projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
media: usb: dvb-usb-v2: rtl28xxu: convert to use i2c_new_client_device()
[linux-block.git] / drivers / md / md.c
CommitLineData
af1a8899 1// SPDX-License-Identifier: GPL-2.0-or-later
1da177e4
LT		 2/*
3 md.c : Multiple Devices driver for Linux
f72ffdd6 4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
1da177e4
LT		 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
32a7627c
N		 23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25
9d48739e
N		 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
1da177e4
LT		 38*/
39
963c555e 40#include <linux/sched/mm.h>
3f07c014 41#include <linux/sched/signal.h>
a6fb0934 42#include <linux/kthread.h>
bff61975 43#include <linux/blkdev.h>
fc974ee2 44#include <linux/badblocks.h>
1da177e4 45#include <linux/sysctl.h>
bff61975 46#include <linux/seq_file.h>
ff01bb48 47#include <linux/fs.h>
d7603b7e 48#include <linux/poll.h>
16f17b39 49#include <linux/ctype.h>
e7d2860b 50#include <linux/string.h>
fb4d8c76
N		 51#include <linux/hdreg.h>
52#include <linux/proc_fs.h>
53#include <linux/random.h>
056075c7 54#include <linux/module.h>
fb4d8c76 55#include <linux/reboot.h>
32a7627c 56#include <linux/file.h>
aa98aa31 57#include <linux/compat.h>
25570727 58#include <linux/delay.h>
bff61975
N		 59#include <linux/raid/md_p.h>
60#include <linux/raid/md_u.h>
5a0e3ad6 61#include <linux/slab.h>
4ad23a97
N		 62#include <linux/percpu-refcount.h>
63
504634f6 64#include <trace/events/block.h>
43b2e5d8 65#include "md.h"
935fe098 66#include "md-bitmap.h"
edb39c9d 67#include "md-cluster.h"
1da177e4 68
1da177e4 69#ifndef MODULE
d710e138 70static void autostart_arrays(int part);
1da177e4
LT		 71#endif
72
01f96c0a
N		 73/* pers_list is a list of registered personalities protected
74 * by pers_lock.
75 * pers_lock does extra service to protect accesses to
76 * mddev->thread when the mutex cannot be held.
77 */
2604b703 78static LIST_HEAD(pers_list);
1da177e4
LT		 79static DEFINE_SPINLOCK(pers_lock);
80
28dec870
KO		 81static struct kobj_type md_ktype;
82
edb39c9d 83struct md_cluster_operations *md_cluster_ops;
589a1c49 84EXPORT_SYMBOL(md_cluster_ops);
2b598ee5 85static struct module *md_cluster_mod;
edb39c9d 86
90b08710 87static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 88static struct workqueue_struct *md_wq;
89static struct workqueue_struct *md_misc_wq;
90b08710 90
746d3207
N		 91static int remove_and_add_spares(struct mddev *mddev,
92 struct md_rdev *this);
5aa61f42 93static void mddev_detach(struct mddev *mddev);
746d3207 94
1e50915f
RB		 95/*
96 * Default number of read corrections we'll attempt on an rdev
97 * before ejecting it from the array. We divide the read error
98 * count by 2 for every hour elapsed between read errors.
99 */
100#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 101/*
102 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
103 * is 1000 KB/sec, so the extra system load does not show up that much.
104 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 105 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 106 * subsystem is idle. There is also an 'absolute maximum' reconstruction
107 * speed limit - in case reconstruction slows down your system despite
108 * idle IO detection.
109 *
110 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 111 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 112 */
113
114static int sysctl_speed_limit_min = 1000;
115static int sysctl_speed_limit_max = 200000;
fd01b88c 116static inline int speed_min(struct mddev *mddev)
88202a0c
N		 117{
118 return mddev->sync_speed_min ?
119 mddev->sync_speed_min : sysctl_speed_limit_min;
120}
121
fd01b88c 122static inline int speed_max(struct mddev *mddev)
88202a0c
N		 123{
124 return mddev->sync_speed_max ?
125 mddev->sync_speed_max : sysctl_speed_limit_max;
126}
1da177e4 127
3e148a32
GJ		 128static int rdev_init_wb(struct md_rdev *rdev)
129{
130 if (rdev->bdev->bd_queue->nr_hw_queues == 1)
131 return 0;
132
133 spin_lock_init(&rdev->wb_list_lock);
134 INIT_LIST_HEAD(&rdev->wb_list);
135 init_waitqueue_head(&rdev->wb_io_wait);
136 set_bit(WBCollisionCheck, &rdev->flags);
137
138 return 1;
139}
140
963c555e
GJ		 141/*
142 * Create wb_info_pool if rdev is the first multi-queue device flaged
143 * with writemostly, also write-behind mode is enabled.
144 */
145void mddev_create_wb_pool(struct mddev *mddev, struct md_rdev *rdev,
146 bool is_suspend)
147{
148 if (mddev->bitmap_info.max_write_behind == 0)
149 return;
150
151 if (!test_bit(WriteMostly, &rdev->flags) || !rdev_init_wb(rdev))
152 return;
153
154 if (mddev->wb_info_pool == NULL) {
155 unsigned int noio_flag;
156
157 if (!is_suspend)
158 mddev_suspend(mddev);
159 noio_flag = memalloc_noio_save();
160 mddev->wb_info_pool = mempool_create_kmalloc_pool(NR_WB_INFOS,
161 sizeof(struct wb_info));
162 memalloc_noio_restore(noio_flag);
163 if (!mddev->wb_info_pool)
164 pr_err("can't alloc memory pool for writemostly\n");
165 if (!is_suspend)
166 mddev_resume(mddev);
167 }
168}
169EXPORT_SYMBOL_GPL(mddev_create_wb_pool);
170
171/*
172 * destroy wb_info_pool if rdev is the last device flaged with WBCollisionCheck.
173 */
174static void mddev_destroy_wb_pool(struct mddev *mddev, struct md_rdev *rdev)
175{
176 if (!test_and_clear_bit(WBCollisionCheck, &rdev->flags))
177 return;
178
179 if (mddev->wb_info_pool) {
180 struct md_rdev *temp;
181 int num = 0;
182
183 /*
184 * Check if other rdevs need wb_info_pool.
185 */
186 rdev_for_each(temp, mddev)
187 if (temp != rdev &&
188 test_bit(WBCollisionCheck, &temp->flags))
189 num++;
190 if (!num) {
191 mddev_suspend(rdev->mddev);
192 mempool_destroy(mddev->wb_info_pool);
193 mddev->wb_info_pool = NULL;
194 mddev_resume(rdev->mddev);
195 }
196 }
197}
198
1da177e4
LT		 199static struct ctl_table_header *raid_table_header;
200
82592c38 201static struct ctl_table raid_table[] = {
1da177e4 202 {
1da177e4
LT		 203 .procname = "speed_limit_min",
204 .data = &sysctl_speed_limit_min,
205 .maxlen = sizeof(int),
80ca3a44 206 .mode = S_IRUGO|S_IWUSR,
6d456111 207 .proc_handler = proc_dointvec,
1da177e4
LT		 208 },
209 {
1da177e4
LT		 210 .procname = "speed_limit_max",
211 .data = &sysctl_speed_limit_max,
212 .maxlen = sizeof(int),
80ca3a44 213 .mode = S_IRUGO|S_IWUSR,
6d456111 214 .proc_handler = proc_dointvec,
1da177e4 215 },
894d2491 216 { }
1da177e4
LT		 217};
218
82592c38 219static struct ctl_table raid_dir_table[] = {
1da177e4 220 {
1da177e4
LT		 221 .procname = "raid",
222 .maxlen = 0,
80ca3a44 223 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 224 .child = raid_table,
225 },
894d2491 226 { }
1da177e4
LT		 227};
228
82592c38 229static struct ctl_table raid_root_table[] = {
1da177e4 230 {
1da177e4
LT		 231 .procname = "dev",
232 .maxlen = 0,
233 .mode = 0555,
234 .child = raid_dir_table,
235 },
894d2491 236 { }
1da177e4
LT		 237};
238
83d5cde4 239static const struct block_device_operations md_fops;
1da177e4 240
f91de92e
N		 241static int start_readonly;
242
78b6350d
N		 243/*
244 * The original mechanism for creating an md device is to create
245 * a device node in /dev and to open it. This causes races with device-close.
246 * The preferred method is to write to the "new_array" module parameter.
247 * This can avoid races.
248 * Setting create_on_open to false disables the original mechanism
249 * so all the races disappear.
250 */
251static bool create_on_open = true;
252
a167f663 253struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 254 struct mddev *mddev)
a167f663 255{
afeee514 256 if (!mddev || !bioset_initialized(&mddev->bio_set))
a167f663
N		 257 return bio_alloc(gfp_mask, nr_iovecs);
258
6251691a 259 return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
a167f663
N		 260}
261EXPORT_SYMBOL_GPL(bio_alloc_mddev);
262
5a85071c
N		 263static struct bio *md_bio_alloc_sync(struct mddev *mddev)
264{
afeee514 265 if (!mddev || !bioset_initialized(&mddev->sync_set))
5a85071c
N		 266 return bio_alloc(GFP_NOIO, 1);
267
afeee514 268 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
5a85071c
N		 269}
270
d7603b7e
N		 271/*
272 * We have a system wide 'event count' that is incremented
273 * on any 'interesting' event, and readers of /proc/mdstat
274 * can use 'poll' or 'select' to find out when the event
275 * count increases.
276 *
277 * Events are:
278 * start array, stop array, error, add device, remove device,
279 * start build, activate spare
280 */
2989ddbd 281static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 282static atomic_t md_event_count;
fd01b88c 283void md_new_event(struct mddev *mddev)
d7603b7e
N		 284{
285 atomic_inc(&md_event_count);
286 wake_up(&md_event_waiters);
287}
29269553 288EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 289
1da177e4
LT		 290/*
291 * Enables to iterate over all existing md arrays
292 * all_mddevs_lock protects this list.
293 */
294static LIST_HEAD(all_mddevs);
295static DEFINE_SPINLOCK(all_mddevs_lock);
296
1da177e4
LT		 297/*
298 * iterates through all used mddevs in the system.
299 * We take care to grab the all_mddevs_lock whenever navigating
300 * the list, and to always hold a refcount when unlocked.
301 * Any code which breaks out of this loop while own
302 * a reference to the current mddev and must mddev_put it.
303 */
fd01b88c 304#define for_each_mddev(_mddev,_tmp) \
1da177e4 305 \
f72ffdd6 306 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 307 _tmp = all_mddevs.next; \
308 _mddev = NULL;}); \
309 ({ if (_tmp != &all_mddevs) \
310 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 311 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 312 if (_mddev) mddev_put(_mddev); \
313 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
314 _tmp != &all_mddevs;}); \
1da177e4 315 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 316 _tmp = _tmp->next;}) \
1da177e4
LT		 317 )
318
409c57f3
N		 319/* Rather than calling directly into the personality make_request function,
320 * IO requests come here first so that we can check if the device is
321 * being suspended pending a reconfiguration.
322 * We hold a refcount over the call to ->make_request. By the time that
323 * call has finished, the bio has been linked into some internal structure
324 * and so is visible to ->quiesce(), so we don't need the refcount any more.
325 */
b3143b9a
N		 326static bool is_suspended(struct mddev *mddev, struct bio *bio)
327{
328 if (mddev->suspended)
329 return true;
330 if (bio_data_dir(bio) != WRITE)
331 return false;
332 if (mddev->suspend_lo >= mddev->suspend_hi)
333 return false;
334 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
335 return false;
336 if (bio_end_sector(bio) < mddev->suspend_lo)
337 return false;
338 return true;
339}
340
393debc2
SL		 341void md_handle_request(struct mddev *mddev, struct bio *bio)
342{
343check_suspended:
344 rcu_read_lock();
b3143b9a 345 if (is_suspended(mddev, bio)) {
393debc2
SL		 346 DEFINE_WAIT(__wait);
347 for (;;) {
348 prepare_to_wait(&mddev->sb_wait, &__wait,
349 TASK_UNINTERRUPTIBLE);
b3143b9a 350 if (!is_suspended(mddev, bio))
393debc2
SL		 351 break;
352 rcu_read_unlock();
353 schedule();
354 rcu_read_lock();
355 }
356 finish_wait(&mddev->sb_wait, &__wait);
357 }
358 atomic_inc(&mddev->active_io);
359 rcu_read_unlock();
360
361 if (!mddev->pers->make_request(mddev, bio)) {
362 atomic_dec(&mddev->active_io);
363 wake_up(&mddev->sb_wait);
364 goto check_suspended;
365 }
366
367 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
368 wake_up(&mddev->sb_wait);
369}
370EXPORT_SYMBOL(md_handle_request);
371
dece1635 372static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 373{
49077326 374 const int rw = bio_data_dir(bio);
ddcf35d3 375 const int sgrp = op_stat_group(bio_op(bio));
fd01b88c 376 struct mddev *mddev = q->queuedata;
e91ece55 377 unsigned int sectors;
49077326 378
62f7b198
GP		 379 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
380 bio_io_error(bio);
381 return BLK_QC_T_NONE;
382 }
383
af67c31f 384 blk_queue_split(q, &bio);
54efd50b 385
274d8cbd 386 if (mddev == NULL || mddev->pers == NULL) {
409c57f3 387 bio_io_error(bio);
dece1635 388 return BLK_QC_T_NONE;
409c57f3 389 }
bbfa57c0 390 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
4246a0b6 391 if (bio_sectors(bio) != 0)
4e4cbee9 392 bio->bi_status = BLK_STS_IOERR;
4246a0b6 393 bio_endio(bio);
dece1635 394 return BLK_QC_T_NONE;
bbfa57c0 395 }
49077326 396
e91ece55
CM		 397 /*
398 * save the sectors now since our bio can
399 * go away inside make_request
400 */
401 sectors = bio_sectors(bio);
9c573de3 402 /* bio could be mergeable after passing to underlayer */
1eff9d32 403 bio->bi_opf &= ~REQ_NOMERGE;
393debc2
SL		 404
405 md_handle_request(mddev, bio);
49077326 406
112f158f
MS		 407 part_stat_lock();
408 part_stat_inc(&mddev->gendisk->part0, ios[sgrp]);
409 part_stat_add(&mddev->gendisk->part0, sectors[sgrp], sectors);
74672d06 410 part_stat_unlock();
49077326 411
dece1635 412 return BLK_QC_T_NONE;
409c57f3
N		 413}
414
9e35b99c
N		 415/* mddev_suspend makes sure no new requests are submitted
416 * to the device, and that any requests that have been submitted
417 * are completely handled.
afa0f557
N		 418 * Once mddev_detach() is called and completes, the module will be
419 * completely unused.
9e35b99c 420 */
fd01b88c 421void mddev_suspend(struct mddev *mddev)
409c57f3 422{
092398dc 423 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
4d5324f7 424 lockdep_assert_held(&mddev->reconfig_mutex);
0dc10e50
MP		 425 if (mddev->suspended++)
426 return;
409c57f3 427 synchronize_rcu();
cc27b0c7 428 wake_up(&mddev->sb_wait);
35bfc521
N		 429 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
430 smp_mb__after_atomic();
409c57f3
N		 431 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
432 mddev->pers->quiesce(mddev, 1);
35bfc521
N		 433 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
434 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
0d9f4f13
JB		 435
436 del_timer_sync(&mddev->safemode_timer);
409c57f3 437}
390ee602 438EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 439
fd01b88c 440void mddev_resume(struct mddev *mddev)
409c57f3 441{
4d5324f7 442 lockdep_assert_held(&mddev->reconfig_mutex);
0dc10e50
MP		 443 if (--mddev->suspended)
444 return;
409c57f3
N		 445 wake_up(&mddev->sb_wait);
446 mddev->pers->quiesce(mddev, 0);
0fd018af 447
47525e59 448 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB		 449 md_wakeup_thread(mddev->thread);
450 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 451}
390ee602 452EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 453
fd01b88c 454int mddev_congested(struct mddev *mddev, int bits)
3fa841d7 455{
5c675f83
N		 456 struct md_personality *pers = mddev->pers;
457 int ret = 0;
458
459 rcu_read_lock();
460 if (mddev->suspended)
461 ret = 1;
462 else if (pers && pers->congested)
463 ret = pers->congested(mddev, bits);
464 rcu_read_unlock();
465 return ret;
466}
467EXPORT_SYMBOL_GPL(mddev_congested);
468static int md_congested(void *data, int bits)
469{
470 struct mddev *mddev = data;
471 return mddev_congested(mddev, bits);
3fa841d7 472}
3fa841d7 473
a2826aa9 474/*
e9c7469b 475 * Generic flush handling for md
a2826aa9
N		 476 */
477
4bc034d3 478static void md_end_flush(struct bio *bio)
a2826aa9 479{
4bc034d3
N		 480 struct md_rdev *rdev = bio->bi_private;
481 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 482
483 rdev_dec_pending(rdev, mddev);
484
4bc034d3
N		 485 if (atomic_dec_and_test(&mddev->flush_pending)) {
486 /* The pre-request flush has finished */
487 queue_work(md_wq, &mddev->flush_work);
a2826aa9 488 }
4bc034d3 489 bio_put(bio);
5a409b4f 490}
a7a07e69 491
4bc034d3
N		 492static void md_submit_flush_data(struct work_struct *ws);
493
494static void submit_flushes(struct work_struct *ws)
a2826aa9 495{
4bc034d3 496 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 497 struct md_rdev *rdev;
a2826aa9 498
2bc13b83 499 mddev->start_flush = ktime_get_boottime();
4bc034d3
N		 500 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
501 atomic_set(&mddev->flush_pending, 1);
a2826aa9 502 rcu_read_lock();
dafb20fa 503 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 504 if (rdev->raid_disk >= 0 &&
505 !test_bit(Faulty, &rdev->flags)) {
506 /* Take two references, one is dropped
507 * when request finishes, one after
508 * we reclaim rcu_read_lock
509 */
510 struct bio *bi;
511 atomic_inc(&rdev->nr_pending);
512 atomic_inc(&rdev->nr_pending);
513 rcu_read_unlock();
b5e1b8ce 514 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
5a409b4f 515 bi->bi_end_io = md_end_flush;
4bc034d3
N		 516 bi->bi_private = rdev;
517 bio_set_dev(bi, rdev->bdev);
70fd7614 518 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
4bc034d3 519 atomic_inc(&mddev->flush_pending);
4e49ea4a 520 submit_bio(bi);
a2826aa9
N		 521 rcu_read_lock();
522 rdev_dec_pending(rdev, mddev);
523 }
524 rcu_read_unlock();
4bc034d3
N		 525 if (atomic_dec_and_test(&mddev->flush_pending))
526 queue_work(md_wq, &mddev->flush_work);
527}
a2826aa9 528
4bc034d3
N		 529static void md_submit_flush_data(struct work_struct *ws)
530{
531 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
532 struct bio *bio = mddev->flush_bio;
533
534 /*
535 * must reset flush_bio before calling into md_handle_request to avoid a
536 * deadlock, because other bios passed md_handle_request suspend check
537 * could wait for this and below md_handle_request could wait for those
538 * bios because of suspend check
539 */
2bc13b83 540 mddev->last_flush = mddev->start_flush;
4bc034d3
N		 541 mddev->flush_bio = NULL;
542 wake_up(&mddev->sb_wait);
543
544 if (bio->bi_iter.bi_size == 0) {
545 /* an empty barrier - all done */
546 bio_endio(bio);
547 } else {
548 bio->bi_opf &= ~REQ_PREFLUSH;
549 md_handle_request(mddev, bio);
a2826aa9 550 }
a2826aa9 551}
4bc034d3 552
775d7831
DJ		 553/*
554 * Manages consolidation of flushes and submitting any flushes needed for
555 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
556 * being finished in another context. Returns false if the flushing is
557 * complete but still needs the I/O portion of the bio to be processed.
558 */
559bool md_flush_request(struct mddev *mddev, struct bio *bio)
4bc034d3 560{
2bc13b83 561 ktime_t start = ktime_get_boottime();
4bc034d3
N		 562 spin_lock_irq(&mddev->lock);
563 wait_event_lock_irq(mddev->sb_wait,
2bc13b83
N		 564 !mddev->flush_bio ||
565 ktime_after(mddev->last_flush, start),
4bc034d3 566 mddev->lock);
2bc13b83
N		 567 if (!ktime_after(mddev->last_flush, start)) {
568 WARN_ON(mddev->flush_bio);
569 mddev->flush_bio = bio;
570 bio = NULL;
571 }
4bc034d3
N		 572 spin_unlock_irq(&mddev->lock);
573
2bc13b83
N		 574 if (!bio) {
575 INIT_WORK(&mddev->flush_work, submit_flushes);
576 queue_work(md_wq, &mddev->flush_work);
577 } else {
578 /* flush was performed for some other bio while we waited. */
579 if (bio->bi_iter.bi_size == 0)
580 /* an empty barrier - all done */
581 bio_endio(bio);
582 else {
583 bio->bi_opf &= ~REQ_PREFLUSH;
775d7831 584 return false;
2bc13b83
N		 585 }
586 }
775d7831 587 return true;
4bc034d3 588}
e9c7469b 589EXPORT_SYMBOL(md_flush_request);
409c57f3 590
fd01b88c 591static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 592{
593 atomic_inc(&mddev->active);
594 return mddev;
595}
596
5fd3a17e 597static void mddev_delayed_delete(struct work_struct *ws);
d3374825 598
fd01b88c 599static void mddev_put(struct mddev *mddev)
1da177e4
LT		 600{
601 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
602 return;
d3374825 603 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 604 mddev->ctime == 0 && !mddev->hold_active) {
605 /* Array is not configured at all, and not held active,
606 * so destroy it */
af8a2434 607 list_del_init(&mddev->all_mddevs);
28dec870
KO		 608
609 /*
610 * Call queue_work inside the spinlock so that
611 * flush_workqueue() after mddev_find will succeed in waiting
612 * for the work to be done.
613 */
614 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
615 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 616 }
617 spin_unlock(&all_mddevs_lock);
1da177e4
LT		 618}
619
8376d3c1 620static void md_safemode_timeout(struct timer_list *t);
25b2edfa 621
fd01b88c 622void mddev_init(struct mddev *mddev)
fafd7fb0 623{
28dec870 624 kobject_init(&mddev->kobj, &md_ktype);
fafd7fb0
N		 625 mutex_init(&mddev->open_mutex);
626 mutex_init(&mddev->reconfig_mutex);
627 mutex_init(&mddev->bitmap_info.mutex);
628 INIT_LIST_HEAD(&mddev->disks);
629 INIT_LIST_HEAD(&mddev->all_mddevs);
8376d3c1 630 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
fafd7fb0
N		 631 atomic_set(&mddev->active, 1);
632 atomic_set(&mddev->openers, 0);
633 atomic_set(&mddev->active_io, 0);
85572d7c 634 spin_lock_init(&mddev->lock);
4bc034d3 635 atomic_set(&mddev->flush_pending, 0);
fafd7fb0
N		 636 init_waitqueue_head(&mddev->sb_wait);
637 init_waitqueue_head(&mddev->recovery_wait);
638 mddev->reshape_position = MaxSector;
2c810cdd 639 mddev->reshape_backwards = 0;
c4a39551 640 mddev->last_sync_action = "none";
fafd7fb0
N		 641 mddev->resync_min = 0;
642 mddev->resync_max = MaxSector;
643 mddev->level = LEVEL_NONE;
644}
390ee602 645EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 646
f72ffdd6 647static struct mddev *mddev_find(dev_t unit)
1da177e4 648{
fd01b88c 649 struct mddev *mddev, *new = NULL;
1da177e4 650
8f5f02c4
N		 651 if (unit && MAJOR(unit) != MD_MAJOR)
652 unit &= ~((1<<MdpMinorShift)-1);
653
1da177e4
LT		 654 retry:
655 spin_lock(&all_mddevs_lock);
efeb53c0
N		 656
657 if (unit) {
658 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
659 if (mddev->unit == unit) {
660 mddev_get(mddev);
661 spin_unlock(&all_mddevs_lock);
662 kfree(new);
663 return mddev;
664 }
665
666 if (new) {
667 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 668 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 669 new->hold_active = UNTIL_IOCTL;
670 return new;
1da177e4 671 }
efeb53c0
N		 672 } else if (new) {
673 /* find an unused unit number */
674 static int next_minor = 512;
675 int start = next_minor;
676 int is_free = 0;
677 int dev = 0;
678 while (!is_free) {
679 dev = MKDEV(MD_MAJOR, next_minor);
680 next_minor++;
681 if (next_minor > MINORMASK)
682 next_minor = 0;
683 if (next_minor == start) {
684 /* Oh dear, all in use. */
685 spin_unlock(&all_mddevs_lock);
686 kfree(new);
687 return NULL;
688 }
f72ffdd6 689
efeb53c0
N		 690 is_free = 1;
691 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
692 if (mddev->unit == dev) {
693 is_free = 0;
694 break;
695 }
696 }
697 new->unit = dev;
698 new->md_minor = MINOR(dev);
699 new->hold_active = UNTIL_STOP;
1da177e4
LT		 700 list_add(&new->all_mddevs, &all_mddevs);
701 spin_unlock(&all_mddevs_lock);
702 return new;
703 }
704 spin_unlock(&all_mddevs_lock);
705
9ffae0cf 706 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 707 if (!new)
708 return NULL;
709
1da177e4
LT		 710 new->unit = unit;
711 if (MAJOR(unit) == MD_MAJOR)
712 new->md_minor = MINOR(unit);
713 else
714 new->md_minor = MINOR(unit) >> MdpMinorShift;
715
fafd7fb0 716 mddev_init(new);
1da177e4 717
1da177e4
LT		 718 goto retry;
719}
720
b6eb127d
N		 721static struct attribute_group md_redundancy_group;
722
5c47daf6 723void mddev_unlock(struct mddev *mddev)
1da177e4 724{
a64c876f 725 if (mddev->to_remove) {
b6eb127d
N		 726 /* These cannot be removed under reconfig_mutex as
727 * an access to the files will try to take reconfig_mutex
728 * while holding the file unremovable, which leads to
729 * a deadlock.
bb4f1e9d
N		 730 * So hold set sysfs_active while the remove in happeing,
731 * and anything else which might set ->to_remove or my
732 * otherwise change the sysfs namespace will fail with
733 * -EBUSY if sysfs_active is still set.
734 * We set sysfs_active under reconfig_mutex and elsewhere
735 * test it under the same mutex to ensure its correct value
736 * is seen.
b6eb127d 737 */
a64c876f
N		 738 struct attribute_group *to_remove = mddev->to_remove;
739 mddev->to_remove = NULL;
bb4f1e9d 740 mddev->sysfs_active = 1;
b6eb127d
N		 741 mutex_unlock(&mddev->reconfig_mutex);
742
00bcb4ac
N		 743 if (mddev->kobj.sd) {
744 if (to_remove != &md_redundancy_group)
745 sysfs_remove_group(&mddev->kobj, to_remove);
746 if (mddev->pers == NULL ||
747 mddev->pers->sync_request == NULL) {
748 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
749 if (mddev->sysfs_action)
750 sysfs_put(mddev->sysfs_action);
751 mddev->sysfs_action = NULL;
752 }
a64c876f 753 }
bb4f1e9d 754 mddev->sysfs_active = 0;
b6eb127d
N		 755 } else
756 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 757
751e67ca
CD		 758 /* As we've dropped the mutex we need a spinlock to
759 * make sure the thread doesn't disappear
01f96c0a
N		 760 */
761 spin_lock(&pers_lock);
005eca5e 762 md_wakeup_thread(mddev->thread);
4d5324f7 763 wake_up(&mddev->sb_wait);
01f96c0a 764 spin_unlock(&pers_lock);
1da177e4 765}
5c47daf6 766EXPORT_SYMBOL_GPL(mddev_unlock);
1da177e4 767
57d051dc 768struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
1ca69c4b
N		 769{
770 struct md_rdev *rdev;
771
772 rdev_for_each_rcu(rdev, mddev)
773 if (rdev->desc_nr == nr)
774 return rdev;
775
776 return NULL;
777}
57d051dc 778EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
1ca69c4b
N		 779
780static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 781{
3cb03002 782 struct md_rdev *rdev;
1da177e4 783
dafb20fa 784 rdev_for_each(rdev, mddev)
1da177e4
LT		 785 if (rdev->bdev->bd_dev == dev)
786 return rdev;
159ec1fc 787
1da177e4
LT		 788 return NULL;
789}
790
1532d9e8 791struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
1ca69c4b
N		 792{
793 struct md_rdev *rdev;
794
795 rdev_for_each_rcu(rdev, mddev)
796 if (rdev->bdev->bd_dev == dev)
797 return rdev;
798
799 return NULL;
800}
1532d9e8 801EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
1ca69c4b 802
84fc4b56 803static struct md_personality *find_pers(int level, char *clevel)
2604b703 804{
84fc4b56 805 struct md_personality *pers;
d9d166c2
N		 806 list_for_each_entry(pers, &pers_list, list) {
807 if (level != LEVEL_NONE && pers->level == level)
2604b703 808 return pers;
d9d166c2
N		 809 if (strcmp(pers->name, clevel)==0)
810 return pers;
811 }
2604b703
N		 812 return NULL;
813}
814
b73df2d3 815/* return the offset of the super block in 512byte sectors */
3cb03002 816static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 817{
57b2caa3 818 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 819 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 820}
821
f72ffdd6 822static int alloc_disk_sb(struct md_rdev *rdev)
1da177e4 823{
1da177e4 824 rdev->sb_page = alloc_page(GFP_KERNEL);
7f0f0d87 825 if (!rdev->sb_page)
ebc24337 826 return -ENOMEM;
1da177e4
LT		 827 return 0;
828}
829
545c8795 830void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT		 831{
832 if (rdev->sb_page) {
2d1f3b5d 833 put_page(rdev->sb_page);
1da177e4
LT		 834 rdev->sb_loaded = 0;
835 rdev->sb_page = NULL;
0f420358 836 rdev->sb_start = 0;
dd8ac336 837 rdev->sectors = 0;
1da177e4 838 }
2699b672
N		 839 if (rdev->bb_page) {
840 put_page(rdev->bb_page);
841 rdev->bb_page = NULL;
842 }
d3b407fb 843 badblocks_exit(&rdev->badblocks);
1da177e4 844}
545c8795 845EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 846
4246a0b6 847static void super_written(struct bio *bio)
7bfa19f2 848{
3cb03002 849 struct md_rdev *rdev = bio->bi_private;
fd01b88c 850 struct mddev *mddev = rdev->mddev;
7bfa19f2 851
4e4cbee9
CH		 852 if (bio->bi_status) {
853 pr_err("md: super_written gets error=%d\n", bio->bi_status);
a9701a30 854 md_error(mddev, rdev);
46533ff7
N		 855 if (!test_bit(Faulty, &rdev->flags)
856 && (bio->bi_opf & MD_FAILFAST)) {
2953079c 857 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
46533ff7
N		 858 set_bit(LastDev, &rdev->flags);
859 }
860 } else
861 clear_bit(LastDev, &rdev->flags);
7bfa19f2 862
a9701a30
N		 863 if (atomic_dec_and_test(&mddev->pending_writes))
864 wake_up(&mddev->sb_wait);
ed3b98c7 865 rdev_dec_pending(rdev, mddev);
f8b58edf 866 bio_put(bio);
7bfa19f2
N		 867}
868
fd01b88c 869void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 870 sector_t sector, int size, struct page *page)
871{
872 /* write first size bytes of page to sector of rdev
873 * Increment mddev->pending_writes before returning
874 * and decrement it on completion, waking up sb_wait
875 * if zero is reached.
876 * If an error occurred, call md_error
877 */
46533ff7
N		 878 struct bio *bio;
879 int ff = 0;
880
4b6c1060
HM		 881 if (!page)
882 return;
883
46533ff7
N		 884 if (test_bit(Faulty, &rdev->flags))
885 return;
886
5a85071c 887 bio = md_bio_alloc_sync(mddev);
7bfa19f2 888
ed3b98c7
SL		 889 atomic_inc(&rdev->nr_pending);
890
74d46992 891 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
4f024f37 892 bio->bi_iter.bi_sector = sector;
7bfa19f2
N		 893 bio_add_page(bio, page, size, 0);
894 bio->bi_private = rdev;
895 bio->bi_end_io = super_written;
46533ff7
N		 896
897 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
898 test_bit(FailFast, &rdev->flags) &&
899 !test_bit(LastDev, &rdev->flags))
900 ff = MD_FAILFAST;
5a8948f8 901 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
a9701a30 902
7bfa19f2 903 atomic_inc(&mddev->pending_writes);
4e49ea4a 904 submit_bio(bio);
a9701a30
N		 905}
906
46533ff7 907int md_super_wait(struct mddev *mddev)
a9701a30 908{
e9c7469b 909 /* wait for all superblock writes that were scheduled to complete */
1967cd56 910 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2953079c 911 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
46533ff7
N		 912 return -EAGAIN;
913 return 0;
7bfa19f2
N		 914}
915
3cb03002 916int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
796a5cf0 917 struct page *page, int op, int op_flags, bool metadata_op)
1da177e4 918{
5a85071c 919 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
1da177e4
LT		 920 int ret;
921
74d46992
CH		 922 if (metadata_op && rdev->meta_bdev)
923 bio_set_dev(bio, rdev->meta_bdev);
924 else
925 bio_set_dev(bio, rdev->bdev);
796a5cf0 926 bio_set_op_attrs(bio, op, op_flags);
ccebd4c4 927 if (metadata_op)
4f024f37 928 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1fdd6fc9
N		 929 else if (rdev->mddev->reshape_position != MaxSector &&
930 (rdev->mddev->reshape_backwards ==
931 (sector >= rdev->mddev->reshape_position)))
4f024f37 932 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
ccebd4c4 933 else
4f024f37 934 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1da177e4 935 bio_add_page(bio, page, size, 0);
4e49ea4a
MC		 936
937 submit_bio_wait(bio);
1da177e4 938
4e4cbee9 939 ret = !bio->bi_status;
1da177e4
LT		 940 bio_put(bio);
941 return ret;
942}
a8745db2 943EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 944
f72ffdd6 945static int read_disk_sb(struct md_rdev *rdev, int size)
1da177e4
LT		 946{
947 char b[BDEVNAME_SIZE];
403df478 948
1da177e4
LT		 949 if (rdev->sb_loaded)
950 return 0;
951
796a5cf0 952 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
1da177e4
LT		 953 goto fail;
954 rdev->sb_loaded = 1;
955 return 0;
956
957fail:
9d48739e
N		 958 pr_err("md: disabled device %s, could not read superblock.\n",
959 bdevname(rdev->bdev,b));
1da177e4
LT		 960 return -EINVAL;
961}
962
e6fd2093 963static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1da177e4 964{
f72ffdd6 965 return sb1->set_uuid0 == sb2->set_uuid0 &&
05710466
AN		 966 sb1->set_uuid1 == sb2->set_uuid1 &&
967 sb1->set_uuid2 == sb2->set_uuid2 &&
968 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 969}
970
e6fd2093 971static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
1da177e4
LT		 972{
973 int ret;
974 mdp_super_t *tmp1, *tmp2;
975
976 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
977 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
978
979 if (!tmp1 || !tmp2) {
980 ret = 0;
1da177e4
LT		 981 goto abort;
982 }
983
984 *tmp1 = *sb1;
985 *tmp2 = *sb2;
986
987 /*
988 * nr_disks is not constant
989 */
990 tmp1->nr_disks = 0;
991 tmp2->nr_disks = 0;
992
ce0c8e05 993 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 994abort:
990a8baf
JJ		 995 kfree(tmp1);
996 kfree(tmp2);
1da177e4
LT		 997 return ret;
998}
999
4d167f09
N		 1000static u32 md_csum_fold(u32 csum)
1001{
1002 csum = (csum & 0xffff) + (csum >> 16);
1003 return (csum & 0xffff) + (csum >> 16);
1004}
1005
f72ffdd6 1006static unsigned int calc_sb_csum(mdp_super_t *sb)
1da177e4 1007{
4d167f09
N		 1008 u64 newcsum = 0;
1009 u32 *sb32 = (u32*)sb;
1010 int i;
1da177e4
LT		 1011 unsigned int disk_csum, csum;
1012
1013 disk_csum = sb->sb_csum;
1014 sb->sb_csum = 0;
4d167f09
N		 1015
1016 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1017 newcsum += sb32[i];
1018 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1019
4d167f09
N		 1020#ifdef CONFIG_ALPHA
1021 /* This used to use csum_partial, which was wrong for several
1022 * reasons including that different results are returned on
1023 * different architectures. It isn't critical that we get exactly
1024 * the same return value as before (we always csum_fold before
1025 * testing, and that removes any differences). However as we
1026 * know that csum_partial always returned a 16bit value on
1027 * alphas, do a fold to maximise conformity to previous behaviour.
1028 */
1029 sb->sb_csum = md_csum_fold(disk_csum);
1030#else
1da177e4 1031 sb->sb_csum = disk_csum;
4d167f09 1032#endif
1da177e4
LT		 1033 return csum;
1034}
1035
1da177e4
LT		 1036/*
1037 * Handle superblock details.
1038 * We want to be able to handle multiple superblock formats
1039 * so we have a common interface to them all, and an array of
1040 * different handlers.
1041 * We rely on user-space to write the initial superblock, and support
1042 * reading and updating of superblocks.
1043 * Interface methods are:
3cb03002 1044 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1045 * loads and validates a superblock on dev.
1046 * if refdev != NULL, compare superblocks on both devices
1047 * Return:
1048 * 0 - dev has a superblock that is compatible with refdev
1049 * 1 - dev has a superblock that is compatible and newer than refdev
1050 * so dev should be used as the refdev in future
1051 * -EINVAL superblock incompatible or invalid
1052 * -othererror e.g. -EIO
1053 *
fd01b88c 1054 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1055 * Verify that dev is acceptable into mddev.
1056 * The first time, mddev->raid_disks will be 0, and data from
1057 * dev should be merged in. Subsequent calls check that dev
1058 * is new enough. Return 0 or -EINVAL
1059 *
fd01b88c 1060 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1061 * Update the superblock for rdev with data in mddev
1062 * This does not write to disc.
1063 *
1064 */
1065
1066struct super_type {
0cd17fec
CW		 1067 char *name;
1068 struct module *owner;
c6563a8c
N		 1069 int (*load_super)(struct md_rdev *rdev,
1070 struct md_rdev *refdev,
0cd17fec 1071 int minor_version);
c6563a8c
N		 1072 int (*validate_super)(struct mddev *mddev,
1073 struct md_rdev *rdev);
1074 void (*sync_super)(struct mddev *mddev,
1075 struct md_rdev *rdev);
3cb03002 1076 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1077 sector_t num_sectors);
c6563a8c
N		 1078 int (*allow_new_offset)(struct md_rdev *rdev,
1079 unsigned long long new_offset);
1da177e4
LT		 1080};
1081
0894cc30
AN		 1082/*
1083 * Check that the given mddev has no bitmap.
1084 *
1085 * This function is called from the run method of all personalities that do not
1086 * support bitmaps. It prints an error message and returns non-zero if mddev
1087 * has a bitmap. Otherwise, it returns 0.
1088 *
1089 */
fd01b88c 1090int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1091{
c3d9714e 1092 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30 1093 return 0;
9d48739e 1094 pr_warn("%s: bitmaps are not supported for %s\n",
0894cc30
AN		 1095 mdname(mddev), mddev->pers->name);
1096 return 1;
1097}
1098EXPORT_SYMBOL(md_check_no_bitmap);
1099
1da177e4 1100/*
f72ffdd6 1101 * load_super for 0.90.0
1da177e4 1102 */
3cb03002 1103static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1104{
1105 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1106 mdp_super_t *sb;
1107 int ret;
228fc7d7 1108 bool spare_disk = true;
1da177e4
LT		 1109
1110 /*
0f420358 1111 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1112 * it's at the end of the disk.
1113 *
1114 * It also happens to be a multiple of 4Kb.
1115 */
57b2caa3 1116 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1117
0002b271 1118 ret = read_disk_sb(rdev, MD_SB_BYTES);
9d48739e
N		 1119 if (ret)
1120 return ret;
1da177e4
LT		 1121
1122 ret = -EINVAL;
1123
1124 bdevname(rdev->bdev, b);
65a06f06 1125 sb = page_address(rdev->sb_page);
1da177e4
LT		 1126
1127 if (sb->md_magic != MD_SB_MAGIC) {
9d48739e 1128 pr_warn("md: invalid raid superblock magic on %s\n", b);
1da177e4
LT		 1129 goto abort;
1130 }
1131
1132 if (sb->major_version != 0 ||
f6705578
N		 1133 sb->minor_version < 90 ||
1134 sb->minor_version > 91) {
9d48739e
N		 1135 pr_warn("Bad version number %d.%d on %s\n",
1136 sb->major_version, sb->minor_version, b);
1da177e4
LT		 1137 goto abort;
1138 }
1139
1140 if (sb->raid_disks <= 0)
1141 goto abort;
1142
4d167f09 1143 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
9d48739e 1144 pr_warn("md: invalid superblock checksum on %s\n", b);
1da177e4
LT		 1145 goto abort;
1146 }
1147
1148 rdev->preferred_minor = sb->md_minor;
1149 rdev->data_offset = 0;
c6563a8c 1150 rdev->new_data_offset = 0;
0002b271 1151 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1152 rdev->badblocks.shift = -1;
1da177e4
LT		 1153
1154 if (sb->level == LEVEL_MULTIPATH)
1155 rdev->desc_nr = -1;
1156 else
1157 rdev->desc_nr = sb->this_disk.number;
1158
228fc7d7
YY		 1159 /* not spare disk, or LEVEL_MULTIPATH */
1160 if (sb->level == LEVEL_MULTIPATH ||
1161 (rdev->desc_nr >= 0 &&
3b7436cc 1162 rdev->desc_nr < MD_SB_DISKS &&
228fc7d7
YY		 1163 sb->disks[rdev->desc_nr].state &
1164 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1165 spare_disk = false;
1166
9a7b2b0f 1167 if (!refdev) {
228fc7d7 1168 if (!spare_disk)
6a5cb53a
YY		 1169 ret = 1;
1170 else
1171 ret = 0;
9a7b2b0f 1172 } else {
1da177e4 1173 __u64 ev1, ev2;
65a06f06 1174 mdp_super_t *refsb = page_address(refdev->sb_page);
e6fd2093 1175 if (!md_uuid_equal(refsb, sb)) {
9d48739e 1176 pr_warn("md: %s has different UUID to %s\n",
1da177e4
LT		 1177 b, bdevname(refdev->bdev,b2));
1178 goto abort;
1179 }
e6fd2093 1180 if (!md_sb_equal(refsb, sb)) {
9d48739e
N		 1181 pr_warn("md: %s has same UUID but different superblock to %s\n",
1182 b, bdevname(refdev->bdev, b2));
1da177e4
LT		 1183 goto abort;
1184 }
1185 ev1 = md_event(sb);
1186 ev2 = md_event(refsb);
6a5cb53a 1187
228fc7d7 1188 if (!spare_disk && ev1 > ev2)
1da177e4 1189 ret = 1;
f72ffdd6 1190 else
1da177e4
LT		 1191 ret = 0;
1192 }
8190e754 1193 rdev->sectors = rdev->sb_start;
667a5313
N		 1194 /* Limit to 4TB as metadata cannot record more than that.
1195 * (not needed for Linear and RAID0 as metadata doesn't
1196 * record this size)
1197 */
72deb455 1198 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
3312c951 1199 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1da177e4 1200
27a7b260 1201 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1202 /* "this cannot possibly happen" ... */
1203 ret = -EINVAL;
1204
1da177e4
LT		 1205 abort:
1206 return ret;
1207}
1208
1209/*
1210 * validate_super for 0.90.0
1211 */
fd01b88c 1212static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1213{
1214 mdp_disk_t *desc;
65a06f06 1215 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1216 __u64 ev1 = md_event(sb);
1da177e4 1217
41158c7e 1218 rdev->raid_disk = -1;
c5d79adb
N		 1219 clear_bit(Faulty, &rdev->flags);
1220 clear_bit(In_sync, &rdev->flags);
8313b8e5 1221 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1222 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1223
1da177e4
LT		 1224 if (mddev->raid_disks == 0) {
1225 mddev->major_version = 0;
1226 mddev->minor_version = sb->minor_version;
1227 mddev->patch_version = sb->patch_version;
e691063a 1228 mddev->external = 0;
9d8f0363 1229 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1230 mddev->ctime = sb->ctime;
1231 mddev->utime = sb->utime;
1232 mddev->level = sb->level;
d9d166c2 1233 mddev->clevel[0] = 0;
1da177e4
LT		 1234 mddev->layout = sb->layout;
1235 mddev->raid_disks = sb->raid_disks;
27a7b260 1236 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1237 mddev->events = ev1;
c3d9714e 1238 mddev->bitmap_info.offset = 0;
6409bb05
N		 1239 mddev->bitmap_info.space = 0;
1240 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1241 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1242 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1243 mddev->reshape_backwards = 0;
1da177e4 1244
f6705578
N		 1245 if (mddev->minor_version >= 91) {
1246 mddev->reshape_position = sb->reshape_position;
1247 mddev->delta_disks = sb->delta_disks;
1248 mddev->new_level = sb->new_level;
1249 mddev->new_layout = sb->new_layout;
664e7c41 1250 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N		 1251 if (mddev->delta_disks < 0)
1252 mddev->reshape_backwards = 1;
f6705578
N		 1253 } else {
1254 mddev->reshape_position = MaxSector;
1255 mddev->delta_disks = 0;
1256 mddev->new_level = mddev->level;
1257 mddev->new_layout = mddev->layout;
664e7c41 1258 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578 1259 }
33f2c35a
N		 1260 if (mddev->level == 0)
1261 mddev->layout = -1;
f6705578 1262
1da177e4
LT		 1263 if (sb->state & (1<<MD_SB_CLEAN))
1264 mddev->recovery_cp = MaxSector;
1265 else {
f72ffdd6 1266 if (sb->events_hi == sb->cp_events_hi &&
1da177e4
LT		 1267 sb->events_lo == sb->cp_events_lo) {
1268 mddev->recovery_cp = sb->recovery_cp;
1269 } else
1270 mddev->recovery_cp = 0;
1271 }
1272
1273 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1274 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1275 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1276 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1277
1278 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1279
1280 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1281 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1282 mddev->bitmap_info.offset =
1283 mddev->bitmap_info.default_offset;
6409bb05 1284 mddev->bitmap_info.space =
c9ad020f 1285 mddev->bitmap_info.default_space;
6409bb05 1286 }
a654b9d8 1287
41158c7e 1288 } else if (mddev->pers == NULL) {
be6800a7
N		 1289 /* Insist on good event counter while assembling, except
1290 * for spares (which don't need an event count) */
1da177e4 1291 ++ev1;
be6800a7
N		 1292 if (sb->disks[rdev->desc_nr].state & (
1293 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
f72ffdd6 1294 if (ev1 < mddev->events)
be6800a7 1295 return -EINVAL;
41158c7e
N		 1296 } else if (mddev->bitmap) {
1297 /* if adding to array with a bitmap, then we can accept an
1298 * older device ... but not too old.
1299 */
41158c7e
N		 1300 if (ev1 < mddev->bitmap->events_cleared)
1301 return 0;
8313b8e5
N		 1302 if (ev1 < mddev->events)
1303 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1304 } else {
1305 if (ev1 < mddev->events)
1306 /* just a hot-add of a new device, leave raid_disk at -1 */
1307 return 0;
1308 }
41158c7e 1309
1da177e4 1310 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1311 desc = sb->disks + rdev->desc_nr;
1312
1313 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1314 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1315 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1316 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1317 set_bit(In_sync, &rdev->flags);
1da177e4 1318 rdev->raid_disk = desc->raid_disk;
f466722c 1319 rdev->saved_raid_disk = desc->raid_disk;
0261cd9f
N		 1320 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1321 /* active but not in sync implies recovery up to
1322 * reshape position. We don't know exactly where
1323 * that is, so set to zero for now */
1324 if (mddev->minor_version >= 91) {
1325 rdev->recovery_offset = 0;
1326 rdev->raid_disk = desc->raid_disk;
1327 }
1da177e4 1328 }
8ddf9efe
N		 1329 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1330 set_bit(WriteMostly, &rdev->flags);
688834e6
N		 1331 if (desc->state & (1<<MD_DISK_FAILFAST))
1332 set_bit(FailFast, &rdev->flags);
41158c7e 1333 } else /* MULTIPATH are always insync */
b2d444d7 1334 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1335 return 0;
1336}
1337
1338/*
1339 * sync_super for 0.90.0
1340 */
fd01b88c 1341static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1342{
1343 mdp_super_t *sb;
3cb03002 1344 struct md_rdev *rdev2;
1da177e4 1345 int next_spare = mddev->raid_disks;
19133a42 1346
1da177e4
LT		 1347 /* make rdev->sb match mddev data..
1348 *
1349 * 1/ zero out disks
1350 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1351 * 3/ any empty disks < next_spare become removed
1352 *
1353 * disks[0] gets initialised to REMOVED because
1354 * we cannot be sure from other fields if it has
1355 * been initialised or not.
1356 */
1357 int i;
1358 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1359
61181565
N		 1360 rdev->sb_size = MD_SB_BYTES;
1361
65a06f06 1362 sb = page_address(rdev->sb_page);
1da177e4
LT		 1363
1364 memset(sb, 0, sizeof(*sb));
1365
1366 sb->md_magic = MD_SB_MAGIC;
1367 sb->major_version = mddev->major_version;
1da177e4
LT		 1368 sb->patch_version = mddev->patch_version;
1369 sb->gvalid_words = 0; /* ignored */
1370 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1371 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1372 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1373 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1374
9ebc6ef1 1375 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1da177e4 1376 sb->level = mddev->level;
58c0fed4 1377 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1378 sb->raid_disks = mddev->raid_disks;
1379 sb->md_minor = mddev->md_minor;
e691063a 1380 sb->not_persistent = 0;
9ebc6ef1 1381 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1da177e4
LT		 1382 sb->state = 0;
1383 sb->events_hi = (mddev->events>>32);
1384 sb->events_lo = (u32)mddev->events;
1385
f6705578
N		 1386 if (mddev->reshape_position == MaxSector)
1387 sb->minor_version = 90;
1388 else {
1389 sb->minor_version = 91;
1390 sb->reshape_position = mddev->reshape_position;
1391 sb->new_level = mddev->new_level;
1392 sb->delta_disks = mddev->delta_disks;
1393 sb->new_layout = mddev->new_layout;
664e7c41 1394 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1395 }
1396 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1397 if (mddev->in_sync)
1398 {
1399 sb->recovery_cp = mddev->recovery_cp;
1400 sb->cp_events_hi = (mddev->events>>32);
1401 sb->cp_events_lo = (u32)mddev->events;
1402 if (mddev->recovery_cp == MaxSector)
1403 sb->state = (1<< MD_SB_CLEAN);
1404 } else
1405 sb->recovery_cp = 0;
1406
1407 sb->layout = mddev->layout;
9d8f0363 1408 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1409
c3d9714e 1410 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1411 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1412
1da177e4 1413 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1414 rdev_for_each(rdev2, mddev) {
1da177e4 1415 mdp_disk_t *d;
86e6ffdd 1416 int desc_nr;
0261cd9f
N		 1417 int is_active = test_bit(In_sync, &rdev2->flags);
1418
1419 if (rdev2->raid_disk >= 0 &&
1420 sb->minor_version >= 91)
1421 /* we have nowhere to store the recovery_offset,
1422 * but if it is not below the reshape_position,
1423 * we can piggy-back on that.
1424 */
1425 is_active = 1;
1426 if (rdev2->raid_disk < 0 ||
1427 test_bit(Faulty, &rdev2->flags))
1428 is_active = 0;
1429 if (is_active)
86e6ffdd 1430 desc_nr = rdev2->raid_disk;
1da177e4 1431 else
86e6ffdd 1432 desc_nr = next_spare++;
19133a42 1433 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1434 d = &sb->disks[rdev2->desc_nr];
1435 nr_disks++;
1436 d->number = rdev2->desc_nr;
1437 d->major = MAJOR(rdev2->bdev->bd_dev);
1438 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1439 if (is_active)
1da177e4
LT		 1440 d->raid_disk = rdev2->raid_disk;
1441 else
1442 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1443 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1444 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1445 else if (is_active) {
1da177e4 1446 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1447 if (test_bit(In_sync, &rdev2->flags))
1448 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1449 active++;
1450 working++;
1451 } else {
1452 d->state = 0;
1453 spare++;
1454 working++;
1455 }
8ddf9efe
N		 1456 if (test_bit(WriteMostly, &rdev2->flags))
1457 d->state |= (1<<MD_DISK_WRITEMOSTLY);
688834e6
N		 1458 if (test_bit(FailFast, &rdev2->flags))
1459 d->state |= (1<<MD_DISK_FAILFAST);
1da177e4 1460 }
1da177e4
LT		 1461 /* now set the "removed" and "faulty" bits on any missing devices */
1462 for (i=0 ; i < mddev->raid_disks ; i++) {
1463 mdp_disk_t *d = &sb->disks[i];
1464 if (d->state == 0 && d->number == 0) {
1465 d->number = i;
1466 d->raid_disk = i;
1467 d->state = (1<<MD_DISK_REMOVED);
1468 d->state |= (1<<MD_DISK_FAULTY);
1469 failed++;
1470 }
1471 }
1472 sb->nr_disks = nr_disks;
1473 sb->active_disks = active;
1474 sb->working_disks = working;
1475 sb->failed_disks = failed;
1476 sb->spare_disks = spare;
1477
1478 sb->this_disk = sb->disks[rdev->desc_nr];
1479 sb->sb_csum = calc_sb_csum(sb);
1480}
1481
0cd17fec
CW		 1482/*
1483 * rdev_size_change for 0.90.0
1484 */
1485static unsigned long long
3cb03002 1486super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1487{
58c0fed4 1488 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1489 return 0; /* component must fit device */
c3d9714e 1490 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1491 return 0; /* can't move bitmap */
57b2caa3 1492 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1493 if (!num_sectors || num_sectors > rdev->sb_start)
1494 num_sectors = rdev->sb_start;
27a7b260
N		 1495 /* Limit to 4TB as metadata cannot record more than that.
1496 * 4TB == 2^32 KB, or 2*2^32 sectors.
1497 */
72deb455 1498 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
3312c951 1499 num_sectors = (sector_t)(2ULL << 32) - 2;
46533ff7
N		 1500 do {
1501 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec 1502 rdev->sb_page);
46533ff7 1503 } while (md_super_wait(rdev->mddev) < 0);
c26a44ed 1504 return num_sectors;
0cd17fec
CW		 1505}
1506
c6563a8c
N		 1507static int
1508super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1509{
1510 /* non-zero offset changes not possible with v0.90 */
1511 return new_offset == 0;
1512}
0cd17fec 1513
1da177e4
LT		 1514/*
1515 * version 1 superblock
1516 */
1517
f72ffdd6 1518static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1da177e4 1519{
1c05b4bc
N		 1520 __le32 disk_csum;
1521 u32 csum;
1da177e4
LT		 1522 unsigned long long newcsum;
1523 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1524 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1525
1526 disk_csum = sb->sb_csum;
1527 sb->sb_csum = 0;
1528 newcsum = 0;
1f3c9907 1529 for (; size >= 4; size -= 4)
1da177e4
LT		 1530 newcsum += le32_to_cpu(*isuper++);
1531
1532 if (size == 2)
1c05b4bc 1533 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1534
1535 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1536 sb->sb_csum = disk_csum;
1537 return cpu_to_le32(csum);
1538}
1539
3cb03002 1540static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1541{
1542 struct mdp_superblock_1 *sb;
1543 int ret;
0f420358 1544 sector_t sb_start;
c6563a8c 1545 sector_t sectors;
1da177e4 1546 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1547 int bmask;
228fc7d7 1548 bool spare_disk = true;
1da177e4
LT		 1549
1550 /*
0f420358 1551 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1552 * It is always aligned to a 4K boundary and
1553 * depeding on minor_version, it can be:
1554 * 0: At least 8K, but less than 12K, from end of device
1555 * 1: At start of device
1556 * 2: 4K from start of device.
1557 */
1558 switch(minor_version) {
1559 case 0:
77304d2a 1560 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1561 sb_start -= 8*2;
1562 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1563 break;
1564 case 1:
0f420358 1565 sb_start = 0;
1da177e4
LT		 1566 break;
1567 case 2:
0f420358 1568 sb_start = 8;
1da177e4
LT		 1569 break;
1570 default:
1571 return -EINVAL;
1572 }
0f420358 1573 rdev->sb_start = sb_start;
1da177e4 1574
0002b271
N		 1575 /* superblock is rarely larger than 1K, but it can be larger,
1576 * and it is safe to read 4k, so we do that
1577 */
1578 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1579 if (ret) return ret;
1580
65a06f06 1581 sb = page_address(rdev->sb_page);
1da177e4
LT		 1582
1583 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1584 sb->major_version != cpu_to_le32(1) ||
1585 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1586 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1587 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1588 return -EINVAL;
1589
1590 if (calc_sb_1_csum(sb) != sb->sb_csum) {
9d48739e 1591 pr_warn("md: invalid superblock checksum on %s\n",
1da177e4
LT		 1592 bdevname(rdev->bdev,b));
1593 return -EINVAL;
1594 }
1595 if (le64_to_cpu(sb->data_size) < 10) {
9d48739e
N		 1596 pr_warn("md: data_size too small on %s\n",
1597 bdevname(rdev->bdev,b));
1da177e4
LT		 1598 return -EINVAL;
1599 }
c6563a8c
N		 1600 if (sb->pad0 ||
1601 sb->pad3[0] ||
1602 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1603 /* Some padding is non-zero, might be a new feature */
1604 return -EINVAL;
e11e93fa 1605
1da177e4
LT		 1606 rdev->preferred_minor = 0xffff;
1607 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N		 1608 rdev->new_data_offset = rdev->data_offset;
1609 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1610 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1611 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1612 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1613
0002b271 1614 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1615 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1616 if (rdev->sb_size & bmask)
a1801f85
N		 1617 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1618
1619 if (minor_version
0f420358 1620 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1621 return -EINVAL;
c6563a8c
N		 1622 if (minor_version
1623 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1624 return -EINVAL;
0002b271 1625
31b65a0d
N		 1626 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1627 rdev->desc_nr = -1;
1628 else
1629 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1630
2699b672
N		 1631 if (!rdev->bb_page) {
1632 rdev->bb_page = alloc_page(GFP_KERNEL);
1633 if (!rdev->bb_page)
1634 return -ENOMEM;
1635 }
1636 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1637 rdev->badblocks.count == 0) {
1638 /* need to load the bad block list.
1639 * Currently we limit it to one page.
1640 */
1641 s32 offset;
1642 sector_t bb_sector;
00485d09 1643 __le64 *bbp;
2699b672
N		 1644 int i;
1645 int sectors = le16_to_cpu(sb->bblog_size);
1646 if (sectors > (PAGE_SIZE / 512))
1647 return -EINVAL;
1648 offset = le32_to_cpu(sb->bblog_offset);
1649 if (offset == 0)
1650 return -EINVAL;
1651 bb_sector = (long long)offset;
1652 if (!sync_page_io(rdev, bb_sector, sectors << 9,
796a5cf0 1653 rdev->bb_page, REQ_OP_READ, 0, true))
2699b672 1654 return -EIO;
00485d09 1655 bbp = (__le64 *)page_address(rdev->bb_page);
2699b672
N		 1656 rdev->badblocks.shift = sb->bblog_shift;
1657 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1658 u64 bb = le64_to_cpu(*bbp);
1659 int count = bb & (0x3ff);
1660 u64 sector = bb >> 10;
1661 sector <<= sb->bblog_shift;
1662 count <<= sb->bblog_shift;
1663 if (bb + 1 == 0)
1664 break;
fc974ee2 1665 if (badblocks_set(&rdev->badblocks, sector, count, 1))
2699b672
N		 1666 return -EINVAL;
1667 }
486adf72
N		 1668 } else if (sb->bblog_offset != 0)
1669 rdev->badblocks.shift = 0;
2699b672 1670
ddc08823
PB		 1671 if ((le32_to_cpu(sb->feature_map) &
1672 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
ea0213e0
AP		 1673 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1674 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1675 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1676 }
1677
33f2c35a
N		 1678 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1679 sb->level != 0)
1680 return -EINVAL;
1681
228fc7d7
YY		 1682 /* not spare disk, or LEVEL_MULTIPATH */
1683 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1684 (rdev->desc_nr >= 0 &&
1685 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1686 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1687 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1688 spare_disk = false;
6a5cb53a 1689
9a7b2b0f 1690 if (!refdev) {
228fc7d7 1691 if (!spare_disk)
6a5cb53a
YY		 1692 ret = 1;
1693 else
1694 ret = 0;
9a7b2b0f 1695 } else {
1da177e4 1696 __u64 ev1, ev2;
65a06f06 1697 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1698
1699 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1700 sb->level != refsb->level ||
1701 sb->layout != refsb->layout ||
1702 sb->chunksize != refsb->chunksize) {
9d48739e 1703 pr_warn("md: %s has strangely different superblock to %s\n",
1da177e4
LT		 1704 bdevname(rdev->bdev,b),
1705 bdevname(refdev->bdev,b2));
1706 return -EINVAL;
1707 }
1708 ev1 = le64_to_cpu(sb->events);
1709 ev2 = le64_to_cpu(refsb->events);
1710
228fc7d7 1711 if (!spare_disk && ev1 > ev2)
8ed75463
N		 1712 ret = 1;
1713 else
1714 ret = 0;
1da177e4 1715 }
c6563a8c
N		 1716 if (minor_version) {
1717 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1718 sectors -= rdev->data_offset;
1719 } else
1720 sectors = rdev->sb_start;
1721 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1722 return -EINVAL;
dd8ac336 1723 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1724 return ret;
1da177e4
LT		 1725}
1726
fd01b88c 1727static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1728{
65a06f06 1729 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1730 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1731
41158c7e 1732 rdev->raid_disk = -1;
c5d79adb
N		 1733 clear_bit(Faulty, &rdev->flags);
1734 clear_bit(In_sync, &rdev->flags);
8313b8e5 1735 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1736 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1737
1da177e4
LT		 1738 if (mddev->raid_disks == 0) {
1739 mddev->major_version = 1;
1740 mddev->patch_version = 0;
e691063a 1741 mddev->external = 0;
9d8f0363 1742 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
9ebc6ef1
DD		 1743 mddev->ctime = le64_to_cpu(sb->ctime);
1744 mddev->utime = le64_to_cpu(sb->utime);
1da177e4 1745 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1746 mddev->clevel[0] = 0;
1da177e4
LT		 1747 mddev->layout = le32_to_cpu(sb->layout);
1748 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1749 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1750 mddev->events = ev1;
c3d9714e 1751 mddev->bitmap_info.offset = 0;
6409bb05
N		 1752 mddev->bitmap_info.space = 0;
1753 /* Default location for bitmap is 1K after superblock
1754 * using 3K - total of 4K
1755 */
c3d9714e 1756 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1757 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N		 1758 mddev->reshape_backwards = 0;
1759
1da177e4
LT		 1760 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1761 memcpy(mddev->uuid, sb->set_uuid, 16);
1762
1763 mddev->max_disks = (4096-256)/2;
a654b9d8 1764
71c0805c 1765 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1766 mddev->bitmap_info.file == NULL) {
c3d9714e
N		 1767 mddev->bitmap_info.offset =
1768 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N		 1769 /* Metadata doesn't record how much space is available.
1770 * For 1.0, we assume we can use up to the superblock
1771 * if before, else to 4K beyond superblock.
1772 * For others, assume no change is possible.
1773 */
1774 if (mddev->minor_version > 0)
1775 mddev->bitmap_info.space = 0;
1776 else if (mddev->bitmap_info.offset > 0)
1777 mddev->bitmap_info.space =
1778 8 - mddev->bitmap_info.offset;
1779 else
1780 mddev->bitmap_info.space =
1781 -mddev->bitmap_info.offset;
1782 }
e11e93fa 1783
f6705578
N		 1784 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1785 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1786 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1787 mddev->new_level = le32_to_cpu(sb->new_level);
1788 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1789 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N		 1790 if (mddev->delta_disks < 0 ||
1791 (mddev->delta_disks == 0 &&
1792 (le32_to_cpu(sb->feature_map)
1793 & MD_FEATURE_RESHAPE_BACKWARDS)))
1794 mddev->reshape_backwards = 1;
f6705578
N		 1795 } else {
1796 mddev->reshape_position = MaxSector;
1797 mddev->delta_disks = 0;
1798 mddev->new_level = mddev->level;
1799 mddev->new_layout = mddev->layout;
664e7c41 1800 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1801 }
1802
33f2c35a
N		 1803 if (mddev->level == 0 &&
1804 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1805 mddev->layout = -1;
1806
486b0f7b 1807 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
a62ab49e 1808 set_bit(MD_HAS_JOURNAL, &mddev->flags);
ea0213e0 1809
ddc08823
PB		 1810 if (le32_to_cpu(sb->feature_map) &
1811 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
ea0213e0
AP		 1812 if (le32_to_cpu(sb->feature_map) &
1813 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1814 return -EINVAL;
ddc08823
PB		 1815 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1816 (le32_to_cpu(sb->feature_map) &
1817 MD_FEATURE_MULTIPLE_PPLS))
1818 return -EINVAL;
ea0213e0
AP		 1819 set_bit(MD_HAS_PPL, &mddev->flags);
1820 }
41158c7e 1821 } else if (mddev->pers == NULL) {
be6800a7
N		 1822 /* Insist of good event counter while assembling, except for
1823 * spares (which don't need an event count) */
1da177e4 1824 ++ev1;
be6800a7
N		 1825 if (rdev->desc_nr >= 0 &&
1826 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
a3dfbdaa
SL		 1827 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1828 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
be6800a7
N		 1829 if (ev1 < mddev->events)
1830 return -EINVAL;
41158c7e
N		 1831 } else if (mddev->bitmap) {
1832 /* If adding to array with a bitmap, then we can accept an
1833 * older device, but not too old.
1834 */
41158c7e
N		 1835 if (ev1 < mddev->bitmap->events_cleared)
1836 return 0;
8313b8e5
N		 1837 if (ev1 < mddev->events)
1838 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N		 1839 } else {
1840 if (ev1 < mddev->events)
1841 /* just a hot-add of a new device, leave raid_disk at -1 */
1842 return 0;
1843 }
1da177e4
LT		 1844 if (mddev->level != LEVEL_MULTIPATH) {
1845 int role;
3673f305
N		 1846 if (rdev->desc_nr < 0 ||
1847 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
c4d4c91b 1848 role = MD_DISK_ROLE_SPARE;
3673f305
N		 1849 rdev->desc_nr = -1;
1850 } else
1851 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4 1852 switch(role) {
c4d4c91b 1853 case MD_DISK_ROLE_SPARE: /* spare */
1da177e4 1854 break;
c4d4c91b 1855 case MD_DISK_ROLE_FAULTY: /* faulty */
b2d444d7 1856 set_bit(Faulty, &rdev->flags);
1da177e4 1857 break;
bac624f3
SL		 1858 case MD_DISK_ROLE_JOURNAL: /* journal device */
1859 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1860 /* journal device without journal feature */
9d48739e 1861 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
bac624f3
SL		 1862 return -EINVAL;
1863 }
1864 set_bit(Journal, &rdev->flags);
3069aa8d 1865 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
9b15603d 1866 rdev->raid_disk = 0;
bac624f3 1867 break;
1da177e4 1868 default:
f466722c 1869 rdev->saved_raid_disk = role;
5fd6c1dc 1870 if ((le32_to_cpu(sb->feature_map) &
f466722c 1871 MD_FEATURE_RECOVERY_OFFSET)) {
5fd6c1dc 1872 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
f466722c
N		 1873 if (!(le32_to_cpu(sb->feature_map) &
1874 MD_FEATURE_RECOVERY_BITMAP))
1875 rdev->saved_raid_disk = -1;
062f5b2a
GJ		 1876 } else {
1877 /*
1878 * If the array is FROZEN, then the device can't
1879 * be in_sync with rest of array.
1880 */
1881 if (!test_bit(MD_RECOVERY_FROZEN,
1882 &mddev->recovery))
1883 set_bit(In_sync, &rdev->flags);
1884 }
1da177e4
LT		 1885 rdev->raid_disk = role;
1886 break;
1887 }
8ddf9efe
N		 1888 if (sb->devflags & WriteMostly1)
1889 set_bit(WriteMostly, &rdev->flags);
688834e6
N		 1890 if (sb->devflags & FailFast1)
1891 set_bit(FailFast, &rdev->flags);
2d78f8c4
N		 1892 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1893 set_bit(Replacement, &rdev->flags);
41158c7e 1894 } else /* MULTIPATH are always insync */
b2d444d7 1895 set_bit(In_sync, &rdev->flags);
41158c7e 1896
1da177e4
LT		 1897 return 0;
1898}
1899
fd01b88c 1900static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1901{
1902 struct mdp_superblock_1 *sb;
3cb03002 1903 struct md_rdev *rdev2;
1da177e4
LT		 1904 int max_dev, i;
1905 /* make rdev->sb match mddev and rdev data. */
1906
65a06f06 1907 sb = page_address(rdev->sb_page);
1da177e4
LT		 1908
1909 sb->feature_map = 0;
1910 sb->pad0 = 0;
5fd6c1dc 1911 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1912 memset(sb->pad3, 0, sizeof(sb->pad3));
1913
1914 sb->utime = cpu_to_le64((__u64)mddev->utime);
1915 sb->events = cpu_to_le64(mddev->events);
1916 if (mddev->in_sync)
1917 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
bd18f646
SL		 1918 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1919 sb->resync_offset = cpu_to_le64(MaxSector);
1da177e4
LT		 1920 else
1921 sb->resync_offset = cpu_to_le64(0);
1922
1c05b4bc 1923 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1924
f0ca340c 1925 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1926 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1927 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1928 sb->level = cpu_to_le32(mddev->level);
1929 sb->layout = cpu_to_le32(mddev->layout);
688834e6
N		 1930 if (test_bit(FailFast, &rdev->flags))
1931 sb->devflags |= FailFast1;
1932 else
1933 sb->devflags &= ~FailFast1;
f0ca340c 1934
aeb9b211
N		 1935 if (test_bit(WriteMostly, &rdev->flags))
1936 sb->devflags |= WriteMostly1;
1937 else
1938 sb->devflags &= ~WriteMostly1;
c6563a8c
N		 1939 sb->data_offset = cpu_to_le64(rdev->data_offset);
1940 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1941
c3d9714e
N		 1942 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1943 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1944 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1945 }
5fd6c1dc 1946
f2076e7d 1947 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
97e4f42d 1948 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1949 sb->feature_map |=
1950 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1951 sb->recovery_offset =
1952 cpu_to_le64(rdev->recovery_offset);
f466722c
N		 1953 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1954 sb->feature_map |=
1955 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
5fd6c1dc 1956 }
3069aa8d
SL		 1957 /* Note: recovery_offset and journal_tail share space */
1958 if (test_bit(Journal, &rdev->flags))
1959 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
2d78f8c4
N		 1960 if (test_bit(Replacement, &rdev->flags))
1961 sb->feature_map |=
1962 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1963
f6705578
N		 1964 if (mddev->reshape_position != MaxSector) {
1965 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1966 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1967 sb->new_layout = cpu_to_le32(mddev->new_layout);
1968 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1969 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1970 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N		 1971 if (mddev->delta_disks == 0 &&
1972 mddev->reshape_backwards)
1973 sb->feature_map
1974 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N		 1975 if (rdev->new_data_offset != rdev->data_offset) {
1976 sb->feature_map
1977 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1978 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1979 - rdev->data_offset));
1980 }
f6705578 1981 }
a654b9d8 1982
3c462c88
GR		 1983 if (mddev_is_clustered(mddev))
1984 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1985
2699b672
N		 1986 if (rdev->badblocks.count == 0)
1987 /* Nothing to do for bad blocks*/ ;
1988 else if (sb->bblog_offset == 0)
1989 /* Cannot record bad blocks on this device */
1990 md_error(mddev, rdev);
1991 else {
1992 struct badblocks *bb = &rdev->badblocks;
ae50640b 1993 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
2699b672
N		 1994 u64 *p = bb->page;
1995 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1996 if (bb->changed) {
1997 unsigned seq;
1998
1999retry:
2000 seq = read_seqbegin(&bb->lock);
2001
2002 memset(bbp, 0xff, PAGE_SIZE);
2003
2004 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 2005 u64 internal_bb = p[i];
2699b672
N		 2006 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2007 | BB_LEN(internal_bb));
35f9ac2d 2008 bbp[i] = cpu_to_le64(store_bb);
2699b672 2009 }
d0962936 2010 bb->changed = 0;
2699b672
N		 2011 if (read_seqretry(&bb->lock, seq))
2012 goto retry;
2013
2014 bb->sector = (rdev->sb_start +
2015 (int)le32_to_cpu(sb->bblog_offset));
2016 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N		 2017 }
2018 }
2019
1da177e4 2020 max_dev = 0;
dafb20fa 2021 rdev_for_each(rdev2, mddev)
1da177e4
LT		 2022 if (rdev2->desc_nr+1 > max_dev)
2023 max_dev = rdev2->desc_nr+1;
a778b73f 2024
70471daf
N		 2025 if (max_dev > le32_to_cpu(sb->max_dev)) {
2026 int bmask;
a778b73f 2027 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 2028 rdev->sb_size = max_dev * 2 + 256;
2029 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2030 if (rdev->sb_size & bmask)
2031 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 2032 } else
2033 max_dev = le32_to_cpu(sb->max_dev);
2034
1da177e4 2035 for (i=0; i<max_dev;i++)
8df72024 2036 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
f72ffdd6 2037
a97b7896
SL		 2038 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2039 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
f72ffdd6 2040
ea0213e0 2041 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
ddc08823
PB		 2042 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2043 sb->feature_map |=
2044 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2045 else
2046 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
ea0213e0
AP		 2047 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2048 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2049 }
2050
dafb20fa 2051 rdev_for_each(rdev2, mddev) {
1da177e4 2052 i = rdev2->desc_nr;
b2d444d7 2053 if (test_bit(Faulty, &rdev2->flags))
c4d4c91b 2054 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
b2d444d7 2055 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 2056 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
a97b7896 2057 else if (test_bit(Journal, &rdev2->flags))
bac624f3 2058 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
93be75ff 2059 else if (rdev2->raid_disk >= 0)
5fd6c1dc 2060 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4 2061 else
c4d4c91b 2062 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1da177e4
LT		 2063 }
2064
1da177e4
LT		 2065 sb->sb_csum = calc_sb_1_csum(sb);
2066}
2067
0cd17fec 2068static unsigned long long
3cb03002 2069super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 2070{
2071 struct mdp_superblock_1 *sb;
15f4a5fd 2072 sector_t max_sectors;
58c0fed4 2073 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 2074 return 0; /* component must fit device */
c6563a8c
N		 2075 if (rdev->data_offset != rdev->new_data_offset)
2076 return 0; /* too confusing */
0f420358 2077 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 2078 /* minor versions 1 and 2; superblock before data */
77304d2a 2079 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 2080 max_sectors -= rdev->data_offset;
2081 if (!num_sectors || num_sectors > max_sectors)
2082 num_sectors = max_sectors;
c3d9714e 2083 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 2084 /* minor version 0 with bitmap we can't move */
2085 return 0;
2086 } else {
2087 /* minor version 0; superblock after data */
0f420358 2088 sector_t sb_start;
77304d2a 2089 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 2090 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 2091 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 2092 if (!num_sectors || num_sectors > max_sectors)
2093 num_sectors = max_sectors;
0f420358 2094 rdev->sb_start = sb_start;
0cd17fec 2095 }
65a06f06 2096 sb = page_address(rdev->sb_page);
15f4a5fd 2097 sb->data_size = cpu_to_le64(num_sectors);
3fb632e4 2098 sb->super_offset = cpu_to_le64(rdev->sb_start);
0cd17fec 2099 sb->sb_csum = calc_sb_1_csum(sb);
46533ff7
N		 2100 do {
2101 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2102 rdev->sb_page);
2103 } while (md_super_wait(rdev->mddev) < 0);
c26a44ed 2104 return num_sectors;
c6563a8c
N		 2105
2106}
2107
2108static int
2109super_1_allow_new_offset(struct md_rdev *rdev,
2110 unsigned long long new_offset)
2111{
2112 /* All necessary checks on new >= old have been done */
2113 struct bitmap *bitmap;
2114 if (new_offset >= rdev->data_offset)
2115 return 1;
2116
2117 /* with 1.0 metadata, there is no metadata to tread on
2118 * so we can always move back */
2119 if (rdev->mddev->minor_version == 0)
2120 return 1;
2121
2122 /* otherwise we must be sure not to step on
2123 * any metadata, so stay:
2124 * 36K beyond start of superblock
2125 * beyond end of badblocks
2126 * beyond write-intent bitmap
2127 */
2128 if (rdev->sb_start + (32+4)*2 > new_offset)
2129 return 0;
2130 bitmap = rdev->mddev->bitmap;
2131 if (bitmap && !rdev->mddev->bitmap_info.file &&
2132 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 2133 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N		 2134 return 0;
2135 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2136 return 0;
2137
2138 return 1;
0cd17fec 2139}
1da177e4 2140
75c96f85 2141static struct super_type super_types[] = {
1da177e4
LT		 2142 [0] = {
2143 .name = "0.90.0",
2144 .owner = THIS_MODULE,
0cd17fec
CW		 2145 .load_super = super_90_load,
2146 .validate_super = super_90_validate,
2147 .sync_super = super_90_sync,
2148 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 2149 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT		 2150 },
2151 [1] = {
2152 .name = "md-1",
2153 .owner = THIS_MODULE,
0cd17fec
CW		 2154 .load_super = super_1_load,
2155 .validate_super = super_1_validate,
2156 .sync_super = super_1_sync,
2157 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 2158 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT		 2159 },
2160};
1da177e4 2161
fd01b88c 2162static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 2163{
2164 if (mddev->sync_super) {
2165 mddev->sync_super(mddev, rdev);
2166 return;
2167 }
2168
2169 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2170
2171 super_types[mddev->major_version].sync_super(mddev, rdev);
2172}
2173
fd01b88c 2174static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 2175{
3cb03002 2176 struct md_rdev *rdev, *rdev2;
1da177e4 2177
4b80991c 2178 rcu_read_lock();
0b020e85
SL		 2179 rdev_for_each_rcu(rdev, mddev1) {
2180 if (test_bit(Faulty, &rdev->flags) ||
2181 test_bit(Journal, &rdev->flags) ||
2182 rdev->raid_disk == -1)
2183 continue;
2184 rdev_for_each_rcu(rdev2, mddev2) {
2185 if (test_bit(Faulty, &rdev2->flags) ||
2186 test_bit(Journal, &rdev2->flags) ||
2187 rdev2->raid_disk == -1)
2188 continue;
7dd5e7c3 2189 if (rdev->bdev->bd_contains ==
4b80991c
N		 2190 rdev2->bdev->bd_contains) {
2191 rcu_read_unlock();
7dd5e7c3 2192 return 1;
4b80991c 2193 }
0b020e85
SL		 2194 }
2195 }
4b80991c 2196 rcu_read_unlock();
1da177e4
LT		 2197 return 0;
2198}
2199
2200static LIST_HEAD(pending_raid_disks);
2201
ac5e7113
AN		 2202/*
2203 * Try to register data integrity profile for an mddev
2204 *
2205 * This is called when an array is started and after a disk has been kicked
2206 * from the array. It only succeeds if all working and active component devices
2207 * are integrity capable with matching profiles.
2208 */
fd01b88c 2209int md_integrity_register(struct mddev *mddev)
ac5e7113 2210{
3cb03002 2211 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 2212
2213 if (list_empty(&mddev->disks))
2214 return 0; /* nothing to do */
629acb6a
JB		 2215 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2216 return 0; /* shouldn't register, or already is */
dafb20fa 2217 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 2218 /* skip spares and non-functional disks */
2219 if (test_bit(Faulty, &rdev->flags))
2220 continue;
2221 if (rdev->raid_disk < 0)
2222 continue;
ac5e7113
AN		 2223 if (!reference) {
2224 /* Use the first rdev as the reference */
2225 reference = rdev;
2226 continue;
2227 }
2228 /* does this rdev's profile match the reference profile? */
2229 if (blk_integrity_compare(reference->bdev->bd_disk,
2230 rdev->bdev->bd_disk) < 0)
2231 return -EINVAL;
2232 }
89078d57
MP		 2233 if (!reference || !bdev_get_integrity(reference->bdev))
2234 return 0;
ac5e7113
AN		 2235 /*
2236 * All component devices are integrity capable and have matching
2237 * profiles, register the common profile for the md device.
2238 */
25520d55
MP		 2239 blk_integrity_register(mddev->gendisk,
2240 bdev_get_integrity(reference->bdev));
2241
9d48739e 2242 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
afeee514 2243 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
9d48739e 2244 pr_err("md: failed to create integrity pool for %s\n",
a91a2785
MP		 2245 mdname(mddev));
2246 return -EINVAL;
2247 }
ac5e7113
AN		 2248 return 0;
2249}
2250EXPORT_SYMBOL(md_integrity_register);
2251
1501efad
DW		 2252/*
2253 * Attempt to add an rdev, but only if it is consistent with the current
2254 * integrity profile
2255 */
2256int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 2257{
2863b9eb 2258 struct blk_integrity *bi_mddev;
1501efad 2259 char name[BDEVNAME_SIZE];
2863b9eb
JB		 2260
2261 if (!mddev->gendisk)
1501efad 2262 return 0;
2863b9eb 2263
2863b9eb 2264 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 2265
ac5e7113 2266 if (!bi_mddev) /* nothing to do */
1501efad
DW		 2267 return 0;
2268
2269 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
9d48739e
N		 2270 pr_err("%s: incompatible integrity profile for %s\n",
2271 mdname(mddev), bdevname(rdev->bdev, name));
1501efad
DW		 2272 return -ENXIO;
2273 }
2274
2275 return 0;
3f9d99c1 2276}
ac5e7113 2277EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2278
f72ffdd6 2279static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
1da177e4 2280{
7dd5e7c3 2281 char b[BDEVNAME_SIZE];
f637b9f9 2282 struct kobject *ko;
5e55e2f5 2283 int err;
1da177e4 2284
11e2ede0
DW		 2285 /* prevent duplicates */
2286 if (find_rdev(mddev, rdev->bdev->bd_dev))
2287 return -EEXIST;
2288
97b20ef7
N		 2289 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2290 mddev->pers)
2291 return -EROFS;
2292
dd8ac336 2293 /* make sure rdev->sectors exceeds mddev->dev_sectors */
f6b6ec5c
SL		 2294 if (!test_bit(Journal, &rdev->flags) &&
2295 rdev->sectors &&
2296 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2297 if (mddev->pers) {
2298 /* Cannot change size, so fail
2299 * If mddev->level <= 0, then we don't care
2300 * about aligning sizes (e.g. linear)
2301 */
2302 if (mddev->level > 0)
2303 return -ENOSPC;
2304 } else
dd8ac336 2305 mddev->dev_sectors = rdev->sectors;
2bf071bf 2306 }
1da177e4
LT		 2307
2308 /* Verify rdev->desc_nr is unique.
2309 * If it is -1, assign a free number, else
2310 * check number is not in use
2311 */
4878e9eb 2312 rcu_read_lock();
1da177e4
LT		 2313 if (rdev->desc_nr < 0) {
2314 int choice = 0;
4878e9eb
N		 2315 if (mddev->pers)
2316 choice = mddev->raid_disks;
57d051dc 2317 while (md_find_rdev_nr_rcu(mddev, choice))
1da177e4
LT		 2318 choice++;
2319 rdev->desc_nr = choice;
2320 } else {
57d051dc 2321 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
4878e9eb 2322 rcu_read_unlock();
1da177e4 2323 return -EBUSY;
4878e9eb 2324 }
1da177e4 2325 }
4878e9eb 2326 rcu_read_unlock();
f6b6ec5c
SL		 2327 if (!test_bit(Journal, &rdev->flags) &&
2328 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
9d48739e
N		 2329 pr_warn("md: %s: array is limited to %d devices\n",
2330 mdname(mddev), mddev->max_disks);
de01dfad
N		 2331 return -EBUSY;
2332 }
19133a42 2333 bdevname(rdev->bdev,b);
90a9befb 2334 strreplace(b, '/', '!');
649316b2 2335
1da177e4 2336 rdev->mddev = mddev;
9d48739e 2337 pr_debug("md: bind<%s>\n", b);
86e6ffdd 2338
963c555e
GJ		 2339 if (mddev->raid_disks)
2340 mddev_create_wb_pool(mddev, rdev, false);
2341
b2d6db58 2342 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2343 goto fail;
86e6ffdd 2344
0762b8bd 2345 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2346 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2347 /* failure here is OK */;
2348 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2349
4b80991c 2350 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2351 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2352
2353 /* May as well allow recovery to be retried once */
5389042f 2354 mddev->recovery_disabled++;
3f9d99c1 2355
1da177e4 2356 return 0;
5e55e2f5
N		 2357
2358 fail:
9d48739e
N		 2359 pr_warn("md: failed to register dev-%s for %s\n",
2360 b, mdname(mddev));
5e55e2f5 2361 return err;
1da177e4
LT		 2362}
2363
177a99b2 2364static void md_delayed_delete(struct work_struct *ws)
5792a285 2365{
3cb03002 2366 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2367 kobject_del(&rdev->kobj);
177a99b2 2368 kobject_put(&rdev->kobj);
5792a285
N		 2369}
2370
f72ffdd6 2371static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2372{
2373 char b[BDEVNAME_SIZE];
403df478 2374
49731baa 2375 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2376 list_del_rcu(&rdev->same_set);
9d48739e 2377 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
963c555e 2378 mddev_destroy_wb_pool(rdev->mddev, rdev);
1da177e4 2379 rdev->mddev = NULL;
86e6ffdd 2380 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2381 sysfs_put(rdev->sysfs_state);
2382 rdev->sysfs_state = NULL;
2230dfe4 2383 rdev->badblocks.count = 0;
5792a285 2384 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2385 * writing to 'remove' to "dev/state". We also need
2386 * to delay it due to rcu usage.
5792a285 2387 */
4b80991c 2388 synchronize_rcu();
177a99b2
N		 2389 INIT_WORK(&rdev->del_work, md_delayed_delete);
2390 kobject_get(&rdev->kobj);
e804ac78 2391 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2392}
2393
2394/*
2395 * prevent the device from being mounted, repartitioned or
2396 * otherwise reused by a RAID array (or any other kernel
2397 * subsystem), by bd_claiming the device.
2398 */
3cb03002 2399static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2400{
2401 int err = 0;
2402 struct block_device *bdev;
2403 char b[BDEVNAME_SIZE];
2404
d4d77629 2405 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2406 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4 2407 if (IS_ERR(bdev)) {
9d48739e 2408 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
1da177e4
LT		 2409 return PTR_ERR(bdev);
2410 }
1da177e4
LT		 2411 rdev->bdev = bdev;
2412 return err;
2413}
2414
3cb03002 2415static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2416{
2417 struct block_device *bdev = rdev->bdev;
2418 rdev->bdev = NULL;
e525fd89 2419 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2420}
2421
2422void md_autodetect_dev(dev_t dev);
2423
f72ffdd6 2424static void export_rdev(struct md_rdev *rdev)
1da177e4
LT		 2425{
2426 char b[BDEVNAME_SIZE];
403df478 2427
9d48739e 2428 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
545c8795 2429 md_rdev_clear(rdev);
1da177e4 2430#ifndef MODULE
d0fae18f
N		 2431 if (test_bit(AutoDetected, &rdev->flags))
2432 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2433#endif
2434 unlock_rdev(rdev);
86e6ffdd 2435 kobject_put(&rdev->kobj);
1da177e4
LT		 2436}
2437
fb56dfef 2438void md_kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT		 2439{
2440 unbind_rdev_from_array(rdev);
2441 export_rdev(rdev);
2442}
fb56dfef 2443EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
1da177e4 2444
fd01b88c 2445static void export_array(struct mddev *mddev)
1da177e4 2446{
0638bb0e 2447 struct md_rdev *rdev;
1da177e4 2448
0638bb0e
N		 2449 while (!list_empty(&mddev->disks)) {
2450 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2451 same_set);
fb56dfef 2452 md_kick_rdev_from_array(rdev);
1da177e4 2453 }
1da177e4
LT		 2454 mddev->raid_disks = 0;
2455 mddev->major_version = 0;
2456}
2457
6497709b
N		 2458static bool set_in_sync(struct mddev *mddev)
2459{
efa4b77b 2460 lockdep_assert_held(&mddev->lock);
4ad23a97
N		 2461 if (!mddev->in_sync) {
2462 mddev->sync_checkers++;
2463 spin_unlock(&mddev->lock);
2464 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2465 spin_lock(&mddev->lock);
2466 if (!mddev->in_sync &&
2467 percpu_ref_is_zero(&mddev->writes_pending)) {
6497709b 2468 mddev->in_sync = 1;
4ad23a97
N		 2469 /*
2470 * Ensure ->in_sync is visible before we clear
2471 * ->sync_checkers.
2472 */
55cc39f3 2473 smp_mb();
6497709b
N		 2474 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2475 sysfs_notify_dirent_safe(mddev->sysfs_state);
2476 }
4ad23a97
N		 2477 if (--mddev->sync_checkers == 0)
2478 percpu_ref_switch_to_percpu(&mddev->writes_pending);
6497709b
N		 2479 }
2480 if (mddev->safemode == 1)
2481 mddev->safemode = 0;
2482 return mddev->in_sync;
2483}
2484
f72ffdd6 2485static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2486{
42543769
N		 2487 /* Update each superblock (in-memory image), but
2488 * if we are allowed to, skip spares which already
2489 * have the right event counter, or have one earlier
2490 * (which would mean they aren't being marked as dirty
2491 * with the rest of the array)
2492 */
3cb03002 2493 struct md_rdev *rdev;
dafb20fa 2494 rdev_for_each(rdev, mddev) {
42543769
N		 2495 if (rdev->sb_events == mddev->events ||
2496 (nospares &&
2497 rdev->raid_disk < 0 &&
42543769
N		 2498 rdev->sb_events+1 == mddev->events)) {
2499 /* Don't update this superblock */
2500 rdev->sb_loaded = 2;
2501 } else {
076f968b 2502 sync_super(mddev, rdev);
42543769
N		 2503 rdev->sb_loaded = 1;
2504 }
1da177e4
LT		 2505 }
2506}
2507
2aa82191
GR		 2508static bool does_sb_need_changing(struct mddev *mddev)
2509{
2510 struct md_rdev *rdev;
2511 struct mdp_superblock_1 *sb;
2512 int role;
2513
2514 /* Find a good rdev */
2515 rdev_for_each(rdev, mddev)
2516 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2517 break;
2518
2519 /* No good device found. */
2520 if (!rdev)
2521 return false;
2522
2523 sb = page_address(rdev->sb_page);
2524 /* Check if a device has become faulty or a spare become active */
2525 rdev_for_each(rdev, mddev) {
2526 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2527 /* Device activated? */
2528 if (role == 0xffff && rdev->raid_disk >=0 &&
2529 !test_bit(Faulty, &rdev->flags))
2530 return true;
2531 /* Device turned faulty? */
2532 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2533 return true;
2534 }
2535
2536 /* Check if any mddev parameters have changed */
2537 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2538 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
13459213 2539 (mddev->layout != le32_to_cpu(sb->layout)) ||
2aa82191
GR		 2540 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2541 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2542 return true;
2543
2544 return false;
2545}
2546
1aee41f6 2547void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2548{
3cb03002 2549 struct md_rdev *rdev;
06d91a5f 2550 int sync_req;
42543769 2551 int nospares = 0;
2699b672 2552 int any_badblocks_changed = 0;
23b63f9f 2553 int ret = -1;
1da177e4 2554
d87f064f
N		 2555 if (mddev->ro) {
2556 if (force_change)
2953079c 2557 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
d87f064f
N		 2558 return;
2559 }
2aa82191 2560
2c97cf13 2561repeat:
2aa82191 2562 if (mddev_is_clustered(mddev)) {
2953079c 2563 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2aa82191 2564 force_change = 1;
2953079c 2565 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
85ad1d13 2566 nospares = 1;
23b63f9f 2567 ret = md_cluster_ops->metadata_update_start(mddev);
2aa82191
GR		 2568 /* Has someone else has updated the sb */
2569 if (!does_sb_need_changing(mddev)) {
23b63f9f
GJ		 2570 if (ret == 0)
2571 md_cluster_ops->metadata_update_cancel(mddev);
2953079c
SL		 2572 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2573 BIT(MD_SB_CHANGE_DEVS) |
2574 BIT(MD_SB_CHANGE_CLEAN));
2aa82191
GR		 2575 return;
2576 }
2577 }
2c97cf13 2578
db0505d3
N		 2579 /*
2580 * First make sure individual recovery_offsets are correct
2581 * curr_resync_completed can only be used during recovery.
2582 * During reshape/resync it might use array-addresses rather
2583 * that device addresses.
2584 */
dafb20fa 2585 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2586 if (rdev->raid_disk >= 0 &&
2587 mddev->delta_disks >= 0 &&
db0505d3
N		 2588 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2589 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2590 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
f2076e7d 2591 !test_bit(Journal, &rdev->flags) &&
3a3a5ddb
N		 2592 !test_bit(In_sync, &rdev->flags) &&
2593 mddev->curr_resync_completed > rdev->recovery_offset)
2594 rdev->recovery_offset = mddev->curr_resync_completed;
2595
f72ffdd6 2596 }
bd52b746 2597 if (!mddev->persistent) {
2953079c
SL		 2598 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2599 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
de393cde 2600 if (!mddev->external) {
2953079c 2601 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
dafb20fa 2602 rdev_for_each(rdev, mddev) {
de393cde 2603 if (rdev->badblocks.changed) {
d0962936 2604 rdev->badblocks.changed = 0;
fc974ee2 2605 ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2606 md_error(mddev, rdev);
2607 }
2608 clear_bit(Blocked, &rdev->flags);
2609 clear_bit(BlockedBadBlocks, &rdev->flags);
2610 wake_up(&rdev->blocked_wait);
2611 }
2612 }
3a3a5ddb
N		 2613 wake_up(&mddev->sb_wait);
2614 return;
2615 }
2616
85572d7c 2617 spin_lock(&mddev->lock);
84692195 2618
9ebc6ef1 2619 mddev->utime = ktime_get_real_seconds();
3a3a5ddb 2620
2953079c 2621 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
850b2b42 2622 force_change = 1;
2953079c 2623 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
850b2b42
N		 2624 /* just a clean<-> dirty transition, possibly leave spares alone,
2625 * though if events isn't the right even/odd, we will have to do
2626 * spares after all
2627 */
2628 nospares = 1;
2629 if (force_change)
2630 nospares = 0;
2631 if (mddev->degraded)
84692195
N		 2632 /* If the array is degraded, then skipping spares is both
2633 * dangerous and fairly pointless.
2634 * Dangerous because a device that was removed from the array
2635 * might have a event_count that still looks up-to-date,
2636 * so it can be re-added without a resync.
2637 * Pointless because if there are any spares to skip,
2638 * then a recovery will happen and soon that array won't
2639 * be degraded any more and the spare can go back to sleep then.
2640 */
850b2b42 2641 nospares = 0;
84692195 2642
06d91a5f 2643 sync_req = mddev->in_sync;
42543769
N		 2644
2645 /* If this is just a dirty<->clean transition, and the array is clean
2646 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2647 if (nospares
42543769 2648 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2649 && mddev->can_decrease_events
2650 && mddev->events != 1) {
42543769 2651 mddev->events--;
a8707c08
N		 2652 mddev->can_decrease_events = 0;
2653 } else {
42543769
N		 2654 /* otherwise we have to go forward and ... */
2655 mddev->events ++;
a8707c08 2656 mddev->can_decrease_events = nospares;
42543769 2657 }
1da177e4 2658
403df478
N		 2659 /*
2660 * This 64-bit counter should never wrap.
2661 * Either we are in around ~1 trillion A.C., assuming
2662 * 1 reboot per second, or we have a bug...
2663 */
2664 WARN_ON(mddev->events == 0);
2699b672 2665
dafb20fa 2666 rdev_for_each(rdev, mddev) {
2699b672
N		 2667 if (rdev->badblocks.changed)
2668 any_badblocks_changed++;
de393cde
N		 2669 if (test_bit(Faulty, &rdev->flags))
2670 set_bit(FaultRecorded, &rdev->flags);
2671 }
2699b672 2672
e691063a 2673 sync_sbs(mddev, nospares);
85572d7c 2674 spin_unlock(&mddev->lock);
1da177e4 2675
36a4e1fe
N		 2676 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2677 mdname(mddev), mddev->in_sync);
1da177e4 2678
504634f6
SL		 2679 if (mddev->queue)
2680 blk_add_trace_msg(mddev->queue, "md md_update_sb");
46533ff7 2681rewrite:
e64e4018 2682 md_bitmap_update_sb(mddev->bitmap);
dafb20fa 2683 rdev_for_each(rdev, mddev) {
1da177e4 2684 char b[BDEVNAME_SIZE];
36a4e1fe 2685
42543769
N		 2686 if (rdev->sb_loaded != 1)
2687 continue; /* no noise on spare devices */
1da177e4 2688
f466722c 2689 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2690 md_super_write(mddev,rdev,
0f420358 2691 rdev->sb_start, rdev->sb_size,
7bfa19f2 2692 rdev->sb_page);
36a4e1fe
N		 2693 pr_debug("md: (write) %s's sb offset: %llu\n",
2694 bdevname(rdev->bdev, b),
2695 (unsigned long long)rdev->sb_start);
42543769 2696 rdev->sb_events = mddev->events;
2699b672
N		 2697 if (rdev->badblocks.size) {
2698 md_super_write(mddev, rdev,
2699 rdev->badblocks.sector,
2700 rdev->badblocks.size << 9,
2701 rdev->bb_page);
2702 rdev->badblocks.size = 0;
2703 }
7bfa19f2 2704
f466722c 2705 } else
36a4e1fe
N		 2706 pr_debug("md: %s (skipping faulty)\n",
2707 bdevname(rdev->bdev, b));
d70ed2e4 2708
7bfa19f2 2709 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2710 /* only need to write one superblock... */
2711 break;
2712 }
46533ff7
N		 2713 if (md_super_wait(mddev) < 0)
2714 goto rewrite;
2953079c 2715 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2716
2c97cf13
GJ		 2717 if (mddev_is_clustered(mddev) && ret == 0)
2718 md_cluster_ops->metadata_update_finish(mddev);
2719
850b2b42 2720 if (mddev->in_sync != sync_req ||
2953079c
SL		 2721 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2722 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
06d91a5f 2723 /* have to write it out again */
06d91a5f 2724 goto repeat;
3d310eb7 2725 wake_up(&mddev->sb_wait);
acb180b0
N		 2726 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2727 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2728
dafb20fa 2729 rdev_for_each(rdev, mddev) {
de393cde
N		 2730 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2731 clear_bit(Blocked, &rdev->flags);
2732
2733 if (any_badblocks_changed)
fc974ee2 2734 ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2735 clear_bit(BlockedBadBlocks, &rdev->flags);
2736 wake_up(&rdev->blocked_wait);
2737 }
1da177e4 2738}
1aee41f6 2739EXPORT_SYMBOL(md_update_sb);
1da177e4 2740
a6da4ef8
GR		 2741static int add_bound_rdev(struct md_rdev *rdev)
2742{
2743 struct mddev *mddev = rdev->mddev;
2744 int err = 0;
87d4d916 2745 bool add_journal = test_bit(Journal, &rdev->flags);
a6da4ef8 2746
87d4d916 2747 if (!mddev->pers->hot_remove_disk || add_journal) {
a6da4ef8
GR		 2748 /* If there is hot_add_disk but no hot_remove_disk
2749 * then added disks for geometry changes,
2750 * and should be added immediately.
2751 */
2752 super_types[mddev->major_version].
2753 validate_super(mddev, rdev);
87d4d916
SL		 2754 if (add_journal)
2755 mddev_suspend(mddev);
a6da4ef8 2756 err = mddev->pers->hot_add_disk(mddev, rdev);
87d4d916
SL		 2757 if (add_journal)
2758 mddev_resume(mddev);
a6da4ef8 2759 if (err) {
db767672 2760 md_kick_rdev_from_array(rdev);
a6da4ef8
GR		 2761 return err;
2762 }
2763 }
2764 sysfs_notify_dirent_safe(rdev->sysfs_state);
2765
2953079c 2766 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
a6da4ef8
GR		 2767 if (mddev->degraded)
2768 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2769 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2770 md_new_event(mddev);
2771 md_wakeup_thread(mddev->thread);
2772 return 0;
2773}
1da177e4 2774
7f6ce769 2775/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2776 * We want to accept with case. For this we use cmd_match.
2777 */
2778static int cmd_match(const char *cmd, const char *str)
2779{
2780 /* See if cmd, written into a sysfs file, matches
2781 * str. They must either be the same, or cmd can
2782 * have a trailing newline
2783 */
2784 while (*cmd && *str && *cmd == *str) {
2785 cmd++;
2786 str++;
2787 }
2788 if (*cmd == '\n')
2789 cmd++;
2790 if (*str || *cmd)
2791 return 0;
2792 return 1;
2793}
2794
86e6ffdd
N		 2795struct rdev_sysfs_entry {
2796 struct attribute attr;
3cb03002
N		 2797 ssize_t (*show)(struct md_rdev *, char *);
2798 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2799};
2800
2801static ssize_t
3cb03002 2802state_show(struct md_rdev *rdev, char *page)
86e6ffdd 2803{
35b785f7 2804 char *sep = ",";
20a49ff6 2805 size_t len = 0;
6aa7de05 2806 unsigned long flags = READ_ONCE(rdev->flags);
86e6ffdd 2807
758bfc8a 2808 if (test_bit(Faulty, &flags) ||
dcbcb486
TM		 2809 (!test_bit(ExternalBbl, &flags) &&
2810 rdev->badblocks.unacked_exist))
35b785f7
TM		 2811 len += sprintf(page+len, "faulty%s", sep);
2812 if (test_bit(In_sync, &flags))
2813 len += sprintf(page+len, "in_sync%s", sep);
2814 if (test_bit(Journal, &flags))
2815 len += sprintf(page+len, "journal%s", sep);
2816 if (test_bit(WriteMostly, &flags))
2817 len += sprintf(page+len, "write_mostly%s", sep);
758bfc8a 2818 if (test_bit(Blocked, &flags) ||
52c64152 2819 (rdev->badblocks.unacked_exist
35b785f7
TM		 2820 && !test_bit(Faulty, &flags)))
2821 len += sprintf(page+len, "blocked%s", sep);
758bfc8a 2822 if (!test_bit(Faulty, &flags) &&
f2076e7d 2823 !test_bit(Journal, &flags) &&
35b785f7
TM		 2824 !test_bit(In_sync, &flags))
2825 len += sprintf(page+len, "spare%s", sep);
2826 if (test_bit(WriteErrorSeen, &flags))
2827 len += sprintf(page+len, "write_error%s", sep);
2828 if (test_bit(WantReplacement, &flags))
2829 len += sprintf(page+len, "want_replacement%s", sep);
2830 if (test_bit(Replacement, &flags))
2831 len += sprintf(page+len, "replacement%s", sep);
2832 if (test_bit(ExternalBbl, &flags))
2833 len += sprintf(page+len, "external_bbl%s", sep);
688834e6
N		 2834 if (test_bit(FailFast, &flags))
2835 len += sprintf(page+len, "failfast%s", sep);
35b785f7
TM		 2836
2837 if (len)
2838 len -= strlen(sep);
2d78f8c4 2839
86e6ffdd
N		 2840 return len+sprintf(page+len, "\n");
2841}
2842
45dc2de1 2843static ssize_t
3cb03002 2844state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2845{
2846 /* can write
de393cde 2847 * faulty - simulates an error
45dc2de1 2848 * remove - disconnects the device
f655675b
N		 2849 * writemostly - sets write_mostly
2850 * -writemostly - clears write_mostly
de393cde
N		 2851 * blocked - sets the Blocked flags
2852 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2853 * insync - sets Insync providing device isn't active
f466722c
N		 2854 * -insync - clear Insync for a device with a slot assigned,
2855 * so that it gets rebuilt based on bitmap
d7a9d443
N		 2856 * write_error - sets WriteErrorSeen
2857 * -write_error - clears WriteErrorSeen
688834e6 2858 * {,-}failfast - set/clear FailFast
45dc2de1
N		 2859 */
2860 int err = -EINVAL;
2861 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2862 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2863 if (test_bit(Faulty, &rdev->flags))
2864 err = 0;
2865 else
2866 err = -EBUSY;
45dc2de1 2867 } else if (cmd_match(buf, "remove")) {
5d881783
SL		 2868 if (rdev->mddev->pers) {
2869 clear_bit(Blocked, &rdev->flags);
2870 remove_and_add_spares(rdev->mddev, rdev);
2871 }
45dc2de1
N		 2872 if (rdev->raid_disk >= 0)
2873 err = -EBUSY;
2874 else {
fd01b88c 2875 struct mddev *mddev = rdev->mddev;
45dc2de1 2876 err = 0;
a9720903
GJ		 2877 if (mddev_is_clustered(mddev))
2878 err = md_cluster_ops->remove_disk(mddev, rdev);
2879
2880 if (err == 0) {
2881 md_kick_rdev_from_array(rdev);
060b0689 2882 if (mddev->pers) {
2953079c 2883 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
060b0689
N		 2884 md_wakeup_thread(mddev->thread);
2885 }
a9720903
GJ		 2886 md_new_event(mddev);
2887 }
45dc2de1 2888 }
f655675b
N		 2889 } else if (cmd_match(buf, "writemostly")) {
2890 set_bit(WriteMostly, &rdev->flags);
963c555e 2891 mddev_create_wb_pool(rdev->mddev, rdev, false);
f655675b
N		 2892 err = 0;
2893 } else if (cmd_match(buf, "-writemostly")) {
963c555e 2894 mddev_destroy_wb_pool(rdev->mddev, rdev);
f655675b 2895 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2896 err = 0;
2897 } else if (cmd_match(buf, "blocked")) {
2898 set_bit(Blocked, &rdev->flags);
2899 err = 0;
2900 } else if (cmd_match(buf, "-blocked")) {
de393cde 2901 if (!test_bit(Faulty, &rdev->flags) &&
dcbcb486 2902 !test_bit(ExternalBbl, &rdev->flags) &&
7da64a0a 2903 rdev->badblocks.unacked_exist) {
de393cde
N		 2904 /* metadata handler doesn't understand badblocks,
2905 * so we need to fail the device
2906 */
2907 md_error(rdev->mddev, rdev);
2908 }
6bfe0b49 2909 clear_bit(Blocked, &rdev->flags);
de393cde 2910 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2911 wake_up(&rdev->blocked_wait);
2912 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2913 md_wakeup_thread(rdev->mddev->thread);
2914
6d56e278
N		 2915 err = 0;
2916 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2917 set_bit(In_sync, &rdev->flags);
f655675b 2918 err = 0;
688834e6
N		 2919 } else if (cmd_match(buf, "failfast")) {
2920 set_bit(FailFast, &rdev->flags);
2921 err = 0;
2922 } else if (cmd_match(buf, "-failfast")) {
2923 clear_bit(FailFast, &rdev->flags);
2924 err = 0;
f2076e7d
SL		 2925 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2926 !test_bit(Journal, &rdev->flags)) {
e1960f8c
N		 2927 if (rdev->mddev->pers == NULL) {
2928 clear_bit(In_sync, &rdev->flags);
2929 rdev->saved_raid_disk = rdev->raid_disk;
2930 rdev->raid_disk = -1;
2931 err = 0;
2932 }
d7a9d443
N		 2933 } else if (cmd_match(buf, "write_error")) {
2934 set_bit(WriteErrorSeen, &rdev->flags);
2935 err = 0;
2936 } else if (cmd_match(buf, "-write_error")) {
2937 clear_bit(WriteErrorSeen, &rdev->flags);
2938 err = 0;
2d78f8c4
N		 2939 } else if (cmd_match(buf, "want_replacement")) {
2940 /* Any non-spare device that is not a replacement can
2941 * become want_replacement at any time, but we then need to
2942 * check if recovery is needed.
2943 */
2944 if (rdev->raid_disk >= 0 &&
f2076e7d 2945 !test_bit(Journal, &rdev->flags) &&
2d78f8c4
N		 2946 !test_bit(Replacement, &rdev->flags))
2947 set_bit(WantReplacement, &rdev->flags);
2948 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2949 md_wakeup_thread(rdev->mddev->thread);
2950 err = 0;
2951 } else if (cmd_match(buf, "-want_replacement")) {
2952 /* Clearing 'want_replacement' is always allowed.
2953 * Once replacements starts it is too late though.
2954 */
2955 err = 0;
2956 clear_bit(WantReplacement, &rdev->flags);
2957 } else if (cmd_match(buf, "replacement")) {
2958 /* Can only set a device as a replacement when array has not
2959 * yet been started. Once running, replacement is automatic
2960 * from spares, or by assigning 'slot'.
2961 */
2962 if (rdev->mddev->pers)
2963 err = -EBUSY;
2964 else {
2965 set_bit(Replacement, &rdev->flags);
2966 err = 0;
2967 }
2968 } else if (cmd_match(buf, "-replacement")) {
2969 /* Similarly, can only clear Replacement before start */
2970 if (rdev->mddev->pers)
2971 err = -EBUSY;
2972 else {
2973 clear_bit(Replacement, &rdev->flags);
2974 err = 0;
2975 }
a6da4ef8 2976 } else if (cmd_match(buf, "re-add")) {
ee37e621
YY		 2977 if (!rdev->mddev->pers)
2978 err = -EINVAL;
2979 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2980 rdev->saved_raid_disk >= 0) {
97f6cd39
GR		 2981 /* clear_bit is performed _after_ all the devices
2982 * have their local Faulty bit cleared. If any writes
2983 * happen in the meantime in the local node, they
2984 * will land in the local bitmap, which will be synced
2985 * by this node eventually
2986 */
2987 if (!mddev_is_clustered(rdev->mddev) ||
2988 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2989 clear_bit(Faulty, &rdev->flags);
2990 err = add_bound_rdev(rdev);
2991 }
a6da4ef8
GR		 2992 } else
2993 err = -EBUSY;
35b785f7
TM		 2994 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2995 set_bit(ExternalBbl, &rdev->flags);
2996 rdev->badblocks.shift = 0;
2997 err = 0;
2998 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2999 clear_bit(ExternalBbl, &rdev->flags);
3000 err = 0;
45dc2de1 3001 }
00bcb4ac
N		 3002 if (!err)
3003 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 3004 return err ? err : len;
3005}
80ca3a44 3006static struct rdev_sysfs_entry rdev_state =
750f199e 3007__ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 3008
4dbcdc75 3009static ssize_t
3cb03002 3010errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 3011{
3012 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3013}
3014
3015static ssize_t
3cb03002 3016errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75 3017{
4c9309c0
AD		 3018 unsigned int n;
3019 int rv;
3020
3021 rv = kstrtouint(buf, 10, &n);
3022 if (rv < 0)
3023 return rv;
3024 atomic_set(&rdev->corrected_errors, n);
3025 return len;
4dbcdc75
N		 3026}
3027static struct rdev_sysfs_entry rdev_errors =
80ca3a44 3028__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 3029
014236d2 3030static ssize_t
3cb03002 3031slot_show(struct md_rdev *rdev, char *page)
014236d2 3032{
f2076e7d
SL		 3033 if (test_bit(Journal, &rdev->flags))
3034 return sprintf(page, "journal\n");
3035 else if (rdev->raid_disk < 0)
014236d2
N		 3036 return sprintf(page, "none\n");
3037 else
3038 return sprintf(page, "%d\n", rdev->raid_disk);
3039}
3040
3041static ssize_t
3cb03002 3042slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2 3043{
4c9309c0 3044 int slot;
c303da6d 3045 int err;
4c9309c0 3046
f2076e7d
SL		 3047 if (test_bit(Journal, &rdev->flags))
3048 return -EBUSY;
014236d2
N		 3049 if (strncmp(buf, "none", 4)==0)
3050 slot = -1;
4c9309c0
AD		 3051 else {
3052 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3053 if (err < 0)
3054 return err;
3055 }
6c2fce2e 3056 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 3057 /* Setting 'slot' on an active array requires also
3058 * updating the 'rd%d' link, and communicating
3059 * with the personality with ->hot_*_disk.
3060 * For now we only support removing
3061 * failed/spare devices. This normally happens automatically,
3062 * but not when the metadata is externally managed.
3063 */
c303da6d
N		 3064 if (rdev->raid_disk == -1)
3065 return -EEXIST;
3066 /* personality does all needed checks */
01393f3d 3067 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 3068 return -EINVAL;
746d3207
N		 3069 clear_bit(Blocked, &rdev->flags);
3070 remove_and_add_spares(rdev->mddev, rdev);
3071 if (rdev->raid_disk >= 0)
3072 return -EBUSY;
c303da6d
N		 3073 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3074 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 3075 } else if (rdev->mddev->pers) {
6c2fce2e 3076 /* Activating a spare .. or possibly reactivating
6d56e278 3077 * if we ever get bitmaps working here.
6c2fce2e 3078 */
cb01c549 3079 int err;
6c2fce2e
NB		 3080
3081 if (rdev->raid_disk != -1)
3082 return -EBUSY;
3083
c6751b2b
N		 3084 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3085 return -EBUSY;
3086
6c2fce2e
NB		 3087 if (rdev->mddev->pers->hot_add_disk == NULL)
3088 return -EINVAL;
3089
ba1b41b6
N		 3090 if (slot >= rdev->mddev->raid_disks &&
3091 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3092 return -ENOSPC;
3093
6c2fce2e
NB		 3094 rdev->raid_disk = slot;
3095 if (test_bit(In_sync, &rdev->flags))
3096 rdev->saved_raid_disk = slot;
3097 else
3098 rdev->saved_raid_disk = -1;
d30519fc 3099 clear_bit(In_sync, &rdev->flags);
8313b8e5 3100 clear_bit(Bitmap_sync, &rdev->flags);
cb01c549
GR		 3101 err = rdev->mddev->pers->
3102 hot_add_disk(rdev->mddev, rdev);
3103 if (err) {
3104 rdev->raid_disk = -1;
3105 return err;
3106 } else
3107 sysfs_notify_dirent_safe(rdev->sysfs_state);
3108 if (sysfs_link_rdev(rdev->mddev, rdev))
3109 /* failure here is OK */;
6c2fce2e 3110 /* don't wakeup anyone, leave that to userspace. */
c303da6d 3111 } else {
ba1b41b6
N		 3112 if (slot >= rdev->mddev->raid_disks &&
3113 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 3114 return -ENOSPC;
3115 rdev->raid_disk = slot;
3116 /* assume it is working */
c5d79adb
N		 3117 clear_bit(Faulty, &rdev->flags);
3118 clear_bit(WriteMostly, &rdev->flags);
c303da6d 3119 set_bit(In_sync, &rdev->flags);
00bcb4ac 3120 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 3121 }
014236d2
N		 3122 return len;
3123}
3124
014236d2 3125static struct rdev_sysfs_entry rdev_slot =
80ca3a44 3126__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 3127
93c8cad0 3128static ssize_t
3cb03002 3129offset_show(struct md_rdev *rdev, char *page)
93c8cad0 3130{
6961ece4 3131 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 3132}
3133
3134static ssize_t
3cb03002 3135offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 3136{
c6563a8c 3137 unsigned long long offset;
b29bebd6 3138 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 3139 return -EINVAL;
8ed0a521 3140 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 3141 return -EBUSY;
dd8ac336 3142 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 3143 /* Must set offset before size, so overlap checks
3144 * can be sane */
3145 return -EBUSY;
93c8cad0 3146 rdev->data_offset = offset;
25f7fd47 3147 rdev->new_data_offset = offset;
93c8cad0
N		 3148 return len;
3149}
3150
3151static struct rdev_sysfs_entry rdev_offset =
80ca3a44 3152__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 3153
c6563a8c
N		 3154static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3155{
3156 return sprintf(page, "%llu\n",
3157 (unsigned long long)rdev->new_data_offset);
3158}
3159
3160static ssize_t new_offset_store(struct md_rdev *rdev,
3161 const char *buf, size_t len)
3162{
3163 unsigned long long new_offset;
3164 struct mddev *mddev = rdev->mddev;
3165
b29bebd6 3166 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N		 3167 return -EINVAL;
3168
f851b60d
N		 3169 if (mddev->sync_thread ||
3170 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
c6563a8c
N		 3171 return -EBUSY;
3172 if (new_offset == rdev->data_offset)
3173 /* reset is always permitted */
3174 ;
3175 else if (new_offset > rdev->data_offset) {
3176 /* must not push array size beyond rdev_sectors */
3177 if (new_offset - rdev->data_offset
3178 + mddev->dev_sectors > rdev->sectors)
3179 return -E2BIG;
3180 }
3181 /* Metadata worries about other space details. */
3182
3183 /* decreasing the offset is inconsistent with a backwards
3184 * reshape.
3185 */
3186 if (new_offset < rdev->data_offset &&
3187 mddev->reshape_backwards)
3188 return -EINVAL;
3189 /* Increasing offset is inconsistent with forwards
3190 * reshape. reshape_direction should be set to
3191 * 'backwards' first.
3192 */
3193 if (new_offset > rdev->data_offset &&
3194 !mddev->reshape_backwards)
3195 return -EINVAL;
3196
3197 if (mddev->pers && mddev->persistent &&
3198 !super_types[mddev->major_version]
3199 .allow_new_offset(rdev, new_offset))
3200 return -E2BIG;
3201 rdev->new_data_offset = new_offset;
3202 if (new_offset > rdev->data_offset)
3203 mddev->reshape_backwards = 1;
3204 else if (new_offset < rdev->data_offset)
3205 mddev->reshape_backwards = 0;
3206
3207 return len;
3208}
3209static struct rdev_sysfs_entry rdev_new_offset =
3210__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3211
83303b61 3212static ssize_t
3cb03002 3213rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 3214{
dd8ac336 3215 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 3216}
3217
c5d79adb
N		 3218static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3219{
3220 /* check if two start/length pairs overlap */
3221 if (s1+l1 <= s2)
3222 return 0;
3223 if (s2+l2 <= s1)
3224 return 0;
3225 return 1;
3226}
3227
b522adcd
DW		 3228static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3229{
3230 unsigned long long blocks;
3231 sector_t new;
3232
b29bebd6 3233 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW		 3234 return -EINVAL;
3235
3236 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3237 return -EINVAL; /* sector conversion overflow */
3238
3239 new = blocks * 2;
3240 if (new != blocks * 2)
3241 return -EINVAL; /* unsigned long long to sector_t overflow */
3242
3243 *sectors = new;
3244 return 0;
3245}
3246
83303b61 3247static ssize_t
3cb03002 3248rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 3249{
fd01b88c 3250 struct mddev *my_mddev = rdev->mddev;
dd8ac336 3251 sector_t oldsectors = rdev->sectors;
b522adcd 3252 sector_t sectors;
27c529bb 3253
f2076e7d
SL		 3254 if (test_bit(Journal, &rdev->flags))
3255 return -EBUSY;
b522adcd 3256 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 3257 return -EINVAL;
c6563a8c
N		 3258 if (rdev->data_offset != rdev->new_data_offset)
3259 return -EINVAL; /* too confusing */
0cd17fec 3260 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 3261 if (my_mddev->persistent) {
dd8ac336
AN		 3262 sectors = super_types[my_mddev->major_version].
3263 rdev_size_change(rdev, sectors);
3264 if (!sectors)
0cd17fec 3265 return -EBUSY;
dd8ac336 3266 } else if (!sectors)
77304d2a 3267 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 3268 rdev->data_offset;
a6468539
N		 3269 if (!my_mddev->pers->resize)
3270 /* Cannot change size for RAID0 or Linear etc */
3271 return -EINVAL;
0cd17fec 3272 }
dd8ac336 3273 if (sectors < my_mddev->dev_sectors)
7d3c6f87 3274 return -EINVAL; /* component must fit device */
0cd17fec 3275
dd8ac336
AN		 3276 rdev->sectors = sectors;
3277 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N		 3278 /* Need to check that all other rdevs with the same
3279 * ->bdev do not overlap. 'rcu' is sufficient to walk
3280 * the rdev lists safely.
3281 * This check does not provide a hard guarantee, it
3282 * just helps avoid dangerous mistakes.
c5d79adb 3283 */
fd01b88c 3284 struct mddev *mddev;
c5d79adb 3285 int overlap = 0;
159ec1fc 3286 struct list_head *tmp;
c5d79adb 3287
8b1afc3d 3288 rcu_read_lock();
29ac4aa3 3289 for_each_mddev(mddev, tmp) {
3cb03002 3290 struct md_rdev *rdev2;
c5d79adb 3291
dafb20fa 3292 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 3293 if (rdev->bdev == rdev2->bdev &&
3294 rdev != rdev2 &&
3295 overlaps(rdev->data_offset, rdev->sectors,
3296 rdev2->data_offset,
3297 rdev2->sectors)) {
c5d79adb
N		 3298 overlap = 1;
3299 break;
3300 }
c5d79adb
N		 3301 if (overlap) {
3302 mddev_put(mddev);
3303 break;
3304 }
3305 }
8b1afc3d 3306 rcu_read_unlock();
c5d79adb
N		 3307 if (overlap) {
3308 /* Someone else could have slipped in a size
3309 * change here, but doing so is just silly.
dd8ac336 3310 * We put oldsectors back because we *know* it is
c5d79adb
N		 3311 * safe, and trust userspace not to race with
3312 * itself
3313 */
dd8ac336 3314 rdev->sectors = oldsectors;
c5d79adb
N		 3315 return -EBUSY;
3316 }
3317 }
83303b61
N		 3318 return len;
3319}
3320
3321static struct rdev_sysfs_entry rdev_size =
80ca3a44 3322__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3323
3cb03002 3324static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 3325{
3326 unsigned long long recovery_start = rdev->recovery_offset;
3327
3328 if (test_bit(In_sync, &rdev->flags) ||
3329 recovery_start == MaxSector)
3330 return sprintf(page, "none\n");
3331
3332 return sprintf(page, "%llu\n", recovery_start);
3333}
3334
3cb03002 3335static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 3336{
3337 unsigned long long recovery_start;
3338
3339 if (cmd_match(buf, "none"))
3340 recovery_start = MaxSector;
b29bebd6 3341 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW		 3342 return -EINVAL;
3343
3344 if (rdev->mddev->pers &&
3345 rdev->raid_disk >= 0)
3346 return -EBUSY;
3347
3348 rdev->recovery_offset = recovery_start;
3349 if (recovery_start == MaxSector)
3350 set_bit(In_sync, &rdev->flags);
3351 else
3352 clear_bit(In_sync, &rdev->flags);
3353 return len;
3354}
3355
3356static struct rdev_sysfs_entry rdev_recovery_start =
3357__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3358
fc974ee2
VV		 3359/* sysfs access to bad-blocks list.
3360 * We present two files.
3361 * 'bad-blocks' lists sector numbers and lengths of ranges that
3362 * are recorded as bad. The list is truncated to fit within
3363 * the one-page limit of sysfs.
3364 * Writing "sector length" to this file adds an acknowledged
3365 * bad block list.
3366 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3367 * been acknowledged. Writing to this file adds bad blocks
3368 * without acknowledging them. This is largely for testing.
3369 */
3cb03002 3370static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3371{
3372 return badblocks_show(&rdev->badblocks, page, 0);
3373}
3cb03002 3374static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3375{
de393cde
N		 3376 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3377 /* Maybe that ack was all we needed */
3378 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3379 wake_up(&rdev->blocked_wait);
3380 return rv;
16c791a5
N		 3381}
3382static struct rdev_sysfs_entry rdev_bad_blocks =
3383__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3384
3cb03002 3385static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 3386{
3387 return badblocks_show(&rdev->badblocks, page, 1);
3388}
3cb03002 3389static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 3390{
3391 return badblocks_store(&rdev->badblocks, page, len, 1);
3392}
3393static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3394__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3395
664aed04
AP		 3396static ssize_t
3397ppl_sector_show(struct md_rdev *rdev, char *page)
3398{
3399 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3400}
3401
3402static ssize_t
3403ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3404{
3405 unsigned long long sector;
3406
3407 if (kstrtoull(buf, 10, &sector) < 0)
3408 return -EINVAL;
3409 if (sector != (sector_t)sector)
3410 return -EINVAL;
3411
3412 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3413 rdev->raid_disk >= 0)
3414 return -EBUSY;
3415
3416 if (rdev->mddev->persistent) {
3417 if (rdev->mddev->major_version == 0)
3418 return -EINVAL;
3419 if ((sector > rdev->sb_start &&
3420 sector - rdev->sb_start > S16_MAX) ||
3421 (sector < rdev->sb_start &&
3422 rdev->sb_start - sector > -S16_MIN))
3423 return -EINVAL;
3424 rdev->ppl.offset = sector - rdev->sb_start;
3425 } else if (!rdev->mddev->external) {
3426 return -EBUSY;
3427 }
3428 rdev->ppl.sector = sector;
3429 return len;
3430}
3431
3432static struct rdev_sysfs_entry rdev_ppl_sector =
3433__ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3434
3435static ssize_t
3436ppl_size_show(struct md_rdev *rdev, char *page)
3437{
3438 return sprintf(page, "%u\n", rdev->ppl.size);
3439}
3440
3441static ssize_t
3442ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3443{
3444 unsigned int size;
3445
3446 if (kstrtouint(buf, 10, &size) < 0)
3447 return -EINVAL;
3448
3449 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3450 rdev->raid_disk >= 0)
3451 return -EBUSY;
3452
3453 if (rdev->mddev->persistent) {
3454 if (rdev->mddev->major_version == 0)
3455 return -EINVAL;
3456 if (size > U16_MAX)
3457 return -EINVAL;
3458 } else if (!rdev->mddev->external) {
3459 return -EBUSY;
3460 }
3461 rdev->ppl.size = size;
3462 return len;
3463}
3464
3465static struct rdev_sysfs_entry rdev_ppl_size =
3466__ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3467
86e6ffdd
N		 3468static struct attribute *rdev_default_attrs[] = {
3469 &rdev_state.attr,
4dbcdc75 3470 &rdev_errors.attr,
014236d2 3471 &rdev_slot.attr,
93c8cad0 3472 &rdev_offset.attr,
c6563a8c 3473 &rdev_new_offset.attr,
83303b61 3474 &rdev_size.attr,
06e3c817 3475 &rdev_recovery_start.attr,
16c791a5
N		 3476 &rdev_bad_blocks.attr,
3477 &rdev_unack_bad_blocks.attr,
664aed04
AP		 3478 &rdev_ppl_sector.attr,
3479 &rdev_ppl_size.attr,
86e6ffdd
N		 3480 NULL,
3481};
3482static ssize_t
3483rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3484{
3485 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3486 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
86e6ffdd
N		 3487
3488 if (!entry->show)
3489 return -EIO;
758bfc8a 3490 if (!rdev->mddev)
168b305b 3491 return -ENODEV;
758bfc8a 3492 return entry->show(rdev, page);
86e6ffdd
N		 3493}
3494
3495static ssize_t
3496rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3497 const char *page, size_t length)
3498{
3499 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3500 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3501 ssize_t rv;
fd01b88c 3502 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3503
3504 if (!entry->store)
3505 return -EIO;
67463acb
N		 3506 if (!capable(CAP_SYS_ADMIN))
3507 return -EACCES;
c42d3240 3508 rv = mddev ? mddev_lock(mddev) : -ENODEV;
ca388059 3509 if (!rv) {
27c529bb 3510 if (rdev->mddev == NULL)
c42d3240 3511 rv = -ENODEV;
27c529bb
N		 3512 else
3513 rv = entry->store(rdev, page, length);
6a51830e 3514 mddev_unlock(mddev);
ca388059
N		 3515 }
3516 return rv;
86e6ffdd
N		 3517}
3518
3519static void rdev_free(struct kobject *ko)
3520{
3cb03002 3521 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3522 kfree(rdev);
3523}
52cf25d0 3524static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3525 .show = rdev_attr_show,
3526 .store = rdev_attr_store,
3527};
3528static struct kobj_type rdev_ktype = {
3529 .release = rdev_free,
3530 .sysfs_ops = &rdev_sysfs_ops,
3531 .default_attrs = rdev_default_attrs,
3532};
3533
3cb03002 3534int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3535{
3536 rdev->desc_nr = -1;
3537 rdev->saved_raid_disk = -1;
3538 rdev->raid_disk = -1;
3539 rdev->flags = 0;
3540 rdev->data_offset = 0;
c6563a8c 3541 rdev->new_data_offset = 0;
e8bb9a83 3542 rdev->sb_events = 0;
0e3ef49e 3543 rdev->last_read_error = 0;
2699b672
N		 3544 rdev->sb_loaded = 0;
3545 rdev->bb_page = NULL;
e8bb9a83
N		 3546 atomic_set(&rdev->nr_pending, 0);
3547 atomic_set(&rdev->read_errors, 0);
3548 atomic_set(&rdev->corrected_errors, 0);
3549
3550 INIT_LIST_HEAD(&rdev->same_set);
3551 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3552
3553 /* Add space to store bad block list.
3554 * This reserves the space even on arrays where it cannot
3555 * be used - I wonder if that matters
3556 */
fc974ee2 3557 return badblocks_init(&rdev->badblocks, 0);
e8bb9a83
N		 3558}
3559EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3560/*
3561 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3562 *
3563 * mark the device faulty if:
3564 *
3565 * - the device is nonexistent (zero size)
3566 * - the device has no valid superblock
3567 *
3568 * a faulty rdev _never_ has rdev->sb set.
3569 */
3cb03002 3570static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3571{
3572 char b[BDEVNAME_SIZE];
3573 int err;
3cb03002 3574 struct md_rdev *rdev;
1da177e4
LT		 3575 sector_t size;
3576
9ffae0cf 3577 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
9d48739e 3578 if (!rdev)
1da177e4 3579 return ERR_PTR(-ENOMEM);
1da177e4 3580
2230dfe4
N		 3581 err = md_rdev_init(rdev);
3582 if (err)
3583 goto abort_free;
3584 err = alloc_disk_sb(rdev);
3585 if (err)
1da177e4
LT		 3586 goto abort_free;
3587
c5d79adb 3588 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3589 if (err)
3590 goto abort_free;
3591
f9cb074b 3592 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3593
77304d2a 3594 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3595 if (!size) {
9d48739e 3596 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
1da177e4
LT		 3597 bdevname(rdev->bdev,b));
3598 err = -EINVAL;
3599 goto abort_free;
3600 }
3601
3602 if (super_format >= 0) {
3603 err = super_types[super_format].
3604 load_super(rdev, NULL, super_minor);
3605 if (err == -EINVAL) {
9d48739e 3606 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
df968c4e 3607 bdevname(rdev->bdev,b),
9d48739e 3608 super_format, super_minor);
1da177e4
LT		 3609 goto abort_free;
3610 }
3611 if (err < 0) {
9d48739e 3612 pr_warn("md: could not read %s's sb, not importing!\n",
1da177e4
LT		 3613 bdevname(rdev->bdev,b));
3614 goto abort_free;
3615 }
3616 }
6bfe0b49 3617
1da177e4
LT		 3618 return rdev;
3619
3620abort_free:
2699b672
N		 3621 if (rdev->bdev)
3622 unlock_rdev(rdev);
545c8795 3623 md_rdev_clear(rdev);
1da177e4
LT		 3624 kfree(rdev);
3625 return ERR_PTR(err);
3626}
3627
3628/*
3629 * Check a full RAID array for plausibility
3630 */
3631
6a5cb53a 3632static int analyze_sbs(struct mddev *mddev)
1da177e4
LT		 3633{
3634 int i;
3cb03002 3635 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3636 char b[BDEVNAME_SIZE];
3637
3638 freshest = NULL;
dafb20fa 3639 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3640 switch (super_types[mddev->major_version].
3641 load_super(rdev, freshest, mddev->minor_version)) {
3642 case 1:
3643 freshest = rdev;
3644 break;
3645 case 0:
3646 break;
3647 default:
9d48739e 3648 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
1da177e4 3649 bdevname(rdev->bdev,b));
fb56dfef 3650 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3651 }
3652
6a5cb53a
YY		 3653 /* Cannot find a valid fresh disk */
3654 if (!freshest) {
3655 pr_warn("md: cannot find a valid disk\n");
3656 return -EINVAL;
3657 }
3658
1da177e4
LT		 3659 super_types[mddev->major_version].
3660 validate_super(mddev, freshest);
3661
3662 i = 0;
dafb20fa 3663 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3664 if (mddev->max_disks &&
3665 (rdev->desc_nr >= mddev->max_disks ||
3666 i > mddev->max_disks)) {
9d48739e
N		 3667 pr_warn("md: %s: %s: only %d devices permitted\n",
3668 mdname(mddev), bdevname(rdev->bdev, b),
3669 mddev->max_disks);
fb56dfef 3670 md_kick_rdev_from_array(rdev);
de01dfad
N		 3671 continue;
3672 }
1aee41f6 3673 if (rdev != freshest) {
1da177e4
LT		 3674 if (super_types[mddev->major_version].
3675 validate_super(mddev, rdev)) {
9d48739e 3676 pr_warn("md: kicking non-fresh %s from array!\n",
1da177e4 3677 bdevname(rdev->bdev,b));
fb56dfef 3678 md_kick_rdev_from_array(rdev);
1da177e4
LT		 3679 continue;
3680 }
1aee41f6 3681 }
1da177e4
LT		 3682 if (mddev->level == LEVEL_MULTIPATH) {
3683 rdev->desc_nr = i++;
3684 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3685 set_bit(In_sync, &rdev->flags);
f2076e7d
SL		 3686 } else if (rdev->raid_disk >=
3687 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3688 !test_bit(Journal, &rdev->flags)) {
a778b73f
N		 3689 rdev->raid_disk = -1;
3690 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3691 }
3692 }
6a5cb53a
YY		 3693
3694 return 0;
1da177e4
LT		 3695}
3696
72e02075
N		 3697/* Read a fixed-point number.
3698 * Numbers in sysfs attributes should be in "standard" units where
3699 * possible, so time should be in seconds.
f72ffdd6 3700 * However we internally use a a much smaller unit such as
72e02075
N		 3701 * milliseconds or jiffies.
3702 * This function takes a decimal number with a possible fractional
3703 * component, and produces an integer which is the result of
3704 * multiplying that number by 10^'scale'.
3705 * all without any floating-point arithmetic.
3706 */
3707int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3708{
3709 unsigned long result = 0;
3710 long decimals = -1;
3711 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3712 if (*cp == '.')
3713 decimals = 0;
3714 else if (decimals < scale) {
3715 unsigned int value;
3716 value = *cp - '0';
3717 result = result * 10 + value;
3718 if (decimals >= 0)
3719 decimals++;
3720 }
3721 cp++;
3722 }
3723 if (*cp == '\n')
3724 cp++;
3725 if (*cp)
3726 return -EINVAL;
3727 if (decimals < 0)
3728 decimals = 0;
cf891607 3729 *res = result * int_pow(10, scale - decimals);
72e02075
N		 3730 return 0;
3731}
3732
16f17b39 3733static ssize_t
fd01b88c 3734safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3735{
3736 int msec = (mddev->safemode_delay*1000)/HZ;
3737 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3738}
3739static ssize_t
fd01b88c 3740safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3741{
16f17b39 3742 unsigned long msec;
97ce0a7f 3743
28c1b9fd 3744 if (mddev_is_clustered(mddev)) {
9d48739e 3745 pr_warn("md: Safemode is disabled for clustered mode\n");
28c1b9fd
GR		 3746 return -EINVAL;
3747 }
3748
72e02075 3749 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3750 return -EINVAL;
16f17b39
N		 3751 if (msec == 0)
3752 mddev->safemode_delay = 0;
3753 else {
19052c0e 3754 unsigned long old_delay = mddev->safemode_delay;
1b30e66f
N		 3755 unsigned long new_delay = (msec*HZ)/1000;
3756
3757 if (new_delay == 0)
3758 new_delay = 1;
3759 mddev->safemode_delay = new_delay;
3760 if (new_delay < old_delay || old_delay == 0)
3761 mod_timer(&mddev->safemode_timer, jiffies+1);
16f17b39
N		 3762 }
3763 return len;
3764}
3765static struct md_sysfs_entry md_safe_delay =
80ca3a44 3766__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3767
eae1701f 3768static ssize_t
fd01b88c 3769level_show(struct mddev *mddev, char *page)
eae1701f 3770{
36d091f4
N		 3771 struct md_personality *p;
3772 int ret;
3773 spin_lock(&mddev->lock);
3774 p = mddev->pers;
d9d166c2 3775 if (p)
36d091f4 3776 ret = sprintf(page, "%s\n", p->name);
d9d166c2 3777 else if (mddev->clevel[0])
36d091f4 3778 ret = sprintf(page, "%s\n", mddev->clevel);
d9d166c2 3779 else if (mddev->level != LEVEL_NONE)
36d091f4 3780 ret = sprintf(page, "%d\n", mddev->level);
d9d166c2 3781 else
36d091f4
N		 3782 ret = 0;
3783 spin_unlock(&mddev->lock);
3784 return ret;
eae1701f
N		 3785}
3786
d9d166c2 3787static ssize_t
fd01b88c 3788level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3789{
f2859af6 3790 char clevel[16];
6791875e
N		 3791 ssize_t rv;
3792 size_t slen = len;
db721d32 3793 struct md_personality *pers, *oldpers;
f2859af6 3794 long level;
db721d32 3795 void *priv, *oldpriv;
3cb03002 3796 struct md_rdev *rdev;
245f46c2 3797
6791875e
N		 3798 if (slen == 0 || slen >= sizeof(clevel))
3799 return -EINVAL;
3800
3801 rv = mddev_lock(mddev);
3802 if (rv)
3803 return rv;
3804
245f46c2 3805 if (mddev->pers == NULL) {
6791875e
N		 3806 strncpy(mddev->clevel, buf, slen);
3807 if (mddev->clevel[slen-1] == '\n')
3808 slen--;
3809 mddev->clevel[slen] = 0;
245f46c2 3810 mddev->level = LEVEL_NONE;
6791875e
N		 3811 rv = len;
3812 goto out_unlock;
245f46c2 3813 }
6791875e 3814 rv = -EROFS;
bd8839e0 3815 if (mddev->ro)
6791875e 3816 goto out_unlock;
245f46c2
N		 3817
3818 /* request to change the personality. Need to ensure:
3819 * - array is not engaged in resync/recovery/reshape
3820 * - old personality can be suspended
3821 * - new personality will access other array.
3822 */
3823
6791875e 3824 rv = -EBUSY;
bb4f1e9d 3825 if (mddev->sync_thread ||
f851b60d 3826 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
bb4f1e9d
N		 3827 mddev->reshape_position != MaxSector ||
3828 mddev->sysfs_active)
6791875e 3829 goto out_unlock;
245f46c2 3830
6791875e 3831 rv = -EINVAL;
245f46c2 3832 if (!mddev->pers->quiesce) {
9d48739e
N		 3833 pr_warn("md: %s: %s does not support online personality change\n",
3834 mdname(mddev), mddev->pers->name);
6791875e 3835 goto out_unlock;
245f46c2
N		 3836 }
3837
3838 /* Now find the new personality */
6791875e
N		 3839 strncpy(clevel, buf, slen);
3840 if (clevel[slen-1] == '\n')
3841 slen--;
3842 clevel[slen] = 0;
b29bebd6 3843 if (kstrtol(clevel, 10, &level))
f2859af6 3844 level = LEVEL_NONE;
245f46c2 3845
f2859af6
DW		 3846 if (request_module("md-%s", clevel) != 0)
3847 request_module("md-level-%s", clevel);
245f46c2 3848 spin_lock(&pers_lock);
f2859af6 3849 pers = find_pers(level, clevel);
245f46c2
N		 3850 if (!pers || !try_module_get(pers->owner)) {
3851 spin_unlock(&pers_lock);
9d48739e 3852 pr_warn("md: personality %s not loaded\n", clevel);
6791875e
N		 3853 rv = -EINVAL;
3854 goto out_unlock;
245f46c2
N		 3855 }
3856 spin_unlock(&pers_lock);
3857
3858 if (pers == mddev->pers) {
3859 /* Nothing to do! */
3860 module_put(pers->owner);
6791875e
N		 3861 rv = len;
3862 goto out_unlock;
245f46c2
N		 3863 }
3864 if (!pers->takeover) {
3865 module_put(pers->owner);
9d48739e
N		 3866 pr_warn("md: %s: %s does not support personality takeover\n",
3867 mdname(mddev), clevel);
6791875e
N		 3868 rv = -EINVAL;
3869 goto out_unlock;
245f46c2
N		 3870 }
3871
dafb20fa 3872 rdev_for_each(rdev, mddev)
e93f68a1
N		 3873 rdev->new_raid_disk = rdev->raid_disk;
3874
245f46c2
N		 3875 /* ->takeover must set new_* and/or delta_disks
3876 * if it succeeds, and may set them when it fails.
3877 */
3878 priv = pers->takeover(mddev);
3879 if (IS_ERR(priv)) {
3880 mddev->new_level = mddev->level;
3881 mddev->new_layout = mddev->layout;
664e7c41 3882 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3883 mddev->raid_disks -= mddev->delta_disks;
3884 mddev->delta_disks = 0;
2c810cdd 3885 mddev->reshape_backwards = 0;
245f46c2 3886 module_put(pers->owner);
9d48739e
N		 3887 pr_warn("md: %s: %s would not accept array\n",
3888 mdname(mddev), clevel);
6791875e
N		 3889 rv = PTR_ERR(priv);
3890 goto out_unlock;
245f46c2
N		 3891 }
3892
3893 /* Looks like we have a winner */
3894 mddev_suspend(mddev);
5aa61f42 3895 mddev_detach(mddev);
36d091f4
N		 3896
3897 spin_lock(&mddev->lock);
db721d32
N		 3898 oldpers = mddev->pers;
3899 oldpriv = mddev->private;
3900 mddev->pers = pers;
3901 mddev->private = priv;
3902 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3903 mddev->level = mddev->new_level;
3904 mddev->layout = mddev->new_layout;
3905 mddev->chunk_sectors = mddev->new_chunk_sectors;
3906 mddev->delta_disks = 0;
3907 mddev->reshape_backwards = 0;
3908 mddev->degraded = 0;
36d091f4 3909 spin_unlock(&mddev->lock);
db721d32
N		 3910
3911 if (oldpers->sync_request == NULL &&
3912 mddev->external) {
3913 /* We are converting from a no-redundancy array
3914 * to a redundancy array and metadata is managed
3915 * externally so we need to be sure that writes
3916 * won't block due to a need to transition
3917 * clean->dirty
3918 * until external management is started.
3919 */
3920 mddev->in_sync = 0;
3921 mddev->safemode_delay = 0;
3922 mddev->safemode = 0;
3923 }
f72ffdd6 3924
db721d32
N		 3925 oldpers->free(mddev, oldpriv);
3926
3927 if (oldpers->sync_request == NULL &&
a64c876f
N		 3928 pers->sync_request != NULL) {
3929 /* need to add the md_redundancy_group */
3930 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
9d48739e
N		 3931 pr_warn("md: cannot register extra attributes for %s\n",
3932 mdname(mddev));
388975cc 3933 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3934 }
db721d32 3935 if (oldpers->sync_request != NULL &&
a64c876f
N		 3936 pers->sync_request == NULL) {
3937 /* need to remove the md_redundancy_group */
3938 if (mddev->to_remove == NULL)
3939 mddev->to_remove = &md_redundancy_group;
3940 }
3941
4cb9da7d
AO		 3942 module_put(oldpers->owner);
3943
dafb20fa 3944 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3945 if (rdev->raid_disk < 0)
3946 continue;
bf2cb0da 3947 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3948 rdev->new_raid_disk = -1;
3949 if (rdev->new_raid_disk == rdev->raid_disk)
3950 continue;
36fad858 3951 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3952 }
dafb20fa 3953 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3954 if (rdev->raid_disk < 0)
3955 continue;
3956 if (rdev->new_raid_disk == rdev->raid_disk)
3957 continue;
3958 rdev->raid_disk = rdev->new_raid_disk;
3959 if (rdev->raid_disk < 0)
3a981b03 3960 clear_bit(In_sync, &rdev->flags);
e93f68a1 3961 else {
36fad858 3962 if (sysfs_link_rdev(mddev, rdev))
9d48739e
N		 3963 pr_warn("md: cannot register rd%d for %s after level change\n",
3964 rdev->raid_disk, mdname(mddev));
3a981b03 3965 }
e93f68a1
N		 3966 }
3967
db721d32 3968 if (pers->sync_request == NULL) {
9af204cf
TM		 3969 /* this is now an array without redundancy, so
3970 * it must always be in_sync
3971 */
3972 mddev->in_sync = 1;
3973 del_timer_sync(&mddev->safemode_timer);
3974 }
02e5f5c0 3975 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3976 pers->run(mddev);
2953079c 3977 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
47525e59 3978 mddev_resume(mddev);
830778a1
N		 3979 if (!mddev->thread)
3980 md_update_sb(mddev, 1);
5cac7861 3981 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3982 md_new_event(mddev);
6791875e
N		 3983 rv = len;
3984out_unlock:
3985 mddev_unlock(mddev);
d9d166c2
N		 3986 return rv;
3987}
3988
3989static struct md_sysfs_entry md_level =
80ca3a44 3990__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3991
d4dbd025 3992static ssize_t
fd01b88c 3993layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3994{
3995 /* just a number, not meaningful for all levels */
08a02ecd
N		 3996 if (mddev->reshape_position != MaxSector &&
3997 mddev->layout != mddev->new_layout)
3998 return sprintf(page, "%d (%d)\n",
3999 mddev->new_layout, mddev->layout);
d4dbd025
N		 4000 return sprintf(page, "%d\n", mddev->layout);
4001}
4002
4003static ssize_t
fd01b88c 4004layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025 4005{
4c9309c0 4006 unsigned int n;
6791875e 4007 int err;
d4dbd025 4008
4c9309c0
AD		 4009 err = kstrtouint(buf, 10, &n);
4010 if (err < 0)
4011 return err;
6791875e
N		 4012 err = mddev_lock(mddev);
4013 if (err)
4014 return err;
d4dbd025 4015
b3546035 4016 if (mddev->pers) {
50ac168a 4017 if (mddev->pers->check_reshape == NULL)
6791875e
N		 4018 err = -EBUSY;
4019 else if (mddev->ro)
4020 err = -EROFS;
4021 else {
4022 mddev->new_layout = n;
4023 err = mddev->pers->check_reshape(mddev);
4024 if (err)
4025 mddev->new_layout = mddev->layout;
597a711b 4026 }
b3546035 4027 } else {
08a02ecd 4028 mddev->new_layout = n;
b3546035
N		 4029 if (mddev->reshape_position == MaxSector)
4030 mddev->layout = n;
4031 }
6791875e
N		 4032 mddev_unlock(mddev);
4033 return err ?: len;
d4dbd025
N		 4034}
4035static struct md_sysfs_entry md_layout =
80ca3a44 4036__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 4037
eae1701f 4038static ssize_t
fd01b88c 4039raid_disks_show(struct mddev *mddev, char *page)
eae1701f 4040{
bb636547
N		 4041 if (mddev->raid_disks == 0)
4042 return 0;
08a02ecd
N		 4043 if (mddev->reshape_position != MaxSector &&
4044 mddev->delta_disks != 0)
4045 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4046 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 4047 return sprintf(page, "%d\n", mddev->raid_disks);
4048}
4049
fd01b88c 4050static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 4051
4052static ssize_t
fd01b88c 4053raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 4054{
4c9309c0 4055 unsigned int n;
6791875e 4056 int err;
da943b99 4057
4c9309c0
AD		 4058 err = kstrtouint(buf, 10, &n);
4059 if (err < 0)
4060 return err;
da943b99 4061
6791875e
N		 4062 err = mddev_lock(mddev);
4063 if (err)
4064 return err;
da943b99 4065 if (mddev->pers)
6791875e 4066 err = update_raid_disks(mddev, n);
08a02ecd 4067 else if (mddev->reshape_position != MaxSector) {
c6563a8c 4068 struct md_rdev *rdev;
08a02ecd 4069 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c 4070
6791875e 4071 err = -EINVAL;
c6563a8c
N		 4072 rdev_for_each(rdev, mddev) {
4073 if (olddisks < n &&
4074 rdev->data_offset < rdev->new_data_offset)
6791875e 4075 goto out_unlock;
c6563a8c
N		 4076 if (olddisks > n &&
4077 rdev->data_offset > rdev->new_data_offset)
6791875e 4078 goto out_unlock;
c6563a8c 4079 }
6791875e 4080 err = 0;
08a02ecd
N		 4081 mddev->delta_disks = n - olddisks;
4082 mddev->raid_disks = n;
2c810cdd 4083 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 4084 } else
da943b99 4085 mddev->raid_disks = n;
6791875e
N		 4086out_unlock:
4087 mddev_unlock(mddev);
4088 return err ? err : len;
da943b99
N		 4089}
4090static struct md_sysfs_entry md_raid_disks =
80ca3a44 4091__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 4092
3b34380a 4093static ssize_t
fd01b88c 4094chunk_size_show(struct mddev *mddev, char *page)
3b34380a 4095{
08a02ecd 4096 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 4097 mddev->chunk_sectors != mddev->new_chunk_sectors)
4098 return sprintf(page, "%d (%d)\n",
4099 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 4100 mddev->chunk_sectors << 9);
4101 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 4102}
4103
4104static ssize_t
fd01b88c 4105chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 4106{
4c9309c0 4107 unsigned long n;
6791875e 4108 int err;
3b34380a 4109
4c9309c0
AD		 4110 err = kstrtoul(buf, 10, &n);
4111 if (err < 0)
4112 return err;
3b34380a 4113
6791875e
N		 4114 err = mddev_lock(mddev);
4115 if (err)
4116 return err;
b3546035 4117 if (mddev->pers) {
50ac168a 4118 if (mddev->pers->check_reshape == NULL)
6791875e
N		 4119 err = -EBUSY;
4120 else if (mddev->ro)
4121 err = -EROFS;
4122 else {
4123 mddev->new_chunk_sectors = n >> 9;
4124 err = mddev->pers->check_reshape(mddev);
4125 if (err)
4126 mddev->new_chunk_sectors = mddev->chunk_sectors;
597a711b 4127 }
b3546035 4128 } else {
664e7c41 4129 mddev->new_chunk_sectors = n >> 9;
b3546035 4130 if (mddev->reshape_position == MaxSector)
9d8f0363 4131 mddev->chunk_sectors = n >> 9;
b3546035 4132 }
6791875e
N		 4133 mddev_unlock(mddev);
4134 return err ?: len;
3b34380a
N		 4135}
4136static struct md_sysfs_entry md_chunk_size =
80ca3a44 4137__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 4138
a94213b1 4139static ssize_t
fd01b88c 4140resync_start_show(struct mddev *mddev, char *page)
a94213b1 4141{
d1a7c503
N		 4142 if (mddev->recovery_cp == MaxSector)
4143 return sprintf(page, "none\n");
a94213b1
N		 4144 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4145}
4146
4147static ssize_t
fd01b88c 4148resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 4149{
4c9309c0 4150 unsigned long long n;
6791875e 4151 int err;
4c9309c0
AD		 4152
4153 if (cmd_match(buf, "none"))
4154 n = MaxSector;
4155 else {
4156 err = kstrtoull(buf, 10, &n);
4157 if (err < 0)
4158 return err;
4159 if (n != (sector_t)n)
4160 return -EINVAL;
4161 }
a94213b1 4162
6791875e
N		 4163 err = mddev_lock(mddev);
4164 if (err)
4165 return err;
b098636c 4166 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6791875e 4167 err = -EBUSY;
a94213b1 4168
6791875e
N		 4169 if (!err) {
4170 mddev->recovery_cp = n;
4171 if (mddev->pers)
2953079c 4172 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
6791875e
N		 4173 }
4174 mddev_unlock(mddev);
4175 return err ?: len;
a94213b1
N		 4176}
4177static struct md_sysfs_entry md_resync_start =
750f199e
N		 4178__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4179 resync_start_show, resync_start_store);
a94213b1 4180
9e653b63
N		 4181/*
4182 * The array state can be:
4183 *
4184 * clear
4185 * No devices, no size, no level
4186 * Equivalent to STOP_ARRAY ioctl
4187 * inactive
4188 * May have some settings, but array is not active
4189 * all IO results in error
4190 * When written, doesn't tear down array, but just stops it
4191 * suspended (not supported yet)
4192 * All IO requests will block. The array can be reconfigured.
910d8cb3 4193 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 4194 * readonly
4195 * no resync can happen. no superblocks get written.
4196 * write requests fail
4197 * read-auto
4198 * like readonly, but behaves like 'clean' on a write request.
4199 *
4200 * clean - no pending writes, but otherwise active.
4201 * When written to inactive array, starts without resync
4202 * If a write request arrives then
4203 * if metadata is known, mark 'dirty' and switch to 'active'.
4204 * if not known, block and switch to write-pending
4205 * If written to an active array that has pending writes, then fails.
4206 * active
4207 * fully active: IO and resync can be happening.
4208 * When written to inactive array, starts with resync
4209 *
4210 * write-pending
4211 * clean, but writes are blocked waiting for 'active' to be written.
4212 *
4213 * active-idle
4214 * like active, but no writes have been seen for a while (100msec).
4215 *
62f7b198
GP		 4216 * broken
4217 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4218 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4219 * when a member is gone, so this state will at least alert the
4220 * user that something is wrong.
9e653b63
N		 4221 */
4222enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
62f7b198 4223 write_pending, active_idle, broken, bad_word};
05381954 4224static char *array_states[] = {
9e653b63 4225 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
62f7b198 4226 "write-pending", "active-idle", "broken", NULL };
9e653b63
N		 4227
4228static int match_word(const char *word, char **list)
4229{
4230 int n;
4231 for (n=0; list[n]; n++)
4232 if (cmd_match(word, list[n]))
4233 break;
4234 return n;
4235}
4236
4237static ssize_t
fd01b88c 4238array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 4239{
4240 enum array_state st = inactive;
4241
62f7b198 4242 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
9e653b63
N		 4243 switch(mddev->ro) {
4244 case 1:
4245 st = readonly;
4246 break;
4247 case 2:
4248 st = read_auto;
4249 break;
4250 case 0:
55cc39f3 4251 spin_lock(&mddev->lock);
2953079c 4252 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
e691063a 4253 st = write_pending;
16f88949
TM		 4254 else if (mddev->in_sync)
4255 st = clean;
9e653b63
N		 4256 else if (mddev->safemode)
4257 st = active_idle;
4258 else
4259 st = active;
55cc39f3 4260 spin_unlock(&mddev->lock);
9e653b63 4261 }
62f7b198
GP		 4262
4263 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4264 st = broken;
4265 } else {
9e653b63
N		 4266 if (list_empty(&mddev->disks) &&
4267 mddev->raid_disks == 0 &&
58c0fed4 4268 mddev->dev_sectors == 0)
9e653b63
N		 4269 st = clear;
4270 else
4271 st = inactive;
4272 }
4273 return sprintf(page, "%s\n", array_states[st]);
4274}
4275
f72ffdd6
N		 4276static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4277static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4278static int do_md_run(struct mddev *mddev);
fd01b88c 4279static int restart_array(struct mddev *mddev);
9e653b63
N		 4280
4281static ssize_t
fd01b88c 4282array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63 4283{
6497709b 4284 int err = 0;
9e653b63 4285 enum array_state st = match_word(buf, array_states);
6791875e
N		 4286
4287 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4288 /* don't take reconfig_mutex when toggling between
4289 * clean and active
4290 */
4291 spin_lock(&mddev->lock);
4292 if (st == active) {
4293 restart_array(mddev);
2953079c 4294 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
91a6c4ad 4295 md_wakeup_thread(mddev->thread);
6791875e 4296 wake_up(&mddev->sb_wait);
6791875e
N		 4297 } else /* st == clean */ {
4298 restart_array(mddev);
6497709b 4299 if (!set_in_sync(mddev))
6791875e
N		 4300 err = -EBUSY;
4301 }
573275b5
TM		 4302 if (!err)
4303 sysfs_notify_dirent_safe(mddev->sysfs_state);
6791875e 4304 spin_unlock(&mddev->lock);
c008f1d3 4305 return err ?: len;
6791875e
N		 4306 }
4307 err = mddev_lock(mddev);
4308 if (err)
4309 return err;
4310 err = -EINVAL;
9e653b63
N		 4311 switch(st) {
4312 case bad_word:
4313 break;
4314 case clear:
4315 /* stopping an active array */
a05b7ea0 4316 err = do_md_stop(mddev, 0, NULL);
9e653b63
N		 4317 break;
4318 case inactive:
4319 /* stopping an active array */
90cf195d 4320 if (mddev->pers)
a05b7ea0 4321 err = do_md_stop(mddev, 2, NULL);
90cf195d 4322 else
e691063a 4323 err = 0; /* already inactive */
9e653b63
N		 4324 break;
4325 case suspended:
4326 break; /* not supported yet */
4327 case readonly:
4328 if (mddev->pers)
a05b7ea0 4329 err = md_set_readonly(mddev, NULL);
9e653b63
N		 4330 else {
4331 mddev->ro = 1;
648b629e 4332 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 4333 err = do_md_run(mddev);
4334 }
4335 break;
4336 case read_auto:
9e653b63 4337 if (mddev->pers) {
80268ee9 4338 if (mddev->ro == 0)
a05b7ea0 4339 err = md_set_readonly(mddev, NULL);
80268ee9 4340 else if (mddev->ro == 1)
648b629e
N		 4341 err = restart_array(mddev);
4342 if (err == 0) {
4343 mddev->ro = 2;
4344 set_disk_ro(mddev->gendisk, 0);
4345 }
9e653b63
N		 4346 } else {
4347 mddev->ro = 2;
4348 err = do_md_run(mddev);
4349 }
4350 break;
4351 case clean:
4352 if (mddev->pers) {
339421de
SL		 4353 err = restart_array(mddev);
4354 if (err)
4355 break;
85572d7c 4356 spin_lock(&mddev->lock);
6497709b 4357 if (!set_in_sync(mddev))
e691063a 4358 err = -EBUSY;
85572d7c 4359 spin_unlock(&mddev->lock);
5bf29597
N		 4360 } else
4361 err = -EINVAL;
9e653b63
N		 4362 break;
4363 case active:
4364 if (mddev->pers) {
339421de
SL		 4365 err = restart_array(mddev);
4366 if (err)
4367 break;
2953079c 4368 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9e653b63
N		 4369 wake_up(&mddev->sb_wait);
4370 err = 0;
4371 } else {
4372 mddev->ro = 0;
648b629e 4373 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 4374 err = do_md_run(mddev);
4375 }
4376 break;
4377 case write_pending:
4378 case active_idle:
62f7b198 4379 case broken:
9e653b63
N		 4380 /* these cannot be set */
4381 break;
4382 }
6791875e
N		 4383
4384 if (!err) {
1d23f178
N		 4385 if (mddev->hold_active == UNTIL_IOCTL)
4386 mddev->hold_active = 0;
00bcb4ac 4387 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4388 }
6791875e
N		 4389 mddev_unlock(mddev);
4390 return err ?: len;
9e653b63 4391}
80ca3a44 4392static struct md_sysfs_entry md_array_state =
750f199e 4393__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 4394
1e50915f 4395static ssize_t
fd01b88c 4396max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 4397 return sprintf(page, "%d\n",
4398 atomic_read(&mddev->max_corr_read_errors));
4399}
4400
4401static ssize_t
fd01b88c 4402max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f 4403{
4c9309c0
AD		 4404 unsigned int n;
4405 int rv;
1e50915f 4406
4c9309c0
AD		 4407 rv = kstrtouint(buf, 10, &n);
4408 if (rv < 0)
4409 return rv;
4410 atomic_set(&mddev->max_corr_read_errors, n);
4411 return len;
1e50915f
RB		 4412}
4413
4414static struct md_sysfs_entry max_corr_read_errors =
4415__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4416 max_corrected_read_errors_store);
4417
6d7ff738 4418static ssize_t
fd01b88c 4419null_show(struct mddev *mddev, char *page)
6d7ff738
N		 4420{
4421 return -EINVAL;
4422}
4423
4424static ssize_t
fd01b88c 4425new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 4426{
4427 /* buf must be %d:%d\n? giving major and minor numbers */
4428 /* The new device is added to the array.
4429 * If the array has a persistent superblock, we read the
4430 * superblock to initialise info and check validity.
4431 * Otherwise, only checking done is that in bind_rdev_to_array,
4432 * which mainly checks size.
4433 */
4434 char *e;
4435 int major = simple_strtoul(buf, &e, 10);
4436 int minor;
4437 dev_t dev;
3cb03002 4438 struct md_rdev *rdev;
6d7ff738
N		 4439 int err;
4440
4441 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4442 return -EINVAL;
4443 minor = simple_strtoul(e+1, &e, 10);
4444 if (*e && *e != '\n')
4445 return -EINVAL;
4446 dev = MKDEV(major, minor);
4447 if (major != MAJOR(dev) ||
4448 minor != MINOR(dev))
4449 return -EOVERFLOW;
4450
6791875e
N		 4451 flush_workqueue(md_misc_wq);
4452
4453 err = mddev_lock(mddev);
4454 if (err)
4455 return err;
6d7ff738
N		 4456 if (mddev->persistent) {
4457 rdev = md_import_device(dev, mddev->major_version,
4458 mddev->minor_version);
4459 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 4460 struct md_rdev *rdev0
4461 = list_entry(mddev->disks.next,
4462 struct md_rdev, same_set);
6d7ff738
N		 4463 err = super_types[mddev->major_version]
4464 .load_super(rdev, rdev0, mddev->minor_version);
4465 if (err < 0)
4466 goto out;
4467 }
c5d79adb
N		 4468 } else if (mddev->external)
4469 rdev = md_import_device(dev, -2, -1);
4470 else
6d7ff738
N		 4471 rdev = md_import_device(dev, -1, -1);
4472
9a8c0fa8
N		 4473 if (IS_ERR(rdev)) {
4474 mddev_unlock(mddev);
6d7ff738 4475 return PTR_ERR(rdev);
9a8c0fa8 4476 }
6d7ff738
N		 4477 err = bind_rdev_to_array(rdev, mddev);
4478 out:
4479 if (err)
4480 export_rdev(rdev);
6791875e 4481 mddev_unlock(mddev);
5492c46e
AO		 4482 if (!err)
4483 md_new_event(mddev);
6d7ff738
N		 4484 return err ? err : len;
4485}
4486
4487static struct md_sysfs_entry md_new_device =
80ca3a44 4488__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4489
9b1d1dac 4490static ssize_t
fd01b88c 4491bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 4492{
4493 char *end;
4494 unsigned long chunk, end_chunk;
6791875e 4495 int err;
9b1d1dac 4496
6791875e
N		 4497 err = mddev_lock(mddev);
4498 if (err)
4499 return err;
9b1d1dac
PC		 4500 if (!mddev->bitmap)
4501 goto out;
4502 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4503 while (*buf) {
4504 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4505 if (buf == end) break;
4506 if (*end == '-') { /* range */
4507 buf = end + 1;
4508 end_chunk = simple_strtoul(buf, &end, 0);
4509 if (buf == end) break;
4510 }
4511 if (*end && !isspace(*end)) break;
e64e4018 4512 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4513 buf = skip_spaces(end);
9b1d1dac 4514 }
e64e4018 4515 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
9b1d1dac 4516out:
6791875e 4517 mddev_unlock(mddev);
9b1d1dac
PC		 4518 return len;
4519}
4520
4521static struct md_sysfs_entry md_bitmap =
4522__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4523
a35b0d69 4524static ssize_t
fd01b88c 4525size_show(struct mddev *mddev, char *page)
a35b0d69 4526{
58c0fed4
AN		 4527 return sprintf(page, "%llu\n",
4528 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 4529}
4530
fd01b88c 4531static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 4532
4533static ssize_t
fd01b88c 4534size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 4535{
4536 /* If array is inactive, we can reduce the component size, but
4537 * not increase it (except from 0).
4538 * If array is active, we can try an on-line resize
4539 */
b522adcd
DW		 4540 sector_t sectors;
4541 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4542
58c0fed4
AN		 4543 if (err < 0)
4544 return err;
6791875e
N		 4545 err = mddev_lock(mddev);
4546 if (err)
4547 return err;
a35b0d69 4548 if (mddev->pers) {
58c0fed4 4549 err = update_size(mddev, sectors);
4ba1e788
XN		 4550 if (err == 0)
4551 md_update_sb(mddev, 1);
a35b0d69 4552 } else {
58c0fed4
AN		 4553 if (mddev->dev_sectors == 0 ||
4554 mddev->dev_sectors > sectors)
4555 mddev->dev_sectors = sectors;
a35b0d69
N		 4556 else
4557 err = -ENOSPC;
4558 }
6791875e 4559 mddev_unlock(mddev);
a35b0d69
N		 4560 return err ? err : len;
4561}
4562
4563static struct md_sysfs_entry md_size =
80ca3a44 4564__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4565
83f0d77a 4566/* Metadata version.
e691063a
N		 4567 * This is one of
4568 * 'none' for arrays with no metadata (good luck...)
4569 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4570 * or N.M for internally known formats
4571 */
4572static ssize_t
fd01b88c 4573metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4574{
4575 if (mddev->persistent)
4576 return sprintf(page, "%d.%d\n",
4577 mddev->major_version, mddev->minor_version);
e691063a
N		 4578 else if (mddev->external)
4579 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4580 else
4581 return sprintf(page, "none\n");
4582}
4583
4584static ssize_t
fd01b88c 4585metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4586{
4587 int major, minor;
4588 char *e;
6791875e 4589 int err;
ea43ddd8
N		 4590 /* Changing the details of 'external' metadata is
4591 * always permitted. Otherwise there must be
4592 * no devices attached to the array.
4593 */
6791875e
N		 4594
4595 err = mddev_lock(mddev);
4596 if (err)
4597 return err;
4598 err = -EBUSY;
ea43ddd8
N		 4599 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4600 ;
4601 else if (!list_empty(&mddev->disks))
6791875e 4602 goto out_unlock;
8bb93aac 4603
6791875e 4604 err = 0;
8bb93aac
N		 4605 if (cmd_match(buf, "none")) {
4606 mddev->persistent = 0;
e691063a
N		 4607 mddev->external = 0;
4608 mddev->major_version = 0;
4609 mddev->minor_version = 90;
6791875e 4610 goto out_unlock;
e691063a
N		 4611 }
4612 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4613 size_t namelen = len-9;
e691063a
N		 4614 if (namelen >= sizeof(mddev->metadata_type))
4615 namelen = sizeof(mddev->metadata_type)-1;
4616 strncpy(mddev->metadata_type, buf+9, namelen);
4617 mddev->metadata_type[namelen] = 0;
4618 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4619 mddev->metadata_type[--namelen] = 0;
4620 mddev->persistent = 0;
4621 mddev->external = 1;
8bb93aac
N		 4622 mddev->major_version = 0;
4623 mddev->minor_version = 90;
6791875e 4624 goto out_unlock;
8bb93aac
N		 4625 }
4626 major = simple_strtoul(buf, &e, 10);
6791875e 4627 err = -EINVAL;
8bb93aac 4628 if (e==buf || *e != '.')
6791875e 4629 goto out_unlock;
8bb93aac
N		 4630 buf = e+1;
4631 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4632 if (e==buf || (*e && *e != '\n') )
6791875e
N		 4633 goto out_unlock;
4634 err = -ENOENT;
50511da3 4635 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
6791875e 4636 goto out_unlock;
8bb93aac
N		 4637 mddev->major_version = major;
4638 mddev->minor_version = minor;
4639 mddev->persistent = 1;
e691063a 4640 mddev->external = 0;
6791875e
N		 4641 err = 0;
4642out_unlock:
4643 mddev_unlock(mddev);
4644 return err ?: len;
8bb93aac
N		 4645}
4646
4647static struct md_sysfs_entry md_metadata =
750f199e 4648__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4649
24dd469d 4650static ssize_t
fd01b88c 4651action_show(struct mddev *mddev, char *page)
24dd469d 4652{
7eec314d 4653 char *type = "idle";
b7b17c9b
N		 4654 unsigned long recovery = mddev->recovery;
4655 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
b6a9ce68 4656 type = "frozen";
b7b17c9b
N		 4657 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4658 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4659 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
ccfcc3c1 4660 type = "reshape";
b7b17c9b
N		 4661 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4662 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
24dd469d 4663 type = "resync";
b7b17c9b 4664 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
24dd469d
N		 4665 type = "check";
4666 else
4667 type = "repair";
b7b17c9b 4668 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
24dd469d 4669 type = "recover";
985ca973
N		 4670 else if (mddev->reshape_position != MaxSector)
4671 type = "reshape";
24dd469d
N		 4672 }
4673 return sprintf(page, "%s\n", type);
4674}
4675
4676static ssize_t
fd01b88c 4677action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4678{
7eec314d
N		 4679 if (!mddev->pers || !mddev->pers->sync_request)
4680 return -EINVAL;
4681
b6a9ce68
N		 4682
4683 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
56ccc112
N		 4684 if (cmd_match(page, "frozen"))
4685 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4686 else
4687 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
8e8e2518
N		 4688 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4689 mddev_lock(mddev) == 0) {
4690 flush_workqueue(md_misc_wq);
4691 if (mddev->sync_thread) {
4692 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6791875e 4693 md_reap_sync_thread(mddev);
6791875e 4694 }
8e8e2518 4695 mddev_unlock(mddev);
7eec314d 4696 }
312045ee 4697 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
24dd469d 4698 return -EBUSY;
72a23c21 4699 else if (cmd_match(page, "resync"))
56ccc112 4700 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4701 else if (cmd_match(page, "recover")) {
56ccc112 4702 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
72a23c21 4703 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
72a23c21 4704 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4705 int err;
4706 if (mddev->pers->start_reshape == NULL)
4707 return -EINVAL;
6791875e
N		 4708 err = mddev_lock(mddev);
4709 if (!err) {
312045ee
N		 4710 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4711 err = -EBUSY;
4712 else {
4713 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4714 err = mddev->pers->start_reshape(mddev);
4715 }
6791875e
N		 4716 mddev_unlock(mddev);
4717 }
16484bf5
N		 4718 if (err)
4719 return err;
a99ac971 4720 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4721 } else {
bce74dac 4722 if (cmd_match(page, "check"))
7eec314d 4723 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4724 else if (!cmd_match(page, "repair"))
7eec314d 4725 return -EINVAL;
56ccc112 4726 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
7eec314d
N		 4727 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4728 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4729 }
48c26ddc
N		 4730 if (mddev->ro == 2) {
4731 /* A write to sync_action is enough to justify
4732 * canceling read-auto mode
4733 */
4734 mddev->ro = 0;
4735 md_wakeup_thread(mddev->sync_thread);
4736 }
03c902e1 4737 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4738 md_wakeup_thread(mddev->thread);
00bcb4ac 4739 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4740 return len;
4741}
4742
c4a39551 4743static struct md_sysfs_entry md_scan_mode =
750f199e 4744__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
c4a39551
JB		 4745
4746static ssize_t
4747last_sync_action_show(struct mddev *mddev, char *page)
4748{
4749 return sprintf(page, "%s\n", mddev->last_sync_action);
4750}
4751
4752static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4753
9d88883e 4754static ssize_t
fd01b88c 4755mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4756{
4757 return sprintf(page, "%llu\n",
7f7583d4
JM		 4758 (unsigned long long)
4759 atomic64_read(&mddev->resync_mismatches));
9d88883e
N		 4760}
4761
80ca3a44 4762static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4763
88202a0c 4764static ssize_t
fd01b88c 4765sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4766{
4767 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4768 mddev->sync_speed_min ? "local": "system");
4769}
4770
4771static ssize_t
fd01b88c 4772sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4773{
4c9309c0
AD		 4774 unsigned int min;
4775 int rv;
4776
88202a0c 4777 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4778 min = 0;
4779 } else {
4780 rv = kstrtouint(buf, 10, &min);
4781 if (rv < 0)
4782 return rv;
4783 if (min == 0)
4784 return -EINVAL;
88202a0c 4785 }
88202a0c
N		 4786 mddev->sync_speed_min = min;
4787 return len;
4788}
4789
4790static struct md_sysfs_entry md_sync_min =
4791__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4792
4793static ssize_t
fd01b88c 4794sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4795{
4796 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4797 mddev->sync_speed_max ? "local": "system");
4798}
4799
4800static ssize_t
fd01b88c 4801sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c 4802{
4c9309c0
AD		 4803 unsigned int max;
4804 int rv;
4805
88202a0c 4806 if (strncmp(buf, "system", 6)==0) {
4c9309c0
AD		 4807 max = 0;
4808 } else {
4809 rv = kstrtouint(buf, 10, &max);
4810 if (rv < 0)
4811 return rv;
4812 if (max == 0)
4813 return -EINVAL;
88202a0c 4814 }
88202a0c
N		 4815 mddev->sync_speed_max = max;
4816 return len;
4817}
4818
4819static struct md_sysfs_entry md_sync_max =
4820__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4821
d7f3d291 4822static ssize_t
fd01b88c 4823degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4824{
4825 return sprintf(page, "%d\n", mddev->degraded);
4826}
4827static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4828
90b08710 4829static ssize_t
fd01b88c 4830sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4831{
4832 return sprintf(page, "%d\n", mddev->parallel_resync);
4833}
4834
4835static ssize_t
fd01b88c 4836sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4837{
4838 long n;
4839
b29bebd6 4840 if (kstrtol(buf, 10, &n))
90b08710
BS		 4841 return -EINVAL;
4842
4843 if (n != 0 && n != 1)
4844 return -EINVAL;
4845
4846 mddev->parallel_resync = n;
4847
4848 if (mddev->sync_thread)
4849 wake_up(&resync_wait);
4850
4851 return len;
4852}
4853
4854/* force parallel resync, even with shared block devices */
4855static struct md_sysfs_entry md_sync_force_parallel =
4856__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4857 sync_force_parallel_show, sync_force_parallel_store);
4858
88202a0c 4859static ssize_t
fd01b88c 4860sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4861{
4862 unsigned long resync, dt, db;
d1a7c503
N		 4863 if (mddev->curr_resync == 0)
4864 return sprintf(page, "none\n");
9687a60c
AN		 4865 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4866 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4867 if (!dt) dt++;
9687a60c
AN		 4868 db = resync - mddev->resync_mark_cnt;
4869 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4870}
4871
80ca3a44 4872static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4873
4874static ssize_t
fd01b88c 4875sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4876{
13ae864b 4877 unsigned long long max_sectors, resync;
88202a0c 4878
acb180b0
N		 4879 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4880 return sprintf(page, "none\n");
4881
72f36d59
N		 4882 if (mddev->curr_resync == 1 ||
4883 mddev->curr_resync == 2)
4884 return sprintf(page, "delayed\n");
4885
c804cdec
N		 4886 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4887 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4888 max_sectors = mddev->resync_max_sectors;
88202a0c 4889 else
58c0fed4 4890 max_sectors = mddev->dev_sectors;
88202a0c 4891
acb180b0 4892 resync = mddev->curr_resync_completed;
13ae864b 4893 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4894}
4895
750f199e
N		 4896static struct md_sysfs_entry md_sync_completed =
4897 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
88202a0c 4898
5e96ee65 4899static ssize_t
fd01b88c 4900min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4901{
4902 return sprintf(page, "%llu\n",
4903 (unsigned long long)mddev->resync_min);
4904}
4905static ssize_t
fd01b88c 4906min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4907{
4908 unsigned long long min;
23da422b 4909 int err;
23da422b 4910
b29bebd6 4911 if (kstrtoull(buf, 10, &min))
5e96ee65 4912 return -EINVAL;
23da422b
N		 4913
4914 spin_lock(&mddev->lock);
4915 err = -EINVAL;
5e96ee65 4916 if (min > mddev->resync_max)
23da422b
N		 4917 goto out_unlock;
4918
4919 err = -EBUSY;
5e96ee65 4920 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4921 goto out_unlock;
5e96ee65 4922
50c37b13
N		 4923 /* Round down to multiple of 4K for safety */
4924 mddev->resync_min = round_down(min, 8);
23da422b 4925 err = 0;
5e96ee65 4926
23da422b
N		 4927out_unlock:
4928 spin_unlock(&mddev->lock);
4929 return err ?: len;
5e96ee65
NB		 4930}
4931
4932static struct md_sysfs_entry md_min_sync =
4933__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4934
c6207277 4935static ssize_t
fd01b88c 4936max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4937{
4938 if (mddev->resync_max == MaxSector)
4939 return sprintf(page, "max\n");
4940 else
4941 return sprintf(page, "%llu\n",
4942 (unsigned long long)mddev->resync_max);
4943}
4944static ssize_t
fd01b88c 4945max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277 4946{
23da422b
N		 4947 int err;
4948 spin_lock(&mddev->lock);
c6207277
N		 4949 if (strncmp(buf, "max", 3) == 0)
4950 mddev->resync_max = MaxSector;
4951 else {
5e96ee65 4952 unsigned long long max;
23da422b
N		 4953 int chunk;
4954
4955 err = -EINVAL;
b29bebd6 4956 if (kstrtoull(buf, 10, &max))
23da422b 4957 goto out_unlock;
5e96ee65 4958 if (max < mddev->resync_min)
23da422b
N		 4959 goto out_unlock;
4960
4961 err = -EBUSY;
c6207277 4962 if (max < mddev->resync_max &&
4d484a4a 4963 mddev->ro == 0 &&
c6207277 4964 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
23da422b 4965 goto out_unlock;
c6207277
N		 4966
4967 /* Must be a multiple of chunk_size */
23da422b
N		 4968 chunk = mddev->chunk_sectors;
4969 if (chunk) {
2ac06c33 4970 sector_t temp = max;
23da422b
N		 4971
4972 err = -EINVAL;
4973 if (sector_div(temp, chunk))
4974 goto out_unlock;
c6207277
N		 4975 }
4976 mddev->resync_max = max;
4977 }
4978 wake_up(&mddev->recovery_wait);
23da422b
N		 4979 err = 0;
4980out_unlock:
4981 spin_unlock(&mddev->lock);
4982 return err ?: len;
c6207277
N		 4983}
4984
4985static struct md_sysfs_entry md_max_sync =
4986__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4987
e464eafd 4988static ssize_t
fd01b88c 4989suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4990{
4991 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4992}
4993
4994static ssize_t
fd01b88c 4995suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 4996{
b03e0ccb 4997 unsigned long long new;
6791875e 4998 int err;
e464eafd 4999
4c9309c0
AD		 5000 err = kstrtoull(buf, 10, &new);
5001 if (err < 0)
5002 return err;
5003 if (new != (sector_t)new)
e464eafd 5004 return -EINVAL;
23ddff37 5005
6791875e
N		 5006 err = mddev_lock(mddev);
5007 if (err)
5008 return err;
5009 err = -EINVAL;
5010 if (mddev->pers == NULL ||
5011 mddev->pers->quiesce == NULL)
5012 goto unlock;
b03e0ccb 5013 mddev_suspend(mddev);
23ddff37 5014 mddev->suspend_lo = new;
b03e0ccb
N		 5015 mddev_resume(mddev);
5016
6791875e
N		 5017 err = 0;
5018unlock:
5019 mddev_unlock(mddev);
5020 return err ?: len;
e464eafd
N		 5021}
5022static struct md_sysfs_entry md_suspend_lo =
5023__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5024
e464eafd 5025static ssize_t
fd01b88c 5026suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 5027{
5028 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5029}
5030
5031static ssize_t
fd01b88c 5032suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd 5033{
b03e0ccb 5034 unsigned long long new;
6791875e 5035 int err;
e464eafd 5036
4c9309c0
AD		 5037 err = kstrtoull(buf, 10, &new);
5038 if (err < 0)
5039 return err;
5040 if (new != (sector_t)new)
e464eafd 5041 return -EINVAL;
23ddff37 5042
6791875e
N		 5043 err = mddev_lock(mddev);
5044 if (err)
5045 return err;
5046 err = -EINVAL;
b03e0ccb 5047 if (mddev->pers == NULL)
6791875e 5048 goto unlock;
b03e0ccb
N		 5049
5050 mddev_suspend(mddev);
23ddff37 5051 mddev->suspend_hi = new;
b03e0ccb
N		 5052 mddev_resume(mddev);
5053
6791875e
N		 5054 err = 0;
5055unlock:
5056 mddev_unlock(mddev);
5057 return err ?: len;
e464eafd
N		 5058}
5059static struct md_sysfs_entry md_suspend_hi =
5060__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5061
08a02ecd 5062static ssize_t
fd01b88c 5063reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 5064{
5065 if (mddev->reshape_position != MaxSector)
5066 return sprintf(page, "%llu\n",
5067 (unsigned long long)mddev->reshape_position);
5068 strcpy(page, "none\n");
5069 return 5;
5070}
5071
5072static ssize_t
fd01b88c 5073reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 5074{
c6563a8c 5075 struct md_rdev *rdev;
4c9309c0 5076 unsigned long long new;
6791875e 5077 int err;
6791875e 5078
4c9309c0
AD		 5079 err = kstrtoull(buf, 10, &new);
5080 if (err < 0)
5081 return err;
5082 if (new != (sector_t)new)
08a02ecd 5083 return -EINVAL;
6791875e
N		 5084 err = mddev_lock(mddev);
5085 if (err)
5086 return err;
5087 err = -EBUSY;
5088 if (mddev->pers)
5089 goto unlock;
08a02ecd
N		 5090 mddev->reshape_position = new;
5091 mddev->delta_disks = 0;
2c810cdd 5092 mddev->reshape_backwards = 0;
08a02ecd
N		 5093 mddev->new_level = mddev->level;
5094 mddev->new_layout = mddev->layout;
664e7c41 5095 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N		 5096 rdev_for_each(rdev, mddev)
5097 rdev->new_data_offset = rdev->data_offset;
6791875e
N		 5098 err = 0;
5099unlock:
5100 mddev_unlock(mddev);
5101 return err ?: len;
08a02ecd
N		 5102}
5103
5104static struct md_sysfs_entry md_reshape_position =
5105__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5106 reshape_position_store);
5107
2c810cdd
N		 5108static ssize_t
5109reshape_direction_show(struct mddev *mddev, char *page)
5110{
5111 return sprintf(page, "%s\n",
5112 mddev->reshape_backwards ? "backwards" : "forwards");
5113}
5114
5115static ssize_t
5116reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5117{
5118 int backwards = 0;
6791875e
N		 5119 int err;
5120
2c810cdd
N		 5121 if (cmd_match(buf, "forwards"))
5122 backwards = 0;
5123 else if (cmd_match(buf, "backwards"))
5124 backwards = 1;
5125 else
5126 return -EINVAL;
5127 if (mddev->reshape_backwards == backwards)
5128 return len;
5129
6791875e
N		 5130 err = mddev_lock(mddev);
5131 if (err)
5132 return err;
2c810cdd
N		 5133 /* check if we are allowed to change */
5134 if (mddev->delta_disks)
6791875e
N		 5135 err = -EBUSY;
5136 else if (mddev->persistent &&
2c810cdd 5137 mddev->major_version == 0)
6791875e
N		 5138 err = -EINVAL;
5139 else
5140 mddev->reshape_backwards = backwards;
5141 mddev_unlock(mddev);
5142 return err ?: len;
2c810cdd
N		 5143}
5144
5145static struct md_sysfs_entry md_reshape_direction =
5146__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5147 reshape_direction_store);
5148
b522adcd 5149static ssize_t
fd01b88c 5150array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 5151{
5152 if (mddev->external_size)
5153 return sprintf(page, "%llu\n",
5154 (unsigned long long)mddev->array_sectors/2);
5155 else
5156 return sprintf(page, "default\n");
5157}
5158
5159static ssize_t
fd01b88c 5160array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 5161{
5162 sector_t sectors;
6791875e
N		 5163 int err;
5164
5165 err = mddev_lock(mddev);
5166 if (err)
5167 return err;
b522adcd 5168
ab5a98b1 5169 /* cluster raid doesn't support change array_sectors */
b670883b
ZL		 5170 if (mddev_is_clustered(mddev)) {
5171 mddev_unlock(mddev);
ab5a98b1 5172 return -EINVAL;
b670883b 5173 }
ab5a98b1 5174
b522adcd
DW		 5175 if (strncmp(buf, "default", 7) == 0) {
5176 if (mddev->pers)
5177 sectors = mddev->pers->size(mddev, 0, 0);
5178 else
5179 sectors = mddev->array_sectors;
5180
5181 mddev->external_size = 0;
5182 } else {
5183 if (strict_blocks_to_sectors(buf, &sectors) < 0)
6791875e
N		 5184 err = -EINVAL;
5185 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5186 err = -E2BIG;
5187 else
5188 mddev->external_size = 1;
b522adcd
DW		 5189 }
5190
6791875e
N		 5191 if (!err) {
5192 mddev->array_sectors = sectors;
5193 if (mddev->pers) {
5194 set_capacity(mddev->gendisk, mddev->array_sectors);
5195 revalidate_disk(mddev->gendisk);
5196 }
cbe6ef1d 5197 }
6791875e
N		 5198 mddev_unlock(mddev);
5199 return err ?: len;
b522adcd
DW		 5200}
5201
5202static struct md_sysfs_entry md_array_size =
5203__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5204 array_size_store);
e464eafd 5205
664aed04
AP		 5206static ssize_t
5207consistency_policy_show(struct mddev *mddev, char *page)
5208{
5209 int ret;
5210
5211 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5212 ret = sprintf(page, "journal\n");
5213 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5214 ret = sprintf(page, "ppl\n");
5215 } else if (mddev->bitmap) {
5216 ret = sprintf(page, "bitmap\n");
5217 } else if (mddev->pers) {
5218 if (mddev->pers->sync_request)
5219 ret = sprintf(page, "resync\n");
5220 else
5221 ret = sprintf(page, "none\n");
5222 } else {
5223 ret = sprintf(page, "unknown\n");
5224 }
5225
5226 return ret;
5227}
5228
5229static ssize_t
5230consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5231{
ba903a3e
AP		 5232 int err = 0;
5233
664aed04 5234 if (mddev->pers) {
ba903a3e
AP		 5235 if (mddev->pers->change_consistency_policy)
5236 err = mddev->pers->change_consistency_policy(mddev, buf);
5237 else
5238 err = -EBUSY;
664aed04
AP		 5239 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5240 set_bit(MD_HAS_PPL, &mddev->flags);
664aed04 5241 } else {
ba903a3e 5242 err = -EINVAL;
664aed04 5243 }
ba903a3e
AP		 5244
5245 return err ? err : len;
664aed04
AP		 5246}
5247
5248static struct md_sysfs_entry md_consistency_policy =
5249__ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5250 consistency_policy_store);
5251
9a567843
GJ		 5252static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5253{
5254 return sprintf(page, "%d\n", mddev->fail_last_dev);
5255}
5256
5257/*
5258 * Setting fail_last_dev to true to allow last device to be forcibly removed
5259 * from RAID1/RAID10.
5260 */
5261static ssize_t
5262fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5263{
5264 int ret;
5265 bool value;
5266
5267 ret = kstrtobool(buf, &value);
5268 if (ret)
5269 return ret;
5270
5271 if (value != mddev->fail_last_dev)
5272 mddev->fail_last_dev = value;
5273
5274 return len;
5275}
5276static struct md_sysfs_entry md_fail_last_dev =
5277__ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5278 fail_last_dev_store);
5279
eae1701f
N		 5280static struct attribute *md_default_attrs[] = {
5281 &md_level.attr,
d4dbd025 5282 &md_layout.attr,
eae1701f 5283 &md_raid_disks.attr,
3b34380a 5284 &md_chunk_size.attr,
a35b0d69 5285 &md_size.attr,
a94213b1 5286 &md_resync_start.attr,
8bb93aac 5287 &md_metadata.attr,
6d7ff738 5288 &md_new_device.attr,
16f17b39 5289 &md_safe_delay.attr,
9e653b63 5290 &md_array_state.attr,
08a02ecd 5291 &md_reshape_position.attr,
2c810cdd 5292 &md_reshape_direction.attr,
b522adcd 5293 &md_array_size.attr,
1e50915f 5294 &max_corr_read_errors.attr,
664aed04 5295 &md_consistency_policy.attr,
9a567843 5296 &md_fail_last_dev.attr,
411036fa
N		 5297 NULL,
5298};
5299
5300static struct attribute *md_redundancy_attrs[] = {
24dd469d 5301 &md_scan_mode.attr,
c4a39551 5302 &md_last_scan_mode.attr,
9d88883e 5303 &md_mismatches.attr,
88202a0c
N		 5304 &md_sync_min.attr,
5305 &md_sync_max.attr,
5306 &md_sync_speed.attr,
90b08710 5307 &md_sync_force_parallel.attr,
88202a0c 5308 &md_sync_completed.attr,
5e96ee65 5309 &md_min_sync.attr,
c6207277 5310 &md_max_sync.attr,
e464eafd
N		 5311 &md_suspend_lo.attr,
5312 &md_suspend_hi.attr,
9b1d1dac 5313 &md_bitmap.attr,
d7f3d291 5314 &md_degraded.attr,
eae1701f
N		 5315 NULL,
5316};
411036fa
N		 5317static struct attribute_group md_redundancy_group = {
5318 .name = NULL,
5319 .attrs = md_redundancy_attrs,
5320};
5321
eae1701f
N		 5322static ssize_t
5323md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5324{
5325 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 5326 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 5327 ssize_t rv;
eae1701f
N		 5328
5329 if (!entry->show)
5330 return -EIO;
af8a2434
N		 5331 spin_lock(&all_mddevs_lock);
5332 if (list_empty(&mddev->all_mddevs)) {
5333 spin_unlock(&all_mddevs_lock);
5334 return -EBUSY;
5335 }
5336 mddev_get(mddev);
5337 spin_unlock(&all_mddevs_lock);
5338
b7b17c9b 5339 rv = entry->show(mddev, page);
af8a2434 5340 mddev_put(mddev);
96de1e66 5341 return rv;
eae1701f
N		 5342}
5343
5344static ssize_t
5345md_attr_store(struct kobject *kobj, struct attribute *attr,
5346 const char *page, size_t length)
5347{
5348 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 5349 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 5350 ssize_t rv;
eae1701f
N		 5351
5352 if (!entry->store)
5353 return -EIO;
67463acb
N		 5354 if (!capable(CAP_SYS_ADMIN))
5355 return -EACCES;
af8a2434
N		 5356 spin_lock(&all_mddevs_lock);
5357 if (list_empty(&mddev->all_mddevs)) {
5358 spin_unlock(&all_mddevs_lock);
5359 return -EBUSY;
5360 }
5361 mddev_get(mddev);
5362 spin_unlock(&all_mddevs_lock);
6791875e 5363 rv = entry->store(mddev, page, length);
af8a2434 5364 mddev_put(mddev);
96de1e66 5365 return rv;
eae1701f
N		 5366}
5367
5368static void md_free(struct kobject *ko)
5369{
fd01b88c 5370 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 5371
5372 if (mddev->sysfs_state)
5373 sysfs_put(mddev->sysfs_state);
5374
d8115c35
BVA		 5375 if (mddev->gendisk)
5376 del_gendisk(mddev->gendisk);
6cd18e71
N		 5377 if (mddev->queue)
5378 blk_cleanup_queue(mddev->queue);
d8115c35 5379 if (mddev->gendisk)
a21d1504 5380 put_disk(mddev->gendisk);
4ad23a97 5381 percpu_ref_exit(&mddev->writes_pending);
a21d1504 5382
28dec870
KO		 5383 bioset_exit(&mddev->bio_set);
5384 bioset_exit(&mddev->sync_set);
eae1701f
N		 5385 kfree(mddev);
5386}
5387
52cf25d0 5388static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 5389 .show = md_attr_show,
5390 .store = md_attr_store,
5391};
5392static struct kobj_type md_ktype = {
5393 .release = md_free,
5394 .sysfs_ops = &md_sysfs_ops,
5395 .default_attrs = md_default_attrs,
5396};
5397
1da177e4
LT		 5398int mdp_major = 0;
5399
5fd3a17e
DW		 5400static void mddev_delayed_delete(struct work_struct *ws)
5401{
fd01b88c 5402 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 5403
43a70507 5404 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 5405 kobject_del(&mddev->kobj);
5406 kobject_put(&mddev->kobj);
5407}
5408
4ad23a97
N		 5409static void no_op(struct percpu_ref *r) {}
5410
a415c0f1
N		 5411int mddev_init_writes_pending(struct mddev *mddev)
5412{
5413 if (mddev->writes_pending.percpu_count_ptr)
5414 return 0;
ddde2af7
RG		 5415 if (percpu_ref_init(&mddev->writes_pending, no_op,
5416 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
a415c0f1
N		 5417 return -ENOMEM;
5418 /* We want to start with the refcount at zero */
5419 percpu_ref_put(&mddev->writes_pending);
5420 return 0;
5421}
5422EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5423
efeb53c0 5424static int md_alloc(dev_t dev, char *name)
1da177e4 5425{
039b7225
N		 5426 /*
5427 * If dev is zero, name is the name of a device to allocate with
5428 * an arbitrary minor number. It will be "md_???"
5429 * If dev is non-zero it must be a device number with a MAJOR of
5430 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5431 * the device is being created by opening a node in /dev.
5432 * If "name" is not NULL, the device is being created by
5433 * writing to /sys/module/md_mod/parameters/new_array.
5434 */
48c9c27b 5435 static DEFINE_MUTEX(disks_mutex);
fd01b88c 5436 struct mddev *mddev = mddev_find(dev);
1da177e4 5437 struct gendisk *disk;
efeb53c0
N		 5438 int partitioned;
5439 int shift;
5440 int unit;
3830c62f 5441 int error;
1da177e4
LT		 5442
5443 if (!mddev)
efeb53c0
N		 5444 return -ENODEV;
5445
5446 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5447 shift = partitioned ? MdpMinorShift : 0;
5448 unit = MINOR(mddev->unit) >> shift;
1da177e4 5449
e804ac78
TH		 5450 /* wait for any previous instance of this device to be
5451 * completely removed (mddev_delayed_delete).
d3374825 5452 */
e804ac78 5453 flush_workqueue(md_misc_wq);
d3374825 5454
48c9c27b 5455 mutex_lock(&disks_mutex);
0909dc44
N		 5456 error = -EEXIST;
5457 if (mddev->gendisk)
5458 goto abort;
efeb53c0 5459
039b7225 5460 if (name && !dev) {
efeb53c0
N		 5461 /* Need to ensure that 'name' is not a duplicate.
5462 */
fd01b88c 5463 struct mddev *mddev2;
efeb53c0
N		 5464 spin_lock(&all_mddevs_lock);
5465
5466 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5467 if (mddev2->gendisk &&
5468 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5469 spin_unlock(&all_mddevs_lock);
0909dc44 5470 goto abort;
efeb53c0
N		 5471 }
5472 spin_unlock(&all_mddevs_lock);
1da177e4 5473 }
039b7225
N		 5474 if (name && dev)
5475 /*
5476 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5477 */
5478 mddev->hold_active = UNTIL_STOP;
8b765398 5479
0909dc44 5480 error = -ENOMEM;
8b765398 5481 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 5482 if (!mddev->queue)
5483 goto abort;
409c57f3
N		 5484 mddev->queue->queuedata = mddev;
5485
409c57f3 5486 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 5487 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 5488
1da177e4
LT		 5489 disk = alloc_disk(1 << shift);
5490 if (!disk) {
8b765398
N		 5491 blk_cleanup_queue(mddev->queue);
5492 mddev->queue = NULL;
0909dc44 5493 goto abort;
1da177e4 5494 }
efeb53c0 5495 disk->major = MAJOR(mddev->unit);
1da177e4 5496 disk->first_minor = unit << shift;
efeb53c0
N		 5497 if (name)
5498 strcpy(disk->disk_name, name);
5499 else if (partitioned)
1da177e4 5500 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 5501 else
1da177e4 5502 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 5503 disk->fops = &md_fops;
5504 disk->private_data = mddev;
5505 disk->queue = mddev->queue;
56883a7e 5506 blk_queue_write_cache(mddev->queue, true, true);
92850bbd 5507 /* Allow extended partitions. This makes the
d3374825 5508 * 'mdp' device redundant, but we can't really
92850bbd
N		 5509 * remove it now.
5510 */
5511 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 5512 mddev->gendisk = disk;
b0140891
N		 5513 /* As soon as we call add_disk(), another thread could get
5514 * through to md_open, so make sure it doesn't get too far
5515 */
5516 mutex_lock(&mddev->open_mutex);
5517 add_disk(disk);
5518
28dec870 5519 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 5520 if (error) {
5521 /* This isn't possible, but as kobject_init_and_add is marked
5522 * __must_check, we must do something with the result
5523 */
9d48739e
N		 5524 pr_debug("md: cannot register %s/md - name in use\n",
5525 disk->disk_name);
0909dc44
N		 5526 error = 0;
5527 }
00bcb4ac
N		 5528 if (mddev->kobj.sd &&
5529 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
9d48739e 5530 pr_debug("pointless warning\n");
b0140891 5531 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 5532 abort:
5533 mutex_unlock(&disks_mutex);
00bcb4ac 5534 if (!error && mddev->kobj.sd) {
3830c62f 5535 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 5536 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 5537 }
d3374825 5538 mddev_put(mddev);
0909dc44 5539 return error;
efeb53c0
N		 5540}
5541
5542static struct kobject *md_probe(dev_t dev, int *part, void *data)
5543{
78b6350d
N		 5544 if (create_on_open)
5545 md_alloc(dev, NULL);
1da177e4
LT		 5546 return NULL;
5547}
5548
e4dca7b7 5549static int add_named_array(const char *val, const struct kernel_param *kp)
efeb53c0 5550{
039b7225
N		 5551 /*
5552 * val must be "md_*" or "mdNNN".
5553 * For "md_*" we allocate an array with a large free minor number, and
efeb53c0 5554 * set the name to val. val must not already be an active name.
039b7225
N		 5555 * For "mdNNN" we allocate an array with the minor number NNN
5556 * which must not already be in use.
efeb53c0
N		 5557 */
5558 int len = strlen(val);
5559 char buf[DISK_NAME_LEN];
039b7225 5560 unsigned long devnum;
efeb53c0
N		 5561
5562 while (len && val[len-1] == '\n')
5563 len--;
5564 if (len >= DISK_NAME_LEN)
5565 return -E2BIG;
5566 strlcpy(buf, val, len+1);
039b7225
N		 5567 if (strncmp(buf, "md_", 3) == 0)
5568 return md_alloc(0, buf);
5569 if (strncmp(buf, "md", 2) == 0 &&
5570 isdigit(buf[2]) &&
5571 kstrtoul(buf+2, 10, &devnum) == 0 &&
5572 devnum <= MINORMASK)
5573 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5574
5575 return -EINVAL;
efeb53c0
N		 5576}
5577
8376d3c1 5578static void md_safemode_timeout(struct timer_list *t)
1da177e4 5579{
8376d3c1 5580 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
1da177e4 5581
4ad23a97
N		 5582 mddev->safemode = 1;
5583 if (mddev->external)
5584 sysfs_notify_dirent_safe(mddev->sysfs_state);
5585
1da177e4
LT		 5586 md_wakeup_thread(mddev->thread);
5587}
5588
6ff8d8ec 5589static int start_dirty_degraded;
1da177e4 5590
fd01b88c 5591int md_run(struct mddev *mddev)
1da177e4 5592{
2604b703 5593 int err;
3cb03002 5594 struct md_rdev *rdev;
84fc4b56 5595 struct md_personality *pers;
1da177e4 5596
a757e64c
N		 5597 if (list_empty(&mddev->disks))
5598 /* cannot run an array with no devices.. */
1da177e4 5599 return -EINVAL;
1da177e4
LT		 5600
5601 if (mddev->pers)
5602 return -EBUSY;
bb4f1e9d
N		 5603 /* Cannot run until previous stop completes properly */
5604 if (mddev->sysfs_active)
5605 return -EBUSY;
b6eb127d 5606
1da177e4
LT		 5607 /*
5608 * Analyze all RAID superblock(s)
5609 */
1ec4a939
N		 5610 if (!mddev->raid_disks) {
5611 if (!mddev->persistent)
5612 return -EINVAL;
6a5cb53a
YY		 5613 err = analyze_sbs(mddev);
5614 if (err)
5615 return -EINVAL;
1ec4a939 5616 }
1da177e4 5617
d9d166c2
N		 5618 if (mddev->level != LEVEL_NONE)
5619 request_module("md-level-%d", mddev->level);
5620 else if (mddev->clevel[0])
5621 request_module("md-%s", mddev->clevel);
1da177e4
LT		 5622
5623 /*
5624 * Drop all container device buffers, from now on
5625 * the only valid external interface is through the md
5626 * device.
1da177e4 5627 */
4b6c1060 5628 mddev->has_superblocks = false;
dafb20fa 5629 rdev_for_each(rdev, mddev) {
b2d444d7 5630 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5631 continue;
5632 sync_blockdev(rdev->bdev);
f98393a6 5633 invalidate_bdev(rdev->bdev);
97b20ef7
N		 5634 if (mddev->ro != 1 &&
5635 (bdev_read_only(rdev->bdev) ||
5636 bdev_read_only(rdev->meta_bdev))) {
5637 mddev->ro = 1;
5638 if (mddev->gendisk)
5639 set_disk_ro(mddev->gendisk, 1);
5640 }
f0d76d70 5641
4b6c1060
HM		 5642 if (rdev->sb_page)
5643 mddev->has_superblocks = true;
5644
f0d76d70
N		 5645 /* perform some consistency tests on the device.
5646 * We don't want the data to overlap the metadata,
58c0fed4 5647 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 5648 */
a6ff7e08
JB		 5649 if (rdev->meta_bdev) {
5650 /* Nothing to check */;
5651 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 5652 if (mddev->dev_sectors &&
5653 rdev->data_offset + mddev->dev_sectors
0f420358 5654 > rdev->sb_start) {
9d48739e
N		 5655 pr_warn("md: %s: data overlaps metadata\n",
5656 mdname(mddev));
f0d76d70
N		 5657 return -EINVAL;
5658 }
5659 } else {
0f420358 5660 if (rdev->sb_start + rdev->sb_size/512
f0d76d70 5661 > rdev->data_offset) {
9d48739e
N		 5662 pr_warn("md: %s: metadata overlaps data\n",
5663 mdname(mddev));
f0d76d70
N		 5664 return -EINVAL;
5665 }
5666 }
00bcb4ac 5667 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 5668 }
5669
afeee514
KO		 5670 if (!bioset_initialized(&mddev->bio_set)) {
5671 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5672 if (err)
5673 return err;
10273170 5674 }
afeee514
KO		 5675 if (!bioset_initialized(&mddev->sync_set)) {
5676 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5677 if (err)
28dec870 5678 return err;
5a85071c 5679 }
a167f663 5680
1da177e4 5681 spin_lock(&pers_lock);
d9d166c2 5682 pers = find_pers(mddev->level, mddev->clevel);
2604b703 5683 if (!pers || !try_module_get(pers->owner)) {
1da177e4 5684 spin_unlock(&pers_lock);
d9d166c2 5685 if (mddev->level != LEVEL_NONE)
9d48739e
N		 5686 pr_warn("md: personality for level %d is not loaded!\n",
5687 mddev->level);
d9d166c2 5688 else
9d48739e
N		 5689 pr_warn("md: personality for level %s is not loaded!\n",
5690 mddev->clevel);
bfc9dfdc
SL		 5691 err = -EINVAL;
5692 goto abort;
1da177e4 5693 }
1da177e4 5694 spin_unlock(&pers_lock);
34817e8c
N		 5695 if (mddev->level != pers->level) {
5696 mddev->level = pers->level;
5697 mddev->new_level = pers->level;
5698 }
d9d166c2 5699 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 5700
f6705578 5701 if (mddev->reshape_position != MaxSector &&
63c70c4f 5702 pers->start_reshape == NULL) {
f6705578 5703 /* This personality cannot handle reshaping... */
f6705578 5704 module_put(pers->owner);
bfc9dfdc
SL		 5705 err = -EINVAL;
5706 goto abort;
f6705578
N		 5707 }
5708
7dd5e7c3
N		 5709 if (pers->sync_request) {
5710 /* Warn if this is a potentially silly
5711 * configuration.
5712 */
5713 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5714 struct md_rdev *rdev2;
7dd5e7c3 5715 int warned = 0;
159ec1fc 5716
dafb20fa
N		 5717 rdev_for_each(rdev, mddev)
5718 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 5719 if (rdev < rdev2 &&
5720 rdev->bdev->bd_contains ==
5721 rdev2->bdev->bd_contains) {
9d48739e
N		 5722 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5723 mdname(mddev),
5724 bdevname(rdev->bdev,b),
5725 bdevname(rdev2->bdev,b2));
7dd5e7c3
N		 5726 warned = 1;
5727 }
5728 }
159ec1fc 5729
7dd5e7c3 5730 if (warned)
9d48739e 5731 pr_warn("True protection against single-disk failure might be compromised.\n");
7dd5e7c3
N		 5732 }
5733
657390d2 5734 mddev->recovery = 0;
58c0fed4
AN		 5735 /* may be over-ridden by personality */
5736 mddev->resync_max_sectors = mddev->dev_sectors;
5737
6ff8d8ec 5738 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5739
0f9552b5 5740 if (start_readonly && mddev->ro == 0)
f91de92e
N		 5741 mddev->ro = 2; /* read-only, but switch on first write */
5742
36d091f4 5743 err = pers->run(mddev);
13e53df3 5744 if (err)
9d48739e 5745 pr_warn("md: pers->run() failed ...\n");
36d091f4 5746 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
9d48739e
N		 5747 WARN_ONCE(!mddev->external_size,
5748 "%s: default size too small, but 'external_size' not in effect?\n",
5749 __func__);
5750 pr_warn("md: invalid array_size %llu > default size %llu\n",
5751 (unsigned long long)mddev->array_sectors / 2,
5752 (unsigned long long)pers->size(mddev, 0, 0) / 2);
b522adcd 5753 err = -EINVAL;
b522adcd 5754 }
36d091f4 5755 if (err == 0 && pers->sync_request &&
ef99bf48 5756 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
f9209a32
GR		 5757 struct bitmap *bitmap;
5758
e64e4018 5759 bitmap = md_bitmap_create(mddev, -1);
f9209a32
GR		 5760 if (IS_ERR(bitmap)) {
5761 err = PTR_ERR(bitmap);
9d48739e
N		 5762 pr_warn("%s: failed to create bitmap (%d)\n",
5763 mdname(mddev), err);
f9209a32
GR		 5764 } else
5765 mddev->bitmap = bitmap;
5766
b15c2e57 5767 }
d494549a
GJ		 5768 if (err)
5769 goto bitmap_abort;
3e148a32
GJ		 5770
5771 if (mddev->bitmap_info.max_write_behind > 0) {
5772 bool creat_pool = false;
5773
5774 rdev_for_each(rdev, mddev) {
5775 if (test_bit(WriteMostly, &rdev->flags) &&
5776 rdev_init_wb(rdev))
5777 creat_pool = true;
5778 }
5779 if (creat_pool && mddev->wb_info_pool == NULL) {
5780 mddev->wb_info_pool =
5781 mempool_create_kmalloc_pool(NR_WB_INFOS,
5782 sizeof(struct wb_info));
5783 if (!mddev->wb_info_pool) {
5784 err = -ENOMEM;
d494549a 5785 goto bitmap_abort;
3e148a32
GJ		 5786 }
5787 }
5788 }
5789
5c675f83 5790 if (mddev->queue) {
bb086a89
SL		 5791 bool nonrot = true;
5792
5793 rdev_for_each(rdev, mddev) {
5794 if (rdev->raid_disk >= 0 &&
5795 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5796 nonrot = false;
5797 break;
5798 }
5799 }
5800 if (mddev->degraded)
5801 nonrot = false;
5802 if (nonrot)
8b904b5b 5803 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
bb086a89 5804 else
8b904b5b 5805 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
dc3b17cc
JK		 5806 mddev->queue->backing_dev_info->congested_data = mddev;
5807 mddev->queue->backing_dev_info->congested_fn = md_congested;
5c675f83 5808 }
36d091f4 5809 if (pers->sync_request) {
00bcb4ac
N		 5810 if (mddev->kobj.sd &&
5811 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
9d48739e
N		 5812 pr_warn("md: cannot register extra attributes for %s\n",
5813 mdname(mddev));
00bcb4ac 5814 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5815 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 5816 mddev->ro = 0;
5817
1e50915f
RB		 5818 atomic_set(&mddev->max_corr_read_errors,
5819 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4 5820 mddev->safemode = 0;
28c1b9fd
GR		 5821 if (mddev_is_clustered(mddev))
5822 mddev->safemode_delay = 0;
5823 else
5824 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5825 mddev->in_sync = 1;
0ca69886 5826 smp_wmb();
36d091f4
N		 5827 spin_lock(&mddev->lock);
5828 mddev->pers = pers;
36d091f4 5829 spin_unlock(&mddev->lock);
dafb20fa 5830 rdev_for_each(rdev, mddev)
36fad858 5831 if (rdev->raid_disk >= 0)
e5b521ee 5832 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
f72ffdd6 5833
a4a3d26d
N		 5834 if (mddev->degraded && !mddev->ro)
5835 /* This ensures that recovering status is reported immediately
5836 * via sysfs - until a lack of spares is confirmed.
5837 */
5838 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
1da177e4 5839 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5840
2953079c 5841 if (mddev->sb_flags)
850b2b42 5842 md_update_sb(mddev, 0);
1da177e4 5843
d7603b7e 5844 md_new_event(mddev);
1da177e4 5845 return 0;
b126194c 5846
d494549a
GJ		 5847bitmap_abort:
5848 mddev_detach(mddev);
5849 if (mddev->private)
5850 pers->free(mddev, mddev->private);
5851 mddev->private = NULL;
5852 module_put(pers->owner);
5853 md_bitmap_destroy(mddev);
b126194c 5854abort:
4bc034d3
N		 5855 bioset_exit(&mddev->bio_set);
5856 bioset_exit(&mddev->sync_set);
b126194c 5857 return err;
1da177e4 5858}
390ee602 5859EXPORT_SYMBOL_GPL(md_run);
1da177e4 5860
fd01b88c 5861static int do_md_run(struct mddev *mddev)
fe60b014
N		 5862{
5863 int err;
5864
9d4b45d6 5865 set_bit(MD_NOT_READY, &mddev->flags);
fe60b014
N		 5866 err = md_run(mddev);
5867 if (err)
5868 goto out;
e64e4018 5869 err = md_bitmap_load(mddev);
69e51b44 5870 if (err) {
e64e4018 5871 md_bitmap_destroy(mddev);
69e51b44
N		 5872 goto out;
5873 }
0fd018af 5874
28c1b9fd
GR		 5875 if (mddev_is_clustered(mddev))
5876 md_allow_write(mddev);
5877
d5d885fd
SL		 5878 /* run start up tasks that require md_thread */
5879 md_start(mddev);
5880
0fd018af
JB		 5881 md_wakeup_thread(mddev->thread);
5882 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5883
fe60b014
N		 5884 set_capacity(mddev->gendisk, mddev->array_sectors);
5885 revalidate_disk(mddev->gendisk);
9d4b45d6 5886 clear_bit(MD_NOT_READY, &mddev->flags);
f0b4f7e2 5887 mddev->changed = 1;
fe60b014 5888 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
9d4b45d6
N		 5889 sysfs_notify_dirent_safe(mddev->sysfs_state);
5890 sysfs_notify_dirent_safe(mddev->sysfs_action);
5891 sysfs_notify(&mddev->kobj, NULL, "degraded");
fe60b014 5892out:
9d4b45d6 5893 clear_bit(MD_NOT_READY, &mddev->flags);
fe60b014
N		 5894 return err;
5895}
5896
d5d885fd
SL		 5897int md_start(struct mddev *mddev)
5898{
5899 int ret = 0;
5900
5901 if (mddev->pers->start) {
5902 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5903 md_wakeup_thread(mddev->thread);
5904 ret = mddev->pers->start(mddev);
5905 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5906 md_wakeup_thread(mddev->sync_thread);
5907 }
5908 return ret;
5909}
5910EXPORT_SYMBOL_GPL(md_start);
5911
fd01b88c 5912static int restart_array(struct mddev *mddev)
1da177e4
LT		 5913{
5914 struct gendisk *disk = mddev->gendisk;
97b20ef7
N		 5915 struct md_rdev *rdev;
5916 bool has_journal = false;
5917 bool has_readonly = false;
1da177e4 5918
80fab1d7 5919 /* Complain if it has no devices */
1da177e4 5920 if (list_empty(&mddev->disks))
80fab1d7
AN		 5921 return -ENXIO;
5922 if (!mddev->pers)
5923 return -EINVAL;
5924 if (!mddev->ro)
5925 return -EBUSY;
339421de 5926
97b20ef7
N		 5927 rcu_read_lock();
5928 rdev_for_each_rcu(rdev, mddev) {
5929 if (test_bit(Journal, &rdev->flags) &&
5930 !test_bit(Faulty, &rdev->flags))
5931 has_journal = true;
5932 if (bdev_read_only(rdev->bdev))
5933 has_readonly = true;
5934 }
5935 rcu_read_unlock();
5936 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
339421de 5937 /* Don't restart rw with journal missing/faulty */
339421de 5938 return -EINVAL;
97b20ef7
N		 5939 if (has_readonly)
5940 return -EROFS;
339421de 5941
80fab1d7
AN		 5942 mddev->safemode = 0;
5943 mddev->ro = 0;
5944 set_disk_ro(disk, 0);
9d48739e 5945 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
80fab1d7
AN		 5946 /* Kick recovery or resync if necessary */
5947 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5948 md_wakeup_thread(mddev->thread);
5949 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5950 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5951 return 0;
1da177e4
LT		 5952}
5953
fd01b88c 5954static void md_clean(struct mddev *mddev)
6177b472
N		 5955{
5956 mddev->array_sectors = 0;
5957 mddev->external_size = 0;
5958 mddev->dev_sectors = 0;
5959 mddev->raid_disks = 0;
5960 mddev->recovery_cp = 0;
5961 mddev->resync_min = 0;
5962 mddev->resync_max = MaxSector;
5963 mddev->reshape_position = MaxSector;
5964 mddev->external = 0;
5965 mddev->persistent = 0;
5966 mddev->level = LEVEL_NONE;
5967 mddev->clevel[0] = 0;
5968 mddev->flags = 0;
2953079c 5969 mddev->sb_flags = 0;
6177b472
N		 5970 mddev->ro = 0;
5971 mddev->metadata_type[0] = 0;
5972 mddev->chunk_sectors = 0;
5973 mddev->ctime = mddev->utime = 0;
5974 mddev->layout = 0;
5975 mddev->max_disks = 0;
5976 mddev->events = 0;
a8707c08 5977 mddev->can_decrease_events = 0;
6177b472 5978 mddev->delta_disks = 0;
2c810cdd 5979 mddev->reshape_backwards = 0;
6177b472
N		 5980 mddev->new_level = LEVEL_NONE;
5981 mddev->new_layout = 0;
5982 mddev->new_chunk_sectors = 0;
5983 mddev->curr_resync = 0;
7f7583d4 5984 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N		 5985 mddev->suspend_lo = mddev->suspend_hi = 0;
5986 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5987 mddev->recovery = 0;
5988 mddev->in_sync = 0;
f0b4f7e2 5989 mddev->changed = 0;
6177b472 5990 mddev->degraded = 0;
6177b472 5991 mddev->safemode = 0;
bd691922 5992 mddev->private = NULL;
c20c33f0 5993 mddev->cluster_info = NULL;
6177b472
N		 5994 mddev->bitmap_info.offset = 0;
5995 mddev->bitmap_info.default_offset = 0;
6409bb05 5996 mddev->bitmap_info.default_space = 0;
6177b472
N		 5997 mddev->bitmap_info.chunksize = 0;
5998 mddev->bitmap_info.daemon_sleep = 0;
5999 mddev->bitmap_info.max_write_behind = 0;
c20c33f0 6000 mddev->bitmap_info.nodes = 0;
6177b472
N		 6001}
6002
fd01b88c 6003static void __md_stop_writes(struct mddev *mddev)
a047e125 6004{
6b6204ee 6005 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
f851b60d 6006 flush_workqueue(md_misc_wq);
a047e125 6007 if (mddev->sync_thread) {
a047e125 6008 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 6009 md_reap_sync_thread(mddev);
a047e125
N		 6010 }
6011
6012 del_timer_sync(&mddev->safemode_timer);
6013
034e33f5
SL		 6014 if (mddev->pers && mddev->pers->quiesce) {
6015 mddev->pers->quiesce(mddev, 1);
6016 mddev->pers->quiesce(mddev, 0);
6017 }
e64e4018 6018 md_bitmap_flush(mddev);
a047e125 6019
b6d428c6 6020 if (mddev->ro == 0 &&
28c1b9fd 6021 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
2953079c 6022 mddev->sb_flags)) {
a047e125 6023 /* mark array as shutdown cleanly */
28c1b9fd
GR		 6024 if (!mddev_is_clustered(mddev))
6025 mddev->in_sync = 1;
a047e125
N		 6026 md_update_sb(mddev, 1);
6027 }
3e148a32
GJ		 6028 mempool_destroy(mddev->wb_info_pool);
6029 mddev->wb_info_pool = NULL;
a047e125 6030}
defad61a 6031
fd01b88c 6032void md_stop_writes(struct mddev *mddev)
defad61a 6033{
29f097c4 6034 mddev_lock_nointr(mddev);
defad61a
N		 6035 __md_stop_writes(mddev);
6036 mddev_unlock(mddev);
6037}
390ee602 6038EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 6039
5aa61f42
N		 6040static void mddev_detach(struct mddev *mddev)
6041{
e64e4018 6042 md_bitmap_wait_behind_writes(mddev);
36d091f4 6043 if (mddev->pers && mddev->pers->quiesce) {
5aa61f42
N		 6044 mddev->pers->quiesce(mddev, 1);
6045 mddev->pers->quiesce(mddev, 0);
6046 }
6047 md_unregister_thread(&mddev->thread);
6048 if (mddev->queue)
6049 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6050}
6051
5eff3c43 6052static void __md_stop(struct mddev *mddev)
6177b472 6053{
36d091f4 6054 struct md_personality *pers = mddev->pers;
e64e4018 6055 md_bitmap_destroy(mddev);
5aa61f42 6056 mddev_detach(mddev);
ee5d004f
N		 6057 /* Ensure ->event_work is done */
6058 flush_workqueue(md_misc_wq);
36d091f4 6059 spin_lock(&mddev->lock);
6177b472 6060 mddev->pers = NULL;
36d091f4
N		 6061 spin_unlock(&mddev->lock);
6062 pers->free(mddev, mddev->private);
bd691922 6063 mddev->private = NULL;
36d091f4
N		 6064 if (pers->sync_request && mddev->to_remove == NULL)
6065 mddev->to_remove = &md_redundancy_group;
6066 module_put(pers->owner);
cca9cf90 6067 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6aaa58c9
JW		 6068}
6069
6070void md_stop(struct mddev *mddev)
6071{
6072 /* stop the array and free an attached data structures.
6073 * This is called from dm-raid
6074 */
6075 __md_stop(mddev);
afeee514
KO		 6076 bioset_exit(&mddev->bio_set);
6077 bioset_exit(&mddev->sync_set);
5eff3c43
N		 6078}
6079
390ee602 6080EXPORT_SYMBOL_GPL(md_stop);
6177b472 6081
a05b7ea0 6082static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N		 6083{
6084 int err = 0;
30b8feb7
N		 6085 int did_freeze = 0;
6086
6087 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6088 did_freeze = 1;
6089 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6090 md_wakeup_thread(mddev->thread);
6091 }
f851b60d 6092 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 6093 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 6094 if (mddev->sync_thread)
30b8feb7
N		 6095 /* Thread might be blocked waiting for metadata update
6096 * which will now never happen */
6097 wake_up_process(mddev->sync_thread->tsk);
f851b60d 6098
2953079c 6099 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
88724bfa 6100 return -EBUSY;
30b8feb7 6101 mddev_unlock(mddev);
f851b60d
N		 6102 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6103 &mddev->recovery));
88724bfa 6104 wait_event(mddev->sb_wait,
2953079c 6105 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
30b8feb7
N		 6106 mddev_lock_nointr(mddev);
6107
a4bd82d0 6108 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6109 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7 6110 mddev->sync_thread ||
af8d8e6f 6111 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9d48739e 6112 pr_warn("md: %s still in use.\n",mdname(mddev));
30b8feb7
N		 6113 if (did_freeze) {
6114 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 6115 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 6116 md_wakeup_thread(mddev->thread);
6117 }
a4bd82d0
N		 6118 err = -EBUSY;
6119 goto out;
6120 }
6121 if (mddev->pers) {
defad61a 6122 __md_stop_writes(mddev);
a4bd82d0
N		 6123
6124 err = -ENXIO;
6125 if (mddev->ro==1)
6126 goto out;
6127 mddev->ro = 1;
6128 set_disk_ro(mddev->gendisk, 1);
6129 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d
N		 6130 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6131 md_wakeup_thread(mddev->thread);
00bcb4ac 6132 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 6133 err = 0;
a4bd82d0
N		 6134 }
6135out:
6136 mutex_unlock(&mddev->open_mutex);
6137 return err;
6138}
6139
9e653b63
N		 6140/* mode:
6141 * 0 - completely stop and dis-assemble array
9e653b63
N		 6142 * 2 - stop but do not disassemble array
6143 */
f72ffdd6 6144static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 6145 struct block_device *bdev)
1da177e4 6146{
1da177e4 6147 struct gendisk *disk = mddev->gendisk;
3cb03002 6148 struct md_rdev *rdev;
30b8feb7
N		 6149 int did_freeze = 0;
6150
6151 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6152 did_freeze = 1;
6153 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6154 md_wakeup_thread(mddev->thread);
6155 }
f851b60d 6156 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
30b8feb7 6157 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
f851b60d 6158 if (mddev->sync_thread)
30b8feb7
N		 6159 /* Thread might be blocked waiting for metadata update
6160 * which will now never happen */
6161 wake_up_process(mddev->sync_thread->tsk);
f851b60d 6162
30b8feb7 6163 mddev_unlock(mddev);
f851b60d
N		 6164 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6165 !test_bit(MD_RECOVERY_RUNNING,
6166 &mddev->recovery)));
30b8feb7 6167 mddev_lock_nointr(mddev);
1da177e4 6168
c8c00a69 6169 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6170 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N		 6171 mddev->sysfs_active ||
6172 mddev->sync_thread ||
af8d8e6f 6173 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
9d48739e 6174 pr_warn("md: %s still in use.\n",mdname(mddev));
6e17b027 6175 mutex_unlock(&mddev->open_mutex);
30b8feb7
N		 6176 if (did_freeze) {
6177 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
45eaf45d 6178 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
30b8feb7
N		 6179 md_wakeup_thread(mddev->thread);
6180 }
260fa034
N		 6181 return -EBUSY;
6182 }
6e17b027 6183 if (mddev->pers) {
a4bd82d0
N		 6184 if (mddev->ro)
6185 set_disk_ro(disk, 0);
409c57f3 6186
defad61a 6187 __md_stop_writes(mddev);
5eff3c43 6188 __md_stop(mddev);
dc3b17cc 6189 mddev->queue->backing_dev_info->congested_fn = NULL;
6177b472 6190
a4bd82d0 6191 /* tell userspace to handle 'inactive' */
00bcb4ac 6192 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 6193
dafb20fa 6194 rdev_for_each(rdev, mddev)
36fad858
NK		 6195 if (rdev->raid_disk >= 0)
6196 sysfs_unlink_rdev(mddev, rdev);
c4647292 6197
a4bd82d0 6198 set_capacity(disk, 0);
6e17b027 6199 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 6200 mddev->changed = 1;
a4bd82d0 6201 revalidate_disk(disk);
0d4ca600 6202
a4bd82d0
N		 6203 if (mddev->ro)
6204 mddev->ro = 0;
6e17b027
N		 6205 } else
6206 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 6207 /*
6208 * Free resources if final stop
6209 */
9e653b63 6210 if (mode == 0) {
9d48739e 6211 pr_info("md: %s stopped.\n", mdname(mddev));
1da177e4 6212
c3d9714e 6213 if (mddev->bitmap_info.file) {
4af1a041
N		 6214 struct file *f = mddev->bitmap_info.file;
6215 spin_lock(&mddev->lock);
c3d9714e 6216 mddev->bitmap_info.file = NULL;
4af1a041
N		 6217 spin_unlock(&mddev->lock);
6218 fput(f);
978f946b 6219 }
c3d9714e 6220 mddev->bitmap_info.offset = 0;
978f946b 6221
1da177e4
LT		 6222 export_array(mddev);
6223
6177b472 6224 md_clean(mddev);
efeb53c0
N		 6225 if (mddev->hold_active == UNTIL_STOP)
6226 mddev->hold_active = 0;
a4bd82d0 6227 }
d7603b7e 6228 md_new_event(mddev);
00bcb4ac 6229 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 6230 return 0;
1da177e4
LT		 6231}
6232
fdee8ae4 6233#ifndef MODULE
fd01b88c 6234static void autorun_array(struct mddev *mddev)
1da177e4 6235{
3cb03002 6236 struct md_rdev *rdev;
1da177e4
LT		 6237 int err;
6238
a757e64c 6239 if (list_empty(&mddev->disks))
1da177e4 6240 return;
1da177e4 6241
9d48739e 6242 pr_info("md: running: ");
1da177e4 6243
dafb20fa 6244 rdev_for_each(rdev, mddev) {
1da177e4 6245 char b[BDEVNAME_SIZE];
9d48739e 6246 pr_cont("<%s>", bdevname(rdev->bdev,b));
1da177e4 6247 }
9d48739e 6248 pr_cont("\n");
1da177e4 6249
d710e138 6250 err = do_md_run(mddev);
1da177e4 6251 if (err) {
9d48739e 6252 pr_warn("md: do_md_run() returned %d\n", err);
a05b7ea0 6253 do_md_stop(mddev, 0, NULL);
1da177e4
LT		 6254 }
6255}
6256
6257/*
6258 * lets try to run arrays based on all disks that have arrived
6259 * until now. (those are in pending_raid_disks)
6260 *
6261 * the method: pick the first pending disk, collect all disks with
6262 * the same UUID, remove all from the pending list and put them into
6263 * the 'same_array' list. Then order this list based on superblock
6264 * update time (freshest comes first), kick out 'old' disks and
6265 * compare superblocks. If everything's fine then run it.
6266 *
6267 * If "unit" is allocated, then bump its reference count
6268 */
6269static void autorun_devices(int part)
6270{
3cb03002 6271 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 6272 struct mddev *mddev;
1da177e4
LT		 6273 char b[BDEVNAME_SIZE];
6274
9d48739e 6275 pr_info("md: autorun ...\n");
1da177e4 6276 while (!list_empty(&pending_raid_disks)) {
e8703fe1 6277 int unit;
1da177e4 6278 dev_t dev;
ad01c9e3 6279 LIST_HEAD(candidates);
1da177e4 6280 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 6281 struct md_rdev, same_set);
1da177e4 6282
9d48739e 6283 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
1da177e4 6284 INIT_LIST_HEAD(&candidates);
159ec1fc 6285 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4 6286 if (super_90_load(rdev, rdev0, 0) >= 0) {
9d48739e
N		 6287 pr_debug("md: adding %s ...\n",
6288 bdevname(rdev->bdev,b));
1da177e4
LT		 6289 list_move(&rdev->same_set, &candidates);
6290 }
6291 /*
6292 * now we have a set of devices, with all of them having
6293 * mostly sane superblocks. It's time to allocate the
6294 * mddev.
6295 */
e8703fe1
N		 6296 if (part) {
6297 dev = MKDEV(mdp_major,
6298 rdev0->preferred_minor << MdpMinorShift);
6299 unit = MINOR(dev) >> MdpMinorShift;
6300 } else {
6301 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6302 unit = MINOR(dev);
6303 }
6304 if (rdev0->preferred_minor != unit) {
9d48739e
N		 6305 pr_warn("md: unit number in %s is bad: %d\n",
6306 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1da177e4
LT		 6307 break;
6308 }
1da177e4
LT		 6309
6310 md_probe(dev, NULL, NULL);
6311 mddev = mddev_find(dev);
9bbbca3a
NB		 6312 if (!mddev || !mddev->gendisk) {
6313 if (mddev)
6314 mddev_put(mddev);
1da177e4
LT		 6315 break;
6316 }
f72ffdd6 6317 if (mddev_lock(mddev))
9d48739e 6318 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
1da177e4
LT		 6319 else if (mddev->raid_disks || mddev->major_version
6320 || !list_empty(&mddev->disks)) {
9d48739e 6321 pr_warn("md: %s already running, cannot run %s\n",
1da177e4
LT		 6322 mdname(mddev), bdevname(rdev0->bdev,b));
6323 mddev_unlock(mddev);
6324 } else {
9d48739e 6325 pr_debug("md: created %s\n", mdname(mddev));
1ec4a939 6326 mddev->persistent = 1;
159ec1fc 6327 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 6328 list_del_init(&rdev->same_set);
6329 if (bind_rdev_to_array(rdev, mddev))
6330 export_rdev(rdev);
6331 }
6332 autorun_array(mddev);
6333 mddev_unlock(mddev);
6334 }
6335 /* on success, candidates will be empty, on error
6336 * it won't...
6337 */
159ec1fc 6338 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 6339 list_del_init(&rdev->same_set);
1da177e4 6340 export_rdev(rdev);
4b80991c 6341 }
1da177e4
LT		 6342 mddev_put(mddev);
6343 }
9d48739e 6344 pr_info("md: ... autorun DONE.\n");
1da177e4 6345}
fdee8ae4 6346#endif /* !MODULE */
1da177e4 6347
f72ffdd6 6348static int get_version(void __user *arg)
1da177e4
LT		 6349{
6350 mdu_version_t ver;
6351
6352 ver.major = MD_MAJOR_VERSION;
6353 ver.minor = MD_MINOR_VERSION;
6354 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6355
6356 if (copy_to_user(arg, &ver, sizeof(ver)))
6357 return -EFAULT;
6358
6359 return 0;
6360}
6361
f72ffdd6 6362static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT		 6363{
6364 mdu_array_info_t info;
a9f326eb 6365 int nr,working,insync,failed,spare;
3cb03002 6366 struct md_rdev *rdev;
1da177e4 6367
1ca69c4b
N		 6368 nr = working = insync = failed = spare = 0;
6369 rcu_read_lock();
6370 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6371 nr++;
b2d444d7 6372 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 6373 failed++;
6374 else {
6375 working++;
b2d444d7 6376 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 6377 insync++;
b347af81
SL		 6378 else if (test_bit(Journal, &rdev->flags))
6379 /* TODO: add journal count to md_u.h */
6380 ;
1da177e4
LT		 6381 else
6382 spare++;
6383 }
6384 }
1ca69c4b 6385 rcu_read_unlock();
1da177e4
LT		 6386
6387 info.major_version = mddev->major_version;
6388 info.minor_version = mddev->minor_version;
6389 info.patch_version = MD_PATCHLEVEL_VERSION;
9ebc6ef1 6390 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1da177e4 6391 info.level = mddev->level;
58c0fed4
AN		 6392 info.size = mddev->dev_sectors / 2;
6393 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 6394 info.size = -1;
1da177e4
LT		 6395 info.nr_disks = nr;
6396 info.raid_disks = mddev->raid_disks;
6397 info.md_minor = mddev->md_minor;
6398 info.not_persistent= !mddev->persistent;
6399
9ebc6ef1 6400 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1da177e4
LT		 6401 info.state = 0;
6402 if (mddev->in_sync)
6403 info.state = (1<<MD_SB_CLEAN);
c3d9714e 6404 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 6405 info.state |= (1<<MD_SB_BITMAP_PRESENT);
ca8895d9
GR		 6406 if (mddev_is_clustered(mddev))
6407 info.state |= (1<<MD_SB_CLUSTERED);
a9f326eb 6408 info.active_disks = insync;
1da177e4
LT		 6409 info.working_disks = working;
6410 info.failed_disks = failed;
6411 info.spare_disks = spare;
6412
6413 info.layout = mddev->layout;
9d8f0363 6414 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 6415
6416 if (copy_to_user(arg, &info, sizeof(info)))
6417 return -EFAULT;
6418
6419 return 0;
6420}
6421
f72ffdd6 6422static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N		 6423{
6424 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
f4ad3d38 6425 char *ptr;
4af1a041 6426 int err;
32a7627c 6427
b6878d9e 6428 file = kzalloc(sizeof(*file), GFP_NOIO);
32a7627c 6429 if (!file)
4af1a041 6430 return -ENOMEM;
32a7627c 6431
4af1a041
N		 6432 err = 0;
6433 spin_lock(&mddev->lock);
25eafe1a
BR		 6434 /* bitmap enabled */
6435 if (mddev->bitmap_info.file) {
6436 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6437 sizeof(file->pathname));
6438 if (IS_ERR(ptr))
6439 err = PTR_ERR(ptr);
6440 else
6441 memmove(file->pathname, ptr,
6442 sizeof(file->pathname)-(ptr-file->pathname));
6443 }
4af1a041 6444 spin_unlock(&mddev->lock);
32a7627c 6445
4af1a041
N		 6446 if (err == 0 &&
6447 copy_to_user(arg, file, sizeof(*file)))
32a7627c 6448 err = -EFAULT;
4af1a041 6449
32a7627c
N		 6450 kfree(file);
6451 return err;
6452}
6453
f72ffdd6 6454static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT		 6455{
6456 mdu_disk_info_t info;
3cb03002 6457 struct md_rdev *rdev;
1da177e4
LT		 6458
6459 if (copy_from_user(&info, arg, sizeof(info)))
6460 return -EFAULT;
6461
1ca69c4b 6462 rcu_read_lock();
57d051dc 6463 rdev = md_find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT		 6464 if (rdev) {
6465 info.major = MAJOR(rdev->bdev->bd_dev);
6466 info.minor = MINOR(rdev->bdev->bd_dev);
6467 info.raid_disk = rdev->raid_disk;
6468 info.state = 0;
b2d444d7 6469 if (test_bit(Faulty, &rdev->flags))
1da177e4 6470 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 6471 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 6472 info.state |= (1<<MD_DISK_ACTIVE);
6473 info.state |= (1<<MD_DISK_SYNC);
6474 }
9efdca16 6475 if (test_bit(Journal, &rdev->flags))
bac624f3 6476 info.state |= (1<<MD_DISK_JOURNAL);
8ddf9efe
N		 6477 if (test_bit(WriteMostly, &rdev->flags))
6478 info.state |= (1<<MD_DISK_WRITEMOSTLY);
688834e6
N		 6479 if (test_bit(FailFast, &rdev->flags))
6480 info.state |= (1<<MD_DISK_FAILFAST);
1da177e4
LT		 6481 } else {
6482 info.major = info.minor = 0;
6483 info.raid_disk = -1;
6484 info.state = (1<<MD_DISK_REMOVED);
6485 }
1ca69c4b 6486 rcu_read_unlock();
1da177e4
LT		 6487
6488 if (copy_to_user(arg, &info, sizeof(info)))
6489 return -EFAULT;
6490
6491 return 0;
6492}
6493
f72ffdd6 6494static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT		 6495{
6496 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 6497 struct md_rdev *rdev;
1da177e4
LT		 6498 dev_t dev = MKDEV(info->major,info->minor);
6499
1aee41f6
GR		 6500 if (mddev_is_clustered(mddev) &&
6501 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
9d48739e
N		 6502 pr_warn("%s: Cannot add to clustered mddev.\n",
6503 mdname(mddev));
1aee41f6
GR		 6504 return -EINVAL;
6505 }
6506
1da177e4
LT		 6507 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6508 return -EOVERFLOW;
6509
6510 if (!mddev->raid_disks) {
6511 int err;
6512 /* expecting a device which has a superblock */
6513 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6514 if (IS_ERR(rdev)) {
9d48739e 6515 pr_warn("md: md_import_device returned %ld\n",
1da177e4
LT		 6516 PTR_ERR(rdev));
6517 return PTR_ERR(rdev);
6518 }
6519 if (!list_empty(&mddev->disks)) {
3cb03002
N		 6520 struct md_rdev *rdev0
6521 = list_entry(mddev->disks.next,
6522 struct md_rdev, same_set);
a9f326eb 6523 err = super_types[mddev->major_version]
1da177e4
LT		 6524 .load_super(rdev, rdev0, mddev->minor_version);
6525 if (err < 0) {
9d48739e 6526 pr_warn("md: %s has different UUID to %s\n",
f72ffdd6 6527 bdevname(rdev->bdev,b),
1da177e4
LT		 6528 bdevname(rdev0->bdev,b2));
6529 export_rdev(rdev);
6530 return -EINVAL;
6531 }
6532 }
6533 err = bind_rdev_to_array(rdev, mddev);
6534 if (err)
6535 export_rdev(rdev);
6536 return err;
6537 }
6538
6539 /*
6540 * add_new_disk can be used once the array is assembled
6541 * to add "hot spares". They must already have a superblock
6542 * written
6543 */
6544 if (mddev->pers) {
6545 int err;
6546 if (!mddev->pers->hot_add_disk) {
9d48739e
N		 6547 pr_warn("%s: personality does not support diskops!\n",
6548 mdname(mddev));
1da177e4
LT		 6549 return -EINVAL;
6550 }
7b1e35f6
N		 6551 if (mddev->persistent)
6552 rdev = md_import_device(dev, mddev->major_version,
6553 mddev->minor_version);
6554 else
6555 rdev = md_import_device(dev, -1, -1);
1da177e4 6556 if (IS_ERR(rdev)) {
9d48739e 6557 pr_warn("md: md_import_device returned %ld\n",
1da177e4
LT		 6558 PTR_ERR(rdev));
6559 return PTR_ERR(rdev);
6560 }
1a855a06 6561 /* set saved_raid_disk if appropriate */
41158c7e
N		 6562 if (!mddev->persistent) {
6563 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 6564 info->raid_disk < mddev->raid_disks) {
41158c7e 6565 rdev->raid_disk = info->raid_disk;
bf572541 6566 set_bit(In_sync, &rdev->flags);
8313b8e5 6567 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 6568 } else
41158c7e 6569 rdev->raid_disk = -1;
f466722c 6570 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N		 6571 } else
6572 super_types[mddev->major_version].
6573 validate_super(mddev, rdev);
bedd86b7 6574 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 6575 rdev->raid_disk != info->raid_disk) {
bedd86b7
N		 6576 /* This was a hot-add request, but events doesn't
6577 * match, so reject it.
6578 */
6579 export_rdev(rdev);
6580 return -EINVAL;
6581 }
6582
b2d444d7 6583 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 6584 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6585 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 6586 else
6587 clear_bit(WriteMostly, &rdev->flags);
688834e6
N		 6588 if (info->state & (1<<MD_DISK_FAILFAST))
6589 set_bit(FailFast, &rdev->flags);
6590 else
6591 clear_bit(FailFast, &rdev->flags);
8ddf9efe 6592
f6b6ec5c
SL		 6593 if (info->state & (1<<MD_DISK_JOURNAL)) {
6594 struct md_rdev *rdev2;
6595 bool has_journal = false;
6596
6597 /* make sure no existing journal disk */
6598 rdev_for_each(rdev2, mddev) {
6599 if (test_bit(Journal, &rdev2->flags)) {
6600 has_journal = true;
6601 break;
6602 }
6603 }
230b55fa 6604 if (has_journal || mddev->bitmap) {
f6b6ec5c
SL		 6605 export_rdev(rdev);
6606 return -EBUSY;
6607 }
bac624f3 6608 set_bit(Journal, &rdev->flags);
f6b6ec5c 6609 }
1aee41f6
GR		 6610 /*
6611 * check whether the device shows up in other nodes
6612 */
6613 if (mddev_is_clustered(mddev)) {
dbb64f86 6614 if (info->state & (1 << MD_DISK_CANDIDATE))
1aee41f6 6615 set_bit(Candidate, &rdev->flags);
dbb64f86 6616 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
1aee41f6 6617 /* --add initiated by this node */
dbb64f86 6618 err = md_cluster_ops->add_new_disk(mddev, rdev);
1aee41f6 6619 if (err) {
1aee41f6
GR		 6620 export_rdev(rdev);
6621 return err;
6622 }
6623 }
6624 }
6625
1da177e4
LT		 6626 rdev->raid_disk = -1;
6627 err = bind_rdev_to_array(rdev, mddev);
dbb64f86 6628
1da177e4
LT		 6629 if (err)
6630 export_rdev(rdev);
dbb64f86
GR		 6631
6632 if (mddev_is_clustered(mddev)) {
e566aef1
GJ		 6633 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6634 if (!err) {
6635 err = md_cluster_ops->new_disk_ack(mddev,
6636 err == 0);
6637 if (err)
6638 md_kick_rdev_from_array(rdev);
6639 }
6640 } else {
dbb64f86
GR		 6641 if (err)
6642 md_cluster_ops->add_new_disk_cancel(mddev);
6643 else
6644 err = add_bound_rdev(rdev);
6645 }
6646
6647 } else if (!err)
a6da4ef8 6648 err = add_bound_rdev(rdev);
dbb64f86 6649
1da177e4
LT		 6650 return err;
6651 }
6652
6653 /* otherwise, add_new_disk is only allowed
6654 * for major_version==0 superblocks
6655 */
6656 if (mddev->major_version != 0) {
9d48739e 6657 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
1da177e4
LT		 6658 return -EINVAL;
6659 }
6660
6661 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6662 int err;
d710e138 6663 rdev = md_import_device(dev, -1, 0);
1da177e4 6664 if (IS_ERR(rdev)) {
9d48739e 6665 pr_warn("md: error, md_import_device() returned %ld\n",
1da177e4
LT		 6666 PTR_ERR(rdev));
6667 return PTR_ERR(rdev);
6668 }
6669 rdev->desc_nr = info->number;
6670 if (info->raid_disk < mddev->raid_disks)
6671 rdev->raid_disk = info->raid_disk;
6672 else
6673 rdev->raid_disk = -1;
6674
1da177e4 6675 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 6676 if (info->state & (1<<MD_DISK_SYNC))
6677 set_bit(In_sync, &rdev->flags);
1da177e4 6678
8ddf9efe
N		 6679 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6680 set_bit(WriteMostly, &rdev->flags);
688834e6
N		 6681 if (info->state & (1<<MD_DISK_FAILFAST))
6682 set_bit(FailFast, &rdev->flags);
8ddf9efe 6683
1da177e4 6684 if (!mddev->persistent) {
9d48739e 6685 pr_debug("md: nonpersistent superblock ...\n");
77304d2a
MS		 6686 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6687 } else
57b2caa3 6688 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 6689 rdev->sectors = rdev->sb_start;
1da177e4 6690
2bf071bf
N		 6691 err = bind_rdev_to_array(rdev, mddev);
6692 if (err) {
6693 export_rdev(rdev);
6694 return err;
6695 }
1da177e4
LT		 6696 }
6697
6698 return 0;
6699}
6700
f72ffdd6 6701static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6702{
6703 char b[BDEVNAME_SIZE];
3cb03002 6704 struct md_rdev *rdev;
1da177e4 6705
c42a0e26
YY		 6706 if (!mddev->pers)
6707 return -ENODEV;
6708
1da177e4
LT		 6709 rdev = find_rdev(mddev, dev);
6710 if (!rdev)
6711 return -ENXIO;
6712
2910ff17
GR		 6713 if (rdev->raid_disk < 0)
6714 goto kick_rdev;
293467aa 6715
3ea8929d
N		 6716 clear_bit(Blocked, &rdev->flags);
6717 remove_and_add_spares(mddev, rdev);
6718
1da177e4
LT		 6719 if (rdev->raid_disk >= 0)
6720 goto busy;
6721
2910ff17 6722kick_rdev:
54a88392 6723 if (mddev_is_clustered(mddev))
88bcfef7
GR		 6724 md_cluster_ops->remove_disk(mddev, rdev);
6725
fb56dfef 6726 md_kick_rdev_from_array(rdev);
2953079c 6727 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
060b0689
N		 6728 if (mddev->thread)
6729 md_wakeup_thread(mddev->thread);
6730 else
6731 md_update_sb(mddev, 1);
d7603b7e 6732 md_new_event(mddev);
1da177e4
LT		 6733
6734 return 0;
6735busy:
9d48739e
N		 6736 pr_debug("md: cannot remove active disk %s from %s ...\n",
6737 bdevname(rdev->bdev,b), mdname(mddev));
1da177e4
LT		 6738 return -EBUSY;
6739}
6740
f72ffdd6 6741static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT		 6742{
6743 char b[BDEVNAME_SIZE];
6744 int err;
3cb03002 6745 struct md_rdev *rdev;
1da177e4
LT		 6746
6747 if (!mddev->pers)
6748 return -ENODEV;
6749
6750 if (mddev->major_version != 0) {
9d48739e 6751 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
1da177e4
LT		 6752 mdname(mddev));
6753 return -EINVAL;
6754 }
6755 if (!mddev->pers->hot_add_disk) {
9d48739e 6756 pr_warn("%s: personality does not support diskops!\n",
1da177e4
LT		 6757 mdname(mddev));
6758 return -EINVAL;
6759 }
6760
d710e138 6761 rdev = md_import_device(dev, -1, 0);
1da177e4 6762 if (IS_ERR(rdev)) {
9d48739e 6763 pr_warn("md: error, md_import_device() returned %ld\n",
1da177e4
LT		 6764 PTR_ERR(rdev));
6765 return -EINVAL;
6766 }
6767
6768 if (mddev->persistent)
57b2caa3 6769 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 6770 else
77304d2a 6771 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 6772
8190e754 6773 rdev->sectors = rdev->sb_start;
1da177e4 6774
b2d444d7 6775 if (test_bit(Faulty, &rdev->flags)) {
9d48739e 6776 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
1da177e4
LT		 6777 bdevname(rdev->bdev,b), mdname(mddev));
6778 err = -EINVAL;
6779 goto abort_export;
6780 }
293467aa 6781
b2d444d7 6782 clear_bit(In_sync, &rdev->flags);
1da177e4 6783 rdev->desc_nr = -1;
5842730d 6784 rdev->saved_raid_disk = -1;
2bf071bf
N		 6785 err = bind_rdev_to_array(rdev, mddev);
6786 if (err)
2aa82191 6787 goto abort_export;
1da177e4
LT		 6788
6789 /*
6790 * The rest should better be atomic, we can have disk failures
6791 * noticed in interrupt contexts ...
6792 */
6793
1da177e4
LT		 6794 rdev->raid_disk = -1;
6795
2953079c 6796 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
060b0689
N		 6797 if (!mddev->thread)
6798 md_update_sb(mddev, 1);
1da177e4
LT		 6799 /*
6800 * Kick recovery, maybe this spare has to be added to the
6801 * array immediately.
6802 */
6803 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6804 md_wakeup_thread(mddev->thread);
d7603b7e 6805 md_new_event(mddev);
1da177e4
LT		 6806 return 0;
6807
1da177e4
LT		 6808abort_export:
6809 export_rdev(rdev);
6810 return err;
6811}
6812
fd01b88c 6813static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 6814{
035328c2 6815 int err = 0;
32a7627c 6816
36fa3063 6817 if (mddev->pers) {
d66b1b39 6818 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N		 6819 return -EBUSY;
6820 if (mddev->recovery || mddev->sync_thread)
6821 return -EBUSY;
6822 /* we should be able to change the bitmap.. */
6823 }
32a7627c 6824
36fa3063 6825 if (fd >= 0) {
035328c2 6826 struct inode *inode;
1e594bb2
N		 6827 struct file *f;
6828
6829 if (mddev->bitmap || mddev->bitmap_info.file)
36fa3063 6830 return -EEXIST; /* cannot add when bitmap is present */
1e594bb2 6831 f = fget(fd);
32a7627c 6832
1e594bb2 6833 if (f == NULL) {
9d48739e
N		 6834 pr_warn("%s: error: failed to get bitmap file\n",
6835 mdname(mddev));
36fa3063
N		 6836 return -EBADF;
6837 }
6838
1e594bb2 6839 inode = f->f_mapping->host;
035328c2 6840 if (!S_ISREG(inode->i_mode)) {
9d48739e
N		 6841 pr_warn("%s: error: bitmap file must be a regular file\n",
6842 mdname(mddev));
035328c2 6843 err = -EBADF;
1e594bb2 6844 } else if (!(f->f_mode & FMODE_WRITE)) {
9d48739e
N		 6845 pr_warn("%s: error: bitmap file must open for write\n",
6846 mdname(mddev));
035328c2
N		 6847 err = -EBADF;
6848 } else if (atomic_read(&inode->i_writecount) != 1) {
9d48739e
N		 6849 pr_warn("%s: error: bitmap file is already in use\n",
6850 mdname(mddev));
035328c2
N		 6851 err = -EBUSY;
6852 }
6853 if (err) {
1e594bb2 6854 fput(f);
36fa3063
N		 6855 return err;
6856 }
1e594bb2 6857 mddev->bitmap_info.file = f;
c3d9714e 6858 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 6859 } else if (mddev->bitmap == NULL)
6860 return -ENOENT; /* cannot remove what isn't there */
6861 err = 0;
6862 if (mddev->pers) {
69e51b44 6863 if (fd >= 0) {
f9209a32
GR		 6864 struct bitmap *bitmap;
6865
e64e4018 6866 bitmap = md_bitmap_create(mddev, -1);
9e1cc0a5 6867 mddev_suspend(mddev);
f9209a32
GR		 6868 if (!IS_ERR(bitmap)) {
6869 mddev->bitmap = bitmap;
e64e4018 6870 err = md_bitmap_load(mddev);
ba599aca
N		 6871 } else
6872 err = PTR_ERR(bitmap);
52a0d49d 6873 if (err) {
e64e4018 6874 md_bitmap_destroy(mddev);
52a0d49d
N		 6875 fd = -1;
6876 }
9e1cc0a5 6877 mddev_resume(mddev);
52a0d49d 6878 } else if (fd < 0) {
9e1cc0a5 6879 mddev_suspend(mddev);
e64e4018 6880 md_bitmap_destroy(mddev);
9e1cc0a5 6881 mddev_resume(mddev);
d7375ab3 6882 }
d7375ab3
N		 6883 }
6884 if (fd < 0) {
4af1a041
N		 6885 struct file *f = mddev->bitmap_info.file;
6886 if (f) {
6887 spin_lock(&mddev->lock);
6888 mddev->bitmap_info.file = NULL;
6889 spin_unlock(&mddev->lock);
6890 fput(f);
6891 }
36fa3063
N		 6892 }
6893
32a7627c
N		 6894 return err;
6895}
6896
1da177e4
LT		 6897/*
6898 * set_array_info is used two different ways
6899 * The original usage is when creating a new array.
6900 * In this usage, raid_disks is > 0 and it together with
6901 * level, size, not_persistent,layout,chunksize determine the
6902 * shape of the array.
6903 * This will always create an array with a type-0.90.0 superblock.
6904 * The newer usage is when assembling an array.
6905 * In this case raid_disks will be 0, and the major_version field is
6906 * use to determine which style super-blocks are to be found on the devices.
6907 * The minor and patch _version numbers are also kept incase the
6908 * super_block handler wishes to interpret them.
6909 */
f72ffdd6 6910static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 6911{
6912
6913 if (info->raid_disks == 0) {
6914 /* just setting version number for superblock loading */
6915 if (info->major_version < 0 ||
50511da3 6916 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 6917 super_types[info->major_version].name == NULL) {
6918 /* maybe try to auto-load a module? */
9d48739e 6919 pr_warn("md: superblock version %d not known\n",
1da177e4
LT		 6920 info->major_version);
6921 return -EINVAL;
6922 }
6923 mddev->major_version = info->major_version;
6924 mddev->minor_version = info->minor_version;
6925 mddev->patch_version = info->patch_version;
3f9d7b0d 6926 mddev->persistent = !info->not_persistent;
cbd19983
N		 6927 /* ensure mddev_put doesn't delete this now that there
6928 * is some minimal configuration.
6929 */
9ebc6ef1 6930 mddev->ctime = ktime_get_real_seconds();
1da177e4
LT		 6931 return 0;
6932 }
6933 mddev->major_version = MD_MAJOR_VERSION;
6934 mddev->minor_version = MD_MINOR_VERSION;
6935 mddev->patch_version = MD_PATCHLEVEL_VERSION;
9ebc6ef1 6936 mddev->ctime = ktime_get_real_seconds();
1da177e4
LT		 6937
6938 mddev->level = info->level;
17115e03 6939 mddev->clevel[0] = 0;
58c0fed4 6940 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 6941 mddev->raid_disks = info->raid_disks;
6942 /* don't set md_minor, it is determined by which /dev/md* was
6943 * openned
6944 */
6945 if (info->state & (1<<MD_SB_CLEAN))
6946 mddev->recovery_cp = MaxSector;
6947 else
6948 mddev->recovery_cp = 0;
6949 mddev->persistent = ! info->not_persistent;
e691063a 6950 mddev->external = 0;
1da177e4
LT		 6951
6952 mddev->layout = info->layout;
33f2c35a
N		 6953 if (mddev->level == 0)
6954 /* Cannot trust RAID0 layout info here */
6955 mddev->layout = -1;
9d8f0363 6956 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4 6957
2953079c 6958 if (mddev->persistent) {
1b3bae49
N		 6959 mddev->max_disks = MD_SB_DISKS;
6960 mddev->flags = 0;
6961 mddev->sb_flags = 0;
2953079c
SL		 6962 }
6963 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
1da177e4 6964
c3d9714e 6965 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6966 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6967 mddev->bitmap_info.offset = 0;
b2a2703c 6968
f6705578
N		 6969 mddev->reshape_position = MaxSector;
6970
1da177e4
LT		 6971 /*
6972 * Generate a 128 bit UUID
6973 */
6974 get_random_bytes(mddev->uuid, 16);
6975
f6705578 6976 mddev->new_level = mddev->level;
664e7c41 6977 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 6978 mddev->new_layout = mddev->layout;
6979 mddev->delta_disks = 0;
2c810cdd 6980 mddev->reshape_backwards = 0;
f6705578 6981
1da177e4
LT		 6982 return 0;
6983}
6984
fd01b88c 6985void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6986{
efa4b77b 6987 lockdep_assert_held(&mddev->reconfig_mutex);
b522adcd
DW		 6988
6989 if (mddev->external_size)
6990 return;
6991
1f403624
DW		 6992 mddev->array_sectors = array_sectors;
6993}
6994EXPORT_SYMBOL(md_set_array_sectors);
6995
fd01b88c 6996static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6997{
3cb03002 6998 struct md_rdev *rdev;
a35b0d69 6999 int rv;
d71f9f88 7000 int fit = (num_sectors == 0);
818da59f 7001 sector_t old_dev_sectors = mddev->dev_sectors;
ab5a98b1 7002
a35b0d69
N		 7003 if (mddev->pers->resize == NULL)
7004 return -EINVAL;
d71f9f88
AN		 7005 /* The "num_sectors" is the number of sectors of each device that
7006 * is used. This can only make sense for arrays with redundancy.
7007 * linear and raid0 always use whatever space is available. We can only
7008 * consider changing this number if no resync or reconstruction is
7009 * happening, and if the new size is acceptable. It must fit before the
0f420358 7010 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 7011 * of each device. If num_sectors is zero, we find the largest size
7012 * that fits.
a35b0d69 7013 */
f851b60d
N		 7014 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7015 mddev->sync_thread)
a35b0d69 7016 return -EBUSY;
bd8839e0
N		 7017 if (mddev->ro)
7018 return -EROFS;
a4a6125a 7019
dafb20fa 7020 rdev_for_each(rdev, mddev) {
dd8ac336 7021 sector_t avail = rdev->sectors;
01ab5662 7022
d71f9f88
AN		 7023 if (fit && (num_sectors == 0 || num_sectors > avail))
7024 num_sectors = avail;
7025 if (avail < num_sectors)
a35b0d69
N		 7026 return -ENOSPC;
7027 }
d71f9f88 7028 rv = mddev->pers->resize(mddev, num_sectors);
c9483634 7029 if (!rv) {
818da59f
GJ		 7030 if (mddev_is_clustered(mddev))
7031 md_cluster_ops->update_size(mddev, old_dev_sectors);
7032 else if (mddev->queue) {
c9483634
GJ		 7033 set_capacity(mddev->gendisk, mddev->array_sectors);
7034 revalidate_disk(mddev->gendisk);
7035 }
7036 }
a35b0d69
N		 7037 return rv;
7038}
7039
fd01b88c 7040static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 7041{
7042 int rv;
c6563a8c 7043 struct md_rdev *rdev;
da943b99 7044 /* change the number of raid disks */
63c70c4f 7045 if (mddev->pers->check_reshape == NULL)
da943b99 7046 return -EINVAL;
bd8839e0
N		 7047 if (mddev->ro)
7048 return -EROFS;
da943b99 7049 if (raid_disks <= 0 ||
233fca36 7050 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 7051 return -EINVAL;
f851b60d
N		 7052 if (mddev->sync_thread ||
7053 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7054 mddev->reshape_position != MaxSector)
da943b99 7055 return -EBUSY;
c6563a8c
N		 7056
7057 rdev_for_each(rdev, mddev) {
7058 if (mddev->raid_disks < raid_disks &&
7059 rdev->data_offset < rdev->new_data_offset)
7060 return -EINVAL;
7061 if (mddev->raid_disks > raid_disks &&
7062 rdev->data_offset > rdev->new_data_offset)
7063 return -EINVAL;
7064 }
7065
63c70c4f 7066 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N		 7067 if (mddev->delta_disks < 0)
7068 mddev->reshape_backwards = 1;
7069 else if (mddev->delta_disks > 0)
7070 mddev->reshape_backwards = 0;
63c70c4f
N		 7071
7072 rv = mddev->pers->check_reshape(mddev);
2c810cdd 7073 if (rv < 0) {
de171cb9 7074 mddev->delta_disks = 0;
2c810cdd
N		 7075 mddev->reshape_backwards = 0;
7076 }
da943b99
N		 7077 return rv;
7078}
7079
1da177e4
LT		 7080/*
7081 * update_array_info is used to change the configuration of an
7082 * on-line array.
7083 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7084 * fields in the info are checked against the array.
7085 * Any differences that cannot be handled will cause an error.
7086 * Normally, only one change can be managed at a time.
7087 */
fd01b88c 7088static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 7089{
7090 int rv = 0;
7091 int cnt = 0;
36fa3063
N		 7092 int state = 0;
7093
7094 /* calculate expected state,ignoring low bits */
c3d9714e 7095 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 7096 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 7097
7098 if (mddev->major_version != info->major_version ||
7099 mddev->minor_version != info->minor_version ||
7100/* mddev->patch_version != info->patch_version || */
7101 mddev->ctime != info->ctime ||
7102 mddev->level != info->level ||
7103/* mddev->layout != info->layout || */
4e023612 7104 mddev->persistent != !info->not_persistent ||
9d8f0363 7105 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 7106 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7107 ((state^info->state) & 0xfffffe00)
7108 )
1da177e4
LT		 7109 return -EINVAL;
7110 /* Check there is only one change */
58c0fed4
AN		 7111 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7112 cnt++;
7113 if (mddev->raid_disks != info->raid_disks)
7114 cnt++;
7115 if (mddev->layout != info->layout)
7116 cnt++;
7117 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7118 cnt++;
7119 if (cnt == 0)
7120 return 0;
7121 if (cnt > 1)
7122 return -EINVAL;
1da177e4
LT		 7123
7124 if (mddev->layout != info->layout) {
7125 /* Change layout
7126 * we don't need to do anything at the md level, the
7127 * personality will take care of it all.
7128 */
50ac168a 7129 if (mddev->pers->check_reshape == NULL)
1da177e4 7130 return -EINVAL;
597a711b
N		 7131 else {
7132 mddev->new_layout = info->layout;
50ac168a 7133 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 7134 if (rv)
7135 mddev->new_layout = mddev->layout;
7136 return rv;
7137 }
1da177e4 7138 }
58c0fed4 7139 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 7140 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 7141
da943b99
N		 7142 if (mddev->raid_disks != info->raid_disks)
7143 rv = update_raid_disks(mddev, info->raid_disks);
7144
36fa3063 7145 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
293467aa
GR		 7146 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7147 rv = -EINVAL;
7148 goto err;
7149 }
7150 if (mddev->recovery || mddev->sync_thread) {
7151 rv = -EBUSY;
7152 goto err;
7153 }
36fa3063 7154 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
f9209a32 7155 struct bitmap *bitmap;
36fa3063 7156 /* add the bitmap */
293467aa
GR		 7157 if (mddev->bitmap) {
7158 rv = -EEXIST;
7159 goto err;
7160 }
7161 if (mddev->bitmap_info.default_offset == 0) {
7162 rv = -EINVAL;
7163 goto err;
7164 }
c3d9714e
N		 7165 mddev->bitmap_info.offset =
7166 mddev->bitmap_info.default_offset;
6409bb05
N		 7167 mddev->bitmap_info.space =
7168 mddev->bitmap_info.default_space;
e64e4018 7169 bitmap = md_bitmap_create(mddev, -1);
9e1cc0a5 7170 mddev_suspend(mddev);
f9209a32
GR		 7171 if (!IS_ERR(bitmap)) {
7172 mddev->bitmap = bitmap;
e64e4018 7173 rv = md_bitmap_load(mddev);
ba599aca
N		 7174 } else
7175 rv = PTR_ERR(bitmap);
36fa3063 7176 if (rv)
e64e4018 7177 md_bitmap_destroy(mddev);
9e1cc0a5 7178 mddev_resume(mddev);
36fa3063
N		 7179 } else {
7180 /* remove the bitmap */
293467aa
GR		 7181 if (!mddev->bitmap) {
7182 rv = -ENOENT;
7183 goto err;
7184 }
7185 if (mddev->bitmap->storage.file) {
7186 rv = -EINVAL;
7187 goto err;
7188 }
f6a2dc64
GJ		 7189 if (mddev->bitmap_info.nodes) {
7190 /* hold PW on all the bitmap lock */
7191 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
9d48739e 7192 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
f6a2dc64
GJ		 7193 rv = -EPERM;
7194 md_cluster_ops->unlock_all_bitmaps(mddev);
7195 goto err;
7196 }
7197
7198 mddev->bitmap_info.nodes = 0;
7199 md_cluster_ops->leave(mddev);
7200 }
9e1cc0a5 7201 mddev_suspend(mddev);
e64e4018 7202 md_bitmap_destroy(mddev);
9e1cc0a5 7203 mddev_resume(mddev);
c3d9714e 7204 mddev->bitmap_info.offset = 0;
36fa3063
N		 7205 }
7206 }
850b2b42 7207 md_update_sb(mddev, 1);
293467aa
GR		 7208 return rv;
7209err:
1da177e4
LT		 7210 return rv;
7211}
7212
fd01b88c 7213static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 7214{
3cb03002 7215 struct md_rdev *rdev;
1ca69c4b 7216 int err = 0;
1da177e4
LT		 7217
7218 if (mddev->pers == NULL)
7219 return -ENODEV;
7220
1ca69c4b 7221 rcu_read_lock();
1532d9e8 7222 rdev = md_find_rdev_rcu(mddev, dev);
1da177e4 7223 if (!rdev)
1ca69c4b
N		 7224 err = -ENODEV;
7225 else {
7226 md_error(mddev, rdev);
7227 if (!test_bit(Faulty, &rdev->flags))
7228 err = -EBUSY;
7229 }
7230 rcu_read_unlock();
7231 return err;
1da177e4
LT		 7232}
7233
2f9618ce
AN		 7234/*
7235 * We have a problem here : there is no easy way to give a CHS
7236 * virtual geometry. We currently pretend that we have a 2 heads
7237 * 4 sectors (with a BIG number of cylinders...). This drives
7238 * dosfs just mad... ;-)
7239 */
a885c8c4
CH		 7240static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7241{
fd01b88c 7242 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 7243
7244 geo->heads = 2;
7245 geo->sectors = 4;
49ce6cea 7246 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 7247 return 0;
7248}
7249
cb335f88
NS		 7250static inline bool md_ioctl_valid(unsigned int cmd)
7251{
7252 switch (cmd) {
7253 case ADD_NEW_DISK:
7254 case BLKROSET:
7255 case GET_ARRAY_INFO:
7256 case GET_BITMAP_FILE:
7257 case GET_DISK_INFO:
7258 case HOT_ADD_DISK:
7259 case HOT_REMOVE_DISK:
cb335f88
NS		 7260 case RAID_AUTORUN:
7261 case RAID_VERSION:
7262 case RESTART_ARRAY_RW:
7263 case RUN_ARRAY:
7264 case SET_ARRAY_INFO:
7265 case SET_BITMAP_FILE:
7266 case SET_DISK_FAULTY:
7267 case STOP_ARRAY:
7268 case STOP_ARRAY_RO:
1aee41f6 7269 case CLUSTERED_DISK_NACK:
cb335f88
NS		 7270 return true;
7271 default:
7272 return false;
7273 }
7274}
7275
a39907fa 7276static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 7277 unsigned int cmd, unsigned long arg)
7278{
7279 int err = 0;
7280 void __user *argp = (void __user *)arg;
fd01b88c 7281 struct mddev *mddev = NULL;
e2218350 7282 int ro;
065e519e 7283 bool did_set_md_closing = false;
1da177e4 7284
cb335f88
NS		 7285 if (!md_ioctl_valid(cmd))
7286 return -ENOTTY;
7287
506c9e44
N		 7288 switch (cmd) {
7289 case RAID_VERSION:
7290 case GET_ARRAY_INFO:
7291 case GET_DISK_INFO:
7292 break;
7293 default:
7294 if (!capable(CAP_SYS_ADMIN))
7295 return -EACCES;
7296 }
1da177e4
LT		 7297
7298 /*
7299 * Commands dealing with the RAID driver but not any
7300 * particular array:
7301 */
c02c0aeb
N		 7302 switch (cmd) {
7303 case RAID_VERSION:
7304 err = get_version(argp);
3adc28d8 7305 goto out;
1da177e4 7306
1da177e4 7307#ifndef MODULE
c02c0aeb
N		 7308 case RAID_AUTORUN:
7309 err = 0;
7310 autostart_arrays(arg);
3adc28d8 7311 goto out;
1da177e4 7312#endif
c02c0aeb 7313 default:;
1da177e4
LT		 7314 }
7315
7316 /*
7317 * Commands creating/starting a new array:
7318 */
7319
a39907fa 7320 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 7321
7322 if (!mddev) {
7323 BUG();
3adc28d8 7324 goto out;
1da177e4
LT		 7325 }
7326
1ca69c4b
N		 7327 /* Some actions do not requires the mutex */
7328 switch (cmd) {
7329 case GET_ARRAY_INFO:
7330 if (!mddev->raid_disks && !mddev->external)
7331 err = -ENODEV;
7332 else
7333 err = get_array_info(mddev, argp);
3adc28d8 7334 goto out;
1ca69c4b
N		 7335
7336 case GET_DISK_INFO:
7337 if (!mddev->raid_disks && !mddev->external)
7338 err = -ENODEV;
7339 else
7340 err = get_disk_info(mddev, argp);
3adc28d8 7341 goto out;
1ca69c4b
N		 7342
7343 case SET_DISK_FAULTY:
7344 err = set_disk_faulty(mddev, new_decode_dev(arg));
3adc28d8 7345 goto out;
4af1a041
N		 7346
7347 case GET_BITMAP_FILE:
7348 err = get_bitmap_file(mddev, argp);
7349 goto out;
7350
1ca69c4b
N		 7351 }
7352
a7a3f08d
N		 7353 if (cmd == ADD_NEW_DISK)
7354 /* need to ensure md_delayed_delete() has completed */
7355 flush_workqueue(md_misc_wq);
7356
90f5f7ad
HR		 7357 if (cmd == HOT_REMOVE_DISK)
7358 /* need to ensure recovery thread has run */
7359 wait_event_interruptible_timeout(mddev->sb_wait,
7360 !test_bit(MD_RECOVERY_NEEDED,
82a301cb 7361 &mddev->recovery),
90f5f7ad 7362 msecs_to_jiffies(5000));
260fa034
N		 7363 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7364 /* Need to flush page cache, and ensure no-one else opens
7365 * and writes
7366 */
7367 mutex_lock(&mddev->open_mutex);
9ba3b7f5 7368 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N		 7369 mutex_unlock(&mddev->open_mutex);
7370 err = -EBUSY;
3adc28d8 7371 goto out;
260fa034 7372 }
065e519e 7373 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
af8d8e6f 7374 set_bit(MD_CLOSING, &mddev->flags);
065e519e 7375 did_set_md_closing = true;
260fa034
N		 7376 mutex_unlock(&mddev->open_mutex);
7377 sync_blockdev(bdev);
7378 }
1da177e4
LT		 7379 err = mddev_lock(mddev);
7380 if (err) {
9d48739e
N		 7381 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7382 err, cmd);
3adc28d8 7383 goto out;
1da177e4
LT		 7384 }
7385
c02c0aeb
N		 7386 if (cmd == SET_ARRAY_INFO) {
7387 mdu_array_info_t info;
7388 if (!arg)
7389 memset(&info, 0, sizeof(info));
7390 else if (copy_from_user(&info, argp, sizeof(info))) {
7391 err = -EFAULT;
3adc28d8 7392 goto unlock;
c02c0aeb
N		 7393 }
7394 if (mddev->pers) {
7395 err = update_array_info(mddev, &info);
7396 if (err) {
9d48739e 7397 pr_warn("md: couldn't update array info. %d\n", err);
3adc28d8 7398 goto unlock;
1da177e4 7399 }
3adc28d8 7400 goto unlock;
c02c0aeb
N		 7401 }
7402 if (!list_empty(&mddev->disks)) {
9d48739e 7403 pr_warn("md: array %s already has disks!\n", mdname(mddev));
c02c0aeb 7404 err = -EBUSY;
3adc28d8 7405 goto unlock;
c02c0aeb
N		 7406 }
7407 if (mddev->raid_disks) {
9d48739e 7408 pr_warn("md: array %s already initialised!\n", mdname(mddev));
c02c0aeb 7409 err = -EBUSY;
3adc28d8 7410 goto unlock;
c02c0aeb
N		 7411 }
7412 err = set_array_info(mddev, &info);
7413 if (err) {
9d48739e 7414 pr_warn("md: couldn't set array info. %d\n", err);
3adc28d8 7415 goto unlock;
c02c0aeb 7416 }
3adc28d8 7417 goto unlock;
1da177e4
LT		 7418 }
7419
7420 /*
7421 * Commands querying/configuring an existing array:
7422 */
32a7627c 7423 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 7424 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 7425 if ((!mddev->raid_disks && !mddev->external)
7426 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7427 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7428 && cmd != GET_BITMAP_FILE) {
1da177e4 7429 err = -ENODEV;
3adc28d8 7430 goto unlock;
1da177e4
LT		 7431 }
7432
7433 /*
7434 * Commands even a read-only array can execute:
7435 */
c02c0aeb 7436 switch (cmd) {
c02c0aeb
N		 7437 case RESTART_ARRAY_RW:
7438 err = restart_array(mddev);
3adc28d8 7439 goto unlock;
1da177e4 7440
c02c0aeb
N		 7441 case STOP_ARRAY:
7442 err = do_md_stop(mddev, 0, bdev);
3adc28d8 7443 goto unlock;
1da177e4 7444
c02c0aeb
N		 7445 case STOP_ARRAY_RO:
7446 err = md_set_readonly(mddev, bdev);
3adc28d8 7447 goto unlock;
1da177e4 7448
3ea8929d
N		 7449 case HOT_REMOVE_DISK:
7450 err = hot_remove_disk(mddev, new_decode_dev(arg));
3adc28d8 7451 goto unlock;
3ea8929d 7452
7ceb17e8
N		 7453 case ADD_NEW_DISK:
7454 /* We can support ADD_NEW_DISK on read-only arrays
466ad292 7455 * only if we are re-adding a preexisting device.
7ceb17e8
N		 7456 * So require mddev->pers and MD_DISK_SYNC.
7457 */
7458 if (mddev->pers) {
7459 mdu_disk_info_t info;
7460 if (copy_from_user(&info, argp, sizeof(info)))
7461 err = -EFAULT;
7462 else if (!(info.state & (1<<MD_DISK_SYNC)))
7463 /* Need to clear read-only for this */
7464 break;
7465 else
7466 err = add_new_disk(mddev, &info);
3adc28d8 7467 goto unlock;
7ceb17e8
N		 7468 }
7469 break;
7470
c02c0aeb
N		 7471 case BLKROSET:
7472 if (get_user(ro, (int __user *)(arg))) {
7473 err = -EFAULT;
3adc28d8 7474 goto unlock;
c02c0aeb
N		 7475 }
7476 err = -EINVAL;
e2218350 7477
c02c0aeb
N		 7478 /* if the bdev is going readonly the value of mddev->ro
7479 * does not matter, no writes are coming
7480 */
7481 if (ro)
3adc28d8 7482 goto unlock;
e2218350 7483
c02c0aeb
N		 7484 /* are we are already prepared for writes? */
7485 if (mddev->ro != 1)
3adc28d8 7486 goto unlock;
e2218350 7487
c02c0aeb
N		 7488 /* transitioning to readauto need only happen for
7489 * arrays that call md_write_start
7490 */
7491 if (mddev->pers) {
7492 err = restart_array(mddev);
7493 if (err == 0) {
7494 mddev->ro = 2;
7495 set_disk_ro(mddev->gendisk, 0);
e2218350 7496 }
c02c0aeb 7497 }
3adc28d8 7498 goto unlock;
1da177e4
LT		 7499 }
7500
7501 /*
7502 * The remaining ioctls are changing the state of the
f91de92e 7503 * superblock, so we do not allow them on read-only arrays.
1da177e4 7504 */
326eb17d 7505 if (mddev->ro && mddev->pers) {
f91de92e
N		 7506 if (mddev->ro == 2) {
7507 mddev->ro = 0;
00bcb4ac 7508 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 7509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N		 7510 /* mddev_unlock will wake thread */
7511 /* If a device failed while we were read-only, we
7512 * need to make sure the metadata is updated now.
7513 */
2953079c 7514 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
f3378b48
N		 7515 mddev_unlock(mddev);
7516 wait_event(mddev->sb_wait,
2953079c
SL		 7517 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7518 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
29f097c4 7519 mddev_lock_nointr(mddev);
f3378b48 7520 }
f91de92e
N		 7521 } else {
7522 err = -EROFS;
3adc28d8 7523 goto unlock;
f91de92e 7524 }
1da177e4
LT		 7525 }
7526
c02c0aeb
N		 7527 switch (cmd) {
7528 case ADD_NEW_DISK:
1da177e4 7529 {
c02c0aeb
N		 7530 mdu_disk_info_t info;
7531 if (copy_from_user(&info, argp, sizeof(info)))
7532 err = -EFAULT;
7533 else
7534 err = add_new_disk(mddev, &info);
3adc28d8 7535 goto unlock;
c02c0aeb 7536 }
1da177e4 7537
1aee41f6
GR		 7538 case CLUSTERED_DISK_NACK:
7539 if (mddev_is_clustered(mddev))
7540 md_cluster_ops->new_disk_ack(mddev, false);
7541 else
7542 err = -EINVAL;
7543 goto unlock;
7544
c02c0aeb
N		 7545 case HOT_ADD_DISK:
7546 err = hot_add_disk(mddev, new_decode_dev(arg));
3adc28d8 7547 goto unlock;
1da177e4 7548
c02c0aeb
N		 7549 case RUN_ARRAY:
7550 err = do_md_run(mddev);
3adc28d8 7551 goto unlock;
1da177e4 7552
c02c0aeb
N		 7553 case SET_BITMAP_FILE:
7554 err = set_bitmap_file(mddev, (int)arg);
3adc28d8 7555 goto unlock;
32a7627c 7556
c02c0aeb
N		 7557 default:
7558 err = -EINVAL;
3adc28d8 7559 goto unlock;
1da177e4
LT		 7560 }
7561
3adc28d8 7562unlock:
d3374825
N		 7563 if (mddev->hold_active == UNTIL_IOCTL &&
7564 err != -EINVAL)
7565 mddev->hold_active = 0;
1da177e4 7566 mddev_unlock(mddev);
3adc28d8 7567out:
065e519e
N		 7568 if(did_set_md_closing)
7569 clear_bit(MD_CLOSING, &mddev->flags);
1da177e4
LT		 7570 return err;
7571}
aa98aa31
AB		 7572#ifdef CONFIG_COMPAT
7573static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7574 unsigned int cmd, unsigned long arg)
7575{
7576 switch (cmd) {
7577 case HOT_REMOVE_DISK:
7578 case HOT_ADD_DISK:
7579 case SET_DISK_FAULTY:
7580 case SET_BITMAP_FILE:
7581 /* These take in integer arg, do not convert */
7582 break;
7583 default:
7584 arg = (unsigned long)compat_ptr(arg);
7585 break;
7586 }
7587
7588 return md_ioctl(bdev, mode, cmd, arg);
7589}
7590#endif /* CONFIG_COMPAT */
1da177e4 7591
a39907fa 7592static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 7593{
7594 /*
7595 * Succeed if we can lock the mddev, which confirms that
7596 * it isn't being stopped right now.
7597 */
fd01b88c 7598 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 7599 int err;
7600
0c098220
YL		 7601 if (!mddev)
7602 return -ENODEV;
7603
d3374825
N		 7604 if (mddev->gendisk != bdev->bd_disk) {
7605 /* we are racing with mddev_put which is discarding this
7606 * bd_disk.
7607 */
7608 mddev_put(mddev);
7609 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 7610 flush_workqueue(md_misc_wq);
d3374825
N		 7611 /* Then retry the open from the top */
7612 return -ERESTARTSYS;
7613 }
7614 BUG_ON(mddev != bdev->bd_disk->private_data);
7615
c8c00a69 7616 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 7617 goto out;
7618
af8d8e6f
GJ		 7619 if (test_bit(MD_CLOSING, &mddev->flags)) {
7620 mutex_unlock(&mddev->open_mutex);
e2342ca8
N		 7621 err = -ENODEV;
7622 goto out;
af8d8e6f
GJ		 7623 }
7624
1da177e4 7625 err = 0;
f2ea68cf 7626 atomic_inc(&mddev->openers);
c8c00a69 7627 mutex_unlock(&mddev->open_mutex);
1da177e4 7628
f0b4f7e2 7629 check_disk_change(bdev);
1da177e4 7630 out:
e2342ca8
N		 7631 if (err)
7632 mddev_put(mddev);
1da177e4
LT		 7633 return err;
7634}
7635
db2a144b 7636static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 7637{
f72ffdd6 7638 struct mddev *mddev = disk->private_data;
1da177e4 7639
52e5f9d1 7640 BUG_ON(!mddev);
f2ea68cf 7641 atomic_dec(&mddev->openers);
1da177e4 7642 mddev_put(mddev);
1da177e4 7643}
f0b4f7e2
N		 7644
7645static int md_media_changed(struct gendisk *disk)
7646{
fd01b88c 7647 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 7648
7649 return mddev->changed;
7650}
7651
7652static int md_revalidate(struct gendisk *disk)
7653{
fd01b88c 7654 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 7655
7656 mddev->changed = 0;
7657 return 0;
7658}
83d5cde4 7659static const struct block_device_operations md_fops =
1da177e4
LT		 7660{
7661 .owner = THIS_MODULE,
a39907fa
AV		 7662 .open = md_open,
7663 .release = md_release,
b492b852 7664 .ioctl = md_ioctl,
aa98aa31
AB		 7665#ifdef CONFIG_COMPAT
7666 .compat_ioctl = md_compat_ioctl,
7667#endif
a885c8c4 7668 .getgeo = md_getgeo,
f0b4f7e2
N		 7669 .media_changed = md_media_changed,
7670 .revalidate_disk= md_revalidate,
1da177e4
LT		 7671};
7672
f72ffdd6 7673static int md_thread(void *arg)
1da177e4 7674{
2b8bf345 7675 struct md_thread *thread = arg;
1da177e4 7676
1da177e4
LT		 7677 /*
7678 * md_thread is a 'system-thread', it's priority should be very
7679 * high. We avoid resource deadlocks individually in each
7680 * raid personality. (RAID5 does preallocation) We also use RR and
7681 * the very same RT priority as kswapd, thus we will never get
7682 * into a priority inversion deadlock.
7683 *
7684 * we definitely have to have equal or higher priority than
7685 * bdflush, otherwise bdflush will deadlock if there are too
7686 * many dirty RAID5 blocks.
7687 */
1da177e4 7688
6985c43f 7689 allow_signal(SIGKILL);
a6fb0934 7690 while (!kthread_should_stop()) {
1da177e4 7691
93588e22
N		 7692 /* We need to wait INTERRUPTIBLE so that
7693 * we don't add to the load-average.
7694 * That means we need to be sure no signals are
7695 * pending
7696 */
7697 if (signal_pending(current))
7698 flush_signals(current);
7699
7700 wait_event_interruptible_timeout
7701 (thread->wqueue,
7702 test_bit(THREAD_WAKEUP, &thread->flags)
ce1ccd07 7703 || kthread_should_stop() || kthread_should_park(),
93588e22 7704 thread->timeout);
1da177e4 7705
6c987910 7706 clear_bit(THREAD_WAKEUP, &thread->flags);
ce1ccd07
SL		 7707 if (kthread_should_park())
7708 kthread_parkme();
6c987910 7709 if (!kthread_should_stop())
4ed8731d 7710 thread->run(thread);
1da177e4 7711 }
a6fb0934 7712
1da177e4
LT		 7713 return 0;
7714}
7715
2b8bf345 7716void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 7717{
7718 if (thread) {
36a4e1fe 7719 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
d1d90147
GJ		 7720 set_bit(THREAD_WAKEUP, &thread->flags);
7721 wake_up(&thread->wqueue);
1da177e4
LT		 7722 }
7723}
6c144d31 7724EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 7725
4ed8731d
SL		 7726struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7727 struct mddev *mddev, const char *name)
1da177e4 7728{
2b8bf345 7729 struct md_thread *thread;
1da177e4 7730
2b8bf345 7731 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 7732 if (!thread)
7733 return NULL;
7734
1da177e4
LT		 7735 init_waitqueue_head(&thread->wqueue);
7736
1da177e4
LT		 7737 thread->run = run;
7738 thread->mddev = mddev;
32a7627c 7739 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 7740 thread->tsk = kthread_run(md_thread, thread,
7741 "%s_%s",
7742 mdname(thread->mddev),
0232605d 7743 name);
a6fb0934 7744 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 7745 kfree(thread);
7746 return NULL;
7747 }
1da177e4
LT		 7748 return thread;
7749}
6c144d31 7750EXPORT_SYMBOL(md_register_thread);
1da177e4 7751
2b8bf345 7752void md_unregister_thread(struct md_thread **threadp)
1da177e4 7753{
2b8bf345 7754 struct md_thread *thread = *threadp;
e0cf8f04
N		 7755 if (!thread)
7756 return;
36a4e1fe 7757 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 7758 /* Locking ensures that mddev_unlock does not wake_up a
7759 * non-existent thread
7760 */
7761 spin_lock(&pers_lock);
7762 *threadp = NULL;
7763 spin_unlock(&pers_lock);
a6fb0934
N		 7764
7765 kthread_stop(thread->tsk);
1da177e4
LT		 7766 kfree(thread);
7767}
6c144d31 7768EXPORT_SYMBOL(md_unregister_thread);
1da177e4 7769
fd01b88c 7770void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 7771{
b2d444d7 7772 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 7773 return;
6bfe0b49 7774
de393cde 7775 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 7776 return;
7777 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 7778 if (mddev->degraded)
7779 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 7780 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 7781 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7782 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7783 md_wakeup_thread(mddev->thread);
768a418d 7784 if (mddev->event_work.func)
e804ac78 7785 queue_work(md_misc_wq, &mddev->event_work);
bb9ef716 7786 md_new_event(mddev);
1da177e4 7787}
6c144d31 7788EXPORT_SYMBOL(md_error);
1da177e4
LT		 7789
7790/* seq_file implementation /proc/mdstat */
7791
7792static void status_unused(struct seq_file *seq)
7793{
7794 int i = 0;
3cb03002 7795 struct md_rdev *rdev;
1da177e4
LT		 7796
7797 seq_printf(seq, "unused devices: ");
7798
159ec1fc 7799 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 7800 char b[BDEVNAME_SIZE];
7801 i++;
7802 seq_printf(seq, "%s ",
7803 bdevname(rdev->bdev,b));
7804 }
7805 if (!i)
7806 seq_printf(seq, "<none>");
7807
7808 seq_printf(seq, "\n");
7809}
7810
f7851be7 7811static int status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 7812{
dd71cf6b 7813 sector_t max_sectors, resync, res;
9642fa73
MT		 7814 unsigned long dt, db = 0;
7815 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7816 int scale, recovery_active;
4588b42e 7817 unsigned int per_milli;
1da177e4 7818
c804cdec
N		 7819 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7820 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 7821 max_sectors = mddev->resync_max_sectors;
1da177e4 7822 else
dd71cf6b 7823 max_sectors = mddev->dev_sectors;
1da177e4 7824
f7851be7
N		 7825 resync = mddev->curr_resync;
7826 if (resync <= 3) {
7827 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7828 /* Still cleaning up */
7829 resync = max_sectors;
d2e2ec82
ND		 7830 } else if (resync > max_sectors)
7831 resync = max_sectors;
7832 else
f7851be7
N		 7833 resync -= atomic_read(&mddev->recovery_active);
7834
7835 if (resync == 0) {
0357ba27
GJ		 7836 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
7837 struct md_rdev *rdev;
7838
7839 rdev_for_each(rdev, mddev)
7840 if (rdev->raid_disk >= 0 &&
7841 !test_bit(Faulty, &rdev->flags) &&
7842 rdev->recovery_offset != MaxSector &&
7843 rdev->recovery_offset) {
7844 seq_printf(seq, "\trecover=REMOTE");
7845 return 1;
7846 }
7847 if (mddev->reshape_position != MaxSector)
7848 seq_printf(seq, "\treshape=REMOTE");
7849 else
7850 seq_printf(seq, "\tresync=REMOTE");
7851 return 1;
7852 }
f7851be7
N		 7853 if (mddev->recovery_cp < MaxSector) {
7854 seq_printf(seq, "\tresync=PENDING");
7855 return 1;
7856 }
7857 return 0;
7858 }
7859 if (resync < 3) {
7860 seq_printf(seq, "\tresync=DELAYED");
7861 return 1;
7862 }
7863
403df478 7864 WARN_ON(max_sectors == 0);
4588b42e 7865 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 7866 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 7867 * u32, as those are the requirements for sector_div.
7868 * Thus 'scale' must be at least 10
7869 */
7870 scale = 10;
7871 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 7872 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 7873 scale++;
7874 }
7875 res = (resync>>scale)*1000;
dd71cf6b 7876 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 7877
7878 per_milli = res;
1da177e4 7879 {
4588b42e 7880 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 7881 seq_printf(seq, "[");
7882 for (i = 0; i < x; i++)
7883 seq_printf(seq, "=");
7884 seq_printf(seq, ">");
7885 for (i = 0; i < y; i++)
7886 seq_printf(seq, ".");
7887 seq_printf(seq, "] ");
7888 }
4588b42e 7889 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 7890 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7891 "reshape" :
61df9d91
N		 7892 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7893 "check" :
7894 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7895 "resync" : "recovery"))),
7896 per_milli/10, per_milli % 10,
dd71cf6b
N		 7897 (unsigned long long) resync/2,
7898 (unsigned long long) max_sectors/2);
1da177e4
LT		 7899
7900 /*
1da177e4
LT		 7901 * dt: time from mark until now
7902 * db: blocks written from mark until now
7903 * rt: remaining time
dd71cf6b 7904 *
9642fa73
MT		 7905 * rt is a sector_t, which is always 64bit now. We are keeping
7906 * the original algorithm, but it is not really necessary.
7907 *
7908 * Original algorithm:
7909 * So we divide before multiply in case it is 32bit and close
7910 * to the limit.
7911 * We scale the divisor (db) by 32 to avoid losing precision
7912 * near the end of resync when the number of remaining sectors
7913 * is close to 'db'.
7914 * We then divide rt by 32 after multiplying by db to compensate.
7915 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 7916 */
7917 dt = ((jiffies - mddev->resync_mark) / HZ);
7918 if (!dt) dt++;
9642fa73
MT		 7919
7920 curr_mark_cnt = mddev->curr_mark_cnt;
7921 recovery_active = atomic_read(&mddev->recovery_active);
7922 resync_mark_cnt = mddev->resync_mark_cnt;
7923
7924 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
7925 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
1da177e4 7926
dd71cf6b 7927 rt = max_sectors - resync; /* number of remaining sectors */
9642fa73 7928 rt = div64_u64(rt, db/32+1);
dd71cf6b
N		 7929 rt *= dt;
7930 rt >>= 5;
7931
7932 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7933 ((unsigned long)rt % 60)/6);
1da177e4 7934
ff4e8d9a 7935 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
f7851be7 7936 return 1;
1da177e4
LT		 7937}
7938
7939static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7940{
7941 struct list_head *tmp;
7942 loff_t l = *pos;
fd01b88c 7943 struct mddev *mddev;
1da177e4
LT		 7944
7945 if (l >= 0x10000)
7946 return NULL;
7947 if (!l--)
7948 /* header */
7949 return (void*)1;
7950
7951 spin_lock(&all_mddevs_lock);
7952 list_for_each(tmp,&all_mddevs)
7953 if (!l--) {
fd01b88c 7954 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 7955 mddev_get(mddev);
7956 spin_unlock(&all_mddevs_lock);
7957 return mddev;
7958 }
7959 spin_unlock(&all_mddevs_lock);
7960 if (!l--)
7961 return (void*)2;/* tail */
7962 return NULL;
7963}
7964
7965static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7966{
7967 struct list_head *tmp;
fd01b88c 7968 struct mddev *next_mddev, *mddev = v;
f72ffdd6 7969
1da177e4
LT		 7970 ++*pos;
7971 if (v == (void*)2)
7972 return NULL;
7973
7974 spin_lock(&all_mddevs_lock);
7975 if (v == (void*)1)
7976 tmp = all_mddevs.next;
7977 else
7978 tmp = mddev->all_mddevs.next;
7979 if (tmp != &all_mddevs)
fd01b88c 7980 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 7981 else {
7982 next_mddev = (void*)2;
7983 *pos = 0x10000;
f72ffdd6 7984 }
1da177e4
LT		 7985 spin_unlock(&all_mddevs_lock);
7986
7987 if (v != (void*)1)
7988 mddev_put(mddev);
7989 return next_mddev;
7990
7991}
7992
7993static void md_seq_stop(struct seq_file *seq, void *v)
7994{
fd01b88c 7995 struct mddev *mddev = v;
1da177e4
LT		 7996
7997 if (mddev && v != (void*)1 && v != (void*)2)
7998 mddev_put(mddev);
7999}
8000
8001static int md_seq_show(struct seq_file *seq, void *v)
8002{
fd01b88c 8003 struct mddev *mddev = v;
dd8ac336 8004 sector_t sectors;
3cb03002 8005 struct md_rdev *rdev;
1da177e4
LT		 8006
8007 if (v == (void*)1) {
84fc4b56 8008 struct md_personality *pers;
1da177e4
LT		 8009 seq_printf(seq, "Personalities : ");
8010 spin_lock(&pers_lock);
2604b703
N		 8011 list_for_each_entry(pers, &pers_list, list)
8012 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 8013
8014 spin_unlock(&pers_lock);
8015 seq_printf(seq, "\n");
f1514638 8016 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 8017 return 0;
8018 }
8019 if (v == (void*)2) {
8020 status_unused(seq);
8021 return 0;
8022 }
8023
36d091f4 8024 spin_lock(&mddev->lock);
1da177e4
LT		 8025 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8026 seq_printf(seq, "%s : %sactive", mdname(mddev),
8027 mddev->pers ? "" : "in");
8028 if (mddev->pers) {
f91de92e 8029 if (mddev->ro==1)
1da177e4 8030 seq_printf(seq, " (read-only)");
f91de92e 8031 if (mddev->ro==2)
52720ae7 8032 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 8033 seq_printf(seq, " %s", mddev->pers->name);
8034 }
8035
dd8ac336 8036 sectors = 0;
f97fcad3
N		 8037 rcu_read_lock();
8038 rdev_for_each_rcu(rdev, mddev) {
1da177e4
LT		 8039 char b[BDEVNAME_SIZE];
8040 seq_printf(seq, " %s[%d]",
8041 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 8042 if (test_bit(WriteMostly, &rdev->flags))
8043 seq_printf(seq, "(W)");
9efdca16
SL		 8044 if (test_bit(Journal, &rdev->flags))
8045 seq_printf(seq, "(J)");
b2d444d7 8046 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8047 seq_printf(seq, "(F)");
8048 continue;
2d78f8c4
N		 8049 }
8050 if (rdev->raid_disk < 0)
b325a32e 8051 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 8052 if (test_bit(Replacement, &rdev->flags))
8053 seq_printf(seq, "(R)");
dd8ac336 8054 sectors += rdev->sectors;
1da177e4 8055 }
f97fcad3 8056 rcu_read_unlock();
1da177e4
LT		 8057
8058 if (!list_empty(&mddev->disks)) {
8059 if (mddev->pers)
8060 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 8061 (unsigned long long)
8062 mddev->array_sectors / 2);
1da177e4
LT		 8063 else
8064 seq_printf(seq, "\n %llu blocks",
dd8ac336 8065 (unsigned long long)sectors / 2);
1da177e4 8066 }
1cd6bf19
N		 8067 if (mddev->persistent) {
8068 if (mddev->major_version != 0 ||
8069 mddev->minor_version != 90) {
8070 seq_printf(seq," super %d.%d",
8071 mddev->major_version,
8072 mddev->minor_version);
8073 }
e691063a
N		 8074 } else if (mddev->external)
8075 seq_printf(seq, " super external:%s",
8076 mddev->metadata_type);
8077 else
1cd6bf19 8078 seq_printf(seq, " super non-persistent");
1da177e4
LT		 8079
8080 if (mddev->pers) {
d710e138 8081 mddev->pers->status(seq, mddev);
f72ffdd6 8082 seq_printf(seq, "\n ");
8e1b39d6 8083 if (mddev->pers->sync_request) {
f7851be7 8084 if (status_resync(seq, mddev))
8e1b39d6 8085 seq_printf(seq, "\n ");
8e1b39d6 8086 }
32a7627c
N		 8087 } else
8088 seq_printf(seq, "\n ");
8089
e64e4018 8090 md_bitmap_status(seq, mddev->bitmap);
1da177e4
LT		 8091
8092 seq_printf(seq, "\n");
8093 }
36d091f4 8094 spin_unlock(&mddev->lock);
f72ffdd6 8095
1da177e4
LT		 8096 return 0;
8097}
8098
110518bc 8099static const struct seq_operations md_seq_ops = {
1da177e4
LT		 8100 .start = md_seq_start,
8101 .next = md_seq_next,
8102 .stop = md_seq_stop,
8103 .show = md_seq_show,
8104};
8105
8106static int md_seq_open(struct inode *inode, struct file *file)
8107{
f1514638 8108 struct seq_file *seq;
1da177e4
LT		 8109 int error;
8110
8111 error = seq_open(file, &md_seq_ops);
d7603b7e 8112 if (error)
f1514638
KS		 8113 return error;
8114
8115 seq = file->private_data;
8116 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 8117 return error;
8118}
8119
e2f23b60 8120static int md_unloading;
afc9a42b 8121static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
d7603b7e 8122{
f1514638 8123 struct seq_file *seq = filp->private_data;
afc9a42b 8124 __poll_t mask;
d7603b7e 8125
e2f23b60 8126 if (md_unloading)
a9a08845 8127 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
d7603b7e
N		 8128 poll_wait(filp, &md_event_waiters, wait);
8129
8130 /* always allow read */
a9a08845 8131 mask = EPOLLIN | EPOLLRDNORM;
d7603b7e 8132
f1514638 8133 if (seq->poll_event != atomic_read(&md_event_count))
a9a08845 8134 mask |= EPOLLERR | EPOLLPRI;
d7603b7e
N		 8135 return mask;
8136}
8137
fa027c2a 8138static const struct file_operations md_seq_fops = {
e24650c2 8139 .owner = THIS_MODULE,
1da177e4
LT		 8140 .open = md_seq_open,
8141 .read = seq_read,
8142 .llseek = seq_lseek,
26e13043 8143 .release = seq_release,
d7603b7e 8144 .poll = mdstat_poll,
1da177e4
LT		 8145};
8146
84fc4b56 8147int register_md_personality(struct md_personality *p)
1da177e4 8148{
9d48739e
N		 8149 pr_debug("md: %s personality registered for level %d\n",
8150 p->name, p->level);
1da177e4 8151 spin_lock(&pers_lock);
2604b703 8152 list_add_tail(&p->list, &pers_list);
1da177e4
LT		 8153 spin_unlock(&pers_lock);
8154 return 0;
8155}
6c144d31 8156EXPORT_SYMBOL(register_md_personality);
1da177e4 8157
84fc4b56 8158int unregister_md_personality(struct md_personality *p)
1da177e4 8159{
9d48739e 8160 pr_debug("md: %s personality unregistered\n", p->name);
1da177e4 8161 spin_lock(&pers_lock);
2604b703 8162 list_del_init(&p->list);
1da177e4
LT		 8163 spin_unlock(&pers_lock);
8164 return 0;
8165}
6c144d31 8166EXPORT_SYMBOL(unregister_md_personality);
1da177e4 8167
6022e75b
N		 8168int register_md_cluster_operations(struct md_cluster_operations *ops,
8169 struct module *module)
edb39c9d 8170{
6022e75b 8171 int ret = 0;
edb39c9d 8172 spin_lock(&pers_lock);
6022e75b
N		 8173 if (md_cluster_ops != NULL)
8174 ret = -EALREADY;
8175 else {
8176 md_cluster_ops = ops;
8177 md_cluster_mod = module;
8178 }
edb39c9d 8179 spin_unlock(&pers_lock);
6022e75b 8180 return ret;
edb39c9d
GR		 8181}
8182EXPORT_SYMBOL(register_md_cluster_operations);
8183
8184int unregister_md_cluster_operations(void)
8185{
8186 spin_lock(&pers_lock);
8187 md_cluster_ops = NULL;
8188 spin_unlock(&pers_lock);
8189 return 0;
8190}
8191EXPORT_SYMBOL(unregister_md_cluster_operations);
8192
8193int md_setup_cluster(struct mddev *mddev, int nodes)
8194{
47a7b0d8
GJ		 8195 if (!md_cluster_ops)
8196 request_module("md-cluster");
edb39c9d 8197 spin_lock(&pers_lock);
47a7b0d8 8198 /* ensure module won't be unloaded */
edb39c9d 8199 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
9d48739e 8200 pr_warn("can't find md-cluster module or get it's reference.\n");
edb39c9d
GR		 8201 spin_unlock(&pers_lock);
8202 return -ENOENT;
8203 }
8204 spin_unlock(&pers_lock);
8205
cf921cc1 8206 return md_cluster_ops->join(mddev, nodes);
edb39c9d
GR		 8207}
8208
8209void md_cluster_stop(struct mddev *mddev)
8210{
c4ce867f
GR		 8211 if (!md_cluster_ops)
8212 return;
edb39c9d
GR		 8213 md_cluster_ops->leave(mddev);
8214 module_put(md_cluster_mod);
8215}
8216
fd01b88c 8217static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 8218{
f72ffdd6 8219 struct md_rdev *rdev;
1da177e4 8220 int idle;
eea1bf38 8221 int curr_events;
1da177e4
LT		 8222
8223 idle = 1;
4b80991c
N		 8224 rcu_read_lock();
8225 rdev_for_each_rcu(rdev, mddev) {
1da177e4 8226 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
59767fbd 8227 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
eea1bf38 8228 atomic_read(&disk->sync_io);
713f6ab1
N		 8229 /* sync IO will cause sync_io to increase before the disk_stats
8230 * as sync_io is counted when a request starts, and
8231 * disk_stats is counted when it completes.
8232 * So resync activity will cause curr_events to be smaller than
8233 * when there was no such activity.
8234 * non-sync IO will cause disk_stat to increase without
8235 * increasing sync_io so curr_events will (eventually)
8236 * be larger than it was before. Once it becomes
8237 * substantially larger, the test below will cause
8238 * the array to appear non-idle, and resync will slow
8239 * down.
8240 * If there is a lot of outstanding resync activity when
8241 * we set last_event to curr_events, then all that activity
8242 * completing might cause the array to appear non-idle
8243 * and resync will be slowed down even though there might
8244 * not have been non-resync activity. This will only
8245 * happen once though. 'last_events' will soon reflect
8246 * the state where there is little or no outstanding
8247 * resync requests, and further resync activity will
8248 * always make curr_events less than last_events.
c0e48521 8249 *
1da177e4 8250 */
eea1bf38 8251 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 8252 rdev->last_events = curr_events;
8253 idle = 0;
8254 }
8255 }
4b80991c 8256 rcu_read_unlock();
1da177e4
LT		 8257 return idle;
8258}
8259
fd01b88c 8260void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 8261{
8262 /* another "blocks" (512byte) blocks have been synced */
8263 atomic_sub(blocks, &mddev->recovery_active);
8264 wake_up(&mddev->recovery_wait);
8265 if (!ok) {
dfc70645 8266 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 8267 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT		 8268 md_wakeup_thread(mddev->thread);
8269 // stop recovery, signal do_sync ....
8270 }
8271}
6c144d31 8272EXPORT_SYMBOL(md_done_sync);
1da177e4 8273
06d91a5f
N		 8274/* md_write_start(mddev, bi)
8275 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 8276 * in superblock) before writing, schedule a superblock update
8277 * and wait for it to complete.
cc27b0c7
N		 8278 * A return value of 'false' means that the write wasn't recorded
8279 * and cannot proceed as the array is being suspend.
06d91a5f 8280 */
cc27b0c7 8281bool md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 8282{
0fd62b86 8283 int did_change = 0;
4b6c1060 8284
06d91a5f 8285 if (bio_data_dir(bi) != WRITE)
cc27b0c7 8286 return true;
06d91a5f 8287
f91de92e
N		 8288 BUG_ON(mddev->ro == 1);
8289 if (mddev->ro == 2) {
8290 /* need to switch to read/write */
8291 mddev->ro = 0;
8292 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8293 md_wakeup_thread(mddev->thread);
25156198 8294 md_wakeup_thread(mddev->sync_thread);
0fd62b86 8295 did_change = 1;
f91de92e 8296 }
4ad23a97
N		 8297 rcu_read_lock();
8298 percpu_ref_get(&mddev->writes_pending);
55cc39f3 8299 smp_mb(); /* Match smp_mb in set_in_sync() */
31a59e34
N		 8300 if (mddev->safemode == 1)
8301 mddev->safemode = 0;
4ad23a97 8302 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
81fe48e9 8303 if (mddev->in_sync || mddev->sync_checkers) {
85572d7c 8304 spin_lock(&mddev->lock);
3d310eb7
N		 8305 if (mddev->in_sync) {
8306 mddev->in_sync = 0;
2953079c
SL		 8307 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8308 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
3d310eb7 8309 md_wakeup_thread(mddev->thread);
0fd62b86 8310 did_change = 1;
3d310eb7 8311 }
85572d7c 8312 spin_unlock(&mddev->lock);
06d91a5f 8313 }
4ad23a97 8314 rcu_read_unlock();
0fd62b86 8315 if (did_change)
00bcb4ac 8316 sysfs_notify_dirent_safe(mddev->sysfs_state);
4b6c1060
HM		 8317 if (!mddev->has_superblocks)
8318 return true;
09a44cc1 8319 wait_event(mddev->sb_wait,
d47c8ad2
N		 8320 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8321 mddev->suspended);
cc27b0c7
N		 8322 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8323 percpu_ref_put(&mddev->writes_pending);
8324 return false;
8325 }
8326 return true;
1da177e4 8327}
6c144d31 8328EXPORT_SYMBOL(md_write_start);
1da177e4 8329
49728050
N		 8330/* md_write_inc can only be called when md_write_start() has
8331 * already been called at least once of the current request.
8332 * It increments the counter and is useful when a single request
8333 * is split into several parts. Each part causes an increment and
8334 * so needs a matching md_write_end().
8335 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8336 * a spinlocked region.
8337 */
8338void md_write_inc(struct mddev *mddev, struct bio *bi)
8339{
8340 if (bio_data_dir(bi) != WRITE)
8341 return;
8342 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
4ad23a97 8343 percpu_ref_get(&mddev->writes_pending);
49728050
N		 8344}
8345EXPORT_SYMBOL(md_write_inc);
8346
fd01b88c 8347void md_write_end(struct mddev *mddev)
1da177e4 8348{
4ad23a97
N		 8349 percpu_ref_put(&mddev->writes_pending);
8350
8351 if (mddev->safemode == 2)
8352 md_wakeup_thread(mddev->thread);
8353 else if (mddev->safemode_delay)
8354 /* The roundup() ensures this only performs locking once
8355 * every ->safemode_delay jiffies
8356 */
8357 mod_timer(&mddev->safemode_timer,
8358 roundup(jiffies, mddev->safemode_delay) +
8359 mddev->safemode_delay);
1da177e4 8360}
4ad23a97 8361
6c144d31 8362EXPORT_SYMBOL(md_write_end);
1da177e4 8363
2a2275d6
N		 8364/* md_allow_write(mddev)
8365 * Calling this ensures that the array is marked 'active' so that writes
8366 * may proceed without blocking. It is important to call this before
8367 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8368 * Must be called with mddev_lock held.
8369 */
2214c260 8370void md_allow_write(struct mddev *mddev)
2a2275d6
N		 8371{
8372 if (!mddev->pers)
2214c260 8373 return;
2a2275d6 8374 if (mddev->ro)
2214c260 8375 return;
1a0fd497 8376 if (!mddev->pers->sync_request)
2214c260 8377 return;
2a2275d6 8378
85572d7c 8379 spin_lock(&mddev->lock);
2a2275d6
N		 8380 if (mddev->in_sync) {
8381 mddev->in_sync = 0;
2953079c
SL		 8382 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8383 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2a2275d6
N		 8384 if (mddev->safemode_delay &&
8385 mddev->safemode == 0)
8386 mddev->safemode = 1;
85572d7c 8387 spin_unlock(&mddev->lock);
2a2275d6 8388 md_update_sb(mddev, 0);
00bcb4ac 8389 sysfs_notify_dirent_safe(mddev->sysfs_state);
2214c260
AP		 8390 /* wait for the dirty state to be recorded in the metadata */
8391 wait_event(mddev->sb_wait,
2214c260 8392 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
2a2275d6 8393 } else
85572d7c 8394 spin_unlock(&mddev->lock);
2a2275d6
N		 8395}
8396EXPORT_SYMBOL_GPL(md_allow_write);
8397
1da177e4
LT		 8398#define SYNC_MARKS 10
8399#define SYNC_MARK_STEP (3*HZ)
54f89341 8400#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 8401void md_do_sync(struct md_thread *thread)
1da177e4 8402{
4ed8731d 8403 struct mddev *mddev = thread->mddev;
fd01b88c 8404 struct mddev *mddev2;
e5b521ee 8405 unsigned int currspeed = 0, window;
ac7e50a3 8406 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 8407 unsigned long mark[SYNC_MARKS];
54f89341 8408 unsigned long update_time;
1da177e4
LT		 8409 sector_t mark_cnt[SYNC_MARKS];
8410 int last_mark,m;
8411 struct list_head *tmp;
8412 sector_t last_check;
57afd89f 8413 int skipped = 0;
3cb03002 8414 struct md_rdev *rdev;
c4a39551 8415 char *desc, *action = NULL;
7c2c57c9 8416 struct blk_plug plug;
41a9a0dc 8417 int ret;
1da177e4
LT		 8418
8419 /* just incase thread restarts... */
d5d885fd
SL		 8420 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8421 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
1da177e4 8422 return;
3991b31e
N		 8423 if (mddev->ro) {/* never try to sync a read-only array */
8424 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 8425 return;
3991b31e 8426 }
1da177e4 8427
41a9a0dc
GJ		 8428 if (mddev_is_clustered(mddev)) {
8429 ret = md_cluster_ops->resync_start(mddev);
8430 if (ret)
8431 goto skip;
8432
bb8bf15b 8433 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
41a9a0dc
GJ		 8434 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8435 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8436 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8437 && ((unsigned long long)mddev->curr_resync_completed
8438 < (unsigned long long)mddev->resync_max_sectors))
8439 goto skip;
8440 }
8441
61df9d91 8442 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 8443 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 8444 desc = "data-check";
c4a39551
JB		 8445 action = "check";
8446 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 8447 desc = "requested-resync";
c4a39551
JB		 8448 action = "repair";
8449 } else
61df9d91
N		 8450 desc = "resync";
8451 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8452 desc = "reshape";
8453 else
8454 desc = "recovery";
8455
c4a39551
JB		 8456 mddev->last_sync_action = action ?: desc;
8457
1da177e4
LT		 8458 /* we overload curr_resync somewhat here.
8459 * 0 == not engaged in resync at all
8460 * 2 == checking that there is no conflict with another sync
8461 * 1 == like 2, but have yielded to allow conflicting resync to
e5b521ee 8462 * commence
1da177e4
LT		 8463 * other == active in resync - this many blocks
8464 *
8465 * Before starting a resync we must have set curr_resync to
8466 * 2, and then checked that every "conflicting" array has curr_resync
8467 * less than ours. When we find one that is the same or higher
8468 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8469 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8470 * This will mean we have to start checking from the beginning again.
8471 *
8472 */
8473
8474 do {
c622ca54 8475 int mddev2_minor = -1;
1da177e4
LT		 8476 mddev->curr_resync = 2;
8477
8478 try_again:
404e4b43 8479 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 8480 goto skip;
29ac4aa3 8481 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 8482 if (mddev2 == mddev)
8483 continue;
90b08710
BS		 8484 if (!mddev->parallel_resync
8485 && mddev2->curr_resync
8486 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 8487 DEFINE_WAIT(wq);
8488 if (mddev < mddev2 && mddev->curr_resync == 2) {
8489 /* arbitrarily yield */
8490 mddev->curr_resync = 1;
8491 wake_up(&resync_wait);
8492 }
8493 if (mddev > mddev2 && mddev->curr_resync == 1)
8494 /* no need to wait here, we can wait the next
8495 * time 'round when curr_resync == 2
8496 */
8497 continue;
9744197c
N		 8498 /* We need to wait 'interruptible' so as not to
8499 * contribute to the load average, and not to
8500 * be caught by 'softlockup'
8501 */
8502 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 8503 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 8504 mddev2->curr_resync >= mddev->curr_resync) {
c622ca54
AP		 8505 if (mddev2_minor != mddev2->md_minor) {
8506 mddev2_minor = mddev2->md_minor;
9d48739e
N		 8507 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8508 desc, mdname(mddev),
8509 mdname(mddev2));
c622ca54 8510 }
1da177e4 8511 mddev_put(mddev2);
9744197c
N		 8512 if (signal_pending(current))
8513 flush_signals(current);
1da177e4
LT		 8514 schedule();
8515 finish_wait(&resync_wait, &wq);
8516 goto try_again;
8517 }
8518 finish_wait(&resync_wait, &wq);
8519 }
8520 }
8521 } while (mddev->curr_resync < 2);
8522
5fd6c1dc 8523 j = 0;
9d88883e 8524 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 8525 /* resync follows the size requested by the personality,
57afd89f 8526 * which defaults to physical size, but can be virtual size
1da177e4
LT		 8527 */
8528 max_sectors = mddev->resync_max_sectors;
7f7583d4 8529 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 8530 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 8531 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8532 j = mddev->resync_min;
8533 else if (!mddev->bitmap)
5fd6c1dc 8534 j = mddev->recovery_cp;
5e96ee65 8535
cb9ee154 8536 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
c804cdec 8537 max_sectors = mddev->resync_max_sectors;
cb9ee154
GJ		 8538 /*
8539 * If the original node aborts reshaping then we continue the
8540 * reshaping, so set j again to avoid restart reshape from the
8541 * first beginning
8542 */
8543 if (mddev_is_clustered(mddev) &&
8544 mddev->reshape_position != MaxSector)
8545 j = mddev->reshape_position;
8546 } else {
1da177e4 8547 /* recovery follows the physical size of devices */
58c0fed4 8548 max_sectors = mddev->dev_sectors;
5fd6c1dc 8549 j = MaxSector;
4e59ca7d 8550 rcu_read_lock();
dafb20fa 8551 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 8552 if (rdev->raid_disk >= 0 &&
f2076e7d 8553 !test_bit(Journal, &rdev->flags) &&
5fd6c1dc
N		 8554 !test_bit(Faulty, &rdev->flags) &&
8555 !test_bit(In_sync, &rdev->flags) &&
8556 rdev->recovery_offset < j)
8557 j = rdev->recovery_offset;
4e59ca7d 8558 rcu_read_unlock();
133d4527
N		 8559
8560 /* If there is a bitmap, we need to make sure all
8561 * writes that started before we added a spare
8562 * complete before we start doing a recovery.
8563 * Otherwise the write might complete and (via
8564 * bitmap_endwrite) set a bit in the bitmap after the
8565 * recovery has checked that bit and skipped that
8566 * region.
8567 */
8568 if (mddev->bitmap) {
8569 mddev->pers->quiesce(mddev, 1);
8570 mddev->pers->quiesce(mddev, 0);
8571 }
5fd6c1dc 8572 }
1da177e4 8573
9d48739e
N		 8574 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8575 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8576 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8577 speed_max(mddev), desc);
1da177e4 8578
eea1bf38 8579 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 8580
57afd89f 8581 io_sectors = 0;
1da177e4
LT		 8582 for (m = 0; m < SYNC_MARKS; m++) {
8583 mark[m] = jiffies;
57afd89f 8584 mark_cnt[m] = io_sectors;
1da177e4
LT		 8585 }
8586 last_mark = 0;
8587 mddev->resync_mark = mark[last_mark];
8588 mddev->resync_mark_cnt = mark_cnt[last_mark];
8589
8590 /*
8591 * Tune reconstruction:
8592 */
e5b521ee 8593 window = 32 * (PAGE_SIZE / 512);
9d48739e
N		 8594 pr_debug("md: using %dk window, over a total of %lluk.\n",
8595 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 8596
8597 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 8598 last_check = 0;
8599
8600 if (j>2) {
9d48739e
N		 8601 pr_debug("md: resuming %s of %s from checkpoint.\n",
8602 desc, mdname(mddev));
1da177e4 8603 mddev->curr_resync = j;
72f36d59
N		 8604 } else
8605 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 8606 mddev->curr_resync_completed = j;
72f36d59
N		 8607 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8608 md_new_event(mddev);
54f89341 8609 update_time = jiffies;
1da177e4 8610
7c2c57c9 8611 blk_start_plug(&plug);
1da177e4 8612 while (j < max_sectors) {
57afd89f 8613 sector_t sectors;
1da177e4 8614
57afd89f 8615 skipped = 0;
97e4f42d 8616
7a91ee1f
N		 8617 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8618 ((mddev->curr_resync > mddev->curr_resync_completed &&
8619 (mddev->curr_resync - mddev->curr_resync_completed)
8620 > (max_sectors >> 4)) ||
54f89341 8621 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f 8622 (j - mddev->curr_resync_completed)*2
c5e19d90
N		 8623 >= mddev->resync_max - mddev->curr_resync_completed ||
8624 mddev->curr_resync_completed > mddev->resync_max
7a91ee1f 8625 )) {
97e4f42d 8626 /* time to update curr_resync_completed */
97e4f42d
N		 8627 wait_event(mddev->recovery_wait,
8628 atomic_read(&mddev->recovery_active) == 0);
75d3da43 8629 mddev->curr_resync_completed = j;
35d78c66 8630 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8631 j > mddev->recovery_cp)
8632 mddev->recovery_cp = j;
54f89341 8633 update_time = jiffies;
2953079c 8634 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
acb180b0 8635 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 8636 }
acb180b0 8637
c91abf5a
N		 8638 while (j >= mddev->resync_max &&
8639 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N		 8640 /* As this condition is controlled by user-space,
8641 * we can block indefinitely, so use '_interruptible'
8642 * to avoid triggering warnings.
8643 */
8644 flush_signals(current); /* just in case */
8645 wait_event_interruptible(mddev->recovery_wait,
8646 mddev->resync_max > j
c91abf5a
N		 8647 || test_bit(MD_RECOVERY_INTR,
8648 &mddev->recovery));
e62e58a5 8649 }
acb180b0 8650
c91abf5a
N		 8651 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8652 break;
acb180b0 8653
09314799 8654 sectors = mddev->pers->sync_request(mddev, j, &skipped);
57afd89f 8655 if (sectors == 0) {
dfc70645 8656 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 8657 break;
1da177e4 8658 }
57afd89f
N		 8659
8660 if (!skipped) { /* actual IO requested */
8661 io_sectors += sectors;
8662 atomic_add(sectors, &mddev->recovery_active);
8663 }
8664
e875ecea
N		 8665 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8666 break;
8667
1da177e4 8668 j += sectors;
5ed1df2e
N		 8669 if (j > max_sectors)
8670 /* when skipping, extra large numbers can be returned. */
8671 j = max_sectors;
72f36d59
N		 8672 if (j > 2)
8673 mddev->curr_resync = j;
ff4e8d9a 8674 mddev->curr_mark_cnt = io_sectors;
d7603b7e 8675 if (last_check == 0)
e875ecea 8676 /* this is the earliest that rebuild will be
d7603b7e
N		 8677 * visible in /proc/mdstat
8678 */
8679 md_new_event(mddev);
57afd89f
N		 8680
8681 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 8682 continue;
8683
57afd89f 8684 last_check = io_sectors;
1da177e4
LT		 8685 repeat:
8686 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8687 /* step marks */
8688 int next = (last_mark+1) % SYNC_MARKS;
8689
8690 mddev->resync_mark = mark[next];
8691 mddev->resync_mark_cnt = mark_cnt[next];
8692 mark[next] = jiffies;
57afd89f 8693 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 8694 last_mark = next;
8695 }
8696
c91abf5a
N		 8697 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8698 break;
1da177e4
LT		 8699
8700 /*
8701 * this loop exits only if either when we are slower than
8702 * the 'hard' speed limit, or the system was IO-idle for
8703 * a jiffy.
8704 * the system might be non-idle CPU-wise, but we only care
8705 * about not overloading the IO subsystem. (things like an
8706 * e2fsck being done on the RAID array should execute fast)
8707 */
1da177e4
LT		 8708 cond_resched();
8709
ac7e50a3
XN		 8710 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8711 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 8712 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 8713
88202a0c 8714 if (currspeed > speed_min(mddev)) {
ac8fa419 8715 if (currspeed > speed_max(mddev)) {
c0e48521 8716 msleep(500);
1da177e4
LT		 8717 goto repeat;
8718 }
ac8fa419
N		 8719 if (!is_mddev_idle(mddev, 0)) {
8720 /*
8721 * Give other IO more of a chance.
8722 * The faster the devices, the less we wait.
8723 */
8724 wait_event(mddev->recovery_wait,
8725 !atomic_read(&mddev->recovery_active));
8726 }
1da177e4
LT		 8727 }
8728 }
9d48739e
N		 8729 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8730 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8731 ? "interrupted" : "done");
1da177e4
LT		 8732 /*
8733 * this also signals 'finished resyncing' to md_stop
8734 */
7c2c57c9 8735 blk_finish_plug(&plug);
1da177e4
LT		 8736 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8737
5ed1df2e
N		 8738 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8739 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
1217e1d1 8740 mddev->curr_resync > 3) {
5ed1df2e
N		 8741 mddev->curr_resync_completed = mddev->curr_resync;
8742 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8743 }
09314799 8744 mddev->pers->sync_request(mddev, max_sectors, &skipped);
1da177e4 8745
dfc70645 8746 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
1217e1d1 8747 mddev->curr_resync > 3) {
5fd6c1dc
N		 8748 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8749 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8750 if (mddev->curr_resync >= mddev->recovery_cp) {
9d48739e
N		 8751 pr_debug("md: checkpointing %s of %s.\n",
8752 desc, mdname(mddev));
0a19caab 8753 if (test_bit(MD_RECOVERY_ERROR,
8754 &mddev->recovery))
8755 mddev->recovery_cp =
8756 mddev->curr_resync_completed;
8757 else
8758 mddev->recovery_cp =
8759 mddev->curr_resync;
5fd6c1dc
N		 8760 }
8761 } else
8762 mddev->recovery_cp = MaxSector;
8763 } else {
8764 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8765 mddev->curr_resync = MaxSector;
db0505d3
N		 8766 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8767 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8768 rcu_read_lock();
8769 rdev_for_each_rcu(rdev, mddev)
8770 if (rdev->raid_disk >= 0 &&
8771 mddev->delta_disks >= 0 &&
8772 !test_bit(Journal, &rdev->flags) &&
8773 !test_bit(Faulty, &rdev->flags) &&
8774 !test_bit(In_sync, &rdev->flags) &&
8775 rdev->recovery_offset < mddev->curr_resync)
8776 rdev->recovery_offset = mddev->curr_resync;
8777 rcu_read_unlock();
8778 }
5fd6c1dc 8779 }
1da177e4 8780 }
db91ff55 8781 skip:
bb8bf15b
GJ		 8782 /* set CHANGE_PENDING here since maybe another update is needed,
8783 * so other nodes are informed. It should be harmless for normal
8784 * raid */
2953079c
SL		 8785 set_mask_bits(&mddev->sb_flags, 0,
8786 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
c186b128 8787
8876391e
BC		 8788 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8789 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8790 mddev->delta_disks > 0 &&
8791 mddev->pers->finish_reshape &&
8792 mddev->pers->size &&
8793 mddev->queue) {
8794 mddev_lock_nointr(mddev);
8795 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8796 mddev_unlock(mddev);
aefb2e5f
GJ		 8797 if (!mddev_is_clustered(mddev)) {
8798 set_capacity(mddev->gendisk, mddev->array_sectors);
8799 revalidate_disk(mddev->gendisk);
8800 }
8876391e
BC		 8801 }
8802
23da422b 8803 spin_lock(&mddev->lock);
c07b70ad
N		 8804 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8805 /* We completed so min/max setting can be forgotten if used. */
8806 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8807 mddev->resync_min = 0;
8808 mddev->resync_max = MaxSector;
8809 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8810 mddev->resync_min = mddev->curr_resync_completed;
f7851be7 8811 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 8812 mddev->curr_resync = 0;
23da422b
N		 8813 spin_unlock(&mddev->lock);
8814
1da177e4 8815 wake_up(&resync_wait);
1da177e4 8816 md_wakeup_thread(mddev->thread);
c6207277 8817 return;
1da177e4 8818}
29269553 8819EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 8820
746d3207
N		 8821static int remove_and_add_spares(struct mddev *mddev,
8822 struct md_rdev *this)
b4c4c7b8 8823{
3cb03002 8824 struct md_rdev *rdev;
b4c4c7b8 8825 int spares = 0;
f2a371c5 8826 int removed = 0;
d787be40 8827 bool remove_some = false;
b4c4c7b8 8828
39772f0a
N		 8829 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8830 /* Mustn't remove devices when resync thread is running */
8831 return 0;
8832
d787be40
N		 8833 rdev_for_each(rdev, mddev) {
8834 if ((this == NULL || rdev == this) &&
8835 rdev->raid_disk >= 0 &&
8836 !test_bit(Blocked, &rdev->flags) &&
8837 test_bit(Faulty, &rdev->flags) &&
8838 atomic_read(&rdev->nr_pending)==0) {
8839 /* Faulty non-Blocked devices with nr_pending == 0
8840 * never get nr_pending incremented,
8841 * never get Faulty cleared, and never get Blocked set.
8842 * So we can synchronize_rcu now rather than once per device
8843 */
8844 remove_some = true;
8845 set_bit(RemoveSynchronized, &rdev->flags);
8846 }
8847 }
8848
8849 if (remove_some)
8850 synchronize_rcu();
8851 rdev_for_each(rdev, mddev) {
746d3207
N		 8852 if ((this == NULL || rdev == this) &&
8853 rdev->raid_disk >= 0 &&
6bfe0b49 8854 !test_bit(Blocked, &rdev->flags) &&
d787be40 8855 ((test_bit(RemoveSynchronized, &rdev->flags) ||
f2076e7d
SL		 8856 (!test_bit(In_sync, &rdev->flags) &&
8857 !test_bit(Journal, &rdev->flags))) &&
d787be40 8858 atomic_read(&rdev->nr_pending)==0)) {
b4c4c7b8 8859 if (mddev->pers->hot_remove_disk(
b8321b68 8860 mddev, rdev) == 0) {
36fad858 8861 sysfs_unlink_rdev(mddev, rdev);
011abdc9 8862 rdev->saved_raid_disk = rdev->raid_disk;
b4c4c7b8 8863 rdev->raid_disk = -1;
f2a371c5 8864 removed++;
b4c4c7b8
N		 8865 }
8866 }
d787be40
N		 8867 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8868 clear_bit(RemoveSynchronized, &rdev->flags);
8869 }
8870
90584fc9
JB		 8871 if (removed && mddev->kobj.sd)
8872 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 8873
2910ff17 8874 if (this && removed)
746d3207
N		 8875 goto no_add;
8876
dafb20fa 8877 rdev_for_each(rdev, mddev) {
2910ff17
GR		 8878 if (this && this != rdev)
8879 continue;
dbb64f86
GR		 8880 if (test_bit(Candidate, &rdev->flags))
8881 continue;
7bfec5f3
N		 8882 if (rdev->raid_disk >= 0 &&
8883 !test_bit(In_sync, &rdev->flags) &&
f2076e7d 8884 !test_bit(Journal, &rdev->flags) &&
7bfec5f3
N		 8885 !test_bit(Faulty, &rdev->flags))
8886 spares++;
7ceb17e8
N		 8887 if (rdev->raid_disk >= 0)
8888 continue;
8889 if (test_bit(Faulty, &rdev->flags))
8890 continue;
f6b6ec5c
SL		 8891 if (!test_bit(Journal, &rdev->flags)) {
8892 if (mddev->ro &&
8893 ! (rdev->saved_raid_disk >= 0 &&
8894 !test_bit(Bitmap_sync, &rdev->flags)))
8895 continue;
7ceb17e8 8896
f6b6ec5c
SL		 8897 rdev->recovery_offset = 0;
8898 }
7ceb17e8
N		 8899 if (mddev->pers->
8900 hot_add_disk(mddev, rdev) == 0) {
8901 if (sysfs_link_rdev(mddev, rdev))
8902 /* failure here is OK */;
f6b6ec5c
SL		 8903 if (!test_bit(Journal, &rdev->flags))
8904 spares++;
7ceb17e8 8905 md_new_event(mddev);
2953079c 8906 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
dfc70645 8907 }
b4c4c7b8 8908 }
746d3207 8909no_add:
6dafab6b 8910 if (removed)
2953079c 8911 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
b4c4c7b8
N		 8912 return spares;
8913}
7ebc0be7 8914
ac05f256
N		 8915static void md_start_sync(struct work_struct *ws)
8916{
8917 struct mddev *mddev = container_of(ws, struct mddev, del_work);
c186b128 8918
ac05f256
N		 8919 mddev->sync_thread = md_register_thread(md_do_sync,
8920 mddev,
8921 "resync");
8922 if (!mddev->sync_thread) {
9d48739e
N		 8923 pr_warn("%s: could not start resync thread...\n",
8924 mdname(mddev));
ac05f256
N		 8925 /* leave the spares where they are, it shouldn't hurt */
8926 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8927 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8928 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8929 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8930 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 8931 wake_up(&resync_wait);
ac05f256
N		 8932 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8933 &mddev->recovery))
8934 if (mddev->sysfs_action)
8935 sysfs_notify_dirent_safe(mddev->sysfs_action);
8936 } else
8937 md_wakeup_thread(mddev->sync_thread);
8938 sysfs_notify_dirent_safe(mddev->sysfs_action);
8939 md_new_event(mddev);
8940}
8941
1da177e4
LT		 8942/*
8943 * This routine is regularly called by all per-raid-array threads to
8944 * deal with generic issues like resync and super-block update.
8945 * Raid personalities that don't have a thread (linear/raid0) do not
8946 * need this as they never do any recovery or update the superblock.
8947 *
8948 * It does not do any resync itself, but rather "forks" off other threads
8949 * to do that as needed.
8950 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8951 * "->recovery" and create a thread at ->sync_thread.
dfc70645 8952 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 8953 * and wakeups up this thread which will reap the thread and finish up.
8954 * This thread also removes any faulty devices (with nr_pending == 0).
8955 *
8956 * The overall approach is:
8957 * 1/ if the superblock needs updating, update it.
8958 * 2/ If a recovery thread is running, don't do anything else.
8959 * 3/ If recovery has finished, clean up, possibly marking spares active.
8960 * 4/ If there are any faulty devices, remove them.
8961 * 5/ If array is degraded, try to add spares devices
8962 * 6/ If array has spares or is not in-sync, start a resync thread.
8963 */
fd01b88c 8964void md_check_recovery(struct mddev *mddev)
1da177e4 8965{
059421e0
N		 8966 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8967 /* Write superblock - thread that called mddev_suspend()
8968 * holds reconfig_mutex for us.
8969 */
8970 set_bit(MD_UPDATING_SB, &mddev->flags);
8971 smp_mb__after_atomic();
8972 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8973 md_update_sb(mddev, 0);
8974 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8975 wake_up(&mddev->sb_wait);
8976 }
8977
68866e42
JB		 8978 if (mddev->suspended)
8979 return;
8980
5f40402d 8981 if (mddev->bitmap)
e64e4018 8982 md_bitmap_daemon_work(mddev);
1da177e4 8983
fca4d848 8984 if (signal_pending(current)) {
31a59e34 8985 if (mddev->pers->sync_request && !mddev->external) {
9d48739e
N		 8986 pr_debug("md: %s in immediate safe mode\n",
8987 mdname(mddev));
fca4d848
N		 8988 mddev->safemode = 2;
8989 }
8990 flush_signals(current);
8991 }
8992
c89a8eee
N		 8993 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8994 return;
1da177e4 8995 if ( ! (
2953079c 8996 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
1da177e4 8997 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 8998 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 8999 (mddev->external == 0 && mddev->safemode == 1) ||
4ad23a97 9000 (mddev->safemode == 2
fca4d848 9001 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 9002 ))
9003 return;
fca4d848 9004
df5b89b3 9005 if (mddev_trylock(mddev)) {
b4c4c7b8 9006 int spares = 0;
480523fe 9007 bool try_set_sync = mddev->safemode != 0;
fca4d848 9008
afc1f55c 9009 if (!mddev->external && mddev->safemode == 1)
33182d15
N		 9010 mddev->safemode = 0;
9011
c89a8eee 9012 if (mddev->ro) {
ab16bfc7
NB		 9013 struct md_rdev *rdev;
9014 if (!mddev->external && mddev->in_sync)
9015 /* 'Blocked' flag not needed as failed devices
9016 * will be recorded if array switched to read/write.
9017 * Leaving it set will prevent the device
9018 * from being removed.
9019 */
9020 rdev_for_each(rdev, mddev)
9021 clear_bit(Blocked, &rdev->flags);
7ceb17e8
N		 9022 /* On a read-only array we can:
9023 * - remove failed devices
9024 * - add already-in_sync devices if the array itself
9025 * is in-sync.
9026 * As we only add devices that are already in-sync,
9027 * we can activate the spares immediately.
c89a8eee 9028 */
7ceb17e8 9029 remove_and_add_spares(mddev, NULL);
8313b8e5
N		 9030 /* There is no thread, but we need to call
9031 * ->spare_active and clear saved_raid_disk
9032 */
2ac295a5 9033 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5 9034 md_reap_sync_thread(mddev);
a4a3d26d 9035 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8313b8e5 9036 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2953079c 9037 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
c89a8eee
N		 9038 goto unlock;
9039 }
9040
659b254f
GJ		 9041 if (mddev_is_clustered(mddev)) {
9042 struct md_rdev *rdev;
9043 /* kick the device if another node issued a
9044 * remove disk.
9045 */
9046 rdev_for_each(rdev, mddev) {
9047 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9048 rdev->raid_disk < 0)
9049 md_kick_rdev_from_array(rdev);
9050 }
9051 }
9052
480523fe 9053 if (try_set_sync && !mddev->external && !mddev->in_sync) {
85572d7c 9054 spin_lock(&mddev->lock);
6497709b 9055 set_in_sync(mddev);
85572d7c 9056 spin_unlock(&mddev->lock);
fca4d848 9057 }
fca4d848 9058
2953079c 9059 if (mddev->sb_flags)
850b2b42 9060 md_update_sb(mddev, 0);
06d91a5f 9061
1da177e4
LT		 9062 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9063 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9064 /* resync/recovery still happening */
9065 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9066 goto unlock;
9067 }
9068 if (mddev->sync_thread) {
a91d5ac0 9069 md_reap_sync_thread(mddev);
1da177e4
LT		 9070 goto unlock;
9071 }
72a23c21
NB		 9072 /* Set RUNNING before clearing NEEDED to avoid
9073 * any transients in the value of "sync_action".
9074 */
72f36d59 9075 mddev->curr_resync_completed = 0;
23da422b 9076 spin_lock(&mddev->lock);
72a23c21 9077 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
23da422b 9078 spin_unlock(&mddev->lock);
24dd469d
N		 9079 /* Clear some bits that don't mean anything, but
9080 * might be left set
9081 */
24dd469d
N		 9082 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9083 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 9084
ed209584
N		 9085 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9086 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 9087 goto not_running;
1da177e4
LT		 9088 /* no recovery is running.
9089 * remove any failed drives, then
9090 * add spares if possible.
72f36d59 9091 * Spares are also removed and re-added, to allow
1da177e4
LT		 9092 * the personality to fail the re-add.
9093 */
1da177e4 9094
b4c4c7b8 9095 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 9096 if (mddev->pers->check_reshape == NULL ||
9097 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 9098 /* Cannot proceed */
ac05f256 9099 goto not_running;
b4c4c7b8 9100 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 9101 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 9102 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N		 9103 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9104 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 9105 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 9106 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 9107 } else if (mddev->recovery_cp < MaxSector) {
9108 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 9109 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 9110 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9111 /* nothing to be done ... */
ac05f256 9112 goto not_running;
24dd469d 9113
1da177e4 9114 if (mddev->pers->sync_request) {
ef99bf48 9115 if (spares) {
a654b9d8
N		 9116 /* We are adding a device or devices to an array
9117 * which has the bitmap stored on all devices.
9118 * So make sure all bitmap pages get written
9119 */
e64e4018 9120 md_bitmap_write_all(mddev->bitmap);
a654b9d8 9121 }
ac05f256
N		 9122 INIT_WORK(&mddev->del_work, md_start_sync);
9123 queue_work(md_misc_wq, &mddev->del_work);
9124 goto unlock;
1da177e4 9125 }
ac05f256 9126 not_running:
72a23c21
NB		 9127 if (!mddev->sync_thread) {
9128 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
f851b60d 9129 wake_up(&resync_wait);
72a23c21
NB		 9130 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9131 &mddev->recovery))
0c3573f1 9132 if (mddev->sysfs_action)
00bcb4ac 9133 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 9134 }
ac05f256
N		 9135 unlock:
9136 wake_up(&mddev->sb_wait);
1da177e4
LT		 9137 mddev_unlock(mddev);
9138 }
9139}
6c144d31 9140EXPORT_SYMBOL(md_check_recovery);
1da177e4 9141
a91d5ac0
JB		 9142void md_reap_sync_thread(struct mddev *mddev)
9143{
9144 struct md_rdev *rdev;
aefb2e5f
GJ		 9145 sector_t old_dev_sectors = mddev->dev_sectors;
9146 bool is_reshaped = false;
a91d5ac0
JB		 9147
9148 /* resync has finished, collect result */
9149 md_unregister_thread(&mddev->sync_thread);
9150 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
0d8ed0e9
GJ		 9151 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9152 mddev->degraded != mddev->raid_disks) {
a91d5ac0
JB		 9153 /* success...*/
9154 /* activate any spares */
9155 if (mddev->pers->spare_active(mddev)) {
9156 sysfs_notify(&mddev->kobj, NULL,
9157 "degraded");
2953079c 9158 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
a91d5ac0
JB		 9159 }
9160 }
9161 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
aefb2e5f 9162 mddev->pers->finish_reshape) {
a91d5ac0 9163 mddev->pers->finish_reshape(mddev);
aefb2e5f
GJ		 9164 if (mddev_is_clustered(mddev))
9165 is_reshaped = true;
9166 }
a91d5ac0
JB		 9167
9168 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 9169 * information must be scrapped.
a91d5ac0 9170 */
f466722c
N		 9171 if (!mddev->degraded)
9172 rdev_for_each(rdev, mddev)
a91d5ac0
JB		 9173 rdev->saved_raid_disk = -1;
9174
9175 md_update_sb(mddev, 1);
2953079c 9176 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
bb8bf15b
GJ		 9177 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9178 * clustered raid */
9179 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9180 md_cluster_ops->resync_finish(mddev);
a91d5ac0 9181 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
ea358cd0 9182 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
a91d5ac0
JB		 9183 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9184 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9185 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9186 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
aefb2e5f
GJ		 9187 /*
9188 * We call md_cluster_ops->update_size here because sync_size could
9189 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9190 * so it is time to update size across cluster.
9191 */
9192 if (mddev_is_clustered(mddev) && is_reshaped
9193 && !test_bit(MD_CLOSING, &mddev->flags))
9194 md_cluster_ops->update_size(mddev, old_dev_sectors);
f851b60d 9195 wake_up(&resync_wait);
a91d5ac0
JB		 9196 /* flag recovery needed just to double check */
9197 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9198 sysfs_notify_dirent_safe(mddev->sysfs_action);
9199 md_new_event(mddev);
9200 if (mddev->event_work.func)
9201 queue_work(md_misc_wq, &mddev->event_work);
9202}
6c144d31 9203EXPORT_SYMBOL(md_reap_sync_thread);
a91d5ac0 9204
fd01b88c 9205void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 9206{
00bcb4ac 9207 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 9208 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 9209 !test_bit(Blocked, &rdev->flags) &&
9210 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 9211 msecs_to_jiffies(5000));
9212 rdev_dec_pending(rdev, mddev);
9213}
9214EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9215
c6563a8c
N		 9216void md_finish_reshape(struct mddev *mddev)
9217{
9218 /* called be personality module when reshape completes. */
9219 struct md_rdev *rdev;
9220
9221 rdev_for_each(rdev, mddev) {
9222 if (rdev->data_offset > rdev->new_data_offset)
9223 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9224 else
9225 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9226 rdev->data_offset = rdev->new_data_offset;
9227 }
9228}
9229EXPORT_SYMBOL(md_finish_reshape);
2230dfe4 9230
fc974ee2 9231/* Bad block management */
2230dfe4 9232
fc974ee2 9233/* Returns 1 on success, 0 on failure */
3cb03002 9234int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 9235 int is_new)
2230dfe4 9236{
85ad1d13 9237 struct mddev *mddev = rdev->mddev;
c6563a8c
N		 9238 int rv;
9239 if (is_new)
9240 s += rdev->new_data_offset;
9241 else
9242 s += rdev->data_offset;
fc974ee2
VV		 9243 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9244 if (rv == 0) {
2230dfe4 9245 /* Make sure they get written out promptly */
35b785f7
TM		 9246 if (test_bit(ExternalBbl, &rdev->flags))
9247 sysfs_notify(&rdev->kobj, NULL,
9248 "unacknowledged_bad_blocks");
8bd2f0a0 9249 sysfs_notify_dirent_safe(rdev->sysfs_state);
2953079c
SL		 9250 set_mask_bits(&mddev->sb_flags, 0,
9251 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
2230dfe4 9252 md_wakeup_thread(rdev->mddev->thread);
fc974ee2
VV		 9253 return 1;
9254 } else
9255 return 0;
2230dfe4
N		 9256}
9257EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9258
c6563a8c
N		 9259int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9260 int is_new)
2230dfe4 9261{
35b785f7 9262 int rv;
c6563a8c
N		 9263 if (is_new)
9264 s += rdev->new_data_offset;
9265 else
9266 s += rdev->data_offset;
35b785f7
TM		 9267 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9268 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9269 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9270 return rv;
2230dfe4
N		 9271}
9272EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9273
75c96f85
AB		 9274static int md_notify_reboot(struct notifier_block *this,
9275 unsigned long code, void *x)
1da177e4
LT		 9276{
9277 struct list_head *tmp;
fd01b88c 9278 struct mddev *mddev;
2dba6a91 9279 int need_delay = 0;
1da177e4 9280
c744a65c
N		 9281 for_each_mddev(mddev, tmp) {
9282 if (mddev_trylock(mddev)) {
30b8aa91
N		 9283 if (mddev->pers)
9284 __md_stop_writes(mddev);
0f62fb22
N		 9285 if (mddev->persistent)
9286 mddev->safemode = 2;
c744a65c 9287 mddev_unlock(mddev);
2dba6a91 9288 }
c744a65c 9289 need_delay = 1;
1da177e4 9290 }
c744a65c
N		 9291 /*
9292 * certain more exotic SCSI devices are known to be
9293 * volatile wrt too early system reboots. While the
9294 * right place to handle this issue is the given
9295 * driver, we do want to have a safe RAID driver ...
9296 */
9297 if (need_delay)
9298 mdelay(1000*1);
9299
1da177e4
LT		 9300 return NOTIFY_DONE;
9301}
9302
75c96f85 9303static struct notifier_block md_notifier = {
1da177e4
LT		 9304 .notifier_call = md_notify_reboot,
9305 .next = NULL,
9306 .priority = INT_MAX, /* before any real devices */
9307};
9308
9309static void md_geninit(void)
9310{
36a4e1fe 9311 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 9312
c7705f34 9313 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 9314}
9315
75c96f85 9316static int __init md_init(void)
1da177e4 9317{
e804ac78
TH		 9318 int ret = -ENOMEM;
9319
ada609ee 9320 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 9321 if (!md_wq)
9322 goto err_wq;
9323
9324 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9325 if (!md_misc_wq)
9326 goto err_misc_wq;
9327
9328 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9329 goto err_md;
9330
9331 if ((ret = register_blkdev(0, "mdp")) < 0)
9332 goto err_mdp;
9333 mdp_major = ret;
9334
af5628f0 9335 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N		 9336 md_probe, NULL, NULL);
9337 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 9338 md_probe, NULL, NULL);
9339
1da177e4 9340 register_reboot_notifier(&md_notifier);
0b4d4147 9341 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 9342
9343 md_geninit();
d710e138 9344 return 0;
1da177e4 9345
e804ac78
TH		 9346err_mdp:
9347 unregister_blkdev(MD_MAJOR, "md");
9348err_md:
9349 destroy_workqueue(md_misc_wq);
9350err_misc_wq:
9351 destroy_workqueue(md_wq);
9352err_wq:
9353 return ret;
9354}
1da177e4 9355
70bcecdb 9356static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
1d7e3e96 9357{
70bcecdb
GR		 9358 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9359 struct md_rdev *rdev2;
9360 int role, ret;
9361 char b[BDEVNAME_SIZE];
1d7e3e96 9362
818da59f
GJ		 9363 /*
9364 * If size is changed in another node then we need to
9365 * do resize as well.
9366 */
9367 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9368 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9369 if (ret)
9370 pr_info("md-cluster: resize failed\n");
9371 else
e64e4018 9372 md_bitmap_update_sb(mddev->bitmap);
818da59f
GJ		 9373 }
9374
70bcecdb
GR		 9375 /* Check for change of roles in the active devices */
9376 rdev_for_each(rdev2, mddev) {
9377 if (test_bit(Faulty, &rdev2->flags))
9378 continue;
9379
9380 /* Check if the roles changed */
9381 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
dbb64f86
GR		 9382
9383 if (test_bit(Candidate, &rdev2->flags)) {
9384 if (role == 0xfffe) {
9385 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9386 md_kick_rdev_from_array(rdev2);
9387 continue;
9388 }
9389 else
9390 clear_bit(Candidate, &rdev2->flags);
9391 }
9392
70bcecdb 9393 if (role != rdev2->raid_disk) {
ca1e98e0
GJ		 9394 /*
9395 * got activated except reshape is happening.
9396 */
9397 if (rdev2->raid_disk == -1 && role != 0xffff &&
9398 !(le32_to_cpu(sb->feature_map) &
9399 MD_FEATURE_RESHAPE_ACTIVE)) {
70bcecdb
GR		 9400 rdev2->saved_raid_disk = role;
9401 ret = remove_and_add_spares(mddev, rdev2);
9402 pr_info("Activated spare: %s\n",
9d48739e 9403 bdevname(rdev2->bdev,b));
a578183e
GJ		 9404 /* wakeup mddev->thread here, so array could
9405 * perform resync with the new activated disk */
9406 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9407 md_wakeup_thread(mddev->thread);
70bcecdb
GR		 9408 }
9409 /* device faulty
9410 * We just want to do the minimum to mark the disk
9411 * as faulty. The recovery is performed by the
9412 * one who initiated the error.
9413 */
9414 if ((role == 0xfffe) || (role == 0xfffd)) {
9415 md_error(mddev, rdev2);
9416 clear_bit(Blocked, &rdev2->flags);
9417 }
9418 }
1d7e3e96 9419 }
70bcecdb 9420
28c1b9fd
GR		 9421 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9422 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
70bcecdb 9423
7564beda
GJ		 9424 /*
9425 * Since mddev->delta_disks has already updated in update_raid_disks,
9426 * so it is time to check reshape.
9427 */
9428 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9429 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9430 /*
9431 * reshape is happening in the remote node, we need to
9432 * update reshape_position and call start_reshape.
9433 */
ed4d0a4e 9434 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
7564beda
GJ		 9435 if (mddev->pers->update_reshape_pos)
9436 mddev->pers->update_reshape_pos(mddev);
9437 if (mddev->pers->start_reshape)
9438 mddev->pers->start_reshape(mddev);
9439 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9440 mddev->reshape_position != MaxSector &&
9441 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9442 /* reshape is just done in another node. */
9443 mddev->reshape_position = MaxSector;
9444 if (mddev->pers->update_reshape_pos)
9445 mddev->pers->update_reshape_pos(mddev);
9446 }
9447
70bcecdb
GR		 9448 /* Finally set the event to be up to date */
9449 mddev->events = le64_to_cpu(sb->events);
9450}
9451
9452static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9453{
9454 int err;
9455 struct page *swapout = rdev->sb_page;
9456 struct mdp_superblock_1 *sb;
9457
9458 /* Store the sb page of the rdev in the swapout temporary
9459 * variable in case we err in the future
9460 */
9461 rdev->sb_page = NULL;
7f0f0d87
N		 9462 err = alloc_disk_sb(rdev);
9463 if (err == 0) {
9464 ClearPageUptodate(rdev->sb_page);
9465 rdev->sb_loaded = 0;
9466 err = super_types[mddev->major_version].
9467 load_super(rdev, NULL, mddev->minor_version);
9468 }
70bcecdb
GR		 9469 if (err < 0) {
9470 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9471 __func__, __LINE__, rdev->desc_nr, err);
7f0f0d87
N		 9472 if (rdev->sb_page)
9473 put_page(rdev->sb_page);
70bcecdb
GR		 9474 rdev->sb_page = swapout;
9475 rdev->sb_loaded = 1;
9476 return err;
1d7e3e96
GR		 9477 }
9478
70bcecdb
GR		 9479 sb = page_address(rdev->sb_page);
9480 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9481 * is not set
9482 */
9483
9484 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9485 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9486
9487 /* The other node finished recovery, call spare_active to set
9488 * device In_sync and mddev->degraded
9489 */
9490 if (rdev->recovery_offset == MaxSector &&
9491 !test_bit(In_sync, &rdev->flags) &&
9492 mddev->pers->spare_active(mddev))
9493 sysfs_notify(&mddev->kobj, NULL, "degraded");
9494
9495 put_page(swapout);
9496 return 0;
9497}
9498
9499void md_reload_sb(struct mddev *mddev, int nr)
9500{
9501 struct md_rdev *rdev;
9502 int err;
9503
9504 /* Find the rdev */
9505 rdev_for_each_rcu(rdev, mddev) {
9506 if (rdev->desc_nr == nr)
9507 break;
9508 }
9509
9510 if (!rdev || rdev->desc_nr != nr) {
9511 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9512 return;
9513 }
9514
9515 err = read_rdev(mddev, rdev);
9516 if (err < 0)
9517 return;
9518
9519 check_sb_changes(mddev, rdev);
9520
9521 /* Read all rdev's to update recovery_offset */
0ea9924a
GJ		 9522 rdev_for_each_rcu(rdev, mddev) {
9523 if (!test_bit(Faulty, &rdev->flags))
9524 read_rdev(mddev, rdev);
9525 }
1d7e3e96
GR		 9526}
9527EXPORT_SYMBOL(md_reload_sb);
9528
1da177e4
LT		 9529#ifndef MODULE
9530
9531/*
9532 * Searches all registered partitions for autorun RAID arrays
9533 * at boot time.
9534 */
4d936ec1 9535
5b1f5bc3 9536static DEFINE_MUTEX(detected_devices_mutex);
4d936ec1
ME		 9537static LIST_HEAD(all_detected_devices);
9538struct detected_devices_node {
9539 struct list_head list;
9540 dev_t dev;
9541};
1da177e4
LT		 9542
9543void md_autodetect_dev(dev_t dev)
9544{
4d936ec1
ME		 9545 struct detected_devices_node *node_detected_dev;
9546
9547 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9548 if (node_detected_dev) {
9549 node_detected_dev->dev = dev;
5b1f5bc3 9550 mutex_lock(&detected_devices_mutex);
4d936ec1 9551 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5b1f5bc3 9552 mutex_unlock(&detected_devices_mutex);
4d936ec1 9553 }
1da177e4
LT		 9554}
9555
1da177e4
LT		 9556static void autostart_arrays(int part)
9557{
3cb03002 9558 struct md_rdev *rdev;
4d936ec1
ME		 9559 struct detected_devices_node *node_detected_dev;
9560 dev_t dev;
9561 int i_scanned, i_passed;
1da177e4 9562
4d936ec1
ME		 9563 i_scanned = 0;
9564 i_passed = 0;
1da177e4 9565
9d48739e 9566 pr_info("md: Autodetecting RAID arrays.\n");
1da177e4 9567
5b1f5bc3 9568 mutex_lock(&detected_devices_mutex);
4d936ec1
ME		 9569 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9570 i_scanned++;
9571 node_detected_dev = list_entry(all_detected_devices.next,
9572 struct detected_devices_node, list);
9573 list_del(&node_detected_dev->list);
9574 dev = node_detected_dev->dev;
9575 kfree(node_detected_dev);
90bcf133 9576 mutex_unlock(&detected_devices_mutex);
df968c4e 9577 rdev = md_import_device(dev,0, 90);
90bcf133 9578 mutex_lock(&detected_devices_mutex);
1da177e4
LT		 9579 if (IS_ERR(rdev))
9580 continue;
9581
403df478 9582 if (test_bit(Faulty, &rdev->flags))
1da177e4 9583 continue;
403df478 9584
d0fae18f 9585 set_bit(AutoDetected, &rdev->flags);
1da177e4 9586 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 9587 i_passed++;
1da177e4 9588 }
5b1f5bc3 9589 mutex_unlock(&detected_devices_mutex);
4d936ec1 9590
9d48739e 9591 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
1da177e4
LT		 9592
9593 autorun_devices(part);
9594}
9595
fdee8ae4 9596#endif /* !MODULE */
1da177e4
LT		 9597
9598static __exit void md_exit(void)
9599{
fd01b88c 9600 struct mddev *mddev;
1da177e4 9601 struct list_head *tmp;
e2f23b60 9602 int delay = 1;
8ab5e4c1 9603
af5628f0 9604 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 9605 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 9606
3dbd8c2e 9607 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 9608 unregister_blkdev(mdp_major, "mdp");
9609 unregister_reboot_notifier(&md_notifier);
9610 unregister_sysctl_table(raid_table_header);
e2f23b60
N		 9611
9612 /* We cannot unload the modules while some process is
9613 * waiting for us in select() or poll() - wake them up
9614 */
9615 md_unloading = 1;
9616 while (waitqueue_active(&md_event_waiters)) {
9617 /* not safe to leave yet */
9618 wake_up(&md_event_waiters);
9619 msleep(delay);
9620 delay += delay;
9621 }
1da177e4 9622 remove_proc_entry("mdstat", NULL);
e2f23b60 9623
29ac4aa3 9624 for_each_mddev(mddev, tmp) {
1da177e4 9625 export_array(mddev);
9356863c 9626 mddev->ctime = 0;
d3374825 9627 mddev->hold_active = 0;
9356863c
N		 9628 /*
9629 * for_each_mddev() will call mddev_put() at the end of each
9630 * iteration. As the mddev is now fully clear, this will
9631 * schedule the mddev for destruction by a workqueue, and the
9632 * destroy_workqueue() below will wait for that to complete.
9633 */
1da177e4 9634 }
e804ac78
TH		 9635 destroy_workqueue(md_misc_wq);
9636 destroy_workqueue(md_wq);
1da177e4
LT		 9637}
9638
685784aa 9639subsys_initcall(md_init);
1da177e4
LT		 9640module_exit(md_exit)
9641
e4dca7b7 9642static int get_ro(char *buffer, const struct kernel_param *kp)
f91de92e
N		 9643{
9644 return sprintf(buffer, "%d", start_readonly);
9645}
e4dca7b7 9646static int set_ro(const char *val, const struct kernel_param *kp)
f91de92e 9647{
4c9309c0 9648 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
f91de92e
N		 9649}
9650
80ca3a44
N		 9651module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9652module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
efeb53c0 9653module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
78b6350d 9654module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
f91de92e 9655
1da177e4 9656MODULE_LICENSE("GPL");
0efb9e61 9657MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 9658MODULE_ALIAS("md");
72008652 9659MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)