projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: make blkdev_get/put() handle exclusive access
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
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>
2a48fc0a 39#include <linux/mutex.h>
1da177e4 40#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 41#include <linux/poll.h>
16f17b39 42#include <linux/ctype.h>
e7d2860b 43#include <linux/string.h>
fb4d8c76
N		 44#include <linux/hdreg.h>
45#include <linux/proc_fs.h>
46#include <linux/random.h>
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
LT		 56
57#define DEBUG 0
58#define dprintk(x...) ((void)(DEBUG && printk(x)))
59
1da177e4 60#ifndef MODULE
d710e138 61static void autostart_arrays(int part);
1da177e4
LT		 62#endif
63
2604b703 64static LIST_HEAD(pers_list);
1da177e4
LT		 65static DEFINE_SPINLOCK(pers_lock);
66
5e56341d
AB		 67static void md_print_devices(void);
68
90b08710 69static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 70static struct workqueue_struct *md_wq;
71static struct workqueue_struct *md_misc_wq;
90b08710 72
5e56341d
AB		 73#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
74
1e50915f
RB		 75/*
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
79 */
80#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 81/*
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 85 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
88 * idle IO detection.
89 *
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 91 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 92 */
93
94static int sysctl_speed_limit_min = 1000;
95static int sysctl_speed_limit_max = 200000;
88202a0c
N		 96static inline int speed_min(mddev_t *mddev)
97{
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
100}
101
102static inline int speed_max(mddev_t *mddev)
103{
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
106}
1da177e4
LT		 107
108static struct ctl_table_header *raid_table_header;
109
110static ctl_table raid_table[] = {
111 {
1da177e4
LT		 112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
80ca3a44 115 .mode = S_IRUGO|S_IWUSR,
6d456111 116 .proc_handler = proc_dointvec,
1da177e4
LT		 117 },
118 {
1da177e4
LT		 119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
6d456111 123 .proc_handler = proc_dointvec,
1da177e4 124 },
894d2491 125 { }
1da177e4
LT		 126};
127
128static ctl_table raid_dir_table[] = {
129 {
1da177e4
LT		 130 .procname = "raid",
131 .maxlen = 0,
80ca3a44 132 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 133 .child = raid_table,
134 },
894d2491 135 { }
1da177e4
LT		 136};
137
138static ctl_table raid_root_table[] = {
139 {
1da177e4
LT		 140 .procname = "dev",
141 .maxlen = 0,
142 .mode = 0555,
143 .child = raid_dir_table,
144 },
894d2491 145 { }
1da177e4
LT		 146};
147
83d5cde4 148static const struct block_device_operations md_fops;
1da177e4 149
f91de92e
N		 150static int start_readonly;
151
a167f663
N		 152/* bio_clone_mddev
153 * like bio_clone, but with a local bio set
154 */
155
156static void mddev_bio_destructor(struct bio *bio)
157{
158 mddev_t *mddev, **mddevp;
159
160 mddevp = (void*)bio;
161 mddev = mddevp[-1];
162
163 bio_free(bio, mddev->bio_set);
164}
165
166struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
167 mddev_t *mddev)
168{
169 struct bio *b;
170 mddev_t **mddevp;
171
172 if (!mddev || !mddev->bio_set)
173 return bio_alloc(gfp_mask, nr_iovecs);
174
175 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
176 mddev->bio_set);
177 if (!b)
178 return NULL;
179 mddevp = (void*)b;
180 mddevp[-1] = mddev;
181 b->bi_destructor = mddev_bio_destructor;
182 return b;
183}
184EXPORT_SYMBOL_GPL(bio_alloc_mddev);
185
186struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
187 mddev_t *mddev)
188{
189 struct bio *b;
190 mddev_t **mddevp;
191
192 if (!mddev || !mddev->bio_set)
193 return bio_clone(bio, gfp_mask);
194
195 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
196 mddev->bio_set);
197 if (!b)
198 return NULL;
199 mddevp = (void*)b;
200 mddevp[-1] = mddev;
201 b->bi_destructor = mddev_bio_destructor;
202 __bio_clone(b, bio);
203 if (bio_integrity(bio)) {
204 int ret;
205
206 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
207
208 if (ret < 0) {
209 bio_put(b);
210 return NULL;
211 }
212 }
213
214 return b;
215}
216EXPORT_SYMBOL_GPL(bio_clone_mddev);
217
d7603b7e
N		 218/*
219 * We have a system wide 'event count' that is incremented
220 * on any 'interesting' event, and readers of /proc/mdstat
221 * can use 'poll' or 'select' to find out when the event
222 * count increases.
223 *
224 * Events are:
225 * start array, stop array, error, add device, remove device,
226 * start build, activate spare
227 */
2989ddbd 228static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 229static atomic_t md_event_count;
29269553 230void md_new_event(mddev_t *mddev)
d7603b7e
N		 231{
232 atomic_inc(&md_event_count);
233 wake_up(&md_event_waiters);
234}
29269553 235EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 236
c331eb04
N		 237/* Alternate version that can be called from interrupts
238 * when calling sysfs_notify isn't needed.
239 */
05381954 240static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 241{
242 atomic_inc(&md_event_count);
243 wake_up(&md_event_waiters);
244}
245
1da177e4
LT		 246/*
247 * Enables to iterate over all existing md arrays
248 * all_mddevs_lock protects this list.
249 */
250static LIST_HEAD(all_mddevs);
251static DEFINE_SPINLOCK(all_mddevs_lock);
252
253
254/*
255 * iterates through all used mddevs in the system.
256 * We take care to grab the all_mddevs_lock whenever navigating
257 * the list, and to always hold a refcount when unlocked.
258 * Any code which breaks out of this loop while own
259 * a reference to the current mddev and must mddev_put it.
260 */
29ac4aa3 261#define for_each_mddev(mddev,tmp) \
1da177e4
LT		 262 \
263 for (({ spin_lock(&all_mddevs_lock); \
264 tmp = all_mddevs.next; \
265 mddev = NULL;}); \
266 ({ if (tmp != &all_mddevs) \
267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
268 spin_unlock(&all_mddevs_lock); \
269 if (mddev) mddev_put(mddev); \
270 mddev = list_entry(tmp, mddev_t, all_mddevs); \
271 tmp != &all_mddevs;}); \
272 ({ spin_lock(&all_mddevs_lock); \
273 tmp = tmp->next;}) \
274 )
275
276
409c57f3
N		 277/* Rather than calling directly into the personality make_request function,
278 * IO requests come here first so that we can check if the device is
279 * being suspended pending a reconfiguration.
280 * We hold a refcount over the call to ->make_request. By the time that
281 * call has finished, the bio has been linked into some internal structure
282 * and so is visible to ->quiesce(), so we don't need the refcount any more.
283 */
284static int md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 285{
49077326 286 const int rw = bio_data_dir(bio);
409c57f3
N		 287 mddev_t *mddev = q->queuedata;
288 int rv;
49077326
N		 289 int cpu;
290
409c57f3
N		 291 if (mddev == NULL || mddev->pers == NULL) {
292 bio_io_error(bio);
293 return 0;
294 }
295 rcu_read_lock();
e9c7469b 296 if (mddev->suspended) {
409c57f3
N		 297 DEFINE_WAIT(__wait);
298 for (;;) {
299 prepare_to_wait(&mddev->sb_wait, &__wait,
300 TASK_UNINTERRUPTIBLE);
e9c7469b 301 if (!mddev->suspended)
409c57f3
N		 302 break;
303 rcu_read_unlock();
304 schedule();
305 rcu_read_lock();
306 }
307 finish_wait(&mddev->sb_wait, &__wait);
308 }
309 atomic_inc(&mddev->active_io);
310 rcu_read_unlock();
49077326 311
21a52c6d 312 rv = mddev->pers->make_request(mddev, bio);
49077326
N		 313
314 cpu = part_stat_lock();
315 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
316 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
317 bio_sectors(bio));
318 part_stat_unlock();
319
409c57f3
N		 320 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
321 wake_up(&mddev->sb_wait);
322
323 return rv;
324}
325
9e35b99c
N		 326/* mddev_suspend makes sure no new requests are submitted
327 * to the device, and that any requests that have been submitted
328 * are completely handled.
329 * Once ->stop is called and completes, the module will be completely
330 * unused.
331 */
390ee602 332void mddev_suspend(mddev_t *mddev)
409c57f3
N		 333{
334 BUG_ON(mddev->suspended);
335 mddev->suspended = 1;
336 synchronize_rcu();
337 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
338 mddev->pers->quiesce(mddev, 1);
409c57f3 339}
390ee602 340EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 341
390ee602 342void mddev_resume(mddev_t *mddev)
409c57f3
N		 343{
344 mddev->suspended = 0;
345 wake_up(&mddev->sb_wait);
346 mddev->pers->quiesce(mddev, 0);
1da177e4 347}
390ee602 348EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 349
3fa841d7
N		 350int mddev_congested(mddev_t *mddev, int bits)
351{
352 return mddev->suspended;
353}
354EXPORT_SYMBOL(mddev_congested);
355
a2826aa9 356/*
e9c7469b 357 * Generic flush handling for md
a2826aa9
N		 358 */
359
e9c7469b 360static void md_end_flush(struct bio *bio, int err)
a2826aa9
N		 361{
362 mdk_rdev_t *rdev = bio->bi_private;
363 mddev_t *mddev = rdev->mddev;
a2826aa9
N		 364
365 rdev_dec_pending(rdev, mddev);
366
367 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 368 /* The pre-request flush has finished */
e804ac78 369 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 370 }
371 bio_put(bio);
372}
373
e9c7469b 374static void submit_flushes(mddev_t *mddev)
a2826aa9
N		 375{
376 mdk_rdev_t *rdev;
377
378 rcu_read_lock();
379 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
380 if (rdev->raid_disk >= 0 &&
381 !test_bit(Faulty, &rdev->flags)) {
382 /* Take two references, one is dropped
383 * when request finishes, one after
384 * we reclaim rcu_read_lock
385 */
386 struct bio *bi;
387 atomic_inc(&rdev->nr_pending);
388 atomic_inc(&rdev->nr_pending);
389 rcu_read_unlock();
a167f663 390 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 391 bi->bi_end_io = md_end_flush;
a2826aa9
N		 392 bi->bi_private = rdev;
393 bi->bi_bdev = rdev->bdev;
394 atomic_inc(&mddev->flush_pending);
e9c7469b 395 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 396 rcu_read_lock();
397 rdev_dec_pending(rdev, mddev);
398 }
399 rcu_read_unlock();
400}
401
e9c7469b 402static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 403{
e9c7469b
TH		 404 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
405 struct bio *bio = mddev->flush_bio;
a2826aa9
N		 406
407 atomic_set(&mddev->flush_pending, 1);
408
e9c7469b 409 if (bio->bi_size == 0)
a2826aa9
N		 410 /* an empty barrier - all done */
411 bio_endio(bio, 0);
412 else {
e9c7469b 413 bio->bi_rw &= ~REQ_FLUSH;
21a52c6d 414 if (mddev->pers->make_request(mddev, bio))
a2826aa9 415 generic_make_request(bio);
a2826aa9
N		 416 }
417 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 418 mddev->flush_bio = NULL;
a2826aa9
N		 419 wake_up(&mddev->sb_wait);
420 }
421}
422
e9c7469b 423void md_flush_request(mddev_t *mddev, struct bio *bio)
a2826aa9
N		 424{
425 spin_lock_irq(&mddev->write_lock);
426 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 427 !mddev->flush_bio,
a2826aa9 428 mddev->write_lock, /*nothing*/);
e9c7469b 429 mddev->flush_bio = bio;
a2826aa9
N		 430 spin_unlock_irq(&mddev->write_lock);
431
432 atomic_set(&mddev->flush_pending, 1);
e9c7469b 433 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
a2826aa9 434
e9c7469b 435 submit_flushes(mddev);
a2826aa9
N		 436
437 if (atomic_dec_and_test(&mddev->flush_pending))
e804ac78 438 queue_work(md_wq, &mddev->flush_work);
a2826aa9 439}
e9c7469b 440EXPORT_SYMBOL(md_flush_request);
409c57f3 441
2ac87401
N		 442/* Support for plugging.
443 * This mirrors the plugging support in request_queue, but does not
444 * require having a whole queue
445 */
446static void plugger_work(struct work_struct *work)
447{
448 struct plug_handle *plug =
449 container_of(work, struct plug_handle, unplug_work);
450 plug->unplug_fn(plug);
451}
452static void plugger_timeout(unsigned long data)
453{
454 struct plug_handle *plug = (void *)data;
455 kblockd_schedule_work(NULL, &plug->unplug_work);
456}
457void plugger_init(struct plug_handle *plug,
458 void (*unplug_fn)(struct plug_handle *))
459{
460 plug->unplug_flag = 0;
461 plug->unplug_fn = unplug_fn;
462 init_timer(&plug->unplug_timer);
463 plug->unplug_timer.function = plugger_timeout;
464 plug->unplug_timer.data = (unsigned long)plug;
465 INIT_WORK(&plug->unplug_work, plugger_work);
466}
467EXPORT_SYMBOL_GPL(plugger_init);
468
469void plugger_set_plug(struct plug_handle *plug)
470{
471 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
472 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
473}
474EXPORT_SYMBOL_GPL(plugger_set_plug);
475
476int plugger_remove_plug(struct plug_handle *plug)
477{
478 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
479 del_timer(&plug->unplug_timer);
480 return 1;
481 } else
482 return 0;
483}
484EXPORT_SYMBOL_GPL(plugger_remove_plug);
485
486
1da177e4
LT		 487static inline mddev_t *mddev_get(mddev_t *mddev)
488{
489 atomic_inc(&mddev->active);
490 return mddev;
491}
492
5fd3a17e 493static void mddev_delayed_delete(struct work_struct *ws);
d3374825 494
1da177e4
LT		 495static void mddev_put(mddev_t *mddev)
496{
a167f663
N		 497 struct bio_set *bs = NULL;
498
1da177e4
LT		 499 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
500 return;
d3374825 501 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 502 mddev->ctime == 0 && !mddev->hold_active) {
503 /* Array is not configured at all, and not held active,
504 * so destroy it */
1da177e4 505 list_del(&mddev->all_mddevs);
a167f663
N		 506 bs = mddev->bio_set;
507 mddev->bio_set = NULL;
d3374825 508 if (mddev->gendisk) {
e804ac78
TH		 509 /* We did a probe so need to clean up. Call
510 * queue_work inside the spinlock so that
511 * flush_workqueue() after mddev_find will
512 * succeed in waiting for the work to be done.
d3374825
N		 513 */
514 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 515 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 516 } else
517 kfree(mddev);
518 }
519 spin_unlock(&all_mddevs_lock);
a167f663
N		 520 if (bs)
521 bioset_free(bs);
1da177e4
LT		 522}
523
390ee602 524void mddev_init(mddev_t *mddev)
fafd7fb0
N		 525{
526 mutex_init(&mddev->open_mutex);
527 mutex_init(&mddev->reconfig_mutex);
528 mutex_init(&mddev->bitmap_info.mutex);
529 INIT_LIST_HEAD(&mddev->disks);
530 INIT_LIST_HEAD(&mddev->all_mddevs);
531 init_timer(&mddev->safemode_timer);
532 atomic_set(&mddev->active, 1);
533 atomic_set(&mddev->openers, 0);
534 atomic_set(&mddev->active_io, 0);
535 spin_lock_init(&mddev->write_lock);
536 atomic_set(&mddev->flush_pending, 0);
537 init_waitqueue_head(&mddev->sb_wait);
538 init_waitqueue_head(&mddev->recovery_wait);
539 mddev->reshape_position = MaxSector;
540 mddev->resync_min = 0;
541 mddev->resync_max = MaxSector;
542 mddev->level = LEVEL_NONE;
543}
390ee602 544EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 545
1da177e4
LT		 546static mddev_t * mddev_find(dev_t unit)
547{
548 mddev_t *mddev, *new = NULL;
549
550 retry:
551 spin_lock(&all_mddevs_lock);
efeb53c0
N		 552
553 if (unit) {
554 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
555 if (mddev->unit == unit) {
556 mddev_get(mddev);
557 spin_unlock(&all_mddevs_lock);
558 kfree(new);
559 return mddev;
560 }
561
562 if (new) {
563 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 564 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 565 new->hold_active = UNTIL_IOCTL;
566 return new;
1da177e4 567 }
efeb53c0
N		 568 } else if (new) {
569 /* find an unused unit number */
570 static int next_minor = 512;
571 int start = next_minor;
572 int is_free = 0;
573 int dev = 0;
574 while (!is_free) {
575 dev = MKDEV(MD_MAJOR, next_minor);
576 next_minor++;
577 if (next_minor > MINORMASK)
578 next_minor = 0;
579 if (next_minor == start) {
580 /* Oh dear, all in use. */
581 spin_unlock(&all_mddevs_lock);
582 kfree(new);
583 return NULL;
584 }
585
586 is_free = 1;
587 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
588 if (mddev->unit == dev) {
589 is_free = 0;
590 break;
591 }
592 }
593 new->unit = dev;
594 new->md_minor = MINOR(dev);
595 new->hold_active = UNTIL_STOP;
1da177e4
LT		 596 list_add(&new->all_mddevs, &all_mddevs);
597 spin_unlock(&all_mddevs_lock);
598 return new;
599 }
600 spin_unlock(&all_mddevs_lock);
601
9ffae0cf 602 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 603 if (!new)
604 return NULL;
605
1da177e4
LT		 606 new->unit = unit;
607 if (MAJOR(unit) == MD_MAJOR)
608 new->md_minor = MINOR(unit);
609 else
610 new->md_minor = MINOR(unit) >> MdpMinorShift;
611
fafd7fb0 612 mddev_init(new);
1da177e4 613
1da177e4
LT		 614 goto retry;
615}
616
617static inline int mddev_lock(mddev_t * mddev)
618{
df5b89b3 619 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 620}
621
b522adcd
DW		 622static inline int mddev_is_locked(mddev_t *mddev)
623{
624 return mutex_is_locked(&mddev->reconfig_mutex);
625}
626
1da177e4
LT		 627static inline int mddev_trylock(mddev_t * mddev)
628{
df5b89b3 629 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 630}
631
b6eb127d
N		 632static struct attribute_group md_redundancy_group;
633
a64c876f 634static void mddev_unlock(mddev_t * mddev)
1da177e4 635{
a64c876f 636 if (mddev->to_remove) {
b6eb127d
N		 637 /* These cannot be removed under reconfig_mutex as
638 * an access to the files will try to take reconfig_mutex
639 * while holding the file unremovable, which leads to
640 * a deadlock.
bb4f1e9d
N		 641 * So hold set sysfs_active while the remove in happeing,
642 * and anything else which might set ->to_remove or my
643 * otherwise change the sysfs namespace will fail with
644 * -EBUSY if sysfs_active is still set.
645 * We set sysfs_active under reconfig_mutex and elsewhere
646 * test it under the same mutex to ensure its correct value
647 * is seen.
b6eb127d 648 */
a64c876f
N		 649 struct attribute_group *to_remove = mddev->to_remove;
650 mddev->to_remove = NULL;
bb4f1e9d 651 mddev->sysfs_active = 1;
b6eb127d
N		 652 mutex_unlock(&mddev->reconfig_mutex);
653
00bcb4ac
N		 654 if (mddev->kobj.sd) {
655 if (to_remove != &md_redundancy_group)
656 sysfs_remove_group(&mddev->kobj, to_remove);
657 if (mddev->pers == NULL ||
658 mddev->pers->sync_request == NULL) {
659 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
660 if (mddev->sysfs_action)
661 sysfs_put(mddev->sysfs_action);
662 mddev->sysfs_action = NULL;
663 }
a64c876f 664 }
bb4f1e9d 665 mddev->sysfs_active = 0;
b6eb127d
N		 666 } else
667 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 668
005eca5e 669 md_wakeup_thread(mddev->thread);
1da177e4
LT		 670}
671
2989ddbd 672static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4 673{
159ec1fc 674 mdk_rdev_t *rdev;
1da177e4 675
159ec1fc 676 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 677 if (rdev->desc_nr == nr)
678 return rdev;
159ec1fc 679
1da177e4
LT		 680 return NULL;
681}
682
683static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
684{
1da177e4
LT		 685 mdk_rdev_t *rdev;
686
159ec1fc 687 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 688 if (rdev->bdev->bd_dev == dev)
689 return rdev;
159ec1fc 690
1da177e4
LT		 691 return NULL;
692}
693
d9d166c2 694static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 695{
696 struct mdk_personality *pers;
d9d166c2
N		 697 list_for_each_entry(pers, &pers_list, list) {
698 if (level != LEVEL_NONE && pers->level == level)
2604b703 699 return pers;
d9d166c2
N		 700 if (strcmp(pers->name, clevel)==0)
701 return pers;
702 }
2604b703
N		 703 return NULL;
704}
705
b73df2d3 706/* return the offset of the super block in 512byte sectors */
77933d72 707static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4 708{
b73df2d3
AN		 709 sector_t num_sectors = bdev->bd_inode->i_size / 512;
710 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 711}
712
1da177e4
LT		 713static int alloc_disk_sb(mdk_rdev_t * rdev)
714{
715 if (rdev->sb_page)
716 MD_BUG();
717
718 rdev->sb_page = alloc_page(GFP_KERNEL);
719 if (!rdev->sb_page) {
720 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 721 return -ENOMEM;
1da177e4
LT		 722 }
723
724 return 0;
725}
726
727static void free_disk_sb(mdk_rdev_t * rdev)
728{
729 if (rdev->sb_page) {
2d1f3b5d 730 put_page(rdev->sb_page);
1da177e4
LT		 731 rdev->sb_loaded = 0;
732 rdev->sb_page = NULL;
0f420358 733 rdev->sb_start = 0;
dd8ac336 734 rdev->sectors = 0;
1da177e4
LT		 735 }
736}
737
738
6712ecf8 739static void super_written(struct bio *bio, int error)
7bfa19f2
N		 740{
741 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 742 mddev_t *mddev = rdev->mddev;
7bfa19f2 743
3a0f5bbb
N		 744 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
745 printk("md: super_written gets error=%d, uptodate=%d\n",
746 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
747 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 748 md_error(mddev, rdev);
3a0f5bbb 749 }
7bfa19f2 750
a9701a30
N		 751 if (atomic_dec_and_test(&mddev->pending_writes))
752 wake_up(&mddev->sb_wait);
f8b58edf 753 bio_put(bio);
7bfa19f2
N		 754}
755
756void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
757 sector_t sector, int size, struct page *page)
758{
759 /* write first size bytes of page to sector of rdev
760 * Increment mddev->pending_writes before returning
761 * and decrement it on completion, waking up sb_wait
762 * if zero is reached.
763 * If an error occurred, call md_error
764 */
a167f663 765 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2
N		 766
767 bio->bi_bdev = rdev->bdev;
768 bio->bi_sector = sector;
769 bio_add_page(bio, page, size, 0);
770 bio->bi_private = rdev;
771 bio->bi_end_io = super_written;
a9701a30 772
7bfa19f2 773 atomic_inc(&mddev->pending_writes);
e9c7469b
TH		 774 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
775 bio);
a9701a30
N		 776}
777
778void md_super_wait(mddev_t *mddev)
779{
e9c7469b 780 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 781 DEFINE_WAIT(wq);
782 for(;;) {
783 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
784 if (atomic_read(&mddev->pending_writes)==0)
785 break;
a9701a30
N		 786 schedule();
787 }
788 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 789}
790
6712ecf8 791static void bi_complete(struct bio *bio, int error)
1da177e4 792{
1da177e4 793 complete((struct completion*)bio->bi_private);
1da177e4
LT		 794}
795
2b193363
N		 796int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
797 struct page *page, int rw)
1da177e4 798{
a167f663 799 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 800 struct completion event;
801 int ret;
802
7b6d91da 803 rw |= REQ_SYNC | REQ_UNPLUG;
1da177e4 804
2b193363 805 bio->bi_bdev = rdev->bdev;
1da177e4
LT		 806 bio->bi_sector = sector;
807 bio_add_page(bio, page, size, 0);
808 init_completion(&event);
809 bio->bi_private = &event;
810 bio->bi_end_io = bi_complete;
811 submit_bio(rw, bio);
812 wait_for_completion(&event);
813
814 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
815 bio_put(bio);
816 return ret;
817}
a8745db2 818EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 819
0002b271 820static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 821{
822 char b[BDEVNAME_SIZE];
823 if (!rdev->sb_page) {
824 MD_BUG();
825 return -EINVAL;
826 }
827 if (rdev->sb_loaded)
828 return 0;
829
830
2b193363 831 if (!sync_page_io(rdev, rdev->sb_start, size, rdev->sb_page, READ))
1da177e4
LT		 832 goto fail;
833 rdev->sb_loaded = 1;
834 return 0;
835
836fail:
837 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
838 bdevname(rdev->bdev,b));
839 return -EINVAL;
840}
841
842static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
843{
05710466
AN		 844 return sb1->set_uuid0 == sb2->set_uuid0 &&
845 sb1->set_uuid1 == sb2->set_uuid1 &&
846 sb1->set_uuid2 == sb2->set_uuid2 &&
847 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 848}
849
1da177e4
LT		 850static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
851{
852 int ret;
853 mdp_super_t *tmp1, *tmp2;
854
855 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
856 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
857
858 if (!tmp1 || !tmp2) {
859 ret = 0;
35020f1a 860 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 861 goto abort;
862 }
863
864 *tmp1 = *sb1;
865 *tmp2 = *sb2;
866
867 /*
868 * nr_disks is not constant
869 */
870 tmp1->nr_disks = 0;
871 tmp2->nr_disks = 0;
872
ce0c8e05 873 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 874abort:
990a8baf
JJ		 875 kfree(tmp1);
876 kfree(tmp2);
1da177e4
LT		 877 return ret;
878}
879
4d167f09
N		 880
881static u32 md_csum_fold(u32 csum)
882{
883 csum = (csum & 0xffff) + (csum >> 16);
884 return (csum & 0xffff) + (csum >> 16);
885}
886
1da177e4
LT		 887static unsigned int calc_sb_csum(mdp_super_t * sb)
888{
4d167f09
N		 889 u64 newcsum = 0;
890 u32 *sb32 = (u32*)sb;
891 int i;
1da177e4
LT		 892 unsigned int disk_csum, csum;
893
894 disk_csum = sb->sb_csum;
895 sb->sb_csum = 0;
4d167f09
N		 896
897 for (i = 0; i < MD_SB_BYTES/4 ; i++)
898 newcsum += sb32[i];
899 csum = (newcsum & 0xffffffff) + (newcsum>>32);
900
901
902#ifdef CONFIG_ALPHA
903 /* This used to use csum_partial, which was wrong for several
904 * reasons including that different results are returned on
905 * different architectures. It isn't critical that we get exactly
906 * the same return value as before (we always csum_fold before
907 * testing, and that removes any differences). However as we
908 * know that csum_partial always returned a 16bit value on
909 * alphas, do a fold to maximise conformity to previous behaviour.
910 */
911 sb->sb_csum = md_csum_fold(disk_csum);
912#else
1da177e4 913 sb->sb_csum = disk_csum;
4d167f09 914#endif
1da177e4
LT		 915 return csum;
916}
917
918
919/*
920 * Handle superblock details.
921 * We want to be able to handle multiple superblock formats
922 * so we have a common interface to them all, and an array of
923 * different handlers.
924 * We rely on user-space to write the initial superblock, and support
925 * reading and updating of superblocks.
926 * Interface methods are:
927 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
928 * loads and validates a superblock on dev.
929 * if refdev != NULL, compare superblocks on both devices
930 * Return:
931 * 0 - dev has a superblock that is compatible with refdev
932 * 1 - dev has a superblock that is compatible and newer than refdev
933 * so dev should be used as the refdev in future
934 * -EINVAL superblock incompatible or invalid
935 * -othererror e.g. -EIO
936 *
937 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
938 * Verify that dev is acceptable into mddev.
939 * The first time, mddev->raid_disks will be 0, and data from
940 * dev should be merged in. Subsequent calls check that dev
941 * is new enough. Return 0 or -EINVAL
942 *
943 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
944 * Update the superblock for rdev with data in mddev
945 * This does not write to disc.
946 *
947 */
948
949struct super_type {
0cd17fec
CW		 950 char *name;
951 struct module *owner;
952 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
953 int minor_version);
954 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
955 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
956 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
15f4a5fd 957 sector_t num_sectors);
1da177e4
LT		 958};
959
0894cc30
AN		 960/*
961 * Check that the given mddev has no bitmap.
962 *
963 * This function is called from the run method of all personalities that do not
964 * support bitmaps. It prints an error message and returns non-zero if mddev
965 * has a bitmap. Otherwise, it returns 0.
966 *
967 */
968int md_check_no_bitmap(mddev_t *mddev)
969{
c3d9714e 970 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 971 return 0;
972 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
973 mdname(mddev), mddev->pers->name);
974 return 1;
975}
976EXPORT_SYMBOL(md_check_no_bitmap);
977
1da177e4
LT		 978/*
979 * load_super for 0.90.0
980 */
981static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
982{
983 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
984 mdp_super_t *sb;
985 int ret;
1da177e4
LT		 986
987 /*
0f420358 988 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 989 * it's at the end of the disk.
990 *
991 * It also happens to be a multiple of 4Kb.
992 */
0f420358 993 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1da177e4 994
0002b271 995 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 996 if (ret) return ret;
997
998 ret = -EINVAL;
999
1000 bdevname(rdev->bdev, b);
1001 sb = (mdp_super_t*)page_address(rdev->sb_page);
1002
1003 if (sb->md_magic != MD_SB_MAGIC) {
1004 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1005 b);
1006 goto abort;
1007 }
1008
1009 if (sb->major_version != 0 ||
f6705578
N		 1010 sb->minor_version < 90 ||
1011 sb->minor_version > 91) {
1da177e4
LT		 1012 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1013 sb->major_version, sb->minor_version,
1014 b);
1015 goto abort;
1016 }
1017
1018 if (sb->raid_disks <= 0)
1019 goto abort;
1020
4d167f09 1021 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1022 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1023 b);
1024 goto abort;
1025 }
1026
1027 rdev->preferred_minor = sb->md_minor;
1028 rdev->data_offset = 0;
0002b271 1029 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 1030
1031 if (sb->level == LEVEL_MULTIPATH)
1032 rdev->desc_nr = -1;
1033 else
1034 rdev->desc_nr = sb->this_disk.number;
1035
9a7b2b0f 1036 if (!refdev) {
1da177e4 1037 ret = 1;
9a7b2b0f 1038 } else {
1da177e4
LT		 1039 __u64 ev1, ev2;
1040 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1041 if (!uuid_equal(refsb, sb)) {
1042 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1043 b, bdevname(refdev->bdev,b2));
1044 goto abort;
1045 }
1046 if (!sb_equal(refsb, sb)) {
1047 printk(KERN_WARNING "md: %s has same UUID"
1048 " but different superblock to %s\n",
1049 b, bdevname(refdev->bdev, b2));
1050 goto abort;
1051 }
1052 ev1 = md_event(sb);
1053 ev2 = md_event(refsb);
1054 if (ev1 > ev2)
1055 ret = 1;
1056 else
1057 ret = 0;
1058 }
8190e754 1059 rdev->sectors = rdev->sb_start;
1da177e4 1060
dd8ac336 1061 if (rdev->sectors < sb->size * 2 && sb->level > 1)
2bf071bf
N		 1062 /* "this cannot possibly happen" ... */
1063 ret = -EINVAL;
1064
1da177e4
LT		 1065 abort:
1066 return ret;
1067}
1068
1069/*
1070 * validate_super for 0.90.0
1071 */
1072static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1073{
1074 mdp_disk_t *desc;
1075 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 1076 __u64 ev1 = md_event(sb);
1da177e4 1077
41158c7e 1078 rdev->raid_disk = -1;
c5d79adb
N		 1079 clear_bit(Faulty, &rdev->flags);
1080 clear_bit(In_sync, &rdev->flags);
1081 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1082
1da177e4
LT		 1083 if (mddev->raid_disks == 0) {
1084 mddev->major_version = 0;
1085 mddev->minor_version = sb->minor_version;
1086 mddev->patch_version = sb->patch_version;
e691063a 1087 mddev->external = 0;
9d8f0363 1088 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1089 mddev->ctime = sb->ctime;
1090 mddev->utime = sb->utime;
1091 mddev->level = sb->level;
d9d166c2 1092 mddev->clevel[0] = 0;
1da177e4
LT		 1093 mddev->layout = sb->layout;
1094 mddev->raid_disks = sb->raid_disks;
58c0fed4 1095 mddev->dev_sectors = sb->size * 2;
07d84d10 1096 mddev->events = ev1;
c3d9714e
N		 1097 mddev->bitmap_info.offset = 0;
1098 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1099
f6705578
N		 1100 if (mddev->minor_version >= 91) {
1101 mddev->reshape_position = sb->reshape_position;
1102 mddev->delta_disks = sb->delta_disks;
1103 mddev->new_level = sb->new_level;
1104 mddev->new_layout = sb->new_layout;
664e7c41 1105 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1106 } else {
1107 mddev->reshape_position = MaxSector;
1108 mddev->delta_disks = 0;
1109 mddev->new_level = mddev->level;
1110 mddev->new_layout = mddev->layout;
664e7c41 1111 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1112 }
1113
1da177e4
LT		 1114 if (sb->state & (1<<MD_SB_CLEAN))
1115 mddev->recovery_cp = MaxSector;
1116 else {
1117 if (sb->events_hi == sb->cp_events_hi &&
1118 sb->events_lo == sb->cp_events_lo) {
1119 mddev->recovery_cp = sb->recovery_cp;
1120 } else
1121 mddev->recovery_cp = 0;
1122 }
1123
1124 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1125 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1126 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1127 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1128
1129 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1130
1131 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1132 mddev->bitmap_info.file == NULL)
1133 mddev->bitmap_info.offset =
1134 mddev->bitmap_info.default_offset;
a654b9d8 1135
41158c7e 1136 } else if (mddev->pers == NULL) {
be6800a7
N		 1137 /* Insist on good event counter while assembling, except
1138 * for spares (which don't need an event count) */
1da177e4 1139 ++ev1;
be6800a7
N		 1140 if (sb->disks[rdev->desc_nr].state & (
1141 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1142 if (ev1 < mddev->events)
1143 return -EINVAL;
41158c7e
N		 1144 } else if (mddev->bitmap) {
1145 /* if adding to array with a bitmap, then we can accept an
1146 * older device ... but not too old.
1147 */
41158c7e
N		 1148 if (ev1 < mddev->bitmap->events_cleared)
1149 return 0;
07d84d10
N		 1150 } else {
1151 if (ev1 < mddev->events)
1152 /* just a hot-add of a new device, leave raid_disk at -1 */
1153 return 0;
1154 }
41158c7e 1155
1da177e4 1156 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1157 desc = sb->disks + rdev->desc_nr;
1158
1159 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1160 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1161 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1162 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1163 set_bit(In_sync, &rdev->flags);
1da177e4 1164 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1165 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1166 /* active but not in sync implies recovery up to
1167 * reshape position. We don't know exactly where
1168 * that is, so set to zero for now */
1169 if (mddev->minor_version >= 91) {
1170 rdev->recovery_offset = 0;
1171 rdev->raid_disk = desc->raid_disk;
1172 }
1da177e4 1173 }
8ddf9efe
N		 1174 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1175 set_bit(WriteMostly, &rdev->flags);
41158c7e 1176 } else /* MULTIPATH are always insync */
b2d444d7 1177 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1178 return 0;
1179}
1180
1181/*
1182 * sync_super for 0.90.0
1183 */
1184static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1185{
1186 mdp_super_t *sb;
1da177e4
LT		 1187 mdk_rdev_t *rdev2;
1188 int next_spare = mddev->raid_disks;
19133a42 1189
1da177e4
LT		 1190
1191 /* make rdev->sb match mddev data..
1192 *
1193 * 1/ zero out disks
1194 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1195 * 3/ any empty disks < next_spare become removed
1196 *
1197 * disks[0] gets initialised to REMOVED because
1198 * we cannot be sure from other fields if it has
1199 * been initialised or not.
1200 */
1201 int i;
1202 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1203
61181565
N		 1204 rdev->sb_size = MD_SB_BYTES;
1205
1da177e4
LT		 1206 sb = (mdp_super_t*)page_address(rdev->sb_page);
1207
1208 memset(sb, 0, sizeof(*sb));
1209
1210 sb->md_magic = MD_SB_MAGIC;
1211 sb->major_version = mddev->major_version;
1da177e4
LT		 1212 sb->patch_version = mddev->patch_version;
1213 sb->gvalid_words = 0; /* ignored */
1214 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1215 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1216 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1217 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1218
1219 sb->ctime = mddev->ctime;
1220 sb->level = mddev->level;
58c0fed4 1221 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1222 sb->raid_disks = mddev->raid_disks;
1223 sb->md_minor = mddev->md_minor;
e691063a 1224 sb->not_persistent = 0;
1da177e4
LT		 1225 sb->utime = mddev->utime;
1226 sb->state = 0;
1227 sb->events_hi = (mddev->events>>32);
1228 sb->events_lo = (u32)mddev->events;
1229
f6705578
N		 1230 if (mddev->reshape_position == MaxSector)
1231 sb->minor_version = 90;
1232 else {
1233 sb->minor_version = 91;
1234 sb->reshape_position = mddev->reshape_position;
1235 sb->new_level = mddev->new_level;
1236 sb->delta_disks = mddev->delta_disks;
1237 sb->new_layout = mddev->new_layout;
664e7c41 1238 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1239 }
1240 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1241 if (mddev->in_sync)
1242 {
1243 sb->recovery_cp = mddev->recovery_cp;
1244 sb->cp_events_hi = (mddev->events>>32);
1245 sb->cp_events_lo = (u32)mddev->events;
1246 if (mddev->recovery_cp == MaxSector)
1247 sb->state = (1<< MD_SB_CLEAN);
1248 } else
1249 sb->recovery_cp = 0;
1250
1251 sb->layout = mddev->layout;
9d8f0363 1252 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1253
c3d9714e 1254 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1255 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1256
1da177e4 1257 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1258 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1259 mdp_disk_t *d;
86e6ffdd 1260 int desc_nr;
0261cd9f
N		 1261 int is_active = test_bit(In_sync, &rdev2->flags);
1262
1263 if (rdev2->raid_disk >= 0 &&
1264 sb->minor_version >= 91)
1265 /* we have nowhere to store the recovery_offset,
1266 * but if it is not below the reshape_position,
1267 * we can piggy-back on that.
1268 */
1269 is_active = 1;
1270 if (rdev2->raid_disk < 0 ||
1271 test_bit(Faulty, &rdev2->flags))
1272 is_active = 0;
1273 if (is_active)
86e6ffdd 1274 desc_nr = rdev2->raid_disk;
1da177e4 1275 else
86e6ffdd 1276 desc_nr = next_spare++;
19133a42 1277 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1278 d = &sb->disks[rdev2->desc_nr];
1279 nr_disks++;
1280 d->number = rdev2->desc_nr;
1281 d->major = MAJOR(rdev2->bdev->bd_dev);
1282 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1283 if (is_active)
1da177e4
LT		 1284 d->raid_disk = rdev2->raid_disk;
1285 else
1286 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1287 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1288 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1289 else if (is_active) {
1da177e4 1290 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1291 if (test_bit(In_sync, &rdev2->flags))
1292 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1293 active++;
1294 working++;
1295 } else {
1296 d->state = 0;
1297 spare++;
1298 working++;
1299 }
8ddf9efe
N		 1300 if (test_bit(WriteMostly, &rdev2->flags))
1301 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1302 }
1da177e4
LT		 1303 /* now set the "removed" and "faulty" bits on any missing devices */
1304 for (i=0 ; i < mddev->raid_disks ; i++) {
1305 mdp_disk_t *d = &sb->disks[i];
1306 if (d->state == 0 && d->number == 0) {
1307 d->number = i;
1308 d->raid_disk = i;
1309 d->state = (1<<MD_DISK_REMOVED);
1310 d->state |= (1<<MD_DISK_FAULTY);
1311 failed++;
1312 }
1313 }
1314 sb->nr_disks = nr_disks;
1315 sb->active_disks = active;
1316 sb->working_disks = working;
1317 sb->failed_disks = failed;
1318 sb->spare_disks = spare;
1319
1320 sb->this_disk = sb->disks[rdev->desc_nr];
1321 sb->sb_csum = calc_sb_csum(sb);
1322}
1323
0cd17fec
CW		 1324/*
1325 * rdev_size_change for 0.90.0
1326 */
1327static unsigned long long
15f4a5fd 1328super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec 1329{
58c0fed4 1330 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1331 return 0; /* component must fit device */
c3d9714e 1332 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1333 return 0; /* can't move bitmap */
0f420358 1334 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
15f4a5fd
AN		 1335 if (!num_sectors || num_sectors > rdev->sb_start)
1336 num_sectors = rdev->sb_start;
0f420358 1337 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1338 rdev->sb_page);
1339 md_super_wait(rdev->mddev);
15f4a5fd 1340 return num_sectors / 2; /* kB for sysfs */
0cd17fec
CW		 1341}
1342
1343
1da177e4
LT		 1344/*
1345 * version 1 superblock
1346 */
1347
1c05b4bc 1348static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1349{
1c05b4bc
N		 1350 __le32 disk_csum;
1351 u32 csum;
1da177e4
LT		 1352 unsigned long long newcsum;
1353 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1354 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1355 int i;
1356
1357 disk_csum = sb->sb_csum;
1358 sb->sb_csum = 0;
1359 newcsum = 0;
1360 for (i=0; size>=4; size -= 4 )
1361 newcsum += le32_to_cpu(*isuper++);
1362
1363 if (size == 2)
1c05b4bc 1364 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1365
1366 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1367 sb->sb_csum = disk_csum;
1368 return cpu_to_le32(csum);
1369}
1370
1371static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1372{
1373 struct mdp_superblock_1 *sb;
1374 int ret;
0f420358 1375 sector_t sb_start;
1da177e4 1376 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1377 int bmask;
1da177e4
LT		 1378
1379 /*
0f420358 1380 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1381 * It is always aligned to a 4K boundary and
1382 * depeding on minor_version, it can be:
1383 * 0: At least 8K, but less than 12K, from end of device
1384 * 1: At start of device
1385 * 2: 4K from start of device.
1386 */
1387 switch(minor_version) {
1388 case 0:
0f420358
AN		 1389 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1390 sb_start -= 8*2;
1391 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1392 break;
1393 case 1:
0f420358 1394 sb_start = 0;
1da177e4
LT		 1395 break;
1396 case 2:
0f420358 1397 sb_start = 8;
1da177e4
LT		 1398 break;
1399 default:
1400 return -EINVAL;
1401 }
0f420358 1402 rdev->sb_start = sb_start;
1da177e4 1403
0002b271
N		 1404 /* superblock is rarely larger than 1K, but it can be larger,
1405 * and it is safe to read 4k, so we do that
1406 */
1407 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1408 if (ret) return ret;
1409
1410
1411 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1412
1413 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1414 sb->major_version != cpu_to_le32(1) ||
1415 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1416 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1417 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1418 return -EINVAL;
1419
1420 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1421 printk("md: invalid superblock checksum on %s\n",
1422 bdevname(rdev->bdev,b));
1423 return -EINVAL;
1424 }
1425 if (le64_to_cpu(sb->data_size) < 10) {
1426 printk("md: data_size too small on %s\n",
1427 bdevname(rdev->bdev,b));
1428 return -EINVAL;
1429 }
e11e93fa 1430
1da177e4
LT		 1431 rdev->preferred_minor = 0xffff;
1432 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1433 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1434
0002b271 1435 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1436 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1437 if (rdev->sb_size & bmask)
a1801f85
N		 1438 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1439
1440 if (minor_version
0f420358 1441 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1442 return -EINVAL;
0002b271 1443
31b65a0d
N		 1444 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1445 rdev->desc_nr = -1;
1446 else
1447 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1448
9a7b2b0f 1449 if (!refdev) {
8ed75463 1450 ret = 1;
9a7b2b0f 1451 } else {
1da177e4
LT		 1452 __u64 ev1, ev2;
1453 struct mdp_superblock_1 *refsb =
1454 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1455
1456 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1457 sb->level != refsb->level ||
1458 sb->layout != refsb->layout ||
1459 sb->chunksize != refsb->chunksize) {
1460 printk(KERN_WARNING "md: %s has strangely different"
1461 " superblock to %s\n",
1462 bdevname(rdev->bdev,b),
1463 bdevname(refdev->bdev,b2));
1464 return -EINVAL;
1465 }
1466 ev1 = le64_to_cpu(sb->events);
1467 ev2 = le64_to_cpu(refsb->events);
1468
1469 if (ev1 > ev2)
8ed75463
N		 1470 ret = 1;
1471 else
1472 ret = 0;
1da177e4 1473 }
a1801f85 1474 if (minor_version)
dd8ac336
AN		 1475 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1476 le64_to_cpu(sb->data_offset);
1da177e4 1477 else
dd8ac336
AN		 1478 rdev->sectors = rdev->sb_start;
1479 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1480 return -EINVAL;
dd8ac336 1481 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1482 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1483 return -EINVAL;
8ed75463 1484 return ret;
1da177e4
LT		 1485}
1486
1487static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1488{
1489 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1490 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1491
41158c7e 1492 rdev->raid_disk = -1;
c5d79adb
N		 1493 clear_bit(Faulty, &rdev->flags);
1494 clear_bit(In_sync, &rdev->flags);
1495 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1496
1da177e4
LT		 1497 if (mddev->raid_disks == 0) {
1498 mddev->major_version = 1;
1499 mddev->patch_version = 0;
e691063a 1500 mddev->external = 0;
9d8f0363 1501 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1502 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1503 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1504 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1505 mddev->clevel[0] = 0;
1da177e4
LT		 1506 mddev->layout = le32_to_cpu(sb->layout);
1507 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1508 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1509 mddev->events = ev1;
c3d9714e
N		 1510 mddev->bitmap_info.offset = 0;
1511 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1512
1513 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1514 memcpy(mddev->uuid, sb->set_uuid, 16);
1515
1516 mddev->max_disks = (4096-256)/2;
a654b9d8 1517
71c0805c 1518 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1519 mddev->bitmap_info.file == NULL )
1520 mddev->bitmap_info.offset =
1521 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1522
f6705578
N		 1523 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1524 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1525 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1526 mddev->new_level = le32_to_cpu(sb->new_level);
1527 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1528 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1529 } else {
1530 mddev->reshape_position = MaxSector;
1531 mddev->delta_disks = 0;
1532 mddev->new_level = mddev->level;
1533 mddev->new_layout = mddev->layout;
664e7c41 1534 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1535 }
1536
41158c7e 1537 } else if (mddev->pers == NULL) {
be6800a7
N		 1538 /* Insist of good event counter while assembling, except for
1539 * spares (which don't need an event count) */
1da177e4 1540 ++ev1;
be6800a7
N		 1541 if (rdev->desc_nr >= 0 &&
1542 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1543 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1544 if (ev1 < mddev->events)
1545 return -EINVAL;
41158c7e
N		 1546 } else if (mddev->bitmap) {
1547 /* If adding to array with a bitmap, then we can accept an
1548 * older device, but not too old.
1549 */
41158c7e
N		 1550 if (ev1 < mddev->bitmap->events_cleared)
1551 return 0;
07d84d10
N		 1552 } else {
1553 if (ev1 < mddev->events)
1554 /* just a hot-add of a new device, leave raid_disk at -1 */
1555 return 0;
1556 }
1da177e4
LT		 1557 if (mddev->level != LEVEL_MULTIPATH) {
1558 int role;
3673f305
N		 1559 if (rdev->desc_nr < 0 ||
1560 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1561 role = 0xffff;
1562 rdev->desc_nr = -1;
1563 } else
1564 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1565 switch(role) {
1566 case 0xffff: /* spare */
1da177e4
LT		 1567 break;
1568 case 0xfffe: /* faulty */
b2d444d7 1569 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1570 break;
1571 default:
5fd6c1dc
N		 1572 if ((le32_to_cpu(sb->feature_map) &
1573 MD_FEATURE_RECOVERY_OFFSET))
1574 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1575 else
1576 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1577 rdev->raid_disk = role;
1578 break;
1579 }
8ddf9efe
N		 1580 if (sb->devflags & WriteMostly1)
1581 set_bit(WriteMostly, &rdev->flags);
41158c7e 1582 } else /* MULTIPATH are always insync */
b2d444d7 1583 set_bit(In_sync, &rdev->flags);
41158c7e 1584
1da177e4
LT		 1585 return 0;
1586}
1587
1588static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1589{
1590 struct mdp_superblock_1 *sb;
1da177e4
LT		 1591 mdk_rdev_t *rdev2;
1592 int max_dev, i;
1593 /* make rdev->sb match mddev and rdev data. */
1594
1595 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1596
1597 sb->feature_map = 0;
1598 sb->pad0 = 0;
5fd6c1dc 1599 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1600 memset(sb->pad1, 0, sizeof(sb->pad1));
1601 memset(sb->pad2, 0, sizeof(sb->pad2));
1602 memset(sb->pad3, 0, sizeof(sb->pad3));
1603
1604 sb->utime = cpu_to_le64((__u64)mddev->utime);
1605 sb->events = cpu_to_le64(mddev->events);
1606 if (mddev->in_sync)
1607 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1608 else
1609 sb->resync_offset = cpu_to_le64(0);
1610
1c05b4bc 1611 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1612
f0ca340c 1613 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1614 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1615 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1616 sb->level = cpu_to_le32(mddev->level);
1617 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1618
c3d9714e
N		 1619 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1620 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1621 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1622 }
5fd6c1dc
N		 1623
1624 if (rdev->raid_disk >= 0 &&
97e4f42d 1625 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1626 sb->feature_map |=
1627 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1628 sb->recovery_offset =
1629 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1630 }
1631
f6705578
N		 1632 if (mddev->reshape_position != MaxSector) {
1633 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1634 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1635 sb->new_layout = cpu_to_le32(mddev->new_layout);
1636 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1637 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1638 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1639 }
a654b9d8 1640
1da177e4 1641 max_dev = 0;
159ec1fc 1642 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1643 if (rdev2->desc_nr+1 > max_dev)
1644 max_dev = rdev2->desc_nr+1;
a778b73f 1645
70471daf
N		 1646 if (max_dev > le32_to_cpu(sb->max_dev)) {
1647 int bmask;
a778b73f 1648 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1649 rdev->sb_size = max_dev * 2 + 256;
1650 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1651 if (rdev->sb_size & bmask)
1652 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1653 } else
1654 max_dev = le32_to_cpu(sb->max_dev);
1655
1da177e4
LT		 1656 for (i=0; i<max_dev;i++)
1657 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1658
159ec1fc 1659 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1660 i = rdev2->desc_nr;
b2d444d7 1661 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1662 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1663 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1664 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1665 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1666 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1667 else
1668 sb->dev_roles[i] = cpu_to_le16(0xffff);
1669 }
1670
1da177e4
LT		 1671 sb->sb_csum = calc_sb_1_csum(sb);
1672}
1673
0cd17fec 1674static unsigned long long
15f4a5fd 1675super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec
CW		 1676{
1677 struct mdp_superblock_1 *sb;
15f4a5fd 1678 sector_t max_sectors;
58c0fed4 1679 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1680 return 0; /* component must fit device */
0f420358 1681 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1682 /* minor versions 1 and 2; superblock before data */
15f4a5fd
AN		 1683 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1684 max_sectors -= rdev->data_offset;
1685 if (!num_sectors || num_sectors > max_sectors)
1686 num_sectors = max_sectors;
c3d9714e 1687 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1688 /* minor version 0 with bitmap we can't move */
1689 return 0;
1690 } else {
1691 /* minor version 0; superblock after data */
0f420358
AN		 1692 sector_t sb_start;
1693 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1694 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1695 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1696 if (!num_sectors || num_sectors > max_sectors)
1697 num_sectors = max_sectors;
0f420358 1698 rdev->sb_start = sb_start;
0cd17fec
CW		 1699 }
1700 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
15f4a5fd 1701 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1702 sb->super_offset = rdev->sb_start;
0cd17fec 1703 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1704 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1705 rdev->sb_page);
1706 md_super_wait(rdev->mddev);
15f4a5fd 1707 return num_sectors / 2; /* kB for sysfs */
0cd17fec 1708}
1da177e4 1709
75c96f85 1710static struct super_type super_types[] = {
1da177e4
LT		 1711 [0] = {
1712 .name = "0.90.0",
1713 .owner = THIS_MODULE,
0cd17fec
CW		 1714 .load_super = super_90_load,
1715 .validate_super = super_90_validate,
1716 .sync_super = super_90_sync,
1717 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1718 },
1719 [1] = {
1720 .name = "md-1",
1721 .owner = THIS_MODULE,
0cd17fec
CW		 1722 .load_super = super_1_load,
1723 .validate_super = super_1_validate,
1724 .sync_super = super_1_sync,
1725 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1726 },
1727};
1da177e4
LT		 1728
1729static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1730{
7dd5e7c3 1731 mdk_rdev_t *rdev, *rdev2;
1da177e4 1732
4b80991c
N		 1733 rcu_read_lock();
1734 rdev_for_each_rcu(rdev, mddev1)
1735 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1736 if (rdev->bdev->bd_contains ==
4b80991c
N		 1737 rdev2->bdev->bd_contains) {
1738 rcu_read_unlock();
7dd5e7c3 1739 return 1;
4b80991c
N		 1740 }
1741 rcu_read_unlock();
1da177e4
LT		 1742 return 0;
1743}
1744
1745static LIST_HEAD(pending_raid_disks);
1746
ac5e7113
AN		 1747/*
1748 * Try to register data integrity profile for an mddev
1749 *
1750 * This is called when an array is started and after a disk has been kicked
1751 * from the array. It only succeeds if all working and active component devices
1752 * are integrity capable with matching profiles.
1753 */
1754int md_integrity_register(mddev_t *mddev)
1755{
1756 mdk_rdev_t *rdev, *reference = NULL;
1757
1758 if (list_empty(&mddev->disks))
1759 return 0; /* nothing to do */
1760 if (blk_get_integrity(mddev->gendisk))
1761 return 0; /* already registered */
1762 list_for_each_entry(rdev, &mddev->disks, same_set) {
1763 /* skip spares and non-functional disks */
1764 if (test_bit(Faulty, &rdev->flags))
1765 continue;
1766 if (rdev->raid_disk < 0)
1767 continue;
1768 /*
1769 * If at least one rdev is not integrity capable, we can not
1770 * enable data integrity for the md device.
1771 */
1772 if (!bdev_get_integrity(rdev->bdev))
1773 return -EINVAL;
1774 if (!reference) {
1775 /* Use the first rdev as the reference */
1776 reference = rdev;
1777 continue;
1778 }
1779 /* does this rdev's profile match the reference profile? */
1780 if (blk_integrity_compare(reference->bdev->bd_disk,
1781 rdev->bdev->bd_disk) < 0)
1782 return -EINVAL;
1783 }
1784 /*
1785 * All component devices are integrity capable and have matching
1786 * profiles, register the common profile for the md device.
1787 */
1788 if (blk_integrity_register(mddev->gendisk,
1789 bdev_get_integrity(reference->bdev)) != 0) {
1790 printk(KERN_ERR "md: failed to register integrity for %s\n",
1791 mdname(mddev));
1792 return -EINVAL;
1793 }
1794 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1795 mdname(mddev));
1796 return 0;
1797}
1798EXPORT_SYMBOL(md_integrity_register);
1799
1800/* Disable data integrity if non-capable/non-matching disk is being added */
1801void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
3f9d99c1 1802{
3f9d99c1 1803 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1804 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1805
ac5e7113 1806 if (!bi_mddev) /* nothing to do */
3f9d99c1 1807 return;
ac5e7113 1808 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1809 return;
ac5e7113
AN		 1810 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1811 rdev->bdev->bd_disk) >= 0)
1812 return;
1813 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1814 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 1815}
ac5e7113 1816EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 1817
1da177e4
LT		 1818static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1819{
7dd5e7c3 1820 char b[BDEVNAME_SIZE];
f637b9f9 1821 struct kobject *ko;
1edf80d3 1822 char *s;
5e55e2f5 1823 int err;
1da177e4
LT		 1824
1825 if (rdev->mddev) {
1826 MD_BUG();
1827 return -EINVAL;
1828 }
11e2ede0
DW		 1829
1830 /* prevent duplicates */
1831 if (find_rdev(mddev, rdev->bdev->bd_dev))
1832 return -EEXIST;
1833
dd8ac336
AN		 1834 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1835 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1836 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 1837 if (mddev->pers) {
1838 /* Cannot change size, so fail
1839 * If mddev->level <= 0, then we don't care
1840 * about aligning sizes (e.g. linear)
1841 */
1842 if (mddev->level > 0)
1843 return -ENOSPC;
1844 } else
dd8ac336 1845 mddev->dev_sectors = rdev->sectors;
2bf071bf 1846 }
1da177e4
LT		 1847
1848 /* Verify rdev->desc_nr is unique.
1849 * If it is -1, assign a free number, else
1850 * check number is not in use
1851 */
1852 if (rdev->desc_nr < 0) {
1853 int choice = 0;
1854 if (mddev->pers) choice = mddev->raid_disks;
1855 while (find_rdev_nr(mddev, choice))
1856 choice++;
1857 rdev->desc_nr = choice;
1858 } else {
1859 if (find_rdev_nr(mddev, rdev->desc_nr))
1860 return -EBUSY;
1861 }
de01dfad
N		 1862 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1863 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1864 mdname(mddev), mddev->max_disks);
1865 return -EBUSY;
1866 }
19133a42 1867 bdevname(rdev->bdev,b);
649316b2 1868 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1869 *s = '!';
649316b2 1870
1da177e4 1871 rdev->mddev = mddev;
19133a42 1872 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1873
b2d6db58 1874 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1875 goto fail;
86e6ffdd 1876
0762b8bd 1877 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 1878 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1879 /* failure here is OK */;
1880 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 1881
4b80991c 1882 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 1883 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 1884
1885 /* May as well allow recovery to be retried once */
1886 mddev->recovery_disabled = 0;
3f9d99c1 1887
1da177e4 1888 return 0;
5e55e2f5
N		 1889
1890 fail:
1891 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1892 b, mdname(mddev));
1893 return err;
1da177e4
LT		 1894}
1895
177a99b2 1896static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1897{
1898 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1899 kobject_del(&rdev->kobj);
177a99b2 1900 kobject_put(&rdev->kobj);
5792a285
N		 1901}
1902
1da177e4
LT		 1903static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1904{
1905 char b[BDEVNAME_SIZE];
1906 if (!rdev->mddev) {
1907 MD_BUG();
1908 return;
1909 }
4b80991c 1910 list_del_rcu(&rdev->same_set);
1da177e4
LT		 1911 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1912 rdev->mddev = NULL;
86e6ffdd 1913 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 1914 sysfs_put(rdev->sysfs_state);
1915 rdev->sysfs_state = NULL;
5792a285 1916 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 1917 * writing to 'remove' to "dev/state". We also need
1918 * to delay it due to rcu usage.
5792a285 1919 */
4b80991c 1920 synchronize_rcu();
177a99b2
N		 1921 INIT_WORK(&rdev->del_work, md_delayed_delete);
1922 kobject_get(&rdev->kobj);
e804ac78 1923 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 1924}
1925
1926/*
1927 * prevent the device from being mounted, repartitioned or
1928 * otherwise reused by a RAID array (or any other kernel
1929 * subsystem), by bd_claiming the device.
1930 */
c5d79adb 1931static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1932{
1933 int err = 0;
1934 struct block_device *bdev;
1935 char b[BDEVNAME_SIZE];
1936
e525fd89
TH		 1937 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1938 shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1939 if (IS_ERR(bdev)) {
1940 printk(KERN_ERR "md: could not open %s.\n",
1941 __bdevname(dev, b));
1942 return PTR_ERR(bdev);
1943 }
c5d79adb
N		 1944 if (!shared)
1945 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1946 rdev->bdev = bdev;
1947 return err;
1948}
1949
1950static void unlock_rdev(mdk_rdev_t *rdev)
1951{
1952 struct block_device *bdev = rdev->bdev;
1953 rdev->bdev = NULL;
1954 if (!bdev)
1955 MD_BUG();
e525fd89 1956 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 1957}
1958
1959void md_autodetect_dev(dev_t dev);
1960
1961static void export_rdev(mdk_rdev_t * rdev)
1962{
1963 char b[BDEVNAME_SIZE];
1964 printk(KERN_INFO "md: export_rdev(%s)\n",
1965 bdevname(rdev->bdev,b));
1966 if (rdev->mddev)
1967 MD_BUG();
1968 free_disk_sb(rdev);
1da177e4 1969#ifndef MODULE
d0fae18f
N		 1970 if (test_bit(AutoDetected, &rdev->flags))
1971 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1972#endif
1973 unlock_rdev(rdev);
86e6ffdd 1974 kobject_put(&rdev->kobj);
1da177e4
LT		 1975}
1976
1977static void kick_rdev_from_array(mdk_rdev_t * rdev)
1978{
1979 unbind_rdev_from_array(rdev);
1980 export_rdev(rdev);
1981}
1982
1983static void export_array(mddev_t *mddev)
1984{
159ec1fc 1985 mdk_rdev_t *rdev, *tmp;
1da177e4 1986
d089c6af 1987 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1988 if (!rdev->mddev) {
1989 MD_BUG();
1990 continue;
1991 }
1992 kick_rdev_from_array(rdev);
1993 }
1994 if (!list_empty(&mddev->disks))
1995 MD_BUG();
1996 mddev->raid_disks = 0;
1997 mddev->major_version = 0;
1998}
1999
2000static void print_desc(mdp_disk_t *desc)
2001{
2002 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2003 desc->major,desc->minor,desc->raid_disk,desc->state);
2004}
2005
cd2ac932 2006static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2007{
2008 int i;
2009
2010 printk(KERN_INFO
2011 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2012 sb->major_version, sb->minor_version, sb->patch_version,
2013 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2014 sb->ctime);
2015 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2016 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2017 sb->md_minor, sb->layout, sb->chunk_size);
2018 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2019 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2020 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2021 sb->failed_disks, sb->spare_disks,
2022 sb->sb_csum, (unsigned long)sb->events_lo);
2023
2024 printk(KERN_INFO);
2025 for (i = 0; i < MD_SB_DISKS; i++) {
2026 mdp_disk_t *desc;
2027
2028 desc = sb->disks + i;
2029 if (desc->number || desc->major || desc->minor ||
2030 desc->raid_disk || (desc->state && (desc->state != 4))) {
2031 printk(" D %2d: ", i);
2032 print_desc(desc);
2033 }
2034 }
2035 printk(KERN_INFO "md: THIS: ");
2036 print_desc(&sb->this_disk);
cd2ac932 2037}
1da177e4 2038
cd2ac932
CR		 2039static void print_sb_1(struct mdp_superblock_1 *sb)
2040{
2041 __u8 *uuid;
2042
2043 uuid = sb->set_uuid;
ad361c98 2044 printk(KERN_INFO
7b75c2f8 2045 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2046 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2047 le32_to_cpu(sb->major_version),
2048 le32_to_cpu(sb->feature_map),
7b75c2f8 2049 uuid,
cd2ac932
CR		 2050 sb->set_name,
2051 (unsigned long long)le64_to_cpu(sb->ctime)
2052 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2053
2054 uuid = sb->device_uuid;
ad361c98
JP		 2055 printk(KERN_INFO
2056 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2057 " RO:%llu\n"
7b75c2f8 2058 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2059 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2060 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2061 le32_to_cpu(sb->level),
2062 (unsigned long long)le64_to_cpu(sb->size),
2063 le32_to_cpu(sb->raid_disks),
2064 le32_to_cpu(sb->layout),
2065 le32_to_cpu(sb->chunksize),
2066 (unsigned long long)le64_to_cpu(sb->data_offset),
2067 (unsigned long long)le64_to_cpu(sb->data_size),
2068 (unsigned long long)le64_to_cpu(sb->super_offset),
2069 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2070 le32_to_cpu(sb->dev_number),
7b75c2f8 2071 uuid,
cd2ac932
CR		 2072 sb->devflags,
2073 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2074 (unsigned long long)le64_to_cpu(sb->events),
2075 (unsigned long long)le64_to_cpu(sb->resync_offset),
2076 le32_to_cpu(sb->sb_csum),
2077 le32_to_cpu(sb->max_dev)
2078 );
1da177e4
LT		 2079}
2080
cd2ac932 2081static void print_rdev(mdk_rdev_t *rdev, int major_version)
1da177e4
LT		 2082{
2083 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2084 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2085 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2086 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2087 rdev->desc_nr);
1da177e4 2088 if (rdev->sb_loaded) {
cd2ac932
CR		 2089 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2090 switch (major_version) {
2091 case 0:
2092 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2093 break;
2094 case 1:
2095 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2096 break;
2097 }
1da177e4
LT		 2098 } else
2099 printk(KERN_INFO "md: no rdev superblock!\n");
2100}
2101
5e56341d 2102static void md_print_devices(void)
1da177e4 2103{
159ec1fc 2104 struct list_head *tmp;
1da177e4
LT		 2105 mdk_rdev_t *rdev;
2106 mddev_t *mddev;
2107 char b[BDEVNAME_SIZE];
2108
2109 printk("\n");
2110 printk("md: **********************************\n");
2111 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2112 printk("md: **********************************\n");
29ac4aa3 2113 for_each_mddev(mddev, tmp) {
1da177e4 2114
32a7627c
N		 2115 if (mddev->bitmap)
2116 bitmap_print_sb(mddev->bitmap);
2117 else
2118 printk("%s: ", mdname(mddev));
159ec1fc 2119 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2120 printk("<%s>", bdevname(rdev->bdev,b));
2121 printk("\n");
2122
159ec1fc 2123 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2124 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2125 }
2126 printk("md: **********************************\n");
2127 printk("\n");
2128}
2129
2130
42543769 2131static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 2132{
42543769
N		 2133 /* Update each superblock (in-memory image), but
2134 * if we are allowed to, skip spares which already
2135 * have the right event counter, or have one earlier
2136 * (which would mean they aren't being marked as dirty
2137 * with the rest of the array)
2138 */
1da177e4 2139 mdk_rdev_t *rdev;
159ec1fc 2140 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2141 if (rdev->sb_events == mddev->events ||
2142 (nospares &&
2143 rdev->raid_disk < 0 &&
42543769
N		 2144 rdev->sb_events+1 == mddev->events)) {
2145 /* Don't update this superblock */
2146 rdev->sb_loaded = 2;
2147 } else {
2148 super_types[mddev->major_version].
2149 sync_super(mddev, rdev);
2150 rdev->sb_loaded = 1;
2151 }
1da177e4
LT		 2152 }
2153}
2154
850b2b42 2155static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 2156{
1da177e4 2157 mdk_rdev_t *rdev;
06d91a5f 2158 int sync_req;
42543769 2159 int nospares = 0;
1da177e4 2160
1da177e4 2161repeat:
3a3a5ddb
N		 2162 /* First make sure individual recovery_offsets are correct */
2163 list_for_each_entry(rdev, &mddev->disks, same_set) {
2164 if (rdev->raid_disk >= 0 &&
2165 mddev->delta_disks >= 0 &&
2166 !test_bit(In_sync, &rdev->flags) &&
2167 mddev->curr_resync_completed > rdev->recovery_offset)
2168 rdev->recovery_offset = mddev->curr_resync_completed;
2169
2170 }
bd52b746 2171 if (!mddev->persistent) {
070dc6dd 2172 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2173 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
d97a41dc
N		 2174 if (!mddev->external)
2175 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3a3a5ddb
N		 2176 wake_up(&mddev->sb_wait);
2177 return;
2178 }
2179
a9701a30 2180 spin_lock_irq(&mddev->write_lock);
84692195 2181
3a3a5ddb
N		 2182 mddev->utime = get_seconds();
2183
850b2b42
N		 2184 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2185 force_change = 1;
2186 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2187 /* just a clean<-> dirty transition, possibly leave spares alone,
2188 * though if events isn't the right even/odd, we will have to do
2189 * spares after all
2190 */
2191 nospares = 1;
2192 if (force_change)
2193 nospares = 0;
2194 if (mddev->degraded)
84692195
N		 2195 /* If the array is degraded, then skipping spares is both
2196 * dangerous and fairly pointless.
2197 * Dangerous because a device that was removed from the array
2198 * might have a event_count that still looks up-to-date,
2199 * so it can be re-added without a resync.
2200 * Pointless because if there are any spares to skip,
2201 * then a recovery will happen and soon that array won't
2202 * be degraded any more and the spare can go back to sleep then.
2203 */
850b2b42 2204 nospares = 0;
84692195 2205
06d91a5f 2206 sync_req = mddev->in_sync;
42543769
N		 2207
2208 /* If this is just a dirty<->clean transition, and the array is clean
2209 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2210 if (nospares
42543769 2211 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2212 && mddev->can_decrease_events
2213 && mddev->events != 1) {
42543769 2214 mddev->events--;
a8707c08
N		 2215 mddev->can_decrease_events = 0;
2216 } else {
42543769
N		 2217 /* otherwise we have to go forward and ... */
2218 mddev->events ++;
a8707c08 2219 mddev->can_decrease_events = nospares;
42543769 2220 }
1da177e4
LT		 2221
2222 if (!mddev->events) {
2223 /*
2224 * oops, this 64-bit counter should never wrap.
2225 * Either we are in around ~1 trillion A.C., assuming
2226 * 1 reboot per second, or we have a bug:
2227 */
2228 MD_BUG();
2229 mddev->events --;
2230 }
e691063a 2231 sync_sbs(mddev, nospares);
a9701a30 2232 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 2233
2234 dprintk(KERN_INFO
2235 "md: updating %s RAID superblock on device (in sync %d)\n",
2236 mdname(mddev),mddev->in_sync);
2237
4ad13663 2238 bitmap_update_sb(mddev->bitmap);
159ec1fc 2239 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 2240 char b[BDEVNAME_SIZE];
2241 dprintk(KERN_INFO "md: ");
42543769
N		 2242 if (rdev->sb_loaded != 1)
2243 continue; /* no noise on spare devices */
b2d444d7 2244 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2245 dprintk("(skipping faulty ");
2246
2247 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 2248 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2249 md_super_write(mddev,rdev,
0f420358 2250 rdev->sb_start, rdev->sb_size,
7bfa19f2
N		 2251 rdev->sb_page);
2252 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2253 bdevname(rdev->bdev,b),
0f420358 2254 (unsigned long long)rdev->sb_start);
42543769 2255 rdev->sb_events = mddev->events;
7bfa19f2 2256
1da177e4
LT		 2257 } else
2258 dprintk(")\n");
7bfa19f2 2259 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2260 /* only need to write one superblock... */
2261 break;
2262 }
a9701a30 2263 md_super_wait(mddev);
850b2b42 2264 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2265
a9701a30 2266 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2267 if (mddev->in_sync != sync_req ||
2268 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2269 /* have to write it out again */
a9701a30 2270 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2271 goto repeat;
2272 }
850b2b42 2273 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2274 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2275 wake_up(&mddev->sb_wait);
acb180b0
N		 2276 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2277 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2278
1da177e4
LT		 2279}
2280
7f6ce769 2281/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2282 * We want to accept with case. For this we use cmd_match.
2283 */
2284static int cmd_match(const char *cmd, const char *str)
2285{
2286 /* See if cmd, written into a sysfs file, matches
2287 * str. They must either be the same, or cmd can
2288 * have a trailing newline
2289 */
2290 while (*cmd && *str && *cmd == *str) {
2291 cmd++;
2292 str++;
2293 }
2294 if (*cmd == '\n')
2295 cmd++;
2296 if (*str || *cmd)
2297 return 0;
2298 return 1;
2299}
2300
86e6ffdd
N		 2301struct rdev_sysfs_entry {
2302 struct attribute attr;
2303 ssize_t (*show)(mdk_rdev_t *, char *);
2304 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2305};
2306
2307static ssize_t
96de1e66 2308state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 2309{
2310 char *sep = "";
20a49ff6 2311 size_t len = 0;
86e6ffdd 2312
b2d444d7 2313 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 2314 len+= sprintf(page+len, "%sfaulty",sep);
2315 sep = ",";
2316 }
b2d444d7 2317 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2318 len += sprintf(page+len, "%sin_sync",sep);
2319 sep = ",";
2320 }
f655675b
N		 2321 if (test_bit(WriteMostly, &rdev->flags)) {
2322 len += sprintf(page+len, "%swrite_mostly",sep);
2323 sep = ",";
2324 }
6bfe0b49
DW		 2325 if (test_bit(Blocked, &rdev->flags)) {
2326 len += sprintf(page+len, "%sblocked", sep);
2327 sep = ",";
2328 }
b2d444d7
N		 2329 if (!test_bit(Faulty, &rdev->flags) &&
2330 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2331 len += sprintf(page+len, "%sspare", sep);
2332 sep = ",";
2333 }
2334 return len+sprintf(page+len, "\n");
2335}
2336
45dc2de1
N		 2337static ssize_t
2338state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2339{
2340 /* can write
2341 * faulty - simulates and error
2342 * remove - disconnects the device
f655675b
N		 2343 * writemostly - sets write_mostly
2344 * -writemostly - clears write_mostly
6bfe0b49
DW		 2345 * blocked - sets the Blocked flag
2346 * -blocked - clears the Blocked flag
6d56e278 2347 * insync - sets Insync providing device isn't active
45dc2de1
N		 2348 */
2349 int err = -EINVAL;
2350 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2351 md_error(rdev->mddev, rdev);
2352 err = 0;
2353 } else if (cmd_match(buf, "remove")) {
2354 if (rdev->raid_disk >= 0)
2355 err = -EBUSY;
2356 else {
2357 mddev_t *mddev = rdev->mddev;
2358 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2359 if (mddev->pers)
2360 md_update_sb(mddev, 1);
45dc2de1
N		 2361 md_new_event(mddev);
2362 err = 0;
2363 }
f655675b
N		 2364 } else if (cmd_match(buf, "writemostly")) {
2365 set_bit(WriteMostly, &rdev->flags);
2366 err = 0;
2367 } else if (cmd_match(buf, "-writemostly")) {
2368 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2369 err = 0;
2370 } else if (cmd_match(buf, "blocked")) {
2371 set_bit(Blocked, &rdev->flags);
2372 err = 0;
2373 } else if (cmd_match(buf, "-blocked")) {
2374 clear_bit(Blocked, &rdev->flags);
2375 wake_up(&rdev->blocked_wait);
2376 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2377 md_wakeup_thread(rdev->mddev->thread);
2378
6d56e278
N		 2379 err = 0;
2380 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2381 set_bit(In_sync, &rdev->flags);
f655675b 2382 err = 0;
45dc2de1 2383 }
00bcb4ac
N		 2384 if (!err)
2385 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2386 return err ? err : len;
2387}
80ca3a44
N		 2388static struct rdev_sysfs_entry rdev_state =
2389__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2390
4dbcdc75
N		 2391static ssize_t
2392errors_show(mdk_rdev_t *rdev, char *page)
2393{
2394 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2395}
2396
2397static ssize_t
2398errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2399{
2400 char *e;
2401 unsigned long n = simple_strtoul(buf, &e, 10);
2402 if (*buf && (*e == 0 || *e == '\n')) {
2403 atomic_set(&rdev->corrected_errors, n);
2404 return len;
2405 }
2406 return -EINVAL;
2407}
2408static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2409__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2410
014236d2
N		 2411static ssize_t
2412slot_show(mdk_rdev_t *rdev, char *page)
2413{
2414 if (rdev->raid_disk < 0)
2415 return sprintf(page, "none\n");
2416 else
2417 return sprintf(page, "%d\n", rdev->raid_disk);
2418}
2419
2420static ssize_t
2421slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2422{
2423 char *e;
c303da6d
N		 2424 int err;
2425 char nm[20];
014236d2
N		 2426 int slot = simple_strtoul(buf, &e, 10);
2427 if (strncmp(buf, "none", 4)==0)
2428 slot = -1;
2429 else if (e==buf || (*e && *e!= '\n'))
2430 return -EINVAL;
6c2fce2e 2431 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2432 /* Setting 'slot' on an active array requires also
2433 * updating the 'rd%d' link, and communicating
2434 * with the personality with ->hot_*_disk.
2435 * For now we only support removing
2436 * failed/spare devices. This normally happens automatically,
2437 * but not when the metadata is externally managed.
2438 */
c303da6d
N		 2439 if (rdev->raid_disk == -1)
2440 return -EEXIST;
2441 /* personality does all needed checks */
2442 if (rdev->mddev->pers->hot_add_disk == NULL)
2443 return -EINVAL;
2444 err = rdev->mddev->pers->
2445 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2446 if (err)
2447 return err;
2448 sprintf(nm, "rd%d", rdev->raid_disk);
2449 sysfs_remove_link(&rdev->mddev->kobj, nm);
b7103107 2450 rdev->raid_disk = -1;
c303da6d
N		 2451 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2452 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e
NB		 2453 } else if (rdev->mddev->pers) {
2454 mdk_rdev_t *rdev2;
6c2fce2e 2455 /* Activating a spare .. or possibly reactivating
6d56e278 2456 * if we ever get bitmaps working here.
6c2fce2e
NB		 2457 */
2458
2459 if (rdev->raid_disk != -1)
2460 return -EBUSY;
2461
2462 if (rdev->mddev->pers->hot_add_disk == NULL)
2463 return -EINVAL;
2464
159ec1fc 2465 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2466 if (rdev2->raid_disk == slot)
2467 return -EEXIST;
2468
2469 rdev->raid_disk = slot;
2470 if (test_bit(In_sync, &rdev->flags))
2471 rdev->saved_raid_disk = slot;
2472 else
2473 rdev->saved_raid_disk = -1;
2474 err = rdev->mddev->pers->
2475 hot_add_disk(rdev->mddev, rdev);
199050ea 2476 if (err) {
6c2fce2e 2477 rdev->raid_disk = -1;
6c2fce2e 2478 return err;
52664732 2479 } else
00bcb4ac 2480 sysfs_notify_dirent_safe(rdev->sysfs_state);
6c2fce2e
NB		 2481 sprintf(nm, "rd%d", rdev->raid_disk);
2482 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
00bcb4ac 2483 /* failure here is OK */;
6c2fce2e 2484 /* don't wakeup anyone, leave that to userspace. */
c303da6d
N		 2485 } else {
2486 if (slot >= rdev->mddev->raid_disks)
2487 return -ENOSPC;
2488 rdev->raid_disk = slot;
2489 /* assume it is working */
c5d79adb
N		 2490 clear_bit(Faulty, &rdev->flags);
2491 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2492 set_bit(In_sync, &rdev->flags);
00bcb4ac 2493 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2494 }
014236d2
N		 2495 return len;
2496}
2497
2498
2499static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2500__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2501
93c8cad0
N		 2502static ssize_t
2503offset_show(mdk_rdev_t *rdev, char *page)
2504{
6961ece4 2505 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2506}
2507
2508static ssize_t
2509offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2510{
2511 char *e;
2512 unsigned long long offset = simple_strtoull(buf, &e, 10);
2513 if (e==buf || (*e && *e != '\n'))
2514 return -EINVAL;
8ed0a521 2515 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2516 return -EBUSY;
dd8ac336 2517 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2518 /* Must set offset before size, so overlap checks
2519 * can be sane */
2520 return -EBUSY;
93c8cad0
N		 2521 rdev->data_offset = offset;
2522 return len;
2523}
2524
2525static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2526__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2527
83303b61
N		 2528static ssize_t
2529rdev_size_show(mdk_rdev_t *rdev, char *page)
2530{
dd8ac336 2531 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2532}
2533
c5d79adb
N		 2534static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2535{
2536 /* check if two start/length pairs overlap */
2537 if (s1+l1 <= s2)
2538 return 0;
2539 if (s2+l2 <= s1)
2540 return 0;
2541 return 1;
2542}
2543
b522adcd
DW		 2544static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2545{
2546 unsigned long long blocks;
2547 sector_t new;
2548
2549 if (strict_strtoull(buf, 10, &blocks) < 0)
2550 return -EINVAL;
2551
2552 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2553 return -EINVAL; /* sector conversion overflow */
2554
2555 new = blocks * 2;
2556 if (new != blocks * 2)
2557 return -EINVAL; /* unsigned long long to sector_t overflow */
2558
2559 *sectors = new;
2560 return 0;
2561}
2562
83303b61
N		 2563static ssize_t
2564rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2565{
27c529bb 2566 mddev_t *my_mddev = rdev->mddev;
dd8ac336 2567 sector_t oldsectors = rdev->sectors;
b522adcd 2568 sector_t sectors;
27c529bb 2569
b522adcd 2570 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2571 return -EINVAL;
0cd17fec 2572 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2573 if (my_mddev->persistent) {
dd8ac336
AN		 2574 sectors = super_types[my_mddev->major_version].
2575 rdev_size_change(rdev, sectors);
2576 if (!sectors)
0cd17fec 2577 return -EBUSY;
dd8ac336
AN		 2578 } else if (!sectors)
2579 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2580 rdev->data_offset;
0cd17fec 2581 }
dd8ac336 2582 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2583 return -EINVAL; /* component must fit device */
0cd17fec 2584
dd8ac336
AN		 2585 rdev->sectors = sectors;
2586 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2587 /* need to check that all other rdevs with the same ->bdev
2588 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2589 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2590 * we have to change it back, we will have the lock again.
2591 */
2592 mddev_t *mddev;
2593 int overlap = 0;
159ec1fc 2594 struct list_head *tmp;
c5d79adb 2595
27c529bb 2596 mddev_unlock(my_mddev);
29ac4aa3 2597 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2598 mdk_rdev_t *rdev2;
2599
2600 mddev_lock(mddev);
159ec1fc 2601 list_for_each_entry(rdev2, &mddev->disks, same_set)
c5d79adb
N		 2602 if (test_bit(AllReserved, &rdev2->flags) ||
2603 (rdev->bdev == rdev2->bdev &&
2604 rdev != rdev2 &&
dd8ac336 2605 overlaps(rdev->data_offset, rdev->sectors,
d07bd3bc 2606 rdev2->data_offset,
dd8ac336 2607 rdev2->sectors))) {
c5d79adb
N		 2608 overlap = 1;
2609 break;
2610 }
2611 mddev_unlock(mddev);
2612 if (overlap) {
2613 mddev_put(mddev);
2614 break;
2615 }
2616 }
27c529bb 2617 mddev_lock(my_mddev);
c5d79adb
N		 2618 if (overlap) {
2619 /* Someone else could have slipped in a size
2620 * change here, but doing so is just silly.
dd8ac336 2621 * We put oldsectors back because we *know* it is
c5d79adb
N		 2622 * safe, and trust userspace not to race with
2623 * itself
2624 */
dd8ac336 2625 rdev->sectors = oldsectors;
c5d79adb
N		 2626 return -EBUSY;
2627 }
2628 }
83303b61
N		 2629 return len;
2630}
2631
2632static struct rdev_sysfs_entry rdev_size =
80ca3a44 2633__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2634
06e3c817
DW		 2635
2636static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2637{
2638 unsigned long long recovery_start = rdev->recovery_offset;
2639
2640 if (test_bit(In_sync, &rdev->flags) ||
2641 recovery_start == MaxSector)
2642 return sprintf(page, "none\n");
2643
2644 return sprintf(page, "%llu\n", recovery_start);
2645}
2646
2647static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2648{
2649 unsigned long long recovery_start;
2650
2651 if (cmd_match(buf, "none"))
2652 recovery_start = MaxSector;
2653 else if (strict_strtoull(buf, 10, &recovery_start))
2654 return -EINVAL;
2655
2656 if (rdev->mddev->pers &&
2657 rdev->raid_disk >= 0)
2658 return -EBUSY;
2659
2660 rdev->recovery_offset = recovery_start;
2661 if (recovery_start == MaxSector)
2662 set_bit(In_sync, &rdev->flags);
2663 else
2664 clear_bit(In_sync, &rdev->flags);
2665 return len;
2666}
2667
2668static struct rdev_sysfs_entry rdev_recovery_start =
2669__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2670
86e6ffdd
N		 2671static struct attribute *rdev_default_attrs[] = {
2672 &rdev_state.attr,
4dbcdc75 2673 &rdev_errors.attr,
014236d2 2674 &rdev_slot.attr,
93c8cad0 2675 &rdev_offset.attr,
83303b61 2676 &rdev_size.attr,
06e3c817 2677 &rdev_recovery_start.attr,
86e6ffdd
N		 2678 NULL,
2679};
2680static ssize_t
2681rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2682{
2683 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2684 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2685 mddev_t *mddev = rdev->mddev;
2686 ssize_t rv;
86e6ffdd
N		 2687
2688 if (!entry->show)
2689 return -EIO;
27c529bb
N		 2690
2691 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2692 if (!rv) {
2693 if (rdev->mddev == NULL)
2694 rv = -EBUSY;
2695 else
2696 rv = entry->show(rdev, page);
2697 mddev_unlock(mddev);
2698 }
2699 return rv;
86e6ffdd
N		 2700}
2701
2702static ssize_t
2703rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2704 const char *page, size_t length)
2705{
2706 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2707 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2708 ssize_t rv;
2709 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2710
2711 if (!entry->store)
2712 return -EIO;
67463acb
N		 2713 if (!capable(CAP_SYS_ADMIN))
2714 return -EACCES;
27c529bb 2715 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2716 if (!rv) {
27c529bb
N		 2717 if (rdev->mddev == NULL)
2718 rv = -EBUSY;
2719 else
2720 rv = entry->store(rdev, page, length);
6a51830e 2721 mddev_unlock(mddev);
ca388059
N		 2722 }
2723 return rv;
86e6ffdd
N		 2724}
2725
2726static void rdev_free(struct kobject *ko)
2727{
2728 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2729 kfree(rdev);
2730}
52cf25d0 2731static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 2732 .show = rdev_attr_show,
2733 .store = rdev_attr_store,
2734};
2735static struct kobj_type rdev_ktype = {
2736 .release = rdev_free,
2737 .sysfs_ops = &rdev_sysfs_ops,
2738 .default_attrs = rdev_default_attrs,
2739};
2740
e8bb9a83
N		 2741void md_rdev_init(mdk_rdev_t *rdev)
2742{
2743 rdev->desc_nr = -1;
2744 rdev->saved_raid_disk = -1;
2745 rdev->raid_disk = -1;
2746 rdev->flags = 0;
2747 rdev->data_offset = 0;
2748 rdev->sb_events = 0;
2749 rdev->last_read_error.tv_sec = 0;
2750 rdev->last_read_error.tv_nsec = 0;
2751 atomic_set(&rdev->nr_pending, 0);
2752 atomic_set(&rdev->read_errors, 0);
2753 atomic_set(&rdev->corrected_errors, 0);
2754
2755 INIT_LIST_HEAD(&rdev->same_set);
2756 init_waitqueue_head(&rdev->blocked_wait);
2757}
2758EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 2759/*
2760 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2761 *
2762 * mark the device faulty if:
2763 *
2764 * - the device is nonexistent (zero size)
2765 * - the device has no valid superblock
2766 *
2767 * a faulty rdev _never_ has rdev->sb set.
2768 */
2769static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2770{
2771 char b[BDEVNAME_SIZE];
2772 int err;
2773 mdk_rdev_t *rdev;
2774 sector_t size;
2775
9ffae0cf 2776 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2777 if (!rdev) {
2778 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2779 return ERR_PTR(-ENOMEM);
2780 }
1da177e4 2781
e8bb9a83 2782 md_rdev_init(rdev);
1da177e4
LT		 2783 if ((err = alloc_disk_sb(rdev)))
2784 goto abort_free;
2785
c5d79adb 2786 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2787 if (err)
2788 goto abort_free;
2789
f9cb074b 2790 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2791
1da177e4
LT		 2792 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2793 if (!size) {
2794 printk(KERN_WARNING
2795 "md: %s has zero or unknown size, marking faulty!\n",
2796 bdevname(rdev->bdev,b));
2797 err = -EINVAL;
2798 goto abort_free;
2799 }
2800
2801 if (super_format >= 0) {
2802 err = super_types[super_format].
2803 load_super(rdev, NULL, super_minor);
2804 if (err == -EINVAL) {
df968c4e
N		 2805 printk(KERN_WARNING
2806 "md: %s does not have a valid v%d.%d "
2807 "superblock, not importing!\n",
2808 bdevname(rdev->bdev,b),
2809 super_format, super_minor);
1da177e4
LT		 2810 goto abort_free;
2811 }
2812 if (err < 0) {
2813 printk(KERN_WARNING
2814 "md: could not read %s's sb, not importing!\n",
2815 bdevname(rdev->bdev,b));
2816 goto abort_free;
2817 }
2818 }
6bfe0b49 2819
1da177e4
LT		 2820 return rdev;
2821
2822abort_free:
2823 if (rdev->sb_page) {
2824 if (rdev->bdev)
2825 unlock_rdev(rdev);
2826 free_disk_sb(rdev);
2827 }
2828 kfree(rdev);
2829 return ERR_PTR(err);
2830}
2831
2832/*
2833 * Check a full RAID array for plausibility
2834 */
2835
2836
a757e64c 2837static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2838{
2839 int i;
159ec1fc 2840 mdk_rdev_t *rdev, *freshest, *tmp;
1da177e4
LT		 2841 char b[BDEVNAME_SIZE];
2842
2843 freshest = NULL;
d089c6af 2844 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2845 switch (super_types[mddev->major_version].
2846 load_super(rdev, freshest, mddev->minor_version)) {
2847 case 1:
2848 freshest = rdev;
2849 break;
2850 case 0:
2851 break;
2852 default:
2853 printk( KERN_ERR \
2854 "md: fatal superblock inconsistency in %s"
2855 " -- removing from array\n",
2856 bdevname(rdev->bdev,b));
2857 kick_rdev_from_array(rdev);
2858 }
2859
2860
2861 super_types[mddev->major_version].
2862 validate_super(mddev, freshest);
2863
2864 i = 0;
d089c6af 2865 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 2866 if (mddev->max_disks &&
2867 (rdev->desc_nr >= mddev->max_disks ||
2868 i > mddev->max_disks)) {
de01dfad
N		 2869 printk(KERN_WARNING
2870 "md: %s: %s: only %d devices permitted\n",
2871 mdname(mddev), bdevname(rdev->bdev, b),
2872 mddev->max_disks);
2873 kick_rdev_from_array(rdev);
2874 continue;
2875 }
1da177e4
LT		 2876 if (rdev != freshest)
2877 if (super_types[mddev->major_version].
2878 validate_super(mddev, rdev)) {
2879 printk(KERN_WARNING "md: kicking non-fresh %s"
2880 " from array!\n",
2881 bdevname(rdev->bdev,b));
2882 kick_rdev_from_array(rdev);
2883 continue;
2884 }
2885 if (mddev->level == LEVEL_MULTIPATH) {
2886 rdev->desc_nr = i++;
2887 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2888 set_bit(In_sync, &rdev->flags);
5e5e3e78 2889 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 2890 rdev->raid_disk = -1;
2891 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2892 }
2893 }
1da177e4
LT		 2894}
2895
72e02075
N		 2896/* Read a fixed-point number.
2897 * Numbers in sysfs attributes should be in "standard" units where
2898 * possible, so time should be in seconds.
2899 * However we internally use a a much smaller unit such as
2900 * milliseconds or jiffies.
2901 * This function takes a decimal number with a possible fractional
2902 * component, and produces an integer which is the result of
2903 * multiplying that number by 10^'scale'.
2904 * all without any floating-point arithmetic.
2905 */
2906int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2907{
2908 unsigned long result = 0;
2909 long decimals = -1;
2910 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2911 if (*cp == '.')
2912 decimals = 0;
2913 else if (decimals < scale) {
2914 unsigned int value;
2915 value = *cp - '0';
2916 result = result * 10 + value;
2917 if (decimals >= 0)
2918 decimals++;
2919 }
2920 cp++;
2921 }
2922 if (*cp == '\n')
2923 cp++;
2924 if (*cp)
2925 return -EINVAL;
2926 if (decimals < 0)
2927 decimals = 0;
2928 while (decimals < scale) {
2929 result *= 10;
2930 decimals ++;
2931 }
2932 *res = result;
2933 return 0;
2934}
2935
2936
19052c0e
N		 2937static void md_safemode_timeout(unsigned long data);
2938
16f17b39
N		 2939static ssize_t
2940safe_delay_show(mddev_t *mddev, char *page)
2941{
2942 int msec = (mddev->safemode_delay*1000)/HZ;
2943 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2944}
2945static ssize_t
2946safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2947{
16f17b39 2948 unsigned long msec;
97ce0a7f 2949
72e02075 2950 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 2951 return -EINVAL;
16f17b39
N		 2952 if (msec == 0)
2953 mddev->safemode_delay = 0;
2954 else {
19052c0e 2955 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 2956 mddev->safemode_delay = (msec*HZ)/1000;
2957 if (mddev->safemode_delay == 0)
2958 mddev->safemode_delay = 1;
19052c0e
N		 2959 if (mddev->safemode_delay < old_delay)
2960 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 2961 }
2962 return len;
2963}
2964static struct md_sysfs_entry md_safe_delay =
80ca3a44 2965__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2966
eae1701f 2967static ssize_t
96de1e66 2968level_show(mddev_t *mddev, char *page)
eae1701f 2969{
2604b703 2970 struct mdk_personality *p = mddev->pers;
d9d166c2 2971 if (p)
eae1701f 2972 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2973 else if (mddev->clevel[0])
2974 return sprintf(page, "%s\n", mddev->clevel);
2975 else if (mddev->level != LEVEL_NONE)
2976 return sprintf(page, "%d\n", mddev->level);
2977 else
2978 return 0;
eae1701f
N		 2979}
2980
d9d166c2
N		 2981static ssize_t
2982level_store(mddev_t *mddev, const char *buf, size_t len)
2983{
f2859af6 2984 char clevel[16];
20a49ff6 2985 ssize_t rv = len;
245f46c2 2986 struct mdk_personality *pers;
f2859af6 2987 long level;
245f46c2 2988 void *priv;
3a981b03 2989 mdk_rdev_t *rdev;
245f46c2
N		 2990
2991 if (mddev->pers == NULL) {
2992 if (len == 0)
2993 return 0;
2994 if (len >= sizeof(mddev->clevel))
2995 return -ENOSPC;
2996 strncpy(mddev->clevel, buf, len);
2997 if (mddev->clevel[len-1] == '\n')
2998 len--;
2999 mddev->clevel[len] = 0;
3000 mddev->level = LEVEL_NONE;
3001 return rv;
3002 }
3003
3004 /* request to change the personality. Need to ensure:
3005 * - array is not engaged in resync/recovery/reshape
3006 * - old personality can be suspended
3007 * - new personality will access other array.
3008 */
3009
bb4f1e9d
N		 3010 if (mddev->sync_thread ||
3011 mddev->reshape_position != MaxSector ||
3012 mddev->sysfs_active)
d9d166c2 3013 return -EBUSY;
245f46c2
N		 3014
3015 if (!mddev->pers->quiesce) {
3016 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3017 mdname(mddev), mddev->pers->name);
3018 return -EINVAL;
3019 }
3020
3021 /* Now find the new personality */
f2859af6 3022 if (len == 0 || len >= sizeof(clevel))
245f46c2 3023 return -EINVAL;
f2859af6
DW		 3024 strncpy(clevel, buf, len);
3025 if (clevel[len-1] == '\n')
d9d166c2 3026 len--;
f2859af6
DW		 3027 clevel[len] = 0;
3028 if (strict_strtol(clevel, 10, &level))
3029 level = LEVEL_NONE;
245f46c2 3030
f2859af6
DW		 3031 if (request_module("md-%s", clevel) != 0)
3032 request_module("md-level-%s", clevel);
245f46c2 3033 spin_lock(&pers_lock);
f2859af6 3034 pers = find_pers(level, clevel);
245f46c2
N		 3035 if (!pers || !try_module_get(pers->owner)) {
3036 spin_unlock(&pers_lock);
f2859af6 3037 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3038 return -EINVAL;
3039 }
3040 spin_unlock(&pers_lock);
3041
3042 if (pers == mddev->pers) {
3043 /* Nothing to do! */
3044 module_put(pers->owner);
3045 return rv;
3046 }
3047 if (!pers->takeover) {
3048 module_put(pers->owner);
3049 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3050 mdname(mddev), clevel);
245f46c2
N		 3051 return -EINVAL;
3052 }
3053
e93f68a1
N		 3054 list_for_each_entry(rdev, &mddev->disks, same_set)
3055 rdev->new_raid_disk = rdev->raid_disk;
3056
245f46c2
N		 3057 /* ->takeover must set new_* and/or delta_disks
3058 * if it succeeds, and may set them when it fails.
3059 */
3060 priv = pers->takeover(mddev);
3061 if (IS_ERR(priv)) {
3062 mddev->new_level = mddev->level;
3063 mddev->new_layout = mddev->layout;
664e7c41 3064 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3065 mddev->raid_disks -= mddev->delta_disks;
3066 mddev->delta_disks = 0;
3067 module_put(pers->owner);
3068 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3069 mdname(mddev), clevel);
245f46c2
N		 3070 return PTR_ERR(priv);
3071 }
3072
3073 /* Looks like we have a winner */
3074 mddev_suspend(mddev);
3075 mddev->pers->stop(mddev);
a64c876f
N		 3076
3077 if (mddev->pers->sync_request == NULL &&
3078 pers->sync_request != NULL) {
3079 /* need to add the md_redundancy_group */
3080 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3081 printk(KERN_WARNING
3082 "md: cannot register extra attributes for %s\n",
3083 mdname(mddev));
19fdb9ee 3084 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3085 }
3086 if (mddev->pers->sync_request != NULL &&
3087 pers->sync_request == NULL) {
3088 /* need to remove the md_redundancy_group */
3089 if (mddev->to_remove == NULL)
3090 mddev->to_remove = &md_redundancy_group;
3091 }
3092
54071b38
TM		 3093 if (mddev->pers->sync_request == NULL &&
3094 mddev->external) {
3095 /* We are converting from a no-redundancy array
3096 * to a redundancy array and metadata is managed
3097 * externally so we need to be sure that writes
3098 * won't block due to a need to transition
3099 * clean->dirty
3100 * until external management is started.
3101 */
3102 mddev->in_sync = 0;
3103 mddev->safemode_delay = 0;
3104 mddev->safemode = 0;
3105 }
3106
e93f68a1
N		 3107 list_for_each_entry(rdev, &mddev->disks, same_set) {
3108 char nm[20];
3109 if (rdev->raid_disk < 0)
3110 continue;
3111 if (rdev->new_raid_disk > mddev->raid_disks)
3112 rdev->new_raid_disk = -1;
3113 if (rdev->new_raid_disk == rdev->raid_disk)
3114 continue;
3115 sprintf(nm, "rd%d", rdev->raid_disk);
3116 sysfs_remove_link(&mddev->kobj, nm);
3117 }
3118 list_for_each_entry(rdev, &mddev->disks, same_set) {
3119 if (rdev->raid_disk < 0)
3120 continue;
3121 if (rdev->new_raid_disk == rdev->raid_disk)
3122 continue;
3123 rdev->raid_disk = rdev->new_raid_disk;
3124 if (rdev->raid_disk < 0)
3a981b03 3125 clear_bit(In_sync, &rdev->flags);
e93f68a1
N		 3126 else {
3127 char nm[20];
3128 sprintf(nm, "rd%d", rdev->raid_disk);
3129 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3130 printk("md: cannot register %s for %s after level change\n",
3131 nm, mdname(mddev));
3a981b03 3132 }
e93f68a1
N		 3133 }
3134
3135 module_put(mddev->pers->owner);
245f46c2
N		 3136 mddev->pers = pers;
3137 mddev->private = priv;
3138 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3139 mddev->level = mddev->new_level;
3140 mddev->layout = mddev->new_layout;
664e7c41 3141 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3142 mddev->delta_disks = 0;
9af204cf
TM		 3143 if (mddev->pers->sync_request == NULL) {
3144 /* this is now an array without redundancy, so
3145 * it must always be in_sync
3146 */
3147 mddev->in_sync = 1;
3148 del_timer_sync(&mddev->safemode_timer);
3149 }
245f46c2
N		 3150 pers->run(mddev);
3151 mddev_resume(mddev);
3152 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3153 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3154 md_wakeup_thread(mddev->thread);
5cac7861 3155 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3156 md_new_event(mddev);
d9d166c2
N		 3157 return rv;
3158}
3159
3160static struct md_sysfs_entry md_level =
80ca3a44 3161__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3162
d4dbd025
N		 3163
3164static ssize_t
3165layout_show(mddev_t *mddev, char *page)
3166{
3167 /* just a number, not meaningful for all levels */
08a02ecd
N		 3168 if (mddev->reshape_position != MaxSector &&
3169 mddev->layout != mddev->new_layout)
3170 return sprintf(page, "%d (%d)\n",
3171 mddev->new_layout, mddev->layout);
d4dbd025
N		 3172 return sprintf(page, "%d\n", mddev->layout);
3173}
3174
3175static ssize_t
3176layout_store(mddev_t *mddev, const char *buf, size_t len)
3177{
3178 char *e;
3179 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3180
3181 if (!*buf || (*e && *e != '\n'))
3182 return -EINVAL;
3183
b3546035
N		 3184 if (mddev->pers) {
3185 int err;
50ac168a 3186 if (mddev->pers->check_reshape == NULL)
b3546035 3187 return -EBUSY;
597a711b 3188 mddev->new_layout = n;
50ac168a 3189 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3190 if (err) {
3191 mddev->new_layout = mddev->layout;
b3546035 3192 return err;
597a711b 3193 }
b3546035 3194 } else {
08a02ecd 3195 mddev->new_layout = n;
b3546035
N		 3196 if (mddev->reshape_position == MaxSector)
3197 mddev->layout = n;
3198 }
d4dbd025
N		 3199 return len;
3200}
3201static struct md_sysfs_entry md_layout =
80ca3a44 3202__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3203
3204
eae1701f 3205static ssize_t
96de1e66 3206raid_disks_show(mddev_t *mddev, char *page)
eae1701f 3207{
bb636547
N		 3208 if (mddev->raid_disks == 0)
3209 return 0;
08a02ecd
N		 3210 if (mddev->reshape_position != MaxSector &&
3211 mddev->delta_disks != 0)
3212 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3213 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3214 return sprintf(page, "%d\n", mddev->raid_disks);
3215}
3216
da943b99
N		 3217static int update_raid_disks(mddev_t *mddev, int raid_disks);
3218
3219static ssize_t
3220raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3221{
da943b99
N		 3222 char *e;
3223 int rv = 0;
3224 unsigned long n = simple_strtoul(buf, &e, 10);
3225
3226 if (!*buf || (*e && *e != '\n'))
3227 return -EINVAL;
3228
3229 if (mddev->pers)
3230 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3231 else if (mddev->reshape_position != MaxSector) {
3232 int olddisks = mddev->raid_disks - mddev->delta_disks;
3233 mddev->delta_disks = n - olddisks;
3234 mddev->raid_disks = n;
3235 } else
da943b99
N		 3236 mddev->raid_disks = n;
3237 return rv ? rv : len;
3238}
3239static struct md_sysfs_entry md_raid_disks =
80ca3a44 3240__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3241
3b34380a
N		 3242static ssize_t
3243chunk_size_show(mddev_t *mddev, char *page)
3244{
08a02ecd 3245 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3246 mddev->chunk_sectors != mddev->new_chunk_sectors)
3247 return sprintf(page, "%d (%d)\n",
3248 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3249 mddev->chunk_sectors << 9);
3250 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3251}
3252
3253static ssize_t
3254chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3255{
3b34380a
N		 3256 char *e;
3257 unsigned long n = simple_strtoul(buf, &e, 10);
3258
3b34380a
N		 3259 if (!*buf || (*e && *e != '\n'))
3260 return -EINVAL;
3261
b3546035
N		 3262 if (mddev->pers) {
3263 int err;
50ac168a 3264 if (mddev->pers->check_reshape == NULL)
b3546035 3265 return -EBUSY;
597a711b 3266 mddev->new_chunk_sectors = n >> 9;
50ac168a 3267 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3268 if (err) {
3269 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3270 return err;
597a711b 3271 }
b3546035 3272 } else {
664e7c41 3273 mddev->new_chunk_sectors = n >> 9;
b3546035 3274 if (mddev->reshape_position == MaxSector)
9d8f0363 3275 mddev->chunk_sectors = n >> 9;
b3546035 3276 }
3b34380a
N		 3277 return len;
3278}
3279static struct md_sysfs_entry md_chunk_size =
80ca3a44 3280__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3281
a94213b1
N		 3282static ssize_t
3283resync_start_show(mddev_t *mddev, char *page)
3284{
d1a7c503
N		 3285 if (mddev->recovery_cp == MaxSector)
3286 return sprintf(page, "none\n");
a94213b1
N		 3287 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3288}
3289
3290static ssize_t
3291resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3292{
a94213b1
N		 3293 char *e;
3294 unsigned long long n = simple_strtoull(buf, &e, 10);
3295
3296 if (mddev->pers)
3297 return -EBUSY;
06e3c817
DW		 3298 if (cmd_match(buf, "none"))
3299 n = MaxSector;
3300 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3301 return -EINVAL;
3302
3303 mddev->recovery_cp = n;
3304 return len;
3305}
3306static struct md_sysfs_entry md_resync_start =
80ca3a44 3307__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3308
9e653b63
N		 3309/*
3310 * The array state can be:
3311 *
3312 * clear
3313 * No devices, no size, no level
3314 * Equivalent to STOP_ARRAY ioctl
3315 * inactive
3316 * May have some settings, but array is not active
3317 * all IO results in error
3318 * When written, doesn't tear down array, but just stops it
3319 * suspended (not supported yet)
3320 * All IO requests will block. The array can be reconfigured.
910d8cb3 3321 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3322 * readonly
3323 * no resync can happen. no superblocks get written.
3324 * write requests fail
3325 * read-auto
3326 * like readonly, but behaves like 'clean' on a write request.
3327 *
3328 * clean - no pending writes, but otherwise active.
3329 * When written to inactive array, starts without resync
3330 * If a write request arrives then
3331 * if metadata is known, mark 'dirty' and switch to 'active'.
3332 * if not known, block and switch to write-pending
3333 * If written to an active array that has pending writes, then fails.
3334 * active
3335 * fully active: IO and resync can be happening.
3336 * When written to inactive array, starts with resync
3337 *
3338 * write-pending
3339 * clean, but writes are blocked waiting for 'active' to be written.
3340 *
3341 * active-idle
3342 * like active, but no writes have been seen for a while (100msec).
3343 *
3344 */
3345enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3346 write_pending, active_idle, bad_word};
05381954 3347static char *array_states[] = {
9e653b63
N		 3348 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3349 "write-pending", "active-idle", NULL };
3350
3351static int match_word(const char *word, char **list)
3352{
3353 int n;
3354 for (n=0; list[n]; n++)
3355 if (cmd_match(word, list[n]))
3356 break;
3357 return n;
3358}
3359
3360static ssize_t
3361array_state_show(mddev_t *mddev, char *page)
3362{
3363 enum array_state st = inactive;
3364
3365 if (mddev->pers)
3366 switch(mddev->ro) {
3367 case 1:
3368 st = readonly;
3369 break;
3370 case 2:
3371 st = read_auto;
3372 break;
3373 case 0:
3374 if (mddev->in_sync)
3375 st = clean;
070dc6dd 3376 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3377 st = write_pending;
9e653b63
N		 3378 else if (mddev->safemode)
3379 st = active_idle;
3380 else
3381 st = active;
3382 }
3383 else {
3384 if (list_empty(&mddev->disks) &&
3385 mddev->raid_disks == 0 &&
58c0fed4 3386 mddev->dev_sectors == 0)
9e653b63
N		 3387 st = clear;
3388 else
3389 st = inactive;
3390 }
3391 return sprintf(page, "%s\n", array_states[st]);
3392}
3393
df5b20cf 3394static int do_md_stop(mddev_t * mddev, int ro, int is_open);
a4bd82d0 3395static int md_set_readonly(mddev_t * mddev, int is_open);
9e653b63
N		 3396static int do_md_run(mddev_t * mddev);
3397static int restart_array(mddev_t *mddev);
3398
3399static ssize_t
3400array_state_store(mddev_t *mddev, const char *buf, size_t len)
3401{
3402 int err = -EINVAL;
3403 enum array_state st = match_word(buf, array_states);
3404 switch(st) {
3405 case bad_word:
3406 break;
3407 case clear:
3408 /* stopping an active array */
f2ea68cf 3409 if (atomic_read(&mddev->openers) > 0)
e691063a 3410 return -EBUSY;
df5b20cf 3411 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3412 break;
3413 case inactive:
3414 /* stopping an active array */
3415 if (mddev->pers) {
f2ea68cf 3416 if (atomic_read(&mddev->openers) > 0)
9e653b63 3417 return -EBUSY;
df5b20cf 3418 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3419 } else
3420 err = 0; /* already inactive */
9e653b63
N		 3421 break;
3422 case suspended:
3423 break; /* not supported yet */
3424 case readonly:
3425 if (mddev->pers)
a4bd82d0 3426 err = md_set_readonly(mddev, 0);
9e653b63
N		 3427 else {
3428 mddev->ro = 1;
648b629e 3429 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3430 err = do_md_run(mddev);
3431 }
3432 break;
3433 case read_auto:
9e653b63 3434 if (mddev->pers) {
80268ee9 3435 if (mddev->ro == 0)
a4bd82d0 3436 err = md_set_readonly(mddev, 0);
80268ee9 3437 else if (mddev->ro == 1)
648b629e
N		 3438 err = restart_array(mddev);
3439 if (err == 0) {
3440 mddev->ro = 2;
3441 set_disk_ro(mddev->gendisk, 0);
3442 }
9e653b63
N		 3443 } else {
3444 mddev->ro = 2;
3445 err = do_md_run(mddev);
3446 }
3447 break;
3448 case clean:
3449 if (mddev->pers) {
3450 restart_array(mddev);
3451 spin_lock_irq(&mddev->write_lock);
3452 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3453 if (mddev->in_sync == 0) {
3454 mddev->in_sync = 1;
31a59e34
N		 3455 if (mddev->safemode == 1)
3456 mddev->safemode = 0;
070dc6dd 3457 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3458 }
3459 err = 0;
3460 } else
3461 err = -EBUSY;
9e653b63 3462 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3463 } else
3464 err = -EINVAL;
9e653b63
N		 3465 break;
3466 case active:
3467 if (mddev->pers) {
3468 restart_array(mddev);
070dc6dd 3469 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3470 wake_up(&mddev->sb_wait);
3471 err = 0;
3472 } else {
3473 mddev->ro = 0;
648b629e 3474 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3475 err = do_md_run(mddev);
3476 }
3477 break;
3478 case write_pending:
3479 case active_idle:
3480 /* these cannot be set */
3481 break;
3482 }
3483 if (err)
3484 return err;
0fd62b86 3485 else {
00bcb4ac 3486 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3487 return len;
0fd62b86 3488 }
9e653b63 3489}
80ca3a44
N		 3490static struct md_sysfs_entry md_array_state =
3491__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3492
1e50915f
RB		 3493static ssize_t
3494max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3495 return sprintf(page, "%d\n",
3496 atomic_read(&mddev->max_corr_read_errors));
3497}
3498
3499static ssize_t
3500max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3501{
3502 char *e;
3503 unsigned long n = simple_strtoul(buf, &e, 10);
3504
3505 if (*buf && (*e == 0 || *e == '\n')) {
3506 atomic_set(&mddev->max_corr_read_errors, n);
3507 return len;
3508 }
3509 return -EINVAL;
3510}
3511
3512static struct md_sysfs_entry max_corr_read_errors =
3513__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3514 max_corrected_read_errors_store);
3515
6d7ff738
N		 3516static ssize_t
3517null_show(mddev_t *mddev, char *page)
3518{
3519 return -EINVAL;
3520}
3521
3522static ssize_t
3523new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3524{
3525 /* buf must be %d:%d\n? giving major and minor numbers */
3526 /* The new device is added to the array.
3527 * If the array has a persistent superblock, we read the
3528 * superblock to initialise info and check validity.
3529 * Otherwise, only checking done is that in bind_rdev_to_array,
3530 * which mainly checks size.
3531 */
3532 char *e;
3533 int major = simple_strtoul(buf, &e, 10);
3534 int minor;
3535 dev_t dev;
3536 mdk_rdev_t *rdev;
3537 int err;
3538
3539 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3540 return -EINVAL;
3541 minor = simple_strtoul(e+1, &e, 10);
3542 if (*e && *e != '\n')
3543 return -EINVAL;
3544 dev = MKDEV(major, minor);
3545 if (major != MAJOR(dev) ||
3546 minor != MINOR(dev))
3547 return -EOVERFLOW;
3548
3549
3550 if (mddev->persistent) {
3551 rdev = md_import_device(dev, mddev->major_version,
3552 mddev->minor_version);
3553 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3554 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3555 mdk_rdev_t, same_set);
3556 err = super_types[mddev->major_version]
3557 .load_super(rdev, rdev0, mddev->minor_version);
3558 if (err < 0)
3559 goto out;
3560 }
c5d79adb
N		 3561 } else if (mddev->external)
3562 rdev = md_import_device(dev, -2, -1);
3563 else
6d7ff738
N		 3564 rdev = md_import_device(dev, -1, -1);
3565
3566 if (IS_ERR(rdev))
3567 return PTR_ERR(rdev);
3568 err = bind_rdev_to_array(rdev, mddev);
3569 out:
3570 if (err)
3571 export_rdev(rdev);
3572 return err ? err : len;
3573}
3574
3575static struct md_sysfs_entry md_new_device =
80ca3a44 3576__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3577
9b1d1dac
PC		 3578static ssize_t
3579bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3580{
3581 char *end;
3582 unsigned long chunk, end_chunk;
3583
3584 if (!mddev->bitmap)
3585 goto out;
3586 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3587 while (*buf) {
3588 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3589 if (buf == end) break;
3590 if (*end == '-') { /* range */
3591 buf = end + 1;
3592 end_chunk = simple_strtoul(buf, &end, 0);
3593 if (buf == end) break;
3594 }
3595 if (*end && !isspace(*end)) break;
3596 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3597 buf = skip_spaces(end);
9b1d1dac
PC		 3598 }
3599 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3600out:
3601 return len;
3602}
3603
3604static struct md_sysfs_entry md_bitmap =
3605__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3606
a35b0d69
N		 3607static ssize_t
3608size_show(mddev_t *mddev, char *page)
3609{
58c0fed4
AN		 3610 return sprintf(page, "%llu\n",
3611 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3612}
3613
d71f9f88 3614static int update_size(mddev_t *mddev, sector_t num_sectors);
a35b0d69
N		 3615
3616static ssize_t
3617size_store(mddev_t *mddev, const char *buf, size_t len)
3618{
3619 /* If array is inactive, we can reduce the component size, but
3620 * not increase it (except from 0).
3621 * If array is active, we can try an on-line resize
3622 */
b522adcd
DW		 3623 sector_t sectors;
3624 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3625
58c0fed4
AN		 3626 if (err < 0)
3627 return err;
a35b0d69 3628 if (mddev->pers) {
58c0fed4 3629 err = update_size(mddev, sectors);
850b2b42 3630 md_update_sb(mddev, 1);
a35b0d69 3631 } else {
58c0fed4
AN		 3632 if (mddev->dev_sectors == 0 ||
3633 mddev->dev_sectors > sectors)
3634 mddev->dev_sectors = sectors;
a35b0d69
N		 3635 else
3636 err = -ENOSPC;
3637 }
3638 return err ? err : len;
3639}
3640
3641static struct md_sysfs_entry md_size =
80ca3a44 3642__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3643
8bb93aac
N		 3644
3645/* Metdata version.
e691063a
N		 3646 * This is one of
3647 * 'none' for arrays with no metadata (good luck...)
3648 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3649 * or N.M for internally known formats
3650 */
3651static ssize_t
3652metadata_show(mddev_t *mddev, char *page)
3653{
3654 if (mddev->persistent)
3655 return sprintf(page, "%d.%d\n",
3656 mddev->major_version, mddev->minor_version);
e691063a
N		 3657 else if (mddev->external)
3658 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3659 else
3660 return sprintf(page, "none\n");
3661}
3662
3663static ssize_t
3664metadata_store(mddev_t *mddev, const char *buf, size_t len)
3665{
3666 int major, minor;
3667 char *e;
ea43ddd8
N		 3668 /* Changing the details of 'external' metadata is
3669 * always permitted. Otherwise there must be
3670 * no devices attached to the array.
3671 */
3672 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3673 ;
3674 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3675 return -EBUSY;
3676
3677 if (cmd_match(buf, "none")) {
3678 mddev->persistent = 0;
e691063a
N		 3679 mddev->external = 0;
3680 mddev->major_version = 0;
3681 mddev->minor_version = 90;
3682 return len;
3683 }
3684 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3685 size_t namelen = len-9;
e691063a
N		 3686 if (namelen >= sizeof(mddev->metadata_type))
3687 namelen = sizeof(mddev->metadata_type)-1;
3688 strncpy(mddev->metadata_type, buf+9, namelen);
3689 mddev->metadata_type[namelen] = 0;
3690 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3691 mddev->metadata_type[--namelen] = 0;
3692 mddev->persistent = 0;
3693 mddev->external = 1;
8bb93aac
N		 3694 mddev->major_version = 0;
3695 mddev->minor_version = 90;
3696 return len;
3697 }
3698 major = simple_strtoul(buf, &e, 10);
3699 if (e==buf || *e != '.')
3700 return -EINVAL;
3701 buf = e+1;
3702 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 3703 if (e==buf || (*e && *e != '\n') )
8bb93aac 3704 return -EINVAL;
50511da3 3705 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 3706 return -ENOENT;
3707 mddev->major_version = major;
3708 mddev->minor_version = minor;
3709 mddev->persistent = 1;
e691063a 3710 mddev->external = 0;
8bb93aac
N		 3711 return len;
3712}
3713
3714static struct md_sysfs_entry md_metadata =
80ca3a44 3715__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 3716
24dd469d 3717static ssize_t
7eec314d 3718action_show(mddev_t *mddev, char *page)
24dd469d 3719{
7eec314d 3720 char *type = "idle";
b6a9ce68
N		 3721 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3722 type = "frozen";
3723 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 3724 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 3725 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3726 type = "reshape";
3727 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 3728 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3729 type = "resync";
3730 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3731 type = "check";
3732 else
3733 type = "repair";
72a23c21 3734 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 3735 type = "recover";
3736 }
3737 return sprintf(page, "%s\n", type);
3738}
3739
3740static ssize_t
7eec314d 3741action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 3742{
7eec314d
N		 3743 if (!mddev->pers || !mddev->pers->sync_request)
3744 return -EINVAL;
3745
b6a9ce68
N		 3746 if (cmd_match(page, "frozen"))
3747 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3748 else
3749 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3750
3751 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 3752 if (mddev->sync_thread) {
3753 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3754 md_unregister_thread(mddev->sync_thread);
3755 mddev->sync_thread = NULL;
3756 mddev->recovery = 0;
3757 }
03c902e1
N		 3758 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3759 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 3760 return -EBUSY;
72a23c21
NB		 3761 else if (cmd_match(page, "resync"))
3762 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3763 else if (cmd_match(page, "recover")) {
3764 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 3765 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 3766 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 3767 int err;
3768 if (mddev->pers->start_reshape == NULL)
3769 return -EINVAL;
3770 err = mddev->pers->start_reshape(mddev);
3771 if (err)
3772 return err;
a99ac971 3773 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 3774 } else {
bce74dac 3775 if (cmd_match(page, "check"))
7eec314d 3776 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 3777 else if (!cmd_match(page, "repair"))
7eec314d
N		 3778 return -EINVAL;
3779 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3780 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 3781 }
03c902e1 3782 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 3783 md_wakeup_thread(mddev->thread);
00bcb4ac 3784 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 3785 return len;
3786}
3787
9d88883e 3788static ssize_t
96de1e66 3789mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 3790{
3791 return sprintf(page, "%llu\n",
3792 (unsigned long long) mddev->resync_mismatches);
3793}
3794
80ca3a44
N		 3795static struct md_sysfs_entry md_scan_mode =
3796__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 3797
96de1e66 3798
80ca3a44 3799static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 3800
88202a0c
N		 3801static ssize_t
3802sync_min_show(mddev_t *mddev, char *page)
3803{
3804 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3805 mddev->sync_speed_min ? "local": "system");
3806}
3807
3808static ssize_t
3809sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3810{
3811 int min;
3812 char *e;
3813 if (strncmp(buf, "system", 6)==0) {
3814 mddev->sync_speed_min = 0;
3815 return len;
3816 }
3817 min = simple_strtoul(buf, &e, 10);
3818 if (buf == e || (*e && *e != '\n') || min <= 0)
3819 return -EINVAL;
3820 mddev->sync_speed_min = min;
3821 return len;
3822}
3823
3824static struct md_sysfs_entry md_sync_min =
3825__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3826
3827static ssize_t
3828sync_max_show(mddev_t *mddev, char *page)
3829{
3830 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3831 mddev->sync_speed_max ? "local": "system");
3832}
3833
3834static ssize_t
3835sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3836{
3837 int max;
3838 char *e;
3839 if (strncmp(buf, "system", 6)==0) {
3840 mddev->sync_speed_max = 0;
3841 return len;
3842 }
3843 max = simple_strtoul(buf, &e, 10);
3844 if (buf == e || (*e && *e != '\n') || max <= 0)
3845 return -EINVAL;
3846 mddev->sync_speed_max = max;
3847 return len;
3848}
3849
3850static struct md_sysfs_entry md_sync_max =
3851__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3852
d7f3d291
IP		 3853static ssize_t
3854degraded_show(mddev_t *mddev, char *page)
3855{
3856 return sprintf(page, "%d\n", mddev->degraded);
3857}
3858static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 3859
90b08710
BS		 3860static ssize_t
3861sync_force_parallel_show(mddev_t *mddev, char *page)
3862{
3863 return sprintf(page, "%d\n", mddev->parallel_resync);
3864}
3865
3866static ssize_t
3867sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3868{
3869 long n;
3870
3871 if (strict_strtol(buf, 10, &n))
3872 return -EINVAL;
3873
3874 if (n != 0 && n != 1)
3875 return -EINVAL;
3876
3877 mddev->parallel_resync = n;
3878
3879 if (mddev->sync_thread)
3880 wake_up(&resync_wait);
3881
3882 return len;
3883}
3884
3885/* force parallel resync, even with shared block devices */
3886static struct md_sysfs_entry md_sync_force_parallel =
3887__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3888 sync_force_parallel_show, sync_force_parallel_store);
3889
88202a0c
N		 3890static ssize_t
3891sync_speed_show(mddev_t *mddev, char *page)
3892{
3893 unsigned long resync, dt, db;
d1a7c503
N		 3894 if (mddev->curr_resync == 0)
3895 return sprintf(page, "none\n");
9687a60c
AN		 3896 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3897 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 3898 if (!dt) dt++;
9687a60c
AN		 3899 db = resync - mddev->resync_mark_cnt;
3900 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 3901}
3902
80ca3a44 3903static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 3904
3905static ssize_t
3906sync_completed_show(mddev_t *mddev, char *page)
3907{
58c0fed4 3908 unsigned long max_sectors, resync;
88202a0c 3909
acb180b0
N		 3910 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3911 return sprintf(page, "none\n");
3912
88202a0c 3913 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 3914 max_sectors = mddev->resync_max_sectors;
88202a0c 3915 else
58c0fed4 3916 max_sectors = mddev->dev_sectors;
88202a0c 3917
acb180b0 3918 resync = mddev->curr_resync_completed;
58c0fed4 3919 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
88202a0c
N		 3920}
3921
80ca3a44 3922static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3923
5e96ee65
NB		 3924static ssize_t
3925min_sync_show(mddev_t *mddev, char *page)
3926{
3927 return sprintf(page, "%llu\n",
3928 (unsigned long long)mddev->resync_min);
3929}
3930static ssize_t
3931min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3932{
3933 unsigned long long min;
3934 if (strict_strtoull(buf, 10, &min))
3935 return -EINVAL;
3936 if (min > mddev->resync_max)
3937 return -EINVAL;
3938 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3939 return -EBUSY;
3940
3941 /* Must be a multiple of chunk_size */
9d8f0363 3942 if (mddev->chunk_sectors) {
2ac06c33 3943 sector_t temp = min;
9d8f0363 3944 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 3945 return -EINVAL;
3946 }
3947 mddev->resync_min = min;
3948
3949 return len;
3950}
3951
3952static struct md_sysfs_entry md_min_sync =
3953__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3954
c6207277
N		 3955static ssize_t
3956max_sync_show(mddev_t *mddev, char *page)
3957{
3958 if (mddev->resync_max == MaxSector)
3959 return sprintf(page, "max\n");
3960 else
3961 return sprintf(page, "%llu\n",
3962 (unsigned long long)mddev->resync_max);
3963}
3964static ssize_t
3965max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3966{
3967 if (strncmp(buf, "max", 3) == 0)
3968 mddev->resync_max = MaxSector;
3969 else {
5e96ee65
NB		 3970 unsigned long long max;
3971 if (strict_strtoull(buf, 10, &max))
3972 return -EINVAL;
3973 if (max < mddev->resync_min)
c6207277
N		 3974 return -EINVAL;
3975 if (max < mddev->resync_max &&
4d484a4a 3976 mddev->ro == 0 &&
c6207277
N		 3977 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3978 return -EBUSY;
3979
3980 /* Must be a multiple of chunk_size */
9d8f0363 3981 if (mddev->chunk_sectors) {
2ac06c33 3982 sector_t temp = max;
9d8f0363 3983 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 3984 return -EINVAL;
3985 }
3986 mddev->resync_max = max;
3987 }
3988 wake_up(&mddev->recovery_wait);
3989 return len;
3990}
3991
3992static struct md_sysfs_entry md_max_sync =
3993__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3994
e464eafd
N		 3995static ssize_t
3996suspend_lo_show(mddev_t *mddev, char *page)
3997{
3998 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3999}
4000
4001static ssize_t
4002suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4003{
4004 char *e;
4005 unsigned long long new = simple_strtoull(buf, &e, 10);
4006
b8d966ef
N		 4007 if (mddev->pers == NULL ||
4008 mddev->pers->quiesce == NULL)
e464eafd
N		 4009 return -EINVAL;
4010 if (buf == e || (*e && *e != '\n'))
4011 return -EINVAL;
4012 if (new >= mddev->suspend_hi ||
4013 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4014 mddev->suspend_lo = new;
4015 mddev->pers->quiesce(mddev, 2);
4016 return len;
4017 } else
4018 return -EINVAL;
4019}
4020static struct md_sysfs_entry md_suspend_lo =
4021__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4022
4023
4024static ssize_t
4025suspend_hi_show(mddev_t *mddev, char *page)
4026{
4027 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4028}
4029
4030static ssize_t
4031suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4032{
4033 char *e;
4034 unsigned long long new = simple_strtoull(buf, &e, 10);
4035
b8d966ef
N		 4036 if (mddev->pers == NULL ||
4037 mddev->pers->quiesce == NULL)
e464eafd
N		 4038 return -EINVAL;
4039 if (buf == e || (*e && *e != '\n'))
4040 return -EINVAL;
4041 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4042 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4043 mddev->suspend_hi = new;
4044 mddev->pers->quiesce(mddev, 1);
4045 mddev->pers->quiesce(mddev, 0);
4046 return len;
4047 } else
4048 return -EINVAL;
4049}
4050static struct md_sysfs_entry md_suspend_hi =
4051__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4052
08a02ecd
N		 4053static ssize_t
4054reshape_position_show(mddev_t *mddev, char *page)
4055{
4056 if (mddev->reshape_position != MaxSector)
4057 return sprintf(page, "%llu\n",
4058 (unsigned long long)mddev->reshape_position);
4059 strcpy(page, "none\n");
4060 return 5;
4061}
4062
4063static ssize_t
4064reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4065{
4066 char *e;
4067 unsigned long long new = simple_strtoull(buf, &e, 10);
4068 if (mddev->pers)
4069 return -EBUSY;
4070 if (buf == e || (*e && *e != '\n'))
4071 return -EINVAL;
4072 mddev->reshape_position = new;
4073 mddev->delta_disks = 0;
4074 mddev->new_level = mddev->level;
4075 mddev->new_layout = mddev->layout;
664e7c41 4076 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4077 return len;
4078}
4079
4080static struct md_sysfs_entry md_reshape_position =
4081__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4082 reshape_position_store);
4083
b522adcd
DW		 4084static ssize_t
4085array_size_show(mddev_t *mddev, char *page)
4086{
4087 if (mddev->external_size)
4088 return sprintf(page, "%llu\n",
4089 (unsigned long long)mddev->array_sectors/2);
4090 else
4091 return sprintf(page, "default\n");
4092}
4093
4094static ssize_t
4095array_size_store(mddev_t *mddev, const char *buf, size_t len)
4096{
4097 sector_t sectors;
4098
4099 if (strncmp(buf, "default", 7) == 0) {
4100 if (mddev->pers)
4101 sectors = mddev->pers->size(mddev, 0, 0);
4102 else
4103 sectors = mddev->array_sectors;
4104
4105 mddev->external_size = 0;
4106 } else {
4107 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4108 return -EINVAL;
4109 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4110 return -E2BIG;
b522adcd
DW		 4111
4112 mddev->external_size = 1;
4113 }
4114
4115 mddev->array_sectors = sectors;
4116 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e
N		 4117 if (mddev->pers)
4118 revalidate_disk(mddev->gendisk);
b522adcd
DW		 4119
4120 return len;
4121}
4122
4123static struct md_sysfs_entry md_array_size =
4124__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4125 array_size_store);
e464eafd 4126
eae1701f
N		 4127static struct attribute *md_default_attrs[] = {
4128 &md_level.attr,
d4dbd025 4129 &md_layout.attr,
eae1701f 4130 &md_raid_disks.attr,
3b34380a 4131 &md_chunk_size.attr,
a35b0d69 4132 &md_size.attr,
a94213b1 4133 &md_resync_start.attr,
8bb93aac 4134 &md_metadata.attr,
6d7ff738 4135 &md_new_device.attr,
16f17b39 4136 &md_safe_delay.attr,
9e653b63 4137 &md_array_state.attr,
08a02ecd 4138 &md_reshape_position.attr,
b522adcd 4139 &md_array_size.attr,
1e50915f 4140 &max_corr_read_errors.attr,
411036fa
N		 4141 NULL,
4142};
4143
4144static struct attribute *md_redundancy_attrs[] = {
24dd469d 4145 &md_scan_mode.attr,
9d88883e 4146 &md_mismatches.attr,
88202a0c
N		 4147 &md_sync_min.attr,
4148 &md_sync_max.attr,
4149 &md_sync_speed.attr,
90b08710 4150 &md_sync_force_parallel.attr,
88202a0c 4151 &md_sync_completed.attr,
5e96ee65 4152 &md_min_sync.attr,
c6207277 4153 &md_max_sync.attr,
e464eafd
N		 4154 &md_suspend_lo.attr,
4155 &md_suspend_hi.attr,
9b1d1dac 4156 &md_bitmap.attr,
d7f3d291 4157 &md_degraded.attr,
eae1701f
N		 4158 NULL,
4159};
411036fa
N		 4160static struct attribute_group md_redundancy_group = {
4161 .name = NULL,
4162 .attrs = md_redundancy_attrs,
4163};
4164
eae1701f
N		 4165
4166static ssize_t
4167md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4168{
4169 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4170 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4171 ssize_t rv;
eae1701f
N		 4172
4173 if (!entry->show)
4174 return -EIO;
5dc5cf7d
IM		 4175 rv = mddev_lock(mddev);
4176 if (!rv) {
4177 rv = entry->show(mddev, page);
4178 mddev_unlock(mddev);
4179 }
96de1e66 4180 return rv;
eae1701f
N		 4181}
4182
4183static ssize_t
4184md_attr_store(struct kobject *kobj, struct attribute *attr,
4185 const char *page, size_t length)
4186{
4187 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4188 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4189 ssize_t rv;
eae1701f
N		 4190
4191 if (!entry->store)
4192 return -EIO;
67463acb
N		 4193 if (!capable(CAP_SYS_ADMIN))
4194 return -EACCES;
5dc5cf7d 4195 rv = mddev_lock(mddev);
d3374825
N		 4196 if (mddev->hold_active == UNTIL_IOCTL)
4197 mddev->hold_active = 0;
5dc5cf7d
IM		 4198 if (!rv) {
4199 rv = entry->store(mddev, page, length);
4200 mddev_unlock(mddev);
4201 }
96de1e66 4202 return rv;
eae1701f
N		 4203}
4204
4205static void md_free(struct kobject *ko)
4206{
4207 mddev_t *mddev = container_of(ko, mddev_t, kobj);
a21d1504
N		 4208
4209 if (mddev->sysfs_state)
4210 sysfs_put(mddev->sysfs_state);
4211
4212 if (mddev->gendisk) {
4213 del_gendisk(mddev->gendisk);
4214 put_disk(mddev->gendisk);
4215 }
4216 if (mddev->queue)
4217 blk_cleanup_queue(mddev->queue);
4218
eae1701f
N		 4219 kfree(mddev);
4220}
4221
52cf25d0 4222static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4223 .show = md_attr_show,
4224 .store = md_attr_store,
4225};
4226static struct kobj_type md_ktype = {
4227 .release = md_free,
4228 .sysfs_ops = &md_sysfs_ops,
4229 .default_attrs = md_default_attrs,
4230};
4231
1da177e4
LT		 4232int mdp_major = 0;
4233
5fd3a17e
DW		 4234static void mddev_delayed_delete(struct work_struct *ws)
4235{
4236 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4237
43a70507 4238 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4239 kobject_del(&mddev->kobj);
4240 kobject_put(&mddev->kobj);
4241}
4242
efeb53c0 4243static int md_alloc(dev_t dev, char *name)
1da177e4 4244{
48c9c27b 4245 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 4246 mddev_t *mddev = mddev_find(dev);
4247 struct gendisk *disk;
efeb53c0
N		 4248 int partitioned;
4249 int shift;
4250 int unit;
3830c62f 4251 int error;
1da177e4
LT		 4252
4253 if (!mddev)
efeb53c0
N		 4254 return -ENODEV;
4255
4256 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4257 shift = partitioned ? MdpMinorShift : 0;
4258 unit = MINOR(mddev->unit) >> shift;
1da177e4 4259
e804ac78
TH		 4260 /* wait for any previous instance of this device to be
4261 * completely removed (mddev_delayed_delete).
d3374825 4262 */
e804ac78 4263 flush_workqueue(md_misc_wq);
d3374825 4264
48c9c27b 4265 mutex_lock(&disks_mutex);
0909dc44
N		 4266 error = -EEXIST;
4267 if (mddev->gendisk)
4268 goto abort;
efeb53c0
N		 4269
4270 if (name) {
4271 /* Need to ensure that 'name' is not a duplicate.
4272 */
4273 mddev_t *mddev2;
4274 spin_lock(&all_mddevs_lock);
4275
4276 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4277 if (mddev2->gendisk &&
4278 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4279 spin_unlock(&all_mddevs_lock);
0909dc44 4280 goto abort;
efeb53c0
N		 4281 }
4282 spin_unlock(&all_mddevs_lock);
1da177e4 4283 }
8b765398 4284
0909dc44 4285 error = -ENOMEM;
8b765398 4286 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4287 if (!mddev->queue)
4288 goto abort;
409c57f3
N		 4289 mddev->queue->queuedata = mddev;
4290
8b765398
N		 4291 /* Can be unlocked because the queue is new: no concurrency */
4292 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
4293
409c57f3 4294 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4295
1da177e4
LT		 4296 disk = alloc_disk(1 << shift);
4297 if (!disk) {
8b765398
N		 4298 blk_cleanup_queue(mddev->queue);
4299 mddev->queue = NULL;
0909dc44 4300 goto abort;
1da177e4 4301 }
efeb53c0 4302 disk->major = MAJOR(mddev->unit);
1da177e4 4303 disk->first_minor = unit << shift;
efeb53c0
N		 4304 if (name)
4305 strcpy(disk->disk_name, name);
4306 else if (partitioned)
1da177e4 4307 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4308 else
1da177e4 4309 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4310 disk->fops = &md_fops;
4311 disk->private_data = mddev;
4312 disk->queue = mddev->queue;
92850bbd 4313 /* Allow extended partitions. This makes the
d3374825 4314 * 'mdp' device redundant, but we can't really
92850bbd
N		 4315 * remove it now.
4316 */
4317 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4
LT		 4318 add_disk(disk);
4319 mddev->gendisk = disk;
ed9e1982
TH		 4320 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4321 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4322 if (error) {
4323 /* This isn't possible, but as kobject_init_and_add is marked
4324 * __must_check, we must do something with the result
4325 */
5e55e2f5
N		 4326 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4327 disk->disk_name);
0909dc44
N		 4328 error = 0;
4329 }
00bcb4ac
N		 4330 if (mddev->kobj.sd &&
4331 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4332 printk(KERN_DEBUG "pointless warning\n");
0909dc44
N		 4333 abort:
4334 mutex_unlock(&disks_mutex);
00bcb4ac 4335 if (!error && mddev->kobj.sd) {
3830c62f 4336 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4337 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4338 }
d3374825 4339 mddev_put(mddev);
0909dc44 4340 return error;
efeb53c0
N		 4341}
4342
4343static struct kobject *md_probe(dev_t dev, int *part, void *data)
4344{
4345 md_alloc(dev, NULL);
1da177e4
LT		 4346 return NULL;
4347}
4348
efeb53c0
N		 4349static int add_named_array(const char *val, struct kernel_param *kp)
4350{
4351 /* val must be "md_*" where * is not all digits.
4352 * We allocate an array with a large free minor number, and
4353 * set the name to val. val must not already be an active name.
4354 */
4355 int len = strlen(val);
4356 char buf[DISK_NAME_LEN];
4357
4358 while (len && val[len-1] == '\n')
4359 len--;
4360 if (len >= DISK_NAME_LEN)
4361 return -E2BIG;
4362 strlcpy(buf, val, len+1);
4363 if (strncmp(buf, "md_", 3) != 0)
4364 return -EINVAL;
4365 return md_alloc(0, buf);
4366}
4367
1da177e4
LT		 4368static void md_safemode_timeout(unsigned long data)
4369{
4370 mddev_t *mddev = (mddev_t *) data;
4371
0fd62b86
NB		 4372 if (!atomic_read(&mddev->writes_pending)) {
4373 mddev->safemode = 1;
4374 if (mddev->external)
00bcb4ac 4375 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4376 }
1da177e4
LT		 4377 md_wakeup_thread(mddev->thread);
4378}
4379
6ff8d8ec 4380static int start_dirty_degraded;
1da177e4 4381
390ee602 4382int md_run(mddev_t *mddev)
1da177e4 4383{
2604b703 4384 int err;
1da177e4 4385 mdk_rdev_t *rdev;
2604b703 4386 struct mdk_personality *pers;
1da177e4 4387
a757e64c
N		 4388 if (list_empty(&mddev->disks))
4389 /* cannot run an array with no devices.. */
1da177e4 4390 return -EINVAL;
1da177e4
LT		 4391
4392 if (mddev->pers)
4393 return -EBUSY;
bb4f1e9d
N		 4394 /* Cannot run until previous stop completes properly */
4395 if (mddev->sysfs_active)
4396 return -EBUSY;
b6eb127d 4397
1da177e4
LT		 4398 /*
4399 * Analyze all RAID superblock(s)
4400 */
1ec4a939
N		 4401 if (!mddev->raid_disks) {
4402 if (!mddev->persistent)
4403 return -EINVAL;
a757e64c 4404 analyze_sbs(mddev);
1ec4a939 4405 }
1da177e4 4406
d9d166c2
N		 4407 if (mddev->level != LEVEL_NONE)
4408 request_module("md-level-%d", mddev->level);
4409 else if (mddev->clevel[0])
4410 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4411
4412 /*
4413 * Drop all container device buffers, from now on
4414 * the only valid external interface is through the md
4415 * device.
1da177e4 4416 */
159ec1fc 4417 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4418 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4419 continue;
4420 sync_blockdev(rdev->bdev);
f98393a6 4421 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4422
4423 /* perform some consistency tests on the device.
4424 * We don't want the data to overlap the metadata,
58c0fed4 4425 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4426 */
0f420358 4427 if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4428 if (mddev->dev_sectors &&
4429 rdev->data_offset + mddev->dev_sectors
0f420358 4430 > rdev->sb_start) {
f0d76d70
N		 4431 printk("md: %s: data overlaps metadata\n",
4432 mdname(mddev));
4433 return -EINVAL;
4434 }
4435 } else {
0f420358 4436 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4437 > rdev->data_offset) {
4438 printk("md: %s: metadata overlaps data\n",
4439 mdname(mddev));
4440 return -EINVAL;
4441 }
4442 }
00bcb4ac 4443 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4444 }
4445
a167f663
N		 4446 if (mddev->bio_set == NULL)
4447 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4448
1da177e4 4449 spin_lock(&pers_lock);
d9d166c2 4450 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4451 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4452 spin_unlock(&pers_lock);
d9d166c2
N		 4453 if (mddev->level != LEVEL_NONE)
4454 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4455 mddev->level);
4456 else
4457 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4458 mddev->clevel);
1da177e4
LT		 4459 return -EINVAL;
4460 }
2604b703 4461 mddev->pers = pers;
1da177e4 4462 spin_unlock(&pers_lock);
34817e8c
N		 4463 if (mddev->level != pers->level) {
4464 mddev->level = pers->level;
4465 mddev->new_level = pers->level;
4466 }
d9d166c2 4467 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4468
f6705578 4469 if (mddev->reshape_position != MaxSector &&
63c70c4f 4470 pers->start_reshape == NULL) {
f6705578
N		 4471 /* This personality cannot handle reshaping... */
4472 mddev->pers = NULL;
4473 module_put(pers->owner);
4474 return -EINVAL;
4475 }
4476
7dd5e7c3
N		 4477 if (pers->sync_request) {
4478 /* Warn if this is a potentially silly
4479 * configuration.
4480 */
4481 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4482 mdk_rdev_t *rdev2;
7dd5e7c3 4483 int warned = 0;
159ec1fc
CR		 4484
4485 list_for_each_entry(rdev, &mddev->disks, same_set)
4486 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4487 if (rdev < rdev2 &&
4488 rdev->bdev->bd_contains ==
4489 rdev2->bdev->bd_contains) {
4490 printk(KERN_WARNING
4491 "%s: WARNING: %s appears to be"
4492 " on the same physical disk as"
4493 " %s.\n",
4494 mdname(mddev),
4495 bdevname(rdev->bdev,b),
4496 bdevname(rdev2->bdev,b2));
4497 warned = 1;
4498 }
4499 }
159ec1fc 4500
7dd5e7c3
N		 4501 if (warned)
4502 printk(KERN_WARNING
4503 "True protection against single-disk"
4504 " failure might be compromised.\n");
4505 }
4506
657390d2 4507 mddev->recovery = 0;
58c0fed4
AN		 4508 /* may be over-ridden by personality */
4509 mddev->resync_max_sectors = mddev->dev_sectors;
4510
6ff8d8ec 4511 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4512
0f9552b5 4513 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4514 mddev->ro = 2; /* read-only, but switch on first write */
4515
b15c2e57 4516 err = mddev->pers->run(mddev);
13e53df3
AN		 4517 if (err)
4518 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4519 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4520 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4521 " but 'external_size' not in effect?\n", __func__);
4522 printk(KERN_ERR
4523 "md: invalid array_size %llu > default size %llu\n",
4524 (unsigned long long)mddev->array_sectors / 2,
4525 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4526 err = -EINVAL;
4527 mddev->pers->stop(mddev);
4528 }
4529 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4530 err = bitmap_create(mddev);
4531 if (err) {
4532 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4533 mdname(mddev), err);
4534 mddev->pers->stop(mddev);
4535 }
4536 }
1da177e4 4537 if (err) {
1da177e4
LT		 4538 module_put(mddev->pers->owner);
4539 mddev->pers = NULL;
32a7627c
N		 4540 bitmap_destroy(mddev);
4541 return err;
1da177e4 4542 }
5e55e2f5 4543 if (mddev->pers->sync_request) {
00bcb4ac
N		 4544 if (mddev->kobj.sd &&
4545 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4546 printk(KERN_WARNING
4547 "md: cannot register extra attributes for %s\n",
4548 mdname(mddev));
00bcb4ac 4549 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4550 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4551 mddev->ro = 0;
4552
1da177e4 4553 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4554 atomic_set(&mddev->max_corr_read_errors,
4555 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4556 mddev->safemode = 0;
4557 mddev->safemode_timer.function = md_safemode_timeout;
4558 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4559 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4560 mddev->in_sync = 1;
86e6ffdd 4561
159ec1fc 4562 list_for_each_entry(rdev, &mddev->disks, same_set)
86e6ffdd
N		 4563 if (rdev->raid_disk >= 0) {
4564 char nm[20];
4565 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5 4566 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
00bcb4ac 4567 /* failure here is OK */;
86e6ffdd 4568 }
1da177e4
LT		 4569
4570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4571
850b2b42
N		 4572 if (mddev->flags)
4573 md_update_sb(mddev, 0);
1da177e4 4574
0b8c9de0
N		 4575 md_wakeup_thread(mddev->thread);
4576 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 4577
d7603b7e 4578 md_new_event(mddev);
00bcb4ac
N		 4579 sysfs_notify_dirent_safe(mddev->sysfs_state);
4580 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4581 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4582 return 0;
4583}
390ee602 4584EXPORT_SYMBOL_GPL(md_run);
1da177e4 4585
fe60b014
N		 4586static int do_md_run(mddev_t *mddev)
4587{
4588 int err;
4589
4590 err = md_run(mddev);
4591 if (err)
4592 goto out;
69e51b44
N		 4593 err = bitmap_load(mddev);
4594 if (err) {
4595 bitmap_destroy(mddev);
4596 goto out;
4597 }
fe60b014
N		 4598 set_capacity(mddev->gendisk, mddev->array_sectors);
4599 revalidate_disk(mddev->gendisk);
4600 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4601out:
4602 return err;
4603}
4604
1da177e4
LT		 4605static int restart_array(mddev_t *mddev)
4606{
4607 struct gendisk *disk = mddev->gendisk;
1da177e4 4608
80fab1d7 4609 /* Complain if it has no devices */
1da177e4 4610 if (list_empty(&mddev->disks))
80fab1d7
AN		 4611 return -ENXIO;
4612 if (!mddev->pers)
4613 return -EINVAL;
4614 if (!mddev->ro)
4615 return -EBUSY;
4616 mddev->safemode = 0;
4617 mddev->ro = 0;
4618 set_disk_ro(disk, 0);
4619 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4620 mdname(mddev));
4621 /* Kick recovery or resync if necessary */
4622 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4623 md_wakeup_thread(mddev->thread);
4624 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4625 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4626 return 0;
1da177e4
LT		 4627}
4628
acc55e22
N		 4629/* similar to deny_write_access, but accounts for our holding a reference
4630 * to the file ourselves */
4631static int deny_bitmap_write_access(struct file * file)
4632{
4633 struct inode *inode = file->f_mapping->host;
4634
4635 spin_lock(&inode->i_lock);
4636 if (atomic_read(&inode->i_writecount) > 1) {
4637 spin_unlock(&inode->i_lock);
4638 return -ETXTBSY;
4639 }
4640 atomic_set(&inode->i_writecount, -1);
4641 spin_unlock(&inode->i_lock);
4642
4643 return 0;
4644}
4645
43a70507 4646void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4647{
4648 struct inode *inode = file->f_mapping->host;
4649
4650 spin_lock(&inode->i_lock);
4651 atomic_set(&inode->i_writecount, 1);
4652 spin_unlock(&inode->i_lock);
4653}
4654
6177b472
N		 4655static void md_clean(mddev_t *mddev)
4656{
4657 mddev->array_sectors = 0;
4658 mddev->external_size = 0;
4659 mddev->dev_sectors = 0;
4660 mddev->raid_disks = 0;
4661 mddev->recovery_cp = 0;
4662 mddev->resync_min = 0;
4663 mddev->resync_max = MaxSector;
4664 mddev->reshape_position = MaxSector;
4665 mddev->external = 0;
4666 mddev->persistent = 0;
4667 mddev->level = LEVEL_NONE;
4668 mddev->clevel[0] = 0;
4669 mddev->flags = 0;
4670 mddev->ro = 0;
4671 mddev->metadata_type[0] = 0;
4672 mddev->chunk_sectors = 0;
4673 mddev->ctime = mddev->utime = 0;
4674 mddev->layout = 0;
4675 mddev->max_disks = 0;
4676 mddev->events = 0;
a8707c08 4677 mddev->can_decrease_events = 0;
6177b472
N		 4678 mddev->delta_disks = 0;
4679 mddev->new_level = LEVEL_NONE;
4680 mddev->new_layout = 0;
4681 mddev->new_chunk_sectors = 0;
4682 mddev->curr_resync = 0;
4683 mddev->resync_mismatches = 0;
4684 mddev->suspend_lo = mddev->suspend_hi = 0;
4685 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4686 mddev->recovery = 0;
4687 mddev->in_sync = 0;
4688 mddev->degraded = 0;
6177b472
N		 4689 mddev->safemode = 0;
4690 mddev->bitmap_info.offset = 0;
4691 mddev->bitmap_info.default_offset = 0;
4692 mddev->bitmap_info.chunksize = 0;
4693 mddev->bitmap_info.daemon_sleep = 0;
4694 mddev->bitmap_info.max_write_behind = 0;
252ac522 4695 mddev->plug = NULL;
6177b472
N		 4696}
4697
390ee602 4698void md_stop_writes(mddev_t *mddev)
a047e125
N		 4699{
4700 if (mddev->sync_thread) {
4701 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4702 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4703 md_unregister_thread(mddev->sync_thread);
4704 mddev->sync_thread = NULL;
4705 }
4706
4707 del_timer_sync(&mddev->safemode_timer);
4708
4709 bitmap_flush(mddev);
4710 md_super_wait(mddev);
4711
4712 if (!mddev->in_sync || mddev->flags) {
4713 /* mark array as shutdown cleanly */
4714 mddev->in_sync = 1;
4715 md_update_sb(mddev, 1);
4716 }
4717}
390ee602 4718EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 4719
390ee602 4720void md_stop(mddev_t *mddev)
6177b472
N		 4721{
4722 mddev->pers->stop(mddev);
4723 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4724 mddev->to_remove = &md_redundancy_group;
4725 module_put(mddev->pers->owner);
4726 mddev->pers = NULL;
cca9cf90 4727 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 4728}
390ee602 4729EXPORT_SYMBOL_GPL(md_stop);
6177b472 4730
a4bd82d0
N		 4731static int md_set_readonly(mddev_t *mddev, int is_open)
4732{
4733 int err = 0;
4734 mutex_lock(&mddev->open_mutex);
4735 if (atomic_read(&mddev->openers) > is_open) {
4736 printk("md: %s still in use.\n",mdname(mddev));
4737 err = -EBUSY;
4738 goto out;
4739 }
4740 if (mddev->pers) {
4741 md_stop_writes(mddev);
4742
4743 err = -ENXIO;
4744 if (mddev->ro==1)
4745 goto out;
4746 mddev->ro = 1;
4747 set_disk_ro(mddev->gendisk, 1);
4748 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 4749 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 4750 err = 0;
4751 }
4752out:
4753 mutex_unlock(&mddev->open_mutex);
4754 return err;
4755}
4756
9e653b63
N		 4757/* mode:
4758 * 0 - completely stop and dis-assemble array
9e653b63
N		 4759 * 2 - stop but do not disassemble array
4760 */
df5b20cf 4761static int do_md_stop(mddev_t * mddev, int mode, int is_open)
1da177e4 4762{
1da177e4 4763 struct gendisk *disk = mddev->gendisk;
c4647292 4764 mdk_rdev_t *rdev;
1da177e4 4765
c8c00a69 4766 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 4767 if (atomic_read(&mddev->openers) > is_open ||
4768 mddev->sysfs_active) {
df5b20cf 4769 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 4770 mutex_unlock(&mddev->open_mutex);
4771 return -EBUSY;
4772 }
1da177e4 4773
6e17b027 4774 if (mddev->pers) {
a4bd82d0
N		 4775 if (mddev->ro)
4776 set_disk_ro(disk, 0);
409c57f3 4777
390ee602 4778 md_stop_writes(mddev);
a4bd82d0
N		 4779 md_stop(mddev);
4780 mddev->queue->merge_bvec_fn = NULL;
4781 mddev->queue->unplug_fn = NULL;
4782 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 4783
a4bd82d0 4784 /* tell userspace to handle 'inactive' */
00bcb4ac 4785 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 4786
a4bd82d0
N		 4787 list_for_each_entry(rdev, &mddev->disks, same_set)
4788 if (rdev->raid_disk >= 0) {
4789 char nm[20];
4790 sprintf(nm, "rd%d", rdev->raid_disk);
4791 sysfs_remove_link(&mddev->kobj, nm);
4792 }
c4647292 4793
a4bd82d0 4794 set_capacity(disk, 0);
6e17b027 4795 mutex_unlock(&mddev->open_mutex);
a4bd82d0 4796 revalidate_disk(disk);
0d4ca600 4797
a4bd82d0
N		 4798 if (mddev->ro)
4799 mddev->ro = 0;
6e17b027
N		 4800 } else
4801 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 4802 /*
4803 * Free resources if final stop
4804 */
9e653b63 4805 if (mode == 0) {
1da177e4
LT		 4806 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4807
978f946b 4808 bitmap_destroy(mddev);
c3d9714e
N		 4809 if (mddev->bitmap_info.file) {
4810 restore_bitmap_write_access(mddev->bitmap_info.file);
4811 fput(mddev->bitmap_info.file);
4812 mddev->bitmap_info.file = NULL;
978f946b 4813 }
c3d9714e 4814 mddev->bitmap_info.offset = 0;
978f946b 4815
1da177e4
LT		 4816 export_array(mddev);
4817
6177b472 4818 md_clean(mddev);
934d9c23 4819 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 4820 if (mddev->hold_active == UNTIL_STOP)
4821 mddev->hold_active = 0;
a4bd82d0 4822 }
3f9d99c1 4823 blk_integrity_unregister(disk);
d7603b7e 4824 md_new_event(mddev);
00bcb4ac 4825 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 4826 return 0;
1da177e4
LT		 4827}
4828
fdee8ae4 4829#ifndef MODULE
1da177e4
LT		 4830static void autorun_array(mddev_t *mddev)
4831{
4832 mdk_rdev_t *rdev;
1da177e4
LT		 4833 int err;
4834
a757e64c 4835 if (list_empty(&mddev->disks))
1da177e4 4836 return;
1da177e4
LT		 4837
4838 printk(KERN_INFO "md: running: ");
4839
159ec1fc 4840 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 4841 char b[BDEVNAME_SIZE];
4842 printk("<%s>", bdevname(rdev->bdev,b));
4843 }
4844 printk("\n");
4845
d710e138 4846 err = do_md_run(mddev);
1da177e4
LT		 4847 if (err) {
4848 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 4849 do_md_stop(mddev, 0, 0);
1da177e4
LT		 4850 }
4851}
4852
4853/*
4854 * lets try to run arrays based on all disks that have arrived
4855 * until now. (those are in pending_raid_disks)
4856 *
4857 * the method: pick the first pending disk, collect all disks with
4858 * the same UUID, remove all from the pending list and put them into
4859 * the 'same_array' list. Then order this list based on superblock
4860 * update time (freshest comes first), kick out 'old' disks and
4861 * compare superblocks. If everything's fine then run it.
4862 *
4863 * If "unit" is allocated, then bump its reference count
4864 */
4865static void autorun_devices(int part)
4866{
159ec1fc 4867 mdk_rdev_t *rdev0, *rdev, *tmp;
1da177e4
LT		 4868 mddev_t *mddev;
4869 char b[BDEVNAME_SIZE];
4870
4871 printk(KERN_INFO "md: autorun ...\n");
4872 while (!list_empty(&pending_raid_disks)) {
e8703fe1 4873 int unit;
1da177e4 4874 dev_t dev;
ad01c9e3 4875 LIST_HEAD(candidates);
1da177e4
LT		 4876 rdev0 = list_entry(pending_raid_disks.next,
4877 mdk_rdev_t, same_set);
4878
4879 printk(KERN_INFO "md: considering %s ...\n",
4880 bdevname(rdev0->bdev,b));
4881 INIT_LIST_HEAD(&candidates);
159ec1fc 4882 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 4883 if (super_90_load(rdev, rdev0, 0) >= 0) {
4884 printk(KERN_INFO "md: adding %s ...\n",
4885 bdevname(rdev->bdev,b));
4886 list_move(&rdev->same_set, &candidates);
4887 }
4888 /*
4889 * now we have a set of devices, with all of them having
4890 * mostly sane superblocks. It's time to allocate the
4891 * mddev.
4892 */
e8703fe1
N		 4893 if (part) {
4894 dev = MKDEV(mdp_major,
4895 rdev0->preferred_minor << MdpMinorShift);
4896 unit = MINOR(dev) >> MdpMinorShift;
4897 } else {
4898 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4899 unit = MINOR(dev);
4900 }
4901 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 4902 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4903 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4904 break;
4905 }
1da177e4
LT		 4906
4907 md_probe(dev, NULL, NULL);
4908 mddev = mddev_find(dev);
9bbbca3a
NB		 4909 if (!mddev || !mddev->gendisk) {
4910 if (mddev)
4911 mddev_put(mddev);
4912 printk(KERN_ERR
1da177e4
LT		 4913 "md: cannot allocate memory for md drive.\n");
4914 break;
4915 }
4916 if (mddev_lock(mddev))
4917 printk(KERN_WARNING "md: %s locked, cannot run\n",
4918 mdname(mddev));
4919 else if (mddev->raid_disks || mddev->major_version
4920 || !list_empty(&mddev->disks)) {
4921 printk(KERN_WARNING
4922 "md: %s already running, cannot run %s\n",
4923 mdname(mddev), bdevname(rdev0->bdev,b));
4924 mddev_unlock(mddev);
4925 } else {
4926 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 4927 mddev->persistent = 1;
159ec1fc 4928 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 4929 list_del_init(&rdev->same_set);
4930 if (bind_rdev_to_array(rdev, mddev))
4931 export_rdev(rdev);
4932 }
4933 autorun_array(mddev);
4934 mddev_unlock(mddev);
4935 }
4936 /* on success, candidates will be empty, on error
4937 * it won't...
4938 */
159ec1fc 4939 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 4940 list_del_init(&rdev->same_set);
1da177e4 4941 export_rdev(rdev);
4b80991c 4942 }
1da177e4
LT		 4943 mddev_put(mddev);
4944 }
4945 printk(KERN_INFO "md: ... autorun DONE.\n");
4946}
fdee8ae4 4947#endif /* !MODULE */
1da177e4 4948
1da177e4
LT		 4949static int get_version(void __user * arg)
4950{
4951 mdu_version_t ver;
4952
4953 ver.major = MD_MAJOR_VERSION;
4954 ver.minor = MD_MINOR_VERSION;
4955 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4956
4957 if (copy_to_user(arg, &ver, sizeof(ver)))
4958 return -EFAULT;
4959
4960 return 0;
4961}
4962
4963static int get_array_info(mddev_t * mddev, void __user * arg)
4964{
4965 mdu_array_info_t info;
a9f326eb 4966 int nr,working,insync,failed,spare;
1da177e4 4967 mdk_rdev_t *rdev;
1da177e4 4968
a9f326eb 4969 nr=working=insync=failed=spare=0;
159ec1fc 4970 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 4971 nr++;
b2d444d7 4972 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4973 failed++;
4974 else {
4975 working++;
b2d444d7 4976 if (test_bit(In_sync, &rdev->flags))
a9f326eb 4977 insync++;
1da177e4
LT		 4978 else
4979 spare++;
4980 }
4981 }
4982
4983 info.major_version = mddev->major_version;
4984 info.minor_version = mddev->minor_version;
4985 info.patch_version = MD_PATCHLEVEL_VERSION;
4986 info.ctime = mddev->ctime;
4987 info.level = mddev->level;
58c0fed4
AN		 4988 info.size = mddev->dev_sectors / 2;
4989 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 4990 info.size = -1;
1da177e4
LT		 4991 info.nr_disks = nr;
4992 info.raid_disks = mddev->raid_disks;
4993 info.md_minor = mddev->md_minor;
4994 info.not_persistent= !mddev->persistent;
4995
4996 info.utime = mddev->utime;
4997 info.state = 0;
4998 if (mddev->in_sync)
4999 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5000 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5001 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5002 info.active_disks = insync;
1da177e4
LT		 5003 info.working_disks = working;
5004 info.failed_disks = failed;
5005 info.spare_disks = spare;
5006
5007 info.layout = mddev->layout;
9d8f0363 5008 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5009
5010 if (copy_to_user(arg, &info, sizeof(info)))
5011 return -EFAULT;
5012
5013 return 0;
5014}
5015
87162a28 5016static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 5017{
5018 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5019 char *ptr, *buf = NULL;
5020 int err = -ENOMEM;
5021
b5470dc5
DW		 5022 if (md_allow_write(mddev))
5023 file = kmalloc(sizeof(*file), GFP_NOIO);
5024 else
5025 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5026
32a7627c
N		 5027 if (!file)
5028 goto out;
5029
5030 /* bitmap disabled, zero the first byte and copy out */
5031 if (!mddev->bitmap || !mddev->bitmap->file) {
5032 file->pathname[0] = '\0';
5033 goto copy_out;
5034 }
5035
5036 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5037 if (!buf)
5038 goto out;
5039
6bcfd601
CH		 5040 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5041 if (IS_ERR(ptr))
32a7627c
N		 5042 goto out;
5043
5044 strcpy(file->pathname, ptr);
5045
5046copy_out:
5047 err = 0;
5048 if (copy_to_user(arg, file, sizeof(*file)))
5049 err = -EFAULT;
5050out:
5051 kfree(buf);
5052 kfree(file);
5053 return err;
5054}
5055
1da177e4
LT		 5056static int get_disk_info(mddev_t * mddev, void __user * arg)
5057{
5058 mdu_disk_info_t info;
1da177e4
LT		 5059 mdk_rdev_t *rdev;
5060
5061 if (copy_from_user(&info, arg, sizeof(info)))
5062 return -EFAULT;
5063
26ef379f 5064 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5065 if (rdev) {
5066 info.major = MAJOR(rdev->bdev->bd_dev);
5067 info.minor = MINOR(rdev->bdev->bd_dev);
5068 info.raid_disk = rdev->raid_disk;
5069 info.state = 0;
b2d444d7 5070 if (test_bit(Faulty, &rdev->flags))
1da177e4 5071 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5072 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5073 info.state |= (1<<MD_DISK_ACTIVE);
5074 info.state |= (1<<MD_DISK_SYNC);
5075 }
8ddf9efe
N		 5076 if (test_bit(WriteMostly, &rdev->flags))
5077 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5078 } else {
5079 info.major = info.minor = 0;
5080 info.raid_disk = -1;
5081 info.state = (1<<MD_DISK_REMOVED);
5082 }
5083
5084 if (copy_to_user(arg, &info, sizeof(info)))
5085 return -EFAULT;
5086
5087 return 0;
5088}
5089
5090static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5091{
5092 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5093 mdk_rdev_t *rdev;
5094 dev_t dev = MKDEV(info->major,info->minor);
5095
5096 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5097 return -EOVERFLOW;
5098
5099 if (!mddev->raid_disks) {
5100 int err;
5101 /* expecting a device which has a superblock */
5102 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5103 if (IS_ERR(rdev)) {
5104 printk(KERN_WARNING
5105 "md: md_import_device returned %ld\n",
5106 PTR_ERR(rdev));
5107 return PTR_ERR(rdev);
5108 }
5109 if (!list_empty(&mddev->disks)) {
5110 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5111 mdk_rdev_t, same_set);
a9f326eb 5112 err = super_types[mddev->major_version]
1da177e4
LT		 5113 .load_super(rdev, rdev0, mddev->minor_version);
5114 if (err < 0) {
5115 printk(KERN_WARNING
5116 "md: %s has different UUID to %s\n",
5117 bdevname(rdev->bdev,b),
5118 bdevname(rdev0->bdev,b2));
5119 export_rdev(rdev);
5120 return -EINVAL;
5121 }
5122 }
5123 err = bind_rdev_to_array(rdev, mddev);
5124 if (err)
5125 export_rdev(rdev);
5126 return err;
5127 }
5128
5129 /*
5130 * add_new_disk can be used once the array is assembled
5131 * to add "hot spares". They must already have a superblock
5132 * written
5133 */
5134 if (mddev->pers) {
5135 int err;
5136 if (!mddev->pers->hot_add_disk) {
5137 printk(KERN_WARNING
5138 "%s: personality does not support diskops!\n",
5139 mdname(mddev));
5140 return -EINVAL;
5141 }
7b1e35f6
N		 5142 if (mddev->persistent)
5143 rdev = md_import_device(dev, mddev->major_version,
5144 mddev->minor_version);
5145 else
5146 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5147 if (IS_ERR(rdev)) {
5148 printk(KERN_WARNING
5149 "md: md_import_device returned %ld\n",
5150 PTR_ERR(rdev));
5151 return PTR_ERR(rdev);
5152 }
41158c7e
N		 5153 /* set save_raid_disk if appropriate */
5154 if (!mddev->persistent) {
5155 if (info->state & (1<<MD_DISK_SYNC) &&
5156 info->raid_disk < mddev->raid_disks)
5157 rdev->raid_disk = info->raid_disk;
5158 else
5159 rdev->raid_disk = -1;
5160 } else
5161 super_types[mddev->major_version].
5162 validate_super(mddev, rdev);
5163 rdev->saved_raid_disk = rdev->raid_disk;
5164
b2d444d7 5165 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5166 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5167 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5168 else
5169 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5170
1da177e4
LT		 5171 rdev->raid_disk = -1;
5172 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5173 if (!err && !mddev->pers->hot_remove_disk) {
5174 /* If there is hot_add_disk but no hot_remove_disk
5175 * then added disks for geometry changes,
5176 * and should be added immediately.
5177 */
5178 super_types[mddev->major_version].
5179 validate_super(mddev, rdev);
5180 err = mddev->pers->hot_add_disk(mddev, rdev);
5181 if (err)
5182 unbind_rdev_from_array(rdev);
5183 }
1da177e4
LT		 5184 if (err)
5185 export_rdev(rdev);
52664732 5186 else
00bcb4ac 5187 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5188
17571284 5189 md_update_sb(mddev, 1);
72a23c21
NB		 5190 if (mddev->degraded)
5191 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5192 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 5193 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5194 return err;
5195 }
5196
5197 /* otherwise, add_new_disk is only allowed
5198 * for major_version==0 superblocks
5199 */
5200 if (mddev->major_version != 0) {
5201 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5202 mdname(mddev));
5203 return -EINVAL;
5204 }
5205
5206 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5207 int err;
d710e138 5208 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5209 if (IS_ERR(rdev)) {
5210 printk(KERN_WARNING
5211 "md: error, md_import_device() returned %ld\n",
5212 PTR_ERR(rdev));
5213 return PTR_ERR(rdev);
5214 }
5215 rdev->desc_nr = info->number;
5216 if (info->raid_disk < mddev->raid_disks)
5217 rdev->raid_disk = info->raid_disk;
5218 else
5219 rdev->raid_disk = -1;
5220
1da177e4 5221 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5222 if (info->state & (1<<MD_DISK_SYNC))
5223 set_bit(In_sync, &rdev->flags);
1da177e4 5224
8ddf9efe
N		 5225 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5226 set_bit(WriteMostly, &rdev->flags);
5227
1da177e4
LT		 5228 if (!mddev->persistent) {
5229 printk(KERN_INFO "md: nonpersistent superblock ...\n");
0f420358 5230 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
1da177e4 5231 } else
0f420358 5232 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
8190e754 5233 rdev->sectors = rdev->sb_start;
1da177e4 5234
2bf071bf
N		 5235 err = bind_rdev_to_array(rdev, mddev);
5236 if (err) {
5237 export_rdev(rdev);
5238 return err;
5239 }
1da177e4
LT		 5240 }
5241
5242 return 0;
5243}
5244
5245static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5246{
5247 char b[BDEVNAME_SIZE];
5248 mdk_rdev_t *rdev;
5249
1da177e4
LT		 5250 rdev = find_rdev(mddev, dev);
5251 if (!rdev)
5252 return -ENXIO;
5253
5254 if (rdev->raid_disk >= 0)
5255 goto busy;
5256
5257 kick_rdev_from_array(rdev);
850b2b42 5258 md_update_sb(mddev, 1);
d7603b7e 5259 md_new_event(mddev);
1da177e4
LT		 5260
5261 return 0;
5262busy:
fdefa4d8 5263 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5264 bdevname(rdev->bdev,b), mdname(mddev));
5265 return -EBUSY;
5266}
5267
5268static int hot_add_disk(mddev_t * mddev, dev_t dev)
5269{
5270 char b[BDEVNAME_SIZE];
5271 int err;
1da177e4
LT		 5272 mdk_rdev_t *rdev;
5273
5274 if (!mddev->pers)
5275 return -ENODEV;
5276
5277 if (mddev->major_version != 0) {
5278 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5279 " version-0 superblocks.\n",
5280 mdname(mddev));
5281 return -EINVAL;
5282 }
5283 if (!mddev->pers->hot_add_disk) {
5284 printk(KERN_WARNING
5285 "%s: personality does not support diskops!\n",
5286 mdname(mddev));
5287 return -EINVAL;
5288 }
5289
d710e138 5290 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5291 if (IS_ERR(rdev)) {
5292 printk(KERN_WARNING
5293 "md: error, md_import_device() returned %ld\n",
5294 PTR_ERR(rdev));
5295 return -EINVAL;
5296 }
5297
5298 if (mddev->persistent)
0f420358 5299 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1da177e4 5300 else
0f420358 5301 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
1da177e4 5302
8190e754 5303 rdev->sectors = rdev->sb_start;
1da177e4 5304
b2d444d7 5305 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5306 printk(KERN_WARNING
5307 "md: can not hot-add faulty %s disk to %s!\n",
5308 bdevname(rdev->bdev,b), mdname(mddev));
5309 err = -EINVAL;
5310 goto abort_export;
5311 }
b2d444d7 5312 clear_bit(In_sync, &rdev->flags);
1da177e4 5313 rdev->desc_nr = -1;
5842730d 5314 rdev->saved_raid_disk = -1;
2bf071bf
N		 5315 err = bind_rdev_to_array(rdev, mddev);
5316 if (err)
5317 goto abort_export;
1da177e4
LT		 5318
5319 /*
5320 * The rest should better be atomic, we can have disk failures
5321 * noticed in interrupt contexts ...
5322 */
5323
1da177e4
LT		 5324 rdev->raid_disk = -1;
5325
850b2b42 5326 md_update_sb(mddev, 1);
1da177e4
LT		 5327
5328 /*
5329 * Kick recovery, maybe this spare has to be added to the
5330 * array immediately.
5331 */
5332 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5333 md_wakeup_thread(mddev->thread);
d7603b7e 5334 md_new_event(mddev);
1da177e4
LT		 5335 return 0;
5336
1da177e4
LT		 5337abort_export:
5338 export_rdev(rdev);
5339 return err;
5340}
5341
32a7627c
N		 5342static int set_bitmap_file(mddev_t *mddev, int fd)
5343{
5344 int err;
5345
36fa3063
N		 5346 if (mddev->pers) {
5347 if (!mddev->pers->quiesce)
5348 return -EBUSY;
5349 if (mddev->recovery || mddev->sync_thread)
5350 return -EBUSY;
5351 /* we should be able to change the bitmap.. */
5352 }
32a7627c 5353
32a7627c 5354
36fa3063
N		 5355 if (fd >= 0) {
5356 if (mddev->bitmap)
5357 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5358 mddev->bitmap_info.file = fget(fd);
32a7627c 5359
c3d9714e 5360 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5361 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5362 mdname(mddev));
5363 return -EBADF;
5364 }
5365
c3d9714e 5366 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5367 if (err) {
5368 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5369 mdname(mddev));
c3d9714e
N		 5370 fput(mddev->bitmap_info.file);
5371 mddev->bitmap_info.file = NULL;
36fa3063
N		 5372 return err;
5373 }
c3d9714e 5374 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5375 } else if (mddev->bitmap == NULL)
5376 return -ENOENT; /* cannot remove what isn't there */
5377 err = 0;
5378 if (mddev->pers) {
5379 mddev->pers->quiesce(mddev, 1);
69e51b44 5380 if (fd >= 0) {
36fa3063 5381 err = bitmap_create(mddev);
69e51b44
N		 5382 if (!err)
5383 err = bitmap_load(mddev);
5384 }
d7375ab3 5385 if (fd < 0 || err) {
36fa3063 5386 bitmap_destroy(mddev);
d7375ab3
N		 5387 fd = -1; /* make sure to put the file */
5388 }
36fa3063 5389 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5390 }
5391 if (fd < 0) {
c3d9714e
N		 5392 if (mddev->bitmap_info.file) {
5393 restore_bitmap_write_access(mddev->bitmap_info.file);
5394 fput(mddev->bitmap_info.file);
acc55e22 5395 }
c3d9714e 5396 mddev->bitmap_info.file = NULL;
36fa3063
N		 5397 }
5398
32a7627c
N		 5399 return err;
5400}
5401
1da177e4
LT		 5402/*
5403 * set_array_info is used two different ways
5404 * The original usage is when creating a new array.
5405 * In this usage, raid_disks is > 0 and it together with
5406 * level, size, not_persistent,layout,chunksize determine the
5407 * shape of the array.
5408 * This will always create an array with a type-0.90.0 superblock.
5409 * The newer usage is when assembling an array.
5410 * In this case raid_disks will be 0, and the major_version field is
5411 * use to determine which style super-blocks are to be found on the devices.
5412 * The minor and patch _version numbers are also kept incase the
5413 * super_block handler wishes to interpret them.
5414 */
5415static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5416{
5417
5418 if (info->raid_disks == 0) {
5419 /* just setting version number for superblock loading */
5420 if (info->major_version < 0 ||
50511da3 5421 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5422 super_types[info->major_version].name == NULL) {
5423 /* maybe try to auto-load a module? */
5424 printk(KERN_INFO
5425 "md: superblock version %d not known\n",
5426 info->major_version);
5427 return -EINVAL;
5428 }
5429 mddev->major_version = info->major_version;
5430 mddev->minor_version = info->minor_version;
5431 mddev->patch_version = info->patch_version;
3f9d7b0d 5432 mddev->persistent = !info->not_persistent;
cbd19983
N		 5433 /* ensure mddev_put doesn't delete this now that there
5434 * is some minimal configuration.
5435 */
5436 mddev->ctime = get_seconds();
1da177e4
LT		 5437 return 0;
5438 }
5439 mddev->major_version = MD_MAJOR_VERSION;
5440 mddev->minor_version = MD_MINOR_VERSION;
5441 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5442 mddev->ctime = get_seconds();
5443
5444 mddev->level = info->level;
17115e03 5445 mddev->clevel[0] = 0;
58c0fed4 5446 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5447 mddev->raid_disks = info->raid_disks;
5448 /* don't set md_minor, it is determined by which /dev/md* was
5449 * openned
5450 */
5451 if (info->state & (1<<MD_SB_CLEAN))
5452 mddev->recovery_cp = MaxSector;
5453 else
5454 mddev->recovery_cp = 0;
5455 mddev->persistent = ! info->not_persistent;
e691063a 5456 mddev->external = 0;
1da177e4
LT		 5457
5458 mddev->layout = info->layout;
9d8f0363 5459 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5460
5461 mddev->max_disks = MD_SB_DISKS;
5462
e691063a
N		 5463 if (mddev->persistent)
5464 mddev->flags = 0;
850b2b42 5465 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5466
c3d9714e
N		 5467 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5468 mddev->bitmap_info.offset = 0;
b2a2703c 5469
f6705578
N		 5470 mddev->reshape_position = MaxSector;
5471
1da177e4
LT		 5472 /*
5473 * Generate a 128 bit UUID
5474 */
5475 get_random_bytes(mddev->uuid, 16);
5476
f6705578 5477 mddev->new_level = mddev->level;
664e7c41 5478 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5479 mddev->new_layout = mddev->layout;
5480 mddev->delta_disks = 0;
5481
1da177e4
LT		 5482 return 0;
5483}
5484
1f403624
DW		 5485void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5486{
b522adcd
DW		 5487 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5488
5489 if (mddev->external_size)
5490 return;
5491
1f403624
DW		 5492 mddev->array_sectors = array_sectors;
5493}
5494EXPORT_SYMBOL(md_set_array_sectors);
5495
d71f9f88 5496static int update_size(mddev_t *mddev, sector_t num_sectors)
a35b0d69 5497{
159ec1fc 5498 mdk_rdev_t *rdev;
a35b0d69 5499 int rv;
d71f9f88 5500 int fit = (num_sectors == 0);
a35b0d69
N		 5501
5502 if (mddev->pers->resize == NULL)
5503 return -EINVAL;
d71f9f88
AN		 5504 /* The "num_sectors" is the number of sectors of each device that
5505 * is used. This can only make sense for arrays with redundancy.
5506 * linear and raid0 always use whatever space is available. We can only
5507 * consider changing this number if no resync or reconstruction is
5508 * happening, and if the new size is acceptable. It must fit before the
0f420358 5509 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5510 * of each device. If num_sectors is zero, we find the largest size
5511 * that fits.
5512
a35b0d69
N		 5513 */
5514 if (mddev->sync_thread)
5515 return -EBUSY;
dba034ee
N		 5516 if (mddev->bitmap)
5517 /* Sorry, cannot grow a bitmap yet, just remove it,
5518 * grow, and re-add.
5519 */
5520 return -EBUSY;
159ec1fc 5521 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5522 sector_t avail = rdev->sectors;
01ab5662 5523
d71f9f88
AN		 5524 if (fit && (num_sectors == 0 || num_sectors > avail))
5525 num_sectors = avail;
5526 if (avail < num_sectors)
a35b0d69
N		 5527 return -ENOSPC;
5528 }
d71f9f88 5529 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5530 if (!rv)
5531 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5532 return rv;
5533}
5534
da943b99
N		 5535static int update_raid_disks(mddev_t *mddev, int raid_disks)
5536{
5537 int rv;
5538 /* change the number of raid disks */
63c70c4f 5539 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5540 return -EINVAL;
5541 if (raid_disks <= 0 ||
233fca36 5542 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5543 return -EINVAL;
63c70c4f 5544 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5545 return -EBUSY;
63c70c4f
N		 5546 mddev->delta_disks = raid_disks - mddev->raid_disks;
5547
5548 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 5549 return rv;
5550}
5551
5552
1da177e4
LT		 5553/*
5554 * update_array_info is used to change the configuration of an
5555 * on-line array.
5556 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5557 * fields in the info are checked against the array.
5558 * Any differences that cannot be handled will cause an error.
5559 * Normally, only one change can be managed at a time.
5560 */
5561static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5562{
5563 int rv = 0;
5564 int cnt = 0;
36fa3063
N		 5565 int state = 0;
5566
5567 /* calculate expected state,ignoring low bits */
c3d9714e 5568 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5569 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5570
5571 if (mddev->major_version != info->major_version ||
5572 mddev->minor_version != info->minor_version ||
5573/* mddev->patch_version != info->patch_version || */
5574 mddev->ctime != info->ctime ||
5575 mddev->level != info->level ||
5576/* mddev->layout != info->layout || */
5577 !mddev->persistent != info->not_persistent||
9d8f0363 5578 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5579 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5580 ((state^info->state) & 0xfffffe00)
5581 )
1da177e4
LT		 5582 return -EINVAL;
5583 /* Check there is only one change */
58c0fed4
AN		 5584 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5585 cnt++;
5586 if (mddev->raid_disks != info->raid_disks)
5587 cnt++;
5588 if (mddev->layout != info->layout)
5589 cnt++;
5590 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5591 cnt++;
5592 if (cnt == 0)
5593 return 0;
5594 if (cnt > 1)
5595 return -EINVAL;
1da177e4
LT		 5596
5597 if (mddev->layout != info->layout) {
5598 /* Change layout
5599 * we don't need to do anything at the md level, the
5600 * personality will take care of it all.
5601 */
50ac168a 5602 if (mddev->pers->check_reshape == NULL)
1da177e4 5603 return -EINVAL;
597a711b
N		 5604 else {
5605 mddev->new_layout = info->layout;
50ac168a 5606 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5607 if (rv)
5608 mddev->new_layout = mddev->layout;
5609 return rv;
5610 }
1da177e4 5611 }
58c0fed4 5612 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5613 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5614
da943b99
N		 5615 if (mddev->raid_disks != info->raid_disks)
5616 rv = update_raid_disks(mddev, info->raid_disks);
5617
36fa3063
N		 5618 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5619 if (mddev->pers->quiesce == NULL)
5620 return -EINVAL;
5621 if (mddev->recovery || mddev->sync_thread)
5622 return -EBUSY;
5623 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5624 /* add the bitmap */
5625 if (mddev->bitmap)
5626 return -EEXIST;
c3d9714e 5627 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5628 return -EINVAL;
c3d9714e
N		 5629 mddev->bitmap_info.offset =
5630 mddev->bitmap_info.default_offset;
36fa3063
N		 5631 mddev->pers->quiesce(mddev, 1);
5632 rv = bitmap_create(mddev);
69e51b44
N		 5633 if (!rv)
5634 rv = bitmap_load(mddev);
36fa3063
N		 5635 if (rv)
5636 bitmap_destroy(mddev);
5637 mddev->pers->quiesce(mddev, 0);
5638 } else {
5639 /* remove the bitmap */
5640 if (!mddev->bitmap)
5641 return -ENOENT;
5642 if (mddev->bitmap->file)
5643 return -EINVAL;
5644 mddev->pers->quiesce(mddev, 1);
5645 bitmap_destroy(mddev);
5646 mddev->pers->quiesce(mddev, 0);
c3d9714e 5647 mddev->bitmap_info.offset = 0;
36fa3063
N		 5648 }
5649 }
850b2b42 5650 md_update_sb(mddev, 1);
1da177e4
LT		 5651 return rv;
5652}
5653
5654static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5655{
5656 mdk_rdev_t *rdev;
5657
5658 if (mddev->pers == NULL)
5659 return -ENODEV;
5660
5661 rdev = find_rdev(mddev, dev);
5662 if (!rdev)
5663 return -ENODEV;
5664
5665 md_error(mddev, rdev);
5666 return 0;
5667}
5668
2f9618ce
AN		 5669/*
5670 * We have a problem here : there is no easy way to give a CHS
5671 * virtual geometry. We currently pretend that we have a 2 heads
5672 * 4 sectors (with a BIG number of cylinders...). This drives
5673 * dosfs just mad... ;-)
5674 */
a885c8c4
CH		 5675static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5676{
5677 mddev_t *mddev = bdev->bd_disk->private_data;
5678
5679 geo->heads = 2;
5680 geo->sectors = 4;
49ce6cea 5681 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 5682 return 0;
5683}
5684
a39907fa 5685static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 5686 unsigned int cmd, unsigned long arg)
5687{
5688 int err = 0;
5689 void __user *argp = (void __user *)arg;
1da177e4 5690 mddev_t *mddev = NULL;
e2218350 5691 int ro;
1da177e4
LT		 5692
5693 if (!capable(CAP_SYS_ADMIN))
5694 return -EACCES;
5695
5696 /*
5697 * Commands dealing with the RAID driver but not any
5698 * particular array:
5699 */
5700 switch (cmd)
5701 {
5702 case RAID_VERSION:
5703 err = get_version(argp);
5704 goto done;
5705
5706 case PRINT_RAID_DEBUG:
5707 err = 0;
5708 md_print_devices();
5709 goto done;
5710
5711#ifndef MODULE
5712 case RAID_AUTORUN:
5713 err = 0;
5714 autostart_arrays(arg);
5715 goto done;
5716#endif
5717 default:;
5718 }
5719
5720 /*
5721 * Commands creating/starting a new array:
5722 */
5723
a39907fa 5724 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 5725
5726 if (!mddev) {
5727 BUG();
5728 goto abort;
5729 }
5730
1da177e4
LT		 5731 err = mddev_lock(mddev);
5732 if (err) {
5733 printk(KERN_INFO
5734 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5735 err, cmd);
5736 goto abort;
5737 }
5738
5739 switch (cmd)
5740 {
5741 case SET_ARRAY_INFO:
5742 {
5743 mdu_array_info_t info;
5744 if (!arg)
5745 memset(&info, 0, sizeof(info));
5746 else if (copy_from_user(&info, argp, sizeof(info))) {
5747 err = -EFAULT;
5748 goto abort_unlock;
5749 }
5750 if (mddev->pers) {
5751 err = update_array_info(mddev, &info);
5752 if (err) {
5753 printk(KERN_WARNING "md: couldn't update"
5754 " array info. %d\n", err);
5755 goto abort_unlock;
5756 }
5757 goto done_unlock;
5758 }
5759 if (!list_empty(&mddev->disks)) {
5760 printk(KERN_WARNING
5761 "md: array %s already has disks!\n",
5762 mdname(mddev));
5763 err = -EBUSY;
5764 goto abort_unlock;
5765 }
5766 if (mddev->raid_disks) {
5767 printk(KERN_WARNING
5768 "md: array %s already initialised!\n",
5769 mdname(mddev));
5770 err = -EBUSY;
5771 goto abort_unlock;
5772 }
5773 err = set_array_info(mddev, &info);
5774 if (err) {
5775 printk(KERN_WARNING "md: couldn't set"
5776 " array info. %d\n", err);
5777 goto abort_unlock;
5778 }
5779 }
5780 goto done_unlock;
5781
5782 default:;
5783 }
5784
5785 /*
5786 * Commands querying/configuring an existing array:
5787 */
32a7627c 5788 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 5789 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 5790 if ((!mddev->raid_disks && !mddev->external)
5791 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5792 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5793 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 5794 err = -ENODEV;
5795 goto abort_unlock;
5796 }
5797
5798 /*
5799 * Commands even a read-only array can execute:
5800 */
5801 switch (cmd)
5802 {
5803 case GET_ARRAY_INFO:
5804 err = get_array_info(mddev, argp);
5805 goto done_unlock;
5806
32a7627c 5807 case GET_BITMAP_FILE:
87162a28 5808 err = get_bitmap_file(mddev, argp);
32a7627c
N		 5809 goto done_unlock;
5810
1da177e4
LT		 5811 case GET_DISK_INFO:
5812 err = get_disk_info(mddev, argp);
5813 goto done_unlock;
5814
5815 case RESTART_ARRAY_RW:
5816 err = restart_array(mddev);
5817 goto done_unlock;
5818
5819 case STOP_ARRAY:
d710e138 5820 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 5821 goto done_unlock;
5822
5823 case STOP_ARRAY_RO:
a4bd82d0 5824 err = md_set_readonly(mddev, 1);
1da177e4
LT		 5825 goto done_unlock;
5826
e2218350
DW		 5827 case BLKROSET:
5828 if (get_user(ro, (int __user *)(arg))) {
5829 err = -EFAULT;
5830 goto done_unlock;
5831 }
5832 err = -EINVAL;
5833
5834 /* if the bdev is going readonly the value of mddev->ro
5835 * does not matter, no writes are coming
5836 */
5837 if (ro)
5838 goto done_unlock;
5839
5840 /* are we are already prepared for writes? */
5841 if (mddev->ro != 1)
5842 goto done_unlock;
5843
5844 /* transitioning to readauto need only happen for
5845 * arrays that call md_write_start
5846 */
5847 if (mddev->pers) {
5848 err = restart_array(mddev);
5849 if (err == 0) {
5850 mddev->ro = 2;
5851 set_disk_ro(mddev->gendisk, 0);
5852 }
5853 }
5854 goto done_unlock;
1da177e4
LT		 5855 }
5856
5857 /*
5858 * The remaining ioctls are changing the state of the
f91de92e
N		 5859 * superblock, so we do not allow them on read-only arrays.
5860 * However non-MD ioctls (e.g. get-size) will still come through
5861 * here and hit the 'default' below, so only disallow
5862 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 5863 */
bb57fc64 5864 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 5865 if (mddev->ro == 2) {
5866 mddev->ro = 0;
00bcb4ac 5867 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 5868 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5869 md_wakeup_thread(mddev->thread);
f91de92e
N		 5870 } else {
5871 err = -EROFS;
5872 goto abort_unlock;
5873 }
1da177e4
LT		 5874 }
5875
5876 switch (cmd)
5877 {
5878 case ADD_NEW_DISK:
5879 {
5880 mdu_disk_info_t info;
5881 if (copy_from_user(&info, argp, sizeof(info)))
5882 err = -EFAULT;
5883 else
5884 err = add_new_disk(mddev, &info);
5885 goto done_unlock;
5886 }
5887
5888 case HOT_REMOVE_DISK:
5889 err = hot_remove_disk(mddev, new_decode_dev(arg));
5890 goto done_unlock;
5891
5892 case HOT_ADD_DISK:
5893 err = hot_add_disk(mddev, new_decode_dev(arg));
5894 goto done_unlock;
5895
5896 case SET_DISK_FAULTY:
5897 err = set_disk_faulty(mddev, new_decode_dev(arg));
5898 goto done_unlock;
5899
5900 case RUN_ARRAY:
d710e138 5901 err = do_md_run(mddev);
1da177e4
LT		 5902 goto done_unlock;
5903
32a7627c
N		 5904 case SET_BITMAP_FILE:
5905 err = set_bitmap_file(mddev, (int)arg);
5906 goto done_unlock;
5907
1da177e4 5908 default:
1da177e4
LT		 5909 err = -EINVAL;
5910 goto abort_unlock;
5911 }
5912
5913done_unlock:
5914abort_unlock:
d3374825
N		 5915 if (mddev->hold_active == UNTIL_IOCTL &&
5916 err != -EINVAL)
5917 mddev->hold_active = 0;
1da177e4
LT		 5918 mddev_unlock(mddev);
5919
5920 return err;
5921done:
5922 if (err)
5923 MD_BUG();
5924abort:
5925 return err;
5926}
aa98aa31
AB		 5927#ifdef CONFIG_COMPAT
5928static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5929 unsigned int cmd, unsigned long arg)
5930{
5931 switch (cmd) {
5932 case HOT_REMOVE_DISK:
5933 case HOT_ADD_DISK:
5934 case SET_DISK_FAULTY:
5935 case SET_BITMAP_FILE:
5936 /* These take in integer arg, do not convert */
5937 break;
5938 default:
5939 arg = (unsigned long)compat_ptr(arg);
5940 break;
5941 }
5942
5943 return md_ioctl(bdev, mode, cmd, arg);
5944}
5945#endif /* CONFIG_COMPAT */
1da177e4 5946
a39907fa 5947static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 5948{
5949 /*
5950 * Succeed if we can lock the mddev, which confirms that
5951 * it isn't being stopped right now.
5952 */
d3374825 5953 mddev_t *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 5954 int err;
5955
d3374825
N		 5956 if (mddev->gendisk != bdev->bd_disk) {
5957 /* we are racing with mddev_put which is discarding this
5958 * bd_disk.
5959 */
5960 mddev_put(mddev);
5961 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 5962 flush_workqueue(md_misc_wq);
d3374825
N		 5963 /* Then retry the open from the top */
5964 return -ERESTARTSYS;
5965 }
5966 BUG_ON(mddev != bdev->bd_disk->private_data);
5967
c8c00a69 5968 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 5969 goto out;
5970
5971 err = 0;
f2ea68cf 5972 atomic_inc(&mddev->openers);
c8c00a69 5973 mutex_unlock(&mddev->open_mutex);
1da177e4 5974
f3b99be1 5975 check_disk_size_change(mddev->gendisk, bdev);
1da177e4
LT		 5976 out:
5977 return err;
5978}
5979
a39907fa 5980static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 5981{
a39907fa 5982 mddev_t *mddev = disk->private_data;
1da177e4 5983
52e5f9d1 5984 BUG_ON(!mddev);
f2ea68cf 5985 atomic_dec(&mddev->openers);
1da177e4
LT		 5986 mddev_put(mddev);
5987
5988 return 0;
5989}
83d5cde4 5990static const struct block_device_operations md_fops =
1da177e4
LT		 5991{
5992 .owner = THIS_MODULE,
a39907fa
AV		 5993 .open = md_open,
5994 .release = md_release,
b492b852 5995 .ioctl = md_ioctl,
aa98aa31
AB		 5996#ifdef CONFIG_COMPAT
5997 .compat_ioctl = md_compat_ioctl,
5998#endif
a885c8c4 5999 .getgeo = md_getgeo,
1da177e4
LT		 6000};
6001
75c96f85 6002static int md_thread(void * arg)
1da177e4
LT		 6003{
6004 mdk_thread_t *thread = arg;
6005
1da177e4
LT		 6006 /*
6007 * md_thread is a 'system-thread', it's priority should be very
6008 * high. We avoid resource deadlocks individually in each
6009 * raid personality. (RAID5 does preallocation) We also use RR and
6010 * the very same RT priority as kswapd, thus we will never get
6011 * into a priority inversion deadlock.
6012 *
6013 * we definitely have to have equal or higher priority than
6014 * bdflush, otherwise bdflush will deadlock if there are too
6015 * many dirty RAID5 blocks.
6016 */
1da177e4 6017
6985c43f 6018 allow_signal(SIGKILL);
a6fb0934 6019 while (!kthread_should_stop()) {
1da177e4 6020
93588e22
N		 6021 /* We need to wait INTERRUPTIBLE so that
6022 * we don't add to the load-average.
6023 * That means we need to be sure no signals are
6024 * pending
6025 */
6026 if (signal_pending(current))
6027 flush_signals(current);
6028
6029 wait_event_interruptible_timeout
6030 (thread->wqueue,
6031 test_bit(THREAD_WAKEUP, &thread->flags)
6032 || kthread_should_stop(),
6033 thread->timeout);
1da177e4
LT		 6034
6035 clear_bit(THREAD_WAKEUP, &thread->flags);
6036
787453c2 6037 thread->run(thread->mddev);
1da177e4 6038 }
a6fb0934 6039
1da177e4
LT		 6040 return 0;
6041}
6042
6043void md_wakeup_thread(mdk_thread_t *thread)
6044{
6045 if (thread) {
6046 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6047 set_bit(THREAD_WAKEUP, &thread->flags);
6048 wake_up(&thread->wqueue);
6049 }
6050}
6051
6052mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6053 const char *name)
6054{
6055 mdk_thread_t *thread;
1da177e4 6056
9ffae0cf 6057 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 6058 if (!thread)
6059 return NULL;
6060
1da177e4
LT		 6061 init_waitqueue_head(&thread->wqueue);
6062
1da177e4
LT		 6063 thread->run = run;
6064 thread->mddev = mddev;
32a7627c 6065 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6066 thread->tsk = kthread_run(md_thread, thread,
6067 "%s_%s",
6068 mdname(thread->mddev),
6069 name ?: mddev->pers->name);
a6fb0934 6070 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6071 kfree(thread);
6072 return NULL;
6073 }
1da177e4
LT		 6074 return thread;
6075}
6076
1da177e4
LT		 6077void md_unregister_thread(mdk_thread_t *thread)
6078{
e0cf8f04
N		 6079 if (!thread)
6080 return;
ba25f9dc 6081 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 6082
6083 kthread_stop(thread->tsk);
1da177e4
LT		 6084 kfree(thread);
6085}
6086
6087void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6088{
6089 if (!mddev) {
6090 MD_BUG();
6091 return;
6092 }
6093
b2d444d7 6094 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6095 return;
6bfe0b49
DW		 6096
6097 if (mddev->external)
6098 set_bit(Blocked, &rdev->flags);
32a7627c 6099/*
1da177e4
LT		 6100 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6101 mdname(mddev),
6102 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6103 __builtin_return_address(0),__builtin_return_address(1),
6104 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 6105*/
d0a0a5ee
AM		 6106 if (!mddev->pers)
6107 return;
1da177e4
LT		 6108 if (!mddev->pers->error_handler)
6109 return;
6110 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6111 if (mddev->degraded)
6112 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6113 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6114 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6115 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6116 md_wakeup_thread(mddev->thread);
768a418d 6117 if (mddev->event_work.func)
e804ac78 6118 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6119 md_new_event_inintr(mddev);
1da177e4
LT		 6120}
6121
6122/* seq_file implementation /proc/mdstat */
6123
6124static void status_unused(struct seq_file *seq)
6125{
6126 int i = 0;
6127 mdk_rdev_t *rdev;
1da177e4
LT		 6128
6129 seq_printf(seq, "unused devices: ");
6130
159ec1fc 6131 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6132 char b[BDEVNAME_SIZE];
6133 i++;
6134 seq_printf(seq, "%s ",
6135 bdevname(rdev->bdev,b));
6136 }
6137 if (!i)
6138 seq_printf(seq, "<none>");
6139
6140 seq_printf(seq, "\n");
6141}
6142
6143
6144static void status_resync(struct seq_file *seq, mddev_t * mddev)
6145{
dd71cf6b
N		 6146 sector_t max_sectors, resync, res;
6147 unsigned long dt, db;
6148 sector_t rt;
4588b42e
N		 6149 int scale;
6150 unsigned int per_milli;
1da177e4 6151
dd71cf6b 6152 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6153
6154 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6155 max_sectors = mddev->resync_max_sectors;
1da177e4 6156 else
dd71cf6b 6157 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6158
6159 /*
6160 * Should not happen.
6161 */
dd71cf6b 6162 if (!max_sectors) {
1da177e4
LT		 6163 MD_BUG();
6164 return;
6165 }
4588b42e 6166 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6167 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6168 * u32, as those are the requirements for sector_div.
6169 * Thus 'scale' must be at least 10
6170 */
6171 scale = 10;
6172 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6173 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6174 scale++;
6175 }
6176 res = (resync>>scale)*1000;
dd71cf6b 6177 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6178
6179 per_milli = res;
1da177e4 6180 {
4588b42e 6181 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6182 seq_printf(seq, "[");
6183 for (i = 0; i < x; i++)
6184 seq_printf(seq, "=");
6185 seq_printf(seq, ">");
6186 for (i = 0; i < y; i++)
6187 seq_printf(seq, ".");
6188 seq_printf(seq, "] ");
6189 }
4588b42e 6190 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6191 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6192 "reshape" :
61df9d91
N		 6193 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6194 "check" :
6195 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6196 "resync" : "recovery"))),
6197 per_milli/10, per_milli % 10,
dd71cf6b
N		 6198 (unsigned long long) resync/2,
6199 (unsigned long long) max_sectors/2);
1da177e4
LT		 6200
6201 /*
1da177e4
LT		 6202 * dt: time from mark until now
6203 * db: blocks written from mark until now
6204 * rt: remaining time
dd71cf6b
N		 6205 *
6206 * rt is a sector_t, so could be 32bit or 64bit.
6207 * So we divide before multiply in case it is 32bit and close
6208 * to the limit.
6209 * We scale the divisor (db) by 32 to avoid loosing precision
6210 * near the end of resync when the number of remaining sectors
6211 * is close to 'db'.
6212 * We then divide rt by 32 after multiplying by db to compensate.
6213 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6214 */
6215 dt = ((jiffies - mddev->resync_mark) / HZ);
6216 if (!dt) dt++;
ff4e8d9a
N		 6217 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6218 - mddev->resync_mark_cnt;
1da177e4 6219
dd71cf6b
N		 6220 rt = max_sectors - resync; /* number of remaining sectors */
6221 sector_div(rt, db/32+1);
6222 rt *= dt;
6223 rt >>= 5;
6224
6225 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6226 ((unsigned long)rt % 60)/6);
1da177e4 6227
ff4e8d9a 6228 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6229}
6230
6231static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6232{
6233 struct list_head *tmp;
6234 loff_t l = *pos;
6235 mddev_t *mddev;
6236
6237 if (l >= 0x10000)
6238 return NULL;
6239 if (!l--)
6240 /* header */
6241 return (void*)1;
6242
6243 spin_lock(&all_mddevs_lock);
6244 list_for_each(tmp,&all_mddevs)
6245 if (!l--) {
6246 mddev = list_entry(tmp, mddev_t, all_mddevs);
6247 mddev_get(mddev);
6248 spin_unlock(&all_mddevs_lock);
6249 return mddev;
6250 }
6251 spin_unlock(&all_mddevs_lock);
6252 if (!l--)
6253 return (void*)2;/* tail */
6254 return NULL;
6255}
6256
6257static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6258{
6259 struct list_head *tmp;
6260 mddev_t *next_mddev, *mddev = v;
6261
6262 ++*pos;
6263 if (v == (void*)2)
6264 return NULL;
6265
6266 spin_lock(&all_mddevs_lock);
6267 if (v == (void*)1)
6268 tmp = all_mddevs.next;
6269 else
6270 tmp = mddev->all_mddevs.next;
6271 if (tmp != &all_mddevs)
6272 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6273 else {
6274 next_mddev = (void*)2;
6275 *pos = 0x10000;
6276 }
6277 spin_unlock(&all_mddevs_lock);
6278
6279 if (v != (void*)1)
6280 mddev_put(mddev);
6281 return next_mddev;
6282
6283}
6284
6285static void md_seq_stop(struct seq_file *seq, void *v)
6286{
6287 mddev_t *mddev = v;
6288
6289 if (mddev && v != (void*)1 && v != (void*)2)
6290 mddev_put(mddev);
6291}
6292
d7603b7e
N		 6293struct mdstat_info {
6294 int event;
6295};
6296
1da177e4
LT		 6297static int md_seq_show(struct seq_file *seq, void *v)
6298{
6299 mddev_t *mddev = v;
dd8ac336 6300 sector_t sectors;
1da177e4 6301 mdk_rdev_t *rdev;
d7603b7e 6302 struct mdstat_info *mi = seq->private;
32a7627c 6303 struct bitmap *bitmap;
1da177e4
LT		 6304
6305 if (v == (void*)1) {
2604b703 6306 struct mdk_personality *pers;
1da177e4
LT		 6307 seq_printf(seq, "Personalities : ");
6308 spin_lock(&pers_lock);
2604b703
N		 6309 list_for_each_entry(pers, &pers_list, list)
6310 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6311
6312 spin_unlock(&pers_lock);
6313 seq_printf(seq, "\n");
d7603b7e 6314 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 6315 return 0;
6316 }
6317 if (v == (void*)2) {
6318 status_unused(seq);
6319 return 0;
6320 }
6321
5dc5cf7d 6322 if (mddev_lock(mddev) < 0)
1da177e4 6323 return -EINTR;
5dc5cf7d 6324
1da177e4
LT		 6325 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6326 seq_printf(seq, "%s : %sactive", mdname(mddev),
6327 mddev->pers ? "" : "in");
6328 if (mddev->pers) {
f91de92e 6329 if (mddev->ro==1)
1da177e4 6330 seq_printf(seq, " (read-only)");
f91de92e 6331 if (mddev->ro==2)
52720ae7 6332 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6333 seq_printf(seq, " %s", mddev->pers->name);
6334 }
6335
dd8ac336 6336 sectors = 0;
159ec1fc 6337 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6338 char b[BDEVNAME_SIZE];
6339 seq_printf(seq, " %s[%d]",
6340 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6341 if (test_bit(WriteMostly, &rdev->flags))
6342 seq_printf(seq, "(W)");
b2d444d7 6343 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6344 seq_printf(seq, "(F)");
6345 continue;
b325a32e
N		 6346 } else if (rdev->raid_disk < 0)
6347 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6348 sectors += rdev->sectors;
1da177e4
LT		 6349 }
6350
6351 if (!list_empty(&mddev->disks)) {
6352 if (mddev->pers)
6353 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6354 (unsigned long long)
6355 mddev->array_sectors / 2);
1da177e4
LT		 6356 else
6357 seq_printf(seq, "\n %llu blocks",
dd8ac336 6358 (unsigned long long)sectors / 2);
1da177e4 6359 }
1cd6bf19
N		 6360 if (mddev->persistent) {
6361 if (mddev->major_version != 0 ||
6362 mddev->minor_version != 90) {
6363 seq_printf(seq," super %d.%d",
6364 mddev->major_version,
6365 mddev->minor_version);
6366 }
e691063a
N		 6367 } else if (mddev->external)
6368 seq_printf(seq, " super external:%s",
6369 mddev->metadata_type);
6370 else
1cd6bf19 6371 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6372
6373 if (mddev->pers) {
d710e138 6374 mddev->pers->status(seq, mddev);
1da177e4 6375 seq_printf(seq, "\n ");
8e1b39d6
N		 6376 if (mddev->pers->sync_request) {
6377 if (mddev->curr_resync > 2) {
d710e138 6378 status_resync(seq, mddev);
8e1b39d6
N		 6379 seq_printf(seq, "\n ");
6380 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6381 seq_printf(seq, "\tresync=DELAYED\n ");
6382 else if (mddev->recovery_cp < MaxSector)
6383 seq_printf(seq, "\tresync=PENDING\n ");
6384 }
32a7627c
N		 6385 } else
6386 seq_printf(seq, "\n ");
6387
6388 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6389 unsigned long chunk_kb;
6390 unsigned long flags;
32a7627c 6391 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6392 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6393 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6394 "%lu%s chunk",
6395 bitmap->pages - bitmap->missing_pages,
6396 bitmap->pages,
6397 (bitmap->pages - bitmap->missing_pages)
6398 << (PAGE_SHIFT - 10),
42a04b50 6399 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6400 chunk_kb ? "KB" : "B");
78d742d8
N		 6401 if (bitmap->file) {
6402 seq_printf(seq, ", file: ");
c32c2f63 6403 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6404 }
78d742d8 6405
32a7627c
N		 6406 seq_printf(seq, "\n");
6407 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6408 }
6409
6410 seq_printf(seq, "\n");
6411 }
6412 mddev_unlock(mddev);
6413
6414 return 0;
6415}
6416
110518bc 6417static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6418 .start = md_seq_start,
6419 .next = md_seq_next,
6420 .stop = md_seq_stop,
6421 .show = md_seq_show,
6422};
6423
6424static int md_seq_open(struct inode *inode, struct file *file)
6425{
6426 int error;
d7603b7e
N		 6427 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6428 if (mi == NULL)
6429 return -ENOMEM;
1da177e4
LT		 6430
6431 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 6432 if (error)
6433 kfree(mi);
6434 else {
6435 struct seq_file *p = file->private_data;
6436 p->private = mi;
6437 mi->event = atomic_read(&md_event_count);
6438 }
1da177e4
LT		 6439 return error;
6440}
6441
d7603b7e
N		 6442static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6443{
6444 struct seq_file *m = filp->private_data;
6445 struct mdstat_info *mi = m->private;
6446 int mask;
6447
6448 poll_wait(filp, &md_event_waiters, wait);
6449
6450 /* always allow read */
6451 mask = POLLIN | POLLRDNORM;
6452
6453 if (mi->event != atomic_read(&md_event_count))
6454 mask |= POLLERR | POLLPRI;
6455 return mask;
6456}
6457
fa027c2a 6458static const struct file_operations md_seq_fops = {
e24650c2 6459 .owner = THIS_MODULE,
1da177e4
LT		 6460 .open = md_seq_open,
6461 .read = seq_read,
6462 .llseek = seq_lseek,
c3f94b40 6463 .release = seq_release_private,
d7603b7e 6464 .poll = mdstat_poll,
1da177e4
LT		 6465};
6466
2604b703 6467int register_md_personality(struct mdk_personality *p)
1da177e4 6468{
1da177e4 6469 spin_lock(&pers_lock);
2604b703
N		 6470 list_add_tail(&p->list, &pers_list);
6471 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6472 spin_unlock(&pers_lock);
6473 return 0;
6474}
6475
2604b703 6476int unregister_md_personality(struct mdk_personality *p)
1da177e4 6477{
2604b703 6478 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6479 spin_lock(&pers_lock);
2604b703 6480 list_del_init(&p->list);
1da177e4
LT		 6481 spin_unlock(&pers_lock);
6482 return 0;
6483}
6484
eea1bf38 6485static int is_mddev_idle(mddev_t *mddev, int init)
1da177e4
LT		 6486{
6487 mdk_rdev_t * rdev;
1da177e4 6488 int idle;
eea1bf38 6489 int curr_events;
1da177e4
LT		 6490
6491 idle = 1;
4b80991c
N		 6492 rcu_read_lock();
6493 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6494 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6495 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6496 (int)part_stat_read(&disk->part0, sectors[1]) -
6497 atomic_read(&disk->sync_io);
713f6ab1
N		 6498 /* sync IO will cause sync_io to increase before the disk_stats
6499 * as sync_io is counted when a request starts, and
6500 * disk_stats is counted when it completes.
6501 * So resync activity will cause curr_events to be smaller than
6502 * when there was no such activity.
6503 * non-sync IO will cause disk_stat to increase without
6504 * increasing sync_io so curr_events will (eventually)
6505 * be larger than it was before. Once it becomes
6506 * substantially larger, the test below will cause
6507 * the array to appear non-idle, and resync will slow
6508 * down.
6509 * If there is a lot of outstanding resync activity when
6510 * we set last_event to curr_events, then all that activity
6511 * completing might cause the array to appear non-idle
6512 * and resync will be slowed down even though there might
6513 * not have been non-resync activity. This will only
6514 * happen once though. 'last_events' will soon reflect
6515 * the state where there is little or no outstanding
6516 * resync requests, and further resync activity will
6517 * always make curr_events less than last_events.
c0e48521 6518 *
1da177e4 6519 */
eea1bf38 6520 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6521 rdev->last_events = curr_events;
6522 idle = 0;
6523 }
6524 }
4b80991c 6525 rcu_read_unlock();
1da177e4
LT		 6526 return idle;
6527}
6528
6529void md_done_sync(mddev_t *mddev, int blocks, int ok)
6530{
6531 /* another "blocks" (512byte) blocks have been synced */
6532 atomic_sub(blocks, &mddev->recovery_active);
6533 wake_up(&mddev->recovery_wait);
6534 if (!ok) {
dfc70645 6535 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6536 md_wakeup_thread(mddev->thread);
6537 // stop recovery, signal do_sync ....
6538 }
6539}
6540
6541
06d91a5f
N		 6542/* md_write_start(mddev, bi)
6543 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6544 * in superblock) before writing, schedule a superblock update
6545 * and wait for it to complete.
06d91a5f 6546 */
3d310eb7 6547void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 6548{
0fd62b86 6549 int did_change = 0;
06d91a5f 6550 if (bio_data_dir(bi) != WRITE)
3d310eb7 6551 return;
06d91a5f 6552
f91de92e
N		 6553 BUG_ON(mddev->ro == 1);
6554 if (mddev->ro == 2) {
6555 /* need to switch to read/write */
6556 mddev->ro = 0;
6557 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6558 md_wakeup_thread(mddev->thread);
25156198 6559 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6560 did_change = 1;
f91de92e 6561 }
06d91a5f 6562 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6563 if (mddev->safemode == 1)
6564 mddev->safemode = 0;
06d91a5f 6565 if (mddev->in_sync) {
a9701a30 6566 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6567 if (mddev->in_sync) {
6568 mddev->in_sync = 0;
850b2b42 6569 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6570 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6571 md_wakeup_thread(mddev->thread);
0fd62b86 6572 did_change = 1;
3d310eb7 6573 }
a9701a30 6574 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6575 }
0fd62b86 6576 if (did_change)
00bcb4ac 6577 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6578 wait_event(mddev->sb_wait,
09a44cc1 6579 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6580}
6581
6582void md_write_end(mddev_t *mddev)
6583{
6584 if (atomic_dec_and_test(&mddev->writes_pending)) {
6585 if (mddev->safemode == 2)
6586 md_wakeup_thread(mddev->thread);
16f17b39 6587 else if (mddev->safemode_delay)
1da177e4
LT		 6588 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6589 }
6590}
6591
2a2275d6
N		 6592/* md_allow_write(mddev)
6593 * Calling this ensures that the array is marked 'active' so that writes
6594 * may proceed without blocking. It is important to call this before
6595 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6596 * Must be called with mddev_lock held.
b5470dc5
DW		 6597 *
6598 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6599 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6600 */
b5470dc5 6601int md_allow_write(mddev_t *mddev)
2a2275d6
N		 6602{
6603 if (!mddev->pers)
b5470dc5 6604 return 0;
2a2275d6 6605 if (mddev->ro)
b5470dc5 6606 return 0;
1a0fd497 6607 if (!mddev->pers->sync_request)
b5470dc5 6608 return 0;
2a2275d6
N		 6609
6610 spin_lock_irq(&mddev->write_lock);
6611 if (mddev->in_sync) {
6612 mddev->in_sync = 0;
6613 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6614 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6615 if (mddev->safemode_delay &&
6616 mddev->safemode == 0)
6617 mddev->safemode = 1;
6618 spin_unlock_irq(&mddev->write_lock);
6619 md_update_sb(mddev, 0);
00bcb4ac 6620 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6621 } else
6622 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6623
070dc6dd 6624 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6625 return -EAGAIN;
6626 else
6627 return 0;
2a2275d6
N		 6628}
6629EXPORT_SYMBOL_GPL(md_allow_write);
6630
b63d7c2e 6631void md_unplug(mddev_t *mddev)
252ac522
N		 6632{
6633 if (mddev->queue)
6634 blk_unplug(mddev->queue);
6635 if (mddev->plug)
6636 mddev->plug->unplug_fn(mddev->plug);
6637}
6638
1da177e4
LT		 6639#define SYNC_MARKS 10
6640#define SYNC_MARK_STEP (3*HZ)
29269553 6641void md_do_sync(mddev_t *mddev)
1da177e4
LT		 6642{
6643 mddev_t *mddev2;
6644 unsigned int currspeed = 0,
6645 window;
57afd89f 6646 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 6647 unsigned long mark[SYNC_MARKS];
6648 sector_t mark_cnt[SYNC_MARKS];
6649 int last_mark,m;
6650 struct list_head *tmp;
6651 sector_t last_check;
57afd89f 6652 int skipped = 0;
5fd6c1dc 6653 mdk_rdev_t *rdev;
61df9d91 6654 char *desc;
1da177e4
LT		 6655
6656 /* just incase thread restarts... */
6657 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6658 return;
5fd6c1dc
N		 6659 if (mddev->ro) /* never try to sync a read-only array */
6660 return;
1da177e4 6661
61df9d91
N		 6662 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6663 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6664 desc = "data-check";
6665 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6666 desc = "requested-resync";
6667 else
6668 desc = "resync";
6669 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6670 desc = "reshape";
6671 else
6672 desc = "recovery";
6673
1da177e4
LT		 6674 /* we overload curr_resync somewhat here.
6675 * 0 == not engaged in resync at all
6676 * 2 == checking that there is no conflict with another sync
6677 * 1 == like 2, but have yielded to allow conflicting resync to
6678 * commense
6679 * other == active in resync - this many blocks
6680 *
6681 * Before starting a resync we must have set curr_resync to
6682 * 2, and then checked that every "conflicting" array has curr_resync
6683 * less than ours. When we find one that is the same or higher
6684 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6685 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6686 * This will mean we have to start checking from the beginning again.
6687 *
6688 */
6689
6690 do {
6691 mddev->curr_resync = 2;
6692
6693 try_again:
404e4b43 6694 if (kthread_should_stop())
6985c43f 6695 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 6696
6697 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 6698 goto skip;
29ac4aa3 6699 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 6700 if (mddev2 == mddev)
6701 continue;
90b08710
BS		 6702 if (!mddev->parallel_resync
6703 && mddev2->curr_resync
6704 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 6705 DEFINE_WAIT(wq);
6706 if (mddev < mddev2 && mddev->curr_resync == 2) {
6707 /* arbitrarily yield */
6708 mddev->curr_resync = 1;
6709 wake_up(&resync_wait);
6710 }
6711 if (mddev > mddev2 && mddev->curr_resync == 1)
6712 /* no need to wait here, we can wait the next
6713 * time 'round when curr_resync == 2
6714 */
6715 continue;
9744197c
N		 6716 /* We need to wait 'interruptible' so as not to
6717 * contribute to the load average, and not to
6718 * be caught by 'softlockup'
6719 */
6720 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 6721 if (!kthread_should_stop() &&
8712e553 6722 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 6723 printk(KERN_INFO "md: delaying %s of %s"
6724 " until %s has finished (they"
1da177e4 6725 " share one or more physical units)\n",
61df9d91 6726 desc, mdname(mddev), mdname(mddev2));
1da177e4 6727 mddev_put(mddev2);
9744197c
N		 6728 if (signal_pending(current))
6729 flush_signals(current);
1da177e4
LT		 6730 schedule();
6731 finish_wait(&resync_wait, &wq);
6732 goto try_again;
6733 }
6734 finish_wait(&resync_wait, &wq);
6735 }
6736 }
6737 } while (mddev->curr_resync < 2);
6738
5fd6c1dc 6739 j = 0;
9d88883e 6740 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 6741 /* resync follows the size requested by the personality,
57afd89f 6742 * which defaults to physical size, but can be virtual size
1da177e4
LT		 6743 */
6744 max_sectors = mddev->resync_max_sectors;
9d88883e 6745 mddev->resync_mismatches = 0;
5fd6c1dc 6746 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 6747 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6748 j = mddev->resync_min;
6749 else if (!mddev->bitmap)
5fd6c1dc 6750 j = mddev->recovery_cp;
5e96ee65 6751
ccfcc3c1 6752 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 6753 max_sectors = mddev->dev_sectors;
5fd6c1dc 6754 else {
1da177e4 6755 /* recovery follows the physical size of devices */
58c0fed4 6756 max_sectors = mddev->dev_sectors;
5fd6c1dc 6757 j = MaxSector;
4e59ca7d
DW		 6758 rcu_read_lock();
6759 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 6760 if (rdev->raid_disk >= 0 &&
6761 !test_bit(Faulty, &rdev->flags) &&
6762 !test_bit(In_sync, &rdev->flags) &&
6763 rdev->recovery_offset < j)
6764 j = rdev->recovery_offset;
4e59ca7d 6765 rcu_read_unlock();
5fd6c1dc 6766 }
1da177e4 6767
61df9d91
N		 6768 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6769 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6770 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 6771 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 6772 "(but not more than %d KB/sec) for %s.\n",
6773 speed_max(mddev), desc);
1da177e4 6774
eea1bf38 6775 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 6776
57afd89f 6777 io_sectors = 0;
1da177e4
LT		 6778 for (m = 0; m < SYNC_MARKS; m++) {
6779 mark[m] = jiffies;
57afd89f 6780 mark_cnt[m] = io_sectors;
1da177e4
LT		 6781 }
6782 last_mark = 0;
6783 mddev->resync_mark = mark[last_mark];
6784 mddev->resync_mark_cnt = mark_cnt[last_mark];
6785
6786 /*
6787 * Tune reconstruction:
6788 */
6789 window = 32*(PAGE_SIZE/512);
6790 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6791 window/2,(unsigned long long) max_sectors/2);
6792
6793 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 6794 last_check = 0;
6795
6796 if (j>2) {
6797 printk(KERN_INFO
61df9d91
N		 6798 "md: resuming %s of %s from checkpoint.\n",
6799 desc, mdname(mddev));
1da177e4
LT		 6800 mddev->curr_resync = j;
6801 }
efa59339 6802 mddev->curr_resync_completed = mddev->curr_resync;
1da177e4
LT		 6803
6804 while (j < max_sectors) {
57afd89f 6805 sector_t sectors;
1da177e4 6806
57afd89f 6807 skipped = 0;
97e4f42d 6808
7a91ee1f
N		 6809 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6810 ((mddev->curr_resync > mddev->curr_resync_completed &&
6811 (mddev->curr_resync - mddev->curr_resync_completed)
6812 > (max_sectors >> 4)) ||
6813 (j - mddev->curr_resync_completed)*2
6814 >= mddev->resync_max - mddev->curr_resync_completed
6815 )) {
97e4f42d 6816 /* time to update curr_resync_completed */
252ac522 6817 md_unplug(mddev);
97e4f42d
N		 6818 wait_event(mddev->recovery_wait,
6819 atomic_read(&mddev->recovery_active) == 0);
6820 mddev->curr_resync_completed =
6821 mddev->curr_resync;
070dc6dd 6822 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 6823 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 6824 }
acb180b0 6825
e62e58a5
N		 6826 while (j >= mddev->resync_max && !kthread_should_stop()) {
6827 /* As this condition is controlled by user-space,
6828 * we can block indefinitely, so use '_interruptible'
6829 * to avoid triggering warnings.
6830 */
6831 flush_signals(current); /* just in case */
6832 wait_event_interruptible(mddev->recovery_wait,
6833 mddev->resync_max > j
6834 || kthread_should_stop());
6835 }
acb180b0
N		 6836
6837 if (kthread_should_stop())
6838 goto interrupted;
6839
57afd89f 6840 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 6841 currspeed < speed_min(mddev));
57afd89f 6842 if (sectors == 0) {
dfc70645 6843 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6844 goto out;
6845 }
57afd89f
N		 6846
6847 if (!skipped) { /* actual IO requested */
6848 io_sectors += sectors;
6849 atomic_add(sectors, &mddev->recovery_active);
6850 }
6851
1da177e4
LT		 6852 j += sectors;
6853 if (j>1) mddev->curr_resync = j;
ff4e8d9a 6854 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 6855 if (last_check == 0)
6856 /* this is the earliers that rebuilt will be
6857 * visible in /proc/mdstat
6858 */
6859 md_new_event(mddev);
57afd89f
N		 6860
6861 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 6862 continue;
6863
57afd89f 6864 last_check = io_sectors;
1da177e4 6865
dfc70645 6866 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4
LT		 6867 break;
6868
6869 repeat:
6870 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6871 /* step marks */
6872 int next = (last_mark+1) % SYNC_MARKS;
6873
6874 mddev->resync_mark = mark[next];
6875 mddev->resync_mark_cnt = mark_cnt[next];
6876 mark[next] = jiffies;
57afd89f 6877 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6878 last_mark = next;
6879 }
6880
6881
c6207277
N		 6882 if (kthread_should_stop())
6883 goto interrupted;
6884
1da177e4
LT		 6885
6886 /*
6887 * this loop exits only if either when we are slower than
6888 * the 'hard' speed limit, or the system was IO-idle for
6889 * a jiffy.
6890 * the system might be non-idle CPU-wise, but we only care
6891 * about not overloading the IO subsystem. (things like an
6892 * e2fsck being done on the RAID array should execute fast)
6893 */
252ac522 6894 md_unplug(mddev);
1da177e4
LT		 6895 cond_resched();
6896
57afd89f
N		 6897 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6898 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 6899
88202a0c
N		 6900 if (currspeed > speed_min(mddev)) {
6901 if ((currspeed > speed_max(mddev)) ||
eea1bf38 6902 !is_mddev_idle(mddev, 0)) {
c0e48521 6903 msleep(500);
1da177e4
LT		 6904 goto repeat;
6905 }
6906 }
6907 }
61df9d91 6908 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 6909 /*
6910 * this also signals 'finished resyncing' to md_stop
6911 */
6912 out:
252ac522 6913 md_unplug(mddev);
1da177e4
LT		 6914
6915 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6916
6917 /* tell personality that we are finished */
57afd89f 6918 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 6919
dfc70645 6920 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 6921 mddev->curr_resync > 2) {
6922 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6923 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6924 if (mddev->curr_resync >= mddev->recovery_cp) {
6925 printk(KERN_INFO
61df9d91
N		 6926 "md: checkpointing %s of %s.\n",
6927 desc, mdname(mddev));
5fd6c1dc
N		 6928 mddev->recovery_cp = mddev->curr_resync;
6929 }
6930 } else
6931 mddev->recovery_cp = MaxSector;
6932 } else {
6933 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6934 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 6935 rcu_read_lock();
6936 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 6937 if (rdev->raid_disk >= 0 &&
70fffd0b 6938 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 6939 !test_bit(Faulty, &rdev->flags) &&
6940 !test_bit(In_sync, &rdev->flags) &&
6941 rdev->recovery_offset < mddev->curr_resync)
6942 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 6943 rcu_read_unlock();
5fd6c1dc 6944 }
1da177e4 6945 }
17571284 6946 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6947
1da177e4 6948 skip:
c07b70ad
N		 6949 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6950 /* We completed so min/max setting can be forgotten if used. */
6951 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6952 mddev->resync_min = 0;
6953 mddev->resync_max = MaxSector;
6954 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6955 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 6956 mddev->curr_resync = 0;
efa59339
N		 6957 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6958 mddev->curr_resync_completed = 0;
c6207277 6959 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 6960 wake_up(&resync_wait);
6961 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6962 md_wakeup_thread(mddev->thread);
c6207277
N		 6963 return;
6964
6965 interrupted:
6966 /*
6967 * got a signal, exit.
6968 */
6969 printk(KERN_INFO
6970 "md: md_do_sync() got signal ... exiting\n");
6971 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6972 goto out;
6973
1da177e4 6974}
29269553 6975EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 6976
6977
b4c4c7b8
N		 6978static int remove_and_add_spares(mddev_t *mddev)
6979{
6980 mdk_rdev_t *rdev;
b4c4c7b8
N		 6981 int spares = 0;
6982
97e4f42d
N		 6983 mddev->curr_resync_completed = 0;
6984
159ec1fc 6985 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 6986 if (rdev->raid_disk >= 0 &&
6bfe0b49 6987 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 6988 (test_bit(Faulty, &rdev->flags) ||
6989 ! test_bit(In_sync, &rdev->flags)) &&
6990 atomic_read(&rdev->nr_pending)==0) {
6991 if (mddev->pers->hot_remove_disk(
6992 mddev, rdev->raid_disk)==0) {
6993 char nm[20];
6994 sprintf(nm,"rd%d", rdev->raid_disk);
6995 sysfs_remove_link(&mddev->kobj, nm);
6996 rdev->raid_disk = -1;
6997 }
6998 }
6999
4044ba58 7000 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
159ec1fc 7001 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7002 if (rdev->raid_disk >= 0 &&
e5427135
DW		 7003 !test_bit(In_sync, &rdev->flags) &&
7004 !test_bit(Blocked, &rdev->flags))
dfc70645 7005 spares++;
b4c4c7b8
N		 7006 if (rdev->raid_disk < 0
7007 && !test_bit(Faulty, &rdev->flags)) {
7008 rdev->recovery_offset = 0;
199050ea
NB		 7009 if (mddev->pers->
7010 hot_add_disk(mddev, rdev) == 0) {
b4c4c7b8
N		 7011 char nm[20];
7012 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 7013 if (sysfs_create_link(&mddev->kobj,
7014 &rdev->kobj, nm))
00bcb4ac 7015 /* failure here is OK */;
b4c4c7b8
N		 7016 spares++;
7017 md_new_event(mddev);
93be75ff 7018 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7019 } else
7020 break;
7021 }
dfc70645 7022 }
b4c4c7b8
N		 7023 }
7024 return spares;
7025}
1da177e4
LT		 7026/*
7027 * This routine is regularly called by all per-raid-array threads to
7028 * deal with generic issues like resync and super-block update.
7029 * Raid personalities that don't have a thread (linear/raid0) do not
7030 * need this as they never do any recovery or update the superblock.
7031 *
7032 * It does not do any resync itself, but rather "forks" off other threads
7033 * to do that as needed.
7034 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7035 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7036 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7037 * and wakeups up this thread which will reap the thread and finish up.
7038 * This thread also removes any faulty devices (with nr_pending == 0).
7039 *
7040 * The overall approach is:
7041 * 1/ if the superblock needs updating, update it.
7042 * 2/ If a recovery thread is running, don't do anything else.
7043 * 3/ If recovery has finished, clean up, possibly marking spares active.
7044 * 4/ If there are any faulty devices, remove them.
7045 * 5/ If array is degraded, try to add spares devices
7046 * 6/ If array has spares or is not in-sync, start a resync thread.
7047 */
7048void md_check_recovery(mddev_t *mddev)
7049{
7050 mdk_rdev_t *rdev;
1da177e4
LT		 7051
7052
5f40402d 7053 if (mddev->bitmap)
aa5cbd10 7054 bitmap_daemon_work(mddev);
1da177e4
LT		 7055
7056 if (mddev->ro)
7057 return;
fca4d848
N		 7058
7059 if (signal_pending(current)) {
31a59e34 7060 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7061 printk(KERN_INFO "md: %s in immediate safe mode\n",
7062 mdname(mddev));
7063 mddev->safemode = 2;
7064 }
7065 flush_signals(current);
7066 }
7067
c89a8eee
N		 7068 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7069 return;
1da177e4 7070 if ( ! (
126925c0 7071 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7072 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7073 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7074 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7075 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7076 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7077 ))
7078 return;
fca4d848 7079
df5b89b3 7080 if (mddev_trylock(mddev)) {
b4c4c7b8 7081 int spares = 0;
fca4d848 7082
c89a8eee
N		 7083 if (mddev->ro) {
7084 /* Only thing we do on a ro array is remove
7085 * failed devices.
7086 */
7087 remove_and_add_spares(mddev);
7088 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7089 goto unlock;
7090 }
7091
31a59e34 7092 if (!mddev->external) {
0fd62b86 7093 int did_change = 0;
31a59e34
N		 7094 spin_lock_irq(&mddev->write_lock);
7095 if (mddev->safemode &&
7096 !atomic_read(&mddev->writes_pending) &&
7097 !mddev->in_sync &&
7098 mddev->recovery_cp == MaxSector) {
7099 mddev->in_sync = 1;
0fd62b86 7100 did_change = 1;
070dc6dd 7101 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7102 }
7103 if (mddev->safemode == 1)
7104 mddev->safemode = 0;
7105 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7106 if (did_change)
00bcb4ac 7107 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7108 }
fca4d848 7109
850b2b42
N		 7110 if (mddev->flags)
7111 md_update_sb(mddev, 0);
06d91a5f 7112
1da177e4
LT		 7113 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7114 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7115 /* resync/recovery still happening */
7116 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7117 goto unlock;
7118 }
7119 if (mddev->sync_thread) {
7120 /* resync has finished, collect result */
7121 md_unregister_thread(mddev->sync_thread);
7122 mddev->sync_thread = NULL;
56ac36d7
DW		 7123 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7124 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1da177e4
LT		 7125 /* success...*/
7126 /* activate any spares */
a99ac971
NB		 7127 if (mddev->pers->spare_active(mddev))
7128 sysfs_notify(&mddev->kobj, NULL,
7129 "degraded");
1da177e4 7130 }
cea9c228
N		 7131 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7132 mddev->pers->finish_reshape)
7133 mddev->pers->finish_reshape(mddev);
850b2b42 7134 md_update_sb(mddev, 1);
41158c7e
N		 7135
7136 /* if array is no-longer degraded, then any saved_raid_disk
7137 * information must be scrapped
7138 */
7139 if (!mddev->degraded)
159ec1fc 7140 list_for_each_entry(rdev, &mddev->disks, same_set)
41158c7e
N		 7141 rdev->saved_raid_disk = -1;
7142
1da177e4
LT		 7143 mddev->recovery = 0;
7144 /* flag recovery needed just to double check */
7145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
00bcb4ac 7146 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7147 md_new_event(mddev);
1da177e4
LT		 7148 goto unlock;
7149 }
72a23c21
NB		 7150 /* Set RUNNING before clearing NEEDED to avoid
7151 * any transients in the value of "sync_action".
7152 */
7153 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7154 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7155 /* Clear some bits that don't mean anything, but
7156 * might be left set
7157 */
24dd469d
N		 7158 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7159 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7160
5fd6c1dc
N		 7161 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7162 goto unlock;
1da177e4
LT		 7163 /* no recovery is running.
7164 * remove any failed drives, then
7165 * add spares if possible.
7166 * Spare are also removed and re-added, to allow
7167 * the personality to fail the re-add.
7168 */
1da177e4 7169
b4c4c7b8 7170 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7171 if (mddev->pers->check_reshape == NULL ||
7172 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7173 /* Cannot proceed */
7174 goto unlock;
7175 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7176 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7177 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7178 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7179 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7180 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7181 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7182 } else if (mddev->recovery_cp < MaxSector) {
7183 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7184 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7185 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7186 /* nothing to be done ... */
1da177e4 7187 goto unlock;
24dd469d 7188
1da177e4 7189 if (mddev->pers->sync_request) {
a654b9d8
N		 7190 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7191 /* We are adding a device or devices to an array
7192 * which has the bitmap stored on all devices.
7193 * So make sure all bitmap pages get written
7194 */
7195 bitmap_write_all(mddev->bitmap);
7196 }
1da177e4
LT		 7197 mddev->sync_thread = md_register_thread(md_do_sync,
7198 mddev,
0da3c619 7199 "resync");
1da177e4
LT		 7200 if (!mddev->sync_thread) {
7201 printk(KERN_ERR "%s: could not start resync"
7202 " thread...\n",
7203 mdname(mddev));
7204 /* leave the spares where they are, it shouldn't hurt */
7205 mddev->recovery = 0;
d7603b7e 7206 } else
1da177e4 7207 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7208 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7209 md_new_event(mddev);
1da177e4
LT		 7210 }
7211 unlock:
72a23c21
NB		 7212 if (!mddev->sync_thread) {
7213 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7214 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7215 &mddev->recovery))
0c3573f1 7216 if (mddev->sysfs_action)
00bcb4ac 7217 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7218 }
1da177e4
LT		 7219 mddev_unlock(mddev);
7220 }
7221}
7222
6bfe0b49
DW		 7223void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7224{
00bcb4ac 7225 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49
DW		 7226 wait_event_timeout(rdev->blocked_wait,
7227 !test_bit(Blocked, &rdev->flags),
7228 msecs_to_jiffies(5000));
7229 rdev_dec_pending(rdev, mddev);
7230}
7231EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7232
75c96f85
AB		 7233static int md_notify_reboot(struct notifier_block *this,
7234 unsigned long code, void *x)
1da177e4
LT		 7235{
7236 struct list_head *tmp;
7237 mddev_t *mddev;
7238
7239 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7240
7241 printk(KERN_INFO "md: stopping all md devices.\n");
7242
29ac4aa3 7243 for_each_mddev(mddev, tmp)
c71d4887 7244 if (mddev_trylock(mddev)) {
2b25000b
N		 7245 /* Force a switch to readonly even array
7246 * appears to still be in use. Hence
7247 * the '100'.
7248 */
a4bd82d0 7249 md_set_readonly(mddev, 100);
c71d4887
NB		 7250 mddev_unlock(mddev);
7251 }
1da177e4
LT		 7252 /*
7253 * certain more exotic SCSI devices are known to be
7254 * volatile wrt too early system reboots. While the
7255 * right place to handle this issue is the given
7256 * driver, we do want to have a safe RAID driver ...
7257 */
7258 mdelay(1000*1);
7259 }
7260 return NOTIFY_DONE;
7261}
7262
75c96f85 7263static struct notifier_block md_notifier = {
1da177e4
LT		 7264 .notifier_call = md_notify_reboot,
7265 .next = NULL,
7266 .priority = INT_MAX, /* before any real devices */
7267};
7268
7269static void md_geninit(void)
7270{
1da177e4
LT		 7271 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7272
c7705f34 7273 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 7274}
7275
75c96f85 7276static int __init md_init(void)
1da177e4 7277{
e804ac78
TH		 7278 int ret = -ENOMEM;
7279
7280 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7281 if (!md_wq)
7282 goto err_wq;
7283
7284 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7285 if (!md_misc_wq)
7286 goto err_misc_wq;
7287
7288 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7289 goto err_md;
7290
7291 if ((ret = register_blkdev(0, "mdp")) < 0)
7292 goto err_mdp;
7293 mdp_major = ret;
7294
3dbd8c2e 7295 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 7296 md_probe, NULL, NULL);
7297 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 7298 md_probe, NULL, NULL);
7299
1da177e4 7300 register_reboot_notifier(&md_notifier);
0b4d4147 7301 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 7302
7303 md_geninit();
d710e138 7304 return 0;
1da177e4 7305
e804ac78
TH		 7306err_mdp:
7307 unregister_blkdev(MD_MAJOR, "md");
7308err_md:
7309 destroy_workqueue(md_misc_wq);
7310err_misc_wq:
7311 destroy_workqueue(md_wq);
7312err_wq:
7313 return ret;
7314}
1da177e4
LT		 7315
7316#ifndef MODULE
7317
7318/*
7319 * Searches all registered partitions for autorun RAID arrays
7320 * at boot time.
7321 */
4d936ec1
ME		 7322
7323static LIST_HEAD(all_detected_devices);
7324struct detected_devices_node {
7325 struct list_head list;
7326 dev_t dev;
7327};
1da177e4
LT		 7328
7329void md_autodetect_dev(dev_t dev)
7330{
4d936ec1
ME		 7331 struct detected_devices_node *node_detected_dev;
7332
7333 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7334 if (node_detected_dev) {
7335 node_detected_dev->dev = dev;
7336 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7337 } else {
7338 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7339 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7340 }
1da177e4
LT		 7341}
7342
7343
7344static void autostart_arrays(int part)
7345{
7346 mdk_rdev_t *rdev;
4d936ec1
ME		 7347 struct detected_devices_node *node_detected_dev;
7348 dev_t dev;
7349 int i_scanned, i_passed;
1da177e4 7350
4d936ec1
ME		 7351 i_scanned = 0;
7352 i_passed = 0;
1da177e4 7353
4d936ec1 7354 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 7355
4d936ec1
ME		 7356 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7357 i_scanned++;
7358 node_detected_dev = list_entry(all_detected_devices.next,
7359 struct detected_devices_node, list);
7360 list_del(&node_detected_dev->list);
7361 dev = node_detected_dev->dev;
7362 kfree(node_detected_dev);
df968c4e 7363 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 7364 if (IS_ERR(rdev))
7365 continue;
7366
b2d444d7 7367 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7368 MD_BUG();
7369 continue;
7370 }
d0fae18f 7371 set_bit(AutoDetected, &rdev->flags);
1da177e4 7372 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 7373 i_passed++;
1da177e4 7374 }
4d936ec1
ME		 7375
7376 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7377 i_scanned, i_passed);
1da177e4
LT		 7378
7379 autorun_devices(part);
7380}
7381
fdee8ae4 7382#endif /* !MODULE */
1da177e4
LT		 7383
7384static __exit void md_exit(void)
7385{
7386 mddev_t *mddev;
7387 struct list_head *tmp;
8ab5e4c1 7388
3dbd8c2e 7389 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 7390 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 7391
3dbd8c2e 7392 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 7393 unregister_blkdev(mdp_major, "mdp");
7394 unregister_reboot_notifier(&md_notifier);
7395 unregister_sysctl_table(raid_table_header);
7396 remove_proc_entry("mdstat", NULL);
29ac4aa3 7397 for_each_mddev(mddev, tmp) {
1da177e4 7398 export_array(mddev);
d3374825 7399 mddev->hold_active = 0;
1da177e4 7400 }
e804ac78
TH		 7401 destroy_workqueue(md_misc_wq);
7402 destroy_workqueue(md_wq);
1da177e4
LT		 7403}
7404
685784aa 7405subsys_initcall(md_init);
1da177e4
LT		 7406module_exit(md_exit)
7407
f91de92e
N		 7408static int get_ro(char *buffer, struct kernel_param *kp)
7409{
7410 return sprintf(buffer, "%d", start_readonly);
7411}
7412static int set_ro(const char *val, struct kernel_param *kp)
7413{
7414 char *e;
7415 int num = simple_strtoul(val, &e, 10);
7416 if (*val && (*e == '\0' || *e == '\n')) {
7417 start_readonly = num;
4dbcdc75 7418 return 0;
f91de92e
N		 7419 }
7420 return -EINVAL;
7421}
7422
80ca3a44
N		 7423module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7424module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 7425
efeb53c0 7426module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 7427
1da177e4
LT		 7428EXPORT_SYMBOL(register_md_personality);
7429EXPORT_SYMBOL(unregister_md_personality);
7430EXPORT_SYMBOL(md_error);
7431EXPORT_SYMBOL(md_done_sync);
7432EXPORT_SYMBOL(md_write_start);
7433EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 7434EXPORT_SYMBOL(md_register_thread);
7435EXPORT_SYMBOL(md_unregister_thread);
7436EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 7437EXPORT_SYMBOL(md_check_recovery);
7438MODULE_LICENSE("GPL");
0efb9e61 7439MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 7440MODULE_ALIAS("md");
72008652 7441MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)