]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - drivers/md/md.c
md: don't allow "-sync" to be set for device in an active array.
[mirror_ubuntu-artful-kernel.git] / drivers / md / md.c
CommitLineData
1da177e4
LT
1/*
2 md.c : Multiple Devices driver for Linux
f72ffdd6 3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
1da177e4
LT
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N
43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N
51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
1da177e4 56
1da177e4 57#ifndef MODULE
d710e138 58static void autostart_arrays(int part);
1da177e4
LT
59#endif
60
01f96c0a
N
61/* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
2604b703 66static LIST_HEAD(pers_list);
1da177e4
LT
67static DEFINE_SPINLOCK(pers_lock);
68
5e56341d
AB
69static void md_print_devices(void);
70
90b08710 71static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH
72static struct workqueue_struct *md_wq;
73static struct workqueue_struct *md_misc_wq;
90b08710 74
746d3207
N
75static int remove_and_add_spares(struct mddev *mddev,
76 struct md_rdev *this);
77
5e56341d
AB
78#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
79
1e50915f
RB
80/*
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
84 */
85#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT
86/*
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 90 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT
91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
93 * idle IO detection.
94 *
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 96 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT
97 */
98
99static int sysctl_speed_limit_min = 1000;
100static int sysctl_speed_limit_max = 200000;
fd01b88c 101static inline int speed_min(struct mddev *mddev)
88202a0c
N
102{
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
105}
106
fd01b88c 107static inline int speed_max(struct mddev *mddev)
88202a0c
N
108{
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
111}
1da177e4
LT
112
113static struct ctl_table_header *raid_table_header;
114
82592c38 115static struct ctl_table raid_table[] = {
1da177e4 116 {
1da177e4
LT
117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
80ca3a44 120 .mode = S_IRUGO|S_IWUSR,
6d456111 121 .proc_handler = proc_dointvec,
1da177e4
LT
122 },
123 {
1da177e4
LT
124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
80ca3a44 127 .mode = S_IRUGO|S_IWUSR,
6d456111 128 .proc_handler = proc_dointvec,
1da177e4 129 },
894d2491 130 { }
1da177e4
LT
131};
132
82592c38 133static struct ctl_table raid_dir_table[] = {
1da177e4 134 {
1da177e4
LT
135 .procname = "raid",
136 .maxlen = 0,
80ca3a44 137 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT
138 .child = raid_table,
139 },
894d2491 140 { }
1da177e4
LT
141};
142
82592c38 143static struct ctl_table raid_root_table[] = {
1da177e4 144 {
1da177e4
LT
145 .procname = "dev",
146 .maxlen = 0,
147 .mode = 0555,
148 .child = raid_dir_table,
149 },
894d2491 150 { }
1da177e4
LT
151};
152
83d5cde4 153static const struct block_device_operations md_fops;
1da177e4 154
f91de92e
N
155static int start_readonly;
156
a167f663
N
157/* bio_clone_mddev
158 * like bio_clone, but with a local bio set
159 */
160
a167f663 161struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 162 struct mddev *mddev)
a167f663
N
163{
164 struct bio *b;
a167f663
N
165
166 if (!mddev || !mddev->bio_set)
167 return bio_alloc(gfp_mask, nr_iovecs);
168
395c72a7 169 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
a167f663
N
170 if (!b)
171 return NULL;
a167f663
N
172 return b;
173}
174EXPORT_SYMBOL_GPL(bio_alloc_mddev);
175
176struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 177 struct mddev *mddev)
a167f663 178{
a167f663
N
179 if (!mddev || !mddev->bio_set)
180 return bio_clone(bio, gfp_mask);
181
bf800ef1 182 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
a167f663
N
183}
184EXPORT_SYMBOL_GPL(bio_clone_mddev);
185
d7603b7e
N
186/*
187 * We have a system wide 'event count' that is incremented
188 * on any 'interesting' event, and readers of /proc/mdstat
189 * can use 'poll' or 'select' to find out when the event
190 * count increases.
191 *
192 * Events are:
193 * start array, stop array, error, add device, remove device,
194 * start build, activate spare
195 */
2989ddbd 196static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 197static atomic_t md_event_count;
fd01b88c 198void md_new_event(struct mddev *mddev)
d7603b7e
N
199{
200 atomic_inc(&md_event_count);
201 wake_up(&md_event_waiters);
202}
29269553 203EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 204
c331eb04
N
205/* Alternate version that can be called from interrupts
206 * when calling sysfs_notify isn't needed.
207 */
fd01b88c 208static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N
209{
210 atomic_inc(&md_event_count);
211 wake_up(&md_event_waiters);
212}
213
1da177e4
LT
214/*
215 * Enables to iterate over all existing md arrays
216 * all_mddevs_lock protects this list.
217 */
218static LIST_HEAD(all_mddevs);
219static DEFINE_SPINLOCK(all_mddevs_lock);
220
1da177e4
LT
221/*
222 * iterates through all used mddevs in the system.
223 * We take care to grab the all_mddevs_lock whenever navigating
224 * the list, and to always hold a refcount when unlocked.
225 * Any code which breaks out of this loop while own
226 * a reference to the current mddev and must mddev_put it.
227 */
fd01b88c 228#define for_each_mddev(_mddev,_tmp) \
1da177e4 229 \
f72ffdd6 230 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N
231 _tmp = all_mddevs.next; \
232 _mddev = NULL;}); \
233 ({ if (_tmp != &all_mddevs) \
234 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 235 spin_unlock(&all_mddevs_lock); \
fd01b88c
N
236 if (_mddev) mddev_put(_mddev); \
237 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
238 _tmp != &all_mddevs;}); \
1da177e4 239 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 240 _tmp = _tmp->next;}) \
1da177e4
LT
241 )
242
409c57f3
N
243/* Rather than calling directly into the personality make_request function,
244 * IO requests come here first so that we can check if the device is
245 * being suspended pending a reconfiguration.
246 * We hold a refcount over the call to ->make_request. By the time that
247 * call has finished, the bio has been linked into some internal structure
248 * and so is visible to ->quiesce(), so we don't need the refcount any more.
249 */
5a7bbad2 250static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 251{
49077326 252 const int rw = bio_data_dir(bio);
fd01b88c 253 struct mddev *mddev = q->queuedata;
49077326 254 int cpu;
e91ece55 255 unsigned int sectors;
49077326 256
0ca69886
N
257 if (mddev == NULL || mddev->pers == NULL
258 || !mddev->ready) {
409c57f3 259 bio_io_error(bio);
5a7bbad2 260 return;
409c57f3 261 }
bbfa57c0
SR
262 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
263 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
264 return;
265 }
0ca69886 266 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 267 rcu_read_lock();
e9c7469b 268 if (mddev->suspended) {
409c57f3
N
269 DEFINE_WAIT(__wait);
270 for (;;) {
271 prepare_to_wait(&mddev->sb_wait, &__wait,
272 TASK_UNINTERRUPTIBLE);
e9c7469b 273 if (!mddev->suspended)
409c57f3
N
274 break;
275 rcu_read_unlock();
276 schedule();
277 rcu_read_lock();
278 }
279 finish_wait(&mddev->sb_wait, &__wait);
280 }
281 atomic_inc(&mddev->active_io);
282 rcu_read_unlock();
49077326 283
e91ece55
CM
284 /*
285 * save the sectors now since our bio can
286 * go away inside make_request
287 */
288 sectors = bio_sectors(bio);
5a7bbad2 289 mddev->pers->make_request(mddev, bio);
49077326
N
290
291 cpu = part_stat_lock();
292 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 293 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N
294 part_stat_unlock();
295
409c57f3
N
296 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
297 wake_up(&mddev->sb_wait);
409c57f3
N
298}
299
9e35b99c
N
300/* mddev_suspend makes sure no new requests are submitted
301 * to the device, and that any requests that have been submitted
302 * are completely handled.
303 * Once ->stop is called and completes, the module will be completely
304 * unused.
305 */
fd01b88c 306void mddev_suspend(struct mddev *mddev)
409c57f3
N
307{
308 BUG_ON(mddev->suspended);
309 mddev->suspended = 1;
310 synchronize_rcu();
311 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
312 mddev->pers->quiesce(mddev, 1);
0d9f4f13
JB
313
314 del_timer_sync(&mddev->safemode_timer);
409c57f3 315}
390ee602 316EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 317
fd01b88c 318void mddev_resume(struct mddev *mddev)
409c57f3
N
319{
320 mddev->suspended = 0;
321 wake_up(&mddev->sb_wait);
322 mddev->pers->quiesce(mddev, 0);
0fd018af 323
47525e59 324 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
0fd018af
JB
325 md_wakeup_thread(mddev->thread);
326 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 327}
390ee602 328EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 329
fd01b88c 330int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N
331{
332 return mddev->suspended;
333}
334EXPORT_SYMBOL(mddev_congested);
335
a2826aa9 336/*
e9c7469b 337 * Generic flush handling for md
a2826aa9
N
338 */
339
e9c7469b 340static void md_end_flush(struct bio *bio, int err)
a2826aa9 341{
3cb03002 342 struct md_rdev *rdev = bio->bi_private;
fd01b88c 343 struct mddev *mddev = rdev->mddev;
a2826aa9
N
344
345 rdev_dec_pending(rdev, mddev);
346
347 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 348 /* The pre-request flush has finished */
e804ac78 349 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N
350 }
351 bio_put(bio);
352}
353
a7a07e69
N
354static void md_submit_flush_data(struct work_struct *ws);
355
a035fc3e 356static void submit_flushes(struct work_struct *ws)
a2826aa9 357{
fd01b88c 358 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 359 struct md_rdev *rdev;
a2826aa9 360
a7a07e69
N
361 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
362 atomic_set(&mddev->flush_pending, 1);
a2826aa9 363 rcu_read_lock();
dafb20fa 364 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N
365 if (rdev->raid_disk >= 0 &&
366 !test_bit(Faulty, &rdev->flags)) {
367 /* Take two references, one is dropped
368 * when request finishes, one after
369 * we reclaim rcu_read_lock
370 */
371 struct bio *bi;
372 atomic_inc(&rdev->nr_pending);
373 atomic_inc(&rdev->nr_pending);
374 rcu_read_unlock();
b5e1b8ce 375 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
e9c7469b 376 bi->bi_end_io = md_end_flush;
a2826aa9
N
377 bi->bi_private = rdev;
378 bi->bi_bdev = rdev->bdev;
379 atomic_inc(&mddev->flush_pending);
e9c7469b 380 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N
381 rcu_read_lock();
382 rdev_dec_pending(rdev, mddev);
383 }
384 rcu_read_unlock();
a7a07e69
N
385 if (atomic_dec_and_test(&mddev->flush_pending))
386 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N
387}
388
e9c7469b 389static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 390{
fd01b88c 391 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 392 struct bio *bio = mddev->flush_bio;
a2826aa9 393
4f024f37 394 if (bio->bi_iter.bi_size == 0)
a2826aa9
N
395 /* an empty barrier - all done */
396 bio_endio(bio, 0);
397 else {
e9c7469b 398 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 399 mddev->pers->make_request(mddev, bio);
a2826aa9 400 }
2b74e12e
N
401
402 mddev->flush_bio = NULL;
403 wake_up(&mddev->sb_wait);
a2826aa9
N
404}
405
fd01b88c 406void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N
407{
408 spin_lock_irq(&mddev->write_lock);
409 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 410 !mddev->flush_bio,
eed8c02e 411 mddev->write_lock);
e9c7469b 412 mddev->flush_bio = bio;
a2826aa9
N
413 spin_unlock_irq(&mddev->write_lock);
414
a035fc3e
N
415 INIT_WORK(&mddev->flush_work, submit_flushes);
416 queue_work(md_wq, &mddev->flush_work);
a2826aa9 417}
e9c7469b 418EXPORT_SYMBOL(md_flush_request);
409c57f3 419
74018dc3 420void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
97658cdd 421{
9cbb1750
N
422 struct mddev *mddev = cb->data;
423 md_wakeup_thread(mddev->thread);
424 kfree(cb);
97658cdd 425}
9cbb1750 426EXPORT_SYMBOL(md_unplug);
2ac87401 427
fd01b88c 428static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT
429{
430 atomic_inc(&mddev->active);
431 return mddev;
432}
433
5fd3a17e 434static void mddev_delayed_delete(struct work_struct *ws);
d3374825 435
fd01b88c 436static void mddev_put(struct mddev *mddev)
1da177e4 437{
a167f663
N
438 struct bio_set *bs = NULL;
439
1da177e4
LT
440 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
441 return;
d3374825 442 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N
443 mddev->ctime == 0 && !mddev->hold_active) {
444 /* Array is not configured at all, and not held active,
445 * so destroy it */
af8a2434 446 list_del_init(&mddev->all_mddevs);
a167f663
N
447 bs = mddev->bio_set;
448 mddev->bio_set = NULL;
d3374825 449 if (mddev->gendisk) {
e804ac78
TH
450 /* We did a probe so need to clean up. Call
451 * queue_work inside the spinlock so that
452 * flush_workqueue() after mddev_find will
453 * succeed in waiting for the work to be done.
d3374825
N
454 */
455 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 456 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N
457 } else
458 kfree(mddev);
459 }
460 spin_unlock(&all_mddevs_lock);
a167f663
N
461 if (bs)
462 bioset_free(bs);
1da177e4
LT
463}
464
fd01b88c 465void mddev_init(struct mddev *mddev)
fafd7fb0
N
466{
467 mutex_init(&mddev->open_mutex);
468 mutex_init(&mddev->reconfig_mutex);
469 mutex_init(&mddev->bitmap_info.mutex);
470 INIT_LIST_HEAD(&mddev->disks);
471 INIT_LIST_HEAD(&mddev->all_mddevs);
472 init_timer(&mddev->safemode_timer);
473 atomic_set(&mddev->active, 1);
474 atomic_set(&mddev->openers, 0);
475 atomic_set(&mddev->active_io, 0);
476 spin_lock_init(&mddev->write_lock);
477 atomic_set(&mddev->flush_pending, 0);
478 init_waitqueue_head(&mddev->sb_wait);
479 init_waitqueue_head(&mddev->recovery_wait);
480 mddev->reshape_position = MaxSector;
2c810cdd 481 mddev->reshape_backwards = 0;
c4a39551 482 mddev->last_sync_action = "none";
fafd7fb0
N
483 mddev->resync_min = 0;
484 mddev->resync_max = MaxSector;
485 mddev->level = LEVEL_NONE;
486}
390ee602 487EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 488
f72ffdd6 489static struct mddev *mddev_find(dev_t unit)
1da177e4 490{
fd01b88c 491 struct mddev *mddev, *new = NULL;
1da177e4 492
8f5f02c4
N
493 if (unit && MAJOR(unit) != MD_MAJOR)
494 unit &= ~((1<<MdpMinorShift)-1);
495
1da177e4
LT
496 retry:
497 spin_lock(&all_mddevs_lock);
efeb53c0
N
498
499 if (unit) {
500 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
501 if (mddev->unit == unit) {
502 mddev_get(mddev);
503 spin_unlock(&all_mddevs_lock);
504 kfree(new);
505 return mddev;
506 }
507
508 if (new) {
509 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 510 spin_unlock(&all_mddevs_lock);
efeb53c0
N
511 new->hold_active = UNTIL_IOCTL;
512 return new;
1da177e4 513 }
efeb53c0
N
514 } else if (new) {
515 /* find an unused unit number */
516 static int next_minor = 512;
517 int start = next_minor;
518 int is_free = 0;
519 int dev = 0;
520 while (!is_free) {
521 dev = MKDEV(MD_MAJOR, next_minor);
522 next_minor++;
523 if (next_minor > MINORMASK)
524 next_minor = 0;
525 if (next_minor == start) {
526 /* Oh dear, all in use. */
527 spin_unlock(&all_mddevs_lock);
528 kfree(new);
529 return NULL;
530 }
f72ffdd6 531
efeb53c0
N
532 is_free = 1;
533 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
534 if (mddev->unit == dev) {
535 is_free = 0;
536 break;
537 }
538 }
539 new->unit = dev;
540 new->md_minor = MINOR(dev);
541 new->hold_active = UNTIL_STOP;
1da177e4
LT
542 list_add(&new->all_mddevs, &all_mddevs);
543 spin_unlock(&all_mddevs_lock);
544 return new;
545 }
546 spin_unlock(&all_mddevs_lock);
547
9ffae0cf 548 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT
549 if (!new)
550 return NULL;
551
1da177e4
LT
552 new->unit = unit;
553 if (MAJOR(unit) == MD_MAJOR)
554 new->md_minor = MINOR(unit);
555 else
556 new->md_minor = MINOR(unit) >> MdpMinorShift;
557
fafd7fb0 558 mddev_init(new);
1da177e4 559
1da177e4
LT
560 goto retry;
561}
562
f72ffdd6 563static inline int __must_check mddev_lock(struct mddev *mddev)
1da177e4 564{
df5b89b3 565 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT
566}
567
29f097c4
N
568/* Sometimes we need to take the lock in a situation where
569 * failure due to interrupts is not acceptable.
570 */
f72ffdd6 571static inline void mddev_lock_nointr(struct mddev *mddev)
29f097c4
N
572{
573 mutex_lock(&mddev->reconfig_mutex);
574}
575
fd01b88c 576static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW
577{
578 return mutex_is_locked(&mddev->reconfig_mutex);
579}
580
f72ffdd6 581static inline int mddev_trylock(struct mddev *mddev)
1da177e4 582{
df5b89b3 583 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT
584}
585
b6eb127d
N
586static struct attribute_group md_redundancy_group;
587
f72ffdd6 588static void mddev_unlock(struct mddev *mddev)
1da177e4 589{
a64c876f 590 if (mddev->to_remove) {
b6eb127d
N
591 /* These cannot be removed under reconfig_mutex as
592 * an access to the files will try to take reconfig_mutex
593 * while holding the file unremovable, which leads to
594 * a deadlock.
bb4f1e9d
N
595 * So hold set sysfs_active while the remove in happeing,
596 * and anything else which might set ->to_remove or my
597 * otherwise change the sysfs namespace will fail with
598 * -EBUSY if sysfs_active is still set.
599 * We set sysfs_active under reconfig_mutex and elsewhere
600 * test it under the same mutex to ensure its correct value
601 * is seen.
b6eb127d 602 */
a64c876f
N
603 struct attribute_group *to_remove = mddev->to_remove;
604 mddev->to_remove = NULL;
bb4f1e9d 605 mddev->sysfs_active = 1;
b6eb127d
N
606 mutex_unlock(&mddev->reconfig_mutex);
607
00bcb4ac
N
608 if (mddev->kobj.sd) {
609 if (to_remove != &md_redundancy_group)
610 sysfs_remove_group(&mddev->kobj, to_remove);
611 if (mddev->pers == NULL ||
612 mddev->pers->sync_request == NULL) {
613 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
614 if (mddev->sysfs_action)
615 sysfs_put(mddev->sysfs_action);
616 mddev->sysfs_action = NULL;
617 }
a64c876f 618 }
bb4f1e9d 619 mddev->sysfs_active = 0;
b6eb127d
N
620 } else
621 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 622
751e67ca
CD
623 /* As we've dropped the mutex we need a spinlock to
624 * make sure the thread doesn't disappear
01f96c0a
N
625 */
626 spin_lock(&pers_lock);
005eca5e 627 md_wakeup_thread(mddev->thread);
01f96c0a 628 spin_unlock(&pers_lock);
1da177e4
LT
629}
630
1ca69c4b
N
631static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
632{
633 struct md_rdev *rdev;
634
635 rdev_for_each_rcu(rdev, mddev)
636 if (rdev->desc_nr == nr)
637 return rdev;
638
639 return NULL;
640}
641
642static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
1da177e4 643{
3cb03002 644 struct md_rdev *rdev;
1da177e4 645
dafb20fa 646 rdev_for_each(rdev, mddev)
1da177e4
LT
647 if (rdev->bdev->bd_dev == dev)
648 return rdev;
159ec1fc 649
1da177e4
LT
650 return NULL;
651}
652
1ca69c4b
N
653static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
654{
655 struct md_rdev *rdev;
656
657 rdev_for_each_rcu(rdev, mddev)
658 if (rdev->bdev->bd_dev == dev)
659 return rdev;
660
661 return NULL;
662}
663
84fc4b56 664static struct md_personality *find_pers(int level, char *clevel)
2604b703 665{
84fc4b56 666 struct md_personality *pers;
d9d166c2
N
667 list_for_each_entry(pers, &pers_list, list) {
668 if (level != LEVEL_NONE && pers->level == level)
2604b703 669 return pers;
d9d166c2
N
670 if (strcmp(pers->name, clevel)==0)
671 return pers;
672 }
2604b703
N
673 return NULL;
674}
675
b73df2d3 676/* return the offset of the super block in 512byte sectors */
3cb03002 677static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 678{
57b2caa3 679 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 680 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT
681}
682
f72ffdd6 683static int alloc_disk_sb(struct md_rdev *rdev)
1da177e4
LT
684{
685 if (rdev->sb_page)
686 MD_BUG();
687
688 rdev->sb_page = alloc_page(GFP_KERNEL);
689 if (!rdev->sb_page) {
690 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 691 return -ENOMEM;
1da177e4
LT
692 }
693
694 return 0;
695}
696
545c8795 697void md_rdev_clear(struct md_rdev *rdev)
1da177e4
LT
698{
699 if (rdev->sb_page) {
2d1f3b5d 700 put_page(rdev->sb_page);
1da177e4
LT
701 rdev->sb_loaded = 0;
702 rdev->sb_page = NULL;
0f420358 703 rdev->sb_start = 0;
dd8ac336 704 rdev->sectors = 0;
1da177e4 705 }
2699b672
N
706 if (rdev->bb_page) {
707 put_page(rdev->bb_page);
708 rdev->bb_page = NULL;
709 }
4fa2f327
N
710 kfree(rdev->badblocks.page);
711 rdev->badblocks.page = NULL;
1da177e4 712}
545c8795 713EXPORT_SYMBOL_GPL(md_rdev_clear);
1da177e4 714
6712ecf8 715static void super_written(struct bio *bio, int error)
7bfa19f2 716{
3cb03002 717 struct md_rdev *rdev = bio->bi_private;
fd01b88c 718 struct mddev *mddev = rdev->mddev;
7bfa19f2 719
3a0f5bbb
N
720 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
721 printk("md: super_written gets error=%d, uptodate=%d\n",
722 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
723 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 724 md_error(mddev, rdev);
3a0f5bbb 725 }
7bfa19f2 726
a9701a30
N
727 if (atomic_dec_and_test(&mddev->pending_writes))
728 wake_up(&mddev->sb_wait);
f8b58edf 729 bio_put(bio);
7bfa19f2
N
730}
731
fd01b88c 732void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N
733 sector_t sector, int size, struct page *page)
734{
735 /* write first size bytes of page to sector of rdev
736 * Increment mddev->pending_writes before returning
737 * and decrement it on completion, waking up sb_wait
738 * if zero is reached.
739 * If an error occurred, call md_error
740 */
a167f663 741 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 742
a6ff7e08 743 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
4f024f37 744 bio->bi_iter.bi_sector = sector;
7bfa19f2
N
745 bio_add_page(bio, page, size, 0);
746 bio->bi_private = rdev;
747 bio->bi_end_io = super_written;
a9701a30 748
7bfa19f2 749 atomic_inc(&mddev->pending_writes);
a5bf4df0 750 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N
751}
752
fd01b88c 753void md_super_wait(struct mddev *mddev)
a9701a30 754{
e9c7469b 755 /* wait for all superblock writes that were scheduled to complete */
1967cd56 756 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
7bfa19f2
N
757}
758
3cb03002 759int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 760 struct page *page, int rw, bool metadata_op)
1da177e4 761{
a167f663 762 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT
763 int ret;
764
a6ff7e08
JB
765 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
766 rdev->meta_bdev : rdev->bdev;
ccebd4c4 767 if (metadata_op)
4f024f37 768 bio->bi_iter.bi_sector = sector + rdev->sb_start;
1fdd6fc9
N
769 else if (rdev->mddev->reshape_position != MaxSector &&
770 (rdev->mddev->reshape_backwards ==
771 (sector >= rdev->mddev->reshape_position)))
4f024f37 772 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
ccebd4c4 773 else
4f024f37 774 bio->bi_iter.bi_sector = sector + rdev->data_offset;
1da177e4 775 bio_add_page(bio, page, size, 0);
c170bbb4 776 submit_bio_wait(rw, bio);
1da177e4
LT
777
778 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
779 bio_put(bio);
780 return ret;
781}
a8745db2 782EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 783
f72ffdd6 784static int read_disk_sb(struct md_rdev *rdev, int size)
1da177e4
LT
785{
786 char b[BDEVNAME_SIZE];
787 if (!rdev->sb_page) {
788 MD_BUG();
789 return -EINVAL;
790 }
791 if (rdev->sb_loaded)
792 return 0;
793
ccebd4c4 794 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT
795 goto fail;
796 rdev->sb_loaded = 1;
797 return 0;
798
799fail:
800 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
801 bdevname(rdev->bdev,b));
802 return -EINVAL;
803}
804
805static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
806{
f72ffdd6 807 return sb1->set_uuid0 == sb2->set_uuid0 &&
05710466
AN
808 sb1->set_uuid1 == sb2->set_uuid1 &&
809 sb1->set_uuid2 == sb2->set_uuid2 &&
810 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT
811}
812
1da177e4
LT
813static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
814{
815 int ret;
816 mdp_super_t *tmp1, *tmp2;
817
818 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
819 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
820
821 if (!tmp1 || !tmp2) {
822 ret = 0;
35020f1a 823 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT
824 goto abort;
825 }
826
827 *tmp1 = *sb1;
828 *tmp2 = *sb2;
829
830 /*
831 * nr_disks is not constant
832 */
833 tmp1->nr_disks = 0;
834 tmp2->nr_disks = 0;
835
ce0c8e05 836 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 837abort:
990a8baf
JJ
838 kfree(tmp1);
839 kfree(tmp2);
1da177e4
LT
840 return ret;
841}
842
4d167f09
N
843static u32 md_csum_fold(u32 csum)
844{
845 csum = (csum & 0xffff) + (csum >> 16);
846 return (csum & 0xffff) + (csum >> 16);
847}
848
f72ffdd6 849static unsigned int calc_sb_csum(mdp_super_t *sb)
1da177e4 850{
4d167f09
N
851 u64 newcsum = 0;
852 u32 *sb32 = (u32*)sb;
853 int i;
1da177e4
LT
854 unsigned int disk_csum, csum;
855
856 disk_csum = sb->sb_csum;
857 sb->sb_csum = 0;
4d167f09
N
858
859 for (i = 0; i < MD_SB_BYTES/4 ; i++)
860 newcsum += sb32[i];
861 csum = (newcsum & 0xffffffff) + (newcsum>>32);
862
4d167f09
N
863#ifdef CONFIG_ALPHA
864 /* This used to use csum_partial, which was wrong for several
865 * reasons including that different results are returned on
866 * different architectures. It isn't critical that we get exactly
867 * the same return value as before (we always csum_fold before
868 * testing, and that removes any differences). However as we
869 * know that csum_partial always returned a 16bit value on
870 * alphas, do a fold to maximise conformity to previous behaviour.
871 */
872 sb->sb_csum = md_csum_fold(disk_csum);
873#else
1da177e4 874 sb->sb_csum = disk_csum;
4d167f09 875#endif
1da177e4
LT
876 return csum;
877}
878
1da177e4
LT
879/*
880 * Handle superblock details.
881 * We want to be able to handle multiple superblock formats
882 * so we have a common interface to them all, and an array of
883 * different handlers.
884 * We rely on user-space to write the initial superblock, and support
885 * reading and updating of superblocks.
886 * Interface methods are:
3cb03002 887 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT
888 * loads and validates a superblock on dev.
889 * if refdev != NULL, compare superblocks on both devices
890 * Return:
891 * 0 - dev has a superblock that is compatible with refdev
892 * 1 - dev has a superblock that is compatible and newer than refdev
893 * so dev should be used as the refdev in future
894 * -EINVAL superblock incompatible or invalid
895 * -othererror e.g. -EIO
896 *
fd01b88c 897 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT
898 * Verify that dev is acceptable into mddev.
899 * The first time, mddev->raid_disks will be 0, and data from
900 * dev should be merged in. Subsequent calls check that dev
901 * is new enough. Return 0 or -EINVAL
902 *
fd01b88c 903 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT
904 * Update the superblock for rdev with data in mddev
905 * This does not write to disc.
906 *
907 */
908
909struct super_type {
0cd17fec
CW
910 char *name;
911 struct module *owner;
c6563a8c
N
912 int (*load_super)(struct md_rdev *rdev,
913 struct md_rdev *refdev,
0cd17fec 914 int minor_version);
c6563a8c
N
915 int (*validate_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 void (*sync_super)(struct mddev *mddev,
918 struct md_rdev *rdev);
3cb03002 919 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 920 sector_t num_sectors);
c6563a8c
N
921 int (*allow_new_offset)(struct md_rdev *rdev,
922 unsigned long long new_offset);
1da177e4
LT
923};
924
0894cc30
AN
925/*
926 * Check that the given mddev has no bitmap.
927 *
928 * This function is called from the run method of all personalities that do not
929 * support bitmaps. It prints an error message and returns non-zero if mddev
930 * has a bitmap. Otherwise, it returns 0.
931 *
932 */
fd01b88c 933int md_check_no_bitmap(struct mddev *mddev)
0894cc30 934{
c3d9714e 935 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN
936 return 0;
937 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
938 mdname(mddev), mddev->pers->name);
939 return 1;
940}
941EXPORT_SYMBOL(md_check_no_bitmap);
942
1da177e4 943/*
f72ffdd6 944 * load_super for 0.90.0
1da177e4 945 */
3cb03002 946static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT
947{
948 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
949 mdp_super_t *sb;
950 int ret;
1da177e4
LT
951
952 /*
0f420358 953 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT
954 * it's at the end of the disk.
955 *
956 * It also happens to be a multiple of 4Kb.
957 */
57b2caa3 958 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 959
0002b271 960 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT
961 if (ret) return ret;
962
963 ret = -EINVAL;
964
965 bdevname(rdev->bdev, b);
65a06f06 966 sb = page_address(rdev->sb_page);
1da177e4
LT
967
968 if (sb->md_magic != MD_SB_MAGIC) {
969 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
970 b);
971 goto abort;
972 }
973
974 if (sb->major_version != 0 ||
f6705578
N
975 sb->minor_version < 90 ||
976 sb->minor_version > 91) {
1da177e4
LT
977 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
978 sb->major_version, sb->minor_version,
979 b);
980 goto abort;
981 }
982
983 if (sb->raid_disks <= 0)
984 goto abort;
985
4d167f09 986 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT
987 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
988 b);
989 goto abort;
990 }
991
992 rdev->preferred_minor = sb->md_minor;
993 rdev->data_offset = 0;
c6563a8c 994 rdev->new_data_offset = 0;
0002b271 995 rdev->sb_size = MD_SB_BYTES;
9f2f3830 996 rdev->badblocks.shift = -1;
1da177e4
LT
997
998 if (sb->level == LEVEL_MULTIPATH)
999 rdev->desc_nr = -1;
1000 else
1001 rdev->desc_nr = sb->this_disk.number;
1002
9a7b2b0f 1003 if (!refdev) {
1da177e4 1004 ret = 1;
9a7b2b0f 1005 } else {
1da177e4 1006 __u64 ev1, ev2;
65a06f06 1007 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT
1008 if (!uuid_equal(refsb, sb)) {
1009 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1010 b, bdevname(refdev->bdev,b2));
1011 goto abort;
1012 }
1013 if (!sb_equal(refsb, sb)) {
1014 printk(KERN_WARNING "md: %s has same UUID"
1015 " but different superblock to %s\n",
1016 b, bdevname(refdev->bdev, b2));
1017 goto abort;
1018 }
1019 ev1 = md_event(sb);
1020 ev2 = md_event(refsb);
1021 if (ev1 > ev2)
1022 ret = 1;
f72ffdd6 1023 else
1da177e4
LT
1024 ret = 0;
1025 }
8190e754 1026 rdev->sectors = rdev->sb_start;
667a5313
N
1027 /* Limit to 4TB as metadata cannot record more than that.
1028 * (not needed for Linear and RAID0 as metadata doesn't
1029 * record this size)
1030 */
1031 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
27a7b260 1032 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1033
27a7b260 1034 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N
1035 /* "this cannot possibly happen" ... */
1036 ret = -EINVAL;
1037
1da177e4
LT
1038 abort:
1039 return ret;
1040}
1041
1042/*
1043 * validate_super for 0.90.0
1044 */
fd01b88c 1045static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1046{
1047 mdp_disk_t *desc;
65a06f06 1048 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1049 __u64 ev1 = md_event(sb);
1da177e4 1050
41158c7e 1051 rdev->raid_disk = -1;
c5d79adb
N
1052 clear_bit(Faulty, &rdev->flags);
1053 clear_bit(In_sync, &rdev->flags);
8313b8e5 1054 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1055 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1056
1da177e4
LT
1057 if (mddev->raid_disks == 0) {
1058 mddev->major_version = 0;
1059 mddev->minor_version = sb->minor_version;
1060 mddev->patch_version = sb->patch_version;
e691063a 1061 mddev->external = 0;
9d8f0363 1062 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT
1063 mddev->ctime = sb->ctime;
1064 mddev->utime = sb->utime;
1065 mddev->level = sb->level;
d9d166c2 1066 mddev->clevel[0] = 0;
1da177e4
LT
1067 mddev->layout = sb->layout;
1068 mddev->raid_disks = sb->raid_disks;
27a7b260 1069 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1070 mddev->events = ev1;
c3d9714e 1071 mddev->bitmap_info.offset = 0;
6409bb05
N
1072 mddev->bitmap_info.space = 0;
1073 /* bitmap can use 60 K after the 4K superblocks */
c3d9714e 1074 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 1075 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
2c810cdd 1076 mddev->reshape_backwards = 0;
1da177e4 1077
f6705578
N
1078 if (mddev->minor_version >= 91) {
1079 mddev->reshape_position = sb->reshape_position;
1080 mddev->delta_disks = sb->delta_disks;
1081 mddev->new_level = sb->new_level;
1082 mddev->new_layout = sb->new_layout;
664e7c41 1083 mddev->new_chunk_sectors = sb->new_chunk >> 9;
2c810cdd
N
1084 if (mddev->delta_disks < 0)
1085 mddev->reshape_backwards = 1;
f6705578
N
1086 } else {
1087 mddev->reshape_position = MaxSector;
1088 mddev->delta_disks = 0;
1089 mddev->new_level = mddev->level;
1090 mddev->new_layout = mddev->layout;
664e7c41 1091 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N
1092 }
1093
1da177e4
LT
1094 if (sb->state & (1<<MD_SB_CLEAN))
1095 mddev->recovery_cp = MaxSector;
1096 else {
f72ffdd6 1097 if (sb->events_hi == sb->cp_events_hi &&
1da177e4
LT
1098 sb->events_lo == sb->cp_events_lo) {
1099 mddev->recovery_cp = sb->recovery_cp;
1100 } else
1101 mddev->recovery_cp = 0;
1102 }
1103
1104 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1105 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1106 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1107 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1108
1109 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N
1110
1111 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
6409bb05 1112 mddev->bitmap_info.file == NULL) {
c3d9714e
N
1113 mddev->bitmap_info.offset =
1114 mddev->bitmap_info.default_offset;
6409bb05 1115 mddev->bitmap_info.space =
c9ad020f 1116 mddev->bitmap_info.default_space;
6409bb05 1117 }
a654b9d8 1118
41158c7e 1119 } else if (mddev->pers == NULL) {
be6800a7
N
1120 /* Insist on good event counter while assembling, except
1121 * for spares (which don't need an event count) */
1da177e4 1122 ++ev1;
be6800a7
N
1123 if (sb->disks[rdev->desc_nr].state & (
1124 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
f72ffdd6 1125 if (ev1 < mddev->events)
be6800a7 1126 return -EINVAL;
41158c7e
N
1127 } else if (mddev->bitmap) {
1128 /* if adding to array with a bitmap, then we can accept an
1129 * older device ... but not too old.
1130 */
41158c7e
N
1131 if (ev1 < mddev->bitmap->events_cleared)
1132 return 0;
8313b8e5
N
1133 if (ev1 < mddev->events)
1134 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N
1135 } else {
1136 if (ev1 < mddev->events)
1137 /* just a hot-add of a new device, leave raid_disk at -1 */
1138 return 0;
1139 }
41158c7e 1140
1da177e4 1141 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT
1142 desc = sb->disks + rdev->desc_nr;
1143
1144 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1145 set_bit(Faulty, &rdev->flags);
7c7546cc
N
1146 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1147 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1148 set_bit(In_sync, &rdev->flags);
1da177e4 1149 rdev->raid_disk = desc->raid_disk;
f466722c 1150 rdev->saved_raid_disk = desc->raid_disk;
0261cd9f
N
1151 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1152 /* active but not in sync implies recovery up to
1153 * reshape position. We don't know exactly where
1154 * that is, so set to zero for now */
1155 if (mddev->minor_version >= 91) {
1156 rdev->recovery_offset = 0;
1157 rdev->raid_disk = desc->raid_disk;
1158 }
1da177e4 1159 }
8ddf9efe
N
1160 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1161 set_bit(WriteMostly, &rdev->flags);
41158c7e 1162 } else /* MULTIPATH are always insync */
b2d444d7 1163 set_bit(In_sync, &rdev->flags);
1da177e4
LT
1164 return 0;
1165}
1166
1167/*
1168 * sync_super for 0.90.0
1169 */
fd01b88c 1170static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1171{
1172 mdp_super_t *sb;
3cb03002 1173 struct md_rdev *rdev2;
1da177e4 1174 int next_spare = mddev->raid_disks;
19133a42 1175
1da177e4
LT
1176 /* make rdev->sb match mddev data..
1177 *
1178 * 1/ zero out disks
1179 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1180 * 3/ any empty disks < next_spare become removed
1181 *
1182 * disks[0] gets initialised to REMOVED because
1183 * we cannot be sure from other fields if it has
1184 * been initialised or not.
1185 */
1186 int i;
1187 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1188
61181565
N
1189 rdev->sb_size = MD_SB_BYTES;
1190
65a06f06 1191 sb = page_address(rdev->sb_page);
1da177e4
LT
1192
1193 memset(sb, 0, sizeof(*sb));
1194
1195 sb->md_magic = MD_SB_MAGIC;
1196 sb->major_version = mddev->major_version;
1da177e4
LT
1197 sb->patch_version = mddev->patch_version;
1198 sb->gvalid_words = 0; /* ignored */
1199 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1200 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1201 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1202 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1203
1204 sb->ctime = mddev->ctime;
1205 sb->level = mddev->level;
58c0fed4 1206 sb->size = mddev->dev_sectors / 2;
1da177e4
LT
1207 sb->raid_disks = mddev->raid_disks;
1208 sb->md_minor = mddev->md_minor;
e691063a 1209 sb->not_persistent = 0;
1da177e4
LT
1210 sb->utime = mddev->utime;
1211 sb->state = 0;
1212 sb->events_hi = (mddev->events>>32);
1213 sb->events_lo = (u32)mddev->events;
1214
f6705578
N
1215 if (mddev->reshape_position == MaxSector)
1216 sb->minor_version = 90;
1217 else {
1218 sb->minor_version = 91;
1219 sb->reshape_position = mddev->reshape_position;
1220 sb->new_level = mddev->new_level;
1221 sb->delta_disks = mddev->delta_disks;
1222 sb->new_layout = mddev->new_layout;
664e7c41 1223 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N
1224 }
1225 mddev->minor_version = sb->minor_version;
1da177e4
LT
1226 if (mddev->in_sync)
1227 {
1228 sb->recovery_cp = mddev->recovery_cp;
1229 sb->cp_events_hi = (mddev->events>>32);
1230 sb->cp_events_lo = (u32)mddev->events;
1231 if (mddev->recovery_cp == MaxSector)
1232 sb->state = (1<< MD_SB_CLEAN);
1233 } else
1234 sb->recovery_cp = 0;
1235
1236 sb->layout = mddev->layout;
9d8f0363 1237 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1238
c3d9714e 1239 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N
1240 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1241
1da177e4 1242 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1243 rdev_for_each(rdev2, mddev) {
1da177e4 1244 mdp_disk_t *d;
86e6ffdd 1245 int desc_nr;
0261cd9f
N
1246 int is_active = test_bit(In_sync, &rdev2->flags);
1247
1248 if (rdev2->raid_disk >= 0 &&
1249 sb->minor_version >= 91)
1250 /* we have nowhere to store the recovery_offset,
1251 * but if it is not below the reshape_position,
1252 * we can piggy-back on that.
1253 */
1254 is_active = 1;
1255 if (rdev2->raid_disk < 0 ||
1256 test_bit(Faulty, &rdev2->flags))
1257 is_active = 0;
1258 if (is_active)
86e6ffdd 1259 desc_nr = rdev2->raid_disk;
1da177e4 1260 else
86e6ffdd 1261 desc_nr = next_spare++;
19133a42 1262 rdev2->desc_nr = desc_nr;
1da177e4
LT
1263 d = &sb->disks[rdev2->desc_nr];
1264 nr_disks++;
1265 d->number = rdev2->desc_nr;
1266 d->major = MAJOR(rdev2->bdev->bd_dev);
1267 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1268 if (is_active)
1da177e4
LT
1269 d->raid_disk = rdev2->raid_disk;
1270 else
1271 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1272 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1273 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1274 else if (is_active) {
1da177e4 1275 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N
1276 if (test_bit(In_sync, &rdev2->flags))
1277 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT
1278 active++;
1279 working++;
1280 } else {
1281 d->state = 0;
1282 spare++;
1283 working++;
1284 }
8ddf9efe
N
1285 if (test_bit(WriteMostly, &rdev2->flags))
1286 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1287 }
1da177e4
LT
1288 /* now set the "removed" and "faulty" bits on any missing devices */
1289 for (i=0 ; i < mddev->raid_disks ; i++) {
1290 mdp_disk_t *d = &sb->disks[i];
1291 if (d->state == 0 && d->number == 0) {
1292 d->number = i;
1293 d->raid_disk = i;
1294 d->state = (1<<MD_DISK_REMOVED);
1295 d->state |= (1<<MD_DISK_FAULTY);
1296 failed++;
1297 }
1298 }
1299 sb->nr_disks = nr_disks;
1300 sb->active_disks = active;
1301 sb->working_disks = working;
1302 sb->failed_disks = failed;
1303 sb->spare_disks = spare;
1304
1305 sb->this_disk = sb->disks[rdev->desc_nr];
1306 sb->sb_csum = calc_sb_csum(sb);
1307}
1308
0cd17fec
CW
1309/*
1310 * rdev_size_change for 0.90.0
1311 */
1312static unsigned long long
3cb03002 1313super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1314{
58c0fed4 1315 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1316 return 0; /* component must fit device */
c3d9714e 1317 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1318 return 0; /* can't move bitmap */
57b2caa3 1319 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN
1320 if (!num_sectors || num_sectors > rdev->sb_start)
1321 num_sectors = rdev->sb_start;
27a7b260
N
1322 /* Limit to 4TB as metadata cannot record more than that.
1323 * 4TB == 2^32 KB, or 2*2^32 sectors.
1324 */
667a5313 1325 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
27a7b260 1326 num_sectors = (2ULL << 32) - 2;
0f420358 1327 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW
1328 rdev->sb_page);
1329 md_super_wait(rdev->mddev);
c26a44ed 1330 return num_sectors;
0cd17fec
CW
1331}
1332
c6563a8c
N
1333static int
1334super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1335{
1336 /* non-zero offset changes not possible with v0.90 */
1337 return new_offset == 0;
1338}
0cd17fec 1339
1da177e4
LT
1340/*
1341 * version 1 superblock
1342 */
1343
f72ffdd6 1344static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1da177e4 1345{
1c05b4bc
N
1346 __le32 disk_csum;
1347 u32 csum;
1da177e4
LT
1348 unsigned long long newcsum;
1349 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1350 __le32 *isuper = (__le32*)sb;
1da177e4
LT
1351
1352 disk_csum = sb->sb_csum;
1353 sb->sb_csum = 0;
1354 newcsum = 0;
1f3c9907 1355 for (; size >= 4; size -= 4)
1da177e4
LT
1356 newcsum += le32_to_cpu(*isuper++);
1357
1358 if (size == 2)
1c05b4bc 1359 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT
1360
1361 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1362 sb->sb_csum = disk_csum;
1363 return cpu_to_le32(csum);
1364}
1365
2699b672
N
1366static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1367 int acknowledged);
3cb03002 1368static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT
1369{
1370 struct mdp_superblock_1 *sb;
1371 int ret;
0f420358 1372 sector_t sb_start;
c6563a8c 1373 sector_t sectors;
1da177e4 1374 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1375 int bmask;
1da177e4
LT
1376
1377 /*
0f420358 1378 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT
1379 * It is always aligned to a 4K boundary and
1380 * depeding on minor_version, it can be:
1381 * 0: At least 8K, but less than 12K, from end of device
1382 * 1: At start of device
1383 * 2: 4K from start of device.
1384 */
1385 switch(minor_version) {
1386 case 0:
77304d2a 1387 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN
1388 sb_start -= 8*2;
1389 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT
1390 break;
1391 case 1:
0f420358 1392 sb_start = 0;
1da177e4
LT
1393 break;
1394 case 2:
0f420358 1395 sb_start = 8;
1da177e4
LT
1396 break;
1397 default:
1398 return -EINVAL;
1399 }
0f420358 1400 rdev->sb_start = sb_start;
1da177e4 1401
0002b271
N
1402 /* superblock is rarely larger than 1K, but it can be larger,
1403 * and it is safe to read 4k, so we do that
1404 */
1405 ret = read_disk_sb(rdev, 4096);
1da177e4
LT
1406 if (ret) return ret;
1407
65a06f06 1408 sb = page_address(rdev->sb_page);
1da177e4
LT
1409
1410 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1411 sb->major_version != cpu_to_le32(1) ||
1412 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1413 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1414 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT
1415 return -EINVAL;
1416
1417 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1418 printk("md: invalid superblock checksum on %s\n",
1419 bdevname(rdev->bdev,b));
1420 return -EINVAL;
1421 }
1422 if (le64_to_cpu(sb->data_size) < 10) {
1423 printk("md: data_size too small on %s\n",
1424 bdevname(rdev->bdev,b));
1425 return -EINVAL;
1426 }
c6563a8c
N
1427 if (sb->pad0 ||
1428 sb->pad3[0] ||
1429 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1430 /* Some padding is non-zero, might be a new feature */
1431 return -EINVAL;
e11e93fa 1432
1da177e4
LT
1433 rdev->preferred_minor = 0xffff;
1434 rdev->data_offset = le64_to_cpu(sb->data_offset);
c6563a8c
N
1435 rdev->new_data_offset = rdev->data_offset;
1436 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1437 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1438 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
4dbcdc75 1439 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1440
0002b271 1441 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1442 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1443 if (rdev->sb_size & bmask)
a1801f85
N
1444 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1445
1446 if (minor_version
0f420358 1447 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1448 return -EINVAL;
c6563a8c
N
1449 if (minor_version
1450 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1451 return -EINVAL;
0002b271 1452
31b65a0d
N
1453 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1454 rdev->desc_nr = -1;
1455 else
1456 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1457
2699b672
N
1458 if (!rdev->bb_page) {
1459 rdev->bb_page = alloc_page(GFP_KERNEL);
1460 if (!rdev->bb_page)
1461 return -ENOMEM;
1462 }
1463 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1464 rdev->badblocks.count == 0) {
1465 /* need to load the bad block list.
1466 * Currently we limit it to one page.
1467 */
1468 s32 offset;
1469 sector_t bb_sector;
1470 u64 *bbp;
1471 int i;
1472 int sectors = le16_to_cpu(sb->bblog_size);
1473 if (sectors > (PAGE_SIZE / 512))
1474 return -EINVAL;
1475 offset = le32_to_cpu(sb->bblog_offset);
1476 if (offset == 0)
1477 return -EINVAL;
1478 bb_sector = (long long)offset;
1479 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1480 rdev->bb_page, READ, true))
1481 return -EIO;
1482 bbp = (u64 *)page_address(rdev->bb_page);
1483 rdev->badblocks.shift = sb->bblog_shift;
1484 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1485 u64 bb = le64_to_cpu(*bbp);
1486 int count = bb & (0x3ff);
1487 u64 sector = bb >> 10;
1488 sector <<= sb->bblog_shift;
1489 count <<= sb->bblog_shift;
1490 if (bb + 1 == 0)
1491 break;
1492 if (md_set_badblocks(&rdev->badblocks,
1493 sector, count, 1) == 0)
1494 return -EINVAL;
1495 }
486adf72
N
1496 } else if (sb->bblog_offset != 0)
1497 rdev->badblocks.shift = 0;
2699b672 1498
9a7b2b0f 1499 if (!refdev) {
8ed75463 1500 ret = 1;
9a7b2b0f 1501 } else {
1da177e4 1502 __u64 ev1, ev2;
65a06f06 1503 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT
1504
1505 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1506 sb->level != refsb->level ||
1507 sb->layout != refsb->layout ||
1508 sb->chunksize != refsb->chunksize) {
1509 printk(KERN_WARNING "md: %s has strangely different"
1510 " superblock to %s\n",
1511 bdevname(rdev->bdev,b),
1512 bdevname(refdev->bdev,b2));
1513 return -EINVAL;
1514 }
1515 ev1 = le64_to_cpu(sb->events);
1516 ev2 = le64_to_cpu(refsb->events);
1517
1518 if (ev1 > ev2)
8ed75463
N
1519 ret = 1;
1520 else
1521 ret = 0;
1da177e4 1522 }
c6563a8c
N
1523 if (minor_version) {
1524 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1525 sectors -= rdev->data_offset;
1526 } else
1527 sectors = rdev->sb_start;
1528 if (sectors < le64_to_cpu(sb->data_size))
1da177e4 1529 return -EINVAL;
dd8ac336 1530 rdev->sectors = le64_to_cpu(sb->data_size);
8ed75463 1531 return ret;
1da177e4
LT
1532}
1533
fd01b88c 1534static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1535{
65a06f06 1536 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1537 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1538
41158c7e 1539 rdev->raid_disk = -1;
c5d79adb
N
1540 clear_bit(Faulty, &rdev->flags);
1541 clear_bit(In_sync, &rdev->flags);
8313b8e5 1542 clear_bit(Bitmap_sync, &rdev->flags);
c5d79adb 1543 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1544
1da177e4
LT
1545 if (mddev->raid_disks == 0) {
1546 mddev->major_version = 1;
1547 mddev->patch_version = 0;
e691063a 1548 mddev->external = 0;
9d8f0363 1549 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT
1550 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1551 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1552 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1553 mddev->clevel[0] = 0;
1da177e4
LT
1554 mddev->layout = le32_to_cpu(sb->layout);
1555 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1556 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1557 mddev->events = ev1;
c3d9714e 1558 mddev->bitmap_info.offset = 0;
6409bb05
N
1559 mddev->bitmap_info.space = 0;
1560 /* Default location for bitmap is 1K after superblock
1561 * using 3K - total of 4K
1562 */
c3d9714e 1563 mddev->bitmap_info.default_offset = 1024 >> 9;
6409bb05 1564 mddev->bitmap_info.default_space = (4096-1024) >> 9;
2c810cdd
N
1565 mddev->reshape_backwards = 0;
1566
1da177e4
LT
1567 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1568 memcpy(mddev->uuid, sb->set_uuid, 16);
1569
1570 mddev->max_disks = (4096-256)/2;
a654b9d8 1571
71c0805c 1572 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
6409bb05 1573 mddev->bitmap_info.file == NULL) {
c3d9714e
N
1574 mddev->bitmap_info.offset =
1575 (__s32)le32_to_cpu(sb->bitmap_offset);
6409bb05
N
1576 /* Metadata doesn't record how much space is available.
1577 * For 1.0, we assume we can use up to the superblock
1578 * if before, else to 4K beyond superblock.
1579 * For others, assume no change is possible.
1580 */
1581 if (mddev->minor_version > 0)
1582 mddev->bitmap_info.space = 0;
1583 else if (mddev->bitmap_info.offset > 0)
1584 mddev->bitmap_info.space =
1585 8 - mddev->bitmap_info.offset;
1586 else
1587 mddev->bitmap_info.space =
1588 -mddev->bitmap_info.offset;
1589 }
e11e93fa 1590
f6705578
N
1591 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1592 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1593 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1594 mddev->new_level = le32_to_cpu(sb->new_level);
1595 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1596 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
2c810cdd
N
1597 if (mddev->delta_disks < 0 ||
1598 (mddev->delta_disks == 0 &&
1599 (le32_to_cpu(sb->feature_map)
1600 & MD_FEATURE_RESHAPE_BACKWARDS)))
1601 mddev->reshape_backwards = 1;
f6705578
N
1602 } else {
1603 mddev->reshape_position = MaxSector;
1604 mddev->delta_disks = 0;
1605 mddev->new_level = mddev->level;
1606 mddev->new_layout = mddev->layout;
664e7c41 1607 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N
1608 }
1609
41158c7e 1610 } else if (mddev->pers == NULL) {
be6800a7
N
1611 /* Insist of good event counter while assembling, except for
1612 * spares (which don't need an event count) */
1da177e4 1613 ++ev1;
be6800a7
N
1614 if (rdev->desc_nr >= 0 &&
1615 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1616 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1617 if (ev1 < mddev->events)
1618 return -EINVAL;
41158c7e
N
1619 } else if (mddev->bitmap) {
1620 /* If adding to array with a bitmap, then we can accept an
1621 * older device, but not too old.
1622 */
41158c7e
N
1623 if (ev1 < mddev->bitmap->events_cleared)
1624 return 0;
8313b8e5
N
1625 if (ev1 < mddev->events)
1626 set_bit(Bitmap_sync, &rdev->flags);
07d84d10
N
1627 } else {
1628 if (ev1 < mddev->events)
1629 /* just a hot-add of a new device, leave raid_disk at -1 */
1630 return 0;
1631 }
1da177e4
LT
1632 if (mddev->level != LEVEL_MULTIPATH) {
1633 int role;
3673f305
N
1634 if (rdev->desc_nr < 0 ||
1635 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1636 role = 0xffff;
1637 rdev->desc_nr = -1;
1638 } else
1639 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT
1640 switch(role) {
1641 case 0xffff: /* spare */
1da177e4
LT
1642 break;
1643 case 0xfffe: /* faulty */
b2d444d7 1644 set_bit(Faulty, &rdev->flags);
1da177e4
LT
1645 break;
1646 default:
f466722c 1647 rdev->saved_raid_disk = role;
5fd6c1dc 1648 if ((le32_to_cpu(sb->feature_map) &
f466722c 1649 MD_FEATURE_RECOVERY_OFFSET)) {
5fd6c1dc 1650 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
f466722c
N
1651 if (!(le32_to_cpu(sb->feature_map) &
1652 MD_FEATURE_RECOVERY_BITMAP))
1653 rdev->saved_raid_disk = -1;
1654 } else
5fd6c1dc 1655 set_bit(In_sync, &rdev->flags);
1da177e4
LT
1656 rdev->raid_disk = role;
1657 break;
1658 }
8ddf9efe
N
1659 if (sb->devflags & WriteMostly1)
1660 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N
1661 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1662 set_bit(Replacement, &rdev->flags);
41158c7e 1663 } else /* MULTIPATH are always insync */
b2d444d7 1664 set_bit(In_sync, &rdev->flags);
41158c7e 1665
1da177e4
LT
1666 return 0;
1667}
1668
fd01b88c 1669static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
1670{
1671 struct mdp_superblock_1 *sb;
3cb03002 1672 struct md_rdev *rdev2;
1da177e4
LT
1673 int max_dev, i;
1674 /* make rdev->sb match mddev and rdev data. */
1675
65a06f06 1676 sb = page_address(rdev->sb_page);
1da177e4
LT
1677
1678 sb->feature_map = 0;
1679 sb->pad0 = 0;
5fd6c1dc 1680 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT
1681 memset(sb->pad3, 0, sizeof(sb->pad3));
1682
1683 sb->utime = cpu_to_le64((__u64)mddev->utime);
1684 sb->events = cpu_to_le64(mddev->events);
1685 if (mddev->in_sync)
1686 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1687 else
1688 sb->resync_offset = cpu_to_le64(0);
1689
1c05b4bc 1690 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1691
f0ca340c 1692 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1693 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1694 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N
1695 sb->level = cpu_to_le32(mddev->level);
1696 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1697
aeb9b211
N
1698 if (test_bit(WriteMostly, &rdev->flags))
1699 sb->devflags |= WriteMostly1;
1700 else
1701 sb->devflags &= ~WriteMostly1;
c6563a8c
N
1702 sb->data_offset = cpu_to_le64(rdev->data_offset);
1703 sb->data_size = cpu_to_le64(rdev->sectors);
aeb9b211 1704
c3d9714e
N
1705 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1706 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1707 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1708 }
5fd6c1dc
N
1709
1710 if (rdev->raid_disk >= 0 &&
97e4f42d 1711 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N
1712 sb->feature_map |=
1713 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1714 sb->recovery_offset =
1715 cpu_to_le64(rdev->recovery_offset);
f466722c
N
1716 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1717 sb->feature_map |=
1718 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
5fd6c1dc 1719 }
2d78f8c4
N
1720 if (test_bit(Replacement, &rdev->flags))
1721 sb->feature_map |=
1722 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1723
f6705578
N
1724 if (mddev->reshape_position != MaxSector) {
1725 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1726 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1727 sb->new_layout = cpu_to_le32(mddev->new_layout);
1728 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1729 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1730 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
2c810cdd
N
1731 if (mddev->delta_disks == 0 &&
1732 mddev->reshape_backwards)
1733 sb->feature_map
1734 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
c6563a8c
N
1735 if (rdev->new_data_offset != rdev->data_offset) {
1736 sb->feature_map
1737 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1738 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1739 - rdev->data_offset));
1740 }
f6705578 1741 }
a654b9d8 1742
2699b672
N
1743 if (rdev->badblocks.count == 0)
1744 /* Nothing to do for bad blocks*/ ;
1745 else if (sb->bblog_offset == 0)
1746 /* Cannot record bad blocks on this device */
1747 md_error(mddev, rdev);
1748 else {
1749 struct badblocks *bb = &rdev->badblocks;
1750 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1751 u64 *p = bb->page;
1752 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1753 if (bb->changed) {
1754 unsigned seq;
1755
1756retry:
1757 seq = read_seqbegin(&bb->lock);
1758
1759 memset(bbp, 0xff, PAGE_SIZE);
1760
1761 for (i = 0 ; i < bb->count ; i++) {
35f9ac2d 1762 u64 internal_bb = p[i];
2699b672
N
1763 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1764 | BB_LEN(internal_bb));
35f9ac2d 1765 bbp[i] = cpu_to_le64(store_bb);
2699b672 1766 }
d0962936 1767 bb->changed = 0;
2699b672
N
1768 if (read_seqretry(&bb->lock, seq))
1769 goto retry;
1770
1771 bb->sector = (rdev->sb_start +
1772 (int)le32_to_cpu(sb->bblog_offset));
1773 bb->size = le16_to_cpu(sb->bblog_size);
2699b672
N
1774 }
1775 }
1776
1da177e4 1777 max_dev = 0;
dafb20fa 1778 rdev_for_each(rdev2, mddev)
1da177e4
LT
1779 if (rdev2->desc_nr+1 > max_dev)
1780 max_dev = rdev2->desc_nr+1;
a778b73f 1781
70471daf
N
1782 if (max_dev > le32_to_cpu(sb->max_dev)) {
1783 int bmask;
a778b73f 1784 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N
1785 rdev->sb_size = max_dev * 2 + 256;
1786 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1787 if (rdev->sb_size & bmask)
1788 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N
1789 } else
1790 max_dev = le32_to_cpu(sb->max_dev);
1791
1da177e4
LT
1792 for (i=0; i<max_dev;i++)
1793 sb->dev_roles[i] = cpu_to_le16(0xfffe);
f72ffdd6 1794
dafb20fa 1795 rdev_for_each(rdev2, mddev) {
1da177e4 1796 i = rdev2->desc_nr;
b2d444d7 1797 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1798 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1799 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1800 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1801 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1802 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT
1803 else
1804 sb->dev_roles[i] = cpu_to_le16(0xffff);
1805 }
1806
1da177e4
LT
1807 sb->sb_csum = calc_sb_1_csum(sb);
1808}
1809
0cd17fec 1810static unsigned long long
3cb03002 1811super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW
1812{
1813 struct mdp_superblock_1 *sb;
15f4a5fd 1814 sector_t max_sectors;
58c0fed4 1815 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1816 return 0; /* component must fit device */
c6563a8c
N
1817 if (rdev->data_offset != rdev->new_data_offset)
1818 return 0; /* too confusing */
0f420358 1819 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1820 /* minor versions 1 and 2; superblock before data */
77304d2a 1821 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN
1822 max_sectors -= rdev->data_offset;
1823 if (!num_sectors || num_sectors > max_sectors)
1824 num_sectors = max_sectors;
c3d9714e 1825 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW
1826 /* minor version 0 with bitmap we can't move */
1827 return 0;
1828 } else {
1829 /* minor version 0; superblock after data */
0f420358 1830 sector_t sb_start;
77304d2a 1831 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1832 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1833 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN
1834 if (!num_sectors || num_sectors > max_sectors)
1835 num_sectors = max_sectors;
0f420358 1836 rdev->sb_start = sb_start;
0cd17fec 1837 }
65a06f06 1838 sb = page_address(rdev->sb_page);
15f4a5fd 1839 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1840 sb->super_offset = rdev->sb_start;
0cd17fec 1841 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1842 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW
1843 rdev->sb_page);
1844 md_super_wait(rdev->mddev);
c26a44ed 1845 return num_sectors;
c6563a8c
N
1846
1847}
1848
1849static int
1850super_1_allow_new_offset(struct md_rdev *rdev,
1851 unsigned long long new_offset)
1852{
1853 /* All necessary checks on new >= old have been done */
1854 struct bitmap *bitmap;
1855 if (new_offset >= rdev->data_offset)
1856 return 1;
1857
1858 /* with 1.0 metadata, there is no metadata to tread on
1859 * so we can always move back */
1860 if (rdev->mddev->minor_version == 0)
1861 return 1;
1862
1863 /* otherwise we must be sure not to step on
1864 * any metadata, so stay:
1865 * 36K beyond start of superblock
1866 * beyond end of badblocks
1867 * beyond write-intent bitmap
1868 */
1869 if (rdev->sb_start + (32+4)*2 > new_offset)
1870 return 0;
1871 bitmap = rdev->mddev->bitmap;
1872 if (bitmap && !rdev->mddev->bitmap_info.file &&
1873 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1ec885cd 1874 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
c6563a8c
N
1875 return 0;
1876 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1877 return 0;
1878
1879 return 1;
0cd17fec 1880}
1da177e4 1881
75c96f85 1882static struct super_type super_types[] = {
1da177e4
LT
1883 [0] = {
1884 .name = "0.90.0",
1885 .owner = THIS_MODULE,
0cd17fec
CW
1886 .load_super = super_90_load,
1887 .validate_super = super_90_validate,
1888 .sync_super = super_90_sync,
1889 .rdev_size_change = super_90_rdev_size_change,
c6563a8c 1890 .allow_new_offset = super_90_allow_new_offset,
1da177e4
LT
1891 },
1892 [1] = {
1893 .name = "md-1",
1894 .owner = THIS_MODULE,
0cd17fec
CW
1895 .load_super = super_1_load,
1896 .validate_super = super_1_validate,
1897 .sync_super = super_1_sync,
1898 .rdev_size_change = super_1_rdev_size_change,
c6563a8c 1899 .allow_new_offset = super_1_allow_new_offset,
1da177e4
LT
1900 },
1901};
1da177e4 1902
fd01b88c 1903static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB
1904{
1905 if (mddev->sync_super) {
1906 mddev->sync_super(mddev, rdev);
1907 return;
1908 }
1909
1910 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1911
1912 super_types[mddev->major_version].sync_super(mddev, rdev);
1913}
1914
fd01b88c 1915static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1916{
3cb03002 1917 struct md_rdev *rdev, *rdev2;
1da177e4 1918
4b80991c
N
1919 rcu_read_lock();
1920 rdev_for_each_rcu(rdev, mddev1)
1921 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1922 if (rdev->bdev->bd_contains ==
4b80991c
N
1923 rdev2->bdev->bd_contains) {
1924 rcu_read_unlock();
7dd5e7c3 1925 return 1;
4b80991c
N
1926 }
1927 rcu_read_unlock();
1da177e4
LT
1928 return 0;
1929}
1930
1931static LIST_HEAD(pending_raid_disks);
1932
ac5e7113
AN
1933/*
1934 * Try to register data integrity profile for an mddev
1935 *
1936 * This is called when an array is started and after a disk has been kicked
1937 * from the array. It only succeeds if all working and active component devices
1938 * are integrity capable with matching profiles.
1939 */
fd01b88c 1940int md_integrity_register(struct mddev *mddev)
ac5e7113 1941{
3cb03002 1942 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN
1943
1944 if (list_empty(&mddev->disks))
1945 return 0; /* nothing to do */
629acb6a
JB
1946 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1947 return 0; /* shouldn't register, or already is */
dafb20fa 1948 rdev_for_each(rdev, mddev) {
ac5e7113
AN
1949 /* skip spares and non-functional disks */
1950 if (test_bit(Faulty, &rdev->flags))
1951 continue;
1952 if (rdev->raid_disk < 0)
1953 continue;
ac5e7113
AN
1954 if (!reference) {
1955 /* Use the first rdev as the reference */
1956 reference = rdev;
1957 continue;
1958 }
1959 /* does this rdev's profile match the reference profile? */
1960 if (blk_integrity_compare(reference->bdev->bd_disk,
1961 rdev->bdev->bd_disk) < 0)
1962 return -EINVAL;
1963 }
89078d57
MP
1964 if (!reference || !bdev_get_integrity(reference->bdev))
1965 return 0;
ac5e7113
AN
1966 /*
1967 * All component devices are integrity capable and have matching
1968 * profiles, register the common profile for the md device.
1969 */
1970 if (blk_integrity_register(mddev->gendisk,
1971 bdev_get_integrity(reference->bdev)) != 0) {
1972 printk(KERN_ERR "md: failed to register integrity for %s\n",
1973 mdname(mddev));
1974 return -EINVAL;
1975 }
a91a2785
MP
1976 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1977 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1978 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1979 mdname(mddev));
1980 return -EINVAL;
1981 }
ac5e7113
AN
1982 return 0;
1983}
1984EXPORT_SYMBOL(md_integrity_register);
1985
1986/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1987void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1988{
2863b9eb
JB
1989 struct blk_integrity *bi_rdev;
1990 struct blk_integrity *bi_mddev;
1991
1992 if (!mddev->gendisk)
1993 return;
1994
1995 bi_rdev = bdev_get_integrity(rdev->bdev);
1996 bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1997
ac5e7113 1998 if (!bi_mddev) /* nothing to do */
3f9d99c1 1999 return;
ac5e7113 2000 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 2001 return;
ac5e7113
AN
2002 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2003 rdev->bdev->bd_disk) >= 0)
2004 return;
2005 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2006 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2007}
ac5e7113 2008EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2009
f72ffdd6 2010static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
1da177e4 2011{
7dd5e7c3 2012 char b[BDEVNAME_SIZE];
f637b9f9 2013 struct kobject *ko;
1edf80d3 2014 char *s;
5e55e2f5 2015 int err;
1da177e4
LT
2016
2017 if (rdev->mddev) {
2018 MD_BUG();
2019 return -EINVAL;
2020 }
11e2ede0
DW
2021
2022 /* prevent duplicates */
2023 if (find_rdev(mddev, rdev->bdev->bd_dev))
2024 return -EEXIST;
2025
dd8ac336
AN
2026 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2027 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2028 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N
2029 if (mddev->pers) {
2030 /* Cannot change size, so fail
2031 * If mddev->level <= 0, then we don't care
2032 * about aligning sizes (e.g. linear)
2033 */
2034 if (mddev->level > 0)
2035 return -ENOSPC;
2036 } else
dd8ac336 2037 mddev->dev_sectors = rdev->sectors;
2bf071bf 2038 }
1da177e4
LT
2039
2040 /* Verify rdev->desc_nr is unique.
2041 * If it is -1, assign a free number, else
2042 * check number is not in use
2043 */
4878e9eb 2044 rcu_read_lock();
1da177e4
LT
2045 if (rdev->desc_nr < 0) {
2046 int choice = 0;
4878e9eb
N
2047 if (mddev->pers)
2048 choice = mddev->raid_disks;
2049 while (find_rdev_nr_rcu(mddev, choice))
1da177e4
LT
2050 choice++;
2051 rdev->desc_nr = choice;
2052 } else {
4878e9eb
N
2053 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2054 rcu_read_unlock();
1da177e4 2055 return -EBUSY;
4878e9eb 2056 }
1da177e4 2057 }
4878e9eb 2058 rcu_read_unlock();
de01dfad
N
2059 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2060 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2061 mdname(mddev), mddev->max_disks);
2062 return -EBUSY;
2063 }
19133a42 2064 bdevname(rdev->bdev,b);
649316b2 2065 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2066 *s = '!';
649316b2 2067
1da177e4 2068 rdev->mddev = mddev;
19133a42 2069 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2070
b2d6db58 2071 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2072 goto fail;
86e6ffdd 2073
0762b8bd 2074 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N
2075 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2076 /* failure here is OK */;
2077 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2078
4b80991c 2079 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2080 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N
2081
2082 /* May as well allow recovery to be retried once */
5389042f 2083 mddev->recovery_disabled++;
3f9d99c1 2084
1da177e4 2085 return 0;
5e55e2f5
N
2086
2087 fail:
2088 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2089 b, mdname(mddev));
2090 return err;
1da177e4
LT
2091}
2092
177a99b2 2093static void md_delayed_delete(struct work_struct *ws)
5792a285 2094{
3cb03002 2095 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2096 kobject_del(&rdev->kobj);
177a99b2 2097 kobject_put(&rdev->kobj);
5792a285
N
2098}
2099
f72ffdd6 2100static void unbind_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT
2101{
2102 char b[BDEVNAME_SIZE];
2103 if (!rdev->mddev) {
2104 MD_BUG();
2105 return;
2106 }
49731baa 2107 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2108 list_del_rcu(&rdev->same_set);
1da177e4
LT
2109 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2110 rdev->mddev = NULL;
86e6ffdd 2111 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N
2112 sysfs_put(rdev->sysfs_state);
2113 rdev->sysfs_state = NULL;
2230dfe4 2114 rdev->badblocks.count = 0;
5792a285 2115 /* We need to delay this, otherwise we can deadlock when
4b80991c
N
2116 * writing to 'remove' to "dev/state". We also need
2117 * to delay it due to rcu usage.
5792a285 2118 */
4b80991c 2119 synchronize_rcu();
177a99b2
N
2120 INIT_WORK(&rdev->del_work, md_delayed_delete);
2121 kobject_get(&rdev->kobj);
e804ac78 2122 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT
2123}
2124
2125/*
2126 * prevent the device from being mounted, repartitioned or
2127 * otherwise reused by a RAID array (or any other kernel
2128 * subsystem), by bd_claiming the device.
2129 */
3cb03002 2130static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT
2131{
2132 int err = 0;
2133 struct block_device *bdev;
2134 char b[BDEVNAME_SIZE];
2135
d4d77629 2136 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2137 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT
2138 if (IS_ERR(bdev)) {
2139 printk(KERN_ERR "md: could not open %s.\n",
2140 __bdevname(dev, b));
2141 return PTR_ERR(bdev);
2142 }
1da177e4
LT
2143 rdev->bdev = bdev;
2144 return err;
2145}
2146
3cb03002 2147static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT
2148{
2149 struct block_device *bdev = rdev->bdev;
2150 rdev->bdev = NULL;
2151 if (!bdev)
2152 MD_BUG();
e525fd89 2153 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT
2154}
2155
2156void md_autodetect_dev(dev_t dev);
2157
f72ffdd6 2158static void export_rdev(struct md_rdev *rdev)
1da177e4
LT
2159{
2160 char b[BDEVNAME_SIZE];
2161 printk(KERN_INFO "md: export_rdev(%s)\n",
2162 bdevname(rdev->bdev,b));
2163 if (rdev->mddev)
2164 MD_BUG();
545c8795 2165 md_rdev_clear(rdev);
1da177e4 2166#ifndef MODULE
d0fae18f
N
2167 if (test_bit(AutoDetected, &rdev->flags))
2168 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT
2169#endif
2170 unlock_rdev(rdev);
86e6ffdd 2171 kobject_put(&rdev->kobj);
1da177e4
LT
2172}
2173
f72ffdd6 2174static void kick_rdev_from_array(struct md_rdev *rdev)
1da177e4
LT
2175{
2176 unbind_rdev_from_array(rdev);
2177 export_rdev(rdev);
2178}
2179
fd01b88c 2180static void export_array(struct mddev *mddev)
1da177e4 2181{
0638bb0e 2182 struct md_rdev *rdev;
1da177e4 2183
0638bb0e
N
2184 while (!list_empty(&mddev->disks)) {
2185 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2186 same_set);
1da177e4
LT
2187 kick_rdev_from_array(rdev);
2188 }
1da177e4
LT
2189 mddev->raid_disks = 0;
2190 mddev->major_version = 0;
2191}
2192
2193static void print_desc(mdp_disk_t *desc)
2194{
2195 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2196 desc->major,desc->minor,desc->raid_disk,desc->state);
2197}
2198
cd2ac932 2199static void print_sb_90(mdp_super_t *sb)
1da177e4
LT
2200{
2201 int i;
2202
f72ffdd6 2203 printk(KERN_INFO
1da177e4
LT
2204 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2205 sb->major_version, sb->minor_version, sb->patch_version,
2206 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2207 sb->ctime);
2208 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2209 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2210 sb->md_minor, sb->layout, sb->chunk_size);
2211 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2212 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2213 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2214 sb->failed_disks, sb->spare_disks,
2215 sb->sb_csum, (unsigned long)sb->events_lo);
2216
2217 printk(KERN_INFO);
2218 for (i = 0; i < MD_SB_DISKS; i++) {
2219 mdp_disk_t *desc;
2220
2221 desc = sb->disks + i;
2222 if (desc->number || desc->major || desc->minor ||
2223 desc->raid_disk || (desc->state && (desc->state != 4))) {
2224 printk(" D %2d: ", i);
2225 print_desc(desc);
2226 }
2227 }
2228 printk(KERN_INFO "md: THIS: ");
2229 print_desc(&sb->this_disk);
cd2ac932 2230}
1da177e4 2231
cd2ac932
CR
2232static void print_sb_1(struct mdp_superblock_1 *sb)
2233{
2234 __u8 *uuid;
2235
2236 uuid = sb->set_uuid;
ad361c98 2237 printk(KERN_INFO
7b75c2f8 2238 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2239 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR
2240 le32_to_cpu(sb->major_version),
2241 le32_to_cpu(sb->feature_map),
7b75c2f8 2242 uuid,
cd2ac932
CR
2243 sb->set_name,
2244 (unsigned long long)le64_to_cpu(sb->ctime)
2245 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2246
2247 uuid = sb->device_uuid;
ad361c98
JP
2248 printk(KERN_INFO
2249 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2250 " RO:%llu\n"
7b75c2f8 2251 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP
2252 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2253 "md: (MaxDev:%u) \n",
cd2ac932
CR
2254 le32_to_cpu(sb->level),
2255 (unsigned long long)le64_to_cpu(sb->size),
2256 le32_to_cpu(sb->raid_disks),
2257 le32_to_cpu(sb->layout),
2258 le32_to_cpu(sb->chunksize),
2259 (unsigned long long)le64_to_cpu(sb->data_offset),
2260 (unsigned long long)le64_to_cpu(sb->data_size),
2261 (unsigned long long)le64_to_cpu(sb->super_offset),
2262 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2263 le32_to_cpu(sb->dev_number),
7b75c2f8 2264 uuid,
cd2ac932
CR
2265 sb->devflags,
2266 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2267 (unsigned long long)le64_to_cpu(sb->events),
2268 (unsigned long long)le64_to_cpu(sb->resync_offset),
2269 le32_to_cpu(sb->sb_csum),
2270 le32_to_cpu(sb->max_dev)
2271 );
1da177e4
LT
2272}
2273
3cb03002 2274static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT
2275{
2276 char b[BDEVNAME_SIZE];
dd8ac336 2277 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
f72ffdd6
N
2278 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2279 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2280 rdev->desc_nr);
1da177e4 2281 if (rdev->sb_loaded) {
cd2ac932
CR
2282 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2283 switch (major_version) {
2284 case 0:
65a06f06 2285 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR
2286 break;
2287 case 1:
65a06f06 2288 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR
2289 break;
2290 }
1da177e4
LT
2291 } else
2292 printk(KERN_INFO "md: no rdev superblock!\n");
2293}
2294
5e56341d 2295static void md_print_devices(void)
1da177e4 2296{
159ec1fc 2297 struct list_head *tmp;
3cb03002 2298 struct md_rdev *rdev;
fd01b88c 2299 struct mddev *mddev;
1da177e4
LT
2300 char b[BDEVNAME_SIZE];
2301
2302 printk("\n");
2303 printk("md: **********************************\n");
2304 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2305 printk("md: **********************************\n");
29ac4aa3 2306 for_each_mddev(mddev, tmp) {
1da177e4 2307
32a7627c
N
2308 if (mddev->bitmap)
2309 bitmap_print_sb(mddev->bitmap);
2310 else
2311 printk("%s: ", mdname(mddev));
dafb20fa 2312 rdev_for_each(rdev, mddev)
1da177e4
LT
2313 printk("<%s>", bdevname(rdev->bdev,b));
2314 printk("\n");
2315
dafb20fa 2316 rdev_for_each(rdev, mddev)
cd2ac932 2317 print_rdev(rdev, mddev->major_version);
1da177e4
LT
2318 }
2319 printk("md: **********************************\n");
2320 printk("\n");
2321}
2322
f72ffdd6 2323static void sync_sbs(struct mddev *mddev, int nospares)
1da177e4 2324{
42543769
N
2325 /* Update each superblock (in-memory image), but
2326 * if we are allowed to, skip spares which already
2327 * have the right event counter, or have one earlier
2328 * (which would mean they aren't being marked as dirty
2329 * with the rest of the array)
2330 */
3cb03002 2331 struct md_rdev *rdev;
dafb20fa 2332 rdev_for_each(rdev, mddev) {
42543769
N
2333 if (rdev->sb_events == mddev->events ||
2334 (nospares &&
2335 rdev->raid_disk < 0 &&
42543769
N
2336 rdev->sb_events+1 == mddev->events)) {
2337 /* Don't update this superblock */
2338 rdev->sb_loaded = 2;
2339 } else {
076f968b 2340 sync_super(mddev, rdev);
42543769
N
2341 rdev->sb_loaded = 1;
2342 }
1da177e4
LT
2343 }
2344}
2345
f72ffdd6 2346static void md_update_sb(struct mddev *mddev, int force_change)
1da177e4 2347{
3cb03002 2348 struct md_rdev *rdev;
06d91a5f 2349 int sync_req;
42543769 2350 int nospares = 0;
2699b672 2351 int any_badblocks_changed = 0;
1da177e4 2352
d87f064f
N
2353 if (mddev->ro) {
2354 if (force_change)
2355 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2356 return;
2357 }
1da177e4 2358repeat:
3a3a5ddb 2359 /* First make sure individual recovery_offsets are correct */
dafb20fa 2360 rdev_for_each(rdev, mddev) {
3a3a5ddb
N
2361 if (rdev->raid_disk >= 0 &&
2362 mddev->delta_disks >= 0 &&
2363 !test_bit(In_sync, &rdev->flags) &&
2364 mddev->curr_resync_completed > rdev->recovery_offset)
2365 rdev->recovery_offset = mddev->curr_resync_completed;
2366
f72ffdd6 2367 }
bd52b746 2368 if (!mddev->persistent) {
070dc6dd 2369 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2370 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2371 if (!mddev->external) {
d97a41dc 2372 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2373 rdev_for_each(rdev, mddev) {
de393cde 2374 if (rdev->badblocks.changed) {
d0962936 2375 rdev->badblocks.changed = 0;
de393cde
N
2376 md_ack_all_badblocks(&rdev->badblocks);
2377 md_error(mddev, rdev);
2378 }
2379 clear_bit(Blocked, &rdev->flags);
2380 clear_bit(BlockedBadBlocks, &rdev->flags);
2381 wake_up(&rdev->blocked_wait);
2382 }
2383 }
3a3a5ddb
N
2384 wake_up(&mddev->sb_wait);
2385 return;
2386 }
2387
a9701a30 2388 spin_lock_irq(&mddev->write_lock);
84692195 2389
3a3a5ddb
N
2390 mddev->utime = get_seconds();
2391
850b2b42
N
2392 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2393 force_change = 1;
2394 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2395 /* just a clean<-> dirty transition, possibly leave spares alone,
2396 * though if events isn't the right even/odd, we will have to do
2397 * spares after all
2398 */
2399 nospares = 1;
2400 if (force_change)
2401 nospares = 0;
2402 if (mddev->degraded)
84692195
N
2403 /* If the array is degraded, then skipping spares is both
2404 * dangerous and fairly pointless.
2405 * Dangerous because a device that was removed from the array
2406 * might have a event_count that still looks up-to-date,
2407 * so it can be re-added without a resync.
2408 * Pointless because if there are any spares to skip,
2409 * then a recovery will happen and soon that array won't
2410 * be degraded any more and the spare can go back to sleep then.
2411 */
850b2b42 2412 nospares = 0;
84692195 2413
06d91a5f 2414 sync_req = mddev->in_sync;
42543769
N
2415
2416 /* If this is just a dirty<->clean transition, and the array is clean
2417 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2418 if (nospares
42543769 2419 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N
2420 && mddev->can_decrease_events
2421 && mddev->events != 1) {
42543769 2422 mddev->events--;
a8707c08
N
2423 mddev->can_decrease_events = 0;
2424 } else {
42543769
N
2425 /* otherwise we have to go forward and ... */
2426 mddev->events ++;
a8707c08 2427 mddev->can_decrease_events = nospares;
42543769 2428 }
1da177e4
LT
2429
2430 if (!mddev->events) {
2431 /*
2432 * oops, this 64-bit counter should never wrap.
2433 * Either we are in around ~1 trillion A.C., assuming
2434 * 1 reboot per second, or we have a bug:
2435 */
2436 MD_BUG();
2437 mddev->events --;
2438 }
2699b672 2439
dafb20fa 2440 rdev_for_each(rdev, mddev) {
2699b672
N
2441 if (rdev->badblocks.changed)
2442 any_badblocks_changed++;
de393cde
N
2443 if (test_bit(Faulty, &rdev->flags))
2444 set_bit(FaultRecorded, &rdev->flags);
2445 }
2699b672 2446
e691063a 2447 sync_sbs(mddev, nospares);
a9701a30 2448 spin_unlock_irq(&mddev->write_lock);
1da177e4 2449
36a4e1fe
N
2450 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2451 mdname(mddev), mddev->in_sync);
1da177e4 2452
4ad13663 2453 bitmap_update_sb(mddev->bitmap);
dafb20fa 2454 rdev_for_each(rdev, mddev) {
1da177e4 2455 char b[BDEVNAME_SIZE];
36a4e1fe 2456
42543769
N
2457 if (rdev->sb_loaded != 1)
2458 continue; /* no noise on spare devices */
1da177e4 2459
f466722c 2460 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2461 md_super_write(mddev,rdev,
0f420358 2462 rdev->sb_start, rdev->sb_size,
7bfa19f2 2463 rdev->sb_page);
36a4e1fe
N
2464 pr_debug("md: (write) %s's sb offset: %llu\n",
2465 bdevname(rdev->bdev, b),
2466 (unsigned long long)rdev->sb_start);
42543769 2467 rdev->sb_events = mddev->events;
2699b672
N
2468 if (rdev->badblocks.size) {
2469 md_super_write(mddev, rdev,
2470 rdev->badblocks.sector,
2471 rdev->badblocks.size << 9,
2472 rdev->bb_page);
2473 rdev->badblocks.size = 0;
2474 }
7bfa19f2 2475
f466722c 2476 } else
36a4e1fe
N
2477 pr_debug("md: %s (skipping faulty)\n",
2478 bdevname(rdev->bdev, b));
d70ed2e4 2479
7bfa19f2 2480 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT
2481 /* only need to write one superblock... */
2482 break;
2483 }
a9701a30 2484 md_super_wait(mddev);
850b2b42 2485 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2486
a9701a30 2487 spin_lock_irq(&mddev->write_lock);
850b2b42
N
2488 if (mddev->in_sync != sync_req ||
2489 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2490 /* have to write it out again */
a9701a30 2491 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N
2492 goto repeat;
2493 }
850b2b42 2494 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2495 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2496 wake_up(&mddev->sb_wait);
acb180b0
N
2497 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2498 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2499
dafb20fa 2500 rdev_for_each(rdev, mddev) {
de393cde
N
2501 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2502 clear_bit(Blocked, &rdev->flags);
2503
2504 if (any_badblocks_changed)
2699b672 2505 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N
2506 clear_bit(BlockedBadBlocks, &rdev->flags);
2507 wake_up(&rdev->blocked_wait);
2508 }
1da177e4
LT
2509}
2510
7f6ce769 2511/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N
2512 * We want to accept with case. For this we use cmd_match.
2513 */
2514static int cmd_match(const char *cmd, const char *str)
2515{
2516 /* See if cmd, written into a sysfs file, matches
2517 * str. They must either be the same, or cmd can
2518 * have a trailing newline
2519 */
2520 while (*cmd && *str && *cmd == *str) {
2521 cmd++;
2522 str++;
2523 }
2524 if (*cmd == '\n')
2525 cmd++;
2526 if (*str || *cmd)
2527 return 0;
2528 return 1;
2529}
2530
86e6ffdd
N
2531struct rdev_sysfs_entry {
2532 struct attribute attr;
3cb03002
N
2533 ssize_t (*show)(struct md_rdev *, char *);
2534 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N
2535};
2536
2537static ssize_t
3cb03002 2538state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N
2539{
2540 char *sep = "";
20a49ff6 2541 size_t len = 0;
86e6ffdd 2542
de393cde
N
2543 if (test_bit(Faulty, &rdev->flags) ||
2544 rdev->badblocks.unacked_exist) {
86e6ffdd
N
2545 len+= sprintf(page+len, "%sfaulty",sep);
2546 sep = ",";
2547 }
b2d444d7 2548 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N
2549 len += sprintf(page+len, "%sin_sync",sep);
2550 sep = ",";
2551 }
f655675b
N
2552 if (test_bit(WriteMostly, &rdev->flags)) {
2553 len += sprintf(page+len, "%swrite_mostly",sep);
2554 sep = ",";
2555 }
de393cde 2556 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N
2557 (rdev->badblocks.unacked_exist
2558 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW
2559 len += sprintf(page+len, "%sblocked", sep);
2560 sep = ",";
2561 }
b2d444d7
N
2562 if (!test_bit(Faulty, &rdev->flags) &&
2563 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N
2564 len += sprintf(page+len, "%sspare", sep);
2565 sep = ",";
2566 }
d7a9d443
N
2567 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2568 len += sprintf(page+len, "%swrite_error", sep);
2569 sep = ",";
2570 }
2d78f8c4
N
2571 if (test_bit(WantReplacement, &rdev->flags)) {
2572 len += sprintf(page+len, "%swant_replacement", sep);
2573 sep = ",";
2574 }
2575 if (test_bit(Replacement, &rdev->flags)) {
2576 len += sprintf(page+len, "%sreplacement", sep);
2577 sep = ",";
2578 }
2579
86e6ffdd
N
2580 return len+sprintf(page+len, "\n");
2581}
2582
45dc2de1 2583static ssize_t
3cb03002 2584state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N
2585{
2586 /* can write
de393cde 2587 * faulty - simulates an error
45dc2de1 2588 * remove - disconnects the device
f655675b
N
2589 * writemostly - sets write_mostly
2590 * -writemostly - clears write_mostly
de393cde
N
2591 * blocked - sets the Blocked flags
2592 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2593 * insync - sets Insync providing device isn't active
f466722c
N
2594 * -insync - clear Insync for a device with a slot assigned,
2595 * so that it gets rebuilt based on bitmap
d7a9d443
N
2596 * write_error - sets WriteErrorSeen
2597 * -write_error - clears WriteErrorSeen
45dc2de1
N
2598 */
2599 int err = -EINVAL;
2600 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2601 md_error(rdev->mddev, rdev);
5ef56c8f
N
2602 if (test_bit(Faulty, &rdev->flags))
2603 err = 0;
2604 else
2605 err = -EBUSY;
45dc2de1
N
2606 } else if (cmd_match(buf, "remove")) {
2607 if (rdev->raid_disk >= 0)
2608 err = -EBUSY;
2609 else {
fd01b88c 2610 struct mddev *mddev = rdev->mddev;
45dc2de1 2611 kick_rdev_from_array(rdev);
3f9d7b0d
N
2612 if (mddev->pers)
2613 md_update_sb(mddev, 1);
45dc2de1
N
2614 md_new_event(mddev);
2615 err = 0;
2616 }
f655675b
N
2617 } else if (cmd_match(buf, "writemostly")) {
2618 set_bit(WriteMostly, &rdev->flags);
2619 err = 0;
2620 } else if (cmd_match(buf, "-writemostly")) {
2621 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW
2622 err = 0;
2623 } else if (cmd_match(buf, "blocked")) {
2624 set_bit(Blocked, &rdev->flags);
2625 err = 0;
2626 } else if (cmd_match(buf, "-blocked")) {
de393cde 2627 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2628 rdev->badblocks.unacked_exist) {
de393cde
N
2629 /* metadata handler doesn't understand badblocks,
2630 * so we need to fail the device
2631 */
2632 md_error(rdev->mddev, rdev);
2633 }
6bfe0b49 2634 clear_bit(Blocked, &rdev->flags);
de393cde 2635 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW
2636 wake_up(&rdev->blocked_wait);
2637 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2638 md_wakeup_thread(rdev->mddev->thread);
2639
6d56e278
N
2640 err = 0;
2641 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2642 set_bit(In_sync, &rdev->flags);
f655675b 2643 err = 0;
f466722c 2644 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
e1960f8c
N
2645 if (rdev->mddev->pers == NULL) {
2646 clear_bit(In_sync, &rdev->flags);
2647 rdev->saved_raid_disk = rdev->raid_disk;
2648 rdev->raid_disk = -1;
2649 err = 0;
2650 }
d7a9d443
N
2651 } else if (cmd_match(buf, "write_error")) {
2652 set_bit(WriteErrorSeen, &rdev->flags);
2653 err = 0;
2654 } else if (cmd_match(buf, "-write_error")) {
2655 clear_bit(WriteErrorSeen, &rdev->flags);
2656 err = 0;
2d78f8c4
N
2657 } else if (cmd_match(buf, "want_replacement")) {
2658 /* Any non-spare device that is not a replacement can
2659 * become want_replacement at any time, but we then need to
2660 * check if recovery is needed.
2661 */
2662 if (rdev->raid_disk >= 0 &&
2663 !test_bit(Replacement, &rdev->flags))
2664 set_bit(WantReplacement, &rdev->flags);
2665 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2666 md_wakeup_thread(rdev->mddev->thread);
2667 err = 0;
2668 } else if (cmd_match(buf, "-want_replacement")) {
2669 /* Clearing 'want_replacement' is always allowed.
2670 * Once replacements starts it is too late though.
2671 */
2672 err = 0;
2673 clear_bit(WantReplacement, &rdev->flags);
2674 } else if (cmd_match(buf, "replacement")) {
2675 /* Can only set a device as a replacement when array has not
2676 * yet been started. Once running, replacement is automatic
2677 * from spares, or by assigning 'slot'.
2678 */
2679 if (rdev->mddev->pers)
2680 err = -EBUSY;
2681 else {
2682 set_bit(Replacement, &rdev->flags);
2683 err = 0;
2684 }
2685 } else if (cmd_match(buf, "-replacement")) {
2686 /* Similarly, can only clear Replacement before start */
2687 if (rdev->mddev->pers)
2688 err = -EBUSY;
2689 else {
2690 clear_bit(Replacement, &rdev->flags);
2691 err = 0;
2692 }
45dc2de1 2693 }
00bcb4ac
N
2694 if (!err)
2695 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N
2696 return err ? err : len;
2697}
80ca3a44
N
2698static struct rdev_sysfs_entry rdev_state =
2699__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2700
4dbcdc75 2701static ssize_t
3cb03002 2702errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N
2703{
2704 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2705}
2706
2707static ssize_t
3cb03002 2708errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N
2709{
2710 char *e;
2711 unsigned long n = simple_strtoul(buf, &e, 10);
2712 if (*buf && (*e == 0 || *e == '\n')) {
2713 atomic_set(&rdev->corrected_errors, n);
2714 return len;
2715 }
2716 return -EINVAL;
2717}
2718static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2719__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2720
014236d2 2721static ssize_t
3cb03002 2722slot_show(struct md_rdev *rdev, char *page)
014236d2
N
2723{
2724 if (rdev->raid_disk < 0)
2725 return sprintf(page, "none\n");
2726 else
2727 return sprintf(page, "%d\n", rdev->raid_disk);
2728}
2729
2730static ssize_t
3cb03002 2731slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N
2732{
2733 char *e;
c303da6d 2734 int err;
014236d2
N
2735 int slot = simple_strtoul(buf, &e, 10);
2736 if (strncmp(buf, "none", 4)==0)
2737 slot = -1;
2738 else if (e==buf || (*e && *e!= '\n'))
2739 return -EINVAL;
6c2fce2e 2740 if (rdev->mddev->pers && slot == -1) {
c303da6d
N
2741 /* Setting 'slot' on an active array requires also
2742 * updating the 'rd%d' link, and communicating
2743 * with the personality with ->hot_*_disk.
2744 * For now we only support removing
2745 * failed/spare devices. This normally happens automatically,
2746 * but not when the metadata is externally managed.
2747 */
c303da6d
N
2748 if (rdev->raid_disk == -1)
2749 return -EEXIST;
2750 /* personality does all needed checks */
01393f3d 2751 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d 2752 return -EINVAL;
746d3207
N
2753 clear_bit(Blocked, &rdev->flags);
2754 remove_and_add_spares(rdev->mddev, rdev);
2755 if (rdev->raid_disk >= 0)
2756 return -EBUSY;
c303da6d
N
2757 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2758 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2759 } else if (rdev->mddev->pers) {
6c2fce2e 2760 /* Activating a spare .. or possibly reactivating
6d56e278 2761 * if we ever get bitmaps working here.
6c2fce2e
NB
2762 */
2763
2764 if (rdev->raid_disk != -1)
2765 return -EBUSY;
2766
c6751b2b
N
2767 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2768 return -EBUSY;
2769
6c2fce2e
NB
2770 if (rdev->mddev->pers->hot_add_disk == NULL)
2771 return -EINVAL;
2772
ba1b41b6
N
2773 if (slot >= rdev->mddev->raid_disks &&
2774 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2775 return -ENOSPC;
2776
6c2fce2e
NB
2777 rdev->raid_disk = slot;
2778 if (test_bit(In_sync, &rdev->flags))
2779 rdev->saved_raid_disk = slot;
2780 else
2781 rdev->saved_raid_disk = -1;
d30519fc 2782 clear_bit(In_sync, &rdev->flags);
8313b8e5 2783 clear_bit(Bitmap_sync, &rdev->flags);
6c2fce2e
NB
2784 err = rdev->mddev->pers->
2785 hot_add_disk(rdev->mddev, rdev);
199050ea 2786 if (err) {
6c2fce2e 2787 rdev->raid_disk = -1;
6c2fce2e 2788 return err;
52664732 2789 } else
00bcb4ac 2790 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2791 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2792 /* failure here is OK */;
6c2fce2e 2793 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2794 } else {
ba1b41b6
N
2795 if (slot >= rdev->mddev->raid_disks &&
2796 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N
2797 return -ENOSPC;
2798 rdev->raid_disk = slot;
2799 /* assume it is working */
c5d79adb
N
2800 clear_bit(Faulty, &rdev->flags);
2801 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2802 set_bit(In_sync, &rdev->flags);
00bcb4ac 2803 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2804 }
014236d2
N
2805 return len;
2806}
2807
014236d2 2808static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2809__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2810
93c8cad0 2811static ssize_t
3cb03002 2812offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2813{
6961ece4 2814 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N
2815}
2816
2817static ssize_t
3cb03002 2818offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0 2819{
c6563a8c 2820 unsigned long long offset;
b29bebd6 2821 if (kstrtoull(buf, 10, &offset) < 0)
93c8cad0 2822 return -EINVAL;
8ed0a521 2823 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2824 return -EBUSY;
dd8ac336 2825 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N
2826 /* Must set offset before size, so overlap checks
2827 * can be sane */
2828 return -EBUSY;
93c8cad0 2829 rdev->data_offset = offset;
25f7fd47 2830 rdev->new_data_offset = offset;
93c8cad0
N
2831 return len;
2832}
2833
2834static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2835__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2836
c6563a8c
N
2837static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2838{
2839 return sprintf(page, "%llu\n",
2840 (unsigned long long)rdev->new_data_offset);
2841}
2842
2843static ssize_t new_offset_store(struct md_rdev *rdev,
2844 const char *buf, size_t len)
2845{
2846 unsigned long long new_offset;
2847 struct mddev *mddev = rdev->mddev;
2848
b29bebd6 2849 if (kstrtoull(buf, 10, &new_offset) < 0)
c6563a8c
N
2850 return -EINVAL;
2851
2852 if (mddev->sync_thread)
2853 return -EBUSY;
2854 if (new_offset == rdev->data_offset)
2855 /* reset is always permitted */
2856 ;
2857 else if (new_offset > rdev->data_offset) {
2858 /* must not push array size beyond rdev_sectors */
2859 if (new_offset - rdev->data_offset
2860 + mddev->dev_sectors > rdev->sectors)
2861 return -E2BIG;
2862 }
2863 /* Metadata worries about other space details. */
2864
2865 /* decreasing the offset is inconsistent with a backwards
2866 * reshape.
2867 */
2868 if (new_offset < rdev->data_offset &&
2869 mddev->reshape_backwards)
2870 return -EINVAL;
2871 /* Increasing offset is inconsistent with forwards
2872 * reshape. reshape_direction should be set to
2873 * 'backwards' first.
2874 */
2875 if (new_offset > rdev->data_offset &&
2876 !mddev->reshape_backwards)
2877 return -EINVAL;
2878
2879 if (mddev->pers && mddev->persistent &&
2880 !super_types[mddev->major_version]
2881 .allow_new_offset(rdev, new_offset))
2882 return -E2BIG;
2883 rdev->new_data_offset = new_offset;
2884 if (new_offset > rdev->data_offset)
2885 mddev->reshape_backwards = 1;
2886 else if (new_offset < rdev->data_offset)
2887 mddev->reshape_backwards = 0;
2888
2889 return len;
2890}
2891static struct rdev_sysfs_entry rdev_new_offset =
2892__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2893
83303b61 2894static ssize_t
3cb03002 2895rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2896{
dd8ac336 2897 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N
2898}
2899
c5d79adb
N
2900static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2901{
2902 /* check if two start/length pairs overlap */
2903 if (s1+l1 <= s2)
2904 return 0;
2905 if (s2+l2 <= s1)
2906 return 0;
2907 return 1;
2908}
2909
b522adcd
DW
2910static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2911{
2912 unsigned long long blocks;
2913 sector_t new;
2914
b29bebd6 2915 if (kstrtoull(buf, 10, &blocks) < 0)
b522adcd
DW
2916 return -EINVAL;
2917
2918 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2919 return -EINVAL; /* sector conversion overflow */
2920
2921 new = blocks * 2;
2922 if (new != blocks * 2)
2923 return -EINVAL; /* unsigned long long to sector_t overflow */
2924
2925 *sectors = new;
2926 return 0;
2927}
2928
83303b61 2929static ssize_t
3cb03002 2930rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2931{
fd01b88c 2932 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2933 sector_t oldsectors = rdev->sectors;
b522adcd 2934 sector_t sectors;
27c529bb 2935
b522adcd 2936 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2937 return -EINVAL;
c6563a8c
N
2938 if (rdev->data_offset != rdev->new_data_offset)
2939 return -EINVAL; /* too confusing */
0cd17fec 2940 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2941 if (my_mddev->persistent) {
dd8ac336
AN
2942 sectors = super_types[my_mddev->major_version].
2943 rdev_size_change(rdev, sectors);
2944 if (!sectors)
0cd17fec 2945 return -EBUSY;
dd8ac336 2946 } else if (!sectors)
77304d2a 2947 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2948 rdev->data_offset;
a6468539
N
2949 if (!my_mddev->pers->resize)
2950 /* Cannot change size for RAID0 or Linear etc */
2951 return -EINVAL;
0cd17fec 2952 }
dd8ac336 2953 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2954 return -EINVAL; /* component must fit device */
0cd17fec 2955
dd8ac336
AN
2956 rdev->sectors = sectors;
2957 if (sectors > oldsectors && my_mddev->external) {
8b1afc3d
N
2958 /* Need to check that all other rdevs with the same
2959 * ->bdev do not overlap. 'rcu' is sufficient to walk
2960 * the rdev lists safely.
2961 * This check does not provide a hard guarantee, it
2962 * just helps avoid dangerous mistakes.
c5d79adb 2963 */
fd01b88c 2964 struct mddev *mddev;
c5d79adb 2965 int overlap = 0;
159ec1fc 2966 struct list_head *tmp;
c5d79adb 2967
8b1afc3d 2968 rcu_read_lock();
29ac4aa3 2969 for_each_mddev(mddev, tmp) {
3cb03002 2970 struct md_rdev *rdev2;
c5d79adb 2971
dafb20fa 2972 rdev_for_each(rdev2, mddev)
f21e9ff7
N
2973 if (rdev->bdev == rdev2->bdev &&
2974 rdev != rdev2 &&
2975 overlaps(rdev->data_offset, rdev->sectors,
2976 rdev2->data_offset,
2977 rdev2->sectors)) {
c5d79adb
N
2978 overlap = 1;
2979 break;
2980 }
c5d79adb
N
2981 if (overlap) {
2982 mddev_put(mddev);
2983 break;
2984 }
2985 }
8b1afc3d 2986 rcu_read_unlock();
c5d79adb
N
2987 if (overlap) {
2988 /* Someone else could have slipped in a size
2989 * change here, but doing so is just silly.
dd8ac336 2990 * We put oldsectors back because we *know* it is
c5d79adb
N
2991 * safe, and trust userspace not to race with
2992 * itself
2993 */
dd8ac336 2994 rdev->sectors = oldsectors;
c5d79adb
N
2995 return -EBUSY;
2996 }
2997 }
83303b61
N
2998 return len;
2999}
3000
3001static struct rdev_sysfs_entry rdev_size =
80ca3a44 3002__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 3003
3cb03002 3004static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW
3005{
3006 unsigned long long recovery_start = rdev->recovery_offset;
3007
3008 if (test_bit(In_sync, &rdev->flags) ||
3009 recovery_start == MaxSector)
3010 return sprintf(page, "none\n");
3011
3012 return sprintf(page, "%llu\n", recovery_start);
3013}
3014
3cb03002 3015static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW
3016{
3017 unsigned long long recovery_start;
3018
3019 if (cmd_match(buf, "none"))
3020 recovery_start = MaxSector;
b29bebd6 3021 else if (kstrtoull(buf, 10, &recovery_start))
06e3c817
DW
3022 return -EINVAL;
3023
3024 if (rdev->mddev->pers &&
3025 rdev->raid_disk >= 0)
3026 return -EBUSY;
3027
3028 rdev->recovery_offset = recovery_start;
3029 if (recovery_start == MaxSector)
3030 set_bit(In_sync, &rdev->flags);
3031 else
3032 clear_bit(In_sync, &rdev->flags);
3033 return len;
3034}
3035
3036static struct rdev_sysfs_entry rdev_recovery_start =
3037__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3038
16c791a5
N
3039static ssize_t
3040badblocks_show(struct badblocks *bb, char *page, int unack);
3041static ssize_t
3042badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3043
3cb03002 3044static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N
3045{
3046 return badblocks_show(&rdev->badblocks, page, 0);
3047}
3cb03002 3048static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 3049{
de393cde
N
3050 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3051 /* Maybe that ack was all we needed */
3052 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3053 wake_up(&rdev->blocked_wait);
3054 return rv;
16c791a5
N
3055}
3056static struct rdev_sysfs_entry rdev_bad_blocks =
3057__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3058
3cb03002 3059static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N
3060{
3061 return badblocks_show(&rdev->badblocks, page, 1);
3062}
3cb03002 3063static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N
3064{
3065 return badblocks_store(&rdev->badblocks, page, len, 1);
3066}
3067static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3068__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3069
86e6ffdd
N
3070static struct attribute *rdev_default_attrs[] = {
3071 &rdev_state.attr,
4dbcdc75 3072 &rdev_errors.attr,
014236d2 3073 &rdev_slot.attr,
93c8cad0 3074 &rdev_offset.attr,
c6563a8c 3075 &rdev_new_offset.attr,
83303b61 3076 &rdev_size.attr,
06e3c817 3077 &rdev_recovery_start.attr,
16c791a5
N
3078 &rdev_bad_blocks.attr,
3079 &rdev_unack_bad_blocks.attr,
86e6ffdd
N
3080 NULL,
3081};
3082static ssize_t
3083rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3084{
3085 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3086 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3087 struct mddev *mddev = rdev->mddev;
27c529bb 3088 ssize_t rv;
86e6ffdd
N
3089
3090 if (!entry->show)
3091 return -EIO;
27c529bb
N
3092
3093 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3094 if (!rv) {
3095 if (rdev->mddev == NULL)
3096 rv = -EBUSY;
3097 else
3098 rv = entry->show(rdev, page);
3099 mddev_unlock(mddev);
3100 }
3101 return rv;
86e6ffdd
N
3102}
3103
3104static ssize_t
3105rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3106 const char *page, size_t length)
3107{
3108 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3109 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3110 ssize_t rv;
fd01b88c 3111 struct mddev *mddev = rdev->mddev;
86e6ffdd
N
3112
3113 if (!entry->store)
3114 return -EIO;
67463acb
N
3115 if (!capable(CAP_SYS_ADMIN))
3116 return -EACCES;
27c529bb 3117 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3118 if (!rv) {
27c529bb
N
3119 if (rdev->mddev == NULL)
3120 rv = -EBUSY;
3121 else
3122 rv = entry->store(rdev, page, length);
6a51830e 3123 mddev_unlock(mddev);
ca388059
N
3124 }
3125 return rv;
86e6ffdd
N
3126}
3127
3128static void rdev_free(struct kobject *ko)
3129{
3cb03002 3130 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N
3131 kfree(rdev);
3132}
52cf25d0 3133static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N
3134 .show = rdev_attr_show,
3135 .store = rdev_attr_store,
3136};
3137static struct kobj_type rdev_ktype = {
3138 .release = rdev_free,
3139 .sysfs_ops = &rdev_sysfs_ops,
3140 .default_attrs = rdev_default_attrs,
3141};
3142
3cb03002 3143int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N
3144{
3145 rdev->desc_nr = -1;
3146 rdev->saved_raid_disk = -1;
3147 rdev->raid_disk = -1;
3148 rdev->flags = 0;
3149 rdev->data_offset = 0;
c6563a8c 3150 rdev->new_data_offset = 0;
e8bb9a83
N
3151 rdev->sb_events = 0;
3152 rdev->last_read_error.tv_sec = 0;
3153 rdev->last_read_error.tv_nsec = 0;
2699b672
N
3154 rdev->sb_loaded = 0;
3155 rdev->bb_page = NULL;
e8bb9a83
N
3156 atomic_set(&rdev->nr_pending, 0);
3157 atomic_set(&rdev->read_errors, 0);
3158 atomic_set(&rdev->corrected_errors, 0);
3159
3160 INIT_LIST_HEAD(&rdev->same_set);
3161 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N
3162
3163 /* Add space to store bad block list.
3164 * This reserves the space even on arrays where it cannot
3165 * be used - I wonder if that matters
3166 */
3167 rdev->badblocks.count = 0;
486adf72 3168 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
2230dfe4
N
3169 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3170 seqlock_init(&rdev->badblocks.lock);
3171 if (rdev->badblocks.page == NULL)
3172 return -ENOMEM;
3173
3174 return 0;
e8bb9a83
N
3175}
3176EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT
3177/*
3178 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3179 *
3180 * mark the device faulty if:
3181 *
3182 * - the device is nonexistent (zero size)
3183 * - the device has no valid superblock
3184 *
3185 * a faulty rdev _never_ has rdev->sb set.
3186 */
3cb03002 3187static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT
3188{
3189 char b[BDEVNAME_SIZE];
3190 int err;
3cb03002 3191 struct md_rdev *rdev;
1da177e4
LT
3192 sector_t size;
3193
9ffae0cf 3194 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT
3195 if (!rdev) {
3196 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3197 return ERR_PTR(-ENOMEM);
3198 }
1da177e4 3199
2230dfe4
N
3200 err = md_rdev_init(rdev);
3201 if (err)
3202 goto abort_free;
3203 err = alloc_disk_sb(rdev);
3204 if (err)
1da177e4
LT
3205 goto abort_free;
3206
c5d79adb 3207 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT
3208 if (err)
3209 goto abort_free;
3210
f9cb074b 3211 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3212
77304d2a 3213 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4 3214 if (!size) {
f72ffdd6 3215 printk(KERN_WARNING
1da177e4
LT
3216 "md: %s has zero or unknown size, marking faulty!\n",
3217 bdevname(rdev->bdev,b));
3218 err = -EINVAL;
3219 goto abort_free;
3220 }
3221
3222 if (super_format >= 0) {
3223 err = super_types[super_format].
3224 load_super(rdev, NULL, super_minor);
3225 if (err == -EINVAL) {
df968c4e
N
3226 printk(KERN_WARNING
3227 "md: %s does not have a valid v%d.%d "
3228 "superblock, not importing!\n",
3229 bdevname(rdev->bdev,b),
3230 super_format, super_minor);
1da177e4
LT
3231 goto abort_free;
3232 }
3233 if (err < 0) {
f72ffdd6 3234 printk(KERN_WARNING
1da177e4
LT
3235 "md: could not read %s's sb, not importing!\n",
3236 bdevname(rdev->bdev,b));
3237 goto abort_free;
3238 }
3239 }
6bfe0b49 3240
1da177e4
LT
3241 return rdev;
3242
3243abort_free:
2699b672
N
3244 if (rdev->bdev)
3245 unlock_rdev(rdev);
545c8795 3246 md_rdev_clear(rdev);
1da177e4
LT
3247 kfree(rdev);
3248 return ERR_PTR(err);
3249}
3250
3251/*
3252 * Check a full RAID array for plausibility
3253 */
3254
f72ffdd6 3255static void analyze_sbs(struct mddev *mddev)
1da177e4
LT
3256{
3257 int i;
3cb03002 3258 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT
3259 char b[BDEVNAME_SIZE];
3260
3261 freshest = NULL;
dafb20fa 3262 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT
3263 switch (super_types[mddev->major_version].
3264 load_super(rdev, freshest, mddev->minor_version)) {
3265 case 1:
3266 freshest = rdev;
3267 break;
3268 case 0:
3269 break;
3270 default:
3271 printk( KERN_ERR \
3272 "md: fatal superblock inconsistency in %s"
f72ffdd6 3273 " -- removing from array\n",
1da177e4
LT
3274 bdevname(rdev->bdev,b));
3275 kick_rdev_from_array(rdev);
3276 }
3277
1da177e4
LT
3278 super_types[mddev->major_version].
3279 validate_super(mddev, freshest);
3280
3281 i = 0;
dafb20fa 3282 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N
3283 if (mddev->max_disks &&
3284 (rdev->desc_nr >= mddev->max_disks ||
3285 i > mddev->max_disks)) {
de01dfad
N
3286 printk(KERN_WARNING
3287 "md: %s: %s: only %d devices permitted\n",
3288 mdname(mddev), bdevname(rdev->bdev, b),
3289 mddev->max_disks);
3290 kick_rdev_from_array(rdev);
3291 continue;
3292 }
1da177e4
LT
3293 if (rdev != freshest)
3294 if (super_types[mddev->major_version].
3295 validate_super(mddev, rdev)) {
3296 printk(KERN_WARNING "md: kicking non-fresh %s"
3297 " from array!\n",
3298 bdevname(rdev->bdev,b));
3299 kick_rdev_from_array(rdev);
3300 continue;
3301 }
3302 if (mddev->level == LEVEL_MULTIPATH) {
3303 rdev->desc_nr = i++;
3304 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3305 set_bit(In_sync, &rdev->flags);
5e5e3e78 3306 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N
3307 rdev->raid_disk = -1;
3308 clear_bit(In_sync, &rdev->flags);
1da177e4
LT
3309 }
3310 }
1da177e4
LT
3311}
3312
72e02075
N
3313/* Read a fixed-point number.
3314 * Numbers in sysfs attributes should be in "standard" units where
3315 * possible, so time should be in seconds.
f72ffdd6 3316 * However we internally use a a much smaller unit such as
72e02075
N
3317 * milliseconds or jiffies.
3318 * This function takes a decimal number with a possible fractional
3319 * component, and produces an integer which is the result of
3320 * multiplying that number by 10^'scale'.
3321 * all without any floating-point arithmetic.
3322 */
3323int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3324{
3325 unsigned long result = 0;
3326 long decimals = -1;
3327 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3328 if (*cp == '.')
3329 decimals = 0;
3330 else if (decimals < scale) {
3331 unsigned int value;
3332 value = *cp - '0';
3333 result = result * 10 + value;
3334 if (decimals >= 0)
3335 decimals++;
3336 }
3337 cp++;
3338 }
3339 if (*cp == '\n')
3340 cp++;
3341 if (*cp)
3342 return -EINVAL;
3343 if (decimals < 0)
3344 decimals = 0;
3345 while (decimals < scale) {
3346 result *= 10;
3347 decimals ++;
3348 }
3349 *res = result;
3350 return 0;
3351}
3352
19052c0e
N
3353static void md_safemode_timeout(unsigned long data);
3354
16f17b39 3355static ssize_t
fd01b88c 3356safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N
3357{
3358 int msec = (mddev->safemode_delay*1000)/HZ;
3359 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3360}
3361static ssize_t
fd01b88c 3362safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3363{
16f17b39 3364 unsigned long msec;
97ce0a7f 3365
72e02075 3366 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3367 return -EINVAL;
16f17b39
N
3368 if (msec == 0)
3369 mddev->safemode_delay = 0;
3370 else {
19052c0e 3371 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N
3372 mddev->safemode_delay = (msec*HZ)/1000;
3373 if (mddev->safemode_delay == 0)
3374 mddev->safemode_delay = 1;
275c51c4 3375 if (mddev->safemode_delay < old_delay || old_delay == 0)
19052c0e 3376 md_safemode_timeout((unsigned long)mddev);
16f17b39
N
3377 }
3378 return len;
3379}
3380static struct md_sysfs_entry md_safe_delay =
80ca3a44 3381__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3382
eae1701f 3383static ssize_t
fd01b88c 3384level_show(struct mddev *mddev, char *page)
eae1701f 3385{
84fc4b56 3386 struct md_personality *p = mddev->pers;
d9d166c2 3387 if (p)
eae1701f 3388 return sprintf(page, "%s\n", p->name);
d9d166c2
N
3389 else if (mddev->clevel[0])
3390 return sprintf(page, "%s\n", mddev->clevel);
3391 else if (mddev->level != LEVEL_NONE)
3392 return sprintf(page, "%d\n", mddev->level);
3393 else
3394 return 0;
eae1701f
N
3395}
3396
d9d166c2 3397static ssize_t
fd01b88c 3398level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3399{
f2859af6 3400 char clevel[16];
20a49ff6 3401 ssize_t rv = len;
84fc4b56 3402 struct md_personality *pers;
f2859af6 3403 long level;
245f46c2 3404 void *priv;
3cb03002 3405 struct md_rdev *rdev;
245f46c2
N
3406
3407 if (mddev->pers == NULL) {
3408 if (len == 0)
3409 return 0;
3410 if (len >= sizeof(mddev->clevel))
3411 return -ENOSPC;
3412 strncpy(mddev->clevel, buf, len);
3413 if (mddev->clevel[len-1] == '\n')
3414 len--;
3415 mddev->clevel[len] = 0;
3416 mddev->level = LEVEL_NONE;
3417 return rv;
3418 }
bd8839e0
N
3419 if (mddev->ro)
3420 return -EROFS;
245f46c2
N
3421
3422 /* request to change the personality. Need to ensure:
3423 * - array is not engaged in resync/recovery/reshape
3424 * - old personality can be suspended
3425 * - new personality will access other array.
3426 */
3427
bb4f1e9d
N
3428 if (mddev->sync_thread ||
3429 mddev->reshape_position != MaxSector ||
3430 mddev->sysfs_active)
d9d166c2 3431 return -EBUSY;
245f46c2
N
3432
3433 if (!mddev->pers->quiesce) {
3434 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3435 mdname(mddev), mddev->pers->name);
3436 return -EINVAL;
3437 }
3438
3439 /* Now find the new personality */
f2859af6 3440 if (len == 0 || len >= sizeof(clevel))
245f46c2 3441 return -EINVAL;
f2859af6
DW
3442 strncpy(clevel, buf, len);
3443 if (clevel[len-1] == '\n')
d9d166c2 3444 len--;
f2859af6 3445 clevel[len] = 0;
b29bebd6 3446 if (kstrtol(clevel, 10, &level))
f2859af6 3447 level = LEVEL_NONE;
245f46c2 3448
f2859af6
DW
3449 if (request_module("md-%s", clevel) != 0)
3450 request_module("md-level-%s", clevel);
245f46c2 3451 spin_lock(&pers_lock);
f2859af6 3452 pers = find_pers(level, clevel);
245f46c2
N
3453 if (!pers || !try_module_get(pers->owner)) {
3454 spin_unlock(&pers_lock);
f2859af6 3455 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N
3456 return -EINVAL;
3457 }
3458 spin_unlock(&pers_lock);
3459
3460 if (pers == mddev->pers) {
3461 /* Nothing to do! */
3462 module_put(pers->owner);
3463 return rv;
3464 }
3465 if (!pers->takeover) {
3466 module_put(pers->owner);
3467 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3468 mdname(mddev), clevel);
245f46c2
N
3469 return -EINVAL;
3470 }
3471
dafb20fa 3472 rdev_for_each(rdev, mddev)
e93f68a1
N
3473 rdev->new_raid_disk = rdev->raid_disk;
3474
245f46c2
N
3475 /* ->takeover must set new_* and/or delta_disks
3476 * if it succeeds, and may set them when it fails.
3477 */
3478 priv = pers->takeover(mddev);
3479 if (IS_ERR(priv)) {
3480 mddev->new_level = mddev->level;
3481 mddev->new_layout = mddev->layout;
664e7c41 3482 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N
3483 mddev->raid_disks -= mddev->delta_disks;
3484 mddev->delta_disks = 0;
2c810cdd 3485 mddev->reshape_backwards = 0;
245f46c2
N
3486 module_put(pers->owner);
3487 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3488 mdname(mddev), clevel);
245f46c2
N
3489 return PTR_ERR(priv);
3490 }
3491
3492 /* Looks like we have a winner */
3493 mddev_suspend(mddev);
3494 mddev->pers->stop(mddev);
f72ffdd6 3495
a64c876f
N
3496 if (mddev->pers->sync_request == NULL &&
3497 pers->sync_request != NULL) {
3498 /* need to add the md_redundancy_group */
3499 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3500 printk(KERN_WARNING
3501 "md: cannot register extra attributes for %s\n",
3502 mdname(mddev));
388975cc 3503 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
f72ffdd6 3504 }
a64c876f
N
3505 if (mddev->pers->sync_request != NULL &&
3506 pers->sync_request == NULL) {
3507 /* need to remove the md_redundancy_group */
3508 if (mddev->to_remove == NULL)
3509 mddev->to_remove = &md_redundancy_group;
3510 }
3511
54071b38
TM
3512 if (mddev->pers->sync_request == NULL &&
3513 mddev->external) {
3514 /* We are converting from a no-redundancy array
3515 * to a redundancy array and metadata is managed
3516 * externally so we need to be sure that writes
3517 * won't block due to a need to transition
3518 * clean->dirty
3519 * until external management is started.
3520 */
3521 mddev->in_sync = 0;
3522 mddev->safemode_delay = 0;
3523 mddev->safemode = 0;
3524 }
3525
dafb20fa 3526 rdev_for_each(rdev, mddev) {
e93f68a1
N
3527 if (rdev->raid_disk < 0)
3528 continue;
bf2cb0da 3529 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N
3530 rdev->new_raid_disk = -1;
3531 if (rdev->new_raid_disk == rdev->raid_disk)
3532 continue;
36fad858 3533 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3534 }
dafb20fa 3535 rdev_for_each(rdev, mddev) {
e93f68a1
N
3536 if (rdev->raid_disk < 0)
3537 continue;
3538 if (rdev->new_raid_disk == rdev->raid_disk)
3539 continue;
3540 rdev->raid_disk = rdev->new_raid_disk;
3541 if (rdev->raid_disk < 0)
3a981b03 3542 clear_bit(In_sync, &rdev->flags);
e93f68a1 3543 else {
36fad858
NK
3544 if (sysfs_link_rdev(mddev, rdev))
3545 printk(KERN_WARNING "md: cannot register rd%d"
3546 " for %s after level change\n",
3547 rdev->raid_disk, mdname(mddev));
3a981b03 3548 }
e93f68a1
N
3549 }
3550
3551 module_put(mddev->pers->owner);
245f46c2
N
3552 mddev->pers = pers;
3553 mddev->private = priv;
3554 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3555 mddev->level = mddev->new_level;
3556 mddev->layout = mddev->new_layout;
664e7c41 3557 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3558 mddev->delta_disks = 0;
2c810cdd 3559 mddev->reshape_backwards = 0;
fee68723 3560 mddev->degraded = 0;
9af204cf
TM
3561 if (mddev->pers->sync_request == NULL) {
3562 /* this is now an array without redundancy, so
3563 * it must always be in_sync
3564 */
3565 mddev->in_sync = 1;
3566 del_timer_sync(&mddev->safemode_timer);
3567 }
02e5f5c0 3568 blk_set_stacking_limits(&mddev->queue->limits);
245f46c2 3569 pers->run(mddev);
245f46c2 3570 set_bit(MD_CHANGE_DEVS, &mddev->flags);
47525e59 3571 mddev_resume(mddev);
830778a1
N
3572 if (!mddev->thread)
3573 md_update_sb(mddev, 1);
5cac7861 3574 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3575 md_new_event(mddev);
d9d166c2
N
3576 return rv;
3577}
3578
3579static struct md_sysfs_entry md_level =
80ca3a44 3580__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3581
d4dbd025 3582static ssize_t
fd01b88c 3583layout_show(struct mddev *mddev, char *page)
d4dbd025
N
3584{
3585 /* just a number, not meaningful for all levels */
08a02ecd
N
3586 if (mddev->reshape_position != MaxSector &&
3587 mddev->layout != mddev->new_layout)
3588 return sprintf(page, "%d (%d)\n",
3589 mddev->new_layout, mddev->layout);
d4dbd025
N
3590 return sprintf(page, "%d\n", mddev->layout);
3591}
3592
3593static ssize_t
fd01b88c 3594layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N
3595{
3596 char *e;
3597 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N
3598
3599 if (!*buf || (*e && *e != '\n'))
3600 return -EINVAL;
3601
b3546035
N
3602 if (mddev->pers) {
3603 int err;
50ac168a 3604 if (mddev->pers->check_reshape == NULL)
b3546035 3605 return -EBUSY;
bd8839e0
N
3606 if (mddev->ro)
3607 return -EROFS;
597a711b 3608 mddev->new_layout = n;
50ac168a 3609 err = mddev->pers->check_reshape(mddev);
597a711b
N
3610 if (err) {
3611 mddev->new_layout = mddev->layout;
b3546035 3612 return err;
597a711b 3613 }
b3546035 3614 } else {
08a02ecd 3615 mddev->new_layout = n;
b3546035
N
3616 if (mddev->reshape_position == MaxSector)
3617 mddev->layout = n;
3618 }
d4dbd025
N
3619 return len;
3620}
3621static struct md_sysfs_entry md_layout =
80ca3a44 3622__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025 3623
eae1701f 3624static ssize_t
fd01b88c 3625raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3626{
bb636547
N
3627 if (mddev->raid_disks == 0)
3628 return 0;
08a02ecd
N
3629 if (mddev->reshape_position != MaxSector &&
3630 mddev->delta_disks != 0)
3631 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3632 mddev->raid_disks - mddev->delta_disks);
eae1701f
N
3633 return sprintf(page, "%d\n", mddev->raid_disks);
3634}
3635
fd01b88c 3636static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N
3637
3638static ssize_t
fd01b88c 3639raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3640{
da943b99
N
3641 char *e;
3642 int rv = 0;
3643 unsigned long n = simple_strtoul(buf, &e, 10);
3644
3645 if (!*buf || (*e && *e != '\n'))
3646 return -EINVAL;
3647
3648 if (mddev->pers)
3649 rv = update_raid_disks(mddev, n);
08a02ecd 3650 else if (mddev->reshape_position != MaxSector) {
c6563a8c 3651 struct md_rdev *rdev;
08a02ecd 3652 int olddisks = mddev->raid_disks - mddev->delta_disks;
c6563a8c
N
3653
3654 rdev_for_each(rdev, mddev) {
3655 if (olddisks < n &&
3656 rdev->data_offset < rdev->new_data_offset)
3657 return -EINVAL;
3658 if (olddisks > n &&
3659 rdev->data_offset > rdev->new_data_offset)
3660 return -EINVAL;
3661 }
08a02ecd
N
3662 mddev->delta_disks = n - olddisks;
3663 mddev->raid_disks = n;
2c810cdd 3664 mddev->reshape_backwards = (mddev->delta_disks < 0);
08a02ecd 3665 } else
da943b99
N
3666 mddev->raid_disks = n;
3667 return rv ? rv : len;
3668}
3669static struct md_sysfs_entry md_raid_disks =
80ca3a44 3670__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3671
3b34380a 3672static ssize_t
fd01b88c 3673chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3674{
08a02ecd 3675 if (mddev->reshape_position != MaxSector &&
664e7c41
AN
3676 mddev->chunk_sectors != mddev->new_chunk_sectors)
3677 return sprintf(page, "%d (%d)\n",
3678 mddev->new_chunk_sectors << 9,
9d8f0363
AN
3679 mddev->chunk_sectors << 9);
3680 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N
3681}
3682
3683static ssize_t
fd01b88c 3684chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3685{
3b34380a
N
3686 char *e;
3687 unsigned long n = simple_strtoul(buf, &e, 10);
3688
3b34380a
N
3689 if (!*buf || (*e && *e != '\n'))
3690 return -EINVAL;
3691
b3546035
N
3692 if (mddev->pers) {
3693 int err;
50ac168a 3694 if (mddev->pers->check_reshape == NULL)
b3546035 3695 return -EBUSY;
bd8839e0
N
3696 if (mddev->ro)
3697 return -EROFS;
597a711b 3698 mddev->new_chunk_sectors = n >> 9;
50ac168a 3699 err = mddev->pers->check_reshape(mddev);
597a711b
N
3700 if (err) {
3701 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3702 return err;
597a711b 3703 }
b3546035 3704 } else {
664e7c41 3705 mddev->new_chunk_sectors = n >> 9;
b3546035 3706 if (mddev->reshape_position == MaxSector)
9d8f0363 3707 mddev->chunk_sectors = n >> 9;
b3546035 3708 }
3b34380a
N
3709 return len;
3710}
3711static struct md_sysfs_entry md_chunk_size =
80ca3a44 3712__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3713
a94213b1 3714static ssize_t
fd01b88c 3715resync_start_show(struct mddev *mddev, char *page)
a94213b1 3716{
d1a7c503
N
3717 if (mddev->recovery_cp == MaxSector)
3718 return sprintf(page, "none\n");
a94213b1
N
3719 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3720}
3721
3722static ssize_t
fd01b88c 3723resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3724{
a94213b1
N
3725 char *e;
3726 unsigned long long n = simple_strtoull(buf, &e, 10);
3727
b098636c 3728 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3729 return -EBUSY;
06e3c817
DW
3730 if (cmd_match(buf, "none"))
3731 n = MaxSector;
3732 else if (!*buf || (*e && *e != '\n'))
a94213b1
N
3733 return -EINVAL;
3734
3735 mddev->recovery_cp = n;
db07d85e
N
3736 if (mddev->pers)
3737 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
a94213b1
N
3738 return len;
3739}
3740static struct md_sysfs_entry md_resync_start =
80ca3a44 3741__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3742
9e653b63
N
3743/*
3744 * The array state can be:
3745 *
3746 * clear
3747 * No devices, no size, no level
3748 * Equivalent to STOP_ARRAY ioctl
3749 * inactive
3750 * May have some settings, but array is not active
3751 * all IO results in error
3752 * When written, doesn't tear down array, but just stops it
3753 * suspended (not supported yet)
3754 * All IO requests will block. The array can be reconfigured.
910d8cb3 3755 * Writing this, if accepted, will block until array is quiescent
9e653b63
N
3756 * readonly
3757 * no resync can happen. no superblocks get written.
3758 * write requests fail
3759 * read-auto
3760 * like readonly, but behaves like 'clean' on a write request.
3761 *
3762 * clean - no pending writes, but otherwise active.
3763 * When written to inactive array, starts without resync
3764 * If a write request arrives then
3765 * if metadata is known, mark 'dirty' and switch to 'active'.
3766 * if not known, block and switch to write-pending
3767 * If written to an active array that has pending writes, then fails.
3768 * active
3769 * fully active: IO and resync can be happening.
3770 * When written to inactive array, starts with resync
3771 *
3772 * write-pending
3773 * clean, but writes are blocked waiting for 'active' to be written.
3774 *
3775 * active-idle
3776 * like active, but no writes have been seen for a while (100msec).
3777 *
3778 */
3779enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3780 write_pending, active_idle, bad_word};
05381954 3781static char *array_states[] = {
9e653b63
N
3782 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3783 "write-pending", "active-idle", NULL };
3784
3785static int match_word(const char *word, char **list)
3786{
3787 int n;
3788 for (n=0; list[n]; n++)
3789 if (cmd_match(word, list[n]))
3790 break;
3791 return n;
3792}
3793
3794static ssize_t
fd01b88c 3795array_state_show(struct mddev *mddev, char *page)
9e653b63
N
3796{
3797 enum array_state st = inactive;
3798
3799 if (mddev->pers)
3800 switch(mddev->ro) {
3801 case 1:
3802 st = readonly;
3803 break;
3804 case 2:
3805 st = read_auto;
3806 break;
3807 case 0:
3808 if (mddev->in_sync)
3809 st = clean;
070dc6dd 3810 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3811 st = write_pending;
9e653b63
N
3812 else if (mddev->safemode)
3813 st = active_idle;
3814 else
3815 st = active;
3816 }
3817 else {
3818 if (list_empty(&mddev->disks) &&
3819 mddev->raid_disks == 0 &&
58c0fed4 3820 mddev->dev_sectors == 0)
9e653b63
N
3821 st = clear;
3822 else
3823 st = inactive;
3824 }
3825 return sprintf(page, "%s\n", array_states[st]);
3826}
3827
f72ffdd6
N
3828static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3829static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3830static int do_md_run(struct mddev *mddev);
fd01b88c 3831static int restart_array(struct mddev *mddev);
9e653b63
N
3832
3833static ssize_t
fd01b88c 3834array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N
3835{
3836 int err = -EINVAL;
3837 enum array_state st = match_word(buf, array_states);
3838 switch(st) {
3839 case bad_word:
3840 break;
3841 case clear:
3842 /* stopping an active array */
a05b7ea0 3843 err = do_md_stop(mddev, 0, NULL);
9e653b63
N
3844 break;
3845 case inactive:
3846 /* stopping an active array */
90cf195d 3847 if (mddev->pers)
a05b7ea0 3848 err = do_md_stop(mddev, 2, NULL);
90cf195d 3849 else
e691063a 3850 err = 0; /* already inactive */
9e653b63
N
3851 break;
3852 case suspended:
3853 break; /* not supported yet */
3854 case readonly:
3855 if (mddev->pers)
a05b7ea0 3856 err = md_set_readonly(mddev, NULL);
9e653b63
N
3857 else {
3858 mddev->ro = 1;
648b629e 3859 set_disk_ro(mddev->gendisk, 1);
9e653b63
N
3860 err = do_md_run(mddev);
3861 }
3862 break;
3863 case read_auto:
9e653b63 3864 if (mddev->pers) {
80268ee9 3865 if (mddev->ro == 0)
a05b7ea0 3866 err = md_set_readonly(mddev, NULL);
80268ee9 3867 else if (mddev->ro == 1)
648b629e
N
3868 err = restart_array(mddev);
3869 if (err == 0) {
3870 mddev->ro = 2;
3871 set_disk_ro(mddev->gendisk, 0);
3872 }
9e653b63
N
3873 } else {
3874 mddev->ro = 2;
3875 err = do_md_run(mddev);
3876 }
3877 break;
3878 case clean:
3879 if (mddev->pers) {
3880 restart_array(mddev);
3881 spin_lock_irq(&mddev->write_lock);
3882 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N
3883 if (mddev->in_sync == 0) {
3884 mddev->in_sync = 1;
31a59e34
N
3885 if (mddev->safemode == 1)
3886 mddev->safemode = 0;
070dc6dd 3887 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N
3888 }
3889 err = 0;
3890 } else
3891 err = -EBUSY;
9e653b63 3892 spin_unlock_irq(&mddev->write_lock);
5bf29597
N
3893 } else
3894 err = -EINVAL;
9e653b63
N
3895 break;
3896 case active:
3897 if (mddev->pers) {
3898 restart_array(mddev);
070dc6dd 3899 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N
3900 wake_up(&mddev->sb_wait);
3901 err = 0;
3902 } else {
3903 mddev->ro = 0;
648b629e 3904 set_disk_ro(mddev->gendisk, 0);
9e653b63
N
3905 err = do_md_run(mddev);
3906 }
3907 break;
3908 case write_pending:
3909 case active_idle:
3910 /* these cannot be set */
3911 break;
3912 }
3913 if (err)
3914 return err;
0fd62b86 3915 else {
1d23f178
N
3916 if (mddev->hold_active == UNTIL_IOCTL)
3917 mddev->hold_active = 0;
00bcb4ac 3918 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3919 return len;
0fd62b86 3920 }
9e653b63 3921}
80ca3a44
N
3922static struct md_sysfs_entry md_array_state =
3923__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3924
1e50915f 3925static ssize_t
fd01b88c 3926max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB
3927 return sprintf(page, "%d\n",
3928 atomic_read(&mddev->max_corr_read_errors));
3929}
3930
3931static ssize_t
fd01b88c 3932max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB
3933{
3934 char *e;
3935 unsigned long n = simple_strtoul(buf, &e, 10);
3936
3937 if (*buf && (*e == 0 || *e == '\n')) {
3938 atomic_set(&mddev->max_corr_read_errors, n);
3939 return len;
3940 }
3941 return -EINVAL;
3942}
3943
3944static struct md_sysfs_entry max_corr_read_errors =
3945__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3946 max_corrected_read_errors_store);
3947
6d7ff738 3948static ssize_t
fd01b88c 3949null_show(struct mddev *mddev, char *page)
6d7ff738
N
3950{
3951 return -EINVAL;
3952}
3953
3954static ssize_t
fd01b88c 3955new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N
3956{
3957 /* buf must be %d:%d\n? giving major and minor numbers */
3958 /* The new device is added to the array.
3959 * If the array has a persistent superblock, we read the
3960 * superblock to initialise info and check validity.
3961 * Otherwise, only checking done is that in bind_rdev_to_array,
3962 * which mainly checks size.
3963 */
3964 char *e;
3965 int major = simple_strtoul(buf, &e, 10);
3966 int minor;
3967 dev_t dev;
3cb03002 3968 struct md_rdev *rdev;
6d7ff738
N
3969 int err;
3970
3971 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3972 return -EINVAL;
3973 minor = simple_strtoul(e+1, &e, 10);
3974 if (*e && *e != '\n')
3975 return -EINVAL;
3976 dev = MKDEV(major, minor);
3977 if (major != MAJOR(dev) ||
3978 minor != MINOR(dev))
3979 return -EOVERFLOW;
3980
6d7ff738
N
3981 if (mddev->persistent) {
3982 rdev = md_import_device(dev, mddev->major_version,
3983 mddev->minor_version);
3984 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N
3985 struct md_rdev *rdev0
3986 = list_entry(mddev->disks.next,
3987 struct md_rdev, same_set);
6d7ff738
N
3988 err = super_types[mddev->major_version]
3989 .load_super(rdev, rdev0, mddev->minor_version);
3990 if (err < 0)
3991 goto out;
3992 }
c5d79adb
N
3993 } else if (mddev->external)
3994 rdev = md_import_device(dev, -2, -1);
3995 else
6d7ff738
N
3996 rdev = md_import_device(dev, -1, -1);
3997
3998 if (IS_ERR(rdev))
3999 return PTR_ERR(rdev);
4000 err = bind_rdev_to_array(rdev, mddev);
4001 out:
4002 if (err)
4003 export_rdev(rdev);
4004 return err ? err : len;
4005}
4006
4007static struct md_sysfs_entry md_new_device =
80ca3a44 4008__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 4009
9b1d1dac 4010static ssize_t
fd01b88c 4011bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC
4012{
4013 char *end;
4014 unsigned long chunk, end_chunk;
4015
4016 if (!mddev->bitmap)
4017 goto out;
4018 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4019 while (*buf) {
4020 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4021 if (buf == end) break;
4022 if (*end == '-') { /* range */
4023 buf = end + 1;
4024 end_chunk = simple_strtoul(buf, &end, 0);
4025 if (buf == end) break;
4026 }
4027 if (*end && !isspace(*end)) break;
4028 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 4029 buf = skip_spaces(end);
9b1d1dac
PC
4030 }
4031 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4032out:
4033 return len;
4034}
4035
4036static struct md_sysfs_entry md_bitmap =
4037__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4038
a35b0d69 4039static ssize_t
fd01b88c 4040size_show(struct mddev *mddev, char *page)
a35b0d69 4041{
58c0fed4
AN
4042 return sprintf(page, "%llu\n",
4043 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N
4044}
4045
fd01b88c 4046static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N
4047
4048static ssize_t
fd01b88c 4049size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N
4050{
4051 /* If array is inactive, we can reduce the component size, but
4052 * not increase it (except from 0).
4053 * If array is active, we can try an on-line resize
4054 */
b522adcd
DW
4055 sector_t sectors;
4056 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 4057
58c0fed4
AN
4058 if (err < 0)
4059 return err;
a35b0d69 4060 if (mddev->pers) {
58c0fed4 4061 err = update_size(mddev, sectors);
850b2b42 4062 md_update_sb(mddev, 1);
a35b0d69 4063 } else {
58c0fed4
AN
4064 if (mddev->dev_sectors == 0 ||
4065 mddev->dev_sectors > sectors)
4066 mddev->dev_sectors = sectors;
a35b0d69
N
4067 else
4068 err = -ENOSPC;
4069 }
4070 return err ? err : len;
4071}
4072
4073static struct md_sysfs_entry md_size =
80ca3a44 4074__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 4075
83f0d77a 4076/* Metadata version.
e691063a
N
4077 * This is one of
4078 * 'none' for arrays with no metadata (good luck...)
4079 * 'external' for arrays with externally managed metadata,
8bb93aac
N
4080 * or N.M for internally known formats
4081 */
4082static ssize_t
fd01b88c 4083metadata_show(struct mddev *mddev, char *page)
8bb93aac
N
4084{
4085 if (mddev->persistent)
4086 return sprintf(page, "%d.%d\n",
4087 mddev->major_version, mddev->minor_version);
e691063a
N
4088 else if (mddev->external)
4089 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N
4090 else
4091 return sprintf(page, "none\n");
4092}
4093
4094static ssize_t
fd01b88c 4095metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N
4096{
4097 int major, minor;
4098 char *e;
ea43ddd8
N
4099 /* Changing the details of 'external' metadata is
4100 * always permitted. Otherwise there must be
4101 * no devices attached to the array.
4102 */
4103 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4104 ;
4105 else if (!list_empty(&mddev->disks))
8bb93aac
N
4106 return -EBUSY;
4107
4108 if (cmd_match(buf, "none")) {
4109 mddev->persistent = 0;
e691063a
N
4110 mddev->external = 0;
4111 mddev->major_version = 0;
4112 mddev->minor_version = 90;
4113 return len;
4114 }
4115 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4116 size_t namelen = len-9;
e691063a
N
4117 if (namelen >= sizeof(mddev->metadata_type))
4118 namelen = sizeof(mddev->metadata_type)-1;
4119 strncpy(mddev->metadata_type, buf+9, namelen);
4120 mddev->metadata_type[namelen] = 0;
4121 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4122 mddev->metadata_type[--namelen] = 0;
4123 mddev->persistent = 0;
4124 mddev->external = 1;
8bb93aac
N
4125 mddev->major_version = 0;
4126 mddev->minor_version = 90;
4127 return len;
4128 }
4129 major = simple_strtoul(buf, &e, 10);
4130 if (e==buf || *e != '.')
4131 return -EINVAL;
4132 buf = e+1;
4133 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4134 if (e==buf || (*e && *e != '\n') )
8bb93aac 4135 return -EINVAL;
50511da3 4136 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N
4137 return -ENOENT;
4138 mddev->major_version = major;
4139 mddev->minor_version = minor;
4140 mddev->persistent = 1;
e691063a 4141 mddev->external = 0;
8bb93aac
N
4142 return len;
4143}
4144
4145static struct md_sysfs_entry md_metadata =
80ca3a44 4146__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4147
24dd469d 4148static ssize_t
fd01b88c 4149action_show(struct mddev *mddev, char *page)
24dd469d 4150{
7eec314d 4151 char *type = "idle";
b6a9ce68
N
4152 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4153 type = "frozen";
4154 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4155 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N
4156 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4157 type = "reshape";
4158 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N
4159 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4160 type = "resync";
4161 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4162 type = "check";
4163 else
4164 type = "repair";
72a23c21 4165 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N
4166 type = "recover";
4167 }
4168 return sprintf(page, "%s\n", type);
4169}
4170
4171static ssize_t
fd01b88c 4172action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4173{
7eec314d
N
4174 if (!mddev->pers || !mddev->pers->sync_request)
4175 return -EINVAL;
4176
b6a9ce68
N
4177 if (cmd_match(page, "frozen"))
4178 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4179 else
4180 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4181
4182 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N
4183 if (mddev->sync_thread) {
4184 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 4185 md_reap_sync_thread(mddev);
7eec314d 4186 }
03c902e1
N
4187 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4188 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4189 return -EBUSY;
72a23c21
NB
4190 else if (cmd_match(page, "resync"))
4191 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4192 else if (cmd_match(page, "recover")) {
4193 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4194 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4195 } else if (cmd_match(page, "reshape")) {
16484bf5
N
4196 int err;
4197 if (mddev->pers->start_reshape == NULL)
4198 return -EINVAL;
4199 err = mddev->pers->start_reshape(mddev);
4200 if (err)
4201 return err;
a99ac971 4202 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4203 } else {
bce74dac 4204 if (cmd_match(page, "check"))
7eec314d 4205 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4206 else if (!cmd_match(page, "repair"))
7eec314d
N
4207 return -EINVAL;
4208 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4209 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4210 }
48c26ddc
N
4211 if (mddev->ro == 2) {
4212 /* A write to sync_action is enough to justify
4213 * canceling read-auto mode
4214 */
4215 mddev->ro = 0;
4216 md_wakeup_thread(mddev->sync_thread);
4217 }
03c902e1 4218 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4219 md_wakeup_thread(mddev->thread);
00bcb4ac 4220 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N
4221 return len;
4222}
4223
c4a39551
JB
4224static struct md_sysfs_entry md_scan_mode =
4225__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4226
4227static ssize_t
4228last_sync_action_show(struct mddev *mddev, char *page)
4229{
4230 return sprintf(page, "%s\n", mddev->last_sync_action);
4231}
4232
4233static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4234
9d88883e 4235static ssize_t
fd01b88c 4236mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N
4237{
4238 return sprintf(page, "%llu\n",
7f7583d4
JM
4239 (unsigned long long)
4240 atomic64_read(&mddev->resync_mismatches));
9d88883e
N
4241}
4242
80ca3a44 4243static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4244
88202a0c 4245static ssize_t
fd01b88c 4246sync_min_show(struct mddev *mddev, char *page)
88202a0c
N
4247{
4248 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4249 mddev->sync_speed_min ? "local": "system");
4250}
4251
4252static ssize_t
fd01b88c 4253sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N
4254{
4255 int min;
4256 char *e;
4257 if (strncmp(buf, "system", 6)==0) {
4258 mddev->sync_speed_min = 0;
4259 return len;
4260 }
4261 min = simple_strtoul(buf, &e, 10);
4262 if (buf == e || (*e && *e != '\n') || min <= 0)
4263 return -EINVAL;
4264 mddev->sync_speed_min = min;
4265 return len;
4266}
4267
4268static struct md_sysfs_entry md_sync_min =
4269__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4270
4271static ssize_t
fd01b88c 4272sync_max_show(struct mddev *mddev, char *page)
88202a0c
N
4273{
4274 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4275 mddev->sync_speed_max ? "local": "system");
4276}
4277
4278static ssize_t
fd01b88c 4279sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N
4280{
4281 int max;
4282 char *e;
4283 if (strncmp(buf, "system", 6)==0) {
4284 mddev->sync_speed_max = 0;
4285 return len;
4286 }
4287 max = simple_strtoul(buf, &e, 10);
4288 if (buf == e || (*e && *e != '\n') || max <= 0)
4289 return -EINVAL;
4290 mddev->sync_speed_max = max;
4291 return len;
4292}
4293
4294static struct md_sysfs_entry md_sync_max =
4295__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4296
d7f3d291 4297static ssize_t
fd01b88c 4298degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP
4299{
4300 return sprintf(page, "%d\n", mddev->degraded);
4301}
4302static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4303
90b08710 4304static ssize_t
fd01b88c 4305sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS
4306{
4307 return sprintf(page, "%d\n", mddev->parallel_resync);
4308}
4309
4310static ssize_t
fd01b88c 4311sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS
4312{
4313 long n;
4314
b29bebd6 4315 if (kstrtol(buf, 10, &n))
90b08710
BS
4316 return -EINVAL;
4317
4318 if (n != 0 && n != 1)
4319 return -EINVAL;
4320
4321 mddev->parallel_resync = n;
4322
4323 if (mddev->sync_thread)
4324 wake_up(&resync_wait);
4325
4326 return len;
4327}
4328
4329/* force parallel resync, even with shared block devices */
4330static struct md_sysfs_entry md_sync_force_parallel =
4331__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4332 sync_force_parallel_show, sync_force_parallel_store);
4333
88202a0c 4334static ssize_t
fd01b88c 4335sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N
4336{
4337 unsigned long resync, dt, db;
d1a7c503
N
4338 if (mddev->curr_resync == 0)
4339 return sprintf(page, "none\n");
9687a60c
AN
4340 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4341 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4342 if (!dt) dt++;
9687a60c
AN
4343 db = resync - mddev->resync_mark_cnt;
4344 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N
4345}
4346
80ca3a44 4347static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N
4348
4349static ssize_t
fd01b88c 4350sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4351{
13ae864b 4352 unsigned long long max_sectors, resync;
88202a0c 4353
acb180b0
N
4354 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4355 return sprintf(page, "none\n");
4356
72f36d59
N
4357 if (mddev->curr_resync == 1 ||
4358 mddev->curr_resync == 2)
4359 return sprintf(page, "delayed\n");
4360
c804cdec
N
4361 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4362 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 4363 max_sectors = mddev->resync_max_sectors;
88202a0c 4364 else
58c0fed4 4365 max_sectors = mddev->dev_sectors;
88202a0c 4366
acb180b0 4367 resync = mddev->curr_resync_completed;
13ae864b 4368 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N
4369}
4370
80ca3a44 4371static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4372
5e96ee65 4373static ssize_t
fd01b88c 4374min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB
4375{
4376 return sprintf(page, "%llu\n",
4377 (unsigned long long)mddev->resync_min);
4378}
4379static ssize_t
fd01b88c 4380min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB
4381{
4382 unsigned long long min;
b29bebd6 4383 if (kstrtoull(buf, 10, &min))
5e96ee65
NB
4384 return -EINVAL;
4385 if (min > mddev->resync_max)
4386 return -EINVAL;
4387 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4388 return -EBUSY;
4389
4390 /* Must be a multiple of chunk_size */
9d8f0363 4391 if (mddev->chunk_sectors) {
2ac06c33 4392 sector_t temp = min;
9d8f0363 4393 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB
4394 return -EINVAL;
4395 }
4396 mddev->resync_min = min;
4397
4398 return len;
4399}
4400
4401static struct md_sysfs_entry md_min_sync =
4402__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4403
c6207277 4404static ssize_t
fd01b88c 4405max_sync_show(struct mddev *mddev, char *page)
c6207277
N
4406{
4407 if (mddev->resync_max == MaxSector)
4408 return sprintf(page, "max\n");
4409 else
4410 return sprintf(page, "%llu\n",
4411 (unsigned long long)mddev->resync_max);
4412}
4413static ssize_t
fd01b88c 4414max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N
4415{
4416 if (strncmp(buf, "max", 3) == 0)
4417 mddev->resync_max = MaxSector;
4418 else {
5e96ee65 4419 unsigned long long max;
b29bebd6 4420 if (kstrtoull(buf, 10, &max))
5e96ee65
NB
4421 return -EINVAL;
4422 if (max < mddev->resync_min)
c6207277
N
4423 return -EINVAL;
4424 if (max < mddev->resync_max &&
4d484a4a 4425 mddev->ro == 0 &&
c6207277
N
4426 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4427 return -EBUSY;
4428
4429 /* Must be a multiple of chunk_size */
9d8f0363 4430 if (mddev->chunk_sectors) {
2ac06c33 4431 sector_t temp = max;
9d8f0363 4432 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N
4433 return -EINVAL;
4434 }
4435 mddev->resync_max = max;
4436 }
4437 wake_up(&mddev->recovery_wait);
4438 return len;
4439}
4440
4441static struct md_sysfs_entry md_max_sync =
4442__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4443
e464eafd 4444static ssize_t
fd01b88c 4445suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N
4446{
4447 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4448}
4449
4450static ssize_t
fd01b88c 4451suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N
4452{
4453 char *e;
4454 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4455 unsigned long long old = mddev->suspend_lo;
e464eafd 4456
f72ffdd6 4457 if (mddev->pers == NULL ||
b8d966ef 4458 mddev->pers->quiesce == NULL)
e464eafd
N
4459 return -EINVAL;
4460 if (buf == e || (*e && *e != '\n'))
4461 return -EINVAL;
23ddff37
N
4462
4463 mddev->suspend_lo = new;
4464 if (new >= old)
4465 /* Shrinking suspended region */
e464eafd 4466 mddev->pers->quiesce(mddev, 2);
23ddff37
N
4467 else {
4468 /* Expanding suspended region - need to wait */
4469 mddev->pers->quiesce(mddev, 1);
4470 mddev->pers->quiesce(mddev, 0);
4471 }
4472 return len;
e464eafd
N
4473}
4474static struct md_sysfs_entry md_suspend_lo =
4475__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4476
e464eafd 4477static ssize_t
fd01b88c 4478suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N
4479{
4480 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4481}
4482
4483static ssize_t
fd01b88c 4484suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N
4485{
4486 char *e;
4487 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4488 unsigned long long old = mddev->suspend_hi;
e464eafd 4489
b8d966ef
N
4490 if (mddev->pers == NULL ||
4491 mddev->pers->quiesce == NULL)
e464eafd
N
4492 return -EINVAL;
4493 if (buf == e || (*e && *e != '\n'))
4494 return -EINVAL;
23ddff37
N
4495
4496 mddev->suspend_hi = new;
4497 if (new <= old)
4498 /* Shrinking suspended region */
4499 mddev->pers->quiesce(mddev, 2);
4500 else {
4501 /* Expanding suspended region - need to wait */
e464eafd
N
4502 mddev->pers->quiesce(mddev, 1);
4503 mddev->pers->quiesce(mddev, 0);
23ddff37
N
4504 }
4505 return len;
e464eafd
N
4506}
4507static struct md_sysfs_entry md_suspend_hi =
4508__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4509
08a02ecd 4510static ssize_t
fd01b88c 4511reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N
4512{
4513 if (mddev->reshape_position != MaxSector)
4514 return sprintf(page, "%llu\n",
4515 (unsigned long long)mddev->reshape_position);
4516 strcpy(page, "none\n");
4517 return 5;
4518}
4519
4520static ssize_t
fd01b88c 4521reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd 4522{
c6563a8c 4523 struct md_rdev *rdev;
08a02ecd
N
4524 char *e;
4525 unsigned long long new = simple_strtoull(buf, &e, 10);
4526 if (mddev->pers)
4527 return -EBUSY;
4528 if (buf == e || (*e && *e != '\n'))
4529 return -EINVAL;
4530 mddev->reshape_position = new;
4531 mddev->delta_disks = 0;
2c810cdd 4532 mddev->reshape_backwards = 0;
08a02ecd
N
4533 mddev->new_level = mddev->level;
4534 mddev->new_layout = mddev->layout;
664e7c41 4535 mddev->new_chunk_sectors = mddev->chunk_sectors;
c6563a8c
N
4536 rdev_for_each(rdev, mddev)
4537 rdev->new_data_offset = rdev->data_offset;
08a02ecd
N
4538 return len;
4539}
4540
4541static struct md_sysfs_entry md_reshape_position =
4542__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4543 reshape_position_store);
4544
2c810cdd
N
4545static ssize_t
4546reshape_direction_show(struct mddev *mddev, char *page)
4547{
4548 return sprintf(page, "%s\n",
4549 mddev->reshape_backwards ? "backwards" : "forwards");
4550}
4551
4552static ssize_t
4553reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4554{
4555 int backwards = 0;
4556 if (cmd_match(buf, "forwards"))
4557 backwards = 0;
4558 else if (cmd_match(buf, "backwards"))
4559 backwards = 1;
4560 else
4561 return -EINVAL;
4562 if (mddev->reshape_backwards == backwards)
4563 return len;
4564
4565 /* check if we are allowed to change */
4566 if (mddev->delta_disks)
4567 return -EBUSY;
4568
4569 if (mddev->persistent &&
4570 mddev->major_version == 0)
4571 return -EINVAL;
4572
4573 mddev->reshape_backwards = backwards;
4574 return len;
4575}
4576
4577static struct md_sysfs_entry md_reshape_direction =
4578__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4579 reshape_direction_store);
4580
b522adcd 4581static ssize_t
fd01b88c 4582array_size_show(struct mddev *mddev, char *page)
b522adcd
DW
4583{
4584 if (mddev->external_size)
4585 return sprintf(page, "%llu\n",
4586 (unsigned long long)mddev->array_sectors/2);
4587 else
4588 return sprintf(page, "default\n");
4589}
4590
4591static ssize_t
fd01b88c 4592array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW
4593{
4594 sector_t sectors;
4595
4596 if (strncmp(buf, "default", 7) == 0) {
4597 if (mddev->pers)
4598 sectors = mddev->pers->size(mddev, 0, 0);
4599 else
4600 sectors = mddev->array_sectors;
4601
4602 mddev->external_size = 0;
4603 } else {
4604 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4605 return -EINVAL;
4606 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4607 return -E2BIG;
b522adcd
DW
4608
4609 mddev->external_size = 1;
4610 }
4611
4612 mddev->array_sectors = sectors;
cbe6ef1d
N
4613 if (mddev->pers) {
4614 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4615 revalidate_disk(mddev->gendisk);
cbe6ef1d 4616 }
b522adcd
DW
4617 return len;
4618}
4619
4620static struct md_sysfs_entry md_array_size =
4621__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4622 array_size_store);
e464eafd 4623
eae1701f
N
4624static struct attribute *md_default_attrs[] = {
4625 &md_level.attr,
d4dbd025 4626 &md_layout.attr,
eae1701f 4627 &md_raid_disks.attr,
3b34380a 4628 &md_chunk_size.attr,
a35b0d69 4629 &md_size.attr,
a94213b1 4630 &md_resync_start.attr,
8bb93aac 4631 &md_metadata.attr,
6d7ff738 4632 &md_new_device.attr,
16f17b39 4633 &md_safe_delay.attr,
9e653b63 4634 &md_array_state.attr,
08a02ecd 4635 &md_reshape_position.attr,
2c810cdd 4636 &md_reshape_direction.attr,
b522adcd 4637 &md_array_size.attr,
1e50915f 4638 &max_corr_read_errors.attr,
411036fa
N
4639 NULL,
4640};
4641
4642static struct attribute *md_redundancy_attrs[] = {
24dd469d 4643 &md_scan_mode.attr,
c4a39551 4644 &md_last_scan_mode.attr,
9d88883e 4645 &md_mismatches.attr,
88202a0c
N
4646 &md_sync_min.attr,
4647 &md_sync_max.attr,
4648 &md_sync_speed.attr,
90b08710 4649 &md_sync_force_parallel.attr,
88202a0c 4650 &md_sync_completed.attr,
5e96ee65 4651 &md_min_sync.attr,
c6207277 4652 &md_max_sync.attr,
e464eafd
N
4653 &md_suspend_lo.attr,
4654 &md_suspend_hi.attr,
9b1d1dac 4655 &md_bitmap.attr,
d7f3d291 4656 &md_degraded.attr,
eae1701f
N
4657 NULL,
4658};
411036fa
N
4659static struct attribute_group md_redundancy_group = {
4660 .name = NULL,
4661 .attrs = md_redundancy_attrs,
4662};
4663
eae1701f
N
4664static ssize_t
4665md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4666{
4667 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4668 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4669 ssize_t rv;
eae1701f
N
4670
4671 if (!entry->show)
4672 return -EIO;
af8a2434
N
4673 spin_lock(&all_mddevs_lock);
4674 if (list_empty(&mddev->all_mddevs)) {
4675 spin_unlock(&all_mddevs_lock);
4676 return -EBUSY;
4677 }
4678 mddev_get(mddev);
4679 spin_unlock(&all_mddevs_lock);
4680
5dc5cf7d
IM
4681 rv = mddev_lock(mddev);
4682 if (!rv) {
4683 rv = entry->show(mddev, page);
4684 mddev_unlock(mddev);
4685 }
af8a2434 4686 mddev_put(mddev);
96de1e66 4687 return rv;
eae1701f
N
4688}
4689
4690static ssize_t
4691md_attr_store(struct kobject *kobj, struct attribute *attr,
4692 const char *page, size_t length)
4693{
4694 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4695 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4696 ssize_t rv;
eae1701f
N
4697
4698 if (!entry->store)
4699 return -EIO;
67463acb
N
4700 if (!capable(CAP_SYS_ADMIN))
4701 return -EACCES;
af8a2434
N
4702 spin_lock(&all_mddevs_lock);
4703 if (list_empty(&mddev->all_mddevs)) {
4704 spin_unlock(&all_mddevs_lock);
4705 return -EBUSY;
4706 }
4707 mddev_get(mddev);
4708 spin_unlock(&all_mddevs_lock);
a7a3f08d
N
4709 if (entry->store == new_dev_store)
4710 flush_workqueue(md_misc_wq);
5dc5cf7d
IM
4711 rv = mddev_lock(mddev);
4712 if (!rv) {
4713 rv = entry->store(mddev, page, length);
4714 mddev_unlock(mddev);
4715 }
af8a2434 4716 mddev_put(mddev);
96de1e66 4717 return rv;
eae1701f
N
4718}
4719
4720static void md_free(struct kobject *ko)
4721{
fd01b88c 4722 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N
4723
4724 if (mddev->sysfs_state)
4725 sysfs_put(mddev->sysfs_state);
4726
4727 if (mddev->gendisk) {
4728 del_gendisk(mddev->gendisk);
4729 put_disk(mddev->gendisk);
4730 }
4731 if (mddev->queue)
4732 blk_cleanup_queue(mddev->queue);
4733
eae1701f
N
4734 kfree(mddev);
4735}
4736
52cf25d0 4737static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N
4738 .show = md_attr_show,
4739 .store = md_attr_store,
4740};
4741static struct kobj_type md_ktype = {
4742 .release = md_free,
4743 .sysfs_ops = &md_sysfs_ops,
4744 .default_attrs = md_default_attrs,
4745};
4746
1da177e4
LT
4747int mdp_major = 0;
4748
5fd3a17e
DW
4749static void mddev_delayed_delete(struct work_struct *ws)
4750{
fd01b88c 4751 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4752
43a70507 4753 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW
4754 kobject_del(&mddev->kobj);
4755 kobject_put(&mddev->kobj);
4756}
4757
efeb53c0 4758static int md_alloc(dev_t dev, char *name)
1da177e4 4759{
48c9c27b 4760 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4761 struct mddev *mddev = mddev_find(dev);
1da177e4 4762 struct gendisk *disk;
efeb53c0
N
4763 int partitioned;
4764 int shift;
4765 int unit;
3830c62f 4766 int error;
1da177e4
LT
4767
4768 if (!mddev)
efeb53c0
N
4769 return -ENODEV;
4770
4771 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4772 shift = partitioned ? MdpMinorShift : 0;
4773 unit = MINOR(mddev->unit) >> shift;
1da177e4 4774
e804ac78
TH
4775 /* wait for any previous instance of this device to be
4776 * completely removed (mddev_delayed_delete).
d3374825 4777 */
e804ac78 4778 flush_workqueue(md_misc_wq);
d3374825 4779
48c9c27b 4780 mutex_lock(&disks_mutex);
0909dc44
N
4781 error = -EEXIST;
4782 if (mddev->gendisk)
4783 goto abort;
efeb53c0
N
4784
4785 if (name) {
4786 /* Need to ensure that 'name' is not a duplicate.
4787 */
fd01b88c 4788 struct mddev *mddev2;
efeb53c0
N
4789 spin_lock(&all_mddevs_lock);
4790
4791 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4792 if (mddev2->gendisk &&
4793 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4794 spin_unlock(&all_mddevs_lock);
0909dc44 4795 goto abort;
efeb53c0
N
4796 }
4797 spin_unlock(&all_mddevs_lock);
1da177e4 4798 }
8b765398 4799
0909dc44 4800 error = -ENOMEM;
8b765398 4801 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N
4802 if (!mddev->queue)
4803 goto abort;
409c57f3
N
4804 mddev->queue->queuedata = mddev;
4805
409c57f3 4806 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4807 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4808
1da177e4
LT
4809 disk = alloc_disk(1 << shift);
4810 if (!disk) {
8b765398
N
4811 blk_cleanup_queue(mddev->queue);
4812 mddev->queue = NULL;
0909dc44 4813 goto abort;
1da177e4 4814 }
efeb53c0 4815 disk->major = MAJOR(mddev->unit);
1da177e4 4816 disk->first_minor = unit << shift;
efeb53c0
N
4817 if (name)
4818 strcpy(disk->disk_name, name);
4819 else if (partitioned)
1da177e4 4820 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4821 else
1da177e4 4822 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT
4823 disk->fops = &md_fops;
4824 disk->private_data = mddev;
4825 disk->queue = mddev->queue;
b0140891 4826 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4827 /* Allow extended partitions. This makes the
d3374825 4828 * 'mdp' device redundant, but we can't really
92850bbd
N
4829 * remove it now.
4830 */
4831 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4832 mddev->gendisk = disk;
b0140891
N
4833 /* As soon as we call add_disk(), another thread could get
4834 * through to md_open, so make sure it doesn't get too far
4835 */
4836 mutex_lock(&mddev->open_mutex);
4837 add_disk(disk);
4838
ed9e1982
TH
4839 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4840 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N
4841 if (error) {
4842 /* This isn't possible, but as kobject_init_and_add is marked
4843 * __must_check, we must do something with the result
4844 */
5e55e2f5
N
4845 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4846 disk->disk_name);
0909dc44
N
4847 error = 0;
4848 }
00bcb4ac
N
4849 if (mddev->kobj.sd &&
4850 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4851 printk(KERN_DEBUG "pointless warning\n");
b0140891 4852 mutex_unlock(&mddev->open_mutex);
0909dc44
N
4853 abort:
4854 mutex_unlock(&disks_mutex);
00bcb4ac 4855 if (!error && mddev->kobj.sd) {
3830c62f 4856 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4857 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4858 }
d3374825 4859 mddev_put(mddev);
0909dc44 4860 return error;
efeb53c0
N
4861}
4862
4863static struct kobject *md_probe(dev_t dev, int *part, void *data)
4864{
4865 md_alloc(dev, NULL);
1da177e4
LT
4866 return NULL;
4867}
4868
efeb53c0
N
4869static int add_named_array(const char *val, struct kernel_param *kp)
4870{
4871 /* val must be "md_*" where * is not all digits.
4872 * We allocate an array with a large free minor number, and
4873 * set the name to val. val must not already be an active name.
4874 */
4875 int len = strlen(val);
4876 char buf[DISK_NAME_LEN];
4877
4878 while (len && val[len-1] == '\n')
4879 len--;
4880 if (len >= DISK_NAME_LEN)
4881 return -E2BIG;
4882 strlcpy(buf, val, len+1);
4883 if (strncmp(buf, "md_", 3) != 0)
4884 return -EINVAL;
4885 return md_alloc(0, buf);
4886}
4887
1da177e4
LT
4888static void md_safemode_timeout(unsigned long data)
4889{
fd01b88c 4890 struct mddev *mddev = (struct mddev *) data;
1da177e4 4891
0fd62b86
NB
4892 if (!atomic_read(&mddev->writes_pending)) {
4893 mddev->safemode = 1;
4894 if (mddev->external)
00bcb4ac 4895 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4896 }
1da177e4
LT
4897 md_wakeup_thread(mddev->thread);
4898}
4899
6ff8d8ec 4900static int start_dirty_degraded;
1da177e4 4901
fd01b88c 4902int md_run(struct mddev *mddev)
1da177e4 4903{
2604b703 4904 int err;
3cb03002 4905 struct md_rdev *rdev;
84fc4b56 4906 struct md_personality *pers;
1da177e4 4907
a757e64c
N
4908 if (list_empty(&mddev->disks))
4909 /* cannot run an array with no devices.. */
1da177e4 4910 return -EINVAL;
1da177e4
LT
4911
4912 if (mddev->pers)
4913 return -EBUSY;
bb4f1e9d
N
4914 /* Cannot run until previous stop completes properly */
4915 if (mddev->sysfs_active)
4916 return -EBUSY;
b6eb127d 4917
1da177e4
LT
4918 /*
4919 * Analyze all RAID superblock(s)
4920 */
1ec4a939
N
4921 if (!mddev->raid_disks) {
4922 if (!mddev->persistent)
4923 return -EINVAL;
a757e64c 4924 analyze_sbs(mddev);
1ec4a939 4925 }
1da177e4 4926
d9d166c2
N
4927 if (mddev->level != LEVEL_NONE)
4928 request_module("md-level-%d", mddev->level);
4929 else if (mddev->clevel[0])
4930 request_module("md-%s", mddev->clevel);
1da177e4
LT
4931
4932 /*
4933 * Drop all container device buffers, from now on
4934 * the only valid external interface is through the md
4935 * device.
1da177e4 4936 */
dafb20fa 4937 rdev_for_each(rdev, mddev) {
b2d444d7 4938 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT
4939 continue;
4940 sync_blockdev(rdev->bdev);
f98393a6 4941 invalidate_bdev(rdev->bdev);
f0d76d70
N
4942
4943 /* perform some consistency tests on the device.
4944 * We don't want the data to overlap the metadata,
58c0fed4 4945 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4946 */
a6ff7e08
JB
4947 if (rdev->meta_bdev) {
4948 /* Nothing to check */;
4949 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN
4950 if (mddev->dev_sectors &&
4951 rdev->data_offset + mddev->dev_sectors
0f420358 4952 > rdev->sb_start) {
f0d76d70
N
4953 printk("md: %s: data overlaps metadata\n",
4954 mdname(mddev));
4955 return -EINVAL;
4956 }
4957 } else {
0f420358 4958 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N
4959 > rdev->data_offset) {
4960 printk("md: %s: metadata overlaps data\n",
4961 mdname(mddev));
4962 return -EINVAL;
4963 }
4964 }
00bcb4ac 4965 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT
4966 }
4967
a167f663 4968 if (mddev->bio_set == NULL)
395c72a7 4969 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
a167f663 4970
1da177e4 4971 spin_lock(&pers_lock);
d9d166c2 4972 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4973 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4974 spin_unlock(&pers_lock);
d9d166c2
N
4975 if (mddev->level != LEVEL_NONE)
4976 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4977 mddev->level);
4978 else
4979 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4980 mddev->clevel);
1da177e4
LT
4981 return -EINVAL;
4982 }
2604b703 4983 mddev->pers = pers;
1da177e4 4984 spin_unlock(&pers_lock);
34817e8c
N
4985 if (mddev->level != pers->level) {
4986 mddev->level = pers->level;
4987 mddev->new_level = pers->level;
4988 }
d9d166c2 4989 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4990
f6705578 4991 if (mddev->reshape_position != MaxSector &&
63c70c4f 4992 pers->start_reshape == NULL) {
f6705578
N
4993 /* This personality cannot handle reshaping... */
4994 mddev->pers = NULL;
4995 module_put(pers->owner);
4996 return -EINVAL;
4997 }
4998
7dd5e7c3
N
4999 if (pers->sync_request) {
5000 /* Warn if this is a potentially silly
5001 * configuration.
5002 */
5003 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5004 struct md_rdev *rdev2;
7dd5e7c3 5005 int warned = 0;
159ec1fc 5006
dafb20fa
N
5007 rdev_for_each(rdev, mddev)
5008 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N
5009 if (rdev < rdev2 &&
5010 rdev->bdev->bd_contains ==
5011 rdev2->bdev->bd_contains) {
5012 printk(KERN_WARNING
5013 "%s: WARNING: %s appears to be"
5014 " on the same physical disk as"
5015 " %s.\n",
5016 mdname(mddev),
5017 bdevname(rdev->bdev,b),
5018 bdevname(rdev2->bdev,b2));
5019 warned = 1;
5020 }
5021 }
159ec1fc 5022
7dd5e7c3
N
5023 if (warned)
5024 printk(KERN_WARNING
5025 "True protection against single-disk"
5026 " failure might be compromised.\n");
5027 }
5028
657390d2 5029 mddev->recovery = 0;
58c0fed4
AN
5030 /* may be over-ridden by personality */
5031 mddev->resync_max_sectors = mddev->dev_sectors;
5032
6ff8d8ec 5033 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 5034
0f9552b5 5035 if (start_readonly && mddev->ro == 0)
f91de92e
N
5036 mddev->ro = 2; /* read-only, but switch on first write */
5037
b15c2e57 5038 err = mddev->pers->run(mddev);
13e53df3
AN
5039 if (err)
5040 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW
5041 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5042 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5043 " but 'external_size' not in effect?\n", __func__);
5044 printk(KERN_ERR
5045 "md: invalid array_size %llu > default size %llu\n",
5046 (unsigned long long)mddev->array_sectors / 2,
5047 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5048 err = -EINVAL;
5049 mddev->pers->stop(mddev);
5050 }
ef99bf48
N
5051 if (err == 0 && mddev->pers->sync_request &&
5052 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
b15c2e57
N
5053 err = bitmap_create(mddev);
5054 if (err) {
5055 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5056 mdname(mddev), err);
5057 mddev->pers->stop(mddev);
5058 }
5059 }
1da177e4 5060 if (err) {
1da177e4
LT
5061 module_put(mddev->pers->owner);
5062 mddev->pers = NULL;
32a7627c
N
5063 bitmap_destroy(mddev);
5064 return err;
1da177e4 5065 }
5e55e2f5 5066 if (mddev->pers->sync_request) {
00bcb4ac
N
5067 if (mddev->kobj.sd &&
5068 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N
5069 printk(KERN_WARNING
5070 "md: cannot register extra attributes for %s\n",
5071 mdname(mddev));
00bcb4ac 5072 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 5073 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N
5074 mddev->ro = 0;
5075
f72ffdd6 5076 atomic_set(&mddev->writes_pending,0);
1e50915f
RB
5077 atomic_set(&mddev->max_corr_read_errors,
5078 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT
5079 mddev->safemode = 0;
5080 mddev->safemode_timer.function = md_safemode_timeout;
5081 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 5082 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 5083 mddev->in_sync = 1;
0ca69886
N
5084 smp_wmb();
5085 mddev->ready = 1;
dafb20fa 5086 rdev_for_each(rdev, mddev)
36fad858
NK
5087 if (rdev->raid_disk >= 0)
5088 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 5089 /* failure here is OK */;
f72ffdd6 5090
1da177e4 5091 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f72ffdd6 5092
7a0a5355 5093 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 5094 md_update_sb(mddev, 0);
1da177e4 5095
d7603b7e 5096 md_new_event(mddev);
00bcb4ac
N
5097 sysfs_notify_dirent_safe(mddev->sysfs_state);
5098 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 5099 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT
5100 return 0;
5101}
390ee602 5102EXPORT_SYMBOL_GPL(md_run);
1da177e4 5103
fd01b88c 5104static int do_md_run(struct mddev *mddev)
fe60b014
N
5105{
5106 int err;
5107
5108 err = md_run(mddev);
5109 if (err)
5110 goto out;
69e51b44
N
5111 err = bitmap_load(mddev);
5112 if (err) {
5113 bitmap_destroy(mddev);
5114 goto out;
5115 }
0fd018af
JB
5116
5117 md_wakeup_thread(mddev->thread);
5118 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5119
fe60b014
N
5120 set_capacity(mddev->gendisk, mddev->array_sectors);
5121 revalidate_disk(mddev->gendisk);
f0b4f7e2 5122 mddev->changed = 1;
fe60b014
N
5123 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5124out:
5125 return err;
5126}
5127
fd01b88c 5128static int restart_array(struct mddev *mddev)
1da177e4
LT
5129{
5130 struct gendisk *disk = mddev->gendisk;
1da177e4 5131
80fab1d7 5132 /* Complain if it has no devices */
1da177e4 5133 if (list_empty(&mddev->disks))
80fab1d7
AN
5134 return -ENXIO;
5135 if (!mddev->pers)
5136 return -EINVAL;
5137 if (!mddev->ro)
5138 return -EBUSY;
5139 mddev->safemode = 0;
5140 mddev->ro = 0;
5141 set_disk_ro(disk, 0);
5142 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5143 mdname(mddev));
5144 /* Kick recovery or resync if necessary */
5145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5146 md_wakeup_thread(mddev->thread);
5147 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5148 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5149 return 0;
1da177e4
LT
5150}
5151
fd01b88c 5152static void md_clean(struct mddev *mddev)
6177b472
N
5153{
5154 mddev->array_sectors = 0;
5155 mddev->external_size = 0;
5156 mddev->dev_sectors = 0;
5157 mddev->raid_disks = 0;
5158 mddev->recovery_cp = 0;
5159 mddev->resync_min = 0;
5160 mddev->resync_max = MaxSector;
5161 mddev->reshape_position = MaxSector;
5162 mddev->external = 0;
5163 mddev->persistent = 0;
5164 mddev->level = LEVEL_NONE;
5165 mddev->clevel[0] = 0;
5166 mddev->flags = 0;
5167 mddev->ro = 0;
5168 mddev->metadata_type[0] = 0;
5169 mddev->chunk_sectors = 0;
5170 mddev->ctime = mddev->utime = 0;
5171 mddev->layout = 0;
5172 mddev->max_disks = 0;
5173 mddev->events = 0;
a8707c08 5174 mddev->can_decrease_events = 0;
6177b472 5175 mddev->delta_disks = 0;
2c810cdd 5176 mddev->reshape_backwards = 0;
6177b472
N
5177 mddev->new_level = LEVEL_NONE;
5178 mddev->new_layout = 0;
5179 mddev->new_chunk_sectors = 0;
5180 mddev->curr_resync = 0;
7f7583d4 5181 atomic64_set(&mddev->resync_mismatches, 0);
6177b472
N
5182 mddev->suspend_lo = mddev->suspend_hi = 0;
5183 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5184 mddev->recovery = 0;
5185 mddev->in_sync = 0;
f0b4f7e2 5186 mddev->changed = 0;
6177b472 5187 mddev->degraded = 0;
6177b472 5188 mddev->safemode = 0;
050b6615 5189 mddev->merge_check_needed = 0;
6177b472
N
5190 mddev->bitmap_info.offset = 0;
5191 mddev->bitmap_info.default_offset = 0;
6409bb05 5192 mddev->bitmap_info.default_space = 0;
6177b472
N
5193 mddev->bitmap_info.chunksize = 0;
5194 mddev->bitmap_info.daemon_sleep = 0;
5195 mddev->bitmap_info.max_write_behind = 0;
5196}
5197
fd01b88c 5198static void __md_stop_writes(struct mddev *mddev)
a047e125 5199{
6b6204ee 5200 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
a047e125 5201 if (mddev->sync_thread) {
a047e125 5202 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
a91d5ac0 5203 md_reap_sync_thread(mddev);
a047e125
N
5204 }
5205
5206 del_timer_sync(&mddev->safemode_timer);
5207
5208 bitmap_flush(mddev);
5209 md_super_wait(mddev);
5210
b6d428c6 5211 if (mddev->ro == 0 &&
7a0a5355 5212 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
a047e125
N
5213 /* mark array as shutdown cleanly */
5214 mddev->in_sync = 1;
5215 md_update_sb(mddev, 1);
5216 }
5217}
defad61a 5218
fd01b88c 5219void md_stop_writes(struct mddev *mddev)
defad61a 5220{
29f097c4 5221 mddev_lock_nointr(mddev);
defad61a
N
5222 __md_stop_writes(mddev);
5223 mddev_unlock(mddev);
5224}
390ee602 5225EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5226
5eff3c43 5227static void __md_stop(struct mddev *mddev)
6177b472 5228{
0ca69886 5229 mddev->ready = 0;
6177b472
N
5230 mddev->pers->stop(mddev);
5231 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5232 mddev->to_remove = &md_redundancy_group;
5233 module_put(mddev->pers->owner);
5234 mddev->pers = NULL;
cca9cf90 5235 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5236}
5eff3c43
N
5237
5238void md_stop(struct mddev *mddev)
5239{
5240 /* stop the array and free an attached data structures.
5241 * This is called from dm-raid
5242 */
5243 __md_stop(mddev);
5244 bitmap_destroy(mddev);
5245 if (mddev->bio_set)
5246 bioset_free(mddev->bio_set);
5247}
5248
390ee602 5249EXPORT_SYMBOL_GPL(md_stop);
6177b472 5250
a05b7ea0 5251static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
a4bd82d0
N
5252{
5253 int err = 0;
30b8feb7
N
5254 int did_freeze = 0;
5255
5256 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5257 did_freeze = 1;
5258 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5259 md_wakeup_thread(mddev->thread);
5260 }
5261 if (mddev->sync_thread) {
5262 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5263 /* Thread might be blocked waiting for metadata update
5264 * which will now never happen */
5265 wake_up_process(mddev->sync_thread->tsk);
5266 }
5267 mddev_unlock(mddev);
5268 wait_event(resync_wait, mddev->sync_thread == NULL);
5269 mddev_lock_nointr(mddev);
5270
a4bd82d0 5271 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5272 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N
5273 mddev->sync_thread ||
5274 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
a4bd82d0 5275 printk("md: %s still in use.\n",mdname(mddev));
30b8feb7
N
5276 if (did_freeze) {
5277 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5278 md_wakeup_thread(mddev->thread);
5279 }
a4bd82d0
N
5280 err = -EBUSY;
5281 goto out;
5282 }
5283 if (mddev->pers) {
defad61a 5284 __md_stop_writes(mddev);
a4bd82d0
N
5285
5286 err = -ENXIO;
5287 if (mddev->ro==1)
5288 goto out;
5289 mddev->ro = 1;
5290 set_disk_ro(mddev->gendisk, 1);
5291 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5292 sysfs_notify_dirent_safe(mddev->sysfs_state);
30b8feb7 5293 err = 0;
a4bd82d0
N
5294 }
5295out:
5296 mutex_unlock(&mddev->open_mutex);
5297 return err;
5298}
5299
9e653b63
N
5300/* mode:
5301 * 0 - completely stop and dis-assemble array
9e653b63
N
5302 * 2 - stop but do not disassemble array
5303 */
f72ffdd6 5304static int do_md_stop(struct mddev *mddev, int mode,
a05b7ea0 5305 struct block_device *bdev)
1da177e4 5306{
1da177e4 5307 struct gendisk *disk = mddev->gendisk;
3cb03002 5308 struct md_rdev *rdev;
30b8feb7
N
5309 int did_freeze = 0;
5310
5311 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5312 did_freeze = 1;
5313 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5314 md_wakeup_thread(mddev->thread);
5315 }
5316 if (mddev->sync_thread) {
5317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5318 /* Thread might be blocked waiting for metadata update
5319 * which will now never happen */
5320 wake_up_process(mddev->sync_thread->tsk);
5321 }
5322 mddev_unlock(mddev);
5323 wait_event(resync_wait, mddev->sync_thread == NULL);
5324 mddev_lock_nointr(mddev);
1da177e4 5325
c8c00a69 5326 mutex_lock(&mddev->open_mutex);
9ba3b7f5 5327 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
30b8feb7
N
5328 mddev->sysfs_active ||
5329 mddev->sync_thread ||
5330 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
df5b20cf 5331 printk("md: %s still in use.\n",mdname(mddev));
6e17b027 5332 mutex_unlock(&mddev->open_mutex);
30b8feb7
N
5333 if (did_freeze) {
5334 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5335 md_wakeup_thread(mddev->thread);
5336 }
260fa034
N
5337 return -EBUSY;
5338 }
6e17b027 5339 if (mddev->pers) {
a4bd82d0
N
5340 if (mddev->ro)
5341 set_disk_ro(disk, 0);
409c57f3 5342
defad61a 5343 __md_stop_writes(mddev);
5eff3c43 5344 __md_stop(mddev);
a4bd82d0 5345 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5346 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5347
a4bd82d0 5348 /* tell userspace to handle 'inactive' */
00bcb4ac 5349 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5350
dafb20fa 5351 rdev_for_each(rdev, mddev)
36fad858
NK
5352 if (rdev->raid_disk >= 0)
5353 sysfs_unlink_rdev(mddev, rdev);
c4647292 5354
a4bd82d0 5355 set_capacity(disk, 0);
6e17b027 5356 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5357 mddev->changed = 1;
a4bd82d0 5358 revalidate_disk(disk);
0d4ca600 5359
a4bd82d0
N
5360 if (mddev->ro)
5361 mddev->ro = 0;
6e17b027
N
5362 } else
5363 mutex_unlock(&mddev->open_mutex);
1da177e4
LT
5364 /*
5365 * Free resources if final stop
5366 */
9e653b63 5367 if (mode == 0) {
1da177e4
LT
5368 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5369
978f946b 5370 bitmap_destroy(mddev);
c3d9714e 5371 if (mddev->bitmap_info.file) {
c3d9714e
N
5372 fput(mddev->bitmap_info.file);
5373 mddev->bitmap_info.file = NULL;
978f946b 5374 }
c3d9714e 5375 mddev->bitmap_info.offset = 0;
978f946b 5376
1da177e4
LT
5377 export_array(mddev);
5378
6177b472 5379 md_clean(mddev);
934d9c23 5380 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N
5381 if (mddev->hold_active == UNTIL_STOP)
5382 mddev->hold_active = 0;
a4bd82d0 5383 }
3f9d99c1 5384 blk_integrity_unregister(disk);
d7603b7e 5385 md_new_event(mddev);
00bcb4ac 5386 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5387 return 0;
1da177e4
LT
5388}
5389
fdee8ae4 5390#ifndef MODULE
fd01b88c 5391static void autorun_array(struct mddev *mddev)
1da177e4 5392{
3cb03002 5393 struct md_rdev *rdev;
1da177e4
LT
5394 int err;
5395
a757e64c 5396 if (list_empty(&mddev->disks))
1da177e4 5397 return;
1da177e4
LT
5398
5399 printk(KERN_INFO "md: running: ");
5400
dafb20fa 5401 rdev_for_each(rdev, mddev) {
1da177e4
LT
5402 char b[BDEVNAME_SIZE];
5403 printk("<%s>", bdevname(rdev->bdev,b));
5404 }
5405 printk("\n");
5406
d710e138 5407 err = do_md_run(mddev);
1da177e4
LT
5408 if (err) {
5409 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
a05b7ea0 5410 do_md_stop(mddev, 0, NULL);
1da177e4
LT
5411 }
5412}
5413
5414/*
5415 * lets try to run arrays based on all disks that have arrived
5416 * until now. (those are in pending_raid_disks)
5417 *
5418 * the method: pick the first pending disk, collect all disks with
5419 * the same UUID, remove all from the pending list and put them into
5420 * the 'same_array' list. Then order this list based on superblock
5421 * update time (freshest comes first), kick out 'old' disks and
5422 * compare superblocks. If everything's fine then run it.
5423 *
5424 * If "unit" is allocated, then bump its reference count
5425 */
5426static void autorun_devices(int part)
5427{
3cb03002 5428 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5429 struct mddev *mddev;
1da177e4
LT
5430 char b[BDEVNAME_SIZE];
5431
5432 printk(KERN_INFO "md: autorun ...\n");
5433 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5434 int unit;
1da177e4 5435 dev_t dev;
ad01c9e3 5436 LIST_HEAD(candidates);
1da177e4 5437 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5438 struct md_rdev, same_set);
1da177e4
LT
5439
5440 printk(KERN_INFO "md: considering %s ...\n",
5441 bdevname(rdev0->bdev,b));
5442 INIT_LIST_HEAD(&candidates);
159ec1fc 5443 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT
5444 if (super_90_load(rdev, rdev0, 0) >= 0) {
5445 printk(KERN_INFO "md: adding %s ...\n",
5446 bdevname(rdev->bdev,b));
5447 list_move(&rdev->same_set, &candidates);
5448 }
5449 /*
5450 * now we have a set of devices, with all of them having
5451 * mostly sane superblocks. It's time to allocate the
5452 * mddev.
5453 */
e8703fe1
N
5454 if (part) {
5455 dev = MKDEV(mdp_major,
5456 rdev0->preferred_minor << MdpMinorShift);
5457 unit = MINOR(dev) >> MdpMinorShift;
5458 } else {
5459 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5460 unit = MINOR(dev);
5461 }
5462 if (rdev0->preferred_minor != unit) {
1da177e4
LT
5463 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5464 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5465 break;
5466 }
1da177e4
LT
5467
5468 md_probe(dev, NULL, NULL);
5469 mddev = mddev_find(dev);
9bbbca3a
NB
5470 if (!mddev || !mddev->gendisk) {
5471 if (mddev)
5472 mddev_put(mddev);
5473 printk(KERN_ERR
1da177e4
LT
5474 "md: cannot allocate memory for md drive.\n");
5475 break;
5476 }
f72ffdd6 5477 if (mddev_lock(mddev))
1da177e4
LT
5478 printk(KERN_WARNING "md: %s locked, cannot run\n",
5479 mdname(mddev));
5480 else if (mddev->raid_disks || mddev->major_version
5481 || !list_empty(&mddev->disks)) {
f72ffdd6 5482 printk(KERN_WARNING
1da177e4
LT
5483 "md: %s already running, cannot run %s\n",
5484 mdname(mddev), bdevname(rdev0->bdev,b));
5485 mddev_unlock(mddev);
5486 } else {
5487 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5488 mddev->persistent = 1;
159ec1fc 5489 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT
5490 list_del_init(&rdev->same_set);
5491 if (bind_rdev_to_array(rdev, mddev))
5492 export_rdev(rdev);
5493 }
5494 autorun_array(mddev);
5495 mddev_unlock(mddev);
5496 }
5497 /* on success, candidates will be empty, on error
5498 * it won't...
5499 */
159ec1fc 5500 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5501 list_del_init(&rdev->same_set);
1da177e4 5502 export_rdev(rdev);
4b80991c 5503 }
1da177e4
LT
5504 mddev_put(mddev);
5505 }
5506 printk(KERN_INFO "md: ... autorun DONE.\n");
5507}
fdee8ae4 5508#endif /* !MODULE */
1da177e4 5509
f72ffdd6 5510static int get_version(void __user *arg)
1da177e4
LT
5511{
5512 mdu_version_t ver;
5513
5514 ver.major = MD_MAJOR_VERSION;
5515 ver.minor = MD_MINOR_VERSION;
5516 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5517
5518 if (copy_to_user(arg, &ver, sizeof(ver)))
5519 return -EFAULT;
5520
5521 return 0;
5522}
5523
f72ffdd6 5524static int get_array_info(struct mddev *mddev, void __user *arg)
1da177e4
LT
5525{
5526 mdu_array_info_t info;
a9f326eb 5527 int nr,working,insync,failed,spare;
3cb03002 5528 struct md_rdev *rdev;
1da177e4 5529
1ca69c4b
N
5530 nr = working = insync = failed = spare = 0;
5531 rcu_read_lock();
5532 rdev_for_each_rcu(rdev, mddev) {
1da177e4 5533 nr++;
b2d444d7 5534 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT
5535 failed++;
5536 else {
5537 working++;
b2d444d7 5538 if (test_bit(In_sync, &rdev->flags))
f72ffdd6 5539 insync++;
1da177e4
LT
5540 else
5541 spare++;
5542 }
5543 }
1ca69c4b 5544 rcu_read_unlock();
1da177e4
LT
5545
5546 info.major_version = mddev->major_version;
5547 info.minor_version = mddev->minor_version;
5548 info.patch_version = MD_PATCHLEVEL_VERSION;
5549 info.ctime = mddev->ctime;
5550 info.level = mddev->level;
58c0fed4
AN
5551 info.size = mddev->dev_sectors / 2;
5552 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5553 info.size = -1;
1da177e4
LT
5554 info.nr_disks = nr;
5555 info.raid_disks = mddev->raid_disks;
5556 info.md_minor = mddev->md_minor;
5557 info.not_persistent= !mddev->persistent;
5558
5559 info.utime = mddev->utime;
5560 info.state = 0;
5561 if (mddev->in_sync)
5562 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5563 if (mddev->bitmap && mddev->bitmap_info.offset)
9bd35920 5564 info.state |= (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5565 info.active_disks = insync;
1da177e4
LT
5566 info.working_disks = working;
5567 info.failed_disks = failed;
5568 info.spare_disks = spare;
5569
5570 info.layout = mddev->layout;
9d8f0363 5571 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT
5572
5573 if (copy_to_user(arg, &info, sizeof(info)))
5574 return -EFAULT;
5575
5576 return 0;
5577}
5578
f72ffdd6 5579static int get_bitmap_file(struct mddev *mddev, void __user * arg)
32a7627c
N
5580{
5581 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5582 char *ptr, *buf = NULL;
5583 int err = -ENOMEM;
5584
60559da4 5585 file = kmalloc(sizeof(*file), GFP_NOIO);
2a2275d6 5586
32a7627c
N
5587 if (!file)
5588 goto out;
5589
5590 /* bitmap disabled, zero the first byte and copy out */
1ec885cd 5591 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
32a7627c
N
5592 file->pathname[0] = '\0';
5593 goto copy_out;
5594 }
5595
5596 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5597 if (!buf)
5598 goto out;
5599
1ec885cd
N
5600 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5601 buf, sizeof(file->pathname));
6bcfd601 5602 if (IS_ERR(ptr))
32a7627c
N
5603 goto out;
5604
5605 strcpy(file->pathname, ptr);
5606
5607copy_out:
5608 err = 0;
5609 if (copy_to_user(arg, file, sizeof(*file)))
5610 err = -EFAULT;
5611out:
5612 kfree(buf);
5613 kfree(file);
5614 return err;
5615}
5616
f72ffdd6 5617static int get_disk_info(struct mddev *mddev, void __user * arg)
1da177e4
LT
5618{
5619 mdu_disk_info_t info;
3cb03002 5620 struct md_rdev *rdev;
1da177e4
LT
5621
5622 if (copy_from_user(&info, arg, sizeof(info)))
5623 return -EFAULT;
5624
1ca69c4b
N
5625 rcu_read_lock();
5626 rdev = find_rdev_nr_rcu(mddev, info.number);
1da177e4
LT
5627 if (rdev) {
5628 info.major = MAJOR(rdev->bdev->bd_dev);
5629 info.minor = MINOR(rdev->bdev->bd_dev);
5630 info.raid_disk = rdev->raid_disk;
5631 info.state = 0;
b2d444d7 5632 if (test_bit(Faulty, &rdev->flags))
1da177e4 5633 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5634 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT
5635 info.state |= (1<<MD_DISK_ACTIVE);
5636 info.state |= (1<<MD_DISK_SYNC);
5637 }
8ddf9efe
N
5638 if (test_bit(WriteMostly, &rdev->flags))
5639 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT
5640 } else {
5641 info.major = info.minor = 0;
5642 info.raid_disk = -1;
5643 info.state = (1<<MD_DISK_REMOVED);
5644 }
1ca69c4b 5645 rcu_read_unlock();
1da177e4
LT
5646
5647 if (copy_to_user(arg, &info, sizeof(info)))
5648 return -EFAULT;
5649
5650 return 0;
5651}
5652
f72ffdd6 5653static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
1da177e4
LT
5654{
5655 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5656 struct md_rdev *rdev;
1da177e4
LT
5657 dev_t dev = MKDEV(info->major,info->minor);
5658
5659 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5660 return -EOVERFLOW;
5661
5662 if (!mddev->raid_disks) {
5663 int err;
5664 /* expecting a device which has a superblock */
5665 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5666 if (IS_ERR(rdev)) {
f72ffdd6 5667 printk(KERN_WARNING
1da177e4
LT
5668 "md: md_import_device returned %ld\n",
5669 PTR_ERR(rdev));
5670 return PTR_ERR(rdev);
5671 }
5672 if (!list_empty(&mddev->disks)) {
3cb03002
N
5673 struct md_rdev *rdev0
5674 = list_entry(mddev->disks.next,
5675 struct md_rdev, same_set);
a9f326eb 5676 err = super_types[mddev->major_version]
1da177e4
LT
5677 .load_super(rdev, rdev0, mddev->minor_version);
5678 if (err < 0) {
f72ffdd6 5679 printk(KERN_WARNING
1da177e4 5680 "md: %s has different UUID to %s\n",
f72ffdd6 5681 bdevname(rdev->bdev,b),
1da177e4
LT
5682 bdevname(rdev0->bdev,b2));
5683 export_rdev(rdev);
5684 return -EINVAL;
5685 }
5686 }
5687 err = bind_rdev_to_array(rdev, mddev);
5688 if (err)
5689 export_rdev(rdev);
5690 return err;
5691 }
5692
5693 /*
5694 * add_new_disk can be used once the array is assembled
5695 * to add "hot spares". They must already have a superblock
5696 * written
5697 */
5698 if (mddev->pers) {
5699 int err;
5700 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5701 printk(KERN_WARNING
1da177e4
LT
5702 "%s: personality does not support diskops!\n",
5703 mdname(mddev));
5704 return -EINVAL;
5705 }
7b1e35f6
N
5706 if (mddev->persistent)
5707 rdev = md_import_device(dev, mddev->major_version,
5708 mddev->minor_version);
5709 else
5710 rdev = md_import_device(dev, -1, -1);
1da177e4 5711 if (IS_ERR(rdev)) {
f72ffdd6 5712 printk(KERN_WARNING
1da177e4
LT
5713 "md: md_import_device returned %ld\n",
5714 PTR_ERR(rdev));
5715 return PTR_ERR(rdev);
5716 }
1a855a06 5717 /* set saved_raid_disk if appropriate */
41158c7e
N
5718 if (!mddev->persistent) {
5719 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5720 info->raid_disk < mddev->raid_disks) {
41158c7e 5721 rdev->raid_disk = info->raid_disk;
bf572541 5722 set_bit(In_sync, &rdev->flags);
8313b8e5 5723 clear_bit(Bitmap_sync, &rdev->flags);
bf572541 5724 } else
41158c7e 5725 rdev->raid_disk = -1;
f466722c 5726 rdev->saved_raid_disk = rdev->raid_disk;
41158c7e
N
5727 } else
5728 super_types[mddev->major_version].
5729 validate_super(mddev, rdev);
bedd86b7 5730 if ((info->state & (1<<MD_DISK_SYNC)) &&
f4563091 5731 rdev->raid_disk != info->raid_disk) {
bedd86b7
N
5732 /* This was a hot-add request, but events doesn't
5733 * match, so reject it.
5734 */
5735 export_rdev(rdev);
5736 return -EINVAL;
5737 }
5738
b2d444d7 5739 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N
5740 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5741 set_bit(WriteMostly, &rdev->flags);
575a80fa
N
5742 else
5743 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5744
1da177e4
LT
5745 rdev->raid_disk = -1;
5746 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N
5747 if (!err && !mddev->pers->hot_remove_disk) {
5748 /* If there is hot_add_disk but no hot_remove_disk
5749 * then added disks for geometry changes,
5750 * and should be added immediately.
5751 */
5752 super_types[mddev->major_version].
5753 validate_super(mddev, rdev);
5754 err = mddev->pers->hot_add_disk(mddev, rdev);
5755 if (err)
5756 unbind_rdev_from_array(rdev);
5757 }
1da177e4
LT
5758 if (err)
5759 export_rdev(rdev);
52664732 5760 else
00bcb4ac 5761 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5762
7ceb17e8 5763 set_bit(MD_CHANGE_DEVS, &mddev->flags);
72a23c21
NB
5764 if (mddev->degraded)
5765 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5766 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5767 if (!err)
5768 md_new_event(mddev);
005eca5e 5769 md_wakeup_thread(mddev->thread);
1da177e4
LT
5770 return err;
5771 }
5772
5773 /* otherwise, add_new_disk is only allowed
5774 * for major_version==0 superblocks
5775 */
5776 if (mddev->major_version != 0) {
5777 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5778 mdname(mddev));
5779 return -EINVAL;
5780 }
5781
5782 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5783 int err;
d710e138 5784 rdev = md_import_device(dev, -1, 0);
1da177e4 5785 if (IS_ERR(rdev)) {
f72ffdd6 5786 printk(KERN_WARNING
1da177e4
LT
5787 "md: error, md_import_device() returned %ld\n",
5788 PTR_ERR(rdev));
5789 return PTR_ERR(rdev);
5790 }
5791 rdev->desc_nr = info->number;
5792 if (info->raid_disk < mddev->raid_disks)
5793 rdev->raid_disk = info->raid_disk;
5794 else
5795 rdev->raid_disk = -1;
5796
1da177e4 5797 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N
5798 if (info->state & (1<<MD_DISK_SYNC))
5799 set_bit(In_sync, &rdev->flags);
1da177e4 5800
8ddf9efe
N
5801 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5802 set_bit(WriteMostly, &rdev->flags);
5803
1da177e4
LT
5804 if (!mddev->persistent) {
5805 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS
5806 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5807 } else
57b2caa3 5808 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5809 rdev->sectors = rdev->sb_start;
1da177e4 5810
2bf071bf
N
5811 err = bind_rdev_to_array(rdev, mddev);
5812 if (err) {
5813 export_rdev(rdev);
5814 return err;
5815 }
1da177e4
LT
5816 }
5817
5818 return 0;
5819}
5820
f72ffdd6 5821static int hot_remove_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT
5822{
5823 char b[BDEVNAME_SIZE];
3cb03002 5824 struct md_rdev *rdev;
1da177e4 5825
1da177e4
LT
5826 rdev = find_rdev(mddev, dev);
5827 if (!rdev)
5828 return -ENXIO;
5829
3ea8929d
N
5830 clear_bit(Blocked, &rdev->flags);
5831 remove_and_add_spares(mddev, rdev);
5832
1da177e4
LT
5833 if (rdev->raid_disk >= 0)
5834 goto busy;
5835
5836 kick_rdev_from_array(rdev);
850b2b42 5837 md_update_sb(mddev, 1);
d7603b7e 5838 md_new_event(mddev);
1da177e4
LT
5839
5840 return 0;
5841busy:
fdefa4d8 5842 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT
5843 bdevname(rdev->bdev,b), mdname(mddev));
5844 return -EBUSY;
5845}
5846
f72ffdd6 5847static int hot_add_disk(struct mddev *mddev, dev_t dev)
1da177e4
LT
5848{
5849 char b[BDEVNAME_SIZE];
5850 int err;
3cb03002 5851 struct md_rdev *rdev;
1da177e4
LT
5852
5853 if (!mddev->pers)
5854 return -ENODEV;
5855
5856 if (mddev->major_version != 0) {
5857 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5858 " version-0 superblocks.\n",
5859 mdname(mddev));
5860 return -EINVAL;
5861 }
5862 if (!mddev->pers->hot_add_disk) {
f72ffdd6 5863 printk(KERN_WARNING
1da177e4
LT
5864 "%s: personality does not support diskops!\n",
5865 mdname(mddev));
5866 return -EINVAL;
5867 }
5868
d710e138 5869 rdev = md_import_device(dev, -1, 0);
1da177e4 5870 if (IS_ERR(rdev)) {
f72ffdd6 5871 printk(KERN_WARNING
1da177e4
LT
5872 "md: error, md_import_device() returned %ld\n",
5873 PTR_ERR(rdev));
5874 return -EINVAL;
5875 }
5876
5877 if (mddev->persistent)
57b2caa3 5878 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5879 else
77304d2a 5880 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5881
8190e754 5882 rdev->sectors = rdev->sb_start;
1da177e4 5883
b2d444d7 5884 if (test_bit(Faulty, &rdev->flags)) {
f72ffdd6 5885 printk(KERN_WARNING
1da177e4
LT
5886 "md: can not hot-add faulty %s disk to %s!\n",
5887 bdevname(rdev->bdev,b), mdname(mddev));
5888 err = -EINVAL;
5889 goto abort_export;
5890 }
b2d444d7 5891 clear_bit(In_sync, &rdev->flags);
1da177e4 5892 rdev->desc_nr = -1;
5842730d 5893 rdev->saved_raid_disk = -1;
2bf071bf
N
5894 err = bind_rdev_to_array(rdev, mddev);
5895 if (err)
5896 goto abort_export;
1da177e4
LT
5897
5898 /*
5899 * The rest should better be atomic, we can have disk failures
5900 * noticed in interrupt contexts ...
5901 */
5902
1da177e4
LT
5903 rdev->raid_disk = -1;
5904
850b2b42 5905 md_update_sb(mddev, 1);
1da177e4
LT
5906
5907 /*
5908 * Kick recovery, maybe this spare has to be added to the
5909 * array immediately.
5910 */
5911 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5912 md_wakeup_thread(mddev->thread);
d7603b7e 5913 md_new_event(mddev);
1da177e4
LT
5914 return 0;
5915
1da177e4
LT
5916abort_export:
5917 export_rdev(rdev);
5918 return err;
5919}
5920
fd01b88c 5921static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c 5922{
035328c2 5923 int err = 0;
32a7627c 5924
36fa3063 5925 if (mddev->pers) {
d66b1b39 5926 if (!mddev->pers->quiesce || !mddev->thread)
36fa3063
N
5927 return -EBUSY;
5928 if (mddev->recovery || mddev->sync_thread)
5929 return -EBUSY;
5930 /* we should be able to change the bitmap.. */
5931 }
32a7627c 5932
36fa3063 5933 if (fd >= 0) {
035328c2 5934 struct inode *inode;
36fa3063
N
5935 if (mddev->bitmap)
5936 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5937 mddev->bitmap_info.file = fget(fd);
32a7627c 5938
c3d9714e 5939 if (mddev->bitmap_info.file == NULL) {
36fa3063
N
5940 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5941 mdname(mddev));
5942 return -EBADF;
5943 }
5944
035328c2
N
5945 inode = mddev->bitmap_info.file->f_mapping->host;
5946 if (!S_ISREG(inode->i_mode)) {
5947 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
5948 mdname(mddev));
5949 err = -EBADF;
5950 } else if (!(mddev->bitmap_info.file->f_mode & FMODE_WRITE)) {
5951 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
5952 mdname(mddev));
5953 err = -EBADF;
5954 } else if (atomic_read(&inode->i_writecount) != 1) {
36fa3063
N
5955 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5956 mdname(mddev));
035328c2
N
5957 err = -EBUSY;
5958 }
5959 if (err) {
c3d9714e
N
5960 fput(mddev->bitmap_info.file);
5961 mddev->bitmap_info.file = NULL;
36fa3063
N
5962 return err;
5963 }
c3d9714e 5964 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N
5965 } else if (mddev->bitmap == NULL)
5966 return -ENOENT; /* cannot remove what isn't there */
5967 err = 0;
5968 if (mddev->pers) {
5969 mddev->pers->quiesce(mddev, 1);
69e51b44 5970 if (fd >= 0) {
36fa3063 5971 err = bitmap_create(mddev);
69e51b44
N
5972 if (!err)
5973 err = bitmap_load(mddev);
5974 }
d7375ab3 5975 if (fd < 0 || err) {
36fa3063 5976 bitmap_destroy(mddev);
d7375ab3
N
5977 fd = -1; /* make sure to put the file */
5978 }
36fa3063 5979 mddev->pers->quiesce(mddev, 0);
d7375ab3
N
5980 }
5981 if (fd < 0) {
035328c2 5982 if (mddev->bitmap_info.file)
c3d9714e 5983 fput(mddev->bitmap_info.file);
c3d9714e 5984 mddev->bitmap_info.file = NULL;
36fa3063
N
5985 }
5986
32a7627c
N
5987 return err;
5988}
5989
1da177e4
LT
5990/*
5991 * set_array_info is used two different ways
5992 * The original usage is when creating a new array.
5993 * In this usage, raid_disks is > 0 and it together with
5994 * level, size, not_persistent,layout,chunksize determine the
5995 * shape of the array.
5996 * This will always create an array with a type-0.90.0 superblock.
5997 * The newer usage is when assembling an array.
5998 * In this case raid_disks will be 0, and the major_version field is
5999 * use to determine which style super-blocks are to be found on the devices.
6000 * The minor and patch _version numbers are also kept incase the
6001 * super_block handler wishes to interpret them.
6002 */
f72ffdd6 6003static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT
6004{
6005
6006 if (info->raid_disks == 0) {
6007 /* just setting version number for superblock loading */
6008 if (info->major_version < 0 ||
50511da3 6009 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT
6010 super_types[info->major_version].name == NULL) {
6011 /* maybe try to auto-load a module? */
f72ffdd6 6012 printk(KERN_INFO
1da177e4
LT
6013 "md: superblock version %d not known\n",
6014 info->major_version);
6015 return -EINVAL;
6016 }
6017 mddev->major_version = info->major_version;
6018 mddev->minor_version = info->minor_version;
6019 mddev->patch_version = info->patch_version;
3f9d7b0d 6020 mddev->persistent = !info->not_persistent;
cbd19983
N
6021 /* ensure mddev_put doesn't delete this now that there
6022 * is some minimal configuration.
6023 */
6024 mddev->ctime = get_seconds();
1da177e4
LT
6025 return 0;
6026 }
6027 mddev->major_version = MD_MAJOR_VERSION;
6028 mddev->minor_version = MD_MINOR_VERSION;
6029 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6030 mddev->ctime = get_seconds();
6031
6032 mddev->level = info->level;
17115e03 6033 mddev->clevel[0] = 0;
58c0fed4 6034 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT
6035 mddev->raid_disks = info->raid_disks;
6036 /* don't set md_minor, it is determined by which /dev/md* was
6037 * openned
6038 */
6039 if (info->state & (1<<MD_SB_CLEAN))
6040 mddev->recovery_cp = MaxSector;
6041 else
6042 mddev->recovery_cp = 0;
6043 mddev->persistent = ! info->not_persistent;
e691063a 6044 mddev->external = 0;
1da177e4
LT
6045
6046 mddev->layout = info->layout;
9d8f0363 6047 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT
6048
6049 mddev->max_disks = MD_SB_DISKS;
6050
e691063a
N
6051 if (mddev->persistent)
6052 mddev->flags = 0;
850b2b42 6053 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6054
c3d9714e 6055 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6409bb05 6056 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
c3d9714e 6057 mddev->bitmap_info.offset = 0;
b2a2703c 6058
f6705578
N
6059 mddev->reshape_position = MaxSector;
6060
1da177e4
LT
6061 /*
6062 * Generate a 128 bit UUID
6063 */
6064 get_random_bytes(mddev->uuid, 16);
6065
f6705578 6066 mddev->new_level = mddev->level;
664e7c41 6067 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N
6068 mddev->new_layout = mddev->layout;
6069 mddev->delta_disks = 0;
2c810cdd 6070 mddev->reshape_backwards = 0;
f6705578 6071
1da177e4
LT
6072 return 0;
6073}
6074
fd01b88c 6075void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 6076{
b522adcd
DW
6077 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6078
6079 if (mddev->external_size)
6080 return;
6081
1f403624
DW
6082 mddev->array_sectors = array_sectors;
6083}
6084EXPORT_SYMBOL(md_set_array_sectors);
6085
fd01b88c 6086static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 6087{
3cb03002 6088 struct md_rdev *rdev;
a35b0d69 6089 int rv;
d71f9f88 6090 int fit = (num_sectors == 0);
a35b0d69
N
6091
6092 if (mddev->pers->resize == NULL)
6093 return -EINVAL;
d71f9f88
AN
6094 /* The "num_sectors" is the number of sectors of each device that
6095 * is used. This can only make sense for arrays with redundancy.
6096 * linear and raid0 always use whatever space is available. We can only
6097 * consider changing this number if no resync or reconstruction is
6098 * happening, and if the new size is acceptable. It must fit before the
0f420358 6099 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN
6100 * of each device. If num_sectors is zero, we find the largest size
6101 * that fits.
a35b0d69
N
6102 */
6103 if (mddev->sync_thread)
6104 return -EBUSY;
bd8839e0
N
6105 if (mddev->ro)
6106 return -EROFS;
a4a6125a 6107
dafb20fa 6108 rdev_for_each(rdev, mddev) {
dd8ac336 6109 sector_t avail = rdev->sectors;
01ab5662 6110
d71f9f88
AN
6111 if (fit && (num_sectors == 0 || num_sectors > avail))
6112 num_sectors = avail;
6113 if (avail < num_sectors)
a35b0d69
N
6114 return -ENOSPC;
6115 }
d71f9f88 6116 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N
6117 if (!rv)
6118 revalidate_disk(mddev->gendisk);
a35b0d69
N
6119 return rv;
6120}
6121
fd01b88c 6122static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N
6123{
6124 int rv;
c6563a8c 6125 struct md_rdev *rdev;
da943b99 6126 /* change the number of raid disks */
63c70c4f 6127 if (mddev->pers->check_reshape == NULL)
da943b99 6128 return -EINVAL;
bd8839e0
N
6129 if (mddev->ro)
6130 return -EROFS;
da943b99 6131 if (raid_disks <= 0 ||
233fca36 6132 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 6133 return -EINVAL;
63c70c4f 6134 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 6135 return -EBUSY;
c6563a8c
N
6136
6137 rdev_for_each(rdev, mddev) {
6138 if (mddev->raid_disks < raid_disks &&
6139 rdev->data_offset < rdev->new_data_offset)
6140 return -EINVAL;
6141 if (mddev->raid_disks > raid_disks &&
6142 rdev->data_offset > rdev->new_data_offset)
6143 return -EINVAL;
6144 }
6145
63c70c4f 6146 mddev->delta_disks = raid_disks - mddev->raid_disks;
2c810cdd
N
6147 if (mddev->delta_disks < 0)
6148 mddev->reshape_backwards = 1;
6149 else if (mddev->delta_disks > 0)
6150 mddev->reshape_backwards = 0;
63c70c4f
N
6151
6152 rv = mddev->pers->check_reshape(mddev);
2c810cdd 6153 if (rv < 0) {
de171cb9 6154 mddev->delta_disks = 0;
2c810cdd
N
6155 mddev->reshape_backwards = 0;
6156 }
da943b99
N
6157 return rv;
6158}
6159
1da177e4
LT
6160/*
6161 * update_array_info is used to change the configuration of an
6162 * on-line array.
6163 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6164 * fields in the info are checked against the array.
6165 * Any differences that cannot be handled will cause an error.
6166 * Normally, only one change can be managed at a time.
6167 */
fd01b88c 6168static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT
6169{
6170 int rv = 0;
6171 int cnt = 0;
36fa3063
N
6172 int state = 0;
6173
6174 /* calculate expected state,ignoring low bits */
c3d9714e 6175 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 6176 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT
6177
6178 if (mddev->major_version != info->major_version ||
6179 mddev->minor_version != info->minor_version ||
6180/* mddev->patch_version != info->patch_version || */
6181 mddev->ctime != info->ctime ||
6182 mddev->level != info->level ||
6183/* mddev->layout != info->layout || */
6184 !mddev->persistent != info->not_persistent||
9d8f0363 6185 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N
6186 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6187 ((state^info->state) & 0xfffffe00)
6188 )
1da177e4
LT
6189 return -EINVAL;
6190 /* Check there is only one change */
58c0fed4
AN
6191 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6192 cnt++;
6193 if (mddev->raid_disks != info->raid_disks)
6194 cnt++;
6195 if (mddev->layout != info->layout)
6196 cnt++;
6197 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6198 cnt++;
6199 if (cnt == 0)
6200 return 0;
6201 if (cnt > 1)
6202 return -EINVAL;
1da177e4
LT
6203
6204 if (mddev->layout != info->layout) {
6205 /* Change layout
6206 * we don't need to do anything at the md level, the
6207 * personality will take care of it all.
6208 */
50ac168a 6209 if (mddev->pers->check_reshape == NULL)
1da177e4 6210 return -EINVAL;
597a711b
N
6211 else {
6212 mddev->new_layout = info->layout;
50ac168a 6213 rv = mddev->pers->check_reshape(mddev);
597a711b
N
6214 if (rv)
6215 mddev->new_layout = mddev->layout;
6216 return rv;
6217 }
1da177e4 6218 }
58c0fed4 6219 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6220 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6221
da943b99
N
6222 if (mddev->raid_disks != info->raid_disks)
6223 rv = update_raid_disks(mddev, info->raid_disks);
6224
36fa3063 6225 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
d66b1b39 6226 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
36fa3063
N
6227 return -EINVAL;
6228 if (mddev->recovery || mddev->sync_thread)
6229 return -EBUSY;
6230 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6231 /* add the bitmap */
6232 if (mddev->bitmap)
6233 return -EEXIST;
c3d9714e 6234 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6235 return -EINVAL;
c3d9714e
N
6236 mddev->bitmap_info.offset =
6237 mddev->bitmap_info.default_offset;
6409bb05
N
6238 mddev->bitmap_info.space =
6239 mddev->bitmap_info.default_space;
36fa3063
N
6240 mddev->pers->quiesce(mddev, 1);
6241 rv = bitmap_create(mddev);
69e51b44
N
6242 if (!rv)
6243 rv = bitmap_load(mddev);
36fa3063
N
6244 if (rv)
6245 bitmap_destroy(mddev);
6246 mddev->pers->quiesce(mddev, 0);
6247 } else {
6248 /* remove the bitmap */
6249 if (!mddev->bitmap)
6250 return -ENOENT;
1ec885cd 6251 if (mddev->bitmap->storage.file)
36fa3063
N
6252 return -EINVAL;
6253 mddev->pers->quiesce(mddev, 1);
6254 bitmap_destroy(mddev);
6255 mddev->pers->quiesce(mddev, 0);
c3d9714e 6256 mddev->bitmap_info.offset = 0;
36fa3063
N
6257 }
6258 }
850b2b42 6259 md_update_sb(mddev, 1);
1da177e4
LT
6260 return rv;
6261}
6262
fd01b88c 6263static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6264{
3cb03002 6265 struct md_rdev *rdev;
1ca69c4b 6266 int err = 0;
1da177e4
LT
6267
6268 if (mddev->pers == NULL)
6269 return -ENODEV;
6270
1ca69c4b
N
6271 rcu_read_lock();
6272 rdev = find_rdev_rcu(mddev, dev);
1da177e4 6273 if (!rdev)
1ca69c4b
N
6274 err = -ENODEV;
6275 else {
6276 md_error(mddev, rdev);
6277 if (!test_bit(Faulty, &rdev->flags))
6278 err = -EBUSY;
6279 }
6280 rcu_read_unlock();
6281 return err;
1da177e4
LT
6282}
6283
2f9618ce
AN
6284/*
6285 * We have a problem here : there is no easy way to give a CHS
6286 * virtual geometry. We currently pretend that we have a 2 heads
6287 * 4 sectors (with a BIG number of cylinders...). This drives
6288 * dosfs just mad... ;-)
6289 */
a885c8c4
CH
6290static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6291{
fd01b88c 6292 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH
6293
6294 geo->heads = 2;
6295 geo->sectors = 4;
49ce6cea 6296 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH
6297 return 0;
6298}
6299
cb335f88
NS
6300static inline bool md_ioctl_valid(unsigned int cmd)
6301{
6302 switch (cmd) {
6303 case ADD_NEW_DISK:
6304 case BLKROSET:
6305 case GET_ARRAY_INFO:
6306 case GET_BITMAP_FILE:
6307 case GET_DISK_INFO:
6308 case HOT_ADD_DISK:
6309 case HOT_REMOVE_DISK:
6310 case PRINT_RAID_DEBUG:
6311 case RAID_AUTORUN:
6312 case RAID_VERSION:
6313 case RESTART_ARRAY_RW:
6314 case RUN_ARRAY:
6315 case SET_ARRAY_INFO:
6316 case SET_BITMAP_FILE:
6317 case SET_DISK_FAULTY:
6318 case STOP_ARRAY:
6319 case STOP_ARRAY_RO:
6320 return true;
6321 default:
6322 return false;
6323 }
6324}
6325
a39907fa 6326static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT
6327 unsigned int cmd, unsigned long arg)
6328{
6329 int err = 0;
6330 void __user *argp = (void __user *)arg;
fd01b88c 6331 struct mddev *mddev = NULL;
e2218350 6332 int ro;
1da177e4 6333
cb335f88
NS
6334 if (!md_ioctl_valid(cmd))
6335 return -ENOTTY;
6336
506c9e44
N
6337 switch (cmd) {
6338 case RAID_VERSION:
6339 case GET_ARRAY_INFO:
6340 case GET_DISK_INFO:
6341 break;
6342 default:
6343 if (!capable(CAP_SYS_ADMIN))
6344 return -EACCES;
6345 }
1da177e4
LT
6346
6347 /*
6348 * Commands dealing with the RAID driver but not any
6349 * particular array:
6350 */
c02c0aeb
N
6351 switch (cmd) {
6352 case RAID_VERSION:
6353 err = get_version(argp);
6354 goto done;
1da177e4 6355
c02c0aeb
N
6356 case PRINT_RAID_DEBUG:
6357 err = 0;
6358 md_print_devices();
6359 goto done;
1da177e4
LT
6360
6361#ifndef MODULE
c02c0aeb
N
6362 case RAID_AUTORUN:
6363 err = 0;
6364 autostart_arrays(arg);
6365 goto done;
1da177e4 6366#endif
c02c0aeb 6367 default:;
1da177e4
LT
6368 }
6369
6370 /*
6371 * Commands creating/starting a new array:
6372 */
6373
a39907fa 6374 mddev = bdev->bd_disk->private_data;
1da177e4
LT
6375
6376 if (!mddev) {
6377 BUG();
6378 goto abort;
6379 }
6380
1ca69c4b
N
6381 /* Some actions do not requires the mutex */
6382 switch (cmd) {
6383 case GET_ARRAY_INFO:
6384 if (!mddev->raid_disks && !mddev->external)
6385 err = -ENODEV;
6386 else
6387 err = get_array_info(mddev, argp);
6388 goto abort;
6389
6390 case GET_DISK_INFO:
6391 if (!mddev->raid_disks && !mddev->external)
6392 err = -ENODEV;
6393 else
6394 err = get_disk_info(mddev, argp);
6395 goto abort;
6396
6397 case SET_DISK_FAULTY:
6398 err = set_disk_faulty(mddev, new_decode_dev(arg));
6399 goto abort;
6400 }
6401
a7a3f08d
N
6402 if (cmd == ADD_NEW_DISK)
6403 /* need to ensure md_delayed_delete() has completed */
6404 flush_workqueue(md_misc_wq);
6405
90f5f7ad
HR
6406 if (cmd == HOT_REMOVE_DISK)
6407 /* need to ensure recovery thread has run */
6408 wait_event_interruptible_timeout(mddev->sb_wait,
6409 !test_bit(MD_RECOVERY_NEEDED,
6410 &mddev->flags),
6411 msecs_to_jiffies(5000));
260fa034
N
6412 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6413 /* Need to flush page cache, and ensure no-one else opens
6414 * and writes
6415 */
6416 mutex_lock(&mddev->open_mutex);
9ba3b7f5 6417 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
260fa034
N
6418 mutex_unlock(&mddev->open_mutex);
6419 err = -EBUSY;
6420 goto abort;
6421 }
6422 set_bit(MD_STILL_CLOSED, &mddev->flags);
6423 mutex_unlock(&mddev->open_mutex);
6424 sync_blockdev(bdev);
6425 }
1da177e4
LT
6426 err = mddev_lock(mddev);
6427 if (err) {
f72ffdd6 6428 printk(KERN_INFO
1da177e4
LT
6429 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6430 err, cmd);
6431 goto abort;
6432 }
6433
c02c0aeb
N
6434 if (cmd == SET_ARRAY_INFO) {
6435 mdu_array_info_t info;
6436 if (!arg)
6437 memset(&info, 0, sizeof(info));
6438 else if (copy_from_user(&info, argp, sizeof(info))) {
6439 err = -EFAULT;
6440 goto abort_unlock;
6441 }
6442 if (mddev->pers) {
6443 err = update_array_info(mddev, &info);
6444 if (err) {
6445 printk(KERN_WARNING "md: couldn't update"
6446 " array info. %d\n", err);
6447 goto abort_unlock;
1da177e4
LT
6448 }
6449 goto done_unlock;
c02c0aeb
N
6450 }
6451 if (!list_empty(&mddev->disks)) {
6452 printk(KERN_WARNING
6453 "md: array %s already has disks!\n",
6454 mdname(mddev));
6455 err = -EBUSY;
6456 goto abort_unlock;
6457 }
6458 if (mddev->raid_disks) {
6459 printk(KERN_WARNING
6460 "md: array %s already initialised!\n",
6461 mdname(mddev));
6462 err = -EBUSY;
6463 goto abort_unlock;
6464 }
6465 err = set_array_info(mddev, &info);
6466 if (err) {
6467 printk(KERN_WARNING "md: couldn't set"
6468 " array info. %d\n", err);
6469 goto abort_unlock;
6470 }
6471 goto done_unlock;
1da177e4
LT
6472 }
6473
6474 /*
6475 * Commands querying/configuring an existing array:
6476 */
32a7627c 6477 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6478 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N
6479 if ((!mddev->raid_disks && !mddev->external)
6480 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6481 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6482 && cmd != GET_BITMAP_FILE) {
1da177e4
LT
6483 err = -ENODEV;
6484 goto abort_unlock;
6485 }
6486
6487 /*
6488 * Commands even a read-only array can execute:
6489 */
c02c0aeb
N
6490 switch (cmd) {
6491 case GET_BITMAP_FILE:
6492 err = get_bitmap_file(mddev, argp);
6493 goto done_unlock;
32a7627c 6494
c02c0aeb
N
6495 case RESTART_ARRAY_RW:
6496 err = restart_array(mddev);
6497 goto done_unlock;
1da177e4 6498
c02c0aeb
N
6499 case STOP_ARRAY:
6500 err = do_md_stop(mddev, 0, bdev);
6501 goto done_unlock;
1da177e4 6502
c02c0aeb
N
6503 case STOP_ARRAY_RO:
6504 err = md_set_readonly(mddev, bdev);
6505 goto done_unlock;
1da177e4 6506
3ea8929d
N
6507 case HOT_REMOVE_DISK:
6508 err = hot_remove_disk(mddev, new_decode_dev(arg));
6509 goto done_unlock;
6510
7ceb17e8
N
6511 case ADD_NEW_DISK:
6512 /* We can support ADD_NEW_DISK on read-only arrays
6513 * on if we are re-adding a preexisting device.
6514 * So require mddev->pers and MD_DISK_SYNC.
6515 */
6516 if (mddev->pers) {
6517 mdu_disk_info_t info;
6518 if (copy_from_user(&info, argp, sizeof(info)))
6519 err = -EFAULT;
6520 else if (!(info.state & (1<<MD_DISK_SYNC)))
6521 /* Need to clear read-only for this */
6522 break;
6523 else
6524 err = add_new_disk(mddev, &info);
6525 goto done_unlock;
6526 }
6527 break;
6528
c02c0aeb
N
6529 case BLKROSET:
6530 if (get_user(ro, (int __user *)(arg))) {
6531 err = -EFAULT;
6532 goto done_unlock;
6533 }
6534 err = -EINVAL;
e2218350 6535
c02c0aeb
N
6536 /* if the bdev is going readonly the value of mddev->ro
6537 * does not matter, no writes are coming
6538 */
6539 if (ro)
6540 goto done_unlock;
e2218350 6541
c02c0aeb
N
6542 /* are we are already prepared for writes? */
6543 if (mddev->ro != 1)
6544 goto done_unlock;
e2218350 6545
c02c0aeb
N
6546 /* transitioning to readauto need only happen for
6547 * arrays that call md_write_start
6548 */
6549 if (mddev->pers) {
6550 err = restart_array(mddev);
6551 if (err == 0) {
6552 mddev->ro = 2;
6553 set_disk_ro(mddev->gendisk, 0);
e2218350 6554 }
c02c0aeb
N
6555 }
6556 goto done_unlock;
1da177e4
LT
6557 }
6558
6559 /*
6560 * The remaining ioctls are changing the state of the
f91de92e
N
6561 * superblock, so we do not allow them on read-only arrays.
6562 * However non-MD ioctls (e.g. get-size) will still come through
6563 * here and hit the 'default' below, so only disallow
6564 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6565 */
bb57fc64 6566 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N
6567 if (mddev->ro == 2) {
6568 mddev->ro = 0;
00bcb4ac 6569 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 6570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
f3378b48
N
6571 /* mddev_unlock will wake thread */
6572 /* If a device failed while we were read-only, we
6573 * need to make sure the metadata is updated now.
6574 */
6575 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6576 mddev_unlock(mddev);
6577 wait_event(mddev->sb_wait,
6578 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6579 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
29f097c4 6580 mddev_lock_nointr(mddev);
f3378b48 6581 }
f91de92e
N
6582 } else {
6583 err = -EROFS;
6584 goto abort_unlock;
6585 }
1da177e4
LT
6586 }
6587
c02c0aeb
N
6588 switch (cmd) {
6589 case ADD_NEW_DISK:
1da177e4 6590 {
c02c0aeb
N
6591 mdu_disk_info_t info;
6592 if (copy_from_user(&info, argp, sizeof(info)))
6593 err = -EFAULT;
6594 else
6595 err = add_new_disk(mddev, &info);
6596 goto done_unlock;
6597 }
1da177e4 6598
c02c0aeb
N
6599 case HOT_ADD_DISK:
6600 err = hot_add_disk(mddev, new_decode_dev(arg));
6601 goto done_unlock;
1da177e4 6602
c02c0aeb
N
6603 case RUN_ARRAY:
6604 err = do_md_run(mddev);
6605 goto done_unlock;
1da177e4 6606
c02c0aeb
N
6607 case SET_BITMAP_FILE:
6608 err = set_bitmap_file(mddev, (int)arg);
6609 goto done_unlock;
32a7627c 6610
c02c0aeb
N
6611 default:
6612 err = -EINVAL;
6613 goto abort_unlock;
1da177e4
LT
6614 }
6615
6616done_unlock:
6617abort_unlock:
d3374825
N
6618 if (mddev->hold_active == UNTIL_IOCTL &&
6619 err != -EINVAL)
6620 mddev->hold_active = 0;
1da177e4
LT
6621 mddev_unlock(mddev);
6622
6623 return err;
6624done:
6625 if (err)
6626 MD_BUG();
6627abort:
6628 return err;
6629}
aa98aa31
AB
6630#ifdef CONFIG_COMPAT
6631static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6632 unsigned int cmd, unsigned long arg)
6633{
6634 switch (cmd) {
6635 case HOT_REMOVE_DISK:
6636 case HOT_ADD_DISK:
6637 case SET_DISK_FAULTY:
6638 case SET_BITMAP_FILE:
6639 /* These take in integer arg, do not convert */
6640 break;
6641 default:
6642 arg = (unsigned long)compat_ptr(arg);
6643 break;
6644 }
6645
6646 return md_ioctl(bdev, mode, cmd, arg);
6647}
6648#endif /* CONFIG_COMPAT */
1da177e4 6649
a39907fa 6650static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT
6651{
6652 /*
6653 * Succeed if we can lock the mddev, which confirms that
6654 * it isn't being stopped right now.
6655 */
fd01b88c 6656 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT
6657 int err;
6658
0c098220
YL
6659 if (!mddev)
6660 return -ENODEV;
6661
d3374825
N
6662 if (mddev->gendisk != bdev->bd_disk) {
6663 /* we are racing with mddev_put which is discarding this
6664 * bd_disk.
6665 */
6666 mddev_put(mddev);
6667 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6668 flush_workqueue(md_misc_wq);
d3374825
N
6669 /* Then retry the open from the top */
6670 return -ERESTARTSYS;
6671 }
6672 BUG_ON(mddev != bdev->bd_disk->private_data);
6673
c8c00a69 6674 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT
6675 goto out;
6676
6677 err = 0;
f2ea68cf 6678 atomic_inc(&mddev->openers);
260fa034 6679 clear_bit(MD_STILL_CLOSED, &mddev->flags);
c8c00a69 6680 mutex_unlock(&mddev->open_mutex);
1da177e4 6681
f0b4f7e2 6682 check_disk_change(bdev);
1da177e4
LT
6683 out:
6684 return err;
6685}
6686
db2a144b 6687static void md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6688{
f72ffdd6 6689 struct mddev *mddev = disk->private_data;
1da177e4 6690
52e5f9d1 6691 BUG_ON(!mddev);
f2ea68cf 6692 atomic_dec(&mddev->openers);
1da177e4 6693 mddev_put(mddev);
1da177e4 6694}
f0b4f7e2
N
6695
6696static int md_media_changed(struct gendisk *disk)
6697{
fd01b88c 6698 struct mddev *mddev = disk->private_data;
f0b4f7e2
N
6699
6700 return mddev->changed;
6701}
6702
6703static int md_revalidate(struct gendisk *disk)
6704{
fd01b88c 6705 struct mddev *mddev = disk->private_data;
f0b4f7e2
N
6706
6707 mddev->changed = 0;
6708 return 0;
6709}
83d5cde4 6710static const struct block_device_operations md_fops =
1da177e4
LT
6711{
6712 .owner = THIS_MODULE,
a39907fa
AV
6713 .open = md_open,
6714 .release = md_release,
b492b852 6715 .ioctl = md_ioctl,
aa98aa31
AB
6716#ifdef CONFIG_COMPAT
6717 .compat_ioctl = md_compat_ioctl,
6718#endif
a885c8c4 6719 .getgeo = md_getgeo,
f0b4f7e2
N
6720 .media_changed = md_media_changed,
6721 .revalidate_disk= md_revalidate,
1da177e4
LT
6722};
6723
f72ffdd6 6724static int md_thread(void *arg)
1da177e4 6725{
2b8bf345 6726 struct md_thread *thread = arg;
1da177e4 6727
1da177e4
LT
6728 /*
6729 * md_thread is a 'system-thread', it's priority should be very
6730 * high. We avoid resource deadlocks individually in each
6731 * raid personality. (RAID5 does preallocation) We also use RR and
6732 * the very same RT priority as kswapd, thus we will never get
6733 * into a priority inversion deadlock.
6734 *
6735 * we definitely have to have equal or higher priority than
6736 * bdflush, otherwise bdflush will deadlock if there are too
6737 * many dirty RAID5 blocks.
6738 */
1da177e4 6739
6985c43f 6740 allow_signal(SIGKILL);
a6fb0934 6741 while (!kthread_should_stop()) {
1da177e4 6742
93588e22
N
6743 /* We need to wait INTERRUPTIBLE so that
6744 * we don't add to the load-average.
6745 * That means we need to be sure no signals are
6746 * pending
6747 */
6748 if (signal_pending(current))
6749 flush_signals(current);
6750
6751 wait_event_interruptible_timeout
6752 (thread->wqueue,
6753 test_bit(THREAD_WAKEUP, &thread->flags)
6754 || kthread_should_stop(),
6755 thread->timeout);
1da177e4 6756
6c987910
N
6757 clear_bit(THREAD_WAKEUP, &thread->flags);
6758 if (!kthread_should_stop())
4ed8731d 6759 thread->run(thread);
1da177e4 6760 }
a6fb0934 6761
1da177e4
LT
6762 return 0;
6763}
6764
2b8bf345 6765void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT
6766{
6767 if (thread) {
36a4e1fe 6768 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT
6769 set_bit(THREAD_WAKEUP, &thread->flags);
6770 wake_up(&thread->wqueue);
6771 }
6772}
6773
4ed8731d
SL
6774struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6775 struct mddev *mddev, const char *name)
1da177e4 6776{
2b8bf345 6777 struct md_thread *thread;
1da177e4 6778
2b8bf345 6779 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT
6780 if (!thread)
6781 return NULL;
6782
1da177e4
LT
6783 init_waitqueue_head(&thread->wqueue);
6784
1da177e4
LT
6785 thread->run = run;
6786 thread->mddev = mddev;
32a7627c 6787 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N
6788 thread->tsk = kthread_run(md_thread, thread,
6789 "%s_%s",
6790 mdname(thread->mddev),
0232605d 6791 name);
a6fb0934 6792 if (IS_ERR(thread->tsk)) {
1da177e4
LT
6793 kfree(thread);
6794 return NULL;
6795 }
1da177e4
LT
6796 return thread;
6797}
6798
2b8bf345 6799void md_unregister_thread(struct md_thread **threadp)
1da177e4 6800{
2b8bf345 6801 struct md_thread *thread = *threadp;
e0cf8f04
N
6802 if (!thread)
6803 return;
36a4e1fe 6804 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N
6805 /* Locking ensures that mddev_unlock does not wake_up a
6806 * non-existent thread
6807 */
6808 spin_lock(&pers_lock);
6809 *threadp = NULL;
6810 spin_unlock(&pers_lock);
a6fb0934
N
6811
6812 kthread_stop(thread->tsk);
1da177e4
LT
6813 kfree(thread);
6814}
6815
fd01b88c 6816void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT
6817{
6818 if (!mddev) {
6819 MD_BUG();
6820 return;
6821 }
6822
b2d444d7 6823 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6824 return;
6bfe0b49 6825
de393cde 6826 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT
6827 return;
6828 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB
6829 if (mddev->degraded)
6830 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6831 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT
6832 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6833 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6834 md_wakeup_thread(mddev->thread);
768a418d 6835 if (mddev->event_work.func)
e804ac78 6836 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6837 md_new_event_inintr(mddev);
1da177e4
LT
6838}
6839
6840/* seq_file implementation /proc/mdstat */
6841
6842static void status_unused(struct seq_file *seq)
6843{
6844 int i = 0;
3cb03002 6845 struct md_rdev *rdev;
1da177e4
LT
6846
6847 seq_printf(seq, "unused devices: ");
6848
159ec1fc 6849 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT
6850 char b[BDEVNAME_SIZE];
6851 i++;
6852 seq_printf(seq, "%s ",
6853 bdevname(rdev->bdev,b));
6854 }
6855 if (!i)
6856 seq_printf(seq, "<none>");
6857
6858 seq_printf(seq, "\n");
6859}
6860
f72ffdd6 6861static void status_resync(struct seq_file *seq, struct mddev *mddev)
1da177e4 6862{
dd71cf6b
N
6863 sector_t max_sectors, resync, res;
6864 unsigned long dt, db;
6865 sector_t rt;
4588b42e
N
6866 int scale;
6867 unsigned int per_milli;
1da177e4 6868
72f36d59
N
6869 if (mddev->curr_resync <= 3)
6870 resync = 0;
6871 else
6872 resync = mddev->curr_resync
6873 - atomic_read(&mddev->recovery_active);
1da177e4 6874
c804cdec
N
6875 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6876 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
dd71cf6b 6877 max_sectors = mddev->resync_max_sectors;
1da177e4 6878 else
dd71cf6b 6879 max_sectors = mddev->dev_sectors;
1da177e4
LT
6880
6881 /*
6882 * Should not happen.
6883 */
dd71cf6b 6884 if (!max_sectors) {
1da177e4
LT
6885 MD_BUG();
6886 return;
6887 }
4588b42e 6888 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6889 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N
6890 * u32, as those are the requirements for sector_div.
6891 * Thus 'scale' must be at least 10
6892 */
6893 scale = 10;
6894 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6895 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N
6896 scale++;
6897 }
6898 res = (resync>>scale)*1000;
dd71cf6b 6899 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N
6900
6901 per_milli = res;
1da177e4 6902 {
4588b42e 6903 int i, x = per_milli/50, y = 20-x;
1da177e4
LT
6904 seq_printf(seq, "[");
6905 for (i = 0; i < x; i++)
6906 seq_printf(seq, "=");
6907 seq_printf(seq, ">");
6908 for (i = 0; i < y; i++)
6909 seq_printf(seq, ".");
6910 seq_printf(seq, "] ");
6911 }
4588b42e 6912 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N
6913 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6914 "reshape" :
61df9d91
N
6915 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6916 "check" :
6917 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6918 "resync" : "recovery"))),
6919 per_milli/10, per_milli % 10,
dd71cf6b
N
6920 (unsigned long long) resync/2,
6921 (unsigned long long) max_sectors/2);
1da177e4
LT
6922
6923 /*
1da177e4
LT
6924 * dt: time from mark until now
6925 * db: blocks written from mark until now
6926 * rt: remaining time
dd71cf6b
N
6927 *
6928 * rt is a sector_t, so could be 32bit or 64bit.
6929 * So we divide before multiply in case it is 32bit and close
6930 * to the limit.
25985edc 6931 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N
6932 * near the end of resync when the number of remaining sectors
6933 * is close to 'db'.
6934 * We then divide rt by 32 after multiplying by db to compensate.
6935 * The '+1' avoids division by zero if db is very small.
1da177e4
LT
6936 */
6937 dt = ((jiffies - mddev->resync_mark) / HZ);
6938 if (!dt) dt++;
ff4e8d9a
N
6939 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6940 - mddev->resync_mark_cnt;
1da177e4 6941
dd71cf6b
N
6942 rt = max_sectors - resync; /* number of remaining sectors */
6943 sector_div(rt, db/32+1);
6944 rt *= dt;
6945 rt >>= 5;
6946
6947 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6948 ((unsigned long)rt % 60)/6);
1da177e4 6949
ff4e8d9a 6950 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT
6951}
6952
6953static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6954{
6955 struct list_head *tmp;
6956 loff_t l = *pos;
fd01b88c 6957 struct mddev *mddev;
1da177e4
LT
6958
6959 if (l >= 0x10000)
6960 return NULL;
6961 if (!l--)
6962 /* header */
6963 return (void*)1;
6964
6965 spin_lock(&all_mddevs_lock);
6966 list_for_each(tmp,&all_mddevs)
6967 if (!l--) {
fd01b88c 6968 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT
6969 mddev_get(mddev);
6970 spin_unlock(&all_mddevs_lock);
6971 return mddev;
6972 }
6973 spin_unlock(&all_mddevs_lock);
6974 if (!l--)
6975 return (void*)2;/* tail */
6976 return NULL;
6977}
6978
6979static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6980{
6981 struct list_head *tmp;
fd01b88c 6982 struct mddev *next_mddev, *mddev = v;
f72ffdd6 6983
1da177e4
LT
6984 ++*pos;
6985 if (v == (void*)2)
6986 return NULL;
6987
6988 spin_lock(&all_mddevs_lock);
6989 if (v == (void*)1)
6990 tmp = all_mddevs.next;
6991 else
6992 tmp = mddev->all_mddevs.next;
6993 if (tmp != &all_mddevs)
fd01b88c 6994 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT
6995 else {
6996 next_mddev = (void*)2;
6997 *pos = 0x10000;
f72ffdd6 6998 }
1da177e4
LT
6999 spin_unlock(&all_mddevs_lock);
7000
7001 if (v != (void*)1)
7002 mddev_put(mddev);
7003 return next_mddev;
7004
7005}
7006
7007static void md_seq_stop(struct seq_file *seq, void *v)
7008{
fd01b88c 7009 struct mddev *mddev = v;
1da177e4
LT
7010
7011 if (mddev && v != (void*)1 && v != (void*)2)
7012 mddev_put(mddev);
7013}
7014
7015static int md_seq_show(struct seq_file *seq, void *v)
7016{
fd01b88c 7017 struct mddev *mddev = v;
dd8ac336 7018 sector_t sectors;
3cb03002 7019 struct md_rdev *rdev;
1da177e4
LT
7020
7021 if (v == (void*)1) {
84fc4b56 7022 struct md_personality *pers;
1da177e4
LT
7023 seq_printf(seq, "Personalities : ");
7024 spin_lock(&pers_lock);
2604b703
N
7025 list_for_each_entry(pers, &pers_list, list)
7026 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT
7027
7028 spin_unlock(&pers_lock);
7029 seq_printf(seq, "\n");
f1514638 7030 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT
7031 return 0;
7032 }
7033 if (v == (void*)2) {
7034 status_unused(seq);
7035 return 0;
7036 }
7037
5dc5cf7d 7038 if (mddev_lock(mddev) < 0)
1da177e4 7039 return -EINTR;
5dc5cf7d 7040
1da177e4
LT
7041 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7042 seq_printf(seq, "%s : %sactive", mdname(mddev),
7043 mddev->pers ? "" : "in");
7044 if (mddev->pers) {
f91de92e 7045 if (mddev->ro==1)
1da177e4 7046 seq_printf(seq, " (read-only)");
f91de92e 7047 if (mddev->ro==2)
52720ae7 7048 seq_printf(seq, " (auto-read-only)");
1da177e4
LT
7049 seq_printf(seq, " %s", mddev->pers->name);
7050 }
7051
dd8ac336 7052 sectors = 0;
dafb20fa 7053 rdev_for_each(rdev, mddev) {
1da177e4
LT
7054 char b[BDEVNAME_SIZE];
7055 seq_printf(seq, " %s[%d]",
7056 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N
7057 if (test_bit(WriteMostly, &rdev->flags))
7058 seq_printf(seq, "(W)");
b2d444d7 7059 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT
7060 seq_printf(seq, "(F)");
7061 continue;
2d78f8c4
N
7062 }
7063 if (rdev->raid_disk < 0)
b325a32e 7064 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N
7065 if (test_bit(Replacement, &rdev->flags))
7066 seq_printf(seq, "(R)");
dd8ac336 7067 sectors += rdev->sectors;
1da177e4
LT
7068 }
7069
7070 if (!list_empty(&mddev->disks)) {
7071 if (mddev->pers)
7072 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN
7073 (unsigned long long)
7074 mddev->array_sectors / 2);
1da177e4
LT
7075 else
7076 seq_printf(seq, "\n %llu blocks",
dd8ac336 7077 (unsigned long long)sectors / 2);
1da177e4 7078 }
1cd6bf19
N
7079 if (mddev->persistent) {
7080 if (mddev->major_version != 0 ||
7081 mddev->minor_version != 90) {
7082 seq_printf(seq," super %d.%d",
7083 mddev->major_version,
7084 mddev->minor_version);
7085 }
e691063a
N
7086 } else if (mddev->external)
7087 seq_printf(seq, " super external:%s",
7088 mddev->metadata_type);
7089 else
1cd6bf19 7090 seq_printf(seq, " super non-persistent");
1da177e4
LT
7091
7092 if (mddev->pers) {
d710e138 7093 mddev->pers->status(seq, mddev);
f72ffdd6 7094 seq_printf(seq, "\n ");
8e1b39d6
N
7095 if (mddev->pers->sync_request) {
7096 if (mddev->curr_resync > 2) {
d710e138 7097 status_resync(seq, mddev);
8e1b39d6 7098 seq_printf(seq, "\n ");
72f36d59 7099 } else if (mddev->curr_resync >= 1)
8e1b39d6
N
7100 seq_printf(seq, "\tresync=DELAYED\n ");
7101 else if (mddev->recovery_cp < MaxSector)
7102 seq_printf(seq, "\tresync=PENDING\n ");
7103 }
32a7627c
N
7104 } else
7105 seq_printf(seq, "\n ");
7106
57148964 7107 bitmap_status(seq, mddev->bitmap);
1da177e4
LT
7108
7109 seq_printf(seq, "\n");
7110 }
7111 mddev_unlock(mddev);
f72ffdd6 7112
1da177e4
LT
7113 return 0;
7114}
7115
110518bc 7116static const struct seq_operations md_seq_ops = {
1da177e4
LT
7117 .start = md_seq_start,
7118 .next = md_seq_next,
7119 .stop = md_seq_stop,
7120 .show = md_seq_show,
7121};
7122
7123static int md_seq_open(struct inode *inode, struct file *file)
7124{
f1514638 7125 struct seq_file *seq;
1da177e4
LT
7126 int error;
7127
7128 error = seq_open(file, &md_seq_ops);
d7603b7e 7129 if (error)
f1514638
KS
7130 return error;
7131
7132 seq = file->private_data;
7133 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT
7134 return error;
7135}
7136
e2f23b60 7137static int md_unloading;
d7603b7e
N
7138static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7139{
f1514638 7140 struct seq_file *seq = filp->private_data;
d7603b7e
N
7141 int mask;
7142
e2f23b60
N
7143 if (md_unloading)
7144 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;;
d7603b7e
N
7145 poll_wait(filp, &md_event_waiters, wait);
7146
7147 /* always allow read */
7148 mask = POLLIN | POLLRDNORM;
7149
f1514638 7150 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N
7151 mask |= POLLERR | POLLPRI;
7152 return mask;
7153}
7154
fa027c2a 7155static const struct file_operations md_seq_fops = {
e24650c2 7156 .owner = THIS_MODULE,
1da177e4
LT
7157 .open = md_seq_open,
7158 .read = seq_read,
7159 .llseek = seq_lseek,
c3f94b40 7160 .release = seq_release_private,
d7603b7e 7161 .poll = mdstat_poll,
1da177e4
LT
7162};
7163
84fc4b56 7164int register_md_personality(struct md_personality *p)
1da177e4 7165{
50bd3774
CY
7166 printk(KERN_INFO "md: %s personality registered for level %d\n",
7167 p->name, p->level);
1da177e4 7168 spin_lock(&pers_lock);
2604b703 7169 list_add_tail(&p->list, &pers_list);
1da177e4
LT
7170 spin_unlock(&pers_lock);
7171 return 0;
7172}
7173
84fc4b56 7174int unregister_md_personality(struct md_personality *p)
1da177e4 7175{
2604b703 7176 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 7177 spin_lock(&pers_lock);
2604b703 7178 list_del_init(&p->list);
1da177e4
LT
7179 spin_unlock(&pers_lock);
7180 return 0;
7181}
7182
fd01b88c 7183static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 7184{
f72ffdd6 7185 struct md_rdev *rdev;
1da177e4 7186 int idle;
eea1bf38 7187 int curr_events;
1da177e4
LT
7188
7189 idle = 1;
4b80991c
N
7190 rcu_read_lock();
7191 rdev_for_each_rcu(rdev, mddev) {
1da177e4 7192 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N
7193 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7194 (int)part_stat_read(&disk->part0, sectors[1]) -
7195 atomic_read(&disk->sync_io);
713f6ab1
N
7196 /* sync IO will cause sync_io to increase before the disk_stats
7197 * as sync_io is counted when a request starts, and
7198 * disk_stats is counted when it completes.
7199 * So resync activity will cause curr_events to be smaller than
7200 * when there was no such activity.
7201 * non-sync IO will cause disk_stat to increase without
7202 * increasing sync_io so curr_events will (eventually)
7203 * be larger than it was before. Once it becomes
7204 * substantially larger, the test below will cause
7205 * the array to appear non-idle, and resync will slow
7206 * down.
7207 * If there is a lot of outstanding resync activity when
7208 * we set last_event to curr_events, then all that activity
7209 * completing might cause the array to appear non-idle
7210 * and resync will be slowed down even though there might
7211 * not have been non-resync activity. This will only
7212 * happen once though. 'last_events' will soon reflect
7213 * the state where there is little or no outstanding
7214 * resync requests, and further resync activity will
7215 * always make curr_events less than last_events.
c0e48521 7216 *
1da177e4 7217 */
eea1bf38 7218 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT
7219 rdev->last_events = curr_events;
7220 idle = 0;
7221 }
7222 }
4b80991c 7223 rcu_read_unlock();
1da177e4
LT
7224 return idle;
7225}
7226
fd01b88c 7227void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT
7228{
7229 /* another "blocks" (512byte) blocks have been synced */
7230 atomic_sub(blocks, &mddev->recovery_active);
7231 wake_up(&mddev->recovery_wait);
7232 if (!ok) {
dfc70645 7233 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
0a19caab 7234 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
1da177e4
LT
7235 md_wakeup_thread(mddev->thread);
7236 // stop recovery, signal do_sync ....
7237 }
7238}
7239
06d91a5f
N
7240/* md_write_start(mddev, bi)
7241 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N
7242 * in superblock) before writing, schedule a superblock update
7243 * and wait for it to complete.
06d91a5f 7244 */
fd01b88c 7245void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 7246{
0fd62b86 7247 int did_change = 0;
06d91a5f 7248 if (bio_data_dir(bi) != WRITE)
3d310eb7 7249 return;
06d91a5f 7250
f91de92e
N
7251 BUG_ON(mddev->ro == 1);
7252 if (mddev->ro == 2) {
7253 /* need to switch to read/write */
7254 mddev->ro = 0;
7255 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7256 md_wakeup_thread(mddev->thread);
25156198 7257 md_wakeup_thread(mddev->sync_thread);
0fd62b86 7258 did_change = 1;
f91de92e 7259 }
06d91a5f 7260 atomic_inc(&mddev->writes_pending);
31a59e34
N
7261 if (mddev->safemode == 1)
7262 mddev->safemode = 0;
06d91a5f 7263 if (mddev->in_sync) {
a9701a30 7264 spin_lock_irq(&mddev->write_lock);
3d310eb7
N
7265 if (mddev->in_sync) {
7266 mddev->in_sync = 0;
850b2b42 7267 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7268 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 7269 md_wakeup_thread(mddev->thread);
0fd62b86 7270 did_change = 1;
3d310eb7 7271 }
a9701a30 7272 spin_unlock_irq(&mddev->write_lock);
06d91a5f 7273 }
0fd62b86 7274 if (did_change)
00bcb4ac 7275 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7276 wait_event(mddev->sb_wait,
09a44cc1 7277 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT
7278}
7279
fd01b88c 7280void md_write_end(struct mddev *mddev)
1da177e4
LT
7281{
7282 if (atomic_dec_and_test(&mddev->writes_pending)) {
7283 if (mddev->safemode == 2)
7284 md_wakeup_thread(mddev->thread);
16f17b39 7285 else if (mddev->safemode_delay)
1da177e4
LT
7286 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7287 }
7288}
7289
2a2275d6
N
7290/* md_allow_write(mddev)
7291 * Calling this ensures that the array is marked 'active' so that writes
7292 * may proceed without blocking. It is important to call this before
7293 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7294 * Must be called with mddev_lock held.
b5470dc5
DW
7295 *
7296 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7297 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7298 */
fd01b88c 7299int md_allow_write(struct mddev *mddev)
2a2275d6
N
7300{
7301 if (!mddev->pers)
b5470dc5 7302 return 0;
2a2275d6 7303 if (mddev->ro)
b5470dc5 7304 return 0;
1a0fd497 7305 if (!mddev->pers->sync_request)
b5470dc5 7306 return 0;
2a2275d6
N
7307
7308 spin_lock_irq(&mddev->write_lock);
7309 if (mddev->in_sync) {
7310 mddev->in_sync = 0;
7311 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7312 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N
7313 if (mddev->safemode_delay &&
7314 mddev->safemode == 0)
7315 mddev->safemode = 1;
7316 spin_unlock_irq(&mddev->write_lock);
7317 md_update_sb(mddev, 0);
00bcb4ac 7318 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N
7319 } else
7320 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7321
070dc6dd 7322 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW
7323 return -EAGAIN;
7324 else
7325 return 0;
2a2275d6
N
7326}
7327EXPORT_SYMBOL_GPL(md_allow_write);
7328
1da177e4
LT
7329#define SYNC_MARKS 10
7330#define SYNC_MARK_STEP (3*HZ)
54f89341 7331#define UPDATE_FREQUENCY (5*60*HZ)
4ed8731d 7332void md_do_sync(struct md_thread *thread)
1da177e4 7333{
4ed8731d 7334 struct mddev *mddev = thread->mddev;
fd01b88c 7335 struct mddev *mddev2;
1da177e4
LT
7336 unsigned int currspeed = 0,
7337 window;
ac7e50a3 7338 sector_t max_sectors,j, io_sectors, recovery_done;
1da177e4 7339 unsigned long mark[SYNC_MARKS];
54f89341 7340 unsigned long update_time;
1da177e4
LT
7341 sector_t mark_cnt[SYNC_MARKS];
7342 int last_mark,m;
7343 struct list_head *tmp;
7344 sector_t last_check;
57afd89f 7345 int skipped = 0;
3cb03002 7346 struct md_rdev *rdev;
c4a39551 7347 char *desc, *action = NULL;
7c2c57c9 7348 struct blk_plug plug;
1da177e4
LT
7349
7350 /* just incase thread restarts... */
7351 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7352 return;
3991b31e
N
7353 if (mddev->ro) {/* never try to sync a read-only array */
7354 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5fd6c1dc 7355 return;
3991b31e 7356 }
1da177e4 7357
61df9d91 7358 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
c4a39551 7359 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
61df9d91 7360 desc = "data-check";
c4a39551
JB
7361 action = "check";
7362 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
61df9d91 7363 desc = "requested-resync";
c4a39551
JB
7364 action = "repair";
7365 } else
61df9d91
N
7366 desc = "resync";
7367 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7368 desc = "reshape";
7369 else
7370 desc = "recovery";
7371
c4a39551
JB
7372 mddev->last_sync_action = action ?: desc;
7373
1da177e4
LT
7374 /* we overload curr_resync somewhat here.
7375 * 0 == not engaged in resync at all
7376 * 2 == checking that there is no conflict with another sync
7377 * 1 == like 2, but have yielded to allow conflicting resync to
7378 * commense
7379 * other == active in resync - this many blocks
7380 *
7381 * Before starting a resync we must have set curr_resync to
7382 * 2, and then checked that every "conflicting" array has curr_resync
7383 * less than ours. When we find one that is the same or higher
7384 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7385 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7386 * This will mean we have to start checking from the beginning again.
7387 *
7388 */
7389
7390 do {
7391 mddev->curr_resync = 2;
7392
7393 try_again:
404e4b43 7394 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7395 goto skip;
29ac4aa3 7396 for_each_mddev(mddev2, tmp) {
1da177e4
LT
7397 if (mddev2 == mddev)
7398 continue;
90b08710
BS
7399 if (!mddev->parallel_resync
7400 && mddev2->curr_resync
7401 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT
7402 DEFINE_WAIT(wq);
7403 if (mddev < mddev2 && mddev->curr_resync == 2) {
7404 /* arbitrarily yield */
7405 mddev->curr_resync = 1;
7406 wake_up(&resync_wait);
7407 }
7408 if (mddev > mddev2 && mddev->curr_resync == 1)
7409 /* no need to wait here, we can wait the next
7410 * time 'round when curr_resync == 2
7411 */
7412 continue;
9744197c
N
7413 /* We need to wait 'interruptible' so as not to
7414 * contribute to the load average, and not to
7415 * be caught by 'softlockup'
7416 */
7417 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
c91abf5a 7418 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8712e553 7419 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N
7420 printk(KERN_INFO "md: delaying %s of %s"
7421 " until %s has finished (they"
1da177e4 7422 " share one or more physical units)\n",
61df9d91 7423 desc, mdname(mddev), mdname(mddev2));
1da177e4 7424 mddev_put(mddev2);
9744197c
N
7425 if (signal_pending(current))
7426 flush_signals(current);
1da177e4
LT
7427 schedule();
7428 finish_wait(&resync_wait, &wq);
7429 goto try_again;
7430 }
7431 finish_wait(&resync_wait, &wq);
7432 }
7433 }
7434 } while (mddev->curr_resync < 2);
7435
5fd6c1dc 7436 j = 0;
9d88883e 7437 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7438 /* resync follows the size requested by the personality,
57afd89f 7439 * which defaults to physical size, but can be virtual size
1da177e4
LT
7440 */
7441 max_sectors = mddev->resync_max_sectors;
7f7583d4 7442 atomic64_set(&mddev->resync_mismatches, 0);
5fd6c1dc 7443 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB
7444 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7445 j = mddev->resync_min;
7446 else if (!mddev->bitmap)
5fd6c1dc 7447 j = mddev->recovery_cp;
5e96ee65 7448
ccfcc3c1 7449 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
c804cdec 7450 max_sectors = mddev->resync_max_sectors;
5fd6c1dc 7451 else {
1da177e4 7452 /* recovery follows the physical size of devices */
58c0fed4 7453 max_sectors = mddev->dev_sectors;
5fd6c1dc 7454 j = MaxSector;
4e59ca7d 7455 rcu_read_lock();
dafb20fa 7456 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N
7457 if (rdev->raid_disk >= 0 &&
7458 !test_bit(Faulty, &rdev->flags) &&
7459 !test_bit(In_sync, &rdev->flags) &&
7460 rdev->recovery_offset < j)
7461 j = rdev->recovery_offset;
4e59ca7d 7462 rcu_read_unlock();
133d4527
N
7463
7464 /* If there is a bitmap, we need to make sure all
7465 * writes that started before we added a spare
7466 * complete before we start doing a recovery.
7467 * Otherwise the write might complete and (via
7468 * bitmap_endwrite) set a bit in the bitmap after the
7469 * recovery has checked that bit and skipped that
7470 * region.
7471 */
7472 if (mddev->bitmap) {
7473 mddev->pers->quiesce(mddev, 1);
7474 mddev->pers->quiesce(mddev, 0);
7475 }
5fd6c1dc 7476 }
1da177e4 7477
61df9d91
N
7478 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7479 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7480 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7481 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N
7482 "(but not more than %d KB/sec) for %s.\n",
7483 speed_max(mddev), desc);
1da177e4 7484
eea1bf38 7485 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7486
57afd89f 7487 io_sectors = 0;
1da177e4
LT
7488 for (m = 0; m < SYNC_MARKS; m++) {
7489 mark[m] = jiffies;
57afd89f 7490 mark_cnt[m] = io_sectors;
1da177e4
LT
7491 }
7492 last_mark = 0;
7493 mddev->resync_mark = mark[last_mark];
7494 mddev->resync_mark_cnt = mark_cnt[last_mark];
7495
7496 /*
7497 * Tune reconstruction:
7498 */
7499 window = 32*(PAGE_SIZE/512);
ac42450c
JB
7500 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7501 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT
7502
7503 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT
7504 last_check = 0;
7505
7506 if (j>2) {
c91abf5a 7507 printk(KERN_INFO
61df9d91
N
7508 "md: resuming %s of %s from checkpoint.\n",
7509 desc, mdname(mddev));
1da177e4 7510 mddev->curr_resync = j;
72f36d59
N
7511 } else
7512 mddev->curr_resync = 3; /* no longer delayed */
75d3da43 7513 mddev->curr_resync_completed = j;
72f36d59
N
7514 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7515 md_new_event(mddev);
54f89341 7516 update_time = jiffies;
1da177e4 7517
7c2c57c9 7518 blk_start_plug(&plug);
1da177e4 7519 while (j < max_sectors) {
57afd89f 7520 sector_t sectors;
1da177e4 7521
57afd89f 7522 skipped = 0;
97e4f42d 7523
7a91ee1f
N
7524 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7525 ((mddev->curr_resync > mddev->curr_resync_completed &&
7526 (mddev->curr_resync - mddev->curr_resync_completed)
7527 > (max_sectors >> 4)) ||
54f89341 7528 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7a91ee1f
N
7529 (j - mddev->curr_resync_completed)*2
7530 >= mddev->resync_max - mddev->curr_resync_completed
7531 )) {
97e4f42d 7532 /* time to update curr_resync_completed */
97e4f42d
N
7533 wait_event(mddev->recovery_wait,
7534 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7535 mddev->curr_resync_completed = j;
35d78c66 7536 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7537 j > mddev->recovery_cp)
7538 mddev->recovery_cp = j;
54f89341 7539 update_time = jiffies;
070dc6dd 7540 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7541 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7542 }
acb180b0 7543
c91abf5a
N
7544 while (j >= mddev->resync_max &&
7545 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
e62e58a5
N
7546 /* As this condition is controlled by user-space,
7547 * we can block indefinitely, so use '_interruptible'
7548 * to avoid triggering warnings.
7549 */
7550 flush_signals(current); /* just in case */
7551 wait_event_interruptible(mddev->recovery_wait,
7552 mddev->resync_max > j
c91abf5a
N
7553 || test_bit(MD_RECOVERY_INTR,
7554 &mddev->recovery));
e62e58a5 7555 }
acb180b0 7556
c91abf5a
N
7557 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7558 break;
acb180b0 7559
57afd89f 7560 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7561 currspeed < speed_min(mddev));
57afd89f 7562 if (sectors == 0) {
dfc70645 7563 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
c91abf5a 7564 break;
1da177e4 7565 }
57afd89f
N
7566
7567 if (!skipped) { /* actual IO requested */
7568 io_sectors += sectors;
7569 atomic_add(sectors, &mddev->recovery_active);
7570 }
7571
e875ecea
N
7572 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7573 break;
7574
1da177e4 7575 j += sectors;
72f36d59
N
7576 if (j > 2)
7577 mddev->curr_resync = j;
ff4e8d9a 7578 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7579 if (last_check == 0)
e875ecea 7580 /* this is the earliest that rebuild will be
d7603b7e
N
7581 * visible in /proc/mdstat
7582 */
7583 md_new_event(mddev);
57afd89f
N
7584
7585 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT
7586 continue;
7587
57afd89f 7588 last_check = io_sectors;
1da177e4
LT
7589 repeat:
7590 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7591 /* step marks */
7592 int next = (last_mark+1) % SYNC_MARKS;
7593
7594 mddev->resync_mark = mark[next];
7595 mddev->resync_mark_cnt = mark_cnt[next];
7596 mark[next] = jiffies;
57afd89f 7597 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT
7598 last_mark = next;
7599 }
7600
c91abf5a
N
7601 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7602 break;
1da177e4
LT
7603
7604 /*
7605 * this loop exits only if either when we are slower than
7606 * the 'hard' speed limit, or the system was IO-idle for
7607 * a jiffy.
7608 * the system might be non-idle CPU-wise, but we only care
7609 * about not overloading the IO subsystem. (things like an
7610 * e2fsck being done on the RAID array should execute fast)
7611 */
1da177e4
LT
7612 cond_resched();
7613
ac7e50a3
XN
7614 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7615 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
57afd89f 7616 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7617
88202a0c
N
7618 if (currspeed > speed_min(mddev)) {
7619 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7620 !is_mddev_idle(mddev, 0)) {
c0e48521 7621 msleep(500);
1da177e4
LT
7622 goto repeat;
7623 }
7624 }
7625 }
c91abf5a
N
7626 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7627 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7628 ? "interrupted" : "done");
1da177e4
LT
7629 /*
7630 * this also signals 'finished resyncing' to md_stop
7631 */
7c2c57c9 7632 blk_finish_plug(&plug);
1da177e4
LT
7633 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7634
7635 /* tell personality that we are finished */
57afd89f 7636 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7637
dfc70645 7638 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N
7639 mddev->curr_resync > 2) {
7640 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7641 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7642 if (mddev->curr_resync >= mddev->recovery_cp) {
7643 printk(KERN_INFO
61df9d91
N
7644 "md: checkpointing %s of %s.\n",
7645 desc, mdname(mddev));
0a19caab 7646 if (test_bit(MD_RECOVERY_ERROR,
7647 &mddev->recovery))
7648 mddev->recovery_cp =
7649 mddev->curr_resync_completed;
7650 else
7651 mddev->recovery_cp =
7652 mddev->curr_resync;
5fd6c1dc
N
7653 }
7654 } else
7655 mddev->recovery_cp = MaxSector;
7656 } else {
7657 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7658 mddev->curr_resync = MaxSector;
4e59ca7d 7659 rcu_read_lock();
dafb20fa 7660 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7661 if (rdev->raid_disk >= 0 &&
70fffd0b 7662 mddev->delta_disks >= 0 &&
5fd6c1dc
N
7663 !test_bit(Faulty, &rdev->flags) &&
7664 !test_bit(In_sync, &rdev->flags) &&
7665 rdev->recovery_offset < mddev->curr_resync)
7666 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7667 rcu_read_unlock();
5fd6c1dc 7668 }
1da177e4 7669 }
db91ff55 7670 skip:
17571284 7671 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7672
c07b70ad
N
7673 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7674 /* We completed so min/max setting can be forgotten if used. */
7675 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7676 mddev->resync_min = 0;
7677 mddev->resync_max = MaxSector;
7678 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7679 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT
7680 mddev->curr_resync = 0;
7681 wake_up(&resync_wait);
7682 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7683 md_wakeup_thread(mddev->thread);
c6207277 7684 return;
1da177e4 7685}
29269553 7686EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7687
746d3207
N
7688static int remove_and_add_spares(struct mddev *mddev,
7689 struct md_rdev *this)
b4c4c7b8 7690{
3cb03002 7691 struct md_rdev *rdev;
b4c4c7b8 7692 int spares = 0;
f2a371c5 7693 int removed = 0;
b4c4c7b8 7694
dafb20fa 7695 rdev_for_each(rdev, mddev)
746d3207
N
7696 if ((this == NULL || rdev == this) &&
7697 rdev->raid_disk >= 0 &&
6bfe0b49 7698 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N
7699 (test_bit(Faulty, &rdev->flags) ||
7700 ! test_bit(In_sync, &rdev->flags)) &&
7701 atomic_read(&rdev->nr_pending)==0) {
7702 if (mddev->pers->hot_remove_disk(
b8321b68 7703 mddev, rdev) == 0) {
36fad858 7704 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7705 rdev->raid_disk = -1;
f2a371c5 7706 removed++;
b4c4c7b8
N
7707 }
7708 }
90584fc9
JB
7709 if (removed && mddev->kobj.sd)
7710 sysfs_notify(&mddev->kobj, NULL, "degraded");
b4c4c7b8 7711
746d3207
N
7712 if (this)
7713 goto no_add;
7714
dafb20fa 7715 rdev_for_each(rdev, mddev) {
7bfec5f3
N
7716 if (rdev->raid_disk >= 0 &&
7717 !test_bit(In_sync, &rdev->flags) &&
7718 !test_bit(Faulty, &rdev->flags))
7719 spares++;
7ceb17e8
N
7720 if (rdev->raid_disk >= 0)
7721 continue;
7722 if (test_bit(Faulty, &rdev->flags))
7723 continue;
7724 if (mddev->ro &&
8313b8e5
N
7725 ! (rdev->saved_raid_disk >= 0 &&
7726 !test_bit(Bitmap_sync, &rdev->flags)))
7ceb17e8
N
7727 continue;
7728
7eb41885
N
7729 if (rdev->saved_raid_disk < 0)
7730 rdev->recovery_offset = 0;
7ceb17e8
N
7731 if (mddev->pers->
7732 hot_add_disk(mddev, rdev) == 0) {
7733 if (sysfs_link_rdev(mddev, rdev))
7734 /* failure here is OK */;
7735 spares++;
7736 md_new_event(mddev);
7737 set_bit(MD_CHANGE_DEVS, &mddev->flags);
dfc70645 7738 }
b4c4c7b8 7739 }
746d3207 7740no_add:
6dafab6b
N
7741 if (removed)
7742 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N
7743 return spares;
7744}
7ebc0be7 7745
ac05f256
N
7746static void md_start_sync(struct work_struct *ws)
7747{
7748 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7749
7750 mddev->sync_thread = md_register_thread(md_do_sync,
7751 mddev,
7752 "resync");
7753 if (!mddev->sync_thread) {
7754 printk(KERN_ERR "%s: could not start resync"
7755 " thread...\n",
7756 mdname(mddev));
7757 /* leave the spares where they are, it shouldn't hurt */
7758 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7759 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7760 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7761 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7762 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7763 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7764 &mddev->recovery))
7765 if (mddev->sysfs_action)
7766 sysfs_notify_dirent_safe(mddev->sysfs_action);
7767 } else
7768 md_wakeup_thread(mddev->sync_thread);
7769 sysfs_notify_dirent_safe(mddev->sysfs_action);
7770 md_new_event(mddev);
7771}
7772
1da177e4
LT
7773/*
7774 * This routine is regularly called by all per-raid-array threads to
7775 * deal with generic issues like resync and super-block update.
7776 * Raid personalities that don't have a thread (linear/raid0) do not
7777 * need this as they never do any recovery or update the superblock.
7778 *
7779 * It does not do any resync itself, but rather "forks" off other threads
7780 * to do that as needed.
7781 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7782 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7783 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT
7784 * and wakeups up this thread which will reap the thread and finish up.
7785 * This thread also removes any faulty devices (with nr_pending == 0).
7786 *
7787 * The overall approach is:
7788 * 1/ if the superblock needs updating, update it.
7789 * 2/ If a recovery thread is running, don't do anything else.
7790 * 3/ If recovery has finished, clean up, possibly marking spares active.
7791 * 4/ If there are any faulty devices, remove them.
7792 * 5/ If array is degraded, try to add spares devices
7793 * 6/ If array has spares or is not in-sync, start a resync thread.
7794 */
fd01b88c 7795void md_check_recovery(struct mddev *mddev)
1da177e4 7796{
68866e42
JB
7797 if (mddev->suspended)
7798 return;
7799
5f40402d 7800 if (mddev->bitmap)
aa5cbd10 7801 bitmap_daemon_work(mddev);
1da177e4 7802
fca4d848 7803 if (signal_pending(current)) {
31a59e34 7804 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N
7805 printk(KERN_INFO "md: %s in immediate safe mode\n",
7806 mdname(mddev));
7807 mddev->safemode = 2;
7808 }
7809 flush_signals(current);
7810 }
7811
c89a8eee
N
7812 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7813 return;
1da177e4 7814 if ( ! (
142d44c3 7815 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7816 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7817 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7818 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N
7819 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7820 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT
7821 ))
7822 return;
fca4d848 7823
df5b89b3 7824 if (mddev_trylock(mddev)) {
b4c4c7b8 7825 int spares = 0;
fca4d848 7826
c89a8eee 7827 if (mddev->ro) {
7ceb17e8
N
7828 /* On a read-only array we can:
7829 * - remove failed devices
7830 * - add already-in_sync devices if the array itself
7831 * is in-sync.
7832 * As we only add devices that are already in-sync,
7833 * we can activate the spares immediately.
c89a8eee 7834 */
7ceb17e8 7835 remove_and_add_spares(mddev, NULL);
8313b8e5
N
7836 /* There is no thread, but we need to call
7837 * ->spare_active and clear saved_raid_disk
7838 */
2ac295a5 7839 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8313b8e5
N
7840 md_reap_sync_thread(mddev);
7841 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
c89a8eee
N
7842 goto unlock;
7843 }
7844
31a59e34 7845 if (!mddev->external) {
0fd62b86 7846 int did_change = 0;
31a59e34
N
7847 spin_lock_irq(&mddev->write_lock);
7848 if (mddev->safemode &&
7849 !atomic_read(&mddev->writes_pending) &&
7850 !mddev->in_sync &&
7851 mddev->recovery_cp == MaxSector) {
7852 mddev->in_sync = 1;
0fd62b86 7853 did_change = 1;
070dc6dd 7854 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N
7855 }
7856 if (mddev->safemode == 1)
7857 mddev->safemode = 0;
7858 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7859 if (did_change)
00bcb4ac 7860 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7861 }
fca4d848 7862
7a0a5355 7863 if (mddev->flags & MD_UPDATE_SB_FLAGS)
850b2b42 7864 md_update_sb(mddev, 0);
06d91a5f 7865
1da177e4
LT
7866 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7867 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7868 /* resync/recovery still happening */
7869 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7870 goto unlock;
7871 }
7872 if (mddev->sync_thread) {
a91d5ac0 7873 md_reap_sync_thread(mddev);
1da177e4
LT
7874 goto unlock;
7875 }
72a23c21
NB
7876 /* Set RUNNING before clearing NEEDED to avoid
7877 * any transients in the value of "sync_action".
7878 */
72f36d59 7879 mddev->curr_resync_completed = 0;
72a23c21 7880 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
24dd469d
N
7881 /* Clear some bits that don't mean anything, but
7882 * might be left set
7883 */
24dd469d
N
7884 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7885 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7886
ed209584
N
7887 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7888 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
ac05f256 7889 goto not_running;
1da177e4
LT
7890 /* no recovery is running.
7891 * remove any failed drives, then
7892 * add spares if possible.
72f36d59 7893 * Spares are also removed and re-added, to allow
1da177e4
LT
7894 * the personality to fail the re-add.
7895 */
1da177e4 7896
b4c4c7b8 7897 if (mddev->reshape_position != MaxSector) {
50ac168a
N
7898 if (mddev->pers->check_reshape == NULL ||
7899 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8 7900 /* Cannot proceed */
ac05f256 7901 goto not_running;
b4c4c7b8 7902 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7903 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
746d3207 7904 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
24dd469d
N
7905 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7906 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7907 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7908 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N
7909 } else if (mddev->recovery_cp < MaxSector) {
7910 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7911 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N
7912 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7913 /* nothing to be done ... */
ac05f256 7914 goto not_running;
24dd469d 7915
1da177e4 7916 if (mddev->pers->sync_request) {
ef99bf48 7917 if (spares) {
a654b9d8
N
7918 /* We are adding a device or devices to an array
7919 * which has the bitmap stored on all devices.
7920 * So make sure all bitmap pages get written
7921 */
7922 bitmap_write_all(mddev->bitmap);
7923 }
ac05f256
N
7924 INIT_WORK(&mddev->del_work, md_start_sync);
7925 queue_work(md_misc_wq, &mddev->del_work);
7926 goto unlock;
1da177e4 7927 }
ac05f256 7928 not_running:
72a23c21
NB
7929 if (!mddev->sync_thread) {
7930 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7931 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7932 &mddev->recovery))
0c3573f1 7933 if (mddev->sysfs_action)
00bcb4ac 7934 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7935 }
ac05f256
N
7936 unlock:
7937 wake_up(&mddev->sb_wait);
1da177e4
LT
7938 mddev_unlock(mddev);
7939 }
7940}
7941
a91d5ac0
JB
7942void md_reap_sync_thread(struct mddev *mddev)
7943{
7944 struct md_rdev *rdev;
7945
7946 /* resync has finished, collect result */
7947 md_unregister_thread(&mddev->sync_thread);
30b8feb7 7948 wake_up(&resync_wait);
a91d5ac0
JB
7949 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7950 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7951 /* success...*/
7952 /* activate any spares */
7953 if (mddev->pers->spare_active(mddev)) {
7954 sysfs_notify(&mddev->kobj, NULL,
7955 "degraded");
7956 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7957 }
7958 }
7959 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7960 mddev->pers->finish_reshape)
7961 mddev->pers->finish_reshape(mddev);
7962
7963 /* If array is no-longer degraded, then any saved_raid_disk
f466722c 7964 * information must be scrapped.
a91d5ac0 7965 */
f466722c
N
7966 if (!mddev->degraded)
7967 rdev_for_each(rdev, mddev)
a91d5ac0
JB
7968 rdev->saved_raid_disk = -1;
7969
7970 md_update_sb(mddev, 1);
7971 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7972 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7973 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7974 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7975 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7976 /* flag recovery needed just to double check */
7977 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7978 sysfs_notify_dirent_safe(mddev->sysfs_action);
7979 md_new_event(mddev);
7980 if (mddev->event_work.func)
7981 queue_work(md_misc_wq, &mddev->event_work);
7982}
7983
fd01b88c 7984void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7985{
00bcb4ac 7986 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7987 wait_event_timeout(rdev->blocked_wait,
de393cde
N
7988 !test_bit(Blocked, &rdev->flags) &&
7989 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW
7990 msecs_to_jiffies(5000));
7991 rdev_dec_pending(rdev, mddev);
7992}
7993EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7994
c6563a8c
N
7995void md_finish_reshape(struct mddev *mddev)
7996{
7997 /* called be personality module when reshape completes. */
7998 struct md_rdev *rdev;
7999
8000 rdev_for_each(rdev, mddev) {
8001 if (rdev->data_offset > rdev->new_data_offset)
8002 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8003 else
8004 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8005 rdev->data_offset = rdev->new_data_offset;
8006 }
8007}
8008EXPORT_SYMBOL(md_finish_reshape);
2230dfe4
N
8009
8010/* Bad block management.
8011 * We can record which blocks on each device are 'bad' and so just
8012 * fail those blocks, or that stripe, rather than the whole device.
8013 * Entries in the bad-block table are 64bits wide. This comprises:
8014 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8015 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8016 * A 'shift' can be set so that larger blocks are tracked and
8017 * consequently larger devices can be covered.
8018 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8019 *
8020 * Locking of the bad-block table uses a seqlock so md_is_badblock
8021 * might need to retry if it is very unlucky.
8022 * We will sometimes want to check for bad blocks in a bi_end_io function,
8023 * so we use the write_seqlock_irq variant.
8024 *
8025 * When looking for a bad block we specify a range and want to
8026 * know if any block in the range is bad. So we binary-search
8027 * to the last range that starts at-or-before the given endpoint,
8028 * (or "before the sector after the target range")
8029 * then see if it ends after the given start.
8030 * We return
8031 * 0 if there are no known bad blocks in the range
8032 * 1 if there are known bad block which are all acknowledged
8033 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8034 * plus the start/length of the first bad section we overlap.
8035 */
8036int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8037 sector_t *first_bad, int *bad_sectors)
8038{
8039 int hi;
ab05613a 8040 int lo;
2230dfe4 8041 u64 *p = bb->page;
ab05613a 8042 int rv;
2230dfe4
N
8043 sector_t target = s + sectors;
8044 unsigned seq;
8045
8046 if (bb->shift > 0) {
8047 /* round the start down, and the end up */
8048 s >>= bb->shift;
8049 target += (1<<bb->shift) - 1;
8050 target >>= bb->shift;
8051 sectors = target - s;
8052 }
8053 /* 'target' is now the first block after the bad range */
8054
8055retry:
8056 seq = read_seqbegin(&bb->lock);
ab05613a 8057 lo = 0;
8058 rv = 0;
2230dfe4
N
8059 hi = bb->count;
8060
8061 /* Binary search between lo and hi for 'target'
8062 * i.e. for the last range that starts before 'target'
8063 */
8064 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8065 * are known not to be the last range before target.
8066 * VARIANT: hi-lo is the number of possible
8067 * ranges, and decreases until it reaches 1
8068 */
8069 while (hi - lo > 1) {
8070 int mid = (lo + hi) / 2;
8071 sector_t a = BB_OFFSET(p[mid]);
8072 if (a < target)
8073 /* This could still be the one, earlier ranges
8074 * could not. */
8075 lo = mid;
8076 else
8077 /* This and later ranges are definitely out. */
8078 hi = mid;
8079 }
8080 /* 'lo' might be the last that started before target, but 'hi' isn't */
8081 if (hi > lo) {
8082 /* need to check all range that end after 's' to see if
8083 * any are unacknowledged.
8084 */
8085 while (lo >= 0 &&
8086 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8087 if (BB_OFFSET(p[lo]) < target) {
8088 /* starts before the end, and finishes after
8089 * the start, so they must overlap
8090 */
8091 if (rv != -1 && BB_ACK(p[lo]))
8092 rv = 1;
8093 else
8094 rv = -1;
8095 *first_bad = BB_OFFSET(p[lo]);
8096 *bad_sectors = BB_LEN(p[lo]);
8097 }
8098 lo--;
8099 }
8100 }
8101
8102 if (read_seqretry(&bb->lock, seq))
8103 goto retry;
8104
8105 return rv;
8106}
8107EXPORT_SYMBOL_GPL(md_is_badblock);
8108
8109/*
8110 * Add a range of bad blocks to the table.
8111 * This might extend the table, or might contract it
8112 * if two adjacent ranges can be merged.
8113 * We binary-search to find the 'insertion' point, then
8114 * decide how best to handle it.
8115 */
8116static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8117 int acknowledged)
8118{
8119 u64 *p;
8120 int lo, hi;
8121 int rv = 1;
905b0297 8122 unsigned long flags;
2230dfe4
N
8123
8124 if (bb->shift < 0)
8125 /* badblocks are disabled */
8126 return 0;
8127
8128 if (bb->shift) {
8129 /* round the start down, and the end up */
8130 sector_t next = s + sectors;
8131 s >>= bb->shift;
8132 next += (1<<bb->shift) - 1;
8133 next >>= bb->shift;
8134 sectors = next - s;
8135 }
8136
905b0297 8137 write_seqlock_irqsave(&bb->lock, flags);
2230dfe4
N
8138
8139 p = bb->page;
8140 lo = 0;
8141 hi = bb->count;
8142 /* Find the last range that starts at-or-before 's' */
8143 while (hi - lo > 1) {
8144 int mid = (lo + hi) / 2;
8145 sector_t a = BB_OFFSET(p[mid]);
8146 if (a <= s)
8147 lo = mid;
8148 else
8149 hi = mid;
8150 }
8151 if (hi > lo && BB_OFFSET(p[lo]) > s)
8152 hi = lo;
8153
8154 if (hi > lo) {
8155 /* we found a range that might merge with the start
8156 * of our new range
8157 */
8158 sector_t a = BB_OFFSET(p[lo]);
8159 sector_t e = a + BB_LEN(p[lo]);
8160 int ack = BB_ACK(p[lo]);
8161 if (e >= s) {
8162 /* Yes, we can merge with a previous range */
8163 if (s == a && s + sectors >= e)
8164 /* new range covers old */
8165 ack = acknowledged;
8166 else
8167 ack = ack && acknowledged;
8168
8169 if (e < s + sectors)
8170 e = s + sectors;
8171 if (e - a <= BB_MAX_LEN) {
8172 p[lo] = BB_MAKE(a, e-a, ack);
8173 s = e;
8174 } else {
8175 /* does not all fit in one range,
8176 * make p[lo] maximal
8177 */
8178 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8179 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8180 s = a + BB_MAX_LEN;
8181 }
8182 sectors = e - s;
8183 }
8184 }
8185 if (sectors && hi < bb->count) {
8186 /* 'hi' points to the first range that starts after 's'.
8187 * Maybe we can merge with the start of that range */
8188 sector_t a = BB_OFFSET(p[hi]);
8189 sector_t e = a + BB_LEN(p[hi]);
8190 int ack = BB_ACK(p[hi]);
8191 if (a <= s + sectors) {
8192 /* merging is possible */
8193 if (e <= s + sectors) {
8194 /* full overlap */
8195 e = s + sectors;
8196 ack = acknowledged;
8197 } else
8198 ack = ack && acknowledged;
8199
8200 a = s;
8201 if (e - a <= BB_MAX_LEN) {
8202 p[hi] = BB_MAKE(a, e-a, ack);
8203 s = e;
8204 } else {
8205 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8206 s = a + BB_MAX_LEN;
8207 }
8208 sectors = e - s;
8209 lo = hi;
8210 hi++;
8211 }
8212 }
8213 if (sectors == 0 && hi < bb->count) {
8214 /* we might be able to combine lo and hi */
8215 /* Note: 's' is at the end of 'lo' */
8216 sector_t a = BB_OFFSET(p[hi]);
8217 int lolen = BB_LEN(p[lo]);
8218 int hilen = BB_LEN(p[hi]);
8219 int newlen = lolen + hilen - (s - a);
8220 if (s >= a && newlen < BB_MAX_LEN) {
8221 /* yes, we can combine them */
8222 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8223 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8224 memmove(p + hi, p + hi + 1,
8225 (bb->count - hi - 1) * 8);
8226 bb->count--;
8227 }
8228 }
8229 while (sectors) {
8230 /* didn't merge (it all).
8231 * Need to add a range just before 'hi' */
8232 if (bb->count >= MD_MAX_BADBLOCKS) {
8233 /* No room for more */
8234 rv = 0;
8235 break;
8236 } else {
8237 int this_sectors = sectors;
8238 memmove(p + hi + 1, p + hi,
8239 (bb->count - hi) * 8);
8240 bb->count++;
8241
8242 if (this_sectors > BB_MAX_LEN)
8243 this_sectors = BB_MAX_LEN;
8244 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8245 sectors -= this_sectors;
8246 s += this_sectors;
8247 }
8248 }
8249
8250 bb->changed = 1;
de393cde
N
8251 if (!acknowledged)
8252 bb->unacked_exist = 1;
905b0297 8253 write_sequnlock_irqrestore(&bb->lock, flags);
2230dfe4
N
8254
8255 return rv;
8256}
8257
3cb03002 8258int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
c6563a8c 8259 int is_new)
2230dfe4 8260{
c6563a8c
N
8261 int rv;
8262 if (is_new)
8263 s += rdev->new_data_offset;
8264 else
8265 s += rdev->data_offset;
8266 rv = md_set_badblocks(&rdev->badblocks,
8267 s, sectors, 0);
2230dfe4
N
8268 if (rv) {
8269 /* Make sure they get written out promptly */
8bd2f0a0 8270 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N
8271 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8272 md_wakeup_thread(rdev->mddev->thread);
8273 }
8274 return rv;
8275}
8276EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8277
8278/*
8279 * Remove a range of bad blocks from the table.
8280 * This may involve extending the table if we spilt a region,
8281 * but it must not fail. So if the table becomes full, we just
8282 * drop the remove request.
8283 */
8284static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8285{
8286 u64 *p;
8287 int lo, hi;
8288 sector_t target = s + sectors;
8289 int rv = 0;
8290
8291 if (bb->shift > 0) {
8292 /* When clearing we round the start up and the end down.
8293 * This should not matter as the shift should align with
8294 * the block size and no rounding should ever be needed.
8295 * However it is better the think a block is bad when it
8296 * isn't than to think a block is not bad when it is.
8297 */
8298 s += (1<<bb->shift) - 1;
8299 s >>= bb->shift;
8300 target >>= bb->shift;
8301 sectors = target - s;
8302 }
8303
8304 write_seqlock_irq(&bb->lock);
8305
8306 p = bb->page;
8307 lo = 0;
8308 hi = bb->count;
8309 /* Find the last range that starts before 'target' */
8310 while (hi - lo > 1) {
8311 int mid = (lo + hi) / 2;
8312 sector_t a = BB_OFFSET(p[mid]);
8313 if (a < target)
8314 lo = mid;
8315 else
8316 hi = mid;
8317 }
8318 if (hi > lo) {
8319 /* p[lo] is the last range that could overlap the
8320 * current range. Earlier ranges could also overlap,
8321 * but only this one can overlap the end of the range.
8322 */
8323 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8324 /* Partial overlap, leave the tail of this range */
8325 int ack = BB_ACK(p[lo]);
8326 sector_t a = BB_OFFSET(p[lo]);
8327 sector_t end = a + BB_LEN(p[lo]);
8328
8329 if (a < s) {
8330 /* we need to split this range */
8331 if (bb->count >= MD_MAX_BADBLOCKS) {
8b32bf5e 8332 rv = -ENOSPC;
2230dfe4
N
8333 goto out;
8334 }
8335 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8336 bb->count++;
8337 p[lo] = BB_MAKE(a, s-a, ack);
8338 lo++;
8339 }
8340 p[lo] = BB_MAKE(target, end - target, ack);
8341 /* there is no longer an overlap */
8342 hi = lo;
8343 lo--;
8344 }
8345 while (lo >= 0 &&
8346 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8347 /* This range does overlap */
8348 if (BB_OFFSET(p[lo]) < s) {
8349 /* Keep the early parts of this range. */
8350 int ack = BB_ACK(p[lo]);
8351 sector_t start = BB_OFFSET(p[lo]);
8352 p[lo] = BB_MAKE(start, s - start, ack);
8353 /* now low doesn't overlap, so.. */
8354 break;
8355 }
8356 lo--;
8357 }
8358 /* 'lo' is strictly before, 'hi' is strictly after,
8359 * anything between needs to be discarded
8360 */
8361 if (hi - lo > 1) {
8362 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8363 bb->count -= (hi - lo - 1);
8364 }
8365 }
8366
8367 bb->changed = 1;
8368out:
8369 write_sequnlock_irq(&bb->lock);
8370 return rv;
8371}
8372
c6563a8c
N
8373int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8374 int is_new)
2230dfe4 8375{
c6563a8c
N
8376 if (is_new)
8377 s += rdev->new_data_offset;
8378 else
8379 s += rdev->data_offset;
2230dfe4 8380 return md_clear_badblocks(&rdev->badblocks,
c6563a8c 8381 s, sectors);
2230dfe4
N
8382}
8383EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8384
8385/*
8386 * Acknowledge all bad blocks in a list.
8387 * This only succeeds if ->changed is clear. It is used by
8388 * in-kernel metadata updates
8389 */
8390void md_ack_all_badblocks(struct badblocks *bb)
8391{
8392 if (bb->page == NULL || bb->changed)
8393 /* no point even trying */
8394 return;
8395 write_seqlock_irq(&bb->lock);
8396
ecb178bb 8397 if (bb->changed == 0 && bb->unacked_exist) {
2230dfe4
N
8398 u64 *p = bb->page;
8399 int i;
8400 for (i = 0; i < bb->count ; i++) {
8401 if (!BB_ACK(p[i])) {
8402 sector_t start = BB_OFFSET(p[i]);
8403 int len = BB_LEN(p[i]);
8404 p[i] = BB_MAKE(start, len, 1);
8405 }
8406 }
de393cde 8407 bb->unacked_exist = 0;
2230dfe4
N
8408 }
8409 write_sequnlock_irq(&bb->lock);
8410}
8411EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8412
16c791a5
N
8413/* sysfs access to bad-blocks list.
8414 * We present two files.
8415 * 'bad-blocks' lists sector numbers and lengths of ranges that
8416 * are recorded as bad. The list is truncated to fit within
8417 * the one-page limit of sysfs.
8418 * Writing "sector length" to this file adds an acknowledged
8419 * bad block list.
8420 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8421 * been acknowledged. Writing to this file adds bad blocks
8422 * without acknowledging them. This is largely for testing.
8423 */
8424
8425static ssize_t
8426badblocks_show(struct badblocks *bb, char *page, int unack)
8427{
8428 size_t len;
8429 int i;
8430 u64 *p = bb->page;
8431 unsigned seq;
8432
8433 if (bb->shift < 0)
8434 return 0;
8435
8436retry:
8437 seq = read_seqbegin(&bb->lock);
8438
8439 len = 0;
8440 i = 0;
8441
8442 while (len < PAGE_SIZE && i < bb->count) {
8443 sector_t s = BB_OFFSET(p[i]);
8444 unsigned int length = BB_LEN(p[i]);
8445 int ack = BB_ACK(p[i]);
8446 i++;
8447
8448 if (unack && ack)
8449 continue;
8450
8451 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8452 (unsigned long long)s << bb->shift,
8453 length << bb->shift);
8454 }
de393cde
N
8455 if (unack && len == 0)
8456 bb->unacked_exist = 0;
16c791a5
N
8457
8458 if (read_seqretry(&bb->lock, seq))
8459 goto retry;
8460
8461 return len;
8462}
8463
8464#define DO_DEBUG 1
8465
8466static ssize_t
8467badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8468{
8469 unsigned long long sector;
8470 int length;
8471 char newline;
8472#ifdef DO_DEBUG
8473 /* Allow clearing via sysfs *only* for testing/debugging.
8474 * Normally only a successful write may clear a badblock
8475 */
8476 int clear = 0;
8477 if (page[0] == '-') {
8478 clear = 1;
8479 page++;
8480 }
8481#endif /* DO_DEBUG */
8482
8483 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8484 case 3:
8485 if (newline != '\n')
8486 return -EINVAL;
8487 case 2:
8488 if (length <= 0)
8489 return -EINVAL;
8490 break;
8491 default:
8492 return -EINVAL;
8493 }
8494
8495#ifdef DO_DEBUG
8496 if (clear) {
8497 md_clear_badblocks(bb, sector, length);
8498 return len;
8499 }
8500#endif /* DO_DEBUG */
8501 if (md_set_badblocks(bb, sector, length, !unack))
8502 return len;
8503 else
8504 return -ENOSPC;
8505}
8506
75c96f85
AB
8507static int md_notify_reboot(struct notifier_block *this,
8508 unsigned long code, void *x)
1da177e4
LT
8509{
8510 struct list_head *tmp;
fd01b88c 8511 struct mddev *mddev;
2dba6a91 8512 int need_delay = 0;
1da177e4 8513
c744a65c
N
8514 for_each_mddev(mddev, tmp) {
8515 if (mddev_trylock(mddev)) {
30b8aa91
N
8516 if (mddev->pers)
8517 __md_stop_writes(mddev);
0f62fb22
N
8518 if (mddev->persistent)
8519 mddev->safemode = 2;
c744a65c 8520 mddev_unlock(mddev);
2dba6a91 8521 }
c744a65c 8522 need_delay = 1;
1da177e4 8523 }
c744a65c
N
8524 /*
8525 * certain more exotic SCSI devices are known to be
8526 * volatile wrt too early system reboots. While the
8527 * right place to handle this issue is the given
8528 * driver, we do want to have a safe RAID driver ...
8529 */
8530 if (need_delay)
8531 mdelay(1000*1);
8532
1da177e4
LT
8533 return NOTIFY_DONE;
8534}
8535
75c96f85 8536static struct notifier_block md_notifier = {
1da177e4
LT
8537 .notifier_call = md_notify_reboot,
8538 .next = NULL,
8539 .priority = INT_MAX, /* before any real devices */
8540};
8541
8542static void md_geninit(void)
8543{
36a4e1fe 8544 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8545
c7705f34 8546 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT
8547}
8548
75c96f85 8549static int __init md_init(void)
1da177e4 8550{
e804ac78
TH
8551 int ret = -ENOMEM;
8552
ada609ee 8553 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH
8554 if (!md_wq)
8555 goto err_wq;
8556
8557 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8558 if (!md_misc_wq)
8559 goto err_misc_wq;
8560
8561 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8562 goto err_md;
8563
8564 if ((ret = register_blkdev(0, "mdp")) < 0)
8565 goto err_mdp;
8566 mdp_major = ret;
8567
af5628f0 8568 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
e8703fe1
N
8569 md_probe, NULL, NULL);
8570 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT
8571 md_probe, NULL, NULL);
8572
1da177e4 8573 register_reboot_notifier(&md_notifier);
0b4d4147 8574 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT
8575
8576 md_geninit();
d710e138 8577 return 0;
1da177e4 8578
e804ac78
TH
8579err_mdp:
8580 unregister_blkdev(MD_MAJOR, "md");
8581err_md:
8582 destroy_workqueue(md_misc_wq);
8583err_misc_wq:
8584 destroy_workqueue(md_wq);
8585err_wq:
8586 return ret;
8587}
1da177e4
LT
8588
8589#ifndef MODULE
8590
8591/*
8592 * Searches all registered partitions for autorun RAID arrays
8593 * at boot time.
8594 */
4d936ec1
ME
8595
8596static LIST_HEAD(all_detected_devices);
8597struct detected_devices_node {
8598 struct list_head list;
8599 dev_t dev;
8600};
1da177e4
LT
8601
8602void md_autodetect_dev(dev_t dev)
8603{
4d936ec1
ME
8604 struct detected_devices_node *node_detected_dev;
8605
8606 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8607 if (node_detected_dev) {
8608 node_detected_dev->dev = dev;
8609 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8610 } else {
8611 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8612 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8613 }
1da177e4
LT
8614}
8615
1da177e4
LT
8616static void autostart_arrays(int part)
8617{
3cb03002 8618 struct md_rdev *rdev;
4d936ec1
ME
8619 struct detected_devices_node *node_detected_dev;
8620 dev_t dev;
8621 int i_scanned, i_passed;
1da177e4 8622
4d936ec1
ME
8623 i_scanned = 0;
8624 i_passed = 0;
1da177e4 8625
4d936ec1 8626 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8627
4d936ec1
ME
8628 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8629 i_scanned++;
8630 node_detected_dev = list_entry(all_detected_devices.next,
8631 struct detected_devices_node, list);
8632 list_del(&node_detected_dev->list);
8633 dev = node_detected_dev->dev;
8634 kfree(node_detected_dev);
df968c4e 8635 rdev = md_import_device(dev,0, 90);
1da177e4
LT
8636 if (IS_ERR(rdev))
8637 continue;
8638
b2d444d7 8639 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT
8640 MD_BUG();
8641 continue;
8642 }
d0fae18f 8643 set_bit(AutoDetected, &rdev->flags);
1da177e4 8644 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8645 i_passed++;
1da177e4 8646 }
4d936ec1
ME
8647
8648 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8649 i_scanned, i_passed);
1da177e4
LT
8650
8651 autorun_devices(part);
8652}
8653
fdee8ae4 8654#endif /* !MODULE */
1da177e4
LT
8655
8656static __exit void md_exit(void)
8657{
fd01b88c 8658 struct mddev *mddev;
1da177e4 8659 struct list_head *tmp;
e2f23b60 8660 int delay = 1;
8ab5e4c1 8661
af5628f0 8662 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
e8703fe1 8663 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8664
3dbd8c2e 8665 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT
8666 unregister_blkdev(mdp_major, "mdp");
8667 unregister_reboot_notifier(&md_notifier);
8668 unregister_sysctl_table(raid_table_header);
e2f23b60
N
8669
8670 /* We cannot unload the modules while some process is
8671 * waiting for us in select() or poll() - wake them up
8672 */
8673 md_unloading = 1;
8674 while (waitqueue_active(&md_event_waiters)) {
8675 /* not safe to leave yet */
8676 wake_up(&md_event_waiters);
8677 msleep(delay);
8678 delay += delay;
8679 }
1da177e4 8680 remove_proc_entry("mdstat", NULL);
e2f23b60 8681
29ac4aa3 8682 for_each_mddev(mddev, tmp) {
1da177e4 8683 export_array(mddev);
d3374825 8684 mddev->hold_active = 0;
1da177e4 8685 }
e804ac78
TH
8686 destroy_workqueue(md_misc_wq);
8687 destroy_workqueue(md_wq);
1da177e4
LT
8688}
8689
685784aa 8690subsys_initcall(md_init);
1da177e4
LT
8691module_exit(md_exit)
8692
f91de92e
N
8693static int get_ro(char *buffer, struct kernel_param *kp)
8694{
8695 return sprintf(buffer, "%d", start_readonly);
8696}
8697static int set_ro(const char *val, struct kernel_param *kp)
8698{
8699 char *e;
8700 int num = simple_strtoul(val, &e, 10);
8701 if (*val && (*e == '\0' || *e == '\n')) {
8702 start_readonly = num;
4dbcdc75 8703 return 0;
f91de92e
N
8704 }
8705 return -EINVAL;
8706}
8707
80ca3a44
N
8708module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8709module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8710
efeb53c0 8711module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8712
1da177e4
LT
8713EXPORT_SYMBOL(register_md_personality);
8714EXPORT_SYMBOL(unregister_md_personality);
8715EXPORT_SYMBOL(md_error);
8716EXPORT_SYMBOL(md_done_sync);
8717EXPORT_SYMBOL(md_write_start);
8718EXPORT_SYMBOL(md_write_end);
1da177e4
LT
8719EXPORT_SYMBOL(md_register_thread);
8720EXPORT_SYMBOL(md_unregister_thread);
8721EXPORT_SYMBOL(md_wakeup_thread);
1da177e4 8722EXPORT_SYMBOL(md_check_recovery);
a91d5ac0 8723EXPORT_SYMBOL(md_reap_sync_thread);
1da177e4 8724MODULE_LICENSE("GPL");
0efb9e61 8725MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8726MODULE_ALIAS("md");
72008652 8727MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);