysfs_entry md_chunk_size =
4258 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4259 
4260 static ssize_t
resync_start_show(struct mddev * mddev,char * page)4261 resync_start_show(struct mddev *mddev, char *page)
4262 {
4263 	if (mddev->recovery_cp == MaxSector)
4264 		return sprintf(page, "none\n");
4265 	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4266 }
4267 
4268 static ssize_t
resync_start_store(struct mddev * mddev,const char * buf,size_t len)4269 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4270 {
4271 	unsigned long long n;
4272 	int err;
4273 
4274 	if (cmd_match(buf, "none"))
4275 		n = MaxSector;
4276 	else {
4277 		err = kstrtoull(buf, 10, &n);
4278 		if (err < 0)
4279 			return err;
4280 		if (n != (sector_t)n)
4281 			return -EINVAL;
4282 	}
4283 
4284 	err = mddev_lock(mddev);
4285 	if (err)
4286 		return err;
4287 	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4288 		err = -EBUSY;
4289 
4290 	if (!err) {
4291 		mddev->recovery_cp = n;
4292 		if (mddev->pers)
4293 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4294 	}
4295 	mddev_unlock(mddev);
4296 	return err ?: len;
4297 }
4298 static struct md_sysfs_entry md_resync_start =
4299 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4300 		resync_start_show, resync_start_store);
4301 
4302 /*
4303  * The array state can be:
4304  *
4305  * clear
4306  *     No devices, no size, no level
4307  *     Equivalent to STOP_ARRAY ioctl
4308  * inactive
4309  *     May have some settings, but array is not active
4310  *        all IO results in error
4311  *     When written, doesn't tear down array, but just stops it
4312  * suspended (not supported yet)
4313  *     All IO requests will block. The array can be reconfigured.
4314  *     Writing this, if accepted, will block until array is quiescent
4315  * readonly
4316  *     no resync can happen.  no superblocks get written.
4317  *     write requests fail
4318  * read-auto
4319  *     like readonly, but behaves like 'clean' on a write request.
4320  *
4321  * clean - no pending writes, but otherwise active.
4322  *     When written to inactive array, starts without resync
4323  *     If a write request arrives then
4324  *       if metadata is known, mark 'dirty' and switch to 'active'.
4325  *       if not known, block and switch to write-pending
4326  *     If written to an active array that has pending writes, then fails.
4327  * active
4328  *     fully active: IO and resync can be happening.
4329  *     When written to inactive array, starts with resync
4330  *
4331  * write-pending
4332  *     clean, but writes are blocked waiting for 'active' to be written.
4333  *
4334  * active-idle
4335  *     like active, but no writes have been seen for a while (100msec).
4336  *
4337  * broken
4338 *     Array is failed. It's useful because mounted-arrays aren't stopped
4339 *     when array is failed, so this state will at least alert the user that
4340 *     something is wrong.
4341  */
4342 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4343 		   write_pending, active_idle, broken, bad_word};
4344 static char *array_states[] = {
4345 	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4346 	"write-pending", "active-idle", "broken", NULL };
4347 
match_word(const char * word,char ** list)4348 static int match_word(const char *word, char **list)
4349 {
4350 	int n;
4351 	for (n=0; list[n]; n++)
4352 		if (cmd_match(word, list[n]))
4353 			break;
4354 	return n;
4355 }
4356 
4357 static ssize_t
array_state_show(struct mddev * mddev,char * page)4358 array_state_show(struct mddev *mddev, char *page)
4359 {
4360 	enum array_state st = inactive;
4361 
4362 	if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4363 		switch(mddev->ro) {
4364 		case MD_RDONLY:
4365 			st = readonly;
4366 			break;
4367 		case MD_AUTO_READ:
4368 			st = read_auto;
4369 			break;
4370 		case MD_RDWR:
4371 			spin_lock(&mddev->lock);
4372 			if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4373 				st = write_pending;
4374 			else if (mddev->in_sync)
4375 				st = clean;
4376 			else if (mddev->safemode)
4377 				st = active_idle;
4378 			else
4379 				st = active;
4380 			spin_unlock(&mddev->lock);
4381 		}
4382 
4383 		if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4384 			st = broken;
4385 	} else {
4386 		if (list_empty(&mddev->disks) &&
4387 		    mddev->raid_disks == 0 &&
4388 		    mddev->dev_sectors == 0)
4389 			st = clear;
4390 		else
4391 			st = inactive;
4392 	}
4393 	return sprintf(page, "%s\n", array_states[st]);
4394 }
4395 
4396 static int do_md_stop(struct mddev *mddev, int ro);
4397 static int md_set_readonly(struct mddev *mddev);
4398 static int restart_array(struct mddev *mddev);
4399 
4400 static ssize_t
array_state_store(struct mddev * mddev,const char * buf,size_t len)4401 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4402 {
4403 	int err = 0;
4404 	enum array_state st = match_word(buf, array_states);
4405 
4406 	/* No lock dependent actions */
4407 	switch (st) {
4408 	case suspended:		/* not supported yet */
4409 	case write_pending:	/* cannot be set */
4410 	case active_idle:	/* cannot be set */
4411 	case broken:		/* cannot be set */
4412 	case bad_word:
4413 		return -EINVAL;
4414 	case clear:
4415 	case readonly:
4416 	case inactive:
4417 	case read_auto:
4418 		if (!mddev->pers || !md_is_rdwr(mddev))
4419 			break;
4420 		/* write sysfs will not open mddev and opener should be 0 */
4421 		err = mddev_set_closing_and_sync_blockdev(mddev, 0);
4422 		if (err)
4423 			return err;
4424 		break;
4425 	default:
4426 		break;
4427 	}
4428 
4429 	if (mddev->pers && (st == active || st == clean) &&
4430 	    mddev->ro != MD_RDONLY) {
4431 		/* don't take reconfig_mutex when toggling between
4432 		 * clean and active
4433 		 */
4434 		spin_lock(&mddev->lock);
4435 		if (st == active) {
4436 			restart_array(mddev);
4437 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4438 			md_wakeup_thread(mddev->thread);
4439 			wake_up(&mddev->sb_wait);
4440 		} else /* st == clean */ {
4441 			restart_array(mddev);
4442 			if (!set_in_sync(mddev))
4443 				err = -EBUSY;
4444 		}
4445 		if (!err)
4446 			sysfs_notify_dirent_safe(mddev->sysfs_state);
4447 		spin_unlock(&mddev->lock);
4448 		return err ?: len;
4449 	}
4450 	err = mddev_lock(mddev);
4451 	if (err)
4452 		return err;
4453 
4454 	switch (st) {
4455 	case inactive:
4456 		/* stop an active array, return 0 otherwise */
4457 		if (mddev->pers)
4458 			err = do_md_stop(mddev, 2);
4459 		break;
4460 	case clear:
4461 		err = do_md_stop(mddev, 0);
4462 		break;
4463 	case readonly:
4464 		if (mddev->pers)
4465 			err = md_set_readonly(mddev);
4466 		else {
4467 			mddev->ro = MD_RDONLY;
4468 			set_disk_ro(mddev->gendisk, 1);
4469 			err = do_md_run(mddev);
4470 		}
4471 		break;
4472 	case read_auto:
4473 		if (mddev->pers) {
4474 			if (md_is_rdwr(mddev))
4475 				err = md_set_readonly(mddev);
4476 			else if (mddev->ro == MD_RDONLY)
4477 				err = restart_array(mddev);
4478 			if (err == 0) {
4479 				mddev->ro = MD_AUTO_READ;
4480 				set_disk_ro(mddev->gendisk, 0);
4481 			}
4482 		} else {
4483 			mddev->ro = MD_AUTO_READ;
4484 			err = do_md_run(mddev);
4485 		}
4486 		break;
4487 	case clean:
4488 		if (mddev->pers) {
4489 			err = restart_array(mddev);
4490 			if (err)
4491 				break;
4492 			spin_lock(&mddev->lock);
4493 			if (!set_in_sync(mddev))
4494 				err = -EBUSY;
4495 			spin_unlock(&mddev->lock);
4496 		} else
4497 			err = -EINVAL;
4498 		break;
4499 	case active:
4500 		if (mddev->pers) {
4501 			err = restart_array(mddev);
4502 			if (err)
4503 				break;
4504 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4505 			wake_up(&mddev->sb_wait);
4506 			err = 0;
4507 		} else {
4508 			mddev->ro = MD_RDWR;
4509 			set_disk_ro(mddev->gendisk, 0);
4510 			err = do_md_run(mddev);
4511 		}
4512 		break;
4513 	default:
4514 		err = -EINVAL;
4515 		break;
4516 	}
4517 
4518 	if (!err) {
4519 		if (mddev->hold_active == UNTIL_IOCTL)
4520 			mddev->hold_active = 0;
4521 		sysfs_notify_dirent_safe(mddev->sysfs_state);
4522 	}
4523 	mddev_unlock(mddev);
4524 
4525 	if (st == readonly || st == read_auto || st == inactive ||
4526 	    (err && st == clear))
4527 		clear_bit(MD_CLOSING, &mddev->flags);
4528 
4529 	return err ?: len;
4530 }
4531 static struct md_sysfs_entry md_array_state =
4532 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4533 
4534 static ssize_t
max_corrected_read_errors_show(struct mddev * mddev,char * page)4535 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4536 	return sprintf(page, "%d\n",
4537 		       atomic_read(&mddev->max_corr_read_errors));
4538 }
4539 
4540 static ssize_t
max_corrected_read_errors_store(struct mddev * mddev,const char * buf,size_t len)4541 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4542 {
4543 	unsigned int n;
4544 	int rv;
4545 
4546 	rv = kstrtouint(buf, 10, &n);
4547 	if (rv < 0)
4548 		return rv;
4549 	if (n > INT_MAX)
4550 		return -EINVAL;
4551 	atomic_set(&mddev->max_corr_read_errors, n);
4552 	return len;
4553 }
4554 
4555 static struct md_sysfs_entry max_corr_read_errors =
4556 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4557 	max_corrected_read_errors_store);
4558 
4559 static ssize_t
null_show(struct mddev * mddev,char * page)4560 null_show(struct mddev *mddev, char *page)
4561 {
4562 	return -EINVAL;
4563 }
4564 
4565 static ssize_t
new_dev_store(struct mddev * mddev,const char * buf,size_t len)4566 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4567 {
4568 	/* buf must be %d:%d\n? giving major and minor numbers */
4569 	/* The new device is added to the array.
4570 	 * If the array has a persistent superblock, we read the
4571 	 * superblock to initialise info and check validity.
4572 	 * Otherwise, only checking done is that in bind_rdev_to_array,
4573 	 * which mainly checks size.
4574 	 */
4575 	char *e;
4576 	int major = simple_strtoul(buf, &e, 10);
4577 	int minor;
4578 	dev_t dev;
4579 	struct md_rdev *rdev;
4580 	int err;
4581 
4582 	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4583 		return -EINVAL;
4584 	minor = simple_strtoul(e+1, &e, 10);
4585 	if (*e && *e != '\n')
4586 		return -EINVAL;
4587 	dev = MKDEV(major, minor);
4588 	if (major != MAJOR(dev) ||
4589 	    minor != MINOR(dev))
4590 		return -EOVERFLOW;
4591 
4592 	err = mddev_suspend_and_lock(mddev);
4593 	if (err)
4594 		return err;
4595 	if (mddev->persistent) {
4596 		rdev = md_import_device(dev, mddev->major_version,
4597 					mddev->minor_version);
4598 		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4599 			struct md_rdev *rdev0
4600 				= list_entry(mddev->disks.next,
4601 					     struct md_rdev, same_set);
4602 			err = super_types[mddev->major_version]
4603 				.load_super(rdev, rdev0, mddev->minor_version);
4604 			if (err < 0)
4605 				goto out;
4606 		}
4607 	} else if (mddev->external)
4608 		rdev = md_import_device(dev, -2, -1);
4609 	else
4610 		rdev = md_import_device(dev, -1, -1);
4611 
4612 	if (IS_ERR(rdev)) {
4613 		mddev_unlock_and_resume(mddev);
4614 		return PTR_ERR(rdev);
4615 	}
4616 	err = bind_rdev_to_array(rdev, mddev);
4617  out:
4618 	if (err)
4619 		export_rdev(rdev, mddev);
4620 	mddev_unlock_and_resume(mddev);
4621 	if (!err)
4622 		md_new_event();
4623 	return err ? err : len;
4624 }
4625 
4626 static struct md_sysfs_entry md_new_device =
4627 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4628 
4629 static ssize_t
bitmap_store(struct mddev * mddev,const char * buf,size_t len)4630 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4631 {
4632 	char *end;
4633 	unsigned long chunk, end_chunk;
4634 	int err;
4635 
4636 	err = mddev_lock(mddev);
4637 	if (err)
4638 		return err;
4639 	if (!mddev->bitmap)
4640 		goto out;
4641 	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4642 	while (*buf) {
4643 		chunk = end_chunk = simple_strtoul(buf, &end, 0);
4644 		if (buf == end)
4645 			break;
4646 
4647 		if (*end == '-') { /* range */
4648 			buf = end + 1;
4649 			end_chunk = simple_strtoul(buf, &end, 0);
4650 			if (buf == end)
4651 				break;
4652 		}
4653 
4654 		if (*end && !isspace(*end))
4655 			break;
4656 
4657 		mddev->bitmap_ops->dirty_bits(mddev, chunk, end_chunk);
4658 		buf = skip_spaces(end);
4659 	}
4660 	mddev->bitmap_ops->unplug(mddev, true); /* flush the bits to disk */
4661 out:
4662 	mddev_unlock(mddev);
4663 	return len;
4664 }
4665 
4666 static struct md_sysfs_entry md_bitmap =
4667 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4668 
4669 static ssize_t
size_show(struct mddev * mddev,char * page)4670 size_show(struct mddev *mddev, char *page)
4671 {
4672 	return sprintf(page, "%llu\n",
4673 		(unsigned long long)mddev->dev_sectors / 2);
4674 }
4675 
4676 static int update_size(struct mddev *mddev, sector_t num_sectors);
4677 
4678 static ssize_t
size_store(struct mddev * mddev,const char * buf,size_t len)4679 size_store(struct mddev *mddev, const char *buf, size_t len)
4680 {
4681 	/* If array is inactive, we can reduce the component size, but
4682 	 * not increase it (except from 0).
4683 	 * If array is active, we can try an on-line resize
4684 	 */
4685 	sector_t sectors;
4686 	int err = strict_blocks_to_sectors(buf, &sectors);
4687 
4688 	if (err < 0)
4689 		return err;
4690 	err = mddev_lock(mddev);
4691 	if (err)
4692 		return err;
4693 	if (mddev->pers) {
4694 		err = update_size(mddev, sectors);
4695 		if (err == 0)
4696 			md_update_sb(mddev, 1);
4697 	} else {
4698 		if (mddev->dev_sectors == 0 ||
4699 		    mddev->dev_sectors > sectors)
4700 			mddev->dev_sectors = sectors;
4701 		else
4702 			err = -ENOSPC;
4703 	}
4704 	mddev_unlock(mddev);
4705 	return err ? err : len;
4706 }
4707 
4708 static struct md_sysfs_entry md_size =
4709 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4710 
4711 /* Metadata version.
4712  * This is one of
4713  *   'none' for arrays with no metadata (good luck...)
4714  *   'external' for arrays with externally managed metadata,
4715  * or N.M for internally known formats
4716  */
4717 static ssize_t
metadata_show(struct mddev * mddev,char * page)4718 metadata_show(struct mddev *mddev, char *page)
4719 {
4720 	if (mddev->persistent)
4721 		return sprintf(page, "%d.%d\n",
4722 			       mddev->major_version, mddev->minor_version);
4723 	else if (mddev->external)
4724 		return sprintf(page, "external:%s\n", mddev->metadata_type);
4725 	else
4726 		return sprintf(page, "none\n");
4727 }
4728 
4729 static ssize_t
metadata_store(struct mddev * mddev,const char * buf,size_t len)4730 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4731 {
4732 	int major, minor;
4733 	char *e;
4734 	int err;
4735 	/* Changing the details of 'external' metadata is
4736 	 * always permitted.  Otherwise there must be
4737 	 * no devices attached to the array.
4738 	 */
4739 
4740 	err = mddev_lock(mddev);
4741 	if (err)
4742 		return err;
4743 	err = -EBUSY;
4744 	if (mddev->external && strncmp(buf, "external:", 9) == 0)
4745 		;
4746 	else if (!list_empty(&mddev->disks))
4747 		goto out_unlock;
4748 
4749 	err = 0;
4750 	if (cmd_match(buf, "none")) {
4751 		mddev->persistent = 0;
4752 		mddev->external = 0;
4753 		mddev->major_version = 0;
4754 		mddev->minor_version = 90;
4755 		goto out_unlock;
4756 	}
4757 	if (strncmp(buf, "external:", 9) == 0) {
4758 		size_t namelen = len-9;
4759 		if (namelen >= sizeof(mddev->metadata_type))
4760 			namelen = sizeof(mddev->metadata_type)-1;
4761 		memcpy(mddev->metadata_type, buf+9, namelen);
4762 		mddev->metadata_type[namelen] = 0;
4763 		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4764 			mddev->metadata_type[--namelen] = 0;
4765 		mddev->persistent = 0;
4766 		mddev->external = 1;
4767 		mddev->major_version = 0;
4768 		mddev->minor_version = 90;
4769 		goto out_unlock;
4770 	}
4771 	major = simple_strtoul(buf, &e, 10);
4772 	err = -EINVAL;
4773 	if (e==buf || *e != '.')
4774 		goto out_unlock;
4775 	buf = e+1;
4776 	minor = simple_strtoul(buf, &e, 10);
4777 	if (e==buf || (*e && *e != '\n') )
4778 		goto out_unlock;
4779 	err = -ENOENT;
4780 	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4781 		goto out_unlock;
4782 	mddev->major_version = major;
4783 	mddev->minor_version = minor;
4784 	mddev->persistent = 1;
4785 	mddev->external = 0;
4786 	err = 0;
4787 out_unlock:
4788 	mddev_unlock(mddev);
4789 	return err ?: len;
4790 }
4791 
4792 static struct md_sysfs_entry md_metadata =
4793 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4794 
md_sync_action(struct mddev * mddev)4795 enum sync_action md_sync_action(struct mddev *mddev)
4796 {
4797 	unsigned long recovery = mddev->recovery;
4798 
4799 	/*
4800 	 * frozen has the highest priority, means running sync_thread will be
4801 	 * stopped immediately, and no new sync_thread can start.
4802 	 */
4803 	if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4804 		return ACTION_FROZEN;
4805 
4806 	/*
4807 	 * read-only array can't register sync_thread, and it can only
4808 	 * add/remove spares.
4809 	 */
4810 	if (!md_is_rdwr(mddev))
4811 		return ACTION_IDLE;
4812 
4813 	/*
4814 	 * idle means no sync_thread is running, and no new sync_thread is
4815 	 * requested.
4816 	 */
4817 	if (!test_bit(MD_RECOVERY_RUNNING, &recovery) &&
4818 	    !test_bit(MD_RECOVERY_NEEDED, &recovery))
4819 		return ACTION_IDLE;
4820 
4821 	if (test_bit(MD_RECOVERY_RESHAPE, &recovery) ||
4822 	    mddev->reshape_position != MaxSector)
4823 		return ACTION_RESHAPE;
4824 
4825 	if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4826 		return ACTION_RECOVER;
4827 
4828 	if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4829 		/*
4830 		 * MD_RECOVERY_CHECK must be paired with
4831 		 * MD_RECOVERY_REQUESTED.
4832 		 */
4833 		if (test_bit(MD_RECOVERY_CHECK, &recovery))
4834 			return ACTION_CHECK;
4835 		if (test_bit(MD_RECOVERY_REQUESTED, &recovery))
4836 			return ACTION_REPAIR;
4837 		return ACTION_RESYNC;
4838 	}
4839 
4840 	/*
4841 	 * MD_RECOVERY_NEEDED or MD_RECOVERY_RUNNING is set, however, no
4842 	 * sync_action is specified.
4843 	 */
4844 	return ACTION_IDLE;
4845 }
4846 
md_sync_action_by_name(const char * page)4847 enum sync_action md_sync_action_by_name(const char *page)
4848 {
4849 	enum sync_action action;
4850 
4851 	for (action = 0; action < NR_SYNC_ACTIONS; ++action) {
4852 		if (cmd_match(page, action_name[action]))
4853 			return action;
4854 	}
4855 
4856 	return NR_SYNC_ACTIONS;
4857 }
4858 
md_sync_action_name(enum sync_action action)4859 const char *md_sync_action_name(enum sync_action action)
4860 {
4861 	return action_name[action];
4862 }
4863 
4864 static ssize_t
action_show(struct mddev * mddev,char * page)4865 action_show(struct mddev *mddev, char *page)
4866 {
4867 	enum sync_action action = md_sync_action(mddev);
4868 
4869 	return sprintf(page, "%s\n", md_sync_action_name(action));
4870 }
4871 
4872 /**
4873  * stop_sync_thread() - wait for sync_thread to stop if it's running.
4874  * @mddev:	the array.
4875  * @locked:	if set, reconfig_mutex will still be held after this function
4876  *		return; if not set, reconfig_mutex will be released after this
4877  *		function return.
4878  */
stop_sync_thread(struct mddev * mddev,bool locked)4879 static void stop_sync_thread(struct mddev *mddev, bool locked)
4880 {
4881 	int sync_seq = atomic_read(&mddev->sync_seq);
4882 
4883 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4884 		if (!locked)
4885 			mddev_unlock(mddev);
4886 		return;
4887 	}
4888 
4889 	mddev_unlock(mddev);
4890 
4891 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4892 	/*
4893 	 * Thread might be blocked waiting for metadata update which will now
4894 	 * never happen
4895 	 */
4896 	md_wakeup_thread_directly(mddev->sync_thread);
4897 	if (work_pending(&mddev->sync_work))
4898 		flush_work(&mddev->sync_work);
4899 
4900 	wait_event(resync_wait,
4901 		   !test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4902 		   (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery) &&
4903 		    sync_seq != atomic_read(&mddev->sync_seq)));
4904 
4905 	if (locked)
4906 		mddev_lock_nointr(mddev);
4907 }
4908 
md_idle_sync_thread(struct mddev * mddev)4909 void md_idle_sync_thread(struct mddev *mddev)
4910 {
4911 	lockdep_assert_held(&mddev->reconfig_mutex);
4912 
4913 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4914 	stop_sync_thread(mddev, true);
4915 }
4916 EXPORT_SYMBOL_GPL(md_idle_sync_thread);
4917 
md_frozen_sync_thread(struct mddev * mddev)4918 void md_frozen_sync_thread(struct mddev *mddev)
4919 {
4920 	lockdep_assert_held(&mddev->reconfig_mutex);
4921 
4922 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4923 	stop_sync_thread(mddev, true);
4924 }
4925 EXPORT_SYMBOL_GPL(md_frozen_sync_thread);
4926 
md_unfrozen_sync_thread(struct mddev * mddev)4927 void md_unfrozen_sync_thread(struct mddev *mddev)
4928 {
4929 	lockdep_assert_held(&mddev->reconfig_mutex);
4930 
4931 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4932 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4933 	md_wakeup_thread(mddev->thread);
4934 	sysfs_notify_dirent_safe(mddev->sysfs_action);
4935 }
4936 EXPORT_SYMBOL_GPL(md_unfrozen_sync_thread);
4937 
mddev_start_reshape(struct mddev * mddev)4938 static int mddev_start_reshape(struct mddev *mddev)
4939 {
4940 	int ret;
4941 
4942 	if (mddev->pers->start_reshape == NULL)
4943 		return -EINVAL;
4944 
4945 	if (mddev->reshape_position == MaxSector ||
4946 	    mddev->pers->check_reshape == NULL ||
4947 	    mddev->pers->check_reshape(mddev)) {
4948 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4949 		ret = mddev->pers->start_reshape(mddev);
4950 		if (ret)
4951 			return ret;
4952 	} else {
4953 		/*
4954 		 * If reshape is still in progress, and md_check_recovery() can
4955 		 * continue to reshape, don't restart reshape because data can
4956 		 * be corrupted for raid456.
4957 		 */
4958 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4959 	}
4960 
4961 	sysfs_notify_dirent_safe(mddev->sysfs_degraded);
4962 	return 0;
4963 }
4964 
4965 static ssize_t
action_store(struct mddev * mddev,const char * page,size_t len)4966 action_store(struct mddev *mddev, const char *page, size_t len)
4967 {
4968 	int ret;
4969 	enum sync_action action;
4970 
4971 	if (!mddev->pers || !mddev->pers->sync_request)
4972 		return -EINVAL;
4973 
4974 retry:
4975 	if (work_busy(&mddev->sync_work))
4976 		flush_work(&mddev->sync_work);
4977 
4978 	ret = mddev_lock(mddev);
4979 	if (ret)
4980 		return ret;
4981 
4982 	if (work_busy(&mddev->sync_work)) {
4983 		mddev_unlock(mddev);
4984 		goto retry;
4985 	}
4986 
4987 	action = md_sync_action_by_name(page);
4988 
4989 	/* TODO: mdadm rely on "idle" to start sync_thread. */
4990 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
4991 		switch (action) {
4992 		case ACTION_FROZEN:
4993 			md_frozen_sync_thread(mddev);
4994 			ret = len;
4995 			goto out;
4996 		case ACTION_IDLE:
4997 			md_idle_sync_thread(mddev);
4998 			break;
4999 		case ACTION_RESHAPE:
5000 		case ACTION_RECOVER:
5001 		case ACTION_CHECK:
5002 		case ACTION_REPAIR:
5003 		case ACTION_RESYNC:
5004 			ret = -EBUSY;
5005 			goto out;
5006 		default:
5007 			ret = -EINVAL;
5008 			goto out;
5009 		}
5010 	} else {
5011 		switch (action) {
5012 		case ACTION_FROZEN:
5013 			set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5014 			ret = len;
5015 			goto out;
5016 		case ACTION_RESHAPE:
5017 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5018 			ret = mddev_start_reshape(mddev);
5019 			if (ret)
5020 				goto out;
5021 			break;
5022 		case ACTION_RECOVER:
5023 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5024 			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5025 			break;
5026 		case ACTION_CHECK:
5027 			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5028 			fallthrough;
5029 		case ACTION_REPAIR:
5030 			set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
5031 			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5032 			fallthrough;
5033 		case ACTION_RESYNC:
5034 		case ACTION_IDLE:
5035 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5036 			break;
5037 		default:
5038 			ret = -EINVAL;
5039 			goto out;
5040 		}
5041 	}
5042 
5043 	if (mddev->ro == MD_AUTO_READ) {
5044 		/* A write to sync_action is enough to justify
5045 		 * canceling read-auto mode
5046 		 */
5047 		mddev->ro = MD_RDWR;
5048 		md_wakeup_thread(mddev->sync_thread);
5049 	}
5050 
5051 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5052 	md_wakeup_thread(mddev->thread);
5053 	sysfs_notify_dirent_safe(mddev->sysfs_action);
5054 	ret = len;
5055 
5056 out:
5057 	mddev_unlock(mddev);
5058 	return ret;
5059 }
5060 
5061 static struct md_sysfs_entry md_scan_mode =
5062 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
5063 
5064 static ssize_t
last_sync_action_show(struct mddev * mddev,char * page)5065 last_sync_action_show(struct mddev *mddev, char *page)
5066 {
5067 	return sprintf(page, "%s\n",
5068 		       md_sync_action_name(mddev->last_sync_action));
5069 }
5070 
5071 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
5072 
5073 static ssize_t
mismatch_cnt_show(struct mddev * mddev,char * page)5074 mismatch_cnt_show(struct mddev *mddev, char *page)
5075 {
5076 	return sprintf(page, "%llu\n",
5077 		       (unsigned long long)
5078 		       atomic64_read(&mddev->resync_mismatches));
5079 }
5080 
5081 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
5082 
5083 static ssize_t
sync_min_show(struct mddev * mddev,char * page)5084 sync_min_show(struct mddev *mddev, char *page)
5085 {
5086 	return sprintf(page, "%d (%s)\n", speed_min(mddev),
5087 		       mddev->sync_speed_min ? "local": "system");
5088 }
5089 
5090 static ssize_t
sync_min_store(struct mddev * mddev,const char * buf,size_t len)5091 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
5092 {
5093 	unsigned int min;
5094 	int rv;
5095 
5096 	if (strncmp(buf, "system", 6)==0) {
5097 		min = 0;
5098 	} else {
5099 		rv = kstrtouint(buf, 10, &min);
5100 		if (rv < 0)
5101 			return rv;
5102 		if (min == 0)
5103 			return -EINVAL;
5104 	}
5105 	mddev->sync_speed_min = min;
5106 	return len;
5107 }
5108 
5109 static struct md_sysfs_entry md_sync_min =
5110 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
5111 
5112 static ssize_t
sync_max_show(struct mddev * mddev,char * page)5113 sync_max_show(struct mddev *mddev, char *page)
5114 {
5115 	return sprintf(page, "%d (%s)\n", speed_max(mddev),
5116 		       mddev->sync_speed_max ? "local": "system");
5117 }
5118 
5119 static ssize_t
sync_max_store(struct mddev * mddev,const char * buf,size_t len)5120 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
5121 {
5122 	unsigned int max;
5123 	int rv;
5124 
5125 	if (strncmp(buf, "system", 6)==0) {
5126 		max = 0;
5127 	} else {
5128 		rv = kstrtouint(buf, 10, &max);
5129 		if (rv < 0)
5130 			return rv;
5131 		if (max == 0)
5132 			return -EINVAL;
5133 	}
5134 	mddev->sync_speed_max = max;
5135 	return len;
5136 }
5137 
5138 static struct md_sysfs_entry md_sync_max =
5139 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
5140 
5141 static ssize_t
degraded_show(struct mddev * mddev,char * page)5142 degraded_show(struct mddev *mddev, char *page)
5143 {
5144 	return sprintf(page, "%d\n", mddev->degraded);
5145 }
5146 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
5147 
5148 static ssize_t
sync_force_parallel_show(struct mddev * mddev,char * page)5149 sync_force_parallel_show(struct mddev *mddev, char *page)
5150 {
5151 	return sprintf(page, "%d\n", mddev->parallel_resync);
5152 }
5153 
5154 static ssize_t
sync_force_parallel_store(struct mddev * mddev,const char * buf,size_t len)5155 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
5156 {
5157 	long n;
5158 
5159 	if (kstrtol(buf, 10, &n))
5160 		return -EINVAL;
5161 
5162 	if (n != 0 && n != 1)
5163 		return -EINVAL;
5164 
5165 	mddev->parallel_resync = n;
5166 
5167 	if (mddev->sync_thread)
5168 		wake_up(&resync_wait);
5169 
5170 	return len;
5171 }
5172 
5173 /* force parallel resync, even with shared block devices */
5174 static struct md_sysfs_entry md_sync_force_parallel =
5175 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
5176        sync_force_parallel_show, sync_force_parallel_store);
5177 
5178 static ssize_t
sync_speed_show(struct mddev * mddev,char * page)5179 sync_speed_show(struct mddev *mddev, char *page)
5180 {
5181 	unsigned long resync, dt, db;
5182 	if (mddev->curr_resync == MD_RESYNC_NONE)
5183 		return sprintf(page, "none\n");
5184 	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
5185 	dt = (jiffies - mddev->resync_mark) / HZ;
5186 	if (!dt) dt++;
5187 	db = resync - mddev->resync_mark_cnt;
5188 	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
5189 }
5190 
5191 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
5192 
5193 static ssize_t
sync_completed_show(struct mddev * mddev,char * page)5194 sync_completed_show(struct mddev *mddev, char *page)
5195 {
5196 	unsigned long long max_sectors, resync;
5197 
5198 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5199 		return sprintf(page, "none\n");
5200 
5201 	if (mddev->curr_resync == MD_RESYNC_YIELDED ||
5202 	    mddev->curr_resync == MD_RESYNC_DELAYED)
5203 		return sprintf(page, "delayed\n");
5204 
5205 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
5206 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5207 		max_sectors = mddev->resync_max_sectors;
5208 	else
5209 		max_sectors = mddev->dev_sectors;
5210 
5211 	resync = mddev->curr_resync_completed;
5212 	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
5213 }
5214 
5215 static struct md_sysfs_entry md_sync_completed =
5216 	__ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
5217 
5218 static ssize_t
min_sync_show(struct mddev * mddev,char * page)5219 min_sync_show(struct mddev *mddev, char *page)
5220 {
5221 	return sprintf(page, "%llu\n",
5222 		       (unsigned long long)mddev->resync_min);
5223 }
5224 static ssize_t
min_sync_store(struct mddev * mddev,const char * buf,size_t len)5225 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5226 {
5227 	unsigned long long min;
5228 	int err;
5229 
5230 	if (kstrtoull(buf, 10, &min))
5231 		return -EINVAL;
5232 
5233 	spin_lock(&mddev->lock);
5234 	err = -EINVAL;
5235 	if (min > mddev->resync_max)
5236 		goto out_unlock;
5237 
5238 	err = -EBUSY;
5239 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5240 		goto out_unlock;
5241 
5242 	/* Round down to multiple of 4K for safety */
5243 	mddev->resync_min = round_down(min, 8);
5244 	err = 0;
5245 
5246 out_unlock:
5247 	spin_unlock(&mddev->lock);
5248 	return err ?: len;
5249 }
5250 
5251 static struct md_sysfs_entry md_min_sync =
5252 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
5253 
5254 static ssize_t
max_sync_show(struct mddev * mddev,char * page)5255 max_sync_show(struct mddev *mddev, char *page)
5256 {
5257 	if (mddev->resync_max == MaxSector)
5258 		return sprintf(page, "max\n");
5259 	else
5260 		return sprintf(page, "%llu\n",
5261 			       (unsigned long long)mddev->resync_max);
5262 }
5263 static ssize_t
max_sync_store(struct mddev * mddev,const char * buf,size_t len)5264 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
5265 {
5266 	int err;
5267 	spin_lock(&mddev->lock);
5268 	if (strncmp(buf, "max", 3) == 0)
5269 		mddev->resync_max = MaxSector;
5270 	else {
5271 		unsigned long long max;
5272 		int chunk;
5273 
5274 		err = -EINVAL;
5275 		if (kstrtoull(buf, 10, &max))
5276 			goto out_unlock;
5277 		if (max < mddev->resync_min)
5278 			goto out_unlock;
5279 
5280 		err = -EBUSY;
5281 		if (max < mddev->resync_max && md_is_rdwr(mddev) &&
5282 		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5283 			goto out_unlock;
5284 
5285 		/* Must be a multiple of chunk_size */
5286 		chunk = mddev->chunk_sectors;
5287 		if (chunk) {
5288 			sector_t temp = max;
5289 
5290 			err = -EINVAL;
5291 			if (sector_div(temp, chunk))
5292 				goto out_unlock;
5293 		}
5294 		mddev->resync_max = max;
5295 	}
5296 	wake_up(&mddev->recovery_wait);
5297 	err = 0;
5298 out_unlock:
5299 	spin_unlock(&mddev->lock);
5300 	return err ?: len;
5301 }
5302 
5303 static struct md_sysfs_entry md_max_sync =
5304 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
5305 
5306 static ssize_t
suspend_lo_show(struct mddev * mddev,char * page)5307 suspend_lo_show(struct mddev *mddev, char *page)
5308 {
5309 	return sprintf(page, "%llu\n",
5310 		       (unsigned long long)READ_ONCE(mddev->suspend_lo));
5311 }
5312 
5313 static ssize_t
suspend_lo_store(struct mddev * mddev,const char * buf,size_t len)5314 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
5315 {
5316 	unsigned long long new;
5317 	int err;
5318 
5319 	err = kstrtoull(buf, 10, &new);
5320 	if (err < 0)
5321 		return err;
5322 	if (new != (sector_t)new)
5323 		return -EINVAL;
5324 
5325 	err = mddev_suspend(mddev, true);
5326 	if (err)
5327 		return err;
5328 
5329 	WRITE_ONCE(mddev->suspend_lo, new);
5330 	mddev_resume(mddev);
5331 
5332 	return len;
5333 }
5334 static struct md_sysfs_entry md_suspend_lo =
5335 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5336 
5337 static ssize_t
suspend_hi_show(struct mddev * mddev,char * page)5338 suspend_hi_show(struct mddev *mddev, char *page)
5339 {
5340 	return sprintf(page, "%llu\n",
5341 		       (unsigned long long)READ_ONCE(mddev->suspend_hi));
5342 }
5343 
5344 static ssize_t
suspend_hi_store(struct mddev * mddev,const char * buf,size_t len)5345 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5346 {
5347 	unsigned long long new;
5348 	int err;
5349 
5350 	err = kstrtoull(buf, 10, &new);
5351 	if (err < 0)
5352 		return err;
5353 	if (new != (sector_t)new)
5354 		return -EINVAL;
5355 
5356 	err = mddev_suspend(mddev, true);
5357 	if (err)
5358 		return err;
5359 
5360 	WRITE_ONCE(mddev->suspend_hi, new);
5361 	mddev_resume(mddev);
5362 
5363 	return len;
5364 }
5365 static struct md_sysfs_entry md_suspend_hi =
5366 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5367 
5368 static ssize_t
reshape_position_show(struct mddev * mddev,char * page)5369 reshape_position_show(struct mddev *mddev, char *page)
5370 {
5371 	if (mddev->reshape_position != MaxSector)
5372 		return sprintf(page, "%llu\n",
5373 			       (unsigned long long)mddev->reshape_position);
5374 	strcpy(page, "none\n");
5375 	return 5;
5376 }
5377 
5378 static ssize_t
reshape_position_store(struct mddev * mddev,const char * buf,size_t len)5379 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5380 {
5381 	struct md_rdev *rdev;
5382 	unsigned long long new;
5383 	int err;
5384 
5385 	err = kstrtoull(buf, 10, &new);
5386 	if (err < 0)
5387 		return err;
5388 	if (new != (sector_t)new)
5389 		return -EINVAL;
5390 	err = mddev_lock(mddev);
5391 	if (err)
5392 		return err;
5393 	err = -EBUSY;
5394 	if (mddev->pers)
5395 		goto unlock;
5396 	mddev->reshape_position = new;
5397 	mddev->delta_disks = 0;
5398 	mddev->reshape_backwards = 0;
5399 	mddev->new_level = mddev->level;
5400 	mddev->new_layout = mddev->layout;
5401 	mddev->new_chunk_sectors = mddev->chunk_sectors;
5402 	rdev_for_each(rdev, mddev)
5403 		rdev->new_data_offset = rdev->data_offset;
5404 	err = 0;
5405 unlock:
5406 	mddev_unlock(mddev);
5407 	return err ?: len;
5408 }
5409 
5410 static struct md_sysfs_entry md_reshape_position =
5411 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5412        reshape_position_store);
5413 
5414 static ssize_t
reshape_direction_show(struct mddev * mddev,char * page)5415 reshape_direction_show(struct mddev *mddev, char *page)
5416 {
5417 	return sprintf(page, "%s\n",
5418 		       mddev->reshape_backwards ? "backwards" : "forwards");
5419 }
5420 
5421 static ssize_t
reshape_direction_store(struct mddev * mddev,const char * buf,size_t len)5422 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5423 {
5424 	int backwards = 0;
5425 	int err;
5426 
5427 	if (cmd_match(buf, "forwards"))
5428 		backwards = 0;
5429 	else if (cmd_match(buf, "backwards"))
5430 		backwards = 1;
5431 	else
5432 		return -EINVAL;
5433 	if (mddev->reshape_backwards == backwards)
5434 		return len;
5435 
5436 	err = mddev_lock(mddev);
5437 	if (err)
5438 		return err;
5439 	/* check if we are allowed to change */
5440 	if (mddev->delta_disks)
5441 		err = -EBUSY;
5442 	else if (mddev->persistent &&
5443 	    mddev->major_version == 0)
5444 		err =  -EINVAL;
5445 	else
5446 		mddev->reshape_backwards = backwards;
5447 	mddev_unlock(mddev);
5448 	return err ?: len;
5449 }
5450 
5451 static struct md_sysfs_entry md_reshape_direction =
5452 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5453        reshape_direction_store);
5454 
5455 static ssize_t
array_size_show(struct mddev * mddev,char * page)5456 array_size_show(struct mddev *mddev, char *page)
5457 {
5458 	if (mddev->external_size)
5459 		return sprintf(page, "%llu\n",
5460 			       (unsigned long long)mddev->array_sectors/2);
5461 	else
5462 		return sprintf(page, "default\n");
5463 }
5464 
5465 static ssize_t
array_size_store(struct mddev * mddev,const char * buf,size_t len)5466 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5467 {
5468 	sector_t sectors;
5469 	int err;
5470 
5471 	err = mddev_lock(mddev);
5472 	if (err)
5473 		return err;
5474 
5475 	/* cluster raid doesn't support change array_sectors */
5476 	if (mddev_is_clustered(mddev)) {
5477 		mddev_unlock(mddev);
5478 		return -EINVAL;
5479 	}
5480 
5481 	if (strncmp(buf, "default", 7) == 0) {
5482 		if (mddev->pers)
5483 			sectors = mddev->pers->size(mddev, 0, 0);
5484 		else
5485 			sectors = mddev->array_sectors;
5486 
5487 		mddev->external_size = 0;
5488 	} else {
5489 		if (strict_blocks_to_sectors(buf, &sectors) < 0)
5490 			err = -EINVAL;
5491 		else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5492 			err = -E2BIG;
5493 		else
5494 			mddev->external_size = 1;
5495 	}
5496 
5497 	if (!err) {
5498 		mddev->array_sectors = sectors;
5499 		if (mddev->pers)
5500 			set_capacity_and_notify(mddev->gendisk,
5501 						mddev->array_sectors);
5502 	}
5503 	mddev_unlock(mddev);
5504 	return err ?: len;
5505 }
5506 
5507 static struct md_sysfs_entry md_array_size =
5508 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5509        array_size_store);
5510 
5511 static ssize_t
consistency_policy_show(struct mddev * mddev,char * page)5512 consistency_policy_show(struct mddev *mddev, char *page)
5513 {
5514 	int ret;
5515 
5516 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5517 		ret = sprintf(page, "journal\n");
5518 	} else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5519 		ret = sprintf(page, "ppl\n");
5520 	} else if (mddev->bitmap) {
5521 		ret = sprintf(page, "bitmap\n");
5522 	} else if (mddev->pers) {
5523 		if (mddev->pers->sync_request)
5524 			ret = sprintf(page, "resync\n");
5525 		else
5526 			ret = sprintf(page, "none\n");
5527 	} else {
5528 		ret = sprintf(page, "unknown\n");
5529 	}
5530 
5531 	return ret;
5532 }
5533 
5534 static ssize_t
consistency_policy_store(struct mddev * mddev,const char * buf,size_t len)5535 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5536 {
5537 	int err = 0;
5538 
5539 	if (mddev->pers) {
5540 		if (mddev->pers->change_consistency_policy)
5541 			err = mddev->pers->change_consistency_policy(mddev, buf);
5542 		else
5543 			err = -EBUSY;
5544 	} else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5545 		set_bit(MD_HAS_PPL, &mddev->flags);
5546 	} else {
5547 		err = -EINVAL;
5548 	}
5549 
5550 	return err ? err : len;
5551 }
5552 
5553 static struct md_sysfs_entry md_consistency_policy =
5554 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5555        consistency_policy_store);
5556 
fail_last_dev_show(struct mddev * mddev,char * page)5557 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5558 {
5559 	return sprintf(page, "%d\n", mddev->fail_last_dev);
5560 }
5561 
5562 /*
5563  * Setting fail_last_dev to true to allow last device to be forcibly removed
5564  * from RAID1/RAID10.
5565  */
5566 static ssize_t
fail_last_dev_store(struct mddev * mddev,const char * buf,size_t len)5567 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5568 {
5569 	int ret;
5570 	bool value;
5571 
5572 	ret = kstrtobool(buf, &value);
5573 	if (ret)
5574 		return ret;
5575 
5576 	if (value != mddev->fail_last_dev)
5577 		mddev->fail_last_dev = value;
5578 
5579 	return len;
5580 }
5581 static struct md_sysfs_entry md_fail_last_dev =
5582 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5583        fail_last_dev_store);
5584 
serialize_policy_show(struct mddev * mddev,char * page)5585 static ssize_t serialize_policy_show(struct mddev *mddev, char *page)
5586 {
5587 	if (mddev->pers == NULL || (mddev->pers->level != 1))
5588 		return sprintf(page, "n/a\n");
5589 	else
5590 		return sprintf(page, "%d\n", mddev->serialize_policy);
5591 }
5592 
5593 /*
5594  * Setting serialize_policy to true to enforce write IO is not reordered
5595  * for raid1.
5596  */
5597 static ssize_t
serialize_policy_store(struct mddev * mddev,const char * buf,size_t len)5598 serialize_policy_store(struct mddev *mddev, const char *buf, size_t len)
5599 {
5600 	int err;
5601 	bool value;
5602 
5603 	err = kstrtobool(buf, &value);
5604 	if (err)
5605 		return err;
5606 
5607 	if (value == mddev->serialize_policy)
5608 		return len;
5609 
5610 	err = mddev_suspend_and_lock(mddev);
5611 	if (err)
5612 		return err;
5613 	if (mddev->pers == NULL || (mddev->pers->level != 1)) {
5614 		pr_err("md: serialize_policy is only effective for raid1\n");
5615 		err = -EINVAL;
5616 		goto unlock;
5617 	}
5618 
5619 	if (value)
5620 		mddev_create_serial_pool(mddev, NULL);
5621 	else
5622 		mddev_destroy_serial_pool(mddev, NULL);
5623 	mddev->serialize_policy = value;
5624 unlock:
5625 	mddev_unlock_and_resume(mddev);
5626 	return err ?: len;
5627 }
5628 
5629 static struct md_sysfs_entry md_serialize_policy =
5630 __ATTR(serialize_policy, S_IRUGO | S_IWUSR, serialize_policy_show,
5631        serialize_policy_store);
5632 
5633 
5634 static struct attribute *md_default_attrs[] = {
5635 	&md_level.attr,
5636 	&md_new_level.attr,
5637 	&md_layout.attr,
5638 	&md_raid_disks.attr,
5639 	&md_uuid.attr,
5640 	&md_chunk_size.attr,
5641 	&md_size.attr,
5642 	&md_resync_start.attr,
5643 	&md_metadata.attr,
5644 	&md_new_device.attr,
5645 	&md_safe_delay.attr,
5646 	&md_array_state.attr,
5647 	&md_reshape_position.attr,
5648 	&md_reshape_direction.attr,
5649 	&md_array_size.attr,
5650 	&max_corr_read_errors.attr,
5651 	&md_consistency_policy.attr,
5652 	&md_fail_last_dev.attr,
5653 	&md_serialize_policy.attr,
5654 	NULL,
5655 };
5656 
5657 static const struct attribute_group md_default_group = {
5658 	.attrs = md_default_attrs,
5659 };
5660 
5661 static struct attribute *md_redundancy_attrs[] = {
5662 	&md_scan_mode.attr,
5663 	&md_last_scan_mode.attr,
5664 	&md_mismatches.attr,
5665 	&md_sync_min.attr,
5666 	&md_sync_max.attr,
5667 	&md_sync_speed.attr,
5668 	&md_sync_force_parallel.attr,
5669 	&md_sync_completed.attr,
5670 	&md_min_sync.attr,
5671 	&md_max_sync.attr,
5672 	&md_suspend_lo.attr,
5673 	&md_suspend_hi.attr,
5674 	&md_bitmap.attr,
5675 	&md_degraded.attr,
5676 	NULL,
5677 };
5678 static const struct attribute_group md_redundancy_group = {
5679 	.name = NULL,
5680 	.attrs = md_redundancy_attrs,
5681 };
5682 
5683 static const struct attribute_group *md_attr_groups[] = {
5684 	&md_default_group,
5685 	&md_bitmap_group,
5686 	NULL,
5687 };
5688 
5689 static ssize_t
md_attr_show(struct kobject * kobj,struct attribute * attr,char * page)5690 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5691 {
5692 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5693 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5694 	ssize_t rv;
5695 
5696 	if (!entry->show)
5697 		return -EIO;
5698 	spin_lock(&all_mddevs_lock);
5699 	if (!mddev_get(mddev)) {
5700 		spin_unlock(&all_mddevs_lock);
5701 		return -EBUSY;
5702 	}
5703 	spin_unlock(&all_mddevs_lock);
5704 
5705 	rv = entry->show(mddev, page);
5706 	mddev_put(mddev);
5707 	return rv;
5708 }
5709 
5710 static ssize_t
md_attr_store(struct kobject * kobj,struct attribute * attr,const char * page,size_t length)5711 md_attr_store(struct kobject *kobj, struct attribute *attr,
5712 	      const char *page, size_t length)
5713 {
5714 	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5715 	struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5716 	ssize_t rv;
5717 
5718 	if (!entry->store)
5719 		return -EIO;
5720 	if (!capable(CAP_SYS_ADMIN))
5721 		return -EACCES;
5722 	spin_lock(&all_mddevs_lock);
5723 	if (!mddev_get(mddev)) {
5724 		spin_unlock(&all_mddevs_lock);
5725 		return -EBUSY;
5726 	}
5727 	spin_unlock(&all_mddevs_lock);
5728 	rv = entry->store(mddev, page, length);
5729 	mddev_put(mddev);
5730 	return rv;
5731 }
5732 
md_kobj_release(struct kobject * ko)5733 static void md_kobj_release(struct kobject *ko)
5734 {
5735 	struct mddev *mddev = container_of(ko, struct mddev, kobj);
5736 
5737 	if (mddev->sysfs_state)
5738 		sysfs_put(mddev->sysfs_state);
5739 	if (mddev->sysfs_level)
5740 		sysfs_put(mddev->sysfs_level);
5741 
5742 	del_gendisk(mddev->gendisk);
5743 	put_disk(mddev->gendisk);
5744 }
5745 
5746 static const struct sysfs_ops md_sysfs_ops = {
5747 	.show	= md_attr_show,
5748 	.store	= md_attr_store,
5749 };
5750 static const struct kobj_type md_ktype = {
5751 	.release	= md_kobj_release,
5752 	.sysfs_ops	= &md_sysfs_ops,
5753 	.default_groups	= md_attr_groups,
5754 };
5755 
5756 int mdp_major = 0;
5757 
5758 /* stack the limit for all rdevs into lim */
mddev_stack_rdev_limits(struct mddev * mddev,struct queue_limits * lim,unsigned int flags)5759 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
5760 		unsigned int flags)
5761 {
5762 	struct md_rdev *rdev;
5763 
5764 	rdev_for_each(rdev, mddev) {
5765 		queue_limits_stack_bdev(lim, rdev->bdev, rdev->data_offset,
5766 					mddev->gendisk->disk_name);
5767 		if ((flags & MDDEV_STACK_INTEGRITY) &&
5768 		    !queue_limits_stack_integrity_bdev(lim, rdev->bdev))
5769 			return -EINVAL;
5770 	}
5771 
5772 	return 0;
5773 }
5774 EXPORT_SYMBOL_GPL(mddev_stack_rdev_limits);
5775 
5776 /* apply the extra stacking limits from a new rdev into mddev */
mddev_stack_new_rdev(struct mddev * mddev,struct md_rdev * rdev)5777 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev)
5778 {
5779 	struct queue_limits lim;
5780 
5781 	if (mddev_is_dm(mddev))
5782 		return 0;
5783 
5784 	lim = queue_limits_start_update(mddev->gendisk->queue);
5785 	queue_limits_stack_bdev(&lim, rdev->bdev, rdev->data_offset,
5786 				mddev->gendisk->disk_name);
5787 
5788 	if (!queue_limits_stack_integrity_bdev(&lim, rdev->bdev)) {
5789 		pr_err("%s: incompatible integrity profile for %pg\n",
5790 		       mdname(mddev), rdev->bdev);
5791 		queue_limits_cancel_update(mddev->gendisk->queue);
5792 		return -ENXIO;
5793 	}
5794 
5795 	return queue_limits_commit_update(mddev->gendisk->queue, &lim);
5796 }
5797 EXPORT_SYMBOL_GPL(mddev_stack_new_rdev);
5798 
5799 /* update the optimal I/O size after a reshape */
mddev_update_io_opt(struct mddev * mddev,unsigned int nr_stripes)5800 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes)
5801 {
5802 	struct queue_limits lim;
5803 
5804 	if (mddev_is_dm(mddev))
5805 		return;
5806 
5807 	/* don't bother updating io_opt if we can't suspend the array */
5808 	if (mddev_suspend(mddev, false) < 0)
5809 		return;
5810 	lim = queue_limits_start_update(mddev->gendisk->queue);
5811 	lim.io_opt = lim.io_min * nr_stripes;
5812 	queue_limits_commit_update(mddev->gendisk->queue, &lim);
5813 	mddev_resume(mddev);
5814 }
5815 EXPORT_SYMBOL_GPL(mddev_update_io_opt);
5816 
mddev_delayed_delete(struct work_struct * ws)5817 static void mddev_delayed_delete(struct work_struct *ws)
5818 {
5819 	struct mddev *mddev = container_of(ws, struct mddev, del_work);
5820 
5821 	kobject_put(&mddev->kobj);
5822 }
5823 
md_init_stacking_limits(struct queue_limits * lim)5824 void md_init_stacking_limits(struct queue_limits *lim)
5825 {
5826 	blk_set_stacking_limits(lim);
5827 	lim->features = BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA |
5828 			BLK_FEAT_IO_STAT | BLK_FEAT_NOWAIT;
5829 }
5830 EXPORT_SYMBOL_GPL(md_init_stacking_limits);
5831 
md_alloc(dev_t dev,char * name)5832 struct mddev *md_alloc(dev_t dev, char *name)
5833 {
5834 	/*
5835 	 * If dev is zero, name is the name of a device to allocate with
5836 	 * an arbitrary minor number.  It will be "md_???"
5837 	 * If dev is non-zero it must be a device number with a MAJOR of
5838 	 * MD_MAJOR or mdp_major.  In this case, if "name" is NULL, then
5839 	 * the device is being created by opening a node in /dev.
5840 	 * If "name" is not NULL, the device is being created by
5841 	 * writing to /sys/module/md_mod/parameters/new_array.
5842 	 */
5843 	static DEFINE_MUTEX(disks_mutex);
5844 	struct mddev *mddev;
5845 	struct gendisk *disk;
5846 	int partitioned;
5847 	int shift;
5848 	int unit;
5849 	int error;
5850 
5851 	/*
5852 	 * Wait for any previous instance of this device to be completely
5853 	 * removed (mddev_delayed_delete).
5854 	 */
5855 	flush_workqueue(md_misc_wq);
5856 
5857 	mutex_lock(&disks_mutex);
5858 	mddev = mddev_alloc(dev);
5859 	if (IS_ERR(mddev)) {
5860 		error = PTR_ERR(mddev);
5861 		goto out_unlock;
5862 	}
5863 
5864 	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5865 	shift = partitioned ? MdpMinorShift : 0;
5866 	unit = MINOR(mddev->unit) >> shift;
5867 
5868 	if (name && !dev) {
5869 		/* Need to ensure that 'name' is not a duplicate.
5870 		 */
5871 		struct mddev *mddev2;
5872 		spin_lock(&all_mddevs_lock);
5873 
5874 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5875 			if (mddev2->gendisk &&
5876 			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
5877 				spin_unlock(&all_mddevs_lock);
5878 				error = -EEXIST;
5879 				goto out_free_mddev;
5880 			}
5881 		spin_unlock(&all_mddevs_lock);
5882 	}
5883 	if (name && dev)
5884 		/*
5885 		 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5886 		 */
5887 		mddev->hold_active = UNTIL_STOP;
5888 
5889 	disk = blk_alloc_disk(NULL, NUMA_NO_NODE);
5890 	if (IS_ERR(disk)) {
5891 		error = PTR_ERR(disk);
5892 		goto out_free_mddev;
5893 	}
5894 
5895 	disk->major = MAJOR(mddev->unit);
5896 	disk->first_minor = unit << shift;
5897 	disk->minors = 1 << shift;
5898 	if (name)
5899 		strcpy(disk->disk_name, name);
5900 	else if (partitioned)
5901 		sprintf(disk->disk_name, "md_d%d", unit);
5902 	else
5903 		sprintf(disk->disk_name, "md%d", unit);
5904 	disk->fops = &md_fops;
5905 	disk->private_data = mddev;
5906 
5907 	disk->events |= DISK_EVENT_MEDIA_CHANGE;
5908 	mddev->gendisk = disk;
5909 	error = add_disk(disk);
5910 	if (error)
5911 		goto out_put_disk;
5912 
5913 	kobject_init(&mddev->kobj, &md_ktype);
5914 	error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5915 	if (error) {
5916 		/*
5917 		 * The disk is already live at this point.  Clear the hold flag
5918 		 * and let mddev_put take care of the deletion, as it isn't any
5919 		 * different from a normal close on last release now.
5920 		 */
5921 		mddev->hold_active = 0;
5922 		mutex_unlock(&disks_mutex);
5923 		mddev_put(mddev);
5924 		return ERR_PTR(error);
5925 	}
5926 
5927 	kobject_uevent(&mddev->kobj, KOBJ_ADD);
5928 	mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5929 	mddev->sysfs_level = sysfs_get_dirent_safe(mddev->kobj.sd, "level");
5930 	mutex_unlock(&disks_mutex);
5931 	return mddev;
5932 
5933 out_put_disk:
5934 	put_disk(disk);
5935 out_free_mddev:
5936 	mddev_free(mddev);
5937 out_unlock:
5938 	mutex_unlock(&disks_mutex);
5939 	return ERR_PTR(error);
5940 }
5941 
md_alloc_and_put(dev_t dev,char * name)5942 static int md_alloc_and_put(dev_t dev, char *name)
5943 {
5944 	struct mddev *mddev = md_alloc(dev, name);
5945 
5946 	if (IS_ERR(mddev))
5947 		return PTR_ERR(mddev);
5948 	mddev_put(mddev);
5949 	return 0;
5950 }
5951 
md_probe(dev_t dev)5952 static void md_probe(dev_t dev)
5953 {
5954 	if (MAJOR(dev) == MD_MAJOR && MINOR(dev) >= 512)
5955 		return;
5956 	if (create_on_open)
5957 		md_alloc_and_put(dev, NULL);
5958 }
5959 
add_named_array(const char * val,const struct kernel_param * kp)5960 static int add_named_array(const char *val, const struct kernel_param *kp)
5961 {
5962 	/*
5963 	 * val must be "md_*" or "mdNNN".
5964 	 * For "md_*" we allocate an array with a large free minor number, and
5965 	 * set the name to val.  val must not already be an active name.
5966 	 * For "mdNNN" we allocate an array with the minor number NNN
5967 	 * which must not already be in use.
5968 	 */
5969 	int len = strlen(val);
5970 	char buf[DISK_NAME_LEN];
5971 	unsigned long devnum;
5972 
5973 	while (len && val[len-1] == '\n')
5974 		len--;
5975 	if (len >= DISK_NAME_LEN)
5976 		return -E2BIG;
5977 	strscpy(buf, val, len+1);
5978 	if (strncmp(buf, "md_", 3) == 0)
5979 		return md_alloc_and_put(0, buf);
5980 	if (strncmp(buf, "md", 2) == 0 &&
5981 	    isdigit(buf[2]) &&
5982 	    kstrtoul(buf+2, 10, &devnum) == 0 &&
5983 	    devnum <= MINORMASK)
5984 		return md_alloc_and_put(MKDEV(MD_MAJOR, devnum), NULL);
5985 
5986 	return -EINVAL;
5987 }
5988 
md_safemode_timeout(struct timer_list * t)5989 static void md_safemode_timeout(struct timer_list *t)
5990 {
5991 	struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5992 
5993 	mddev->safemode = 1;
5994 	if (mddev->external)
5995 		sysfs_notify_dirent_safe(mddev->sysfs_state);
5996 
5997 	md_wakeup_thread(mddev->thread);
5998 }
5999 
6000 static int start_dirty_degraded;
6001 
md_run(struct mddev * mddev)6002 int md_run(struct mddev *mddev)
6003 {
6004 	int err;
6005 	struct md_rdev *rdev;
6006 	struct md_personality *pers;
6007 	bool nowait = true;
6008 
6009 	if (list_empty(&mddev->disks))
6010 		/* cannot run an array with no devices.. */
6011 		return -EINVAL;
6012 
6013 	if (mddev->pers)
6014 		return -EBUSY;
6015 	/* Cannot run until previous stop completes properly */
6016 	if (mddev->sysfs_active)
6017 		return -EBUSY;
6018 
6019 	/*
6020 	 * Analyze all RAID superblock(s)
6021 	 */
6022 	if (!mddev->raid_disks) {
6023 		if (!mddev->persistent)
6024 			return -EINVAL;
6025 		err = analyze_sbs(mddev);
6026 		if (err)
6027 			return -EINVAL;
6028 	}
6029 
6030 	if (mddev->level != LEVEL_NONE)
6031 		request_module("md-level-%d", mddev->level);
6032 	else if (mddev->clevel[0])
6033 		request_module("md-%s", mddev->clevel);
6034 
6035 	/*
6036 	 * Drop all container device buffers, from now on
6037 	 * the only valid external interface is through the md
6038 	 * device.
6039 	 */
6040 	mddev->has_superblocks = false;
6041 	rdev_for_each(rdev, mddev) {
6042 		if (test_bit(Faulty, &rdev->flags))
6043 			continue;
6044 		sync_blockdev(rdev->bdev);
6045 		invalidate_bdev(rdev->bdev);
6046 		if (mddev->ro != MD_RDONLY && rdev_read_only(rdev)) {
6047 			mddev->ro = MD_RDONLY;
6048 			if (!mddev_is_dm(mddev))
6049 				set_disk_ro(mddev->gendisk, 1);
6050 		}
6051 
6052 		if (rdev->sb_page)
6053 			mddev->has_superblocks = true;
6054 
6055 		/* perform some consistency tests on the device.
6056 		 * We don't want the data to overlap the metadata,
6057 		 * Internal Bitmap issues have been handled elsewhere.
6058 		 */
6059 		if (rdev->meta_bdev) {
6060 			/* Nothing to check */;
6061 		} else if (rdev->data_offset < rdev->sb_start) {
6062 			if (mddev->dev_sectors &&
6063 			    rdev->data_offset + mddev->dev_sectors
6064 			    > rdev->sb_start) {
6065 				pr_warn("md: %s: data overlaps metadata\n",
6066 					mdname(mddev));
6067 				return -EINVAL;
6068 			}
6069 		} else {
6070 			if (rdev->sb_start + rdev->sb_size/512
6071 			    > rdev->data_offset) {
6072 				pr_warn("md: %s: metadata overlaps data\n",
6073 					mdname(mddev));
6074 				return -EINVAL;
6075 			}
6076 		}
6077 		sysfs_notify_dirent_safe(rdev->sysfs_state);
6078 		nowait = nowait && bdev_nowait(rdev->bdev);
6079 	}
6080 
6081 	if (!bioset_initialized(&mddev->bio_set)) {
6082 		err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
6083 		if (err)
6084 			return err;
6085 	}
6086 	if (!bioset_initialized(&mddev->sync_set)) {
6087 		err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
6088 		if (err)
6089 			goto exit_bio_set;
6090 	}
6091 
6092 	if (!bioset_initialized(&mddev->io_clone_set)) {
6093 		err = bioset_init(&mddev->io_clone_set, BIO_POOL_SIZE,
6094 				  offsetof(struct md_io_clone, bio_clone), 0);
6095 		if (err)
6096 			goto exit_sync_set;
6097 	}
6098 
6099 	spin_lock(&pers_lock);
6100 	pers = find_pers(mddev->level, mddev->clevel);
6101 	if (!pers || !try_module_get(pers->owner)) {
6102 		spin_unlock(&pers_lock);
6103 		if (mddev->level != LEVEL_NONE)
6104 			pr_warn("md: personality for level %d is not loaded!\n",
6105 				mddev->level);
6106 		else
6107 			pr_warn("md: personality for level %s is not loaded!\n",
6108 				mddev->clevel);
6109 		err = -EINVAL;
6110 		goto abort;
6111 	}
6112 	spin_unlock(&pers_lock);
6113 	if (mddev->level != pers->level) {
6114 		mddev->level = pers->level;
6115 		mddev->new_level = pers->level;
6116 	}
6117 	strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
6118 
6119 	if (mddev->reshape_position != MaxSector &&
6120 	    pers->start_reshape == NULL) {
6121 		/* This personality cannot handle reshaping... */
6122 		module_put(pers->owner);
6123 		err = -EINVAL;
6124 		goto abort;
6125 	}
6126 
6127 	if (pers->sync_request) {
6128 		/* Warn if this is a potentially silly
6129 		 * configuration.
6130 		 */
6131 		struct md_rdev *rdev2;
6132 		int warned = 0;
6133 
6134 		rdev_for_each(rdev, mddev)
6135 			rdev_for_each(rdev2, mddev) {
6136 				if (rdev < rdev2 &&
6137 				    rdev->bdev->bd_disk ==
6138 				    rdev2->bdev->bd_disk) {
6139 					pr_warn("%s: WARNING: %pg appears to be on the same physical disk as %pg.\n",
6140 						mdname(mddev),
6141 						rdev->bdev,
6142 						rdev2->bdev);
6143 					warned = 1;
6144 				}
6145 			}
6146 
6147 		if (warned)
6148 			pr_warn("True protection against single-disk failure might be compromised.\n");
6149 	}
6150 
6151 	/* dm-raid expect sync_thread to be frozen until resume */
6152 	if (mddev->gendisk)
6153 		mddev->recovery = 0;
6154 
6155 	/* may be over-ridden by personality */
6156 	mddev->resync_max_sectors = mddev->dev_sectors;
6157 
6158 	mddev->ok_start_degraded = start_dirty_degraded;
6159 
6160 	if (start_readonly && md_is_rdwr(mddev))
6161 		mddev->ro = MD_AUTO_READ; /* read-only, but switch on first write */
6162 
6163 	err = pers->run(mddev);
6164 	if (err)
6165 		pr_warn("md: pers->run() failed ...\n");
6166 	else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
6167 		WARN_ONCE(!mddev->external_size,
6168 			  "%s: default size too small, but 'external_size' not in effect?\n",
6169 			  __func__);
6170 		pr_warn("md: invalid array_size %llu > default size %llu\n",
6171 			(unsigned long long)mddev->array_sectors / 2,
6172 			(unsigned long long)pers->size(mddev, 0, 0) / 2);
6173 		err = -EINVAL;
6174 	}
6175 	if (err == 0 && pers->sync_request &&
6176 	    (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
6177 		err = mddev->bitmap_ops->create(mddev, -1);
6178 		if (err)
6179 			pr_warn("%s: failed to create bitmap (%d)\n",
6180 				mdname(mddev), err);
6181 	}
6182 	if (err)
6183 		goto bitmap_abort;
6184 
6185 	if (mddev->bitmap_info.max_write_behind > 0) {
6186 		bool create_pool = false;
6187 
6188 		rdev_for_each(rdev, mddev) {
6189 			if (test_bit(WriteMostly, &rdev->flags) &&
6190 			    rdev_init_serial(rdev))
6191 				create_pool = true;
6192 		}
6193 		if (create_pool && mddev->serial_info_pool == NULL) {
6194 			mddev->serial_info_pool =
6195 				mempool_create_kmalloc_pool(NR_SERIAL_INFOS,
6196 						    sizeof(struct serial_info));
6197 			if (!mddev->serial_info_pool) {
6198 				err = -ENOMEM;
6199 				goto bitmap_abort;
6200 			}
6201 		}
6202 	}
6203 
6204 	if (pers->sync_request) {
6205 		if (mddev->kobj.sd &&
6206 		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
6207 			pr_warn("md: cannot register extra attributes for %s\n",
6208 				mdname(mddev));
6209 		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
6210 		mddev->sysfs_completed = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_completed");
6211 		mddev->sysfs_degraded = sysfs_get_dirent_safe(mddev->kobj.sd, "degraded");
6212 	} else if (mddev->ro == MD_AUTO_READ)
6213 		mddev->ro = MD_RDWR;
6214 
6215 	atomic_set(&mddev->max_corr_read_errors,
6216 		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
6217 	mddev->safemode = 0;
6218 	if (mddev_is_clustered(mddev))
6219 		mddev->safemode_delay = 0;
6220 	else
6221 		mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
6222 	mddev->in_sync = 1;
6223 	smp_wmb();
6224 	spin_lock(&mddev->lock);
6225 	mddev->pers = pers;
6226 	spin_unlock(&mddev->lock);
6227 	rdev_for_each(rdev, mddev)
6228 		if (rdev->raid_disk >= 0)
6229 			sysfs_link_rdev(mddev, rdev); /* failure here is OK */
6230 
6231 	if (mddev->degraded && md_is_rdwr(mddev))
6232 		/* This ensures that recovering status is reported immediately
6233 		 * via sysfs - until a lack of spares is confirmed.
6234 		 */
6235 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6236 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6237 
6238 	if (mddev->sb_flags)
6239 		md_update_sb(mddev, 0);
6240 
6241 	md_new_event();
6242 	return 0;
6243 
6244 bitmap_abort:
6245 	mddev_detach(mddev);
6246 	if (mddev->private)
6247 		pers->free(mddev, mddev->private);
6248 	mddev->private = NULL;
6249 	module_put(pers->owner);
6250 	mddev->bitmap_ops->destroy(mddev);
6251 abort:
6252 	bioset_exit(&mddev->io_clone_set);
6253 exit_sync_set:
6254 	bioset_exit(&mddev->sync_set);
6255 exit_bio_set:
6256 	bioset_exit(&mddev->bio_set);
6257 	return err;
6258 }
6259 EXPORT_SYMBOL_GPL(md_run);
6260 
do_md_run(struct mddev * mddev)6261 int do_md_run(struct mddev *mddev)
6262 {
6263 	int err;
6264 
6265 	set_bit(MD_NOT_READY, &mddev->flags);
6266 	err = md_run(mddev);
6267 	if (err)
6268 		goto out;
6269 
6270 	err = mddev->bitmap_ops->load(mddev);
6271 	if (err) {
6272 		mddev->bitmap_ops->destroy(mddev);
6273 		goto out;
6274 	}
6275 
6276 	if (mddev_is_clustered(mddev))
6277 		md_allow_write(mddev);
6278 
6279 	/* run start up tasks that require md_thread */
6280 	md_start(mddev);
6281 
6282 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
6283 
6284 	set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
6285 	clear_bit(MD_NOT_READY, &mddev->flags);
6286 	mddev->changed = 1;
6287 	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
6288 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6289 	sysfs_notify_dirent_safe(mddev->sysfs_action);
6290 	sysfs_notify_dirent_safe(mddev->sysfs_degraded);
6291 out:
6292 	clear_bit(MD_NOT_READY, &mddev->flags);
6293 	return err;
6294 }
6295 
md_start(struct mddev * mddev)6296 int md_start(struct mddev *mddev)
6297 {
6298 	int ret = 0;
6299 
6300 	if (mddev->pers->start) {
6301 		set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6302 		ret = mddev->pers->start(mddev);
6303 		clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
6304 		md_wakeup_thread(mddev->sync_thread);
6305 	}
6306 	return ret;
6307 }
6308 EXPORT_SYMBOL_GPL(md_start);
6309 
restart_array(struct mddev * mddev)6310 static int restart_array(struct mddev *mddev)
6311 {
6312 	struct gendisk *disk = mddev->gendisk;
6313 	struct md_rdev *rdev;
6314 	bool has_journal = false;
6315 	bool has_readonly = false;
6316 
6317 	/* Complain if it has no devices */
6318 	if (list_empty(&mddev->disks))
6319 		return -ENXIO;
6320 	if (!mddev->pers)
6321 		return -EINVAL;
6322 	if (md_is_rdwr(mddev))
6323 		return -EBUSY;
6324 
6325 	rcu_read_lock();
6326 	rdev_for_each_rcu(rdev, mddev) {
6327 		if (test_bit(Journal, &rdev->flags) &&
6328 		    !test_bit(Faulty, &rdev->flags))
6329 			has_journal = true;
6330 		if (rdev_read_only(rdev))
6331 			has_readonly = true;
6332 	}
6333 	rcu_read_unlock();
6334 	if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
6335 		/* Don't restart rw with journal missing/faulty */
6336 			return -EINVAL;
6337 	if (has_readonly)
6338 		return -EROFS;
6339 
6340 	mddev->safemode = 0;
6341 	mddev->ro = MD_RDWR;
6342 	set_disk_ro(disk, 0);
6343 	pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
6344 	/* Kick recovery or resync if necessary */
6345 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6346 	md_wakeup_thread(mddev->sync_thread);
6347 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6348 	return 0;
6349 }
6350 
md_clean(struct mddev * mddev)6351 static void md_clean(struct mddev *mddev)
6352 {
6353 	mddev->array_sectors = 0;
6354 	mddev->external_size = 0;
6355 	mddev->dev_sectors = 0;
6356 	mddev->raid_disks = 0;
6357 	mddev->recovery_cp = 0;
6358 	mddev->resync_min = 0;
6359 	mddev->resync_max = MaxSector;
6360 	mddev->reshape_position = MaxSector;
6361 	/* we still need mddev->external in export_rdev, do not clear it yet */
6362 	mddev->persistent = 0;
6363 	mddev->level = LEVEL_NONE;
6364 	mddev->clevel[0] = 0;
6365 	/*
6366 	 * Don't clear MD_CLOSING, or mddev can be opened again.
6367 	 * 'hold_active != 0' means mddev is still in the creation
6368 	 * process and will be used later.
6369 	 */
6370 	if (mddev->hold_active)
6371 		mddev->flags = 0;
6372 	else
6373 		mddev->flags &= BIT_ULL_MASK(MD_CLOSING);
6374 	mddev->sb_flags = 0;
6375 	mddev->ro = MD_RDWR;
6376 	mddev->metadata_type[0] = 0;
6377 	mddev->chunk_sectors = 0;
6378 	mddev->ctime = mddev->utime = 0;
6379 	mddev->layout = 0;
6380 	mddev->max_disks = 0;
6381 	mddev->events = 0;
6382 	mddev->can_decrease_events = 0;
6383 	mddev->delta_disks = 0;
6384 	mddev->reshape_backwards = 0;
6385 	mddev->new_level = LEVEL_NONE;
6386 	mddev->new_layout = 0;
6387 	mddev->new_chunk_sectors = 0;
6388 	mddev->curr_resync = MD_RESYNC_NONE;
6389 	atomic64_set(&mddev->resync_mismatches, 0);
6390 	mddev->suspend_lo = mddev->suspend_hi = 0;
6391 	mddev->sync_speed_min = mddev->sync_speed_max = 0;
6392 	mddev->recovery = 0;
6393 	mddev->in_sync = 0;
6394 	mddev->changed = 0;
6395 	mddev->degraded = 0;
6396 	mddev->safemode = 0;
6397 	mddev->private = NULL;
6398 	mddev->cluster_info = NULL;
6399 	mddev->bitmap_info.offset = 0;
6400 	mddev->bitmap_info.default_offset = 0;
6401 	mddev->bitmap_info.default_space = 0;
6402 	mddev->bitmap_info.chunksize = 0;
6403 	mddev->bitmap_info.daemon_sleep = 0;
6404 	mddev->bitmap_info.max_write_behind = 0;
6405 	mddev->bitmap_info.nodes = 0;
6406 }
6407 
__md_stop_writes(struct mddev * mddev)6408 static void __md_stop_writes(struct mddev *mddev)
6409 {
6410 	del_timer_sync(&mddev->safemode_timer);
6411 
6412 	if (mddev->pers && mddev->pers->quiesce) {
6413 		mddev->pers->quiesce(mddev, 1);
6414 		mddev->pers->quiesce(mddev, 0);
6415 	}
6416 
6417 	mddev->bitmap_ops->flush(mddev);
6418 
6419 	if (md_is_rdwr(mddev) &&
6420 	    ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6421 	     mddev->sb_flags)) {
6422 		/* mark array as shutdown cleanly */
6423 		if (!mddev_is_clustered(mddev))
6424 			mddev->in_sync = 1;
6425 		md_update_sb(mddev, 1);
6426 	}
6427 	/* disable policy to guarantee rdevs free resources for serialization */
6428 	mddev->serialize_policy = 0;
6429 	mddev_destroy_serial_pool(mddev, NULL);
6430 }
6431 
md_stop_writes(struct mddev * mddev)6432 void md_stop_writes(struct mddev *mddev)
6433 {
6434 	mddev_lock_nointr(mddev);
6435 	set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6436 	stop_sync_thread(mddev, true);
6437 	__md_stop_writes(mddev);
6438 	mddev_unlock(mddev);
6439 }
6440 EXPORT_SYMBOL_GPL(md_stop_writes);
6441 
mddev_detach(struct mddev * mddev)6442 static void mddev_detach(struct mddev *mddev)
6443 {
6444 	mddev->bitmap_ops->wait_behind_writes(mddev);
6445 	if (mddev->pers && mddev->pers->quiesce && !is_md_suspended(mddev)) {
6446 		mddev->pers->quiesce(mddev, 1);
6447 		mddev->pers->quiesce(mddev, 0);
6448 	}
6449 	md_unregister_thread(mddev, &mddev->thread);
6450 
6451 	/* the unplug fn references 'conf' */
6452 	if (!mddev_is_dm(mddev))
6453 		blk_sync_queue(mddev->gendisk->queue);
6454 }
6455 
__md_stop(struct mddev * mddev)6456 static void __md_stop(struct mddev *mddev)
6457 {
6458 	struct md_personality *pers = mddev->pers;
6459 
6460 	mddev->bitmap_ops->destroy(mddev);
6461 	mddev_detach(mddev);
6462 	spin_lock(&mddev->lock);
6463 	mddev->pers = NULL;
6464 	spin_unlock(&mddev->lock);
6465 	if (mddev->private)
6466 		pers->free(mddev, mddev->private);
6467 	mddev->private = NULL;
6468 	if (pers->sync_request && mddev->to_remove == NULL)
6469 		mddev->to_remove = &md_redundancy_group;
6470 	module_put(pers->owner);
6471 	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6472 
6473 	bioset_exit(&mddev->bio_set);
6474 	bioset_exit(&mddev->sync_set);
6475 	bioset_exit(&mddev->io_clone_set);
6476 }
6477 
md_stop(struct mddev * mddev)6478 void md_stop(struct mddev *mddev)
6479 {
6480 	lockdep_assert_held(&mddev->reconfig_mutex);
6481 
6482 	/* stop the array and free an attached data structures.
6483 	 * This is called from dm-raid
6484 	 */
6485 	__md_stop_writes(mddev);
6486 	__md_stop(mddev);
6487 }
6488 
6489 EXPORT_SYMBOL_GPL(md_stop);
6490 
6491 /* ensure 'mddev->pers' exist before calling md_set_readonly() */
md_set_readonly(struct mddev * mddev)6492 static int md_set_readonly(struct mddev *mddev)
6493 {
6494 	int err = 0;
6495 	int did_freeze = 0;
6496 
6497 	if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6498 		return -EBUSY;
6499 
6500 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6501 		did_freeze = 1;
6502 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6503 	}
6504 
6505 	stop_sync_thread(mddev, false);
6506 	wait_event(mddev->sb_wait,
6507 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6508 	mddev_lock_nointr(mddev);
6509 
6510 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6511 		pr_warn("md: %s still in use.\n",mdname(mddev));
6512 		err = -EBUSY;
6513 		goto out;
6514 	}
6515 
6516 	__md_stop_writes(mddev);
6517 
6518 	if (mddev->ro == MD_RDONLY) {
6519 		err  = -ENXIO;
6520 		goto out;
6521 	}
6522 
6523 	mddev->ro = MD_RDONLY;
6524 	set_disk_ro(mddev->gendisk, 1);
6525 
6526 out:
6527 	if (!err || did_freeze) {
6528 		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6529 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6530 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6531 	}
6532 
6533 	return err;
6534 }
6535 
6536 /* mode:
6537  *   0 - completely stop and dis-assemble array
6538  *   2 - stop but do not disassemble array
6539  */
do_md_stop(struct mddev * mddev,int mode)6540 static int do_md_stop(struct mddev *mddev, int mode)
6541 {
6542 	struct gendisk *disk = mddev->gendisk;
6543 	struct md_rdev *rdev;
6544 	int did_freeze = 0;
6545 
6546 	if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6547 		did_freeze = 1;
6548 		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6549 	}
6550 
6551 	stop_sync_thread(mddev, true);
6552 
6553 	if (mddev->sysfs_active ||
6554 	    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6555 		pr_warn("md: %s still in use.\n",mdname(mddev));
6556 		if (did_freeze) {
6557 			clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6558 			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6559 		}
6560 		return -EBUSY;
6561 	}
6562 	if (mddev->pers) {
6563 		if (!md_is_rdwr(mddev))
6564 			set_disk_ro(disk, 0);
6565 
6566 		__md_stop_writes(mddev);
6567 		__md_stop(mddev);
6568 
6569 		/* tell userspace to handle 'inactive' */
6570 		sysfs_notify_dirent_safe(mddev->sysfs_state);
6571 
6572 		rdev_for_each(rdev, mddev)
6573 			if (rdev->raid_disk >= 0)
6574 				sysfs_unlink_rdev(mddev, rdev);
6575 
6576 		set_capacity_and_notify(disk, 0);
6577 		mddev->changed = 1;
6578 
6579 		if (!md_is_rdwr(mddev))
6580 			mddev->ro = MD_RDWR;
6581 	}
6582 	/*
6583 	 * Free resources if final stop
6584 	 */
6585 	if (mode == 0) {
6586 		pr_info("md: %s stopped.\n", mdname(mddev));
6587 
6588 		if (mddev->bitmap_info.file) {
6589 			struct file *f = mddev->bitmap_info.file;
6590 			spin_lock(&mddev->lock);
6591 			mddev->bitmap_info.file = NULL;
6592 			spin_unlock(&mddev->lock);
6593 			fput(f);
6594 		}
6595 		mddev->bitmap_info.offset = 0;
6596 
6597 		export_array(mddev);
6598 
6599 		md_clean(mddev);
6600 		if (mddev->hold_active == UNTIL_STOP)
6601 			mddev->hold_active = 0;
6602 	}
6603 	md_new_event();
6604 	sysfs_notify_dirent_safe(mddev->sysfs_state);
6605 	return 0;
6606 }
6607 
6608 #ifndef MODULE
autorun_array(struct mddev * mddev)6609 static void autorun_array(struct mddev *mddev)
6610 {
6611 	struct md_rdev *rdev;
6612 	int err;
6613 
6614 	if (list_empty(&mddev->disks))
6615 		return;
6616 
6617 	pr_info("md: running: ");
6618 
6619 	rdev_for_each(rdev, mddev) {
6620 		pr_cont("<%pg>", rdev->bdev);
6621 	}
6622 	pr_cont("\n");
6623 
6624 	err = do_md_run(mddev);
6625 	if (err) {
6626 		pr_warn("md: do_md_run() returned %d\n", err);
6627 		do_md_stop(mddev, 0);
6628 	}
6629 }
6630 
6631 /*
6632  * lets try to run arrays based on all disks that have arrived
6633  * until now. (those are in pending_raid_disks)
6634  *
6635  * the method: pick the first pending disk, collect all disks with
6636  * the same UUID, remove all from the pending list and put them into
6637  * the 'same_array' list. Then order this list based on superblock
6638  * update time (freshest comes first), kick out 'old' disks and
6639  * compare superblocks. If everything's fine then run it.
6640  *
6641  * If "unit" is allocated, then bump its reference count
6642  */
autorun_devices(int part)6643 static void autorun_devices(int part)
6644 {
6645 	struct md_rdev *rdev0, *rdev, *tmp;
6646 	struct mddev *mddev;
6647 
6648 	pr_info("md: autorun ...\n");
6649 	while (!list_empty(&pending_raid_disks)) {
6650 		int unit;
6651 		dev_t dev;
6652 		LIST_HEAD(candidates);
6653 		rdev0 = list_entry(pending_raid_disks.next,
6654 					 struct md_rdev, same_set);
6655 
6656 		pr_debug("md: considering %pg ...\n", rdev0->bdev);
6657 		INIT_LIST_HEAD(&candidates);
6658 		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6659 			if (super_90_load(rdev, rdev0, 0) >= 0) {
6660 				pr_debug("md:  adding %pg ...\n",
6661 					 rdev->bdev);
6662 				list_move(&rdev->same_set, &candidates);
6663 			}
6664 		/*
6665 		 * now we have a set of devices, with all of them having
6666 		 * mostly sane superblocks. It's time to allocate the
6667 		 * mddev.
6668 		 */
6669 		if (part) {
6670 			dev = MKDEV(mdp_major,
6671 				    rdev0->preferred_minor << MdpMinorShift);
6672 			unit = MINOR(dev) >> MdpMinorShift;
6673 		} else {
6674 			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6675 			unit = MINOR(dev);
6676 		}
6677 		if (rdev0->preferred_minor != unit) {
6678 			pr_warn("md: unit number in %pg is bad: %d\n",
6679 				rdev0->bdev, rdev0->preferred_minor);
6680 			break;
6681 		}
6682 
6683 		mddev = md_alloc(dev, NULL);
6684 		if (IS_ERR(mddev))
6685 			break;
6686 
6687 		if (mddev_suspend_and_lock(mddev))
6688 			pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6689 		else if (mddev->raid_disks || mddev->major_version
6690 			 || !list_empty(&mddev->disks)) {
6691 			pr_warn("md: %s already running, cannot run %pg\n",
6692 				mdname(mddev), rdev0->bdev);
6693 			mddev_unlock_and_resume(mddev);
6694 		} else {
6695 			pr_debug("md: created %s\n", mdname(mddev));
6696 			mddev->persistent = 1;
6697 			rdev_for_each_list(rdev, tmp, &candidates) {
6698 				list_del_init(&rdev->same_set);
6699 				if (bind_rdev_to_array(rdev, mddev))
6700 					export_rdev(rdev, mddev);
6701 			}
6702 			autorun_array(mddev);
6703 			mddev_unlock_and_resume(mddev);
6704 		}
6705 		/* on success, candidates will be empty, on error
6706 		 * it won't...
6707 		 */
6708 		rdev_for_each_list(rdev, tmp, &candidates) {
6709 			list_del_init(&rdev->same_set);
6710 			export_rdev(rdev, mddev);
6711 		}
6712 		mddev_put(mddev);
6713 	}
6714 	pr_info("md: ... autorun DONE.\n");
6715 }
6716 #endif /* !MODULE */
6717 
get_version(void __user * arg)6718 static int get_version(void __user *arg)
6719 {
6720 	mdu_version_t ver;
6721 
6722 	ver.major = MD_MAJOR_VERSION;
6723 	ver.minor = MD_MINOR_VERSION;
6724 	ver.patchlevel = MD_PATCHLEVEL_VERSION;
6725 
6726 	if (copy_to_user(arg, &ver, sizeof(ver)))
6727 		return -EFAULT;
6728 
6729 	return 0;
6730 }
6731 
get_array_info(struct mddev * mddev,void __user * arg)6732 static int get_array_info(struct mddev *mddev, void __user *arg)
6733 {
6734 	mdu_array_info_t info;
6735 	int nr,working,insync,failed,spare;
6736 	struct md_rdev *rdev;
6737 
6738 	nr = working = insync = failed = spare = 0;
6739 	rcu_read_lock();
6740 	rdev_for_each_rcu(rdev, mddev) {
6741 		nr++;
6742 		if (test_bit(Faulty, &rdev->flags))
6743 			failed++;
6744 		else {
6745 			working++;
6746 			if (test_bit(In_sync, &rdev->flags))
6747 				insync++;
6748 			else if (test_bit(Journal, &rdev->flags))
6749 				/* TODO: add journal count to md_u.h */
6750 				;
6751 			else
6752 				spare++;
6753 		}
6754 	}
6755 	rcu_read_unlock();
6756 
6757 	info.major_version = mddev->major_version;
6758 	info.minor_version = mddev->minor_version;
6759 	info.patch_version = MD_PATCHLEVEL_VERSION;
6760 	info.ctime         = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6761 	info.level         = mddev->level;
6762 	info.size          = mddev->dev_sectors / 2;
6763 	if (info.size != mddev->dev_sectors / 2) /* overflow */
6764 		info.size = -1;
6765 	info.nr_disks      = nr;
6766 	info.raid_disks    = mddev->raid_disks;
6767 	info.md_minor      = mddev->md_minor;
6768 	info.not_persistent= !mddev->persistent;
6769 
6770 	info.utime         = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6771 	info.state         = 0;
6772 	if (mddev->in_sync)
6773 		info.state = (1<<MD_SB_CLEAN);
6774 	if (mddev->bitmap && mddev->bitmap_info.offset)
6775 		info.state |= (1<<MD_SB_BITMAP_PRESENT);
6776 	if (mddev_is_clustered(mddev))
6777 		info.state |= (1<<MD_SB_CLUSTERED);
6778 	info.active_disks  = insync;
6779 	info.working_disks = working;
6780 	info.failed_disks  = failed;
6781 	info.spare_disks   = spare;
6782 
6783 	info.layout        = mddev->layout;
6784 	info.chunk_size    = mddev->chunk_sectors << 9;
6785 
6786 	if (copy_to_user(arg, &info, sizeof(info)))
6787 		return -EFAULT;
6788 
6789 	return 0;
6790 }
6791 
get_bitmap_file(struct mddev * mddev,void __user * arg)6792 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6793 {
6794 	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6795 	char *ptr;
6796 	int err;
6797 
6798 	file = kzalloc(sizeof(*file), GFP_NOIO);
6799 	if (!file)
6800 		return -ENOMEM;
6801 
6802 	err = 0;
6803 	spin_lock(&mddev->lock);
6804 	/* bitmap enabled */
6805 	if (mddev->bitmap_info.file) {
6806 		ptr = file_path(mddev->bitmap_info.file, file->pathname,
6807 				sizeof(file->pathname));
6808 		if (IS_ERR(ptr))
6809 			err = PTR_ERR(ptr);
6810 		else
6811 			memmove(file->pathname, ptr,
6812 				sizeof(file->pathname)-(ptr-file->pathname));
6813 	}
6814 	spin_unlock(&mddev->lock);
6815 
6816 	if (err == 0 &&
6817 	    copy_to_user(arg, file, sizeof(*file)))
6818 		err = -EFAULT;
6819 
6820 	kfree(file);
6821 	return err;
6822 }
6823 
get_disk_info(struct mddev * mddev,void __user * arg)6824 static int get_disk_info(struct mddev *mddev, void __user * arg)
6825 {
6826 	mdu_disk_info_t info;
6827 	struct md_rdev *rdev;
6828 
6829 	if (copy_from_user(&info, arg, sizeof(info)))
6830 		return -EFAULT;
6831 
6832 	rcu_read_lock();
6833 	rdev = md_find_rdev_nr_rcu(mddev, info.number);
6834 	if (rdev) {
6835 		info.major = MAJOR(rdev->bdev->bd_dev);
6836 		info.minor = MINOR(rdev->bdev->bd_dev);
6837 		info.raid_disk = rdev->raid_disk;
6838 		info.state = 0;
6839 		if (test_bit(Faulty, &rdev->flags))
6840 			info.state |= (1<<MD_DISK_FAULTY);
6841 		else if (test_bit(In_sync, &rdev->flags)) {
6842 			info.state |= (1<<MD_DISK_ACTIVE);
6843 			info.state |= (1<<MD_DISK_SYNC);
6844 		}
6845 		if (test_bit(Journal, &rdev->flags))
6846 			info.state |= (1<<MD_DISK_JOURNAL);
6847 		if (test_bit(WriteMostly, &rdev->flags))
6848 			info.state |= (1<<MD_DISK_WRITEMOSTLY);
6849 		if (test_bit(FailFast, &rdev->flags))
6850 			info.state |= (1<<MD_DISK_FAILFAST);
6851 	} else {
6852 		info.major = info.minor = 0;
6853 		info.raid_disk = -1;
6854 		info.state = (1<<MD_DISK_REMOVED);
6855 	}
6856 	rcu_read_unlock();
6857 
6858 	if (copy_to_user(arg, &info, sizeof(info)))
6859 		return -EFAULT;
6860 
6861 	return 0;
6862 }
6863 
md_add_new_disk(struct mddev * mddev,struct mdu_disk_info_s * info)6864 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info)
6865 {
6866 	struct md_rdev *rdev;
6867 	dev_t dev = MKDEV(info->major,info->minor);
6868 
6869 	if (mddev_is_clustered(mddev) &&
6870 		!(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6871 		pr_warn("%s: Cannot add to clustered mddev.\n",
6872 			mdname(mddev));
6873 		return -EINVAL;
6874 	}
6875 
6876 	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6877 		return -EOVERFLOW;
6878 
6879 	if (!mddev->raid_disks) {
6880 		int err;
6881 		/* expecting a device which has a superblock */
6882 		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6883 		if (IS_ERR(rdev)) {
6884 			pr_warn("md: md_import_device returned %ld\n",
6885 				PTR_ERR(rdev));
6886 			return PTR_ERR(rdev);
6887 		}
6888 		if (!list_empty(&mddev->disks)) {
6889 			struct md_rdev *rdev0
6890 				= list_entry(mddev->disks.next,
6891 					     struct md_rdev, same_set);
6892 			err = super_types[mddev->major_version]
6893 				.load_super(rdev, rdev0, mddev->minor_version);
6894 			if (err < 0) {
6895 				pr_warn("md: %pg has different UUID to %pg\n",
6896 					rdev->bdev,
6897 					rdev0->bdev);
6898 				export_rdev(rdev, mddev);
6899 				return -EINVAL;
6900 			}
6901 		}
6902 		err = bind_rdev_to_array(rdev, mddev);
6903 		if (err)
6904 			export_rdev(rdev, mddev);
6905 		return err;
6906 	}
6907 
6908 	/*
6909 	 * md_add_new_disk can be used once the array is assembled
6910 	 * to add "hot spares".  They must already have a superblock
6911 	 * written
6912 	 */
6913 	if (mddev->pers) {
6914 		int err;
6915 		if (!mddev->pers->hot_add_disk) {
6916 			pr_warn("%s: personality does not support diskops!\n",
6917 				mdname(mddev));
6918 			return -EINVAL;
6919 		}
6920 		if (mddev->persistent)
6921 			rdev = md_import_device(dev, mddev->major_version,
6922 						mddev->minor_version);
6923 		else
6924 			rdev = md_import_device(dev, -1, -1);
6925 		if (IS_ERR(rdev)) {
6926 			pr_warn("md: md_import_device returned %ld\n",
6927 				PTR_ERR(rdev));
6928 			return PTR_ERR(rdev);
6929 		}
6930 		/* set saved_raid_disk if appropriate */
6931 		if (!mddev->persistent) {
6932 			if (info->state & (1<<MD_DISK_SYNC)  &&
6933 			    info->raid_disk < mddev->raid_disks) {
6934 				rdev->raid_disk = info->raid_disk;
6935 				clear_bit(Bitmap_sync, &rdev->flags);
6936 			} else
6937 				rdev->raid_disk = -1;
6938 			rdev->saved_raid_disk = rdev->raid_disk;
6939 		} else
6940 			super_types[mddev->major_version].
6941 				validate_super(mddev, NULL/*freshest*/, rdev);
6942 		if ((info->state & (1<<MD_DISK_SYNC)) &&
6943 		     rdev->raid_disk != info->raid_disk) {
6944 			/* This was a hot-add request, but events doesn't
6945 			 * match, so reject it.
6946 			 */
6947 			export_rdev(rdev, mddev);
6948 			return -EINVAL;
6949 		}
6950 
6951 		clear_bit(In_sync, &rdev->flags); /* just to be sure */
6952 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6953 			set_bit(WriteMostly, &rdev->flags);
6954 		else
6955 			clear_bit(WriteMostly, &rdev->flags);
6956 		if (info->state & (1<<MD_DISK_FAILFAST))
6957 			set_bit(FailFast, &rdev->flags);
6958 		else
6959 			clear_bit(FailFast, &rdev->flags);
6960 
6961 		if (info->state & (1<<MD_DISK_JOURNAL)) {
6962 			struct md_rdev *rdev2;
6963 			bool has_journal = false;
6964 
6965 			/* make sure no existing journal disk */
6966 			rdev_for_each(rdev2, mddev) {
6967 				if (test_bit(Journal, &rdev2->flags)) {
6968 					has_journal = true;
6969 					break;
6970 				}
6971 			}
6972 			if (has_journal || mddev->bitmap) {
6973 				export_rdev(rdev, mddev);
6974 				return -EBUSY;
6975 			}
6976 			set_bit(Journal, &rdev->flags);
6977 		}
6978 		/*
6979 		 * check whether the device shows up in other nodes
6980 		 */
6981 		if (mddev_is_clustered(mddev)) {
6982 			if (info->state & (1 << MD_DISK_CANDIDATE))
6983 				set_bit(Candidate, &rdev->flags);
6984 			else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6985 				/* --add initiated by this node */
6986 				err = md_cluster_ops->add_new_disk(mddev, rdev);
6987 				if (err) {
6988 					export_rdev(rdev, mddev);
6989 					return err;
6990 				}
6991 			}
6992 		}
6993 
6994 		rdev->raid_disk = -1;
6995 		err = bind_rdev_to_array(rdev, mddev);
6996 
6997 		if (err)
6998 			export_rdev(rdev, mddev);
6999 
7000 		if (mddev_is_clustered(mddev)) {
7001 			if (info->state & (1 << MD_DISK_CANDIDATE)) {
7002 				if (!err) {
7003 					err = md_cluster_ops->new_disk_ack(mddev,
7004 						err == 0);
7005 					if (err)
7006 						md_kick_rdev_from_array(rdev);
7007 				}
7008 			} else {
7009 				if (err)
7010 					md_cluster_ops->add_new_disk_cancel(mddev);
7011 				else
7012 					err = add_bound_rdev(rdev);
7013 			}
7014 
7015 		} else if (!err)
7016 			err = add_bound_rdev(rdev);
7017 
7018 		return err;
7019 	}
7020 
7021 	/* otherwise, md_add_new_disk is only allowed
7022 	 * for major_version==0 superblocks
7023 	 */
7024 	if (mddev->major_version != 0) {
7025 		pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
7026 		return -EINVAL;
7027 	}
7028 
7029 	if (!(info->state & (1<<MD_DISK_FAULTY))) {
7030 		int err;
7031 		rdev = md_import_device(dev, -1, 0);
7032 		if (IS_ERR(rdev)) {
7033 			pr_warn("md: error, md_import_device() returned %ld\n",
7034 				PTR_ERR(rdev));
7035 			return PTR_ERR(rdev);
7036 		}
7037 		rdev->desc_nr = info->number;
7038 		if (info->raid_disk < mddev->raid_disks)
7039 			rdev->raid_disk = info->raid_disk;
7040 		else
7041 			rdev->raid_disk = -1;
7042 
7043 		if (rdev->raid_disk < mddev->raid_disks)
7044 			if (info->state & (1<<MD_DISK_SYNC))
7045 				set_bit(In_sync, &rdev->flags);
7046 
7047 		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
7048 			set_bit(WriteMostly, &rdev->flags);
7049 		if (info->state & (1<<MD_DISK_FAILFAST))
7050 			set_bit(FailFast, &rdev->flags);
7051 
7052 		if (!mddev->persistent) {
7053 			pr_debug("md: nonpersistent superblock ...\n");
7054 			rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7055 		} else
7056 			rdev->sb_start = calc_dev_sboffset(rdev);
7057 		rdev->sectors = rdev->sb_start;
7058 
7059 		err = bind_rdev_to_array(rdev, mddev);
7060 		if (err) {
7061 			export_rdev(rdev, mddev);
7062 			return err;
7063 		}
7064 	}
7065 
7066 	return 0;
7067 }
7068 
hot_remove_disk(struct mddev * mddev,dev_t dev)7069 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
7070 {
7071 	struct md_rdev *rdev;
7072 
7073 	if (!mddev->pers)
7074 		return -ENODEV;
7075 
7076 	rdev = find_rdev(mddev, dev);
7077 	if (!rdev)
7078 		return -ENXIO;
7079 
7080 	if (rdev->raid_disk < 0)
7081 		goto kick_rdev;
7082 
7083 	clear_bit(Blocked, &rdev->flags);
7084 	remove_and_add_spares(mddev, rdev);
7085 
7086 	if (rdev->raid_disk >= 0)
7087 		goto busy;
7088 
7089 kick_rdev:
7090 	if (mddev_is_clustered(mddev)) {
7091 		if (md_cluster_ops->remove_disk(mddev, rdev))
7092 			goto busy;
7093 	}
7094 
7095 	md_kick_rdev_from_array(rdev);
7096 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7097 	if (!mddev->thread)
7098 		md_update_sb(mddev, 1);
7099 	md_new_event();
7100 
7101 	return 0;
7102 busy:
7103 	pr_debug("md: cannot remove active disk %pg from %s ...\n",
7104 		 rdev->bdev, mdname(mddev));
7105 	return -EBUSY;
7106 }
7107 
hot_add_disk(struct mddev * mddev,dev_t dev)7108 static int hot_add_disk(struct mddev *mddev, dev_t dev)
7109 {
7110 	int err;
7111 	struct md_rdev *rdev;
7112 
7113 	if (!mddev->pers)
7114 		return -ENODEV;
7115 
7116 	if (mddev->major_version != 0) {
7117 		pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
7118 			mdname(mddev));
7119 		return -EINVAL;
7120 	}
7121 	if (!mddev->pers->hot_add_disk) {
7122 		pr_warn("%s: personality does not support diskops!\n",
7123 			mdname(mddev));
7124 		return -EINVAL;
7125 	}
7126 
7127 	rdev = md_import_device(dev, -1, 0);
7128 	if (IS_ERR(rdev)) {
7129 		pr_warn("md: error, md_import_device() returned %ld\n",
7130 			PTR_ERR(rdev));
7131 		return -EINVAL;
7132 	}
7133 
7134 	if (mddev->persistent)
7135 		rdev->sb_start = calc_dev_sboffset(rdev);
7136 	else
7137 		rdev->sb_start = bdev_nr_sectors(rdev->bdev);
7138 
7139 	rdev->sectors = rdev->sb_start;
7140 
7141 	if (test_bit(Faulty, &rdev->flags)) {
7142 		pr_warn("md: can not hot-add faulty %pg disk to %s!\n",
7143 			rdev->bdev, mdname(mddev));
7144 		err = -EINVAL;
7145 		goto abort_export;
7146 	}
7147 
7148 	clear_bit(In_sync, &rdev->flags);
7149 	rdev->desc_nr = -1;
7150 	rdev->saved_raid_disk = -1;
7151 	err = bind_rdev_to_array(rdev, mddev);
7152 	if (err)
7153 		goto abort_export;
7154 
7155 	/*
7156 	 * The rest should better be atomic, we can have disk failures
7157 	 * noticed in interrupt contexts ...
7158 	 */
7159 
7160 	rdev->raid_disk = -1;
7161 
7162 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7163 	if (!mddev->thread)
7164 		md_update_sb(mddev, 1);
7165 	/*
7166 	 * Kick recovery, maybe this spare has to be added to the
7167 	 * array immediately.
7168 	 */
7169 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7170 	md_new_event();
7171 	return 0;
7172 
7173 abort_export:
7174 	export_rdev(rdev, mddev);
7175 	return err;
7176 }
7177 
set_bitmap_file(struct mddev * mddev,int fd)7178 static int set_bitmap_file(struct mddev *mddev, int fd)
7179 {
7180 	int err = 0;
7181 
7182 	if (mddev->pers) {
7183 		if (!mddev->pers->quiesce || !mddev->thread)
7184 			return -EBUSY;
7185 		if (mddev->recovery || mddev->sync_thread)
7186 			return -EBUSY;
7187 		/* we should be able to change the bitmap.. */
7188 	}
7189 
7190 	if (fd >= 0) {
7191 		struct inode *inode;
7192 		struct file *f;
7193 
7194 		if (mddev->bitmap || mddev->bitmap_info.file)
7195 			return -EEXIST; /* cannot add when bitmap is present */
7196 
7197 		if (!IS_ENABLED(CONFIG_MD_BITMAP_FILE)) {
7198 			pr_warn("%s: bitmap files not supported by this kernel\n",
7199 				mdname(mddev));
7200 			return -EINVAL;
7201 		}
7202 		pr_warn("%s: using deprecated bitmap file support\n",
7203 			mdname(mddev));
7204 
7205 		f = fget(fd);
7206 
7207 		if (f == NULL) {
7208 			pr_warn("%s: error: failed to get bitmap file\n",
7209 				mdname(mddev));
7210 			return -EBADF;
7211 		}
7212 
7213 		inode = f->f_mapping->host;
7214 		if (!S_ISREG(inode->i_mode)) {
7215 			pr_warn("%s: error: bitmap file must be a regular file\n",
7216 				mdname(mddev));
7217 			err = -EBADF;
7218 		} else if (!(f->f_mode & FMODE_WRITE)) {
7219 			pr_warn("%s: error: bitmap file must open for write\n",
7220 				mdname(mddev));
7221 			err = -EBADF;
7222 		} else if (atomic_read(&inode->i_writecount) != 1) {
7223 			pr_warn("%s: error: bitmap file is already in use\n",
7224 				mdname(mddev));
7225 			err = -EBUSY;
7226 		}
7227 		if (err) {
7228 			fput(f);
7229 			return err;
7230 		}
7231 		mddev->bitmap_info.file = f;
7232 		mddev->bitmap_info.offset = 0; /* file overrides offset */
7233 	} else if (mddev->bitmap == NULL)
7234 		return -ENOENT; /* cannot remove what isn't there */
7235 	err = 0;
7236 	if (mddev->pers) {
7237 		if (fd >= 0) {
7238 			err = mddev->bitmap_ops->create(mddev, -1);
7239 			if (!err)
7240 				err = mddev->bitmap_ops->load(mddev);
7241 
7242 			if (err) {
7243 				mddev->bitmap_ops->destroy(mddev);
7244 				fd = -1;
7245 			}
7246 		} else if (fd < 0) {
7247 			mddev->bitmap_ops->destroy(mddev);
7248 		}
7249 	}
7250 
7251 	if (fd < 0) {
7252 		struct file *f = mddev->bitmap_info.file;
7253 		if (f) {
7254 			spin_lock(&mddev->lock);
7255 			mddev->bitmap_info.file = NULL;
7256 			spin_unlock(&mddev->lock);
7257 			fput(f);
7258 		}
7259 	}
7260 
7261 	return err;
7262 }
7263 
7264 /*
7265  * md_set_array_info is used two different ways
7266  * The original usage is when creating a new array.
7267  * In this usage, raid_disks is > 0 and it together with
7268  *  level, size, not_persistent,layout,chunksize determine the
7269  *  shape of the array.
7270  *  This will always create an array with a type-0.90.0 superblock.
7271  * The newer usage is when assembling an array.
7272  *  In this case raid_disks will be 0, and the major_version field is
7273  *  use to determine which style super-blocks are to be found on the devices.
7274  *  The minor and patch _version numbers are also kept incase the
7275  *  super_block handler wishes to interpret them.
7276  */
md_set_array_info(struct mddev * mddev,struct mdu_array_info_s * info)7277 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info)
7278 {
7279 	if (info->raid_disks == 0) {
7280 		/* just setting version number for superblock loading */
7281 		if (info->major_version < 0 ||
7282 		    info->major_version >= ARRAY_SIZE(super_types) ||
7283 		    super_types[info->major_version].name == NULL) {
7284 			/* maybe try to auto-load a module? */
7285 			pr_warn("md: superblock version %d not known\n",
7286 				info->major_version);
7287 			return -EINVAL;
7288 		}
7289 		mddev->major_version = info->major_version;
7290 		mddev->minor_version = info->minor_version;
7291 		mddev->patch_version = info->patch_version;
7292 		mddev->persistent = !info->not_persistent;
7293 		/* ensure mddev_put doesn't delete this now that there
7294 		 * is some minimal configuration.
7295 		 */
7296 		mddev->ctime         = ktime_get_real_seconds();
7297 		return 0;
7298 	}
7299 	mddev->major_version = MD_MAJOR_VERSION;
7300 	mddev->minor_version = MD_MINOR_VERSION;
7301 	mddev->patch_version = MD_PATCHLEVEL_VERSION;
7302 	mddev->ctime         = ktime_get_real_seconds();
7303 
7304 	mddev->level         = info->level;
7305 	mddev->clevel[0]     = 0;
7306 	mddev->dev_sectors   = 2 * (sector_t)info->size;
7307 	mddev->raid_disks    = info->raid_disks;
7308 	/* don't set md_minor, it is determined by which /dev/md* was
7309 	 * openned
7310 	 */
7311 	if (info->state & (1<<MD_SB_CLEAN))
7312 		mddev->recovery_cp = MaxSector;
7313 	else
7314 		mddev->recovery_cp = 0;
7315 	mddev->persistent    = ! info->not_persistent;
7316 	mddev->external	     = 0;
7317 
7318 	mddev->layout        = info->layout;
7319 	if (mddev->level == 0)
7320 		/* Cannot trust RAID0 layout info here */
7321 		mddev->layout = -1;
7322 	mddev->chunk_sectors = info->chunk_size >> 9;
7323 
7324 	if (mddev->persistent) {
7325 		mddev->max_disks = MD_SB_DISKS;
7326 		mddev->flags = 0;
7327 		mddev->sb_flags = 0;
7328 	}
7329 	set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
7330 
7331 	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
7332 	mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
7333 	mddev->bitmap_info.offset = 0;
7334 
7335 	mddev->reshape_position = MaxSector;
7336 
7337 	/*
7338 	 * Generate a 128 bit UUID
7339 	 */
7340 	get_random_bytes(mddev->uuid, 16);
7341 
7342 	mddev->new_level = mddev->level;
7343 	mddev->new_chunk_sectors = mddev->chunk_sectors;
7344 	mddev->new_layout = mddev->layout;
7345 	mddev->delta_disks = 0;
7346 	mddev->reshape_backwards = 0;
7347 
7348 	return 0;
7349 }
7350 
md_set_array_sectors(struct mddev * mddev,sector_t array_sectors)7351 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
7352 {
7353 	lockdep_assert_held(&mddev->reconfig_mutex);
7354 
7355 	if (mddev->external_size)
7356 		return;
7357 
7358 	mddev->array_sectors = array_sectors;
7359 }
7360 EXPORT_SYMBOL(md_set_array_sectors);
7361 
update_size(struct mddev * mddev,sector_t num_sectors)7362 static int update_size(struct mddev *mddev, sector_t num_sectors)
7363 {
7364 	struct md_rdev *rdev;
7365 	int rv;
7366 	int fit = (num_sectors == 0);
7367 	sector_t old_dev_sectors = mddev->dev_sectors;
7368 
7369 	if (mddev->pers->resize == NULL)
7370 		return -EINVAL;
7371 	/* The "num_sectors" is the number of sectors of each device that
7372 	 * is used.  This can only make sense for arrays with redundancy.
7373 	 * linear and raid0 always use whatever space is available. We can only
7374 	 * consider changing this number if no resync or reconstruction is
7375 	 * happening, and if the new size is acceptable. It must fit before the
7376 	 * sb_start or, if that is <data_offset, it must fit before the size
7377 	 * of each device.  If num_sectors is zero, we find the largest size
7378 	 * that fits.
7379 	 */
7380 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
7381 		return -EBUSY;
7382 	if (!md_is_rdwr(mddev))
7383 		return -EROFS;
7384 
7385 	rdev_for_each(rdev, mddev) {
7386 		sector_t avail = rdev->sectors;
7387 
7388 		if (fit && (num_sectors == 0 || num_sectors > avail))
7389 			num_sectors = avail;
7390 		if (avail < num_sectors)
7391 			return -ENOSPC;
7392 	}
7393 	rv = mddev->pers->resize(mddev, num_sectors);
7394 	if (!rv) {
7395 		if (mddev_is_clustered(mddev))
7396 			md_cluster_ops->update_size(mddev, old_dev_sectors);
7397 		else if (!mddev_is_dm(mddev))
7398 			set_capacity_and_notify(mddev->gendisk,
7399 						mddev->array_sectors);
7400 	}
7401 	return rv;
7402 }
7403 
update_raid_disks(struct mddev * mddev,int raid_disks)7404 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7405 {
7406 	int rv;
7407 	struct md_rdev *rdev;
7408 	/* change the number of raid disks */
7409 	if (mddev->pers->check_reshape == NULL)
7410 		return -EINVAL;
7411 	if (!md_is_rdwr(mddev))
7412 		return -EROFS;
7413 	if (raid_disks <= 0 ||
7414 	    (mddev->max_disks && raid_disks >= mddev->max_disks))
7415 		return -EINVAL;
7416 	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7417 	    test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) ||
7418 	    mddev->reshape_position != MaxSector)
7419 		return -EBUSY;
7420 
7421 	rdev_for_each(rdev, mddev) {
7422 		if (mddev->raid_disks < raid_disks &&
7423 		    rdev->data_offset < rdev->new_data_offset)
7424 			return -EINVAL;
7425 		if (mddev->raid_disks > raid_disks &&
7426 		    rdev->data_offset > rdev->new_data_offset)
7427 			return -EINVAL;
7428 	}
7429 
7430 	mddev->delta_disks = raid_disks - mddev->raid_disks;
7431 	if (mddev->delta_disks < 0)
7432 		mddev->reshape_backwards = 1;
7433 	else if (mddev->delta_disks > 0)
7434 		mddev->reshape_backwards = 0;
7435 
7436 	rv = mddev->pers->check_reshape(mddev);
7437 	if (rv < 0) {
7438 		mddev->delta_disks = 0;
7439 		mddev->reshape_backwards = 0;
7440 	}
7441 	return rv;
7442 }
7443 
7444 /*
7445  * update_array_info is used to change the configuration of an
7446  * on-line array.
7447  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7448  * fields in the info are checked against the array.
7449  * Any differences that cannot be handled will cause an error.
7450  * Normally, only one change can be managed at a time.
7451  */
update_array_info(struct mddev * mddev,mdu_array_info_t * info)7452 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7453 {
7454 	int rv = 0;
7455 	int cnt = 0;
7456 	int state = 0;
7457 
7458 	/* calculate expected state,ignoring low bits */
7459 	if (mddev->bitmap && mddev->bitmap_info.offset)
7460 		state |= (1 << MD_SB_BITMAP_PRESENT);
7461 
7462 	if (mddev->major_version != info->major_version ||
7463 	    mddev->minor_version != info->minor_version ||
7464 /*	    mddev->patch_version != info->patch_version || */
7465 	    mddev->ctime         != info->ctime         ||
7466 	    mddev->level         != info->level         ||
7467 /*	    mddev->layout        != info->layout        || */
7468 	    mddev->persistent	 != !info->not_persistent ||
7469 	    mddev->chunk_sectors != info->chunk_size >> 9 ||
7470 	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7471 	    ((state^info->state) & 0xfffffe00)
7472 		)
7473 		return -EINVAL;
7474 	/* Check there is only one change */
7475 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7476 		cnt++;
7477 	if (mddev->raid_disks != info->raid_disks)
7478 		cnt++;
7479 	if (mddev->layout != info->layout)
7480 		cnt++;
7481 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7482 		cnt++;
7483 	if (cnt == 0)
7484 		return 0;
7485 	if (cnt > 1)
7486 		return -EINVAL;
7487 
7488 	if (mddev->layout != info->layout) {
7489 		/* Change layout
7490 		 * we don't need to do anything at the md level, the
7491 		 * personality will take care of it all.
7492 		 */
7493 		if (mddev->pers->check_reshape == NULL)
7494 			return -EINVAL;
7495 		else {
7496 			mddev->new_layout = info->layout;
7497 			rv = mddev->pers->check_reshape(mddev);
7498 			if (rv)
7499 				mddev->new_layout = mddev->layout;
7500 			return rv;
7501 		}
7502 	}
7503 	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7504 		rv = update_size(mddev, (sector_t)info->size * 2);
7505 
7506 	if (mddev->raid_disks    != info->raid_disks)
7507 		rv = update_raid_disks(mddev, info->raid_disks);
7508 
7509 	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7510 		if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7511 			rv = -EINVAL;
7512 			goto err;
7513 		}
7514 		if (mddev->recovery || mddev->sync_thread) {
7515 			rv = -EBUSY;
7516 			goto err;
7517 		}
7518 		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7519 			/* add the bitmap */
7520 			if (mddev->bitmap) {
7521 				rv = -EEXIST;
7522 				goto err;
7523 			}
7524 			if (mddev->bitmap_info.default_offset == 0) {
7525 				rv = -EINVAL;
7526 				goto err;
7527 			}
7528 			mddev->bitmap_info.offset =
7529 				mddev->bitmap_info.default_offset;
7530 			mddev->bitmap_info.space =
7531 				mddev->bitmap_info.default_space;
7532 			rv = mddev->bitmap_ops->create(mddev, -1);
7533 			if (!rv)
7534 				rv = mddev->bitmap_ops->load(mddev);
7535 
7536 			if (rv)
7537 				mddev->bitmap_ops->destroy(mddev);
7538 		} else {
7539 			struct md_bitmap_stats stats;
7540 
7541 			rv = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
7542 			if (rv)
7543 				goto err;
7544 
7545 			if (stats.file) {
7546 				rv = -EINVAL;
7547 				goto err;
7548 			}
7549 
7550 			if (mddev->bitmap_info.nodes) {
7551 				/* hold PW on all the bitmap lock */
7552 				if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7553 					pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7554 					rv = -EPERM;
7555 					md_cluster_ops->unlock_all_bitmaps(mddev);
7556 					goto err;
7557 				}
7558 
7559 				mddev->bitmap_info.nodes = 0;
7560 				md_cluster_ops->leave(mddev);
7561 				module_put(md_cluster_mod);
7562 				mddev->safemode_delay = DEFAULT_SAFEMODE_DELAY;
7563 			}
7564 			mddev->bitmap_ops->destroy(mddev);
7565 			mddev->bitmap_info.offset = 0;
7566 		}
7567 	}
7568 	md_update_sb(mddev, 1);
7569 	return rv;
7570 err:
7571 	return rv;
7572 }
7573 
set_disk_faulty(struct mddev * mddev,dev_t dev)7574 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7575 {
7576 	struct md_rdev *rdev;
7577 	int err = 0;
7578 
7579 	if (mddev->pers == NULL)
7580 		return -ENODEV;
7581 
7582 	rcu_read_lock();
7583 	rdev = md_find_rdev_rcu(mddev, dev);
7584 	if (!rdev)
7585 		err =  -ENODEV;
7586 	else {
7587 		md_error(mddev, rdev);
7588 		if (test_bit(MD_BROKEN, &mddev->flags))
7589 			err = -EBUSY;
7590 	}
7591 	rcu_read_unlock();
7592 	return err;
7593 }
7594 
7595 /*
7596  * We have a problem here : there is no easy way to give a CHS
7597  * virtual geometry. We currently pretend that we have a 2 heads
7598  * 4 sectors (with a BIG number of cylinders...). This drives
7599  * dosfs just mad... ;-)
7600  */
md_getgeo(struct block_device * bdev,struct hd_geometry * geo)7601 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7602 {
7603 	struct mddev *mddev = bdev->bd_disk->private_data;
7604 
7605 	geo->heads = 2;
7606 	geo->sectors = 4;
7607 	geo->cylinders = mddev->array_sectors / 8;
7608 	return 0;
7609 }
7610 
md_ioctl_valid(unsigned int cmd)7611 static inline int md_ioctl_valid(unsigned int cmd)
7612 {
7613 	switch (cmd) {
7614 	case GET_ARRAY_INFO:
7615 	case GET_DISK_INFO:
7616 	case RAID_VERSION:
7617 		return 0;
7618 	case ADD_NEW_DISK:
7619 	case GET_BITMAP_FILE:
7620 	case HOT_ADD_DISK:
7621 	case HOT_REMOVE_DISK:
7622 	case RESTART_ARRAY_RW:
7623 	case RUN_ARRAY:
7624 	case SET_ARRAY_INFO:
7625 	case SET_BITMAP_FILE:
7626 	case SET_DISK_FAULTY:
7627 	case STOP_ARRAY:
7628 	case STOP_ARRAY_RO:
7629 	case CLUSTERED_DISK_NACK:
7630 		if (!capable(CAP_SYS_ADMIN))
7631 			return -EACCES;
7632 		return 0;
7633 	default:
7634 		return -ENOTTY;
7635 	}
7636 }
7637 
md_ioctl_need_suspend(unsigned int cmd)7638 static bool md_ioctl_need_suspend(unsigned int cmd)
7639 {
7640 	switch (cmd) {
7641 	case ADD_NEW_DISK:
7642 	case HOT_ADD_DISK:
7643 	case HOT_REMOVE_DISK:
7644 	case SET_BITMAP_FILE:
7645 	case SET_ARRAY_INFO:
7646 		return true;
7647 	default:
7648 		return false;
7649 	}
7650 }
7651 
__md_set_array_info(struct mddev * mddev,void __user * argp)7652 static int __md_set_array_info(struct mddev *mddev, void __user *argp)
7653 {
7654 	mdu_array_info_t info;
7655 	int err;
7656 
7657 	if (!argp)
7658 		memset(&info, 0, sizeof(info));
7659 	else if (copy_from_user(&info, argp, sizeof(info)))
7660 		return -EFAULT;
7661 
7662 	if (mddev->pers) {
7663 		err = update_array_info(mddev, &info);
7664 		if (err)
7665 			pr_warn("md: couldn't update array info. %d\n", err);
7666 		return err;
7667 	}
7668 
7669 	if (!list_empty(&mddev->disks)) {
7670 		pr_warn("md: array %s already has disks!\n", mdname(mddev));
7671 		return -EBUSY;
7672 	}
7673 
7674 	if (mddev->raid_disks) {
7675 		pr_warn("md: array %s already initialised!\n", mdname(mddev));
7676 		return -EBUSY;
7677 	}
7678 
7679 	err = md_set_array_info(mddev, &info);
7680 	if (err)
7681 		pr_warn("md: couldn't set array info. %d\n", err);
7682 
7683 	return err;
7684 }
7685 
md_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)7686 static int md_ioctl(struct block_device *bdev, blk_mode_t mode,
7687 			unsigned int cmd, unsigned long arg)
7688 {
7689 	int err = 0;
7690 	void __user *argp = (void __user *)arg;
7691 	struct mddev *mddev = NULL;
7692 
7693 	err = md_ioctl_valid(cmd);
7694 	if (err)
7695 		return err;
7696 
7697 	/*
7698 	 * Commands dealing with the RAID driver but not any
7699 	 * particular array:
7700 	 */
7701 	if (cmd == RAID_VERSION)
7702 		return get_version(argp);
7703 
7704 	/*
7705 	 * Commands creating/starting a new array:
7706 	 */
7707 
7708 	mddev = bdev->bd_disk->private_data;
7709 
7710 	/* Some actions do not requires the mutex */
7711 	switch (cmd) {
7712 	case GET_ARRAY_INFO:
7713 		if (!mddev->raid_disks && !mddev->external)
7714 			return -ENODEV;
7715 		return get_array_info(mddev, argp);
7716 
7717 	case GET_DISK_INFO:
7718 		if (!mddev->raid_disks && !mddev->external)
7719 			return -ENODEV;
7720 		return get_disk_info(mddev, argp);
7721 
7722 	case SET_DISK_FAULTY:
7723 		return set_disk_faulty(mddev, new_decode_dev(arg));
7724 
7725 	case GET_BITMAP_FILE:
7726 		return get_bitmap_file(mddev, argp);
7727 	}
7728 
7729 	if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7730 		/* Need to flush page cache, and ensure no-one else opens
7731 		 * and writes
7732 		 */
7733 		err = mddev_set_closing_and_sync_blockdev(mddev, 1);
7734 		if (err)
7735 			return err;
7736 	}
7737 
7738 	if (!md_is_rdwr(mddev))
7739 		flush_work(&mddev->sync_work);
7740 
7741 	err = md_ioctl_need_suspend(cmd) ? mddev_suspend_and_lock(mddev) :
7742 					   mddev_lock(mddev);
7743 	if (err) {
7744 		pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7745 			 err, cmd);
7746 		goto out;
7747 	}
7748 
7749 	if (cmd == SET_ARRAY_INFO) {
7750 		err = __md_set_array_info(mddev, argp);
7751 		goto unlock;
7752 	}
7753 
7754 	/*
7755 	 * Commands querying/configuring an existing array:
7756 	 */
7757 	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7758 	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7759 	if ((!mddev->raid_disks && !mddev->external)
7760 	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7761 	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7762 	    && cmd != GET_BITMAP_FILE) {
7763 		err = -ENODEV;
7764 		goto unlock;
7765 	}
7766 
7767 	/*
7768 	 * Commands even a read-only array can execute:
7769 	 */
7770 	switch (cmd) {
7771 	case RESTART_ARRAY_RW:
7772 		err = restart_array(mddev);
7773 		goto unlock;
7774 
7775 	case STOP_ARRAY:
7776 		err = do_md_stop(mddev, 0);
7777 		goto unlock;
7778 
7779 	case STOP_ARRAY_RO:
7780 		if (mddev->pers)
7781 			err = md_set_readonly(mddev);
7782 		goto unlock;
7783 
7784 	case HOT_REMOVE_DISK:
7785 		err = hot_remove_disk(mddev, new_decode_dev(arg));
7786 		goto unlock;
7787 
7788 	case ADD_NEW_DISK:
7789 		/* We can support ADD_NEW_DISK on read-only arrays
7790 		 * only if we are re-adding a preexisting device.
7791 		 * So require mddev->pers and MD_DISK_SYNC.
7792 		 */
7793 		if (mddev->pers) {
7794 			mdu_disk_info_t info;
7795 			if (copy_from_user(&info, argp, sizeof(info)))
7796 				err = -EFAULT;
7797 			else if (!(info.state & (1<<MD_DISK_SYNC)))
7798 				/* Need to clear read-only for this */
7799 				break;
7800 			else
7801 				err = md_add_new_disk(mddev, &info);
7802 			goto unlock;
7803 		}
7804 		break;
7805 	}
7806 
7807 	/*
7808 	 * The remaining ioctls are changing the state of the
7809 	 * superblock, so we do not allow them on read-only arrays.
7810 	 */
7811 	if (!md_is_rdwr(mddev) && mddev->pers) {
7812 		if (mddev->ro != MD_AUTO_READ) {
7813 			err = -EROFS;
7814 			goto unlock;
7815 		}
7816 		mddev->ro = MD_RDWR;
7817 		sysfs_notify_dirent_safe(mddev->sysfs_state);
7818 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7819 		/* mddev_unlock will wake thread */
7820 		/* If a device failed while we were read-only, we
7821 		 * need to make sure the metadata is updated now.
7822 		 */
7823 		if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7824 			mddev_unlock(mddev);
7825 			wait_event(mddev->sb_wait,
7826 				   !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7827 				   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7828 			mddev_lock_nointr(mddev);
7829 		}
7830 	}
7831 
7832 	switch (cmd) {
7833 	case ADD_NEW_DISK:
7834 	{
7835 		mdu_disk_info_t info;
7836 		if (copy_from_user(&info, argp, sizeof(info)))
7837 			err = -EFAULT;
7838 		else
7839 			err = md_add_new_disk(mddev, &info);
7840 		goto unlock;
7841 	}
7842 
7843 	case CLUSTERED_DISK_NACK:
7844 		if (mddev_is_clustered(mddev))
7845 			md_cluster_ops->new_disk_ack(mddev, false);
7846 		else
7847 			err = -EINVAL;
7848 		goto unlock;
7849 
7850 	case HOT_ADD_DISK:
7851 		err = hot_add_disk(mddev, new_decode_dev(arg));
7852 		goto unlock;
7853 
7854 	case RUN_ARRAY:
7855 		err = do_md_run(mddev);
7856 		goto unlock;
7857 
7858 	case SET_BITMAP_FILE:
7859 		err = set_bitmap_file(mddev, (int)arg);
7860 		goto unlock;
7861 
7862 	default:
7863 		err = -EINVAL;
7864 		goto unlock;
7865 	}
7866 
7867 unlock:
7868 	if (mddev->hold_active == UNTIL_IOCTL &&
7869 	    err != -EINVAL)
7870 		mddev->hold_active = 0;
7871 
7872 	md_ioctl_need_suspend(cmd) ? mddev_unlock_and_resume(mddev) :
7873 				     mddev_unlock(mddev);
7874 
7875 out:
7876 	if (cmd == STOP_ARRAY_RO || (err && cmd == STOP_ARRAY))
7877 		clear_bit(MD_CLOSING, &mddev->flags);
7878 	return err;
7879 }
7880 #ifdef CONFIG_COMPAT
md_compat_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)7881 static int md_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
7882 		    unsigned int cmd, unsigned long arg)
7883 {
7884 	switch (cmd) {
7885 	case HOT_REMOVE_DISK:
7886 	case HOT_ADD_DISK:
7887 	case SET_DISK_FAULTY:
7888 	case SET_BITMAP_FILE:
7889 		/* These take in integer arg, do not convert */
7890 		break;
7891 	default:
7892 		arg = (unsigned long)compat_ptr(arg);
7893 		break;
7894 	}
7895 
7896 	return md_ioctl(bdev, mode, cmd, arg);
7897 }
7898 #endif /* CONFIG_COMPAT */
7899 
md_set_read_only(struct block_device * bdev,bool ro)7900 static int md_set_read_only(struct block_device *bdev, bool ro)
7901 {
7902 	struct mddev *mddev = bdev->bd_disk->private_data;
7903 	int err;
7904 
7905 	err = mddev_lock(mddev);
7906 	if (err)
7907 		return err;
7908 
7909 	if (!mddev->raid_disks && !mddev->external) {
7910 		err = -ENODEV;
7911 		goto out_unlock;
7912 	}
7913 
7914 	/*
7915 	 * Transitioning to read-auto need only happen for arrays that call
7916 	 * md_write_start and which are not ready for writes yet.
7917 	 */
7918 	if (!ro && mddev->ro == MD_RDONLY && mddev->pers) {
7919 		err = restart_array(mddev);
7920 		if (err)
7921 			goto out_unlock;
7922 		mddev->ro = MD_AUTO_READ;
7923 	}
7924 
7925 out_unlock:
7926 	mddev_unlock(mddev);
7927 	return err;
7928 }
7929 
md_open(struct gendisk * disk,blk_mode_t mode)7930 static int md_open(struct gendisk *disk, blk_mode_t mode)
7931 {
7932 	struct mddev *mddev;
7933 	int err;
7934 
7935 	spin_lock(&all_mddevs_lock);
7936 	mddev = mddev_get(disk->private_data);
7937 	spin_unlock(&all_mddevs_lock);
7938 	if (!mddev)
7939 		return -ENODEV;
7940 
7941 	err = mutex_lock_interruptible(&mddev->open_mutex);
7942 	if (err)
7943 		goto out;
7944 
7945 	err = -ENODEV;
7946 	if (test_bit(MD_CLOSING, &mddev->flags))
7947 		goto out_unlock;
7948 
7949 	atomic_inc(&mddev->openers);
7950 	mutex_unlock(&mddev->open_mutex);
7951 
7952 	disk_check_media_change(disk);
7953 	return 0;
7954 
7955 out_unlock:
7956 	mutex_unlock(&mddev->open_mutex);
7957 out:
7958 	mddev_put(mddev);
7959 	return err;
7960 }
7961 
md_release(struct gendisk * disk)7962 static void md_release(struct gendisk *disk)
7963 {
7964 	struct mddev *mddev = disk->private_data;
7965 
7966 	BUG_ON(!mddev);
7967 	atomic_dec(&mddev->openers);
7968 	mddev_put(mddev);
7969 }
7970 
md_check_events(struct gendisk * disk,unsigned int clearing)7971 static unsigned int md_check_events(struct gendisk *disk, unsigned int clearing)
7972 {
7973 	struct mddev *mddev = disk->private_data;
7974 	unsigned int ret = 0;
7975 
7976 	if (mddev->changed)
7977 		ret = DISK_EVENT_MEDIA_CHANGE;
7978 	mddev->changed = 0;
7979 	return ret;
7980 }
7981 
md_free_disk(struct gendisk * disk)7982 static void md_free_disk(struct gendisk *disk)
7983 {
7984 	struct mddev *mddev = disk->private_data;
7985 
7986 	mddev_free(mddev);
7987 }
7988 
7989 const struct block_device_operations md_fops =
7990 {
7991 	.owner		= THIS_MODULE,
7992 	.submit_bio	= md_submit_bio,
7993 	.open		= md_open,
7994 	.release	= md_release,
7995 	.ioctl		= md_ioctl,
7996 #ifdef CONFIG_COMPAT
7997 	.compat_ioctl	= md_compat_ioctl,
7998 #endif
7999 	.getgeo		= md_getgeo,
8000 	.check_events	= md_check_events,
8001 	.set_read_only	= md_set_read_only,
8002 	.free_disk	= md_free_disk,
8003 };
8004 
md_thread(void * arg)8005 static int md_thread(void *arg)
8006 {
8007 	struct md_thread *thread = arg;
8008 
8009 	/*
8010 	 * md_thread is a 'system-thread', it's priority should be very
8011 	 * high. We avoid resource deadlocks individually in each
8012 	 * raid personality. (RAID5 does preallocation) We also use RR and
8013 	 * the very same RT priority as kswapd, thus we will never get
8014 	 * into a priority inversion deadlock.
8015 	 *
8016 	 * we definitely have to have equal or higher priority than
8017 	 * bdflush, otherwise bdflush will deadlock if there are too
8018 	 * many dirty RAID5 blocks.
8019 	 */
8020 
8021 	allow_signal(SIGKILL);
8022 	while (!kthread_should_stop()) {
8023 
8024 		/* We need to wait INTERRUPTIBLE so that
8025 		 * we don't add to the load-average.
8026 		 * That means we need to be sure no signals are
8027 		 * pending
8028 		 */
8029 		if (signal_pending(current))
8030 			flush_signals(current);
8031 
8032 		wait_event_interruptible_timeout
8033 			(thread->wqueue,
8034 			 test_bit(THREAD_WAKEUP, &thread->flags)
8035 			 || kthread_should_stop() || kthread_should_park(),
8036 			 thread->timeout);
8037 
8038 		clear_bit(THREAD_WAKEUP, &thread->flags);
8039 		if (kthread_should_park())
8040 			kthread_parkme();
8041 		if (!kthread_should_stop())
8042 			thread->run(thread);
8043 	}
8044 
8045 	return 0;
8046 }
8047 
md_wakeup_thread_directly(struct md_thread __rcu * thread)8048 static void md_wakeup_thread_directly(struct md_thread __rcu *thread)
8049 {
8050 	struct md_thread *t;
8051 
8052 	rcu_read_lock();
8053 	t = rcu_dereference(thread);
8054 	if (t)
8055 		wake_up_process(t->tsk);
8056 	rcu_read_unlock();
8057 }
8058 
md_wakeup_thread(struct md_thread __rcu * thread)8059 void md_wakeup_thread(struct md_thread __rcu *thread)
8060 {
8061 	struct md_thread *t;
8062 
8063 	rcu_read_lock();
8064 	t = rcu_dereference(thread);
8065 	if (t) {
8066 		pr_debug("md: waking up MD thread %s.\n", t->tsk->comm);
8067 		set_bit(THREAD_WAKEUP, &t->flags);
8068 		if (wq_has_sleeper(&t->wqueue))
8069 			wake_up(&t->wqueue);
8070 	}
8071 	rcu_read_unlock();
8072 }
8073 EXPORT_SYMBOL(md_wakeup_thread);
8074 
md_register_thread(void (* run)(struct md_thread *),struct mddev * mddev,const char * name)8075 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
8076 		struct mddev *mddev, const char *name)
8077 {
8078 	struct md_thread *thread;
8079 
8080 	thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
8081 	if (!thread)
8082 		return NULL;
8083 
8084 	init_waitqueue_head(&thread->wqueue);
8085 
8086 	thread->run = run;
8087 	thread->mddev = mddev;
8088 	thread->timeout = MAX_SCHEDULE_TIMEOUT;
8089 	thread->tsk = kthread_run(md_thread, thread,
8090 				  "%s_%s",
8091 				  mdname(thread->mddev),
8092 				  name);
8093 	if (IS_ERR(thread->tsk)) {
8094 		kfree(thread);
8095 		return NULL;
8096 	}
8097 	return thread;
8098 }
8099 EXPORT_SYMBOL(md_register_thread);
8100 
md_unregister_thread(struct mddev * mddev,struct md_thread __rcu ** threadp)8101 void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp)
8102 {
8103 	struct md_thread *thread = rcu_dereference_protected(*threadp,
8104 					lockdep_is_held(&mddev->reconfig_mutex));
8105 
8106 	if (!thread)
8107 		return;
8108 
8109 	rcu_assign_pointer(*threadp, NULL);
8110 	synchronize_rcu();
8111 
8112 	pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
8113 	kthread_stop(thread->tsk);
8114 	kfree(thread);
8115 }
8116 EXPORT_SYMBOL(md_unregister_thread);
8117 
md_error(struct mddev * mddev,struct md_rdev * rdev)8118 void md_error(struct mddev *mddev, struct md_rdev *rdev)
8119 {
8120 	if (!rdev || test_bit(Faulty, &rdev->flags))
8121 		return;
8122 
8123 	if (!mddev->pers || !mddev->pers->error_handler)
8124 		return;
8125 	mddev->pers->error_handler(mddev, rdev);
8126 
8127 	if (mddev->pers->level == 0)
8128 		return;
8129 
8130 	if (mddev->degraded && !test_bit(MD_BROKEN, &mddev->flags))
8131 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8132 	sysfs_notify_dirent_safe(rdev->sysfs_state);
8133 	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8134 	if (!test_bit(MD_BROKEN, &mddev->flags)) {
8135 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8136 		md_wakeup_thread(mddev->thread);
8137 	}
8138 	if (mddev->event_work.func)
8139 		queue_work(md_misc_wq, &mddev->event_work);
8140 	md_new_event();
8141 }
8142 EXPORT_SYMBOL(md_error);
8143 
8144 /* seq_file implementation /proc/mdstat */
8145 
status_unused(struct seq_file * seq)8146 static void status_unused(struct seq_file *seq)
8147 {
8148 	int i = 0;
8149 	struct md_rdev *rdev;
8150 
8151 	seq_printf(seq, "unused devices: ");
8152 
8153 	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
8154 		i++;
8155 		seq_printf(seq, "%pg ", rdev->bdev);
8156 	}
8157 	if (!i)
8158 		seq_printf(seq, "<none>");
8159 
8160 	seq_printf(seq, "\n");
8161 }
8162 
status_personalities(struct seq_file * seq)8163 static void status_personalities(struct seq_file *seq)
8164 {
8165 	struct md_personality *pers;
8166 
8167 	seq_puts(seq, "Personalities : ");
8168 	spin_lock(&pers_lock);
8169 	list_for_each_entry(pers, &pers_list, list)
8170 		seq_printf(seq, "[%s] ", pers->name);
8171 
8172 	spin_unlock(&pers_lock);
8173 	seq_puts(seq, "\n");
8174 }
8175 
status_resync(struct seq_file * seq,struct mddev * mddev)8176 static int status_resync(struct seq_file *seq, struct mddev *mddev)
8177 {
8178 	sector_t max_sectors, resync, res;
8179 	unsigned long dt, db = 0;
8180 	sector_t rt, curr_mark_cnt, resync_mark_cnt;
8181 	int scale, recovery_active;
8182 	unsigned int per_milli;
8183 
8184 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8185 	    test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8186 		max_sectors = mddev->resync_max_sectors;
8187 	else
8188 		max_sectors = mddev->dev_sectors;
8189 
8190 	resync = mddev->curr_resync;
8191 	if (resync < MD_RESYNC_ACTIVE) {
8192 		if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8193 			/* Still cleaning up */
8194 			resync = max_sectors;
8195 	} else if (resync > max_sectors) {
8196 		resync = max_sectors;
8197 	} else {
8198 		res = atomic_read(&mddev->recovery_active);
8199 		/*
8200 		 * Resync has started, but the subtraction has overflowed or
8201 		 * yielded one of the special values. Force it to active to
8202 		 * ensure the status reports an active resync.
8203 		 */
8204 		if (resync < res || resync - res < MD_RESYNC_ACTIVE)
8205 			resync = MD_RESYNC_ACTIVE;
8206 		else
8207 			resync -= res;
8208 	}
8209 
8210 	if (resync == MD_RESYNC_NONE) {
8211 		if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
8212 			struct md_rdev *rdev;
8213 
8214 			rdev_for_each(rdev, mddev)
8215 				if (rdev->raid_disk >= 0 &&
8216 				    !test_bit(Faulty, &rdev->flags) &&
8217 				    rdev->recovery_offset != MaxSector &&
8218 				    rdev->recovery_offset) {
8219 					seq_printf(seq, "\trecover=REMOTE");
8220 					return 1;
8221 				}
8222 			if (mddev->reshape_position != MaxSector)
8223 				seq_printf(seq, "\treshape=REMOTE");
8224 			else
8225 				seq_printf(seq, "\tresync=REMOTE");
8226 			return 1;
8227 		}
8228 		if (mddev->recovery_cp < MaxSector) {
8229 			seq_printf(seq, "\tresync=PENDING");
8230 			return 1;
8231 		}
8232 		return 0;
8233 	}
8234 	if (resync < MD_RESYNC_ACTIVE) {
8235 		seq_printf(seq, "\tresync=DELAYED");
8236 		return 1;
8237 	}
8238 
8239 	WARN_ON(max_sectors == 0);
8240 	/* Pick 'scale' such that (resync>>scale)*1000 will fit
8241 	 * in a sector_t, and (max_sectors>>scale) will fit in a
8242 	 * u32, as those are the requirements for sector_div.
8243 	 * Thus 'scale' must be at least 10
8244 	 */
8245 	scale = 10;
8246 	if (sizeof(sector_t) > sizeof(unsigned long)) {
8247 		while ( max_sectors/2 > (1ULL<<(scale+32)))
8248 			scale++;
8249 	}
8250 	res = (resync>>scale)*1000;
8251 	sector_div(res, (u32)((max_sectors>>scale)+1));
8252 
8253 	per_milli = res;
8254 	{
8255 		int i, x = per_milli/50, y = 20-x;
8256 		seq_printf(seq, "[");
8257 		for (i = 0; i < x; i++)
8258 			seq_printf(seq, "=");
8259 		seq_printf(seq, ">");
8260 		for (i = 0; i < y; i++)
8261 			seq_printf(seq, ".");
8262 		seq_printf(seq, "] ");
8263 	}
8264 	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
8265 		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
8266 		    "reshape" :
8267 		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
8268 		     "check" :
8269 		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
8270 		      "resync" : "recovery"))),
8271 		   per_milli/10, per_milli % 10,
8272 		   (unsigned long long) resync/2,
8273 		   (unsigned long long) max_sectors/2);
8274 
8275 	/*
8276 	 * dt: time from mark until now
8277 	 * db: blocks written from mark until now
8278 	 * rt: remaining time
8279 	 *
8280 	 * rt is a sector_t, which is always 64bit now. We are keeping
8281 	 * the original algorithm, but it is not really necessary.
8282 	 *
8283 	 * Original algorithm:
8284 	 *   So we divide before multiply in case it is 32bit and close
8285 	 *   to the limit.
8286 	 *   We scale the divisor (db) by 32 to avoid losing precision
8287 	 *   near the end of resync when the number of remaining sectors
8288 	 *   is close to 'db'.
8289 	 *   We then divide rt by 32 after multiplying by db to compensate.
8290 	 *   The '+1' avoids division by zero if db is very small.
8291 	 */
8292 	dt = ((jiffies - mddev->resync_mark) / HZ);
8293 	if (!dt) dt++;
8294 
8295 	curr_mark_cnt = mddev->curr_mark_cnt;
8296 	recovery_active = atomic_read(&mddev->recovery_active);
8297 	resync_mark_cnt = mddev->resync_mark_cnt;
8298 
8299 	if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
8300 		db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
8301 
8302 	rt = max_sectors - resync;    /* number of remaining sectors */
8303 	rt = div64_u64(rt, db/32+1);
8304 	rt *= dt;
8305 	rt >>= 5;
8306 
8307 	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
8308 		   ((unsigned long)rt % 60)/6);
8309 
8310 	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
8311 	return 1;
8312 }
8313 
md_seq_start(struct seq_file * seq,loff_t * pos)8314 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
8315 	__acquires(&all_mddevs_lock)
8316 {
8317 	seq->poll_event = atomic_read(&md_event_count);
8318 	spin_lock(&all_mddevs_lock);
8319 
8320 	return seq_list_start_head(&all_mddevs, *pos);
8321 }
8322 
md_seq_next(struct seq_file * seq,void * v,loff_t * pos)8323 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
8324 {
8325 	return seq_list_next(v, &all_mddevs, pos);
8326 }
8327 
md_seq_stop(struct seq_file * seq,void * v)8328 static void md_seq_stop(struct seq_file *seq, void *v)
8329 	__releases(&all_mddevs_lock)
8330 {
8331 	spin_unlock(&all_mddevs_lock);
8332 }
8333 
md_bitmap_status(struct seq_file * seq,struct mddev * mddev)8334 static void md_bitmap_status(struct seq_file *seq, struct mddev *mddev)
8335 {
8336 	struct md_bitmap_stats stats;
8337 	unsigned long used_pages;
8338 	unsigned long chunk_kb;
8339 	int err;
8340 
8341 	err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats);
8342 	if (err)
8343 		return;
8344 
8345 	chunk_kb = mddev->bitmap_info.chunksize >> 10;
8346 	used_pages = stats.pages - stats.missing_pages;
8347 
8348 	seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], %lu%s chunk",
8349 		   used_pages, stats.pages, used_pages << (PAGE_SHIFT - 10),
8350 		   chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
8351 		   chunk_kb ? "KB" : "B");
8352 
8353 	if (stats.file) {
8354 		seq_puts(seq, ", file: ");
8355 		seq_file_path(seq, stats.file, " \t\n");
8356 	}
8357 
8358 	seq_putc(seq, '\n');
8359 }
8360 
md_seq_show(struct seq_file * seq,void * v)8361 static int md_seq_show(struct seq_file *seq, void *v)
8362 {
8363 	struct mddev *mddev;
8364 	sector_t sectors;
8365 	struct md_rdev *rdev;
8366 
8367 	if (v == &all_mddevs) {
8368 		status_personalities(seq);
8369 		if (list_empty(&all_mddevs))
8370 			status_unused(seq);
8371 		return 0;
8372 	}
8373 
8374 	mddev = list_entry(v, struct mddev, all_mddevs);
8375 	if (!mddev_get(mddev))
8376 		return 0;
8377 
8378 	spin_unlock(&all_mddevs_lock);
8379 	spin_lock(&mddev->lock);
8380 	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8381 		seq_printf(seq, "%s : ", mdname(mddev));
8382 		if (mddev->pers) {
8383 			if (test_bit(MD_BROKEN, &mddev->flags))
8384 				seq_printf(seq, "broken");
8385 			else
8386 				seq_printf(seq, "active");
8387 			if (mddev->ro == MD_RDONLY)
8388 				seq_printf(seq, " (read-only)");
8389 			if (mddev->ro == MD_AUTO_READ)
8390 				seq_printf(seq, " (auto-read-only)");
8391 			seq_printf(seq, " %s", mddev->pers->name);
8392 		} else {
8393 			seq_printf(seq, "inactive");
8394 		}
8395 
8396 		sectors = 0;
8397 		rcu_read_lock();
8398 		rdev_for_each_rcu(rdev, mddev) {
8399 			seq_printf(seq, " %pg[%d]", rdev->bdev, rdev->desc_nr);
8400 
8401 			if (test_bit(WriteMostly, &rdev->flags))
8402 				seq_printf(seq, "(W)");
8403 			if (test_bit(Journal, &rdev->flags))
8404 				seq_printf(seq, "(J)");
8405 			if (test_bit(Faulty, &rdev->flags)) {
8406 				seq_printf(seq, "(F)");
8407 				continue;
8408 			}
8409 			if (rdev->raid_disk < 0)
8410 				seq_printf(seq, "(S)"); /* spare */
8411 			if (test_bit(Replacement, &rdev->flags))
8412 				seq_printf(seq, "(R)");
8413 			sectors += rdev->sectors;
8414 		}
8415 		rcu_read_unlock();
8416 
8417 		if (!list_empty(&mddev->disks)) {
8418 			if (mddev->pers)
8419 				seq_printf(seq, "\n      %llu blocks",
8420 					   (unsigned long long)
8421 					   mddev->array_sectors / 2);
8422 			else
8423 				seq_printf(seq, "\n      %llu blocks",
8424 					   (unsigned long long)sectors / 2);
8425 		}
8426 		if (mddev->persistent) {
8427 			if (mddev->major_version != 0 ||
8428 			    mddev->minor_version != 90) {
8429 				seq_printf(seq," super %d.%d",
8430 					   mddev->major_version,
8431 					   mddev->minor_version);
8432 			}
8433 		} else if (mddev->external)
8434 			seq_printf(seq, " super external:%s",
8435 				   mddev->metadata_type);
8436 		else
8437 			seq_printf(seq, " super non-persistent");
8438 
8439 		if (mddev->pers) {
8440 			mddev->pers->status(seq, mddev);
8441 			seq_printf(seq, "\n      ");
8442 			if (mddev->pers->sync_request) {
8443 				if (status_resync(seq, mddev))
8444 					seq_printf(seq, "\n      ");
8445 			}
8446 		} else
8447 			seq_printf(seq, "\n       ");
8448 
8449 		md_bitmap_status(seq, mddev);
8450 
8451 		seq_printf(seq, "\n");
8452 	}
8453 	spin_unlock(&mddev->lock);
8454 	spin_lock(&all_mddevs_lock);
8455 
8456 	if (mddev == list_last_entry(&all_mddevs, struct mddev, all_mddevs))
8457 		status_unused(seq);
8458 
8459 	if (atomic_dec_and_test(&mddev->active))
8460 		__mddev_put(mddev);
8461 
8462 	return 0;
8463 }
8464 
8465 static const struct seq_operations md_seq_ops = {
8466 	.start  = md_seq_start,
8467 	.next   = md_seq_next,
8468 	.stop   = md_seq_stop,
8469 	.show   = md_seq_show,
8470 };
8471 
md_seq_open(struct inode * inode,struct file * file)8472 static int md_seq_open(struct inode *inode, struct file *file)
8473 {
8474 	struct seq_file *seq;
8475 	int error;
8476 
8477 	error = seq_open(file, &md_seq_ops);
8478 	if (error)
8479 		return error;
8480 
8481 	seq = file->private_data;
8482 	seq->poll_event = atomic_read(&md_event_count);
8483 	return error;
8484 }
8485 
8486 static int md_unloading;
mdstat_poll(struct file * filp,poll_table * wait)8487 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8488 {
8489 	struct seq_file *seq = filp->private_data;
8490 	__poll_t mask;
8491 
8492 	if (md_unloading)
8493 		return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8494 	poll_wait(filp, &md_event_waiters, wait);
8495 
8496 	/* always allow read */
8497 	mask = EPOLLIN | EPOLLRDNORM;
8498 
8499 	if (seq->poll_event != atomic_read(&md_event_count))
8500 		mask |= EPOLLERR | EPOLLPRI;
8501 	return mask;
8502 }
8503 
8504 static const struct proc_ops mdstat_proc_ops = {
8505 	.proc_open	= md_seq_open,
8506 	.proc_read	= seq_read,
8507 	.proc_lseek	= seq_lseek,
8508 	.proc_release	= seq_release,
8509 	.proc_poll	= mdstat_poll,
8510 };
8511 
register_md_personality(struct md_personality * p)8512 int register_md_personality(struct md_personality *p)
8513 {
8514 	pr_debug("md: %s personality registered for level %d\n",
8515 		 p->name, p->level);
8516 	spin_lock(&pers_lock);
8517 	list_add_tail(&p->list, &pers_list);
8518 	spin_unlock(&pers_lock);
8519 	return 0;
8520 }
8521 EXPORT_SYMBOL(register_md_personality);
8522 
unregister_md_personality(struct md_personality * p)8523 int unregister_md_personality(struct md_personality *p)
8524 {
8525 	pr_debug("md: %s personality unregistered\n", p->name);
8526 	spin_lock(&pers_lock);
8527 	list_del_init(&p->list);
8528 	spin_unlock(&pers_lock);
8529 	return 0;
8530 }
8531 EXPORT_SYMBOL(unregister_md_personality);
8532 
register_md_cluster_operations(const struct md_cluster_operations * ops,struct module * module)8533 int register_md_cluster_operations(const struct md_cluster_operations *ops,
8534 				   struct module *module)
8535 {
8536 	int ret = 0;
8537 	spin_lock(&pers_lock);
8538 	if (md_cluster_ops != NULL)
8539 		ret = -EALREADY;
8540 	else {
8541 		md_cluster_ops = ops;
8542 		md_cluster_mod = module;
8543 	}
8544 	spin_unlock(&pers_lock);
8545 	return ret;
8546 }
8547 EXPORT_SYMBOL(register_md_cluster_operations);
8548 
unregister_md_cluster_operations(void)8549 int unregister_md_cluster_operations(void)
8550 {
8551 	spin_lock(&pers_lock);
8552 	md_cluster_ops = NULL;
8553 	spin_unlock(&pers_lock);
8554 	return 0;
8555 }
8556 EXPORT_SYMBOL(unregister_md_cluster_operations);
8557 
md_setup_cluster(struct mddev * mddev,int nodes)8558 int md_setup_cluster(struct mddev *mddev, int nodes)
8559 {
8560 	int ret;
8561 	if (!md_cluster_ops)
8562 		request_module("md-cluster");
8563 	spin_lock(&pers_lock);
8564 	/* ensure module won't be unloaded */
8565 	if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8566 		pr_warn("can't find md-cluster module or get its reference.\n");
8567 		spin_unlock(&pers_lock);
8568 		return -ENOENT;
8569 	}
8570 	spin_unlock(&pers_lock);
8571 
8572 	ret = md_cluster_ops->join(mddev, nodes);
8573 	if (!ret)
8574 		mddev->safemode_delay = 0;
8575 	return ret;
8576 }
8577 
md_cluster_stop(struct mddev * mddev)8578 void md_cluster_stop(struct mddev *mddev)
8579 {
8580 	if (!md_cluster_ops)
8581 		return;
8582 	md_cluster_ops->leave(mddev);
8583 	module_put(md_cluster_mod);
8584 }
8585 
is_mddev_idle(struct mddev * mddev,int init)8586 static int is_mddev_idle(struct mddev *mddev, int init)
8587 {
8588 	struct md_rdev *rdev;
8589 	int idle;
8590 	int curr_events;
8591 
8592 	idle = 1;
8593 	rcu_read_lock();
8594 	rdev_for_each_rcu(rdev, mddev) {
8595 		struct gendisk *disk = rdev->bdev->bd_disk;
8596 
8597 		if (!init && !blk_queue_io_stat(disk->queue))
8598 			continue;
8599 
8600 		curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
8601 			      atomic_read(&disk->sync_io);
8602 		/* sync IO will cause sync_io to increase before the disk_stats
8603 		 * as sync_io is counted when a request starts, and
8604 		 * disk_stats is counted when it completes.
8605 		 * So resync activity will cause curr_events to be smaller than
8606 		 * when there was no such activity.
8607 		 * non-sync IO will cause disk_stat to increase without
8608 		 * increasing sync_io so curr_events will (eventually)
8609 		 * be larger than it was before.  Once it becomes
8610 		 * substantially larger, the test below will cause
8611 		 * the array to appear non-idle, and resync will slow
8612 		 * down.
8613 		 * If there is a lot of outstanding resync activity when
8614 		 * we set last_event to curr_events, then all that activity
8615 		 * completing might cause the array to appear non-idle
8616 		 * and resync will be slowed down even though there might
8617 		 * not have been non-resync activity.  This will only
8618 		 * happen once though.  'last_events' will soon reflect
8619 		 * the state where there is little or no outstanding
8620 		 * resync requests, and further resync activity will
8621 		 * always make curr_events less than last_events.
8622 		 *
8623 		 */
8624 		if (init || curr_events - rdev->last_events > 64) {
8625 			rdev->last_events = curr_events;
8626 			idle = 0;
8627 		}
8628 	}
8629 	rcu_read_unlock();
8630 	return idle;
8631 }
8632 
md_done_sync(struct mddev * mddev,int blocks,int ok)8633 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8634 {
8635 	/* another "blocks" (512byte) blocks have been synced */
8636 	atomic_sub(blocks, &mddev->recovery_active);
8637 	wake_up(&mddev->recovery_wait);
8638 	if (!ok) {
8639 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8640 		set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8641 		md_wakeup_thread(mddev->thread);
8642 		// stop recovery, signal do_sync ....
8643 	}
8644 }
8645 EXPORT_SYMBOL(md_done_sync);
8646 
8647 /* md_write_start(mddev, bi)
8648  * If we need to update some array metadata (e.g. 'active' flag
8649  * in superblock) before writing, schedule a superblock update
8650  * and wait for it to complete.
8651  * A return value of 'false' means that the write wasn't recorded
8652  * and cannot proceed as the array is being suspend.
8653  */
md_write_start(struct mddev * mddev,struct bio * bi)8654 void md_write_start(struct mddev *mddev, struct bio *bi)
8655 {
8656 	int did_change = 0;
8657 
8658 	if (bio_data_dir(bi) != WRITE)
8659 		return;
8660 
8661 	BUG_ON(mddev->ro == MD_RDONLY);
8662 	if (mddev->ro == MD_AUTO_READ) {
8663 		/* need to switch to read/write */
8664 		mddev->ro = MD_RDWR;
8665 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8666 		md_wakeup_thread(mddev->thread);
8667 		md_wakeup_thread(mddev->sync_thread);
8668 		did_change = 1;
8669 	}
8670 	rcu_read_lock();
8671 	percpu_ref_get(&mddev->writes_pending);
8672 	smp_mb(); /* Match smp_mb in set_in_sync() */
8673 	if (mddev->safemode == 1)
8674 		mddev->safemode = 0;
8675 	/* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8676 	if (mddev->in_sync || mddev->sync_checkers) {
8677 		spin_lock(&mddev->lock);
8678 		if (mddev->in_sync) {
8679 			mddev->in_sync = 0;
8680 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8681 			set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8682 			md_wakeup_thread(mddev->thread);
8683 			did_change = 1;
8684 		}
8685 		spin_unlock(&mddev->lock);
8686 	}
8687 	rcu_read_unlock();
8688 	if (did_change)
8689 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8690 	if (!mddev->has_superblocks)
8691 		return;
8692 	wait_event(mddev->sb_wait,
8693 		   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8694 }
8695 EXPORT_SYMBOL(md_write_start);
8696 
8697 /* md_write_inc can only be called when md_write_start() has
8698  * already been called at least once of the current request.
8699  * It increments the counter and is useful when a single request
8700  * is split into several parts.  Each part causes an increment and
8701  * so needs a matching md_write_end().
8702  * Unlike md_write_start(), it is safe to call md_write_inc() inside
8703  * a spinlocked region.
8704  */
md_write_inc(struct mddev * mddev,struct bio * bi)8705 void md_write_inc(struct mddev *mddev, struct bio *bi)
8706 {
8707 	if (bio_data_dir(bi) != WRITE)
8708 		return;
8709 	WARN_ON_ONCE(mddev->in_sync || !md_is_rdwr(mddev));
8710 	percpu_ref_get(&mddev->writes_pending);
8711 }
8712 EXPORT_SYMBOL(md_write_inc);
8713 
md_write_end(struct mddev * mddev)8714 void md_write_end(struct mddev *mddev)
8715 {
8716 	percpu_ref_put(&mddev->writes_pending);
8717 
8718 	if (mddev->safemode == 2)
8719 		md_wakeup_thread(mddev->thread);
8720 	else if (mddev->safemode_delay)
8721 		/* The roundup() ensures this only performs locking once
8722 		 * every ->safemode_delay jiffies
8723 		 */
8724 		mod_timer(&mddev->safemode_timer,
8725 			  roundup(jiffies, mddev->safemode_delay) +
8726 			  mddev->safemode_delay);
8727 }
8728 
8729 EXPORT_SYMBOL(md_write_end);
8730 
8731 /* 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)8732 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
8733 			struct bio *bio, sector_t start, sector_t size)
8734 {
8735 	struct bio *discard_bio = NULL;
8736 
8737 	if (__blkdev_issue_discard(rdev->bdev, start, size, GFP_NOIO,
8738 			&discard_bio) || !discard_bio)
8739 		return;
8740 
8741 	bio_chain(discard_bio, bio);
8742 	bio_clone_blkg_association(discard_bio, bio);
8743 	mddev_trace_remap(mddev, discard_bio, bio->bi_iter.bi_sector);
8744 	submit_bio_noacct(discard_bio);
8745 }
8746 EXPORT_SYMBOL_GPL(md_submit_discard_bio);
8747 
md_end_clone_io(struct bio * bio)8748 static void md_end_clone_io(struct bio *bio)
8749 {
8750 	struct md_io_clone *md_io_clone = bio->bi_private;
8751 	struct bio *orig_bio = md_io_clone->orig_bio;
8752 	struct mddev *mddev = md_io_clone->mddev;
8753 
8754 	if (bio->bi_status && !orig_bio->bi_status)
8755 		orig_bio->bi_status = bio->bi_status;
8756 
8757 	if (md_io_clone->start_time)
8758 		bio_end_io_acct(orig_bio, md_io_clone->start_time);
8759 
8760 	bio_put(bio);
8761 	bio_endio(orig_bio);
8762 	percpu_ref_put(&mddev->active_io);
8763 }
8764 
md_clone_bio(struct mddev * mddev,struct bio ** bio)8765 static void md_clone_bio(struct mddev *mddev, struct bio **bio)
8766 {
8767 	struct block_device *bdev = (*bio)->bi_bdev;
8768 	struct md_io_clone *md_io_clone;
8769 	struct bio *clone =
8770 		bio_alloc_clone(bdev, *bio, GFP_NOIO, &mddev->io_clone_set);
8771 
8772 	md_io_clone = container_of(clone, struct md_io_clone, bio_clone);
8773 	md_io_clone->orig_bio = *bio;
8774 	md_io_clone->mddev = mddev;
8775 	if (blk_queue_io_stat(bdev->bd_disk->queue))
8776 		md_io_clone->start_time = bio_start_io_acct(*bio);
8777 
8778 	clone->bi_end_io = md_end_clone_io;
8779 	clone->bi_private = md_io_clone;
8780 	*bio = clone;
8781 }
8782 
md_account_bio(struct mddev * mddev,struct bio ** bio)8783 void md_account_bio(struct mddev *mddev, struct bio **bio)
8784 {
8785 	percpu_ref_get(&mddev->active_io);
8786 	md_clone_bio(mddev, bio);
8787 }
8788 EXPORT_SYMBOL_GPL(md_account_bio);
8789 
md_free_cloned_bio(struct bio * bio)8790 void md_free_cloned_bio(struct bio *bio)
8791 {
8792 	struct md_io_clone *md_io_clone = bio->bi_private;
8793 	struct bio *orig_bio = md_io_clone->orig_bio;
8794 	struct mddev *mddev = md_io_clone->mddev;
8795 
8796 	if (bio->bi_status && !orig_bio->bi_status)
8797 		orig_bio->bi_status = bio->bi_status;
8798 
8799 	if (md_io_clone->start_time)
8800 		bio_end_io_acct(orig_bio, md_io_clone->start_time);
8801 
8802 	bio_put(bio);
8803 	percpu_ref_put(&mddev->active_io);
8804 }
8805 EXPORT_SYMBOL_GPL(md_free_cloned_bio);
8806 
8807 /* md_allow_write(mddev)
8808  * Calling this ensures that the array is marked 'active' so that writes
8809  * may proceed without blocking.  It is important to call this before
8810  * attempting a GFP_KERNEL allocation while holding the mddev lock.
8811  * Must be called with mddev_lock held.
8812  */
md_allow_write(struct mddev * mddev)8813 void md_allow_write(struct mddev *mddev)
8814 {
8815 	if (!mddev->pers)
8816 		return;
8817 	if (!md_is_rdwr(mddev))
8818 		return;
8819 	if (!mddev->pers->sync_request)
8820 		return;
8821 
8822 	spin_lock(&mddev->lock);
8823 	if (mddev->in_sync) {
8824 		mddev->in_sync = 0;
8825 		set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8826 		set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8827 		if (mddev->safemode_delay &&
8828 		    mddev->safemode == 0)
8829 			mddev->safemode = 1;
8830 		spin_unlock(&mddev->lock);
8831 		md_update_sb(mddev, 0);
8832 		sysfs_notify_dirent_safe(mddev->sysfs_state);
8833 		/* wait for the dirty state to be recorded in the metadata */
8834 		wait_event(mddev->sb_wait,
8835 			   !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8836 	} else
8837 		spin_unlock(&mddev->lock);
8838 }
8839 EXPORT_SYMBOL_GPL(md_allow_write);
8840 
md_sync_max_sectors(struct mddev * mddev,enum sync_action action)8841 static sector_t md_sync_max_sectors(struct mddev *mddev,
8842 				    enum sync_action action)
8843 {
8844 	switch (action) {
8845 	case ACTION_RESYNC:
8846 	case ACTION_CHECK:
8847 	case ACTION_REPAIR:
8848 		atomic64_set(&mddev->resync_mismatches, 0);
8849 		fallthrough;
8850 	case ACTION_RESHAPE:
8851 		return mddev->resync_max_sectors;
8852 	case ACTION_RECOVER:
8853 		return mddev->dev_sectors;
8854 	default:
8855 		return 0;
8856 	}
8857 }
8858 
md_sync_position(struct mddev * mddev,enum sync_action action)8859 static sector_t md_sync_position(struct mddev *mddev, enum sync_action action)
8860 {
8861 	sector_t start = 0;
8862 	struct md_rdev *rdev;
8863 
8864 	switch (action) {
8865 	case ACTION_CHECK:
8866 	case ACTION_REPAIR:
8867 		return mddev->resync_min;
8868 	case ACTION_RESYNC:
8869 		if (!mddev->bitmap)
8870 			return mddev->recovery_cp;
8871 		return 0;
8872 	case ACTION_RESHAPE:
8873 		/*
8874 		 * If the original node aborts reshaping then we continue the
8875 		 * reshaping, so set again to avoid restart reshape from the
8876 		 * first beginning
8877 		 */
8878 		if (mddev_is_clustered(mddev) &&
8879 		    mddev->reshape_position != MaxSector)
8880 			return mddev->reshape_position;
8881 		return 0;
8882 	case ACTION_RECOVER:
8883 		start = MaxSector;
8884 		rcu_read_lock();
8885 		rdev_for_each_rcu(rdev, mddev)
8886 			if (rdev->raid_disk >= 0 &&
8887 			    !test_bit(Journal, &rdev->flags) &&
8888 			    !test_bit(Faulty, &rdev->flags) &&
8889 			    !test_bit(In_sync, &rdev->flags) &&
8890 			    rdev->recovery_offset < start)
8891 				start = rdev->recovery_offset;
8892 		rcu_read_unlock();
8893 
8894 		/* If there is a bitmap, we need to make sure all
8895 		 * writes that started before we added a spare
8896 		 * complete before we start doing a recovery.
8897 		 * Otherwise the write might complete and (via
8898 		 * bitmap_endwrite) set a bit in the bitmap after the
8899 		 * recovery has checked that bit and skipped that
8900 		 * region.
8901 		 */
8902 		if (mddev->bitmap) {
8903 			mddev->pers->quiesce(mddev, 1);
8904 			mddev->pers->quiesce(mddev, 0);
8905 		}
8906 		return start;
8907 	default:
8908 		return MaxSector;
8909 	}
8910 }
8911 
8912 #define SYNC_MARKS	10
8913 #define	SYNC_MARK_STEP	(3*HZ)
8914 #define UPDATE_FREQUENCY (5*60*HZ)
md_do_sync(struct md_thread * thread)8915 void md_do_sync(struct md_thread *thread)
8916 {
8917 	struct mddev *mddev = thread->mddev;
8918 	struct mddev *mddev2;
8919 	unsigned int currspeed = 0, window;
8920 	sector_t max_sectors,j, io_sectors, recovery_done;
8921 	unsigned long mark[SYNC_MARKS];
8922 	unsigned long update_time;
8923 	sector_t mark_cnt[SYNC_MARKS];
8924 	int last_mark,m;
8925 	sector_t last_check;
8926 	int skipped = 0;
8927 	struct md_rdev *rdev;
8928 	enum sync_action action;
8929 	const char *desc;
8930 	struct blk_plug plug;
8931 	int ret;
8932 
8933 	/* just incase thread restarts... */
8934 	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8935 		return;
8936 
8937 	if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8938 		goto skip;
8939 
8940 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
8941 	    !md_is_rdwr(mddev)) {/* never try to sync a read-only array */
8942 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8943 		goto skip;
8944 	}
8945 
8946 	if (mddev_is_clustered(mddev)) {
8947 		ret = md_cluster_ops->resync_start(mddev);
8948 		if (ret)
8949 			goto skip;
8950 
8951 		set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8952 		if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8953 			test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8954 			test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8955 		     && ((unsigned long long)mddev->curr_resync_completed
8956 			 < (unsigned long long)mddev->resync_max_sectors))
8957 			goto skip;
8958 	}
8959 
8960 	action = md_sync_action(mddev);
8961 	desc = md_sync_action_name(action);
8962 	mddev->last_sync_action = action;
8963 
8964 	/*
8965 	 * Before starting a resync we must have set curr_resync to
8966 	 * 2, and then checked that every "conflicting" array has curr_resync
8967 	 * less than ours.  When we find one that is the same or higher
8968 	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
8969 	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8970 	 * This will mean we have to start checking from the beginning again.
8971 	 *
8972 	 */
8973 	if (mddev_is_clustered(mddev))
8974 		md_cluster_ops->resync_start_notify(mddev);
8975 	do {
8976 		int mddev2_minor = -1;
8977 		mddev->curr_resync = MD_RESYNC_DELAYED;
8978 
8979 	try_again:
8980 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8981 			goto skip;
8982 		spin_lock(&all_mddevs_lock);
8983 		list_for_each_entry(mddev2, &all_mddevs, all_mddevs) {
8984 			if (test_bit(MD_DELETED, &mddev2->flags))
8985 				continue;
8986 			if (mddev2 == mddev)
8987 				continue;
8988 			if (!mddev->parallel_resync
8989 			&&  mddev2->curr_resync
8990 			&&  match_mddev_units(mddev, mddev2)) {
8991 				DEFINE_WAIT(wq);
8992 				if (mddev < mddev2 &&
8993 				    mddev->curr_resync == MD_RESYNC_DELAYED) {
8994 					/* arbitrarily yield */
8995 					mddev->curr_resync = MD_RESYNC_YIELDED;
8996 					wake_up(&resync_wait);
8997 				}
8998 				if (mddev > mddev2 &&
8999 				    mddev->curr_resync == MD_RESYNC_YIELDED)
9000 					/* no need to wait here, we can wait the next
9001 					 * time 'round when curr_resync == 2
9002 					 */
9003 					continue;
9004 				/* We need to wait 'interruptible' so as not to
9005 				 * contribute to the load average, and not to
9006 				 * be caught by 'softlockup'
9007 				 */
9008 				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
9009 				if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9010 				    mddev2->curr_resync >= mddev->curr_resync) {
9011 					if (mddev2_minor != mddev2->md_minor) {
9012 						mddev2_minor = mddev2->md_minor;
9013 						pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
9014 							desc, mdname(mddev),
9015 							mdname(mddev2));
9016 					}
9017 					spin_unlock(&all_mddevs_lock);
9018 
9019 					if (signal_pending(current))
9020 						flush_signals(current);
9021 					schedule();
9022 					finish_wait(&resync_wait, &wq);
9023 					goto try_again;
9024 				}
9025 				finish_wait(&resync_wait, &wq);
9026 			}
9027 		}
9028 		spin_unlock(&all_mddevs_lock);
9029 	} while (mddev->curr_resync < MD_RESYNC_DELAYED);
9030 
9031 	max_sectors = md_sync_max_sectors(mddev, action);
9032 	j = md_sync_position(mddev, action);
9033 
9034 	pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
9035 	pr_debug("md: minimum _guaranteed_  speed: %d KB/sec/disk.\n", speed_min(mddev));
9036 	pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
9037 		 speed_max(mddev), desc);
9038 
9039 	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
9040 
9041 	io_sectors = 0;
9042 	for (m = 0; m < SYNC_MARKS; m++) {
9043 		mark[m] = jiffies;
9044 		mark_cnt[m] = io_sectors;
9045 	}
9046 	last_mark = 0;
9047 	mddev->resync_mark = mark[last_mark];
9048 	mddev->resync_mark_cnt = mark_cnt[last_mark];
9049 
9050 	/*
9051 	 * Tune reconstruction:
9052 	 */
9053 	window = 32 * (PAGE_SIZE / 512);
9054 	pr_debug("md: using %dk window, over a total of %lluk.\n",
9055 		 window/2, (unsigned long long)max_sectors/2);
9056 
9057 	atomic_set(&mddev->recovery_active, 0);
9058 	last_check = 0;
9059 
9060 	if (j >= MD_RESYNC_ACTIVE) {
9061 		pr_debug("md: resuming %s of %s from checkpoint.\n",
9062 			 desc, mdname(mddev));
9063 		mddev->curr_resync = j;
9064 	} else
9065 		mddev->curr_resync = MD_RESYNC_ACTIVE; /* no longer delayed */
9066 	mddev->curr_resync_completed = j;
9067 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9068 	md_new_event();
9069 	update_time = jiffies;
9070 
9071 	blk_start_plug(&plug);
9072 	while (j < max_sectors) {
9073 		sector_t sectors;
9074 
9075 		skipped = 0;
9076 
9077 		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9078 		    ((mddev->curr_resync > mddev->curr_resync_completed &&
9079 		      (mddev->curr_resync - mddev->curr_resync_completed)
9080 		      > (max_sectors >> 4)) ||
9081 		     time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
9082 		     (j - mddev->curr_resync_completed)*2
9083 		     >= mddev->resync_max - mddev->curr_resync_completed ||
9084 		     mddev->curr_resync_completed > mddev->resync_max
9085 			    )) {
9086 			/* time to update curr_resync_completed */
9087 			wait_event(mddev->recovery_wait,
9088 				   atomic_read(&mddev->recovery_active) == 0);
9089 			mddev->curr_resync_completed = j;
9090 			if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
9091 			    j > mddev->recovery_cp)
9092 				mddev->recovery_cp = j;
9093 			update_time = jiffies;
9094 			set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
9095 			sysfs_notify_dirent_safe(mddev->sysfs_completed);
9096 		}
9097 
9098 		while (j >= mddev->resync_max &&
9099 		       !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9100 			/* As this condition is controlled by user-space,
9101 			 * we can block indefinitely, so use '_interruptible'
9102 			 * to avoid triggering warnings.
9103 			 */
9104 			flush_signals(current); /* just in case */
9105 			wait_event_interruptible(mddev->recovery_wait,
9106 						 mddev->resync_max > j
9107 						 || test_bit(MD_RECOVERY_INTR,
9108 							     &mddev->recovery));
9109 		}
9110 
9111 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9112 			break;
9113 
9114 		sectors = mddev->pers->sync_request(mddev, j, max_sectors,
9115 						    &skipped);
9116 		if (sectors == 0) {
9117 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9118 			break;
9119 		}
9120 
9121 		if (!skipped) { /* actual IO requested */
9122 			io_sectors += sectors;
9123 			atomic_add(sectors, &mddev->recovery_active);
9124 		}
9125 
9126 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9127 			break;
9128 
9129 		j += sectors;
9130 		if (j > max_sectors)
9131 			/* when skipping, extra large numbers can be returned. */
9132 			j = max_sectors;
9133 		if (j >= MD_RESYNC_ACTIVE)
9134 			mddev->curr_resync = j;
9135 		mddev->curr_mark_cnt = io_sectors;
9136 		if (last_check == 0)
9137 			/* this is the earliest that rebuild will be
9138 			 * visible in /proc/mdstat
9139 			 */
9140 			md_new_event();
9141 
9142 		if (last_check + window > io_sectors || j == max_sectors)
9143 			continue;
9144 
9145 		last_check = io_sectors;
9146 	repeat:
9147 		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
9148 			/* step marks */
9149 			int next = (last_mark+1) % SYNC_MARKS;
9150 
9151 			mddev->resync_mark = mark[next];
9152 			mddev->resync_mark_cnt = mark_cnt[next];
9153 			mark[next] = jiffies;
9154 			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
9155 			last_mark = next;
9156 		}
9157 
9158 		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9159 			break;
9160 
9161 		/*
9162 		 * this loop exits only if either when we are slower than
9163 		 * the 'hard' speed limit, or the system was IO-idle for
9164 		 * a jiffy.
9165 		 * the system might be non-idle CPU-wise, but we only care
9166 		 * about not overloading the IO subsystem. (things like an
9167 		 * e2fsck being done on the RAID array should execute fast)
9168 		 */
9169 		cond_resched();
9170 
9171 		recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
9172 		currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
9173 			/((jiffies-mddev->resync_mark)/HZ +1) +1;
9174 
9175 		if (currspeed > speed_min(mddev)) {
9176 			if (currspeed > speed_max(mddev)) {
9177 				msleep(500);
9178 				goto repeat;
9179 			}
9180 			if (!is_mddev_idle(mddev, 0)) {
9181 				/*
9182 				 * Give other IO more of a chance.
9183 				 * The faster the devices, the less we wait.
9184 				 */
9185 				wait_event(mddev->recovery_wait,
9186 					   !atomic_read(&mddev->recovery_active));
9187 			}
9188 		}
9189 	}
9190 	pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
9191 		test_bit(MD_RECOVERY_INTR, &mddev->recovery)
9192 		? "interrupted" : "done");
9193 	/*
9194 	 * this also signals 'finished resyncing' to md_stop
9195 	 */
9196 	blk_finish_plug(&plug);
9197 	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
9198 
9199 	if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9200 	    !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9201 	    mddev->curr_resync >= MD_RESYNC_ACTIVE) {
9202 		mddev->curr_resync_completed = mddev->curr_resync;
9203 		sysfs_notify_dirent_safe(mddev->sysfs_completed);
9204 	}
9205 	mddev->pers->sync_request(mddev, max_sectors, max_sectors, &skipped);
9206 
9207 	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
9208 	    mddev->curr_resync > MD_RESYNC_ACTIVE) {
9209 		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
9210 			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9211 				if (mddev->curr_resync >= mddev->recovery_cp) {
9212 					pr_debug("md: checkpointing %s of %s.\n",
9213 						 desc, mdname(mddev));
9214 					if (test_bit(MD_RECOVERY_ERROR,
9215 						&mddev->recovery))
9216 						mddev->recovery_cp =
9217 							mddev->curr_resync_completed;
9218 					else
9219 						mddev->recovery_cp =
9220 							mddev->curr_resync;
9221 				}
9222 			} else
9223 				mddev->recovery_cp = MaxSector;
9224 		} else {
9225 			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
9226 				mddev->curr_resync = MaxSector;
9227 			if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9228 			    test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
9229 				rcu_read_lock();
9230 				rdev_for_each_rcu(rdev, mddev)
9231 					if (rdev->raid_disk >= 0 &&
9232 					    mddev->delta_disks >= 0 &&
9233 					    !test_bit(Journal, &rdev->flags) &&
9234 					    !test_bit(Faulty, &rdev->flags) &&
9235 					    !test_bit(In_sync, &rdev->flags) &&
9236 					    rdev->recovery_offset < mddev->curr_resync)
9237 						rdev->recovery_offset = mddev->curr_resync;
9238 				rcu_read_unlock();
9239 			}
9240 		}
9241 	}
9242  skip:
9243 	/* set CHANGE_PENDING here since maybe another update is needed,
9244 	 * so other nodes are informed. It should be harmless for normal
9245 	 * raid */
9246 	set_mask_bits(&mddev->sb_flags, 0,
9247 		      BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
9248 
9249 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9250 			!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9251 			mddev->delta_disks > 0 &&
9252 			mddev->pers->finish_reshape &&
9253 			mddev->pers->size &&
9254 			!mddev_is_dm(mddev)) {
9255 		mddev_lock_nointr(mddev);
9256 		md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
9257 		mddev_unlock(mddev);
9258 		if (!mddev_is_clustered(mddev))
9259 			set_capacity_and_notify(mddev->gendisk,
9260 						mddev->array_sectors);
9261 	}
9262 
9263 	spin_lock(&mddev->lock);
9264 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
9265 		/* We completed so min/max setting can be forgotten if used. */
9266 		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9267 			mddev->resync_min = 0;
9268 		mddev->resync_max = MaxSector;
9269 	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
9270 		mddev->resync_min = mddev->curr_resync_completed;
9271 	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
9272 	mddev->curr_resync = MD_RESYNC_NONE;
9273 	spin_unlock(&mddev->lock);
9274 
9275 	wake_up(&resync_wait);
9276 	md_wakeup_thread(mddev->thread);
9277 	return;
9278 }
9279 EXPORT_SYMBOL_GPL(md_do_sync);
9280 
rdev_removeable(struct md_rdev * rdev)9281 static bool rdev_removeable(struct md_rdev *rdev)
9282 {
9283 	/* rdev is not used. */
9284 	if (rdev->raid_disk < 0)
9285 		return false;
9286 
9287 	/* There are still inflight io, don't remove this rdev. */
9288 	if (atomic_read(&rdev->nr_pending))
9289 		return false;
9290 
9291 	/*
9292 	 * An error occurred but has not yet been acknowledged by the metadata
9293 	 * handler, don't remove this rdev.
9294 	 */
9295 	if (test_bit(Blocked, &rdev->flags))
9296 		return false;
9297 
9298 	/* Fautly rdev is not used, it's safe to remove it. */
9299 	if (test_bit(Faulty, &rdev->flags))
9300 		return true;
9301 
9302 	/* Journal disk can only be removed if it's faulty. */
9303 	if (test_bit(Journal, &rdev->flags))
9304 		return false;
9305 
9306 	/*
9307 	 * 'In_sync' is cleared while 'raid_disk' is valid, which means
9308 	 * replacement has just become active from pers->spare_active(), and
9309 	 * then pers->hot_remove_disk() will replace this rdev with replacement.
9310 	 */
9311 	if (!test_bit(In_sync, &rdev->flags))
9312 		return true;
9313 
9314 	return false;
9315 }
9316 
rdev_is_spare(struct md_rdev * rdev)9317 static bool rdev_is_spare(struct md_rdev *rdev)
9318 {
9319 	return !test_bit(Candidate, &rdev->flags) && rdev->raid_disk >= 0 &&
9320 	       !test_bit(In_sync, &rdev->flags) &&
9321 	       !test_bit(Journal, &rdev->flags) &&
9322 	       !test_bit(Faulty, &rdev->flags);
9323 }
9324 
rdev_addable(struct md_rdev * rdev)9325 static bool rdev_addable(struct md_rdev *rdev)
9326 {
9327 	/* rdev is already used, don't add it again. */
9328 	if (test_bit(Candidate, &rdev->flags) || rdev->raid_disk >= 0 ||
9329 	    test_bit(Faulty, &rdev->flags))
9330 		return false;
9331 
9332 	/* Allow to add journal disk. */
9333 	if (test_bit(Journal, &rdev->flags))
9334 		return true;
9335 
9336 	/* Allow to add if array is read-write. */
9337 	if (md_is_rdwr(rdev->mddev))
9338 		return true;
9339 
9340 	/*
9341 	 * For read-only array, only allow to readd a rdev. And if bitmap is
9342 	 * used, don't allow to readd a rdev that is too old.
9343 	 */
9344 	if (rdev->saved_raid_disk >= 0 && !test_bit(Bitmap_sync, &rdev->flags))
9345 		return true;
9346 
9347 	return false;
9348 }
9349 
md_spares_need_change(struct mddev * mddev)9350 static bool md_spares_need_change(struct mddev *mddev)
9351 {
9352 	struct md_rdev *rdev;
9353 
9354 	rcu_read_lock();
9355 	rdev_for_each_rcu(rdev, mddev) {
9356 		if (rdev_removeable(rdev) || rdev_addable(rdev)) {
9357 			rcu_read_unlock();
9358 			return true;
9359 		}
9360 	}
9361 	rcu_read_unlock();
9362 	return false;
9363 }
9364 
remove_and_add_spares(struct mddev * mddev,struct md_rdev * this)9365 static int remove_and_add_spares(struct mddev *mddev,
9366 				 struct md_rdev *this)
9367 {
9368 	struct md_rdev *rdev;
9369 	int spares = 0;
9370 	int removed = 0;
9371 
9372 	if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
9373 		/* Mustn't remove devices when resync thread is running */
9374 		return 0;
9375 
9376 	rdev_for_each(rdev, mddev) {
9377 		if ((this == NULL || rdev == this) && rdev_removeable(rdev) &&
9378 		    !mddev->pers->hot_remove_disk(mddev, rdev)) {
9379 			sysfs_unlink_rdev(mddev, rdev);
9380 			rdev->saved_raid_disk = rdev->raid_disk;
9381 			rdev->raid_disk = -1;
9382 			removed++;
9383 		}
9384 	}
9385 
9386 	if (removed && mddev->kobj.sd)
9387 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9388 
9389 	if (this && removed)
9390 		goto no_add;
9391 
9392 	rdev_for_each(rdev, mddev) {
9393 		if (this && this != rdev)
9394 			continue;
9395 		if (rdev_is_spare(rdev))
9396 			spares++;
9397 		if (!rdev_addable(rdev))
9398 			continue;
9399 		if (!test_bit(Journal, &rdev->flags))
9400 			rdev->recovery_offset = 0;
9401 		if (mddev->pers->hot_add_disk(mddev, rdev) == 0) {
9402 			/* failure here is OK */
9403 			sysfs_link_rdev(mddev, rdev);
9404 			if (!test_bit(Journal, &rdev->flags))
9405 				spares++;
9406 			md_new_event();
9407 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9408 		}
9409 	}
9410 no_add:
9411 	if (removed)
9412 		set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9413 	return spares;
9414 }
9415 
md_choose_sync_action(struct mddev * mddev,int * spares)9416 static bool md_choose_sync_action(struct mddev *mddev, int *spares)
9417 {
9418 	/* Check if reshape is in progress first. */
9419 	if (mddev->reshape_position != MaxSector) {
9420 		if (mddev->pers->check_reshape == NULL ||
9421 		    mddev->pers->check_reshape(mddev) != 0)
9422 			return false;
9423 
9424 		set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9425 		clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9426 		return true;
9427 	}
9428 
9429 	/*
9430 	 * Remove any failed drives, then add spares if possible. Spares are
9431 	 * also removed and re-added, to allow the personality to fail the
9432 	 * re-add.
9433 	 */
9434 	*spares = remove_and_add_spares(mddev, NULL);
9435 	if (*spares) {
9436 		clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9437 		clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9438 		clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9439 
9440 		/* Start new recovery. */
9441 		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9442 		return true;
9443 	}
9444 
9445 	/* Check if recovery is in progress. */
9446 	if (mddev->recovery_cp < MaxSector) {
9447 		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9448 		clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9449 		return true;
9450 	}
9451 
9452 	/* Delay to choose resync/check/repair in md_do_sync(). */
9453 	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9454 		return true;
9455 
9456 	/* Nothing to be done */
9457 	return false;
9458 }
9459 
md_start_sync(struct work_struct * ws)9460 static void md_start_sync(struct work_struct *ws)
9461 {
9462 	struct mddev *mddev = container_of(ws, struct mddev, sync_work);
9463 	int spares = 0;
9464 	bool suspend = false;
9465 	char *name;
9466 
9467 	/*
9468 	 * If reshape is still in progress, spares won't be added or removed
9469 	 * from conf until reshape is done.
9470 	 */
9471 	if (mddev->reshape_position == MaxSector &&
9472 	    md_spares_need_change(mddev)) {
9473 		suspend = true;
9474 		mddev_suspend(mddev, false);
9475 	}
9476 
9477 	mddev_lock_nointr(mddev);
9478 	if (!md_is_rdwr(mddev)) {
9479 		/*
9480 		 * On a read-only array we can:
9481 		 * - remove failed devices
9482 		 * - add already-in_sync devices if the array itself is in-sync.
9483 		 * As we only add devices that are already in-sync, we can
9484 		 * activate the spares immediately.
9485 		 */
9486 		remove_and_add_spares(mddev, NULL);
9487 		goto not_running;
9488 	}
9489 
9490 	if (!md_choose_sync_action(mddev, &spares))
9491 		goto not_running;
9492 
9493 	if (!mddev->pers->sync_request)
9494 		goto not_running;
9495 
9496 	/*
9497 	 * We are adding a device or devices to an array which has the bitmap
9498 	 * stored on all devices. So make sure all bitmap pages get written.
9499 	 */
9500 	if (spares)
9501 		mddev->bitmap_ops->write_all(mddev);
9502 
9503 	name = test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ?
9504 			"reshape" : "resync";
9505 	rcu_assign_pointer(mddev->sync_thread,
9506 			   md_register_thread(md_do_sync, mddev, name));
9507 	if (!mddev->sync_thread) {
9508 		pr_warn("%s: could not start resync thread...\n",
9509 			mdname(mddev));
9510 		/* leave the spares where they are, it shouldn't hurt */
9511 		goto not_running;
9512 	}
9513 
9514 	mddev_unlock(mddev);
9515 	/*
9516 	 * md_start_sync was triggered by MD_RECOVERY_NEEDED, so we should
9517 	 * not set it again. Otherwise, we may cause issue like this one:
9518 	 *     https://bugzilla.kernel.org/show_bug.cgi?id=218200
9519 	 * Therefore, use __mddev_resume(mddev, false).
9520 	 */
9521 	if (suspend)
9522 		__mddev_resume(mddev, false);
9523 	md_wakeup_thread(mddev->sync_thread);
9524 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9525 	md_new_event();
9526 	return;
9527 
9528 not_running:
9529 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9530 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9531 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9532 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9533 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9534 	mddev_unlock(mddev);
9535 	/*
9536 	 * md_start_sync was triggered by MD_RECOVERY_NEEDED, so we should
9537 	 * not set it again. Otherwise, we may cause issue like this one:
9538 	 *     https://bugzilla.kernel.org/show_bug.cgi?id=218200
9539 	 * Therefore, use __mddev_resume(mddev, false).
9540 	 */
9541 	if (suspend)
9542 		__mddev_resume(mddev, false);
9543 
9544 	wake_up(&resync_wait);
9545 	if (test_and_clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
9546 	    mddev->sysfs_action)
9547 		sysfs_notify_dirent_safe(mddev->sysfs_action);
9548 }
9549 
unregister_sync_thread(struct mddev * mddev)9550 static void unregister_sync_thread(struct mddev *mddev)
9551 {
9552 	if (!test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9553 		/* resync/recovery still happening */
9554 		clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9555 		return;
9556 	}
9557 
9558 	if (WARN_ON_ONCE(!mddev->sync_thread))
9559 		return;
9560 
9561 	md_reap_sync_thread(mddev);
9562 }
9563 
9564 /*
9565  * This routine is regularly called by all per-raid-array threads to
9566  * deal with generic issues like resync and super-block update.
9567  * Raid personalities that don't have a thread (linear/raid0) do not
9568  * need this as they never do any recovery or update the superblock.
9569  *
9570  * It does not do any resync itself, but rather "forks" off other threads
9571  * to do that as needed.
9572  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
9573  * "->recovery" and create a thread at ->sync_thread.
9574  * When the thread finishes it sets MD_RECOVERY_DONE
9575  * and wakeups up this thread which will reap the thread and finish up.
9576  * This thread also removes any faulty devices (with nr_pending == 0).
9577  *
9578  * The overall approach is:
9579  *  1/ if the superblock needs updating, update it.
9580  *  2/ If a recovery thread is running, don't do anything else.
9581  *  3/ If recovery has finished, clean up, possibly marking spares active.
9582  *  4/ If there are any faulty devices, remove them.
9583  *  5/ If array is degraded, try to add spares devices
9584  *  6/ If array has spares or is not in-sync, start a resync thread.
9585  */
md_check_recovery(struct mddev * mddev)9586 void md_check_recovery(struct mddev *mddev)
9587 {
9588 	if (mddev->bitmap)
9589 		mddev->bitmap_ops->daemon_work(mddev);
9590 
9591 	if (signal_pending(current)) {
9592 		if (mddev->pers->sync_request && !mddev->external) {
9593 			pr_debug("md: %s in immediate safe mode\n",
9594 				 mdname(mddev));
9595 			mddev->safemode = 2;
9596 		}
9597 		flush_signals(current);
9598 	}
9599 
9600 	if (!md_is_rdwr(mddev) &&
9601 	    !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) &&
9602 	    !test_bit(MD_RECOVERY_DONE, &mddev->recovery))
9603 		return;
9604 	if ( ! (
9605 		(mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
9606 		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9607 		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
9608 		(mddev->external == 0 && mddev->safemode == 1) ||
9609 		(mddev->safemode == 2
9610 		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9611 		))
9612 		return;
9613 
9614 	if (mddev_trylock(mddev)) {
9615 		bool try_set_sync = mddev->safemode != 0;
9616 
9617 		if (!mddev->external && mddev->safemode == 1)
9618 			mddev->safemode = 0;
9619 
9620 		if (!md_is_rdwr(mddev)) {
9621 			struct md_rdev *rdev;
9622 
9623 			if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9624 				unregister_sync_thread(mddev);
9625 				goto unlock;
9626 			}
9627 
9628 			if (!mddev->external && mddev->in_sync)
9629 				/*
9630 				 * 'Blocked' flag not needed as failed devices
9631 				 * will be recorded if array switched to read/write.
9632 				 * Leaving it set will prevent the device
9633 				 * from being removed.
9634 				 */
9635 				rdev_for_each(rdev, mddev)
9636 					clear_bit(Blocked, &rdev->flags);
9637 
9638 			/*
9639 			 * There is no thread, but we need to call
9640 			 * ->spare_active and clear saved_raid_disk
9641 			 */
9642 			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9643 			md_reap_sync_thread(mddev);
9644 
9645 			/*
9646 			 * Let md_start_sync() to remove and add rdevs to the
9647 			 * array.
9648 			 */
9649 			if (md_spares_need_change(mddev)) {
9650 				set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9651 				queue_work(md_misc_wq, &mddev->sync_work);
9652 			}
9653 
9654 			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9655 			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9656 			clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9657 
9658 			goto unlock;
9659 		}
9660 
9661 		if (mddev_is_clustered(mddev)) {
9662 			struct md_rdev *rdev, *tmp;
9663 			/* kick the device if another node issued a
9664 			 * remove disk.
9665 			 */
9666 			rdev_for_each_safe(rdev, tmp, mddev) {
9667 				if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9668 						rdev->raid_disk < 0)
9669 					md_kick_rdev_from_array(rdev);
9670 			}
9671 		}
9672 
9673 		if (try_set_sync && !mddev->external && !mddev->in_sync) {
9674 			spin_lock(&mddev->lock);
9675 			set_in_sync(mddev);
9676 			spin_unlock(&mddev->lock);
9677 		}
9678 
9679 		if (mddev->sb_flags)
9680 			md_update_sb(mddev, 0);
9681 
9682 		/*
9683 		 * Never start a new sync thread if MD_RECOVERY_RUNNING is
9684 		 * still set.
9685 		 */
9686 		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9687 			unregister_sync_thread(mddev);
9688 			goto unlock;
9689 		}
9690 
9691 		/* Set RUNNING before clearing NEEDED to avoid
9692 		 * any transients in the value of "sync_action".
9693 		 */
9694 		mddev->curr_resync_completed = 0;
9695 		spin_lock(&mddev->lock);
9696 		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9697 		spin_unlock(&mddev->lock);
9698 		/* Clear some bits that don't mean anything, but
9699 		 * might be left set
9700 		 */
9701 		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9702 		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9703 
9704 		if (test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) &&
9705 		    !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
9706 			queue_work(md_misc_wq, &mddev->sync_work);
9707 		} else {
9708 			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9709 			wake_up(&resync_wait);
9710 		}
9711 
9712 	unlock:
9713 		wake_up(&mddev->sb_wait);
9714 		mddev_unlock(mddev);
9715 	}
9716 }
9717 EXPORT_SYMBOL(md_check_recovery);
9718 
md_reap_sync_thread(struct mddev * mddev)9719 void md_reap_sync_thread(struct mddev *mddev)
9720 {
9721 	struct md_rdev *rdev;
9722 	sector_t old_dev_sectors = mddev->dev_sectors;
9723 	bool is_reshaped = false;
9724 
9725 	/* resync has finished, collect result */
9726 	md_unregister_thread(mddev, &mddev->sync_thread);
9727 	atomic_inc(&mddev->sync_seq);
9728 
9729 	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9730 	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9731 	    mddev->degraded != mddev->raid_disks) {
9732 		/* success...*/
9733 		/* activate any spares */
9734 		if (mddev->pers->spare_active(mddev)) {
9735 			sysfs_notify_dirent_safe(mddev->sysfs_degraded);
9736 			set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9737 		}
9738 	}
9739 	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9740 	    mddev->pers->finish_reshape) {
9741 		mddev->pers->finish_reshape(mddev);
9742 		if (mddev_is_clustered(mddev))
9743 			is_reshaped = true;
9744 	}
9745 
9746 	/* If array is no-longer degraded, then any saved_raid_disk
9747 	 * information must be scrapped.
9748 	 */
9749 	if (!mddev->degraded)
9750 		rdev_for_each(rdev, mddev)
9751 			rdev->saved_raid_disk = -1;
9752 
9753 	md_update_sb(mddev, 1);
9754 	/* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9755 	 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9756 	 * clustered raid */
9757 	if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9758 		md_cluster_ops->resync_finish(mddev);
9759 	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9760 	clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9761 	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9762 	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9763 	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9764 	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9765 	/*
9766 	 * We call md_cluster_ops->update_size here because sync_size could
9767 	 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9768 	 * so it is time to update size across cluster.
9769 	 */
9770 	if (mddev_is_clustered(mddev) && is_reshaped
9771 				      && !test_bit(MD_CLOSING, &mddev->flags))
9772 		md_cluster_ops->update_size(mddev, old_dev_sectors);
9773 	/* flag recovery needed just to double check */
9774 	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9775 	sysfs_notify_dirent_safe(mddev->sysfs_completed);
9776 	sysfs_notify_dirent_safe(mddev->sysfs_action);
9777 	md_new_event();
9778 	if (mddev->event_work.func)
9779 		queue_work(md_misc_wq, &mddev->event_work);
9780 	wake_up(&resync_wait);
9781 }
9782 EXPORT_SYMBOL(md_reap_sync_thread);
9783 
md_wait_for_blocked_rdev(struct md_rdev * rdev,struct mddev * mddev)9784 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9785 {
9786 	sysfs_notify_dirent_safe(rdev->sysfs_state);
9787 	wait_event_timeout(rdev->blocked_wait, !rdev_blocked(rdev),
9788 			   msecs_to_jiffies(5000));
9789 	rdev_dec_pending(rdev, mddev);
9790 }
9791 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9792 
md_finish_reshape(struct mddev * mddev)9793 void md_finish_reshape(struct mddev *mddev)
9794 {
9795 	/* called be personality module when reshape completes. */
9796 	struct md_rdev *rdev;
9797 
9798 	rdev_for_each(rdev, mddev) {
9799 		if (rdev->data_offset > rdev->new_data_offset)
9800 			rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9801 		else
9802 			rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9803 		rdev->data_offset = rdev->new_data_offset;
9804 	}
9805 }
9806 EXPORT_SYMBOL(md_finish_reshape);
9807 
9808 /* Bad block management */
9809 
9810 /* Returns 1 on success, 0 on failure */
rdev_set_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9811 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9812 		       int is_new)
9813 {
9814 	struct mddev *mddev = rdev->mddev;
9815 	int rv;
9816 
9817 	/*
9818 	 * Recording new badblocks for faulty rdev will force unnecessary
9819 	 * super block updating. This is fragile for external management because
9820 	 * userspace daemon may trying to remove this device and deadlock may
9821 	 * occur. This will be probably solved in the mdadm, but it is safer to
9822 	 * avoid it.
9823 	 */
9824 	if (test_bit(Faulty, &rdev->flags))
9825 		return 1;
9826 
9827 	if (is_new)
9828 		s += rdev->new_data_offset;
9829 	else
9830 		s += rdev->data_offset;
9831 	rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9832 	if (rv == 0) {
9833 		/* Make sure they get written out promptly */
9834 		if (test_bit(ExternalBbl, &rdev->flags))
9835 			sysfs_notify_dirent_safe(rdev->sysfs_unack_badblocks);
9836 		sysfs_notify_dirent_safe(rdev->sysfs_state);
9837 		set_mask_bits(&mddev->sb_flags, 0,
9838 			      BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9839 		md_wakeup_thread(rdev->mddev->thread);
9840 		return 1;
9841 	} else
9842 		return 0;
9843 }
9844 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9845 
rdev_clear_badblocks(struct md_rdev * rdev,sector_t s,int sectors,int is_new)9846 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9847 			 int is_new)
9848 {
9849 	int rv;
9850 	if (is_new)
9851 		s += rdev->new_data_offset;
9852 	else
9853 		s += rdev->data_offset;
9854 	rv = badblocks_clear(&rdev->badblocks, s, sectors);
9855 	if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9856 		sysfs_notify_dirent_safe(rdev->sysfs_badblocks);
9857 	return rv;
9858 }
9859 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9860 
md_notify_reboot(struct notifier_block * this,unsigned long code,void * x)9861 static int md_notify_reboot(struct notifier_block *this,
9862 			    unsigned long code, void *x)
9863 {
9864 	struct mddev *mddev, *n;
9865 	int need_delay = 0;
9866 
9867 	spin_lock(&all_mddevs_lock);
9868 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
9869 		if (!mddev_get(mddev))
9870 			continue;
9871 		spin_unlock(&all_mddevs_lock);
9872 		if (mddev_trylock(mddev)) {
9873 			if (mddev->pers)
9874 				__md_stop_writes(mddev);
9875 			if (mddev->persistent)
9876 				mddev->safemode = 2;
9877 			mddev_unlock(mddev);
9878 		}
9879 		need_delay = 1;
9880 		mddev_put(mddev);
9881 		spin_lock(&all_mddevs_lock);
9882 	}
9883 	spin_unlock(&all_mddevs_lock);
9884 
9885 	/*
9886 	 * certain more exotic SCSI devices are known to be
9887 	 * volatile wrt too early system reboots. While the
9888 	 * right place to handle this issue is the given
9889 	 * driver, we do want to have a safe RAID driver ...
9890 	 */
9891 	if (need_delay)
9892 		msleep(1000);
9893 
9894 	return NOTIFY_DONE;
9895 }
9896 
9897 static struct notifier_block md_notifier = {
9898 	.notifier_call	= md_notify_reboot,
9899 	.next		= NULL,
9900 	.priority	= INT_MAX, /* before any real devices */
9901 };
9902 
md_geninit(void)9903 static void md_geninit(void)
9904 {
9905 	pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9906 
9907 	proc_create("mdstat", S_IRUGO, NULL, &mdstat_proc_ops);
9908 }
9909 
md_init(void)9910 static int __init md_init(void)
9911 {
9912 	int ret = -ENOMEM;
9913 
9914 	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9915 	if (!md_wq)
9916 		goto err_wq;
9917 
9918 	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9919 	if (!md_misc_wq)
9920 		goto err_misc_wq;
9921 
9922 	md_bitmap_wq = alloc_workqueue("md_bitmap", WQ_MEM_RECLAIM | WQ_UNBOUND,
9923 				       0);
9924 	if (!md_bitmap_wq)
9925 		goto err_bitmap_wq;
9926 
9927 	ret = __register_blkdev(MD_MAJOR, "md", md_probe);
9928 	if (ret < 0)
9929 		goto err_md;
9930 
9931 	ret = __register_blkdev(0, "mdp", md_probe);
9932 	if (ret < 0)
9933 		goto err_mdp;
9934 	mdp_major = ret;
9935 
9936 	register_reboot_notifier(&md_notifier);
9937 	raid_table_header = register_sysctl("dev/raid", raid_table);
9938 
9939 	md_geninit();
9940 	return 0;
9941 
9942 err_mdp:
9943 	unregister_blkdev(MD_MAJOR, "md");
9944 err_md:
9945 	destroy_workqueue(md_bitmap_wq);
9946 err_bitmap_wq:
9947 	destroy_workqueue(md_misc_wq);
9948 err_misc_wq:
9949 	destroy_workqueue(md_wq);
9950 err_wq:
9951 	return ret;
9952 }
9953 
check_sb_changes(struct mddev * mddev,struct md_rdev * rdev)9954 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9955 {
9956 	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9957 	struct md_rdev *rdev2, *tmp;
9958 	int role, ret;
9959 
9960 	/*
9961 	 * If size is changed in another node then we need to
9962 	 * do resize as well.
9963 	 */
9964 	if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9965 		ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9966 		if (ret)
9967 			pr_info("md-cluster: resize failed\n");
9968 		else
9969 			mddev->bitmap_ops->update_sb(mddev->bitmap);
9970 	}
9971 
9972 	/* Check for change of roles in the active devices */
9973 	rdev_for_each_safe(rdev2, tmp, mddev) {
9974 		if (test_bit(Faulty, &rdev2->flags))
9975 			continue;
9976 
9977 		/* Check if the roles changed */
9978 		role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9979 
9980 		if (test_bit(Candidate, &rdev2->flags)) {
9981 			if (role == MD_DISK_ROLE_FAULTY) {
9982 				pr_info("md: Removing Candidate device %pg because add failed\n",
9983 					rdev2->bdev);
9984 				md_kick_rdev_from_array(rdev2);
9985 				continue;
9986 			}
9987 			else
9988 				clear_bit(Candidate, &rdev2->flags);
9989 		}
9990 
9991 		if (role != rdev2->raid_disk) {
9992 			/*
9993 			 * got activated except reshape is happening.
9994 			 */
9995 			if (rdev2->raid_disk == -1 && role != MD_DISK_ROLE_SPARE &&
9996 			    !(le32_to_cpu(sb->feature_map) &
9997 			      MD_FEATURE_RESHAPE_ACTIVE) &&
9998 			    !md_cluster_ops->resync_status_get(mddev)) {
9999 				/*
10000 				 * -1 to make raid1_add_disk() set conf->fullsync
10001 				 * to 1. This could avoid skipping sync when the
10002 				 * remote node is down during resyncing.
10003 				 */
10004 				if ((le32_to_cpu(sb->feature_map)
10005 				    & MD_FEATURE_RECOVERY_OFFSET))
10006 					rdev2->saved_raid_disk = -1;
10007 				else
10008 					rdev2->saved_raid_disk = role;
10009 				ret = remove_and_add_spares(mddev, rdev2);
10010 				pr_info("Activated spare: %pg\n",
10011 					rdev2->bdev);
10012 				/* wakeup mddev->thread here, so array could
10013 				 * perform resync with the new activated disk */
10014 				set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
10015 				md_wakeup_thread(mddev->thread);
10016 			}
10017 			/* device faulty
10018 			 * We just want to do the minimum to mark the disk
10019 			 * as faulty. The recovery is performed by the
10020 			 * one who initiated the error.
10021 			 */
10022 			if (role == MD_DISK_ROLE_FAULTY ||
10023 			    role == MD_DISK_ROLE_JOURNAL) {
10024 				md_error(mddev, rdev2);
10025 				clear_bit(Blocked, &rdev2->flags);
10026 			}
10027 		}
10028 	}
10029 
10030 	if (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) {
10031 		ret = update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
10032 		if (ret)
10033 			pr_warn("md: updating array disks failed. %d\n", ret);
10034 	}
10035 
10036 	/*
10037 	 * Since mddev->delta_disks has already updated in update_raid_disks,
10038 	 * so it is time to check reshape.
10039 	 */
10040 	if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
10041 	    (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
10042 		/*
10043 		 * reshape is happening in the remote node, we need to
10044 		 * update reshape_position and call start_reshape.
10045 		 */
10046 		mddev->reshape_position = le64_to_cpu(sb->reshape_position);
10047 		if (mddev->pers->update_reshape_pos)
10048 			mddev->pers->update_reshape_pos(mddev);
10049 		if (mddev->pers->start_reshape)
10050 			mddev->pers->start_reshape(mddev);
10051 	} else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
10052 		   mddev->reshape_position != MaxSector &&
10053 		   !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
10054 		/* reshape is just done in another node. */
10055 		mddev->reshape_position = MaxSector;
10056 		if (mddev->pers->update_reshape_pos)
10057 			mddev->pers->update_reshape_pos(mddev);
10058 	}
10059 
10060 	/* Finally set the event to be up to date */
10061 	mddev->events = le64_to_cpu(sb->events);
10062 }
10063 
read_rdev(struct mddev * mddev,struct md_rdev * rdev)10064 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
10065 {
10066 	int err;
10067 	struct page *swapout = rdev->sb_page;
10068 	struct mdp_superblock_1 *sb;
10069 
10070 	/* Store the sb page of the rdev in the swapout temporary
10071 	 * variable in case we err in the future
10072 	 */
10073 	rdev->sb_page = NULL;
10074 	err = alloc_disk_sb(rdev);
10075 	if (err == 0) {
10076 		ClearPageUptodate(rdev->sb_page);
10077 		rdev->sb_loaded = 0;
10078 		err = super_types[mddev->major_version].
10079 			load_super(rdev, NULL, mddev->minor_version);
10080 	}
10081 	if (err < 0) {
10082 		pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
10083 				__func__, __LINE__, rdev->desc_nr, err);
10084 		if (rdev->sb_page)
10085 			put_page(rdev->sb_page);
10086 		rdev->sb_page = swapout;
10087 		rdev->sb_loaded = 1;
10088 		return err;
10089 	}
10090 
10091 	sb = page_address(rdev->sb_page);
10092 	/* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
10093 	 * is not set
10094 	 */
10095 
10096 	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
10097 		rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
10098 
10099 	/* The other node finished recovery, call spare_active to set
10100 	 * device In_sync and mddev->degraded
10101 	 */
10102 	if (rdev->recovery_offset == MaxSector &&
10103 	    !test_bit(In_sync, &rdev->flags) &&
10104 	    mddev->pers->spare_active(mddev))
10105 		sysfs_notify_dirent_safe(mddev->sysfs_degraded);
10106 
10107 	put_page(swapout);
10108 	return 0;
10109 }
10110 
md_reload_sb(struct mddev * mddev,int nr)10111 void md_reload_sb(struct mddev *mddev, int nr)
10112 {
10113 	struct md_rdev *rdev = NULL, *iter;
10114 	int err;
10115 
10116 	/* Find the rdev */
10117 	rdev_for_each_rcu(iter, mddev) {
10118 		if (iter->desc_nr == nr) {
10119 			rdev = iter;
10120 			break;
10121 		}
10122 	}
10123 
10124 	if (!rdev) {
10125 		pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
10126 		return;
10127 	}
10128 
10129 	err = read_rdev(mddev, rdev);
10130 	if (err < 0)
10131 		return;
10132 
10133 	check_sb_changes(mddev, rdev);
10134 
10135 	/* Read all rdev's to update recovery_offset */
10136 	rdev_for_each_rcu(rdev, mddev) {
10137 		if (!test_bit(Faulty, &rdev->flags))
10138 			read_rdev(mddev, rdev);
10139 	}
10140 }
10141 EXPORT_SYMBOL(md_reload_sb);
10142 
10143 #ifndef MODULE
10144 
10145 /*
10146  * Searches all registered partitions for autorun RAID arrays
10147  * at boot time.
10148  */
10149 
10150 static DEFINE_MUTEX(detected_devices_mutex);
10151 static LIST_HEAD(all_detected_devices);
10152 struct detected_devices_node {
10153 	struct list_head list;
10154 	dev_t dev;
10155 };
10156 
md_autodetect_dev(dev_t dev)10157 void md_autodetect_dev(dev_t dev)
10158 {
10159 	struct detected_devices_node *node_detected_dev;
10160 
10161 	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
10162 	if (node_detected_dev) {
10163 		node_detected_dev->dev = dev;
10164 		mutex_lock(&detected_devices_mutex);
10165 		list_add_tail(&node_detected_dev->list, &all_detected_devices);
10166 		mutex_unlock(&detected_devices_mutex);
10167 	}
10168 }
10169 
md_autostart_arrays(int part)10170 void md_autostart_arrays(int part)
10171 {
10172 	struct md_rdev *rdev;
10173 	struct detected_devices_node *node_detected_dev;
10174 	dev_t dev;
10175 	int i_scanned, i_passed;
10176 
10177 	i_scanned = 0;
10178 	i_passed = 0;
10179 
10180 	pr_info("md: Autodetecting RAID arrays.\n");
10181 
10182 	mutex_lock(&detected_devices_mutex);
10183 	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
10184 		i_scanned++;
10185 		node_detected_dev = list_entry(all_detected_devices.next,
10186 					struct detected_devices_node, list);
10187 		list_del(&node_detected_dev->list);
10188 		dev = node_detected_dev->dev;
10189 		kfree(node_detected_dev);
10190 		mutex_unlock(&detected_devices_mutex);
10191 		rdev = md_import_device(dev,0, 90);
10192 		mutex_lock(&detected_devices_mutex);
10193 		if (IS_ERR(rdev))
10194 			continue;
10195 
10196 		if (test_bit(Faulty, &rdev->flags))
10197 			continue;
10198 
10199 		set_bit(AutoDetected, &rdev->flags);
10200 		list_add(&rdev->same_set, &pending_raid_disks);
10201 		i_passed++;
10202 	}
10203 	mutex_unlock(&detected_devices_mutex);
10204 
10205 	pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
10206 
10207 	autorun_devices(part);
10208 }
10209 
10210 #endif /* !MODULE */
10211 
md_exit(void)10212 static __exit void md_exit(void)
10213 {
10214 	struct mddev *mddev, *n;
10215 	int delay = 1;
10216 
10217 	unregister_blkdev(MD_MAJOR,"md");
10218 	unregister_blkdev(mdp_major, "mdp");
10219 	unregister_reboot_notifier(&md_notifier);
10220 	unregister_sysctl_table(raid_table_header);
10221 
10222 	/* We cannot unload the modules while some process is
10223 	 * waiting for us in select() or poll() - wake them up
10224 	 */
10225 	md_unloading = 1;
10226 	while (waitqueue_active(&md_event_waiters)) {
10227 		/* not safe to leave yet */
10228 		wake_up(&md_event_waiters);
10229 		msleep(delay);
10230 		delay += delay;
10231 	}
10232 	remove_proc_entry("mdstat", NULL);
10233 
10234 	spin_lock(&all_mddevs_lock);
10235 	list_for_each_entry_safe(mddev, n, &all_mddevs, all_mddevs) {
10236 		if (!mddev_get(mddev))
10237 			continue;
10238 		spin_unlock(&all_mddevs_lock);
10239 		export_array(mddev);
10240 		mddev->ctime = 0;
10241 		mddev->hold_active = 0;
10242 		/*
10243 		 * As the mddev is now fully clear, mddev_put will schedule
10244 		 * the mddev for destruction by a workqueue, and the
10245 		 * destroy_workqueue() below will wait for that to complete.
10246 		 */
10247 		mddev_put(mddev);
10248 		spin_lock(&all_mddevs_lock);
10249 	}
10250 	spin_unlock(&all_mddevs_lock);
10251 
10252 	destroy_workqueue(md_misc_wq);
10253 	destroy_workqueue(md_bitmap_wq);
10254 	destroy_workqueue(md_wq);
10255 }
10256 
10257 subsys_initcall(md_init);
module_exit(md_exit)10258 module_exit(md_exit)
10259 
10260 static int get_ro(char *buffer, const struct kernel_param *kp)
10261 {
10262 	return sprintf(buffer, "%d\n", start_readonly);
10263 }
set_ro(const char * val,const struct kernel_param * kp)10264 static int set_ro(const char *val, const struct kernel_param *kp)
10265 {
10266 	return kstrtouint(val, 10, (unsigned int *)&start_readonly);
10267 }
10268 
10269 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
10270 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
10271 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
10272 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
10273 
10274 MODULE_LICENSE("GPL");
10275 MODULE_DESCRIPTION("MD RAID framework");
10276 MODULE_ALIAS("md");
10277 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
10278