projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: Be more careful about clearing flags bit in ->recovery
[GitHub/mt8127/android_kernel_alcatel_ttab.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
0ca69886
N		 291 if (mddev == NULL || mddev->pers == NULL
292 || !mddev->ready) {
409c57f3
N		 293 bio_io_error(bio);
294 return 0;
295 }
0ca69886 296 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 297 rcu_read_lock();
e9c7469b 298 if (mddev->suspended) {
409c57f3
N		 299 DEFINE_WAIT(__wait);
300 for (;;) {
301 prepare_to_wait(&mddev->sb_wait, &__wait,
302 TASK_UNINTERRUPTIBLE);
e9c7469b 303 if (!mddev->suspended)
409c57f3
N		 304 break;
305 rcu_read_unlock();
306 schedule();
307 rcu_read_lock();
308 }
309 finish_wait(&mddev->sb_wait, &__wait);
310 }
311 atomic_inc(&mddev->active_io);
312 rcu_read_unlock();
49077326 313
21a52c6d 314 rv = mddev->pers->make_request(mddev, bio);
49077326
N		 315
316 cpu = part_stat_lock();
317 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
318 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
319 bio_sectors(bio));
320 part_stat_unlock();
321
409c57f3
N		 322 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
323 wake_up(&mddev->sb_wait);
324
325 return rv;
326}
327
9e35b99c
N		 328/* mddev_suspend makes sure no new requests are submitted
329 * to the device, and that any requests that have been submitted
330 * are completely handled.
331 * Once ->stop is called and completes, the module will be completely
332 * unused.
333 */
390ee602 334void mddev_suspend(mddev_t *mddev)
409c57f3
N		 335{
336 BUG_ON(mddev->suspended);
337 mddev->suspended = 1;
338 synchronize_rcu();
339 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
340 mddev->pers->quiesce(mddev, 1);
409c57f3 341}
390ee602 342EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 343
390ee602 344void mddev_resume(mddev_t *mddev)
409c57f3
N		 345{
346 mddev->suspended = 0;
347 wake_up(&mddev->sb_wait);
348 mddev->pers->quiesce(mddev, 0);
1da177e4 349}
390ee602 350EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 351
3fa841d7
N		 352int mddev_congested(mddev_t *mddev, int bits)
353{
354 return mddev->suspended;
355}
356EXPORT_SYMBOL(mddev_congested);
357
a2826aa9 358/*
e9c7469b 359 * Generic flush handling for md
a2826aa9
N		 360 */
361
e9c7469b 362static void md_end_flush(struct bio *bio, int err)
a2826aa9
N		 363{
364 mdk_rdev_t *rdev = bio->bi_private;
365 mddev_t *mddev = rdev->mddev;
a2826aa9
N		 366
367 rdev_dec_pending(rdev, mddev);
368
369 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 370 /* The pre-request flush has finished */
e804ac78 371 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 372 }
373 bio_put(bio);
374}
375
a7a07e69
N		 376static void md_submit_flush_data(struct work_struct *ws);
377
a035fc3e 378static void submit_flushes(struct work_struct *ws)
a2826aa9 379{
a035fc3e 380 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
a2826aa9
N		 381 mdk_rdev_t *rdev;
382
a7a07e69
N		 383 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
384 atomic_set(&mddev->flush_pending, 1);
a2826aa9
N		 385 rcu_read_lock();
386 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
387 if (rdev->raid_disk >= 0 &&
388 !test_bit(Faulty, &rdev->flags)) {
389 /* Take two references, one is dropped
390 * when request finishes, one after
391 * we reclaim rcu_read_lock
392 */
393 struct bio *bi;
394 atomic_inc(&rdev->nr_pending);
395 atomic_inc(&rdev->nr_pending);
396 rcu_read_unlock();
a167f663 397 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 398 bi->bi_end_io = md_end_flush;
a2826aa9
N		 399 bi->bi_private = rdev;
400 bi->bi_bdev = rdev->bdev;
401 atomic_inc(&mddev->flush_pending);
e9c7469b 402 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 403 rcu_read_lock();
404 rdev_dec_pending(rdev, mddev);
405 }
406 rcu_read_unlock();
a7a07e69
N		 407 if (atomic_dec_and_test(&mddev->flush_pending))
408 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 409}
410
e9c7469b 411static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 412{
e9c7469b
TH		 413 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
414 struct bio *bio = mddev->flush_bio;
a2826aa9 415
e9c7469b 416 if (bio->bi_size == 0)
a2826aa9
N		 417 /* an empty barrier - all done */
418 bio_endio(bio, 0);
419 else {
e9c7469b 420 bio->bi_rw &= ~REQ_FLUSH;
21a52c6d 421 if (mddev->pers->make_request(mddev, bio))
a2826aa9 422 generic_make_request(bio);
a2826aa9 423 }
2b74e12e
N		 424
425 mddev->flush_bio = NULL;
426 wake_up(&mddev->sb_wait);
a2826aa9
N		 427}
428
e9c7469b 429void md_flush_request(mddev_t *mddev, struct bio *bio)
a2826aa9
N		 430{
431 spin_lock_irq(&mddev->write_lock);
432 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 433 !mddev->flush_bio,
a2826aa9 434 mddev->write_lock, /*nothing*/);
e9c7469b 435 mddev->flush_bio = bio;
a2826aa9
N		 436 spin_unlock_irq(&mddev->write_lock);
437
a035fc3e
N		 438 INIT_WORK(&mddev->flush_work, submit_flushes);
439 queue_work(md_wq, &mddev->flush_work);
a2826aa9 440}
e9c7469b 441EXPORT_SYMBOL(md_flush_request);
409c57f3 442
2ac87401
N		 443/* Support for plugging.
444 * This mirrors the plugging support in request_queue, but does not
445 * require having a whole queue
446 */
447static void plugger_work(struct work_struct *work)
448{
449 struct plug_handle *plug =
450 container_of(work, struct plug_handle, unplug_work);
451 plug->unplug_fn(plug);
452}
453static void plugger_timeout(unsigned long data)
454{
455 struct plug_handle *plug = (void *)data;
456 kblockd_schedule_work(NULL, &plug->unplug_work);
457}
458void plugger_init(struct plug_handle *plug,
459 void (*unplug_fn)(struct plug_handle *))
460{
461 plug->unplug_flag = 0;
462 plug->unplug_fn = unplug_fn;
463 init_timer(&plug->unplug_timer);
464 plug->unplug_timer.function = plugger_timeout;
465 plug->unplug_timer.data = (unsigned long)plug;
466 INIT_WORK(&plug->unplug_work, plugger_work);
467}
468EXPORT_SYMBOL_GPL(plugger_init);
469
470void plugger_set_plug(struct plug_handle *plug)
471{
472 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
473 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
474}
475EXPORT_SYMBOL_GPL(plugger_set_plug);
476
477int plugger_remove_plug(struct plug_handle *plug)
478{
479 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
480 del_timer(&plug->unplug_timer);
481 return 1;
482 } else
483 return 0;
484}
485EXPORT_SYMBOL_GPL(plugger_remove_plug);
486
487
1da177e4
LT		 488static inline mddev_t *mddev_get(mddev_t *mddev)
489{
490 atomic_inc(&mddev->active);
491 return mddev;
492}
493
5fd3a17e 494static void mddev_delayed_delete(struct work_struct *ws);
d3374825 495
1da177e4
LT		 496static void mddev_put(mddev_t *mddev)
497{
a167f663
N		 498 struct bio_set *bs = NULL;
499
1da177e4
LT		 500 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
501 return;
d3374825 502 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 503 mddev->ctime == 0 && !mddev->hold_active) {
504 /* Array is not configured at all, and not held active,
505 * so destroy it */
1da177e4 506 list_del(&mddev->all_mddevs);
a167f663
N		 507 bs = mddev->bio_set;
508 mddev->bio_set = NULL;
d3374825 509 if (mddev->gendisk) {
e804ac78
TH		 510 /* We did a probe so need to clean up. Call
511 * queue_work inside the spinlock so that
512 * flush_workqueue() after mddev_find will
513 * succeed in waiting for the work to be done.
d3374825
N		 514 */
515 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 516 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 517 } else
518 kfree(mddev);
519 }
520 spin_unlock(&all_mddevs_lock);
a167f663
N		 521 if (bs)
522 bioset_free(bs);
1da177e4
LT		 523}
524
390ee602 525void mddev_init(mddev_t *mddev)
fafd7fb0
N		 526{
527 mutex_init(&mddev->open_mutex);
528 mutex_init(&mddev->reconfig_mutex);
529 mutex_init(&mddev->bitmap_info.mutex);
530 INIT_LIST_HEAD(&mddev->disks);
531 INIT_LIST_HEAD(&mddev->all_mddevs);
532 init_timer(&mddev->safemode_timer);
533 atomic_set(&mddev->active, 1);
534 atomic_set(&mddev->openers, 0);
535 atomic_set(&mddev->active_io, 0);
536 spin_lock_init(&mddev->write_lock);
537 atomic_set(&mddev->flush_pending, 0);
538 init_waitqueue_head(&mddev->sb_wait);
539 init_waitqueue_head(&mddev->recovery_wait);
540 mddev->reshape_position = MaxSector;
541 mddev->resync_min = 0;
542 mddev->resync_max = MaxSector;
543 mddev->level = LEVEL_NONE;
544}
390ee602 545EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 546
1da177e4
LT		 547static mddev_t * mddev_find(dev_t unit)
548{
549 mddev_t *mddev, *new = NULL;
550
551 retry:
552 spin_lock(&all_mddevs_lock);
efeb53c0
N		 553
554 if (unit) {
555 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
556 if (mddev->unit == unit) {
557 mddev_get(mddev);
558 spin_unlock(&all_mddevs_lock);
559 kfree(new);
560 return mddev;
561 }
562
563 if (new) {
564 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 565 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 566 new->hold_active = UNTIL_IOCTL;
567 return new;
1da177e4 568 }
efeb53c0
N		 569 } else if (new) {
570 /* find an unused unit number */
571 static int next_minor = 512;
572 int start = next_minor;
573 int is_free = 0;
574 int dev = 0;
575 while (!is_free) {
576 dev = MKDEV(MD_MAJOR, next_minor);
577 next_minor++;
578 if (next_minor > MINORMASK)
579 next_minor = 0;
580 if (next_minor == start) {
581 /* Oh dear, all in use. */
582 spin_unlock(&all_mddevs_lock);
583 kfree(new);
584 return NULL;
585 }
586
587 is_free = 1;
588 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
589 if (mddev->unit == dev) {
590 is_free = 0;
591 break;
592 }
593 }
594 new->unit = dev;
595 new->md_minor = MINOR(dev);
596 new->hold_active = UNTIL_STOP;
1da177e4
LT		 597 list_add(&new->all_mddevs, &all_mddevs);
598 spin_unlock(&all_mddevs_lock);
599 return new;
600 }
601 spin_unlock(&all_mddevs_lock);
602
9ffae0cf 603 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 604 if (!new)
605 return NULL;
606
1da177e4
LT		 607 new->unit = unit;
608 if (MAJOR(unit) == MD_MAJOR)
609 new->md_minor = MINOR(unit);
610 else
611 new->md_minor = MINOR(unit) >> MdpMinorShift;
612
fafd7fb0 613 mddev_init(new);
1da177e4 614
1da177e4
LT		 615 goto retry;
616}
617
618static inline int mddev_lock(mddev_t * mddev)
619{
df5b89b3 620 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 621}
622
b522adcd
DW		 623static inline int mddev_is_locked(mddev_t *mddev)
624{
625 return mutex_is_locked(&mddev->reconfig_mutex);
626}
627
1da177e4
LT		 628static inline int mddev_trylock(mddev_t * mddev)
629{
df5b89b3 630 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 631}
632
b6eb127d
N		 633static struct attribute_group md_redundancy_group;
634
a64c876f 635static void mddev_unlock(mddev_t * mddev)
1da177e4 636{
a64c876f 637 if (mddev->to_remove) {
b6eb127d
N		 638 /* These cannot be removed under reconfig_mutex as
639 * an access to the files will try to take reconfig_mutex
640 * while holding the file unremovable, which leads to
641 * a deadlock.
bb4f1e9d
N		 642 * So hold set sysfs_active while the remove in happeing,
643 * and anything else which might set ->to_remove or my
644 * otherwise change the sysfs namespace will fail with
645 * -EBUSY if sysfs_active is still set.
646 * We set sysfs_active under reconfig_mutex and elsewhere
647 * test it under the same mutex to ensure its correct value
648 * is seen.
b6eb127d 649 */
a64c876f
N		 650 struct attribute_group *to_remove = mddev->to_remove;
651 mddev->to_remove = NULL;
bb4f1e9d 652 mddev->sysfs_active = 1;
b6eb127d
N		 653 mutex_unlock(&mddev->reconfig_mutex);
654
00bcb4ac
N		 655 if (mddev->kobj.sd) {
656 if (to_remove != &md_redundancy_group)
657 sysfs_remove_group(&mddev->kobj, to_remove);
658 if (mddev->pers == NULL ||
659 mddev->pers->sync_request == NULL) {
660 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
661 if (mddev->sysfs_action)
662 sysfs_put(mddev->sysfs_action);
663 mddev->sysfs_action = NULL;
664 }
a64c876f 665 }
bb4f1e9d 666 mddev->sysfs_active = 0;
b6eb127d
N		 667 } else
668 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 669
005eca5e 670 md_wakeup_thread(mddev->thread);
1da177e4
LT		 671}
672
2989ddbd 673static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4 674{
159ec1fc 675 mdk_rdev_t *rdev;
1da177e4 676
159ec1fc 677 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 678 if (rdev->desc_nr == nr)
679 return rdev;
159ec1fc 680
1da177e4
LT		 681 return NULL;
682}
683
684static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
685{
1da177e4
LT		 686 mdk_rdev_t *rdev;
687
159ec1fc 688 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 689 if (rdev->bdev->bd_dev == dev)
690 return rdev;
159ec1fc 691
1da177e4
LT		 692 return NULL;
693}
694
d9d166c2 695static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 696{
697 struct mdk_personality *pers;
d9d166c2
N		 698 list_for_each_entry(pers, &pers_list, list) {
699 if (level != LEVEL_NONE && pers->level == level)
2604b703 700 return pers;
d9d166c2
N		 701 if (strcmp(pers->name, clevel)==0)
702 return pers;
703 }
2604b703
N		 704 return NULL;
705}
706
b73df2d3 707/* return the offset of the super block in 512byte sectors */
77933d72 708static inline sector_t calc_dev_sboffset(struct block_device *bdev)
1da177e4 709{
77304d2a 710 sector_t num_sectors = i_size_read(bdev->bd_inode) / 512;
b73df2d3 711 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 712}
713
1da177e4
LT		 714static int alloc_disk_sb(mdk_rdev_t * rdev)
715{
716 if (rdev->sb_page)
717 MD_BUG();
718
719 rdev->sb_page = alloc_page(GFP_KERNEL);
720 if (!rdev->sb_page) {
721 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 722 return -ENOMEM;
1da177e4
LT		 723 }
724
725 return 0;
726}
727
728static void free_disk_sb(mdk_rdev_t * rdev)
729{
730 if (rdev->sb_page) {
2d1f3b5d 731 put_page(rdev->sb_page);
1da177e4
LT		 732 rdev->sb_loaded = 0;
733 rdev->sb_page = NULL;
0f420358 734 rdev->sb_start = 0;
dd8ac336 735 rdev->sectors = 0;
1da177e4
LT		 736 }
737}
738
739
6712ecf8 740static void super_written(struct bio *bio, int error)
7bfa19f2
N		 741{
742 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 743 mddev_t *mddev = rdev->mddev;
7bfa19f2 744
3a0f5bbb
N		 745 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
746 printk("md: super_written gets error=%d, uptodate=%d\n",
747 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
748 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 749 md_error(mddev, rdev);
3a0f5bbb 750 }
7bfa19f2 751
a9701a30
N		 752 if (atomic_dec_and_test(&mddev->pending_writes))
753 wake_up(&mddev->sb_wait);
f8b58edf 754 bio_put(bio);
7bfa19f2
N		 755}
756
757void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
758 sector_t sector, int size, struct page *page)
759{
760 /* write first size bytes of page to sector of rdev
761 * Increment mddev->pending_writes before returning
762 * and decrement it on completion, waking up sb_wait
763 * if zero is reached.
764 * If an error occurred, call md_error
765 */
a167f663 766 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2
N		 767
768 bio->bi_bdev = rdev->bdev;
769 bio->bi_sector = sector;
770 bio_add_page(bio, page, size, 0);
771 bio->bi_private = rdev;
772 bio->bi_end_io = super_written;
a9701a30 773
7bfa19f2 774 atomic_inc(&mddev->pending_writes);
e9c7469b
TH		 775 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
776 bio);
a9701a30
N		 777}
778
779void md_super_wait(mddev_t *mddev)
780{
e9c7469b 781 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 782 DEFINE_WAIT(wq);
783 for(;;) {
784 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
785 if (atomic_read(&mddev->pending_writes)==0)
786 break;
a9701a30
N		 787 schedule();
788 }
789 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 790}
791
6712ecf8 792static void bi_complete(struct bio *bio, int error)
1da177e4 793{
1da177e4 794 complete((struct completion*)bio->bi_private);
1da177e4
LT		 795}
796
2b193363
N		 797int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
798 struct page *page, int rw)
1da177e4 799{
a167f663 800 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 801 struct completion event;
802 int ret;
803
7b6d91da 804 rw |= REQ_SYNC | REQ_UNPLUG;
1da177e4 805
2b193363 806 bio->bi_bdev = rdev->bdev;
1da177e4
LT		 807 bio->bi_sector = sector;
808 bio_add_page(bio, page, size, 0);
809 init_completion(&event);
810 bio->bi_private = &event;
811 bio->bi_end_io = bi_complete;
812 submit_bio(rw, bio);
813 wait_for_completion(&event);
814
815 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
816 bio_put(bio);
817 return ret;
818}
a8745db2 819EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 820
0002b271 821static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 822{
823 char b[BDEVNAME_SIZE];
824 if (!rdev->sb_page) {
825 MD_BUG();
826 return -EINVAL;
827 }
828 if (rdev->sb_loaded)
829 return 0;
830
831
2b193363 832 if (!sync_page_io(rdev, rdev->sb_start, size, rdev->sb_page, READ))
1da177e4
LT		 833 goto fail;
834 rdev->sb_loaded = 1;
835 return 0;
836
837fail:
838 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
839 bdevname(rdev->bdev,b));
840 return -EINVAL;
841}
842
843static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
844{
05710466
AN		 845 return sb1->set_uuid0 == sb2->set_uuid0 &&
846 sb1->set_uuid1 == sb2->set_uuid1 &&
847 sb1->set_uuid2 == sb2->set_uuid2 &&
848 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 849}
850
1da177e4
LT		 851static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
852{
853 int ret;
854 mdp_super_t *tmp1, *tmp2;
855
856 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
857 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
858
859 if (!tmp1 || !tmp2) {
860 ret = 0;
35020f1a 861 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 862 goto abort;
863 }
864
865 *tmp1 = *sb1;
866 *tmp2 = *sb2;
867
868 /*
869 * nr_disks is not constant
870 */
871 tmp1->nr_disks = 0;
872 tmp2->nr_disks = 0;
873
ce0c8e05 874 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 875abort:
990a8baf
JJ		 876 kfree(tmp1);
877 kfree(tmp2);
1da177e4
LT		 878 return ret;
879}
880
4d167f09
N		 881
882static u32 md_csum_fold(u32 csum)
883{
884 csum = (csum & 0xffff) + (csum >> 16);
885 return (csum & 0xffff) + (csum >> 16);
886}
887
1da177e4
LT		 888static unsigned int calc_sb_csum(mdp_super_t * sb)
889{
4d167f09
N		 890 u64 newcsum = 0;
891 u32 *sb32 = (u32*)sb;
892 int i;
1da177e4
LT		 893 unsigned int disk_csum, csum;
894
895 disk_csum = sb->sb_csum;
896 sb->sb_csum = 0;
4d167f09
N		 897
898 for (i = 0; i < MD_SB_BYTES/4 ; i++)
899 newcsum += sb32[i];
900 csum = (newcsum & 0xffffffff) + (newcsum>>32);
901
902
903#ifdef CONFIG_ALPHA
904 /* This used to use csum_partial, which was wrong for several
905 * reasons including that different results are returned on
906 * different architectures. It isn't critical that we get exactly
907 * the same return value as before (we always csum_fold before
908 * testing, and that removes any differences). However as we
909 * know that csum_partial always returned a 16bit value on
910 * alphas, do a fold to maximise conformity to previous behaviour.
911 */
912 sb->sb_csum = md_csum_fold(disk_csum);
913#else
1da177e4 914 sb->sb_csum = disk_csum;
4d167f09 915#endif
1da177e4
LT		 916 return csum;
917}
918
919
920/*
921 * Handle superblock details.
922 * We want to be able to handle multiple superblock formats
923 * so we have a common interface to them all, and an array of
924 * different handlers.
925 * We rely on user-space to write the initial superblock, and support
926 * reading and updating of superblocks.
927 * Interface methods are:
928 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
929 * loads and validates a superblock on dev.
930 * if refdev != NULL, compare superblocks on both devices
931 * Return:
932 * 0 - dev has a superblock that is compatible with refdev
933 * 1 - dev has a superblock that is compatible and newer than refdev
934 * so dev should be used as the refdev in future
935 * -EINVAL superblock incompatible or invalid
936 * -othererror e.g. -EIO
937 *
938 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
939 * Verify that dev is acceptable into mddev.
940 * The first time, mddev->raid_disks will be 0, and data from
941 * dev should be merged in. Subsequent calls check that dev
942 * is new enough. Return 0 or -EINVAL
943 *
944 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
945 * Update the superblock for rdev with data in mddev
946 * This does not write to disc.
947 *
948 */
949
950struct super_type {
0cd17fec
CW		 951 char *name;
952 struct module *owner;
953 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
954 int minor_version);
955 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
956 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
957 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
15f4a5fd 958 sector_t num_sectors);
1da177e4
LT		 959};
960
0894cc30
AN		 961/*
962 * Check that the given mddev has no bitmap.
963 *
964 * This function is called from the run method of all personalities that do not
965 * support bitmaps. It prints an error message and returns non-zero if mddev
966 * has a bitmap. Otherwise, it returns 0.
967 *
968 */
969int md_check_no_bitmap(mddev_t *mddev)
970{
c3d9714e 971 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 972 return 0;
973 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
974 mdname(mddev), mddev->pers->name);
975 return 1;
976}
977EXPORT_SYMBOL(md_check_no_bitmap);
978
1da177e4
LT		 979/*
980 * load_super for 0.90.0
981 */
982static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
983{
984 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
985 mdp_super_t *sb;
986 int ret;
1da177e4
LT		 987
988 /*
0f420358 989 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 990 * it's at the end of the disk.
991 *
992 * It also happens to be a multiple of 4Kb.
993 */
0f420358 994 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1da177e4 995
0002b271 996 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 997 if (ret) return ret;
998
999 ret = -EINVAL;
1000
1001 bdevname(rdev->bdev, b);
1002 sb = (mdp_super_t*)page_address(rdev->sb_page);
1003
1004 if (sb->md_magic != MD_SB_MAGIC) {
1005 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1006 b);
1007 goto abort;
1008 }
1009
1010 if (sb->major_version != 0 ||
f6705578
N		 1011 sb->minor_version < 90 ||
1012 sb->minor_version > 91) {
1da177e4
LT		 1013 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1014 sb->major_version, sb->minor_version,
1015 b);
1016 goto abort;
1017 }
1018
1019 if (sb->raid_disks <= 0)
1020 goto abort;
1021
4d167f09 1022 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1023 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1024 b);
1025 goto abort;
1026 }
1027
1028 rdev->preferred_minor = sb->md_minor;
1029 rdev->data_offset = 0;
0002b271 1030 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 1031
1032 if (sb->level == LEVEL_MULTIPATH)
1033 rdev->desc_nr = -1;
1034 else
1035 rdev->desc_nr = sb->this_disk.number;
1036
9a7b2b0f 1037 if (!refdev) {
1da177e4 1038 ret = 1;
9a7b2b0f 1039 } else {
1da177e4
LT		 1040 __u64 ev1, ev2;
1041 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1042 if (!uuid_equal(refsb, sb)) {
1043 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1044 b, bdevname(refdev->bdev,b2));
1045 goto abort;
1046 }
1047 if (!sb_equal(refsb, sb)) {
1048 printk(KERN_WARNING "md: %s has same UUID"
1049 " but different superblock to %s\n",
1050 b, bdevname(refdev->bdev, b2));
1051 goto abort;
1052 }
1053 ev1 = md_event(sb);
1054 ev2 = md_event(refsb);
1055 if (ev1 > ev2)
1056 ret = 1;
1057 else
1058 ret = 0;
1059 }
8190e754 1060 rdev->sectors = rdev->sb_start;
1da177e4 1061
dd8ac336 1062 if (rdev->sectors < sb->size * 2 && sb->level > 1)
2bf071bf
N		 1063 /* "this cannot possibly happen" ... */
1064 ret = -EINVAL;
1065
1da177e4
LT		 1066 abort:
1067 return ret;
1068}
1069
1070/*
1071 * validate_super for 0.90.0
1072 */
1073static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1074{
1075 mdp_disk_t *desc;
1076 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 1077 __u64 ev1 = md_event(sb);
1da177e4 1078
41158c7e 1079 rdev->raid_disk = -1;
c5d79adb
N		 1080 clear_bit(Faulty, &rdev->flags);
1081 clear_bit(In_sync, &rdev->flags);
1082 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1083
1da177e4
LT		 1084 if (mddev->raid_disks == 0) {
1085 mddev->major_version = 0;
1086 mddev->minor_version = sb->minor_version;
1087 mddev->patch_version = sb->patch_version;
e691063a 1088 mddev->external = 0;
9d8f0363 1089 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1090 mddev->ctime = sb->ctime;
1091 mddev->utime = sb->utime;
1092 mddev->level = sb->level;
d9d166c2 1093 mddev->clevel[0] = 0;
1da177e4
LT		 1094 mddev->layout = sb->layout;
1095 mddev->raid_disks = sb->raid_disks;
58c0fed4 1096 mddev->dev_sectors = sb->size * 2;
07d84d10 1097 mddev->events = ev1;
c3d9714e
N		 1098 mddev->bitmap_info.offset = 0;
1099 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1100
f6705578
N		 1101 if (mddev->minor_version >= 91) {
1102 mddev->reshape_position = sb->reshape_position;
1103 mddev->delta_disks = sb->delta_disks;
1104 mddev->new_level = sb->new_level;
1105 mddev->new_layout = sb->new_layout;
664e7c41 1106 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1107 } else {
1108 mddev->reshape_position = MaxSector;
1109 mddev->delta_disks = 0;
1110 mddev->new_level = mddev->level;
1111 mddev->new_layout = mddev->layout;
664e7c41 1112 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1113 }
1114
1da177e4
LT		 1115 if (sb->state & (1<<MD_SB_CLEAN))
1116 mddev->recovery_cp = MaxSector;
1117 else {
1118 if (sb->events_hi == sb->cp_events_hi &&
1119 sb->events_lo == sb->cp_events_lo) {
1120 mddev->recovery_cp = sb->recovery_cp;
1121 } else
1122 mddev->recovery_cp = 0;
1123 }
1124
1125 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1126 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1127 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1128 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1129
1130 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1131
1132 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1133 mddev->bitmap_info.file == NULL)
1134 mddev->bitmap_info.offset =
1135 mddev->bitmap_info.default_offset;
a654b9d8 1136
41158c7e 1137 } else if (mddev->pers == NULL) {
be6800a7
N		 1138 /* Insist on good event counter while assembling, except
1139 * for spares (which don't need an event count) */
1da177e4 1140 ++ev1;
be6800a7
N		 1141 if (sb->disks[rdev->desc_nr].state & (
1142 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1143 if (ev1 < mddev->events)
1144 return -EINVAL;
41158c7e
N		 1145 } else if (mddev->bitmap) {
1146 /* if adding to array with a bitmap, then we can accept an
1147 * older device ... but not too old.
1148 */
41158c7e
N		 1149 if (ev1 < mddev->bitmap->events_cleared)
1150 return 0;
07d84d10
N		 1151 } else {
1152 if (ev1 < mddev->events)
1153 /* just a hot-add of a new device, leave raid_disk at -1 */
1154 return 0;
1155 }
41158c7e 1156
1da177e4 1157 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1158 desc = sb->disks + rdev->desc_nr;
1159
1160 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1161 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1162 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1163 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1164 set_bit(In_sync, &rdev->flags);
1da177e4 1165 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1166 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1167 /* active but not in sync implies recovery up to
1168 * reshape position. We don't know exactly where
1169 * that is, so set to zero for now */
1170 if (mddev->minor_version >= 91) {
1171 rdev->recovery_offset = 0;
1172 rdev->raid_disk = desc->raid_disk;
1173 }
1da177e4 1174 }
8ddf9efe
N		 1175 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1176 set_bit(WriteMostly, &rdev->flags);
41158c7e 1177 } else /* MULTIPATH are always insync */
b2d444d7 1178 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1179 return 0;
1180}
1181
1182/*
1183 * sync_super for 0.90.0
1184 */
1185static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1186{
1187 mdp_super_t *sb;
1da177e4
LT		 1188 mdk_rdev_t *rdev2;
1189 int next_spare = mddev->raid_disks;
19133a42 1190
1da177e4
LT		 1191
1192 /* make rdev->sb match mddev data..
1193 *
1194 * 1/ zero out disks
1195 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1196 * 3/ any empty disks < next_spare become removed
1197 *
1198 * disks[0] gets initialised to REMOVED because
1199 * we cannot be sure from other fields if it has
1200 * been initialised or not.
1201 */
1202 int i;
1203 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1204
61181565
N		 1205 rdev->sb_size = MD_SB_BYTES;
1206
1da177e4
LT		 1207 sb = (mdp_super_t*)page_address(rdev->sb_page);
1208
1209 memset(sb, 0, sizeof(*sb));
1210
1211 sb->md_magic = MD_SB_MAGIC;
1212 sb->major_version = mddev->major_version;
1da177e4
LT		 1213 sb->patch_version = mddev->patch_version;
1214 sb->gvalid_words = 0; /* ignored */
1215 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1216 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1217 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1218 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1219
1220 sb->ctime = mddev->ctime;
1221 sb->level = mddev->level;
58c0fed4 1222 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1223 sb->raid_disks = mddev->raid_disks;
1224 sb->md_minor = mddev->md_minor;
e691063a 1225 sb->not_persistent = 0;
1da177e4
LT		 1226 sb->utime = mddev->utime;
1227 sb->state = 0;
1228 sb->events_hi = (mddev->events>>32);
1229 sb->events_lo = (u32)mddev->events;
1230
f6705578
N		 1231 if (mddev->reshape_position == MaxSector)
1232 sb->minor_version = 90;
1233 else {
1234 sb->minor_version = 91;
1235 sb->reshape_position = mddev->reshape_position;
1236 sb->new_level = mddev->new_level;
1237 sb->delta_disks = mddev->delta_disks;
1238 sb->new_layout = mddev->new_layout;
664e7c41 1239 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1240 }
1241 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1242 if (mddev->in_sync)
1243 {
1244 sb->recovery_cp = mddev->recovery_cp;
1245 sb->cp_events_hi = (mddev->events>>32);
1246 sb->cp_events_lo = (u32)mddev->events;
1247 if (mddev->recovery_cp == MaxSector)
1248 sb->state = (1<< MD_SB_CLEAN);
1249 } else
1250 sb->recovery_cp = 0;
1251
1252 sb->layout = mddev->layout;
9d8f0363 1253 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1254
c3d9714e 1255 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1256 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1257
1da177e4 1258 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1259 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1260 mdp_disk_t *d;
86e6ffdd 1261 int desc_nr;
0261cd9f
N		 1262 int is_active = test_bit(In_sync, &rdev2->flags);
1263
1264 if (rdev2->raid_disk >= 0 &&
1265 sb->minor_version >= 91)
1266 /* we have nowhere to store the recovery_offset,
1267 * but if it is not below the reshape_position,
1268 * we can piggy-back on that.
1269 */
1270 is_active = 1;
1271 if (rdev2->raid_disk < 0 ||
1272 test_bit(Faulty, &rdev2->flags))
1273 is_active = 0;
1274 if (is_active)
86e6ffdd 1275 desc_nr = rdev2->raid_disk;
1da177e4 1276 else
86e6ffdd 1277 desc_nr = next_spare++;
19133a42 1278 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1279 d = &sb->disks[rdev2->desc_nr];
1280 nr_disks++;
1281 d->number = rdev2->desc_nr;
1282 d->major = MAJOR(rdev2->bdev->bd_dev);
1283 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1284 if (is_active)
1da177e4
LT		 1285 d->raid_disk = rdev2->raid_disk;
1286 else
1287 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1288 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1289 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1290 else if (is_active) {
1da177e4 1291 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1292 if (test_bit(In_sync, &rdev2->flags))
1293 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1294 active++;
1295 working++;
1296 } else {
1297 d->state = 0;
1298 spare++;
1299 working++;
1300 }
8ddf9efe
N		 1301 if (test_bit(WriteMostly, &rdev2->flags))
1302 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1303 }
1da177e4
LT		 1304 /* now set the "removed" and "faulty" bits on any missing devices */
1305 for (i=0 ; i < mddev->raid_disks ; i++) {
1306 mdp_disk_t *d = &sb->disks[i];
1307 if (d->state == 0 && d->number == 0) {
1308 d->number = i;
1309 d->raid_disk = i;
1310 d->state = (1<<MD_DISK_REMOVED);
1311 d->state |= (1<<MD_DISK_FAULTY);
1312 failed++;
1313 }
1314 }
1315 sb->nr_disks = nr_disks;
1316 sb->active_disks = active;
1317 sb->working_disks = working;
1318 sb->failed_disks = failed;
1319 sb->spare_disks = spare;
1320
1321 sb->this_disk = sb->disks[rdev->desc_nr];
1322 sb->sb_csum = calc_sb_csum(sb);
1323}
1324
0cd17fec
CW		 1325/*
1326 * rdev_size_change for 0.90.0
1327 */
1328static unsigned long long
15f4a5fd 1329super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec 1330{
58c0fed4 1331 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1332 return 0; /* component must fit device */
c3d9714e 1333 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1334 return 0; /* can't move bitmap */
0f420358 1335 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
15f4a5fd
AN		 1336 if (!num_sectors || num_sectors > rdev->sb_start)
1337 num_sectors = rdev->sb_start;
0f420358 1338 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1339 rdev->sb_page);
1340 md_super_wait(rdev->mddev);
c26a44ed 1341 return num_sectors;
0cd17fec
CW		 1342}
1343
1344
1da177e4
LT		 1345/*
1346 * version 1 superblock
1347 */
1348
1c05b4bc 1349static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1350{
1c05b4bc
N		 1351 __le32 disk_csum;
1352 u32 csum;
1da177e4
LT		 1353 unsigned long long newcsum;
1354 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1355 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1356 int i;
1357
1358 disk_csum = sb->sb_csum;
1359 sb->sb_csum = 0;
1360 newcsum = 0;
1361 for (i=0; size>=4; size -= 4 )
1362 newcsum += le32_to_cpu(*isuper++);
1363
1364 if (size == 2)
1c05b4bc 1365 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1366
1367 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1368 sb->sb_csum = disk_csum;
1369 return cpu_to_le32(csum);
1370}
1371
1372static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1373{
1374 struct mdp_superblock_1 *sb;
1375 int ret;
0f420358 1376 sector_t sb_start;
1da177e4 1377 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1378 int bmask;
1da177e4
LT		 1379
1380 /*
0f420358 1381 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1382 * It is always aligned to a 4K boundary and
1383 * depeding on minor_version, it can be:
1384 * 0: At least 8K, but less than 12K, from end of device
1385 * 1: At start of device
1386 * 2: 4K from start of device.
1387 */
1388 switch(minor_version) {
1389 case 0:
77304d2a 1390 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1391 sb_start -= 8*2;
1392 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1393 break;
1394 case 1:
0f420358 1395 sb_start = 0;
1da177e4
LT		 1396 break;
1397 case 2:
0f420358 1398 sb_start = 8;
1da177e4
LT		 1399 break;
1400 default:
1401 return -EINVAL;
1402 }
0f420358 1403 rdev->sb_start = sb_start;
1da177e4 1404
0002b271
N		 1405 /* superblock is rarely larger than 1K, but it can be larger,
1406 * and it is safe to read 4k, so we do that
1407 */
1408 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1409 if (ret) return ret;
1410
1411
1412 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1413
1414 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1415 sb->major_version != cpu_to_le32(1) ||
1416 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1417 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1418 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1419 return -EINVAL;
1420
1421 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1422 printk("md: invalid superblock checksum on %s\n",
1423 bdevname(rdev->bdev,b));
1424 return -EINVAL;
1425 }
1426 if (le64_to_cpu(sb->data_size) < 10) {
1427 printk("md: data_size too small on %s\n",
1428 bdevname(rdev->bdev,b));
1429 return -EINVAL;
1430 }
e11e93fa 1431
1da177e4
LT		 1432 rdev->preferred_minor = 0xffff;
1433 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1434 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1435
0002b271 1436 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1437 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1438 if (rdev->sb_size & bmask)
a1801f85
N		 1439 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1440
1441 if (minor_version
0f420358 1442 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1443 return -EINVAL;
0002b271 1444
31b65a0d
N		 1445 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1446 rdev->desc_nr = -1;
1447 else
1448 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1449
9a7b2b0f 1450 if (!refdev) {
8ed75463 1451 ret = 1;
9a7b2b0f 1452 } else {
1da177e4
LT		 1453 __u64 ev1, ev2;
1454 struct mdp_superblock_1 *refsb =
1455 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1456
1457 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1458 sb->level != refsb->level ||
1459 sb->layout != refsb->layout ||
1460 sb->chunksize != refsb->chunksize) {
1461 printk(KERN_WARNING "md: %s has strangely different"
1462 " superblock to %s\n",
1463 bdevname(rdev->bdev,b),
1464 bdevname(refdev->bdev,b2));
1465 return -EINVAL;
1466 }
1467 ev1 = le64_to_cpu(sb->events);
1468 ev2 = le64_to_cpu(refsb->events);
1469
1470 if (ev1 > ev2)
8ed75463
N		 1471 ret = 1;
1472 else
1473 ret = 0;
1da177e4 1474 }
a1801f85 1475 if (minor_version)
77304d2a 1476 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1477 le64_to_cpu(sb->data_offset);
1da177e4 1478 else
dd8ac336
AN		 1479 rdev->sectors = rdev->sb_start;
1480 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1481 return -EINVAL;
dd8ac336 1482 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1483 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1484 return -EINVAL;
8ed75463 1485 return ret;
1da177e4
LT		 1486}
1487
1488static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1489{
1490 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1491 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1492
41158c7e 1493 rdev->raid_disk = -1;
c5d79adb
N		 1494 clear_bit(Faulty, &rdev->flags);
1495 clear_bit(In_sync, &rdev->flags);
1496 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1497
1da177e4
LT		 1498 if (mddev->raid_disks == 0) {
1499 mddev->major_version = 1;
1500 mddev->patch_version = 0;
e691063a 1501 mddev->external = 0;
9d8f0363 1502 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1503 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1504 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1505 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1506 mddev->clevel[0] = 0;
1da177e4
LT		 1507 mddev->layout = le32_to_cpu(sb->layout);
1508 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1509 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1510 mddev->events = ev1;
c3d9714e
N		 1511 mddev->bitmap_info.offset = 0;
1512 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1513
1514 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1515 memcpy(mddev->uuid, sb->set_uuid, 16);
1516
1517 mddev->max_disks = (4096-256)/2;
a654b9d8 1518
71c0805c 1519 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1520 mddev->bitmap_info.file == NULL )
1521 mddev->bitmap_info.offset =
1522 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1523
f6705578
N		 1524 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1525 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1526 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1527 mddev->new_level = le32_to_cpu(sb->new_level);
1528 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1529 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1530 } else {
1531 mddev->reshape_position = MaxSector;
1532 mddev->delta_disks = 0;
1533 mddev->new_level = mddev->level;
1534 mddev->new_layout = mddev->layout;
664e7c41 1535 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1536 }
1537
41158c7e 1538 } else if (mddev->pers == NULL) {
be6800a7
N		 1539 /* Insist of good event counter while assembling, except for
1540 * spares (which don't need an event count) */
1da177e4 1541 ++ev1;
be6800a7
N		 1542 if (rdev->desc_nr >= 0 &&
1543 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1544 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1545 if (ev1 < mddev->events)
1546 return -EINVAL;
41158c7e
N		 1547 } else if (mddev->bitmap) {
1548 /* If adding to array with a bitmap, then we can accept an
1549 * older device, but not too old.
1550 */
41158c7e
N		 1551 if (ev1 < mddev->bitmap->events_cleared)
1552 return 0;
07d84d10
N		 1553 } else {
1554 if (ev1 < mddev->events)
1555 /* just a hot-add of a new device, leave raid_disk at -1 */
1556 return 0;
1557 }
1da177e4
LT		 1558 if (mddev->level != LEVEL_MULTIPATH) {
1559 int role;
3673f305
N		 1560 if (rdev->desc_nr < 0 ||
1561 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1562 role = 0xffff;
1563 rdev->desc_nr = -1;
1564 } else
1565 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1566 switch(role) {
1567 case 0xffff: /* spare */
1da177e4
LT		 1568 break;
1569 case 0xfffe: /* faulty */
b2d444d7 1570 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1571 break;
1572 default:
5fd6c1dc
N		 1573 if ((le32_to_cpu(sb->feature_map) &
1574 MD_FEATURE_RECOVERY_OFFSET))
1575 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1576 else
1577 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1578 rdev->raid_disk = role;
1579 break;
1580 }
8ddf9efe
N		 1581 if (sb->devflags & WriteMostly1)
1582 set_bit(WriteMostly, &rdev->flags);
41158c7e 1583 } else /* MULTIPATH are always insync */
b2d444d7 1584 set_bit(In_sync, &rdev->flags);
41158c7e 1585
1da177e4
LT		 1586 return 0;
1587}
1588
1589static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1590{
1591 struct mdp_superblock_1 *sb;
1da177e4
LT		 1592 mdk_rdev_t *rdev2;
1593 int max_dev, i;
1594 /* make rdev->sb match mddev and rdev data. */
1595
1596 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1597
1598 sb->feature_map = 0;
1599 sb->pad0 = 0;
5fd6c1dc 1600 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1601 memset(sb->pad1, 0, sizeof(sb->pad1));
1602 memset(sb->pad2, 0, sizeof(sb->pad2));
1603 memset(sb->pad3, 0, sizeof(sb->pad3));
1604
1605 sb->utime = cpu_to_le64((__u64)mddev->utime);
1606 sb->events = cpu_to_le64(mddev->events);
1607 if (mddev->in_sync)
1608 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1609 else
1610 sb->resync_offset = cpu_to_le64(0);
1611
1c05b4bc 1612 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1613
f0ca340c 1614 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1615 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1616 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1617 sb->level = cpu_to_le32(mddev->level);
1618 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1619
c3d9714e
N		 1620 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1621 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1622 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1623 }
5fd6c1dc
N		 1624
1625 if (rdev->raid_disk >= 0 &&
97e4f42d 1626 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1627 sb->feature_map |=
1628 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1629 sb->recovery_offset =
1630 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1631 }
1632
f6705578
N		 1633 if (mddev->reshape_position != MaxSector) {
1634 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1635 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1636 sb->new_layout = cpu_to_le32(mddev->new_layout);
1637 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1638 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1639 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1640 }
a654b9d8 1641
1da177e4 1642 max_dev = 0;
159ec1fc 1643 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1644 if (rdev2->desc_nr+1 > max_dev)
1645 max_dev = rdev2->desc_nr+1;
a778b73f 1646
70471daf
N		 1647 if (max_dev > le32_to_cpu(sb->max_dev)) {
1648 int bmask;
a778b73f 1649 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1650 rdev->sb_size = max_dev * 2 + 256;
1651 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1652 if (rdev->sb_size & bmask)
1653 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1654 } else
1655 max_dev = le32_to_cpu(sb->max_dev);
1656
1da177e4
LT		 1657 for (i=0; i<max_dev;i++)
1658 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1659
159ec1fc 1660 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1661 i = rdev2->desc_nr;
b2d444d7 1662 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1663 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1664 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1665 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1666 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1667 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1668 else
1669 sb->dev_roles[i] = cpu_to_le16(0xffff);
1670 }
1671
1da177e4
LT		 1672 sb->sb_csum = calc_sb_1_csum(sb);
1673}
1674
0cd17fec 1675static unsigned long long
15f4a5fd 1676super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec
CW		 1677{
1678 struct mdp_superblock_1 *sb;
15f4a5fd 1679 sector_t max_sectors;
58c0fed4 1680 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1681 return 0; /* component must fit device */
0f420358 1682 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1683 /* minor versions 1 and 2; superblock before data */
77304d2a 1684 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1685 max_sectors -= rdev->data_offset;
1686 if (!num_sectors || num_sectors > max_sectors)
1687 num_sectors = max_sectors;
c3d9714e 1688 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1689 /* minor version 0 with bitmap we can't move */
1690 return 0;
1691 } else {
1692 /* minor version 0; superblock after data */
0f420358 1693 sector_t sb_start;
77304d2a 1694 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1695 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1696 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1697 if (!num_sectors || num_sectors > max_sectors)
1698 num_sectors = max_sectors;
0f420358 1699 rdev->sb_start = sb_start;
0cd17fec
CW		 1700 }
1701 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
15f4a5fd 1702 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1703 sb->super_offset = rdev->sb_start;
0cd17fec 1704 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1705 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1706 rdev->sb_page);
1707 md_super_wait(rdev->mddev);
c26a44ed 1708 return num_sectors;
0cd17fec 1709}
1da177e4 1710
75c96f85 1711static struct super_type super_types[] = {
1da177e4
LT		 1712 [0] = {
1713 .name = "0.90.0",
1714 .owner = THIS_MODULE,
0cd17fec
CW		 1715 .load_super = super_90_load,
1716 .validate_super = super_90_validate,
1717 .sync_super = super_90_sync,
1718 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1719 },
1720 [1] = {
1721 .name = "md-1",
1722 .owner = THIS_MODULE,
0cd17fec
CW		 1723 .load_super = super_1_load,
1724 .validate_super = super_1_validate,
1725 .sync_super = super_1_sync,
1726 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1727 },
1728};
1da177e4
LT		 1729
1730static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1731{
7dd5e7c3 1732 mdk_rdev_t *rdev, *rdev2;
1da177e4 1733
4b80991c
N		 1734 rcu_read_lock();
1735 rdev_for_each_rcu(rdev, mddev1)
1736 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1737 if (rdev->bdev->bd_contains ==
4b80991c
N		 1738 rdev2->bdev->bd_contains) {
1739 rcu_read_unlock();
7dd5e7c3 1740 return 1;
4b80991c
N		 1741 }
1742 rcu_read_unlock();
1da177e4
LT		 1743 return 0;
1744}
1745
1746static LIST_HEAD(pending_raid_disks);
1747
ac5e7113
AN		 1748/*
1749 * Try to register data integrity profile for an mddev
1750 *
1751 * This is called when an array is started and after a disk has been kicked
1752 * from the array. It only succeeds if all working and active component devices
1753 * are integrity capable with matching profiles.
1754 */
1755int md_integrity_register(mddev_t *mddev)
1756{
1757 mdk_rdev_t *rdev, *reference = NULL;
1758
1759 if (list_empty(&mddev->disks))
1760 return 0; /* nothing to do */
1761 if (blk_get_integrity(mddev->gendisk))
1762 return 0; /* already registered */
1763 list_for_each_entry(rdev, &mddev->disks, same_set) {
1764 /* skip spares and non-functional disks */
1765 if (test_bit(Faulty, &rdev->flags))
1766 continue;
1767 if (rdev->raid_disk < 0)
1768 continue;
1769 /*
1770 * If at least one rdev is not integrity capable, we can not
1771 * enable data integrity for the md device.
1772 */
1773 if (!bdev_get_integrity(rdev->bdev))
1774 return -EINVAL;
1775 if (!reference) {
1776 /* Use the first rdev as the reference */
1777 reference = rdev;
1778 continue;
1779 }
1780 /* does this rdev's profile match the reference profile? */
1781 if (blk_integrity_compare(reference->bdev->bd_disk,
1782 rdev->bdev->bd_disk) < 0)
1783 return -EINVAL;
1784 }
1785 /*
1786 * All component devices are integrity capable and have matching
1787 * profiles, register the common profile for the md device.
1788 */
1789 if (blk_integrity_register(mddev->gendisk,
1790 bdev_get_integrity(reference->bdev)) != 0) {
1791 printk(KERN_ERR "md: failed to register integrity for %s\n",
1792 mdname(mddev));
1793 return -EINVAL;
1794 }
1795 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1796 mdname(mddev));
1797 return 0;
1798}
1799EXPORT_SYMBOL(md_integrity_register);
1800
1801/* Disable data integrity if non-capable/non-matching disk is being added */
1802void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
3f9d99c1 1803{
3f9d99c1 1804 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1805 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1806
ac5e7113 1807 if (!bi_mddev) /* nothing to do */
3f9d99c1 1808 return;
ac5e7113 1809 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1810 return;
ac5e7113
AN		 1811 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1812 rdev->bdev->bd_disk) >= 0)
1813 return;
1814 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1815 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 1816}
ac5e7113 1817EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 1818
1da177e4
LT		 1819static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1820{
7dd5e7c3 1821 char b[BDEVNAME_SIZE];
f637b9f9 1822 struct kobject *ko;
1edf80d3 1823 char *s;
5e55e2f5 1824 int err;
1da177e4
LT		 1825
1826 if (rdev->mddev) {
1827 MD_BUG();
1828 return -EINVAL;
1829 }
11e2ede0
DW		 1830
1831 /* prevent duplicates */
1832 if (find_rdev(mddev, rdev->bdev->bd_dev))
1833 return -EEXIST;
1834
dd8ac336
AN		 1835 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1836 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1837 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 1838 if (mddev->pers) {
1839 /* Cannot change size, so fail
1840 * If mddev->level <= 0, then we don't care
1841 * about aligning sizes (e.g. linear)
1842 */
1843 if (mddev->level > 0)
1844 return -ENOSPC;
1845 } else
dd8ac336 1846 mddev->dev_sectors = rdev->sectors;
2bf071bf 1847 }
1da177e4
LT		 1848
1849 /* Verify rdev->desc_nr is unique.
1850 * If it is -1, assign a free number, else
1851 * check number is not in use
1852 */
1853 if (rdev->desc_nr < 0) {
1854 int choice = 0;
1855 if (mddev->pers) choice = mddev->raid_disks;
1856 while (find_rdev_nr(mddev, choice))
1857 choice++;
1858 rdev->desc_nr = choice;
1859 } else {
1860 if (find_rdev_nr(mddev, rdev->desc_nr))
1861 return -EBUSY;
1862 }
de01dfad
N		 1863 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1864 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1865 mdname(mddev), mddev->max_disks);
1866 return -EBUSY;
1867 }
19133a42 1868 bdevname(rdev->bdev,b);
649316b2 1869 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1870 *s = '!';
649316b2 1871
1da177e4 1872 rdev->mddev = mddev;
19133a42 1873 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1874
b2d6db58 1875 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1876 goto fail;
86e6ffdd 1877
0762b8bd 1878 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 1879 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1880 /* failure here is OK */;
1881 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 1882
4b80991c 1883 list_add_rcu(&rdev->same_set, &mddev->disks);
c5d79adb 1884 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
4044ba58
N		 1885
1886 /* May as well allow recovery to be retried once */
1887 mddev->recovery_disabled = 0;
3f9d99c1 1888
1da177e4 1889 return 0;
5e55e2f5
N		 1890
1891 fail:
1892 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1893 b, mdname(mddev));
1894 return err;
1da177e4
LT		 1895}
1896
177a99b2 1897static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1898{
1899 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1900 kobject_del(&rdev->kobj);
177a99b2 1901 kobject_put(&rdev->kobj);
5792a285
N		 1902}
1903
1da177e4
LT		 1904static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1905{
1906 char b[BDEVNAME_SIZE];
1907 if (!rdev->mddev) {
1908 MD_BUG();
1909 return;
1910 }
5463c790 1911 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
4b80991c 1912 list_del_rcu(&rdev->same_set);
1da177e4
LT		 1913 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1914 rdev->mddev = NULL;
86e6ffdd 1915 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 1916 sysfs_put(rdev->sysfs_state);
1917 rdev->sysfs_state = NULL;
5792a285 1918 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 1919 * writing to 'remove' to "dev/state". We also need
1920 * to delay it due to rcu usage.
5792a285 1921 */
4b80991c 1922 synchronize_rcu();
177a99b2
N		 1923 INIT_WORK(&rdev->del_work, md_delayed_delete);
1924 kobject_get(&rdev->kobj);
e804ac78 1925 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 1926}
1927
1928/*
1929 * prevent the device from being mounted, repartitioned or
1930 * otherwise reused by a RAID array (or any other kernel
1931 * subsystem), by bd_claiming the device.
1932 */
c5d79adb 1933static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1934{
1935 int err = 0;
1936 struct block_device *bdev;
1937 char b[BDEVNAME_SIZE];
1938
2e7b651d 1939 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1940 if (IS_ERR(bdev)) {
1941 printk(KERN_ERR "md: could not open %s.\n",
1942 __bdevname(dev, b));
1943 return PTR_ERR(bdev);
1944 }
c5d79adb 1945 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1946 if (err) {
1947 printk(KERN_ERR "md: could not bd_claim %s.\n",
1948 bdevname(bdev, b));
9a1c3542 1949 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1950 return err;
1951 }
c5d79adb
N		 1952 if (!shared)
1953 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1954 rdev->bdev = bdev;
1955 return err;
1956}
1957
1958static void unlock_rdev(mdk_rdev_t *rdev)
1959{
1960 struct block_device *bdev = rdev->bdev;
1961 rdev->bdev = NULL;
1962 if (!bdev)
1963 MD_BUG();
1964 bd_release(bdev);
9a1c3542 1965 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1966}
1967
1968void md_autodetect_dev(dev_t dev);
1969
1970static void export_rdev(mdk_rdev_t * rdev)
1971{
1972 char b[BDEVNAME_SIZE];
1973 printk(KERN_INFO "md: export_rdev(%s)\n",
1974 bdevname(rdev->bdev,b));
1975 if (rdev->mddev)
1976 MD_BUG();
1977 free_disk_sb(rdev);
1da177e4 1978#ifndef MODULE
d0fae18f
N		 1979 if (test_bit(AutoDetected, &rdev->flags))
1980 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1981#endif
1982 unlock_rdev(rdev);
86e6ffdd 1983 kobject_put(&rdev->kobj);
1da177e4
LT		 1984}
1985
1986static void kick_rdev_from_array(mdk_rdev_t * rdev)
1987{
1988 unbind_rdev_from_array(rdev);
1989 export_rdev(rdev);
1990}
1991
1992static void export_array(mddev_t *mddev)
1993{
159ec1fc 1994 mdk_rdev_t *rdev, *tmp;
1da177e4 1995
d089c6af 1996 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1997 if (!rdev->mddev) {
1998 MD_BUG();
1999 continue;
2000 }
2001 kick_rdev_from_array(rdev);
2002 }
2003 if (!list_empty(&mddev->disks))
2004 MD_BUG();
2005 mddev->raid_disks = 0;
2006 mddev->major_version = 0;
2007}
2008
2009static void print_desc(mdp_disk_t *desc)
2010{
2011 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2012 desc->major,desc->minor,desc->raid_disk,desc->state);
2013}
2014
cd2ac932 2015static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2016{
2017 int i;
2018
2019 printk(KERN_INFO
2020 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2021 sb->major_version, sb->minor_version, sb->patch_version,
2022 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2023 sb->ctime);
2024 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2025 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2026 sb->md_minor, sb->layout, sb->chunk_size);
2027 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2028 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2029 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2030 sb->failed_disks, sb->spare_disks,
2031 sb->sb_csum, (unsigned long)sb->events_lo);
2032
2033 printk(KERN_INFO);
2034 for (i = 0; i < MD_SB_DISKS; i++) {
2035 mdp_disk_t *desc;
2036
2037 desc = sb->disks + i;
2038 if (desc->number || desc->major || desc->minor ||
2039 desc->raid_disk || (desc->state && (desc->state != 4))) {
2040 printk(" D %2d: ", i);
2041 print_desc(desc);
2042 }
2043 }
2044 printk(KERN_INFO "md: THIS: ");
2045 print_desc(&sb->this_disk);
cd2ac932 2046}
1da177e4 2047
cd2ac932
CR		 2048static void print_sb_1(struct mdp_superblock_1 *sb)
2049{
2050 __u8 *uuid;
2051
2052 uuid = sb->set_uuid;
ad361c98 2053 printk(KERN_INFO
7b75c2f8 2054 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2055 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2056 le32_to_cpu(sb->major_version),
2057 le32_to_cpu(sb->feature_map),
7b75c2f8 2058 uuid,
cd2ac932
CR		 2059 sb->set_name,
2060 (unsigned long long)le64_to_cpu(sb->ctime)
2061 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2062
2063 uuid = sb->device_uuid;
ad361c98
JP		 2064 printk(KERN_INFO
2065 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2066 " RO:%llu\n"
7b75c2f8 2067 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2068 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2069 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2070 le32_to_cpu(sb->level),
2071 (unsigned long long)le64_to_cpu(sb->size),
2072 le32_to_cpu(sb->raid_disks),
2073 le32_to_cpu(sb->layout),
2074 le32_to_cpu(sb->chunksize),
2075 (unsigned long long)le64_to_cpu(sb->data_offset),
2076 (unsigned long long)le64_to_cpu(sb->data_size),
2077 (unsigned long long)le64_to_cpu(sb->super_offset),
2078 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2079 le32_to_cpu(sb->dev_number),
7b75c2f8 2080 uuid,
cd2ac932
CR		 2081 sb->devflags,
2082 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2083 (unsigned long long)le64_to_cpu(sb->events),
2084 (unsigned long long)le64_to_cpu(sb->resync_offset),
2085 le32_to_cpu(sb->sb_csum),
2086 le32_to_cpu(sb->max_dev)
2087 );
1da177e4
LT		 2088}
2089
cd2ac932 2090static void print_rdev(mdk_rdev_t *rdev, int major_version)
1da177e4
LT		 2091{
2092 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2093 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2094 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2095 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2096 rdev->desc_nr);
1da177e4 2097 if (rdev->sb_loaded) {
cd2ac932
CR		 2098 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2099 switch (major_version) {
2100 case 0:
2101 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2102 break;
2103 case 1:
2104 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2105 break;
2106 }
1da177e4
LT		 2107 } else
2108 printk(KERN_INFO "md: no rdev superblock!\n");
2109}
2110
5e56341d 2111static void md_print_devices(void)
1da177e4 2112{
159ec1fc 2113 struct list_head *tmp;
1da177e4
LT		 2114 mdk_rdev_t *rdev;
2115 mddev_t *mddev;
2116 char b[BDEVNAME_SIZE];
2117
2118 printk("\n");
2119 printk("md: **********************************\n");
2120 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2121 printk("md: **********************************\n");
29ac4aa3 2122 for_each_mddev(mddev, tmp) {
1da177e4 2123
32a7627c
N		 2124 if (mddev->bitmap)
2125 bitmap_print_sb(mddev->bitmap);
2126 else
2127 printk("%s: ", mdname(mddev));
159ec1fc 2128 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2129 printk("<%s>", bdevname(rdev->bdev,b));
2130 printk("\n");
2131
159ec1fc 2132 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2133 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2134 }
2135 printk("md: **********************************\n");
2136 printk("\n");
2137}
2138
2139
42543769 2140static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 2141{
42543769
N		 2142 /* Update each superblock (in-memory image), but
2143 * if we are allowed to, skip spares which already
2144 * have the right event counter, or have one earlier
2145 * (which would mean they aren't being marked as dirty
2146 * with the rest of the array)
2147 */
1da177e4 2148 mdk_rdev_t *rdev;
159ec1fc 2149 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2150 if (rdev->sb_events == mddev->events ||
2151 (nospares &&
2152 rdev->raid_disk < 0 &&
42543769
N		 2153 rdev->sb_events+1 == mddev->events)) {
2154 /* Don't update this superblock */
2155 rdev->sb_loaded = 2;
2156 } else {
2157 super_types[mddev->major_version].
2158 sync_super(mddev, rdev);
2159 rdev->sb_loaded = 1;
2160 }
1da177e4
LT		 2161 }
2162}
2163
850b2b42 2164static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 2165{
1da177e4 2166 mdk_rdev_t *rdev;
06d91a5f 2167 int sync_req;
42543769 2168 int nospares = 0;
1da177e4 2169
1da177e4 2170repeat:
3a3a5ddb
N		 2171 /* First make sure individual recovery_offsets are correct */
2172 list_for_each_entry(rdev, &mddev->disks, same_set) {
2173 if (rdev->raid_disk >= 0 &&
2174 mddev->delta_disks >= 0 &&
2175 !test_bit(In_sync, &rdev->flags) &&
2176 mddev->curr_resync_completed > rdev->recovery_offset)
2177 rdev->recovery_offset = mddev->curr_resync_completed;
2178
2179 }
bd52b746 2180 if (!mddev->persistent) {
070dc6dd 2181 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2182 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
d97a41dc
N		 2183 if (!mddev->external)
2184 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3a3a5ddb
N		 2185 wake_up(&mddev->sb_wait);
2186 return;
2187 }
2188
a9701a30 2189 spin_lock_irq(&mddev->write_lock);
84692195 2190
3a3a5ddb
N		 2191 mddev->utime = get_seconds();
2192
850b2b42
N		 2193 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2194 force_change = 1;
2195 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2196 /* just a clean<-> dirty transition, possibly leave spares alone,
2197 * though if events isn't the right even/odd, we will have to do
2198 * spares after all
2199 */
2200 nospares = 1;
2201 if (force_change)
2202 nospares = 0;
2203 if (mddev->degraded)
84692195
N		 2204 /* If the array is degraded, then skipping spares is both
2205 * dangerous and fairly pointless.
2206 * Dangerous because a device that was removed from the array
2207 * might have a event_count that still looks up-to-date,
2208 * so it can be re-added without a resync.
2209 * Pointless because if there are any spares to skip,
2210 * then a recovery will happen and soon that array won't
2211 * be degraded any more and the spare can go back to sleep then.
2212 */
850b2b42 2213 nospares = 0;
84692195 2214
06d91a5f 2215 sync_req = mddev->in_sync;
42543769
N		 2216
2217 /* If this is just a dirty<->clean transition, and the array is clean
2218 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2219 if (nospares
42543769 2220 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2221 && mddev->can_decrease_events
2222 && mddev->events != 1) {
42543769 2223 mddev->events--;
a8707c08
N		 2224 mddev->can_decrease_events = 0;
2225 } else {
42543769
N		 2226 /* otherwise we have to go forward and ... */
2227 mddev->events ++;
a8707c08 2228 mddev->can_decrease_events = nospares;
42543769 2229 }
1da177e4
LT		 2230
2231 if (!mddev->events) {
2232 /*
2233 * oops, this 64-bit counter should never wrap.
2234 * Either we are in around ~1 trillion A.C., assuming
2235 * 1 reboot per second, or we have a bug:
2236 */
2237 MD_BUG();
2238 mddev->events --;
2239 }
e691063a 2240 sync_sbs(mddev, nospares);
a9701a30 2241 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 2242
2243 dprintk(KERN_INFO
2244 "md: updating %s RAID superblock on device (in sync %d)\n",
2245 mdname(mddev),mddev->in_sync);
2246
4ad13663 2247 bitmap_update_sb(mddev->bitmap);
159ec1fc 2248 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 2249 char b[BDEVNAME_SIZE];
2250 dprintk(KERN_INFO "md: ");
42543769
N		 2251 if (rdev->sb_loaded != 1)
2252 continue; /* no noise on spare devices */
b2d444d7 2253 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2254 dprintk("(skipping faulty ");
2255
2256 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 2257 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2258 md_super_write(mddev,rdev,
0f420358 2259 rdev->sb_start, rdev->sb_size,
7bfa19f2
N		 2260 rdev->sb_page);
2261 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2262 bdevname(rdev->bdev,b),
0f420358 2263 (unsigned long long)rdev->sb_start);
42543769 2264 rdev->sb_events = mddev->events;
7bfa19f2 2265
1da177e4
LT		 2266 } else
2267 dprintk(")\n");
7bfa19f2 2268 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2269 /* only need to write one superblock... */
2270 break;
2271 }
a9701a30 2272 md_super_wait(mddev);
850b2b42 2273 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2274
a9701a30 2275 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2276 if (mddev->in_sync != sync_req ||
2277 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2278 /* have to write it out again */
a9701a30 2279 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2280 goto repeat;
2281 }
850b2b42 2282 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2283 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2284 wake_up(&mddev->sb_wait);
acb180b0
N		 2285 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2286 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2287
1da177e4
LT		 2288}
2289
7f6ce769 2290/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2291 * We want to accept with case. For this we use cmd_match.
2292 */
2293static int cmd_match(const char *cmd, const char *str)
2294{
2295 /* See if cmd, written into a sysfs file, matches
2296 * str. They must either be the same, or cmd can
2297 * have a trailing newline
2298 */
2299 while (*cmd && *str && *cmd == *str) {
2300 cmd++;
2301 str++;
2302 }
2303 if (*cmd == '\n')
2304 cmd++;
2305 if (*str || *cmd)
2306 return 0;
2307 return 1;
2308}
2309
86e6ffdd
N		 2310struct rdev_sysfs_entry {
2311 struct attribute attr;
2312 ssize_t (*show)(mdk_rdev_t *, char *);
2313 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2314};
2315
2316static ssize_t
96de1e66 2317state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 2318{
2319 char *sep = "";
20a49ff6 2320 size_t len = 0;
86e6ffdd 2321
b2d444d7 2322 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 2323 len+= sprintf(page+len, "%sfaulty",sep);
2324 sep = ",";
2325 }
b2d444d7 2326 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2327 len += sprintf(page+len, "%sin_sync",sep);
2328 sep = ",";
2329 }
f655675b
N		 2330 if (test_bit(WriteMostly, &rdev->flags)) {
2331 len += sprintf(page+len, "%swrite_mostly",sep);
2332 sep = ",";
2333 }
6bfe0b49
DW		 2334 if (test_bit(Blocked, &rdev->flags)) {
2335 len += sprintf(page+len, "%sblocked", sep);
2336 sep = ",";
2337 }
b2d444d7
N		 2338 if (!test_bit(Faulty, &rdev->flags) &&
2339 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2340 len += sprintf(page+len, "%sspare", sep);
2341 sep = ",";
2342 }
2343 return len+sprintf(page+len, "\n");
2344}
2345
45dc2de1
N		 2346static ssize_t
2347state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2348{
2349 /* can write
2350 * faulty - simulates and error
2351 * remove - disconnects the device
f655675b
N		 2352 * writemostly - sets write_mostly
2353 * -writemostly - clears write_mostly
6bfe0b49
DW		 2354 * blocked - sets the Blocked flag
2355 * -blocked - clears the Blocked flag
6d56e278 2356 * insync - sets Insync providing device isn't active
45dc2de1
N		 2357 */
2358 int err = -EINVAL;
2359 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2360 md_error(rdev->mddev, rdev);
2361 err = 0;
2362 } else if (cmd_match(buf, "remove")) {
2363 if (rdev->raid_disk >= 0)
2364 err = -EBUSY;
2365 else {
2366 mddev_t *mddev = rdev->mddev;
2367 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2368 if (mddev->pers)
2369 md_update_sb(mddev, 1);
45dc2de1
N		 2370 md_new_event(mddev);
2371 err = 0;
2372 }
f655675b
N		 2373 } else if (cmd_match(buf, "writemostly")) {
2374 set_bit(WriteMostly, &rdev->flags);
2375 err = 0;
2376 } else if (cmd_match(buf, "-writemostly")) {
2377 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2378 err = 0;
2379 } else if (cmd_match(buf, "blocked")) {
2380 set_bit(Blocked, &rdev->flags);
2381 err = 0;
2382 } else if (cmd_match(buf, "-blocked")) {
2383 clear_bit(Blocked, &rdev->flags);
2384 wake_up(&rdev->blocked_wait);
2385 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2386 md_wakeup_thread(rdev->mddev->thread);
2387
6d56e278
N		 2388 err = 0;
2389 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2390 set_bit(In_sync, &rdev->flags);
f655675b 2391 err = 0;
45dc2de1 2392 }
00bcb4ac
N		 2393 if (!err)
2394 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2395 return err ? err : len;
2396}
80ca3a44
N		 2397static struct rdev_sysfs_entry rdev_state =
2398__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2399
4dbcdc75
N		 2400static ssize_t
2401errors_show(mdk_rdev_t *rdev, char *page)
2402{
2403 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2404}
2405
2406static ssize_t
2407errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2408{
2409 char *e;
2410 unsigned long n = simple_strtoul(buf, &e, 10);
2411 if (*buf && (*e == 0 || *e == '\n')) {
2412 atomic_set(&rdev->corrected_errors, n);
2413 return len;
2414 }
2415 return -EINVAL;
2416}
2417static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2418__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2419
014236d2
N		 2420static ssize_t
2421slot_show(mdk_rdev_t *rdev, char *page)
2422{
2423 if (rdev->raid_disk < 0)
2424 return sprintf(page, "none\n");
2425 else
2426 return sprintf(page, "%d\n", rdev->raid_disk);
2427}
2428
2429static ssize_t
2430slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2431{
2432 char *e;
c303da6d
N		 2433 int err;
2434 char nm[20];
014236d2
N		 2435 int slot = simple_strtoul(buf, &e, 10);
2436 if (strncmp(buf, "none", 4)==0)
2437 slot = -1;
2438 else if (e==buf || (*e && *e!= '\n'))
2439 return -EINVAL;
6c2fce2e 2440 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2441 /* Setting 'slot' on an active array requires also
2442 * updating the 'rd%d' link, and communicating
2443 * with the personality with ->hot_*_disk.
2444 * For now we only support removing
2445 * failed/spare devices. This normally happens automatically,
2446 * but not when the metadata is externally managed.
2447 */
c303da6d
N		 2448 if (rdev->raid_disk == -1)
2449 return -EEXIST;
2450 /* personality does all needed checks */
2451 if (rdev->mddev->pers->hot_add_disk == NULL)
2452 return -EINVAL;
2453 err = rdev->mddev->pers->
2454 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2455 if (err)
2456 return err;
2457 sprintf(nm, "rd%d", rdev->raid_disk);
2458 sysfs_remove_link(&rdev->mddev->kobj, nm);
b7103107 2459 rdev->raid_disk = -1;
c303da6d
N		 2460 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2461 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e
NB		 2462 } else if (rdev->mddev->pers) {
2463 mdk_rdev_t *rdev2;
6c2fce2e 2464 /* Activating a spare .. or possibly reactivating
6d56e278 2465 * if we ever get bitmaps working here.
6c2fce2e
NB		 2466 */
2467
2468 if (rdev->raid_disk != -1)
2469 return -EBUSY;
2470
2471 if (rdev->mddev->pers->hot_add_disk == NULL)
2472 return -EINVAL;
2473
159ec1fc 2474 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2475 if (rdev2->raid_disk == slot)
2476 return -EEXIST;
2477
2478 rdev->raid_disk = slot;
2479 if (test_bit(In_sync, &rdev->flags))
2480 rdev->saved_raid_disk = slot;
2481 else
2482 rdev->saved_raid_disk = -1;
2483 err = rdev->mddev->pers->
2484 hot_add_disk(rdev->mddev, rdev);
199050ea 2485 if (err) {
6c2fce2e 2486 rdev->raid_disk = -1;
6c2fce2e 2487 return err;
52664732 2488 } else
00bcb4ac 2489 sysfs_notify_dirent_safe(rdev->sysfs_state);
6c2fce2e
NB		 2490 sprintf(nm, "rd%d", rdev->raid_disk);
2491 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
00bcb4ac 2492 /* failure here is OK */;
6c2fce2e 2493 /* don't wakeup anyone, leave that to userspace. */
c303da6d
N		 2494 } else {
2495 if (slot >= rdev->mddev->raid_disks)
2496 return -ENOSPC;
2497 rdev->raid_disk = slot;
2498 /* assume it is working */
c5d79adb
N		 2499 clear_bit(Faulty, &rdev->flags);
2500 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2501 set_bit(In_sync, &rdev->flags);
00bcb4ac 2502 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2503 }
014236d2
N		 2504 return len;
2505}
2506
2507
2508static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2509__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2510
93c8cad0
N		 2511static ssize_t
2512offset_show(mdk_rdev_t *rdev, char *page)
2513{
6961ece4 2514 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2515}
2516
2517static ssize_t
2518offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2519{
2520 char *e;
2521 unsigned long long offset = simple_strtoull(buf, &e, 10);
2522 if (e==buf || (*e && *e != '\n'))
2523 return -EINVAL;
8ed0a521 2524 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2525 return -EBUSY;
dd8ac336 2526 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2527 /* Must set offset before size, so overlap checks
2528 * can be sane */
2529 return -EBUSY;
93c8cad0
N		 2530 rdev->data_offset = offset;
2531 return len;
2532}
2533
2534static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2535__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2536
83303b61
N		 2537static ssize_t
2538rdev_size_show(mdk_rdev_t *rdev, char *page)
2539{
dd8ac336 2540 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2541}
2542
c5d79adb
N		 2543static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2544{
2545 /* check if two start/length pairs overlap */
2546 if (s1+l1 <= s2)
2547 return 0;
2548 if (s2+l2 <= s1)
2549 return 0;
2550 return 1;
2551}
2552
b522adcd
DW		 2553static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2554{
2555 unsigned long long blocks;
2556 sector_t new;
2557
2558 if (strict_strtoull(buf, 10, &blocks) < 0)
2559 return -EINVAL;
2560
2561 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2562 return -EINVAL; /* sector conversion overflow */
2563
2564 new = blocks * 2;
2565 if (new != blocks * 2)
2566 return -EINVAL; /* unsigned long long to sector_t overflow */
2567
2568 *sectors = new;
2569 return 0;
2570}
2571
83303b61
N		 2572static ssize_t
2573rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2574{
27c529bb 2575 mddev_t *my_mddev = rdev->mddev;
dd8ac336 2576 sector_t oldsectors = rdev->sectors;
b522adcd 2577 sector_t sectors;
27c529bb 2578
b522adcd 2579 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2580 return -EINVAL;
0cd17fec 2581 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2582 if (my_mddev->persistent) {
dd8ac336
AN		 2583 sectors = super_types[my_mddev->major_version].
2584 rdev_size_change(rdev, sectors);
2585 if (!sectors)
0cd17fec 2586 return -EBUSY;
dd8ac336 2587 } else if (!sectors)
77304d2a 2588 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2589 rdev->data_offset;
0cd17fec 2590 }
dd8ac336 2591 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2592 return -EINVAL; /* component must fit device */
0cd17fec 2593
dd8ac336
AN		 2594 rdev->sectors = sectors;
2595 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2596 /* need to check that all other rdevs with the same ->bdev
2597 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2598 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2599 * we have to change it back, we will have the lock again.
2600 */
2601 mddev_t *mddev;
2602 int overlap = 0;
159ec1fc 2603 struct list_head *tmp;
c5d79adb 2604
27c529bb 2605 mddev_unlock(my_mddev);
29ac4aa3 2606 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2607 mdk_rdev_t *rdev2;
2608
2609 mddev_lock(mddev);
159ec1fc 2610 list_for_each_entry(rdev2, &mddev->disks, same_set)
c5d79adb
N		 2611 if (test_bit(AllReserved, &rdev2->flags) ||
2612 (rdev->bdev == rdev2->bdev &&
2613 rdev != rdev2 &&
dd8ac336 2614 overlaps(rdev->data_offset, rdev->sectors,
d07bd3bc 2615 rdev2->data_offset,
dd8ac336 2616 rdev2->sectors))) {
c5d79adb
N		 2617 overlap = 1;
2618 break;
2619 }
2620 mddev_unlock(mddev);
2621 if (overlap) {
2622 mddev_put(mddev);
2623 break;
2624 }
2625 }
27c529bb 2626 mddev_lock(my_mddev);
c5d79adb
N		 2627 if (overlap) {
2628 /* Someone else could have slipped in a size
2629 * change here, but doing so is just silly.
dd8ac336 2630 * We put oldsectors back because we *know* it is
c5d79adb
N		 2631 * safe, and trust userspace not to race with
2632 * itself
2633 */
dd8ac336 2634 rdev->sectors = oldsectors;
c5d79adb
N		 2635 return -EBUSY;
2636 }
2637 }
83303b61
N		 2638 return len;
2639}
2640
2641static struct rdev_sysfs_entry rdev_size =
80ca3a44 2642__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2643
06e3c817
DW		 2644
2645static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2646{
2647 unsigned long long recovery_start = rdev->recovery_offset;
2648
2649 if (test_bit(In_sync, &rdev->flags) ||
2650 recovery_start == MaxSector)
2651 return sprintf(page, "none\n");
2652
2653 return sprintf(page, "%llu\n", recovery_start);
2654}
2655
2656static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2657{
2658 unsigned long long recovery_start;
2659
2660 if (cmd_match(buf, "none"))
2661 recovery_start = MaxSector;
2662 else if (strict_strtoull(buf, 10, &recovery_start))
2663 return -EINVAL;
2664
2665 if (rdev->mddev->pers &&
2666 rdev->raid_disk >= 0)
2667 return -EBUSY;
2668
2669 rdev->recovery_offset = recovery_start;
2670 if (recovery_start == MaxSector)
2671 set_bit(In_sync, &rdev->flags);
2672 else
2673 clear_bit(In_sync, &rdev->flags);
2674 return len;
2675}
2676
2677static struct rdev_sysfs_entry rdev_recovery_start =
2678__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2679
86e6ffdd
N		 2680static struct attribute *rdev_default_attrs[] = {
2681 &rdev_state.attr,
4dbcdc75 2682 &rdev_errors.attr,
014236d2 2683 &rdev_slot.attr,
93c8cad0 2684 &rdev_offset.attr,
83303b61 2685 &rdev_size.attr,
06e3c817 2686 &rdev_recovery_start.attr,
86e6ffdd
N		 2687 NULL,
2688};
2689static ssize_t
2690rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2691{
2692 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2693 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2694 mddev_t *mddev = rdev->mddev;
2695 ssize_t rv;
86e6ffdd
N		 2696
2697 if (!entry->show)
2698 return -EIO;
27c529bb
N		 2699
2700 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2701 if (!rv) {
2702 if (rdev->mddev == NULL)
2703 rv = -EBUSY;
2704 else
2705 rv = entry->show(rdev, page);
2706 mddev_unlock(mddev);
2707 }
2708 return rv;
86e6ffdd
N		 2709}
2710
2711static ssize_t
2712rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2713 const char *page, size_t length)
2714{
2715 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2716 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2717 ssize_t rv;
2718 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2719
2720 if (!entry->store)
2721 return -EIO;
67463acb
N		 2722 if (!capable(CAP_SYS_ADMIN))
2723 return -EACCES;
27c529bb 2724 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2725 if (!rv) {
27c529bb
N		 2726 if (rdev->mddev == NULL)
2727 rv = -EBUSY;
2728 else
2729 rv = entry->store(rdev, page, length);
6a51830e 2730 mddev_unlock(mddev);
ca388059
N		 2731 }
2732 return rv;
86e6ffdd
N		 2733}
2734
2735static void rdev_free(struct kobject *ko)
2736{
2737 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2738 kfree(rdev);
2739}
52cf25d0 2740static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 2741 .show = rdev_attr_show,
2742 .store = rdev_attr_store,
2743};
2744static struct kobj_type rdev_ktype = {
2745 .release = rdev_free,
2746 .sysfs_ops = &rdev_sysfs_ops,
2747 .default_attrs = rdev_default_attrs,
2748};
2749
e8bb9a83
N		 2750void md_rdev_init(mdk_rdev_t *rdev)
2751{
2752 rdev->desc_nr = -1;
2753 rdev->saved_raid_disk = -1;
2754 rdev->raid_disk = -1;
2755 rdev->flags = 0;
2756 rdev->data_offset = 0;
2757 rdev->sb_events = 0;
2758 rdev->last_read_error.tv_sec = 0;
2759 rdev->last_read_error.tv_nsec = 0;
2760 atomic_set(&rdev->nr_pending, 0);
2761 atomic_set(&rdev->read_errors, 0);
2762 atomic_set(&rdev->corrected_errors, 0);
2763
2764 INIT_LIST_HEAD(&rdev->same_set);
2765 init_waitqueue_head(&rdev->blocked_wait);
2766}
2767EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 2768/*
2769 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2770 *
2771 * mark the device faulty if:
2772 *
2773 * - the device is nonexistent (zero size)
2774 * - the device has no valid superblock
2775 *
2776 * a faulty rdev _never_ has rdev->sb set.
2777 */
2778static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2779{
2780 char b[BDEVNAME_SIZE];
2781 int err;
2782 mdk_rdev_t *rdev;
2783 sector_t size;
2784
9ffae0cf 2785 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2786 if (!rdev) {
2787 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2788 return ERR_PTR(-ENOMEM);
2789 }
1da177e4 2790
e8bb9a83 2791 md_rdev_init(rdev);
1da177e4
LT		 2792 if ((err = alloc_disk_sb(rdev)))
2793 goto abort_free;
2794
c5d79adb 2795 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2796 if (err)
2797 goto abort_free;
2798
f9cb074b 2799 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2800
77304d2a 2801 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 2802 if (!size) {
2803 printk(KERN_WARNING
2804 "md: %s has zero or unknown size, marking faulty!\n",
2805 bdevname(rdev->bdev,b));
2806 err = -EINVAL;
2807 goto abort_free;
2808 }
2809
2810 if (super_format >= 0) {
2811 err = super_types[super_format].
2812 load_super(rdev, NULL, super_minor);
2813 if (err == -EINVAL) {
df968c4e
N		 2814 printk(KERN_WARNING
2815 "md: %s does not have a valid v%d.%d "
2816 "superblock, not importing!\n",
2817 bdevname(rdev->bdev,b),
2818 super_format, super_minor);
1da177e4
LT		 2819 goto abort_free;
2820 }
2821 if (err < 0) {
2822 printk(KERN_WARNING
2823 "md: could not read %s's sb, not importing!\n",
2824 bdevname(rdev->bdev,b));
2825 goto abort_free;
2826 }
2827 }
6bfe0b49 2828
1da177e4
LT		 2829 return rdev;
2830
2831abort_free:
2832 if (rdev->sb_page) {
2833 if (rdev->bdev)
2834 unlock_rdev(rdev);
2835 free_disk_sb(rdev);
2836 }
2837 kfree(rdev);
2838 return ERR_PTR(err);
2839}
2840
2841/*
2842 * Check a full RAID array for plausibility
2843 */
2844
2845
a757e64c 2846static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2847{
2848 int i;
159ec1fc 2849 mdk_rdev_t *rdev, *freshest, *tmp;
1da177e4
LT		 2850 char b[BDEVNAME_SIZE];
2851
2852 freshest = NULL;
d089c6af 2853 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2854 switch (super_types[mddev->major_version].
2855 load_super(rdev, freshest, mddev->minor_version)) {
2856 case 1:
2857 freshest = rdev;
2858 break;
2859 case 0:
2860 break;
2861 default:
2862 printk( KERN_ERR \
2863 "md: fatal superblock inconsistency in %s"
2864 " -- removing from array\n",
2865 bdevname(rdev->bdev,b));
2866 kick_rdev_from_array(rdev);
2867 }
2868
2869
2870 super_types[mddev->major_version].
2871 validate_super(mddev, freshest);
2872
2873 i = 0;
d089c6af 2874 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 2875 if (mddev->max_disks &&
2876 (rdev->desc_nr >= mddev->max_disks ||
2877 i > mddev->max_disks)) {
de01dfad
N		 2878 printk(KERN_WARNING
2879 "md: %s: %s: only %d devices permitted\n",
2880 mdname(mddev), bdevname(rdev->bdev, b),
2881 mddev->max_disks);
2882 kick_rdev_from_array(rdev);
2883 continue;
2884 }
1da177e4
LT		 2885 if (rdev != freshest)
2886 if (super_types[mddev->major_version].
2887 validate_super(mddev, rdev)) {
2888 printk(KERN_WARNING "md: kicking non-fresh %s"
2889 " from array!\n",
2890 bdevname(rdev->bdev,b));
2891 kick_rdev_from_array(rdev);
2892 continue;
2893 }
2894 if (mddev->level == LEVEL_MULTIPATH) {
2895 rdev->desc_nr = i++;
2896 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2897 set_bit(In_sync, &rdev->flags);
5e5e3e78 2898 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 2899 rdev->raid_disk = -1;
2900 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2901 }
2902 }
1da177e4
LT		 2903}
2904
72e02075
N		 2905/* Read a fixed-point number.
2906 * Numbers in sysfs attributes should be in "standard" units where
2907 * possible, so time should be in seconds.
2908 * However we internally use a a much smaller unit such as
2909 * milliseconds or jiffies.
2910 * This function takes a decimal number with a possible fractional
2911 * component, and produces an integer which is the result of
2912 * multiplying that number by 10^'scale'.
2913 * all without any floating-point arithmetic.
2914 */
2915int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2916{
2917 unsigned long result = 0;
2918 long decimals = -1;
2919 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2920 if (*cp == '.')
2921 decimals = 0;
2922 else if (decimals < scale) {
2923 unsigned int value;
2924 value = *cp - '0';
2925 result = result * 10 + value;
2926 if (decimals >= 0)
2927 decimals++;
2928 }
2929 cp++;
2930 }
2931 if (*cp == '\n')
2932 cp++;
2933 if (*cp)
2934 return -EINVAL;
2935 if (decimals < 0)
2936 decimals = 0;
2937 while (decimals < scale) {
2938 result *= 10;
2939 decimals ++;
2940 }
2941 *res = result;
2942 return 0;
2943}
2944
2945
19052c0e
N		 2946static void md_safemode_timeout(unsigned long data);
2947
16f17b39
N		 2948static ssize_t
2949safe_delay_show(mddev_t *mddev, char *page)
2950{
2951 int msec = (mddev->safemode_delay*1000)/HZ;
2952 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2953}
2954static ssize_t
2955safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2956{
16f17b39 2957 unsigned long msec;
97ce0a7f 2958
72e02075 2959 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 2960 return -EINVAL;
16f17b39
N		 2961 if (msec == 0)
2962 mddev->safemode_delay = 0;
2963 else {
19052c0e 2964 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 2965 mddev->safemode_delay = (msec*HZ)/1000;
2966 if (mddev->safemode_delay == 0)
2967 mddev->safemode_delay = 1;
19052c0e
N		 2968 if (mddev->safemode_delay < old_delay)
2969 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 2970 }
2971 return len;
2972}
2973static struct md_sysfs_entry md_safe_delay =
80ca3a44 2974__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2975
eae1701f 2976static ssize_t
96de1e66 2977level_show(mddev_t *mddev, char *page)
eae1701f 2978{
2604b703 2979 struct mdk_personality *p = mddev->pers;
d9d166c2 2980 if (p)
eae1701f 2981 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2982 else if (mddev->clevel[0])
2983 return sprintf(page, "%s\n", mddev->clevel);
2984 else if (mddev->level != LEVEL_NONE)
2985 return sprintf(page, "%d\n", mddev->level);
2986 else
2987 return 0;
eae1701f
N		 2988}
2989
d9d166c2
N		 2990static ssize_t
2991level_store(mddev_t *mddev, const char *buf, size_t len)
2992{
f2859af6 2993 char clevel[16];
20a49ff6 2994 ssize_t rv = len;
245f46c2 2995 struct mdk_personality *pers;
f2859af6 2996 long level;
245f46c2 2997 void *priv;
3a981b03 2998 mdk_rdev_t *rdev;
245f46c2
N		 2999
3000 if (mddev->pers == NULL) {
3001 if (len == 0)
3002 return 0;
3003 if (len >= sizeof(mddev->clevel))
3004 return -ENOSPC;
3005 strncpy(mddev->clevel, buf, len);
3006 if (mddev->clevel[len-1] == '\n')
3007 len--;
3008 mddev->clevel[len] = 0;
3009 mddev->level = LEVEL_NONE;
3010 return rv;
3011 }
3012
3013 /* request to change the personality. Need to ensure:
3014 * - array is not engaged in resync/recovery/reshape
3015 * - old personality can be suspended
3016 * - new personality will access other array.
3017 */
3018
bb4f1e9d
N		 3019 if (mddev->sync_thread ||
3020 mddev->reshape_position != MaxSector ||
3021 mddev->sysfs_active)
d9d166c2 3022 return -EBUSY;
245f46c2
N		 3023
3024 if (!mddev->pers->quiesce) {
3025 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3026 mdname(mddev), mddev->pers->name);
3027 return -EINVAL;
3028 }
3029
3030 /* Now find the new personality */
f2859af6 3031 if (len == 0 || len >= sizeof(clevel))
245f46c2 3032 return -EINVAL;
f2859af6
DW		 3033 strncpy(clevel, buf, len);
3034 if (clevel[len-1] == '\n')
d9d166c2 3035 len--;
f2859af6
DW		 3036 clevel[len] = 0;
3037 if (strict_strtol(clevel, 10, &level))
3038 level = LEVEL_NONE;
245f46c2 3039
f2859af6
DW		 3040 if (request_module("md-%s", clevel) != 0)
3041 request_module("md-level-%s", clevel);
245f46c2 3042 spin_lock(&pers_lock);
f2859af6 3043 pers = find_pers(level, clevel);
245f46c2
N		 3044 if (!pers || !try_module_get(pers->owner)) {
3045 spin_unlock(&pers_lock);
f2859af6 3046 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3047 return -EINVAL;
3048 }
3049 spin_unlock(&pers_lock);
3050
3051 if (pers == mddev->pers) {
3052 /* Nothing to do! */
3053 module_put(pers->owner);
3054 return rv;
3055 }
3056 if (!pers->takeover) {
3057 module_put(pers->owner);
3058 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3059 mdname(mddev), clevel);
245f46c2
N		 3060 return -EINVAL;
3061 }
3062
e93f68a1
N		 3063 list_for_each_entry(rdev, &mddev->disks, same_set)
3064 rdev->new_raid_disk = rdev->raid_disk;
3065
245f46c2
N		 3066 /* ->takeover must set new_* and/or delta_disks
3067 * if it succeeds, and may set them when it fails.
3068 */
3069 priv = pers->takeover(mddev);
3070 if (IS_ERR(priv)) {
3071 mddev->new_level = mddev->level;
3072 mddev->new_layout = mddev->layout;
664e7c41 3073 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3074 mddev->raid_disks -= mddev->delta_disks;
3075 mddev->delta_disks = 0;
3076 module_put(pers->owner);
3077 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3078 mdname(mddev), clevel);
245f46c2
N		 3079 return PTR_ERR(priv);
3080 }
3081
3082 /* Looks like we have a winner */
3083 mddev_suspend(mddev);
3084 mddev->pers->stop(mddev);
a64c876f
N		 3085
3086 if (mddev->pers->sync_request == NULL &&
3087 pers->sync_request != NULL) {
3088 /* need to add the md_redundancy_group */
3089 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3090 printk(KERN_WARNING
3091 "md: cannot register extra attributes for %s\n",
3092 mdname(mddev));
19fdb9ee 3093 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3094 }
3095 if (mddev->pers->sync_request != NULL &&
3096 pers->sync_request == NULL) {
3097 /* need to remove the md_redundancy_group */
3098 if (mddev->to_remove == NULL)
3099 mddev->to_remove = &md_redundancy_group;
3100 }
3101
54071b38
TM		 3102 if (mddev->pers->sync_request == NULL &&
3103 mddev->external) {
3104 /* We are converting from a no-redundancy array
3105 * to a redundancy array and metadata is managed
3106 * externally so we need to be sure that writes
3107 * won't block due to a need to transition
3108 * clean->dirty
3109 * until external management is started.
3110 */
3111 mddev->in_sync = 0;
3112 mddev->safemode_delay = 0;
3113 mddev->safemode = 0;
3114 }
3115
e93f68a1
N		 3116 list_for_each_entry(rdev, &mddev->disks, same_set) {
3117 char nm[20];
3118 if (rdev->raid_disk < 0)
3119 continue;
3120 if (rdev->new_raid_disk > mddev->raid_disks)
3121 rdev->new_raid_disk = -1;
3122 if (rdev->new_raid_disk == rdev->raid_disk)
3123 continue;
3124 sprintf(nm, "rd%d", rdev->raid_disk);
3125 sysfs_remove_link(&mddev->kobj, nm);
3126 }
3127 list_for_each_entry(rdev, &mddev->disks, same_set) {
3128 if (rdev->raid_disk < 0)
3129 continue;
3130 if (rdev->new_raid_disk == rdev->raid_disk)
3131 continue;
3132 rdev->raid_disk = rdev->new_raid_disk;
3133 if (rdev->raid_disk < 0)
3a981b03 3134 clear_bit(In_sync, &rdev->flags);
e93f68a1
N		 3135 else {
3136 char nm[20];
3137 sprintf(nm, "rd%d", rdev->raid_disk);
3138 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3139 printk("md: cannot register %s for %s after level change\n",
3140 nm, mdname(mddev));
3a981b03 3141 }
e93f68a1
N		 3142 }
3143
3144 module_put(mddev->pers->owner);
245f46c2
N		 3145 mddev->pers = pers;
3146 mddev->private = priv;
3147 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3148 mddev->level = mddev->new_level;
3149 mddev->layout = mddev->new_layout;
664e7c41 3150 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3151 mddev->delta_disks = 0;
9af204cf
TM		 3152 if (mddev->pers->sync_request == NULL) {
3153 /* this is now an array without redundancy, so
3154 * it must always be in_sync
3155 */
3156 mddev->in_sync = 1;
3157 del_timer_sync(&mddev->safemode_timer);
3158 }
245f46c2
N		 3159 pers->run(mddev);
3160 mddev_resume(mddev);
3161 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3162 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3163 md_wakeup_thread(mddev->thread);
5cac7861 3164 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3165 md_new_event(mddev);
d9d166c2
N		 3166 return rv;
3167}
3168
3169static struct md_sysfs_entry md_level =
80ca3a44 3170__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3171
d4dbd025
N		 3172
3173static ssize_t
3174layout_show(mddev_t *mddev, char *page)
3175{
3176 /* just a number, not meaningful for all levels */
08a02ecd
N		 3177 if (mddev->reshape_position != MaxSector &&
3178 mddev->layout != mddev->new_layout)
3179 return sprintf(page, "%d (%d)\n",
3180 mddev->new_layout, mddev->layout);
d4dbd025
N		 3181 return sprintf(page, "%d\n", mddev->layout);
3182}
3183
3184static ssize_t
3185layout_store(mddev_t *mddev, const char *buf, size_t len)
3186{
3187 char *e;
3188 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3189
3190 if (!*buf || (*e && *e != '\n'))
3191 return -EINVAL;
3192
b3546035
N		 3193 if (mddev->pers) {
3194 int err;
50ac168a 3195 if (mddev->pers->check_reshape == NULL)
b3546035 3196 return -EBUSY;
597a711b 3197 mddev->new_layout = n;
50ac168a 3198 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3199 if (err) {
3200 mddev->new_layout = mddev->layout;
b3546035 3201 return err;
597a711b 3202 }
b3546035 3203 } else {
08a02ecd 3204 mddev->new_layout = n;
b3546035
N		 3205 if (mddev->reshape_position == MaxSector)
3206 mddev->layout = n;
3207 }
d4dbd025
N		 3208 return len;
3209}
3210static struct md_sysfs_entry md_layout =
80ca3a44 3211__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3212
3213
eae1701f 3214static ssize_t
96de1e66 3215raid_disks_show(mddev_t *mddev, char *page)
eae1701f 3216{
bb636547
N		 3217 if (mddev->raid_disks == 0)
3218 return 0;
08a02ecd
N		 3219 if (mddev->reshape_position != MaxSector &&
3220 mddev->delta_disks != 0)
3221 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3222 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3223 return sprintf(page, "%d\n", mddev->raid_disks);
3224}
3225
da943b99
N		 3226static int update_raid_disks(mddev_t *mddev, int raid_disks);
3227
3228static ssize_t
3229raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3230{
da943b99
N		 3231 char *e;
3232 int rv = 0;
3233 unsigned long n = simple_strtoul(buf, &e, 10);
3234
3235 if (!*buf || (*e && *e != '\n'))
3236 return -EINVAL;
3237
3238 if (mddev->pers)
3239 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3240 else if (mddev->reshape_position != MaxSector) {
3241 int olddisks = mddev->raid_disks - mddev->delta_disks;
3242 mddev->delta_disks = n - olddisks;
3243 mddev->raid_disks = n;
3244 } else
da943b99
N		 3245 mddev->raid_disks = n;
3246 return rv ? rv : len;
3247}
3248static struct md_sysfs_entry md_raid_disks =
80ca3a44 3249__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3250
3b34380a
N		 3251static ssize_t
3252chunk_size_show(mddev_t *mddev, char *page)
3253{
08a02ecd 3254 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3255 mddev->chunk_sectors != mddev->new_chunk_sectors)
3256 return sprintf(page, "%d (%d)\n",
3257 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3258 mddev->chunk_sectors << 9);
3259 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3260}
3261
3262static ssize_t
3263chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3264{
3b34380a
N		 3265 char *e;
3266 unsigned long n = simple_strtoul(buf, &e, 10);
3267
3b34380a
N		 3268 if (!*buf || (*e && *e != '\n'))
3269 return -EINVAL;
3270
b3546035
N		 3271 if (mddev->pers) {
3272 int err;
50ac168a 3273 if (mddev->pers->check_reshape == NULL)
b3546035 3274 return -EBUSY;
597a711b 3275 mddev->new_chunk_sectors = n >> 9;
50ac168a 3276 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3277 if (err) {
3278 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3279 return err;
597a711b 3280 }
b3546035 3281 } else {
664e7c41 3282 mddev->new_chunk_sectors = n >> 9;
b3546035 3283 if (mddev->reshape_position == MaxSector)
9d8f0363 3284 mddev->chunk_sectors = n >> 9;
b3546035 3285 }
3b34380a
N		 3286 return len;
3287}
3288static struct md_sysfs_entry md_chunk_size =
80ca3a44 3289__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3290
a94213b1
N		 3291static ssize_t
3292resync_start_show(mddev_t *mddev, char *page)
3293{
d1a7c503
N		 3294 if (mddev->recovery_cp == MaxSector)
3295 return sprintf(page, "none\n");
a94213b1
N		 3296 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3297}
3298
3299static ssize_t
3300resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3301{
a94213b1
N		 3302 char *e;
3303 unsigned long long n = simple_strtoull(buf, &e, 10);
3304
3305 if (mddev->pers)
3306 return -EBUSY;
06e3c817
DW		 3307 if (cmd_match(buf, "none"))
3308 n = MaxSector;
3309 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3310 return -EINVAL;
3311
3312 mddev->recovery_cp = n;
3313 return len;
3314}
3315static struct md_sysfs_entry md_resync_start =
80ca3a44 3316__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3317
9e653b63
N		 3318/*
3319 * The array state can be:
3320 *
3321 * clear
3322 * No devices, no size, no level
3323 * Equivalent to STOP_ARRAY ioctl
3324 * inactive
3325 * May have some settings, but array is not active
3326 * all IO results in error
3327 * When written, doesn't tear down array, but just stops it
3328 * suspended (not supported yet)
3329 * All IO requests will block. The array can be reconfigured.
910d8cb3 3330 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3331 * readonly
3332 * no resync can happen. no superblocks get written.
3333 * write requests fail
3334 * read-auto
3335 * like readonly, but behaves like 'clean' on a write request.
3336 *
3337 * clean - no pending writes, but otherwise active.
3338 * When written to inactive array, starts without resync
3339 * If a write request arrives then
3340 * if metadata is known, mark 'dirty' and switch to 'active'.
3341 * if not known, block and switch to write-pending
3342 * If written to an active array that has pending writes, then fails.
3343 * active
3344 * fully active: IO and resync can be happening.
3345 * When written to inactive array, starts with resync
3346 *
3347 * write-pending
3348 * clean, but writes are blocked waiting for 'active' to be written.
3349 *
3350 * active-idle
3351 * like active, but no writes have been seen for a while (100msec).
3352 *
3353 */
3354enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3355 write_pending, active_idle, bad_word};
05381954 3356static char *array_states[] = {
9e653b63
N		 3357 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3358 "write-pending", "active-idle", NULL };
3359
3360static int match_word(const char *word, char **list)
3361{
3362 int n;
3363 for (n=0; list[n]; n++)
3364 if (cmd_match(word, list[n]))
3365 break;
3366 return n;
3367}
3368
3369static ssize_t
3370array_state_show(mddev_t *mddev, char *page)
3371{
3372 enum array_state st = inactive;
3373
3374 if (mddev->pers)
3375 switch(mddev->ro) {
3376 case 1:
3377 st = readonly;
3378 break;
3379 case 2:
3380 st = read_auto;
3381 break;
3382 case 0:
3383 if (mddev->in_sync)
3384 st = clean;
070dc6dd 3385 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3386 st = write_pending;
9e653b63
N		 3387 else if (mddev->safemode)
3388 st = active_idle;
3389 else
3390 st = active;
3391 }
3392 else {
3393 if (list_empty(&mddev->disks) &&
3394 mddev->raid_disks == 0 &&
58c0fed4 3395 mddev->dev_sectors == 0)
9e653b63
N		 3396 st = clear;
3397 else
3398 st = inactive;
3399 }
3400 return sprintf(page, "%s\n", array_states[st]);
3401}
3402
df5b20cf 3403static int do_md_stop(mddev_t * mddev, int ro, int is_open);
a4bd82d0 3404static int md_set_readonly(mddev_t * mddev, int is_open);
9e653b63
N		 3405static int do_md_run(mddev_t * mddev);
3406static int restart_array(mddev_t *mddev);
3407
3408static ssize_t
3409array_state_store(mddev_t *mddev, const char *buf, size_t len)
3410{
3411 int err = -EINVAL;
3412 enum array_state st = match_word(buf, array_states);
3413 switch(st) {
3414 case bad_word:
3415 break;
3416 case clear:
3417 /* stopping an active array */
f2ea68cf 3418 if (atomic_read(&mddev->openers) > 0)
e691063a 3419 return -EBUSY;
df5b20cf 3420 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3421 break;
3422 case inactive:
3423 /* stopping an active array */
3424 if (mddev->pers) {
f2ea68cf 3425 if (atomic_read(&mddev->openers) > 0)
9e653b63 3426 return -EBUSY;
df5b20cf 3427 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3428 } else
3429 err = 0; /* already inactive */
9e653b63
N		 3430 break;
3431 case suspended:
3432 break; /* not supported yet */
3433 case readonly:
3434 if (mddev->pers)
a4bd82d0 3435 err = md_set_readonly(mddev, 0);
9e653b63
N		 3436 else {
3437 mddev->ro = 1;
648b629e 3438 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3439 err = do_md_run(mddev);
3440 }
3441 break;
3442 case read_auto:
9e653b63 3443 if (mddev->pers) {
80268ee9 3444 if (mddev->ro == 0)
a4bd82d0 3445 err = md_set_readonly(mddev, 0);
80268ee9 3446 else if (mddev->ro == 1)
648b629e
N		 3447 err = restart_array(mddev);
3448 if (err == 0) {
3449 mddev->ro = 2;
3450 set_disk_ro(mddev->gendisk, 0);
3451 }
9e653b63
N		 3452 } else {
3453 mddev->ro = 2;
3454 err = do_md_run(mddev);
3455 }
3456 break;
3457 case clean:
3458 if (mddev->pers) {
3459 restart_array(mddev);
3460 spin_lock_irq(&mddev->write_lock);
3461 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3462 if (mddev->in_sync == 0) {
3463 mddev->in_sync = 1;
31a59e34
N		 3464 if (mddev->safemode == 1)
3465 mddev->safemode = 0;
070dc6dd 3466 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3467 }
3468 err = 0;
3469 } else
3470 err = -EBUSY;
9e653b63 3471 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3472 } else
3473 err = -EINVAL;
9e653b63
N		 3474 break;
3475 case active:
3476 if (mddev->pers) {
3477 restart_array(mddev);
070dc6dd 3478 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3479 wake_up(&mddev->sb_wait);
3480 err = 0;
3481 } else {
3482 mddev->ro = 0;
648b629e 3483 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3484 err = do_md_run(mddev);
3485 }
3486 break;
3487 case write_pending:
3488 case active_idle:
3489 /* these cannot be set */
3490 break;
3491 }
3492 if (err)
3493 return err;
0fd62b86 3494 else {
00bcb4ac 3495 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3496 return len;
0fd62b86 3497 }
9e653b63 3498}
80ca3a44
N		 3499static struct md_sysfs_entry md_array_state =
3500__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3501
1e50915f
RB		 3502static ssize_t
3503max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3504 return sprintf(page, "%d\n",
3505 atomic_read(&mddev->max_corr_read_errors));
3506}
3507
3508static ssize_t
3509max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3510{
3511 char *e;
3512 unsigned long n = simple_strtoul(buf, &e, 10);
3513
3514 if (*buf && (*e == 0 || *e == '\n')) {
3515 atomic_set(&mddev->max_corr_read_errors, n);
3516 return len;
3517 }
3518 return -EINVAL;
3519}
3520
3521static struct md_sysfs_entry max_corr_read_errors =
3522__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3523 max_corrected_read_errors_store);
3524
6d7ff738
N		 3525static ssize_t
3526null_show(mddev_t *mddev, char *page)
3527{
3528 return -EINVAL;
3529}
3530
3531static ssize_t
3532new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3533{
3534 /* buf must be %d:%d\n? giving major and minor numbers */
3535 /* The new device is added to the array.
3536 * If the array has a persistent superblock, we read the
3537 * superblock to initialise info and check validity.
3538 * Otherwise, only checking done is that in bind_rdev_to_array,
3539 * which mainly checks size.
3540 */
3541 char *e;
3542 int major = simple_strtoul(buf, &e, 10);
3543 int minor;
3544 dev_t dev;
3545 mdk_rdev_t *rdev;
3546 int err;
3547
3548 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3549 return -EINVAL;
3550 minor = simple_strtoul(e+1, &e, 10);
3551 if (*e && *e != '\n')
3552 return -EINVAL;
3553 dev = MKDEV(major, minor);
3554 if (major != MAJOR(dev) ||
3555 minor != MINOR(dev))
3556 return -EOVERFLOW;
3557
3558
3559 if (mddev->persistent) {
3560 rdev = md_import_device(dev, mddev->major_version,
3561 mddev->minor_version);
3562 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3563 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3564 mdk_rdev_t, same_set);
3565 err = super_types[mddev->major_version]
3566 .load_super(rdev, rdev0, mddev->minor_version);
3567 if (err < 0)
3568 goto out;
3569 }
c5d79adb
N		 3570 } else if (mddev->external)
3571 rdev = md_import_device(dev, -2, -1);
3572 else
6d7ff738
N		 3573 rdev = md_import_device(dev, -1, -1);
3574
3575 if (IS_ERR(rdev))
3576 return PTR_ERR(rdev);
3577 err = bind_rdev_to_array(rdev, mddev);
3578 out:
3579 if (err)
3580 export_rdev(rdev);
3581 return err ? err : len;
3582}
3583
3584static struct md_sysfs_entry md_new_device =
80ca3a44 3585__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3586
9b1d1dac
PC		 3587static ssize_t
3588bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3589{
3590 char *end;
3591 unsigned long chunk, end_chunk;
3592
3593 if (!mddev->bitmap)
3594 goto out;
3595 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3596 while (*buf) {
3597 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3598 if (buf == end) break;
3599 if (*end == '-') { /* range */
3600 buf = end + 1;
3601 end_chunk = simple_strtoul(buf, &end, 0);
3602 if (buf == end) break;
3603 }
3604 if (*end && !isspace(*end)) break;
3605 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3606 buf = skip_spaces(end);
9b1d1dac
PC		 3607 }
3608 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3609out:
3610 return len;
3611}
3612
3613static struct md_sysfs_entry md_bitmap =
3614__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3615
a35b0d69
N		 3616static ssize_t
3617size_show(mddev_t *mddev, char *page)
3618{
58c0fed4
AN		 3619 return sprintf(page, "%llu\n",
3620 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3621}
3622
d71f9f88 3623static int update_size(mddev_t *mddev, sector_t num_sectors);
a35b0d69
N		 3624
3625static ssize_t
3626size_store(mddev_t *mddev, const char *buf, size_t len)
3627{
3628 /* If array is inactive, we can reduce the component size, but
3629 * not increase it (except from 0).
3630 * If array is active, we can try an on-line resize
3631 */
b522adcd
DW		 3632 sector_t sectors;
3633 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3634
58c0fed4
AN		 3635 if (err < 0)
3636 return err;
a35b0d69 3637 if (mddev->pers) {
58c0fed4 3638 err = update_size(mddev, sectors);
850b2b42 3639 md_update_sb(mddev, 1);
a35b0d69 3640 } else {
58c0fed4
AN		 3641 if (mddev->dev_sectors == 0 ||
3642 mddev->dev_sectors > sectors)
3643 mddev->dev_sectors = sectors;
a35b0d69
N		 3644 else
3645 err = -ENOSPC;
3646 }
3647 return err ? err : len;
3648}
3649
3650static struct md_sysfs_entry md_size =
80ca3a44 3651__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3652
8bb93aac
N		 3653
3654/* Metdata version.
e691063a
N		 3655 * This is one of
3656 * 'none' for arrays with no metadata (good luck...)
3657 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3658 * or N.M for internally known formats
3659 */
3660static ssize_t
3661metadata_show(mddev_t *mddev, char *page)
3662{
3663 if (mddev->persistent)
3664 return sprintf(page, "%d.%d\n",
3665 mddev->major_version, mddev->minor_version);
e691063a
N		 3666 else if (mddev->external)
3667 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3668 else
3669 return sprintf(page, "none\n");
3670}
3671
3672static ssize_t
3673metadata_store(mddev_t *mddev, const char *buf, size_t len)
3674{
3675 int major, minor;
3676 char *e;
ea43ddd8
N		 3677 /* Changing the details of 'external' metadata is
3678 * always permitted. Otherwise there must be
3679 * no devices attached to the array.
3680 */
3681 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3682 ;
3683 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3684 return -EBUSY;
3685
3686 if (cmd_match(buf, "none")) {
3687 mddev->persistent = 0;
e691063a
N		 3688 mddev->external = 0;
3689 mddev->major_version = 0;
3690 mddev->minor_version = 90;
3691 return len;
3692 }
3693 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3694 size_t namelen = len-9;
e691063a
N		 3695 if (namelen >= sizeof(mddev->metadata_type))
3696 namelen = sizeof(mddev->metadata_type)-1;
3697 strncpy(mddev->metadata_type, buf+9, namelen);
3698 mddev->metadata_type[namelen] = 0;
3699 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3700 mddev->metadata_type[--namelen] = 0;
3701 mddev->persistent = 0;
3702 mddev->external = 1;
8bb93aac
N		 3703 mddev->major_version = 0;
3704 mddev->minor_version = 90;
3705 return len;
3706 }
3707 major = simple_strtoul(buf, &e, 10);
3708 if (e==buf || *e != '.')
3709 return -EINVAL;
3710 buf = e+1;
3711 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 3712 if (e==buf || (*e && *e != '\n') )
8bb93aac 3713 return -EINVAL;
50511da3 3714 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 3715 return -ENOENT;
3716 mddev->major_version = major;
3717 mddev->minor_version = minor;
3718 mddev->persistent = 1;
e691063a 3719 mddev->external = 0;
8bb93aac
N		 3720 return len;
3721}
3722
3723static struct md_sysfs_entry md_metadata =
80ca3a44 3724__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 3725
24dd469d 3726static ssize_t
7eec314d 3727action_show(mddev_t *mddev, char *page)
24dd469d 3728{
7eec314d 3729 char *type = "idle";
b6a9ce68
N		 3730 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3731 type = "frozen";
3732 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 3733 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 3734 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3735 type = "reshape";
3736 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 3737 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3738 type = "resync";
3739 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3740 type = "check";
3741 else
3742 type = "repair";
72a23c21 3743 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 3744 type = "recover";
3745 }
3746 return sprintf(page, "%s\n", type);
3747}
3748
7ebc0be7
N		 3749static void reap_sync_thread(mddev_t *mddev);
3750
24dd469d 3751static ssize_t
7eec314d 3752action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 3753{
7eec314d
N		 3754 if (!mddev->pers || !mddev->pers->sync_request)
3755 return -EINVAL;
3756
b6a9ce68
N		 3757 if (cmd_match(page, "frozen"))
3758 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3759 else
3760 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3761
3762 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 3763 if (mddev->sync_thread) {
3764 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 3765 reap_sync_thread(mddev);
7eec314d 3766 }
03c902e1
N		 3767 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3768 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 3769 return -EBUSY;
72a23c21
NB		 3770 else if (cmd_match(page, "resync"))
3771 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3772 else if (cmd_match(page, "recover")) {
3773 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 3774 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 3775 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 3776 int err;
3777 if (mddev->pers->start_reshape == NULL)
3778 return -EINVAL;
3779 err = mddev->pers->start_reshape(mddev);
3780 if (err)
3781 return err;
a99ac971 3782 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 3783 } else {
bce74dac 3784 if (cmd_match(page, "check"))
7eec314d 3785 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 3786 else if (!cmd_match(page, "repair"))
7eec314d
N		 3787 return -EINVAL;
3788 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3789 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 3790 }
03c902e1 3791 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 3792 md_wakeup_thread(mddev->thread);
00bcb4ac 3793 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 3794 return len;
3795}
3796
9d88883e 3797static ssize_t
96de1e66 3798mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 3799{
3800 return sprintf(page, "%llu\n",
3801 (unsigned long long) mddev->resync_mismatches);
3802}
3803
80ca3a44
N		 3804static struct md_sysfs_entry md_scan_mode =
3805__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 3806
96de1e66 3807
80ca3a44 3808static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 3809
88202a0c
N		 3810static ssize_t
3811sync_min_show(mddev_t *mddev, char *page)
3812{
3813 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3814 mddev->sync_speed_min ? "local": "system");
3815}
3816
3817static ssize_t
3818sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3819{
3820 int min;
3821 char *e;
3822 if (strncmp(buf, "system", 6)==0) {
3823 mddev->sync_speed_min = 0;
3824 return len;
3825 }
3826 min = simple_strtoul(buf, &e, 10);
3827 if (buf == e || (*e && *e != '\n') || min <= 0)
3828 return -EINVAL;
3829 mddev->sync_speed_min = min;
3830 return len;
3831}
3832
3833static struct md_sysfs_entry md_sync_min =
3834__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3835
3836static ssize_t
3837sync_max_show(mddev_t *mddev, char *page)
3838{
3839 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3840 mddev->sync_speed_max ? "local": "system");
3841}
3842
3843static ssize_t
3844sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3845{
3846 int max;
3847 char *e;
3848 if (strncmp(buf, "system", 6)==0) {
3849 mddev->sync_speed_max = 0;
3850 return len;
3851 }
3852 max = simple_strtoul(buf, &e, 10);
3853 if (buf == e || (*e && *e != '\n') || max <= 0)
3854 return -EINVAL;
3855 mddev->sync_speed_max = max;
3856 return len;
3857}
3858
3859static struct md_sysfs_entry md_sync_max =
3860__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3861
d7f3d291
IP		 3862static ssize_t
3863degraded_show(mddev_t *mddev, char *page)
3864{
3865 return sprintf(page, "%d\n", mddev->degraded);
3866}
3867static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 3868
90b08710
BS		 3869static ssize_t
3870sync_force_parallel_show(mddev_t *mddev, char *page)
3871{
3872 return sprintf(page, "%d\n", mddev->parallel_resync);
3873}
3874
3875static ssize_t
3876sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3877{
3878 long n;
3879
3880 if (strict_strtol(buf, 10, &n))
3881 return -EINVAL;
3882
3883 if (n != 0 && n != 1)
3884 return -EINVAL;
3885
3886 mddev->parallel_resync = n;
3887
3888 if (mddev->sync_thread)
3889 wake_up(&resync_wait);
3890
3891 return len;
3892}
3893
3894/* force parallel resync, even with shared block devices */
3895static struct md_sysfs_entry md_sync_force_parallel =
3896__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3897 sync_force_parallel_show, sync_force_parallel_store);
3898
88202a0c
N		 3899static ssize_t
3900sync_speed_show(mddev_t *mddev, char *page)
3901{
3902 unsigned long resync, dt, db;
d1a7c503
N		 3903 if (mddev->curr_resync == 0)
3904 return sprintf(page, "none\n");
9687a60c
AN		 3905 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3906 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 3907 if (!dt) dt++;
9687a60c
AN		 3908 db = resync - mddev->resync_mark_cnt;
3909 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 3910}
3911
80ca3a44 3912static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 3913
3914static ssize_t
3915sync_completed_show(mddev_t *mddev, char *page)
3916{
58c0fed4 3917 unsigned long max_sectors, resync;
88202a0c 3918
acb180b0
N		 3919 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3920 return sprintf(page, "none\n");
3921
88202a0c 3922 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 3923 max_sectors = mddev->resync_max_sectors;
88202a0c 3924 else
58c0fed4 3925 max_sectors = mddev->dev_sectors;
88202a0c 3926
acb180b0 3927 resync = mddev->curr_resync_completed;
58c0fed4 3928 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
88202a0c
N		 3929}
3930
80ca3a44 3931static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3932
5e96ee65
NB		 3933static ssize_t
3934min_sync_show(mddev_t *mddev, char *page)
3935{
3936 return sprintf(page, "%llu\n",
3937 (unsigned long long)mddev->resync_min);
3938}
3939static ssize_t
3940min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3941{
3942 unsigned long long min;
3943 if (strict_strtoull(buf, 10, &min))
3944 return -EINVAL;
3945 if (min > mddev->resync_max)
3946 return -EINVAL;
3947 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3948 return -EBUSY;
3949
3950 /* Must be a multiple of chunk_size */
9d8f0363 3951 if (mddev->chunk_sectors) {
2ac06c33 3952 sector_t temp = min;
9d8f0363 3953 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 3954 return -EINVAL;
3955 }
3956 mddev->resync_min = min;
3957
3958 return len;
3959}
3960
3961static struct md_sysfs_entry md_min_sync =
3962__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3963
c6207277
N		 3964static ssize_t
3965max_sync_show(mddev_t *mddev, char *page)
3966{
3967 if (mddev->resync_max == MaxSector)
3968 return sprintf(page, "max\n");
3969 else
3970 return sprintf(page, "%llu\n",
3971 (unsigned long long)mddev->resync_max);
3972}
3973static ssize_t
3974max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3975{
3976 if (strncmp(buf, "max", 3) == 0)
3977 mddev->resync_max = MaxSector;
3978 else {
5e96ee65
NB		 3979 unsigned long long max;
3980 if (strict_strtoull(buf, 10, &max))
3981 return -EINVAL;
3982 if (max < mddev->resync_min)
c6207277
N		 3983 return -EINVAL;
3984 if (max < mddev->resync_max &&
4d484a4a 3985 mddev->ro == 0 &&
c6207277
N		 3986 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3987 return -EBUSY;
3988
3989 /* Must be a multiple of chunk_size */
9d8f0363 3990 if (mddev->chunk_sectors) {
2ac06c33 3991 sector_t temp = max;
9d8f0363 3992 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 3993 return -EINVAL;
3994 }
3995 mddev->resync_max = max;
3996 }
3997 wake_up(&mddev->recovery_wait);
3998 return len;
3999}
4000
4001static struct md_sysfs_entry md_max_sync =
4002__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4003
e464eafd
N		 4004static ssize_t
4005suspend_lo_show(mddev_t *mddev, char *page)
4006{
4007 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4008}
4009
4010static ssize_t
4011suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4012{
4013 char *e;
4014 unsigned long long new = simple_strtoull(buf, &e, 10);
4015
b8d966ef
N		 4016 if (mddev->pers == NULL ||
4017 mddev->pers->quiesce == NULL)
e464eafd
N		 4018 return -EINVAL;
4019 if (buf == e || (*e && *e != '\n'))
4020 return -EINVAL;
4021 if (new >= mddev->suspend_hi ||
4022 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
4023 mddev->suspend_lo = new;
4024 mddev->pers->quiesce(mddev, 2);
4025 return len;
4026 } else
4027 return -EINVAL;
4028}
4029static struct md_sysfs_entry md_suspend_lo =
4030__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4031
4032
4033static ssize_t
4034suspend_hi_show(mddev_t *mddev, char *page)
4035{
4036 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4037}
4038
4039static ssize_t
4040suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4041{
4042 char *e;
4043 unsigned long long new = simple_strtoull(buf, &e, 10);
4044
b8d966ef
N		 4045 if (mddev->pers == NULL ||
4046 mddev->pers->quiesce == NULL)
e464eafd
N		 4047 return -EINVAL;
4048 if (buf == e || (*e && *e != '\n'))
4049 return -EINVAL;
4050 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
4051 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
4052 mddev->suspend_hi = new;
4053 mddev->pers->quiesce(mddev, 1);
4054 mddev->pers->quiesce(mddev, 0);
4055 return len;
4056 } else
4057 return -EINVAL;
4058}
4059static struct md_sysfs_entry md_suspend_hi =
4060__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4061
08a02ecd
N		 4062static ssize_t
4063reshape_position_show(mddev_t *mddev, char *page)
4064{
4065 if (mddev->reshape_position != MaxSector)
4066 return sprintf(page, "%llu\n",
4067 (unsigned long long)mddev->reshape_position);
4068 strcpy(page, "none\n");
4069 return 5;
4070}
4071
4072static ssize_t
4073reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4074{
4075 char *e;
4076 unsigned long long new = simple_strtoull(buf, &e, 10);
4077 if (mddev->pers)
4078 return -EBUSY;
4079 if (buf == e || (*e && *e != '\n'))
4080 return -EINVAL;
4081 mddev->reshape_position = new;
4082 mddev->delta_disks = 0;
4083 mddev->new_level = mddev->level;
4084 mddev->new_layout = mddev->layout;
664e7c41 4085 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4086 return len;
4087}
4088
4089static struct md_sysfs_entry md_reshape_position =
4090__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4091 reshape_position_store);
4092
b522adcd
DW		 4093static ssize_t
4094array_size_show(mddev_t *mddev, char *page)
4095{
4096 if (mddev->external_size)
4097 return sprintf(page, "%llu\n",
4098 (unsigned long long)mddev->array_sectors/2);
4099 else
4100 return sprintf(page, "default\n");
4101}
4102
4103static ssize_t
4104array_size_store(mddev_t *mddev, const char *buf, size_t len)
4105{
4106 sector_t sectors;
4107
4108 if (strncmp(buf, "default", 7) == 0) {
4109 if (mddev->pers)
4110 sectors = mddev->pers->size(mddev, 0, 0);
4111 else
4112 sectors = mddev->array_sectors;
4113
4114 mddev->external_size = 0;
4115 } else {
4116 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4117 return -EINVAL;
4118 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4119 return -E2BIG;
b522adcd
DW		 4120
4121 mddev->external_size = 1;
4122 }
4123
4124 mddev->array_sectors = sectors;
4125 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e
N		 4126 if (mddev->pers)
4127 revalidate_disk(mddev->gendisk);
b522adcd
DW		 4128
4129 return len;
4130}
4131
4132static struct md_sysfs_entry md_array_size =
4133__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4134 array_size_store);
e464eafd 4135
eae1701f
N		 4136static struct attribute *md_default_attrs[] = {
4137 &md_level.attr,
d4dbd025 4138 &md_layout.attr,
eae1701f 4139 &md_raid_disks.attr,
3b34380a 4140 &md_chunk_size.attr,
a35b0d69 4141 &md_size.attr,
a94213b1 4142 &md_resync_start.attr,
8bb93aac 4143 &md_metadata.attr,
6d7ff738 4144 &md_new_device.attr,
16f17b39 4145 &md_safe_delay.attr,
9e653b63 4146 &md_array_state.attr,
08a02ecd 4147 &md_reshape_position.attr,
b522adcd 4148 &md_array_size.attr,
1e50915f 4149 &max_corr_read_errors.attr,
411036fa
N		 4150 NULL,
4151};
4152
4153static struct attribute *md_redundancy_attrs[] = {
24dd469d 4154 &md_scan_mode.attr,
9d88883e 4155 &md_mismatches.attr,
88202a0c
N		 4156 &md_sync_min.attr,
4157 &md_sync_max.attr,
4158 &md_sync_speed.attr,
90b08710 4159 &md_sync_force_parallel.attr,
88202a0c 4160 &md_sync_completed.attr,
5e96ee65 4161 &md_min_sync.attr,
c6207277 4162 &md_max_sync.attr,
e464eafd
N		 4163 &md_suspend_lo.attr,
4164 &md_suspend_hi.attr,
9b1d1dac 4165 &md_bitmap.attr,
d7f3d291 4166 &md_degraded.attr,
eae1701f
N		 4167 NULL,
4168};
411036fa
N		 4169static struct attribute_group md_redundancy_group = {
4170 .name = NULL,
4171 .attrs = md_redundancy_attrs,
4172};
4173
eae1701f
N		 4174
4175static ssize_t
4176md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4177{
4178 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4179 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4180 ssize_t rv;
eae1701f
N		 4181
4182 if (!entry->show)
4183 return -EIO;
5dc5cf7d
IM		 4184 rv = mddev_lock(mddev);
4185 if (!rv) {
4186 rv = entry->show(mddev, page);
4187 mddev_unlock(mddev);
4188 }
96de1e66 4189 return rv;
eae1701f
N		 4190}
4191
4192static ssize_t
4193md_attr_store(struct kobject *kobj, struct attribute *attr,
4194 const char *page, size_t length)
4195{
4196 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4197 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4198 ssize_t rv;
eae1701f
N		 4199
4200 if (!entry->store)
4201 return -EIO;
67463acb
N		 4202 if (!capable(CAP_SYS_ADMIN))
4203 return -EACCES;
5dc5cf7d 4204 rv = mddev_lock(mddev);
d3374825
N		 4205 if (mddev->hold_active == UNTIL_IOCTL)
4206 mddev->hold_active = 0;
5dc5cf7d
IM		 4207 if (!rv) {
4208 rv = entry->store(mddev, page, length);
4209 mddev_unlock(mddev);
4210 }
96de1e66 4211 return rv;
eae1701f
N		 4212}
4213
4214static void md_free(struct kobject *ko)
4215{
4216 mddev_t *mddev = container_of(ko, mddev_t, kobj);
a21d1504
N		 4217
4218 if (mddev->sysfs_state)
4219 sysfs_put(mddev->sysfs_state);
4220
4221 if (mddev->gendisk) {
4222 del_gendisk(mddev->gendisk);
4223 put_disk(mddev->gendisk);
4224 }
4225 if (mddev->queue)
4226 blk_cleanup_queue(mddev->queue);
4227
eae1701f
N		 4228 kfree(mddev);
4229}
4230
52cf25d0 4231static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4232 .show = md_attr_show,
4233 .store = md_attr_store,
4234};
4235static struct kobj_type md_ktype = {
4236 .release = md_free,
4237 .sysfs_ops = &md_sysfs_ops,
4238 .default_attrs = md_default_attrs,
4239};
4240
1da177e4
LT		 4241int mdp_major = 0;
4242
5fd3a17e
DW		 4243static void mddev_delayed_delete(struct work_struct *ws)
4244{
4245 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4246
43a70507 4247 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4248 kobject_del(&mddev->kobj);
4249 kobject_put(&mddev->kobj);
4250}
4251
efeb53c0 4252static int md_alloc(dev_t dev, char *name)
1da177e4 4253{
48c9c27b 4254 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 4255 mddev_t *mddev = mddev_find(dev);
4256 struct gendisk *disk;
efeb53c0
N		 4257 int partitioned;
4258 int shift;
4259 int unit;
3830c62f 4260 int error;
1da177e4
LT		 4261
4262 if (!mddev)
efeb53c0
N		 4263 return -ENODEV;
4264
4265 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4266 shift = partitioned ? MdpMinorShift : 0;
4267 unit = MINOR(mddev->unit) >> shift;
1da177e4 4268
e804ac78
TH		 4269 /* wait for any previous instance of this device to be
4270 * completely removed (mddev_delayed_delete).
d3374825 4271 */
e804ac78 4272 flush_workqueue(md_misc_wq);
d3374825 4273
48c9c27b 4274 mutex_lock(&disks_mutex);
0909dc44
N		 4275 error = -EEXIST;
4276 if (mddev->gendisk)
4277 goto abort;
efeb53c0
N		 4278
4279 if (name) {
4280 /* Need to ensure that 'name' is not a duplicate.
4281 */
4282 mddev_t *mddev2;
4283 spin_lock(&all_mddevs_lock);
4284
4285 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4286 if (mddev2->gendisk &&
4287 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4288 spin_unlock(&all_mddevs_lock);
0909dc44 4289 goto abort;
efeb53c0
N		 4290 }
4291 spin_unlock(&all_mddevs_lock);
1da177e4 4292 }
8b765398 4293
0909dc44 4294 error = -ENOMEM;
8b765398 4295 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4296 if (!mddev->queue)
4297 goto abort;
409c57f3
N		 4298 mddev->queue->queuedata = mddev;
4299
409c57f3 4300 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4301
1da177e4
LT		 4302 disk = alloc_disk(1 << shift);
4303 if (!disk) {
8b765398
N		 4304 blk_cleanup_queue(mddev->queue);
4305 mddev->queue = NULL;
0909dc44 4306 goto abort;
1da177e4 4307 }
efeb53c0 4308 disk->major = MAJOR(mddev->unit);
1da177e4 4309 disk->first_minor = unit << shift;
efeb53c0
N		 4310 if (name)
4311 strcpy(disk->disk_name, name);
4312 else if (partitioned)
1da177e4 4313 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4314 else
1da177e4 4315 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4316 disk->fops = &md_fops;
4317 disk->private_data = mddev;
4318 disk->queue = mddev->queue;
92850bbd 4319 /* Allow extended partitions. This makes the
d3374825 4320 * 'mdp' device redundant, but we can't really
92850bbd
N		 4321 * remove it now.
4322 */
4323 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4
LT		 4324 add_disk(disk);
4325 mddev->gendisk = disk;
ed9e1982
TH		 4326 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4327 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4328 if (error) {
4329 /* This isn't possible, but as kobject_init_and_add is marked
4330 * __must_check, we must do something with the result
4331 */
5e55e2f5
N		 4332 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4333 disk->disk_name);
0909dc44
N		 4334 error = 0;
4335 }
00bcb4ac
N		 4336 if (mddev->kobj.sd &&
4337 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4338 printk(KERN_DEBUG "pointless warning\n");
be20e6c6
DW		 4339
4340 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
0909dc44
N		 4341 abort:
4342 mutex_unlock(&disks_mutex);
00bcb4ac 4343 if (!error && mddev->kobj.sd) {
3830c62f 4344 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4345 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4346 }
d3374825 4347 mddev_put(mddev);
0909dc44 4348 return error;
efeb53c0
N		 4349}
4350
4351static struct kobject *md_probe(dev_t dev, int *part, void *data)
4352{
4353 md_alloc(dev, NULL);
1da177e4
LT		 4354 return NULL;
4355}
4356
efeb53c0
N		 4357static int add_named_array(const char *val, struct kernel_param *kp)
4358{
4359 /* val must be "md_*" where * is not all digits.
4360 * We allocate an array with a large free minor number, and
4361 * set the name to val. val must not already be an active name.
4362 */
4363 int len = strlen(val);
4364 char buf[DISK_NAME_LEN];
4365
4366 while (len && val[len-1] == '\n')
4367 len--;
4368 if (len >= DISK_NAME_LEN)
4369 return -E2BIG;
4370 strlcpy(buf, val, len+1);
4371 if (strncmp(buf, "md_", 3) != 0)
4372 return -EINVAL;
4373 return md_alloc(0, buf);
4374}
4375
1da177e4
LT		 4376static void md_safemode_timeout(unsigned long data)
4377{
4378 mddev_t *mddev = (mddev_t *) data;
4379
0fd62b86
NB		 4380 if (!atomic_read(&mddev->writes_pending)) {
4381 mddev->safemode = 1;
4382 if (mddev->external)
00bcb4ac 4383 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4384 }
1da177e4
LT		 4385 md_wakeup_thread(mddev->thread);
4386}
4387
6ff8d8ec 4388static int start_dirty_degraded;
1da177e4 4389
390ee602 4390int md_run(mddev_t *mddev)
1da177e4 4391{
2604b703 4392 int err;
1da177e4 4393 mdk_rdev_t *rdev;
2604b703 4394 struct mdk_personality *pers;
1da177e4 4395
a757e64c
N		 4396 if (list_empty(&mddev->disks))
4397 /* cannot run an array with no devices.. */
1da177e4 4398 return -EINVAL;
1da177e4
LT		 4399
4400 if (mddev->pers)
4401 return -EBUSY;
bb4f1e9d
N		 4402 /* Cannot run until previous stop completes properly */
4403 if (mddev->sysfs_active)
4404 return -EBUSY;
b6eb127d 4405
1da177e4
LT		 4406 /*
4407 * Analyze all RAID superblock(s)
4408 */
1ec4a939
N		 4409 if (!mddev->raid_disks) {
4410 if (!mddev->persistent)
4411 return -EINVAL;
a757e64c 4412 analyze_sbs(mddev);
1ec4a939 4413 }
1da177e4 4414
d9d166c2
N		 4415 if (mddev->level != LEVEL_NONE)
4416 request_module("md-level-%d", mddev->level);
4417 else if (mddev->clevel[0])
4418 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4419
4420 /*
4421 * Drop all container device buffers, from now on
4422 * the only valid external interface is through the md
4423 * device.
1da177e4 4424 */
159ec1fc 4425 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4426 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4427 continue;
4428 sync_blockdev(rdev->bdev);
f98393a6 4429 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4430
4431 /* perform some consistency tests on the device.
4432 * We don't want the data to overlap the metadata,
58c0fed4 4433 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4434 */
0f420358 4435 if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4436 if (mddev->dev_sectors &&
4437 rdev->data_offset + mddev->dev_sectors
0f420358 4438 > rdev->sb_start) {
f0d76d70
N		 4439 printk("md: %s: data overlaps metadata\n",
4440 mdname(mddev));
4441 return -EINVAL;
4442 }
4443 } else {
0f420358 4444 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4445 > rdev->data_offset) {
4446 printk("md: %s: metadata overlaps data\n",
4447 mdname(mddev));
4448 return -EINVAL;
4449 }
4450 }
00bcb4ac 4451 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4452 }
4453
a167f663
N		 4454 if (mddev->bio_set == NULL)
4455 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4456
1da177e4 4457 spin_lock(&pers_lock);
d9d166c2 4458 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4459 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4460 spin_unlock(&pers_lock);
d9d166c2
N		 4461 if (mddev->level != LEVEL_NONE)
4462 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4463 mddev->level);
4464 else
4465 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4466 mddev->clevel);
1da177e4
LT		 4467 return -EINVAL;
4468 }
2604b703 4469 mddev->pers = pers;
1da177e4 4470 spin_unlock(&pers_lock);
34817e8c
N		 4471 if (mddev->level != pers->level) {
4472 mddev->level = pers->level;
4473 mddev->new_level = pers->level;
4474 }
d9d166c2 4475 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4476
f6705578 4477 if (mddev->reshape_position != MaxSector &&
63c70c4f 4478 pers->start_reshape == NULL) {
f6705578
N		 4479 /* This personality cannot handle reshaping... */
4480 mddev->pers = NULL;
4481 module_put(pers->owner);
4482 return -EINVAL;
4483 }
4484
7dd5e7c3
N		 4485 if (pers->sync_request) {
4486 /* Warn if this is a potentially silly
4487 * configuration.
4488 */
4489 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4490 mdk_rdev_t *rdev2;
7dd5e7c3 4491 int warned = 0;
159ec1fc
CR		 4492
4493 list_for_each_entry(rdev, &mddev->disks, same_set)
4494 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4495 if (rdev < rdev2 &&
4496 rdev->bdev->bd_contains ==
4497 rdev2->bdev->bd_contains) {
4498 printk(KERN_WARNING
4499 "%s: WARNING: %s appears to be"
4500 " on the same physical disk as"
4501 " %s.\n",
4502 mdname(mddev),
4503 bdevname(rdev->bdev,b),
4504 bdevname(rdev2->bdev,b2));
4505 warned = 1;
4506 }
4507 }
159ec1fc 4508
7dd5e7c3
N		 4509 if (warned)
4510 printk(KERN_WARNING
4511 "True protection against single-disk"
4512 " failure might be compromised.\n");
4513 }
4514
657390d2 4515 mddev->recovery = 0;
58c0fed4
AN		 4516 /* may be over-ridden by personality */
4517 mddev->resync_max_sectors = mddev->dev_sectors;
4518
6ff8d8ec 4519 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4520
0f9552b5 4521 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4522 mddev->ro = 2; /* read-only, but switch on first write */
4523
b15c2e57 4524 err = mddev->pers->run(mddev);
13e53df3
AN		 4525 if (err)
4526 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4527 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4528 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4529 " but 'external_size' not in effect?\n", __func__);
4530 printk(KERN_ERR
4531 "md: invalid array_size %llu > default size %llu\n",
4532 (unsigned long long)mddev->array_sectors / 2,
4533 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4534 err = -EINVAL;
4535 mddev->pers->stop(mddev);
4536 }
4537 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4538 err = bitmap_create(mddev);
4539 if (err) {
4540 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4541 mdname(mddev), err);
4542 mddev->pers->stop(mddev);
4543 }
4544 }
1da177e4 4545 if (err) {
1da177e4
LT		 4546 module_put(mddev->pers->owner);
4547 mddev->pers = NULL;
32a7627c
N		 4548 bitmap_destroy(mddev);
4549 return err;
1da177e4 4550 }
5e55e2f5 4551 if (mddev->pers->sync_request) {
00bcb4ac
N		 4552 if (mddev->kobj.sd &&
4553 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4554 printk(KERN_WARNING
4555 "md: cannot register extra attributes for %s\n",
4556 mdname(mddev));
00bcb4ac 4557 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4558 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4559 mddev->ro = 0;
4560
1da177e4 4561 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4562 atomic_set(&mddev->max_corr_read_errors,
4563 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4564 mddev->safemode = 0;
4565 mddev->safemode_timer.function = md_safemode_timeout;
4566 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4567 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4568 mddev->in_sync = 1;
0ca69886
N		 4569 smp_wmb();
4570 mddev->ready = 1;
159ec1fc 4571 list_for_each_entry(rdev, &mddev->disks, same_set)
86e6ffdd
N		 4572 if (rdev->raid_disk >= 0) {
4573 char nm[20];
4574 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5 4575 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
00bcb4ac 4576 /* failure here is OK */;
86e6ffdd 4577 }
1da177e4
LT		 4578
4579 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4580
850b2b42
N		 4581 if (mddev->flags)
4582 md_update_sb(mddev, 0);
1da177e4 4583
0b8c9de0
N		 4584 md_wakeup_thread(mddev->thread);
4585 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 4586
d7603b7e 4587 md_new_event(mddev);
00bcb4ac
N		 4588 sysfs_notify_dirent_safe(mddev->sysfs_state);
4589 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4590 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4591 return 0;
4592}
390ee602 4593EXPORT_SYMBOL_GPL(md_run);
1da177e4 4594
fe60b014
N		 4595static int do_md_run(mddev_t *mddev)
4596{
4597 int err;
4598
4599 err = md_run(mddev);
4600 if (err)
4601 goto out;
69e51b44
N		 4602 err = bitmap_load(mddev);
4603 if (err) {
4604 bitmap_destroy(mddev);
4605 goto out;
4606 }
fe60b014
N		 4607 set_capacity(mddev->gendisk, mddev->array_sectors);
4608 revalidate_disk(mddev->gendisk);
4609 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4610out:
4611 return err;
4612}
4613
1da177e4
LT		 4614static int restart_array(mddev_t *mddev)
4615{
4616 struct gendisk *disk = mddev->gendisk;
1da177e4 4617
80fab1d7 4618 /* Complain if it has no devices */
1da177e4 4619 if (list_empty(&mddev->disks))
80fab1d7
AN		 4620 return -ENXIO;
4621 if (!mddev->pers)
4622 return -EINVAL;
4623 if (!mddev->ro)
4624 return -EBUSY;
4625 mddev->safemode = 0;
4626 mddev->ro = 0;
4627 set_disk_ro(disk, 0);
4628 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4629 mdname(mddev));
4630 /* Kick recovery or resync if necessary */
4631 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4632 md_wakeup_thread(mddev->thread);
4633 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4634 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4635 return 0;
1da177e4
LT		 4636}
4637
acc55e22
N		 4638/* similar to deny_write_access, but accounts for our holding a reference
4639 * to the file ourselves */
4640static int deny_bitmap_write_access(struct file * file)
4641{
4642 struct inode *inode = file->f_mapping->host;
4643
4644 spin_lock(&inode->i_lock);
4645 if (atomic_read(&inode->i_writecount) > 1) {
4646 spin_unlock(&inode->i_lock);
4647 return -ETXTBSY;
4648 }
4649 atomic_set(&inode->i_writecount, -1);
4650 spin_unlock(&inode->i_lock);
4651
4652 return 0;
4653}
4654
43a70507 4655void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4656{
4657 struct inode *inode = file->f_mapping->host;
4658
4659 spin_lock(&inode->i_lock);
4660 atomic_set(&inode->i_writecount, 1);
4661 spin_unlock(&inode->i_lock);
4662}
4663
6177b472
N		 4664static void md_clean(mddev_t *mddev)
4665{
4666 mddev->array_sectors = 0;
4667 mddev->external_size = 0;
4668 mddev->dev_sectors = 0;
4669 mddev->raid_disks = 0;
4670 mddev->recovery_cp = 0;
4671 mddev->resync_min = 0;
4672 mddev->resync_max = MaxSector;
4673 mddev->reshape_position = MaxSector;
4674 mddev->external = 0;
4675 mddev->persistent = 0;
4676 mddev->level = LEVEL_NONE;
4677 mddev->clevel[0] = 0;
4678 mddev->flags = 0;
4679 mddev->ro = 0;
4680 mddev->metadata_type[0] = 0;
4681 mddev->chunk_sectors = 0;
4682 mddev->ctime = mddev->utime = 0;
4683 mddev->layout = 0;
4684 mddev->max_disks = 0;
4685 mddev->events = 0;
a8707c08 4686 mddev->can_decrease_events = 0;
6177b472
N		 4687 mddev->delta_disks = 0;
4688 mddev->new_level = LEVEL_NONE;
4689 mddev->new_layout = 0;
4690 mddev->new_chunk_sectors = 0;
4691 mddev->curr_resync = 0;
4692 mddev->resync_mismatches = 0;
4693 mddev->suspend_lo = mddev->suspend_hi = 0;
4694 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4695 mddev->recovery = 0;
4696 mddev->in_sync = 0;
4697 mddev->degraded = 0;
6177b472
N		 4698 mddev->safemode = 0;
4699 mddev->bitmap_info.offset = 0;
4700 mddev->bitmap_info.default_offset = 0;
4701 mddev->bitmap_info.chunksize = 0;
4702 mddev->bitmap_info.daemon_sleep = 0;
4703 mddev->bitmap_info.max_write_behind = 0;
252ac522 4704 mddev->plug = NULL;
6177b472
N		 4705}
4706
defad61a 4707static void __md_stop_writes(mddev_t *mddev)
a047e125
N		 4708{
4709 if (mddev->sync_thread) {
4710 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4711 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4712 reap_sync_thread(mddev);
a047e125
N		 4713 }
4714
4715 del_timer_sync(&mddev->safemode_timer);
4716
4717 bitmap_flush(mddev);
4718 md_super_wait(mddev);
4719
4720 if (!mddev->in_sync || mddev->flags) {
4721 /* mark array as shutdown cleanly */
4722 mddev->in_sync = 1;
4723 md_update_sb(mddev, 1);
4724 }
4725}
defad61a
N		 4726
4727void md_stop_writes(mddev_t *mddev)
4728{
4729 mddev_lock(mddev);
4730 __md_stop_writes(mddev);
4731 mddev_unlock(mddev);
4732}
390ee602 4733EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 4734
390ee602 4735void md_stop(mddev_t *mddev)
6177b472 4736{
0ca69886 4737 mddev->ready = 0;
6177b472
N		 4738 mddev->pers->stop(mddev);
4739 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4740 mddev->to_remove = &md_redundancy_group;
4741 module_put(mddev->pers->owner);
4742 mddev->pers = NULL;
cca9cf90 4743 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 4744}
390ee602 4745EXPORT_SYMBOL_GPL(md_stop);
6177b472 4746
a4bd82d0
N		 4747static int md_set_readonly(mddev_t *mddev, int is_open)
4748{
4749 int err = 0;
4750 mutex_lock(&mddev->open_mutex);
4751 if (atomic_read(&mddev->openers) > is_open) {
4752 printk("md: %s still in use.\n",mdname(mddev));
4753 err = -EBUSY;
4754 goto out;
4755 }
4756 if (mddev->pers) {
defad61a 4757 __md_stop_writes(mddev);
a4bd82d0
N		 4758
4759 err = -ENXIO;
4760 if (mddev->ro==1)
4761 goto out;
4762 mddev->ro = 1;
4763 set_disk_ro(mddev->gendisk, 1);
4764 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 4765 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 4766 err = 0;
4767 }
4768out:
4769 mutex_unlock(&mddev->open_mutex);
4770 return err;
4771}
4772
9e653b63
N		 4773/* mode:
4774 * 0 - completely stop and dis-assemble array
9e653b63
N		 4775 * 2 - stop but do not disassemble array
4776 */
df5b20cf 4777static int do_md_stop(mddev_t * mddev, int mode, int is_open)
1da177e4 4778{
1da177e4 4779 struct gendisk *disk = mddev->gendisk;
c4647292 4780 mdk_rdev_t *rdev;
1da177e4 4781
c8c00a69 4782 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 4783 if (atomic_read(&mddev->openers) > is_open ||
4784 mddev->sysfs_active) {
df5b20cf 4785 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 4786 mutex_unlock(&mddev->open_mutex);
4787 return -EBUSY;
4788 }
1da177e4 4789
6e17b027 4790 if (mddev->pers) {
a4bd82d0
N		 4791 if (mddev->ro)
4792 set_disk_ro(disk, 0);
409c57f3 4793
defad61a 4794 __md_stop_writes(mddev);
a4bd82d0
N		 4795 md_stop(mddev);
4796 mddev->queue->merge_bvec_fn = NULL;
4797 mddev->queue->unplug_fn = NULL;
4798 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 4799
a4bd82d0 4800 /* tell userspace to handle 'inactive' */
00bcb4ac 4801 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 4802
a4bd82d0
N		 4803 list_for_each_entry(rdev, &mddev->disks, same_set)
4804 if (rdev->raid_disk >= 0) {
4805 char nm[20];
4806 sprintf(nm, "rd%d", rdev->raid_disk);
4807 sysfs_remove_link(&mddev->kobj, nm);
4808 }
c4647292 4809
a4bd82d0 4810 set_capacity(disk, 0);
6e17b027 4811 mutex_unlock(&mddev->open_mutex);
a4bd82d0 4812 revalidate_disk(disk);
0d4ca600 4813
a4bd82d0
N		 4814 if (mddev->ro)
4815 mddev->ro = 0;
6e17b027
N		 4816 } else
4817 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 4818 /*
4819 * Free resources if final stop
4820 */
9e653b63 4821 if (mode == 0) {
1da177e4
LT		 4822 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4823
978f946b 4824 bitmap_destroy(mddev);
c3d9714e
N		 4825 if (mddev->bitmap_info.file) {
4826 restore_bitmap_write_access(mddev->bitmap_info.file);
4827 fput(mddev->bitmap_info.file);
4828 mddev->bitmap_info.file = NULL;
978f946b 4829 }
c3d9714e 4830 mddev->bitmap_info.offset = 0;
978f946b 4831
1da177e4
LT		 4832 export_array(mddev);
4833
6177b472 4834 md_clean(mddev);
934d9c23 4835 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 4836 if (mddev->hold_active == UNTIL_STOP)
4837 mddev->hold_active = 0;
a4bd82d0 4838 }
3f9d99c1 4839 blk_integrity_unregister(disk);
d7603b7e 4840 md_new_event(mddev);
00bcb4ac 4841 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 4842 return 0;
1da177e4
LT		 4843}
4844
fdee8ae4 4845#ifndef MODULE
1da177e4
LT		 4846static void autorun_array(mddev_t *mddev)
4847{
4848 mdk_rdev_t *rdev;
1da177e4
LT		 4849 int err;
4850
a757e64c 4851 if (list_empty(&mddev->disks))
1da177e4 4852 return;
1da177e4
LT		 4853
4854 printk(KERN_INFO "md: running: ");
4855
159ec1fc 4856 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 4857 char b[BDEVNAME_SIZE];
4858 printk("<%s>", bdevname(rdev->bdev,b));
4859 }
4860 printk("\n");
4861
d710e138 4862 err = do_md_run(mddev);
1da177e4
LT		 4863 if (err) {
4864 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 4865 do_md_stop(mddev, 0, 0);
1da177e4
LT		 4866 }
4867}
4868
4869/*
4870 * lets try to run arrays based on all disks that have arrived
4871 * until now. (those are in pending_raid_disks)
4872 *
4873 * the method: pick the first pending disk, collect all disks with
4874 * the same UUID, remove all from the pending list and put them into
4875 * the 'same_array' list. Then order this list based on superblock
4876 * update time (freshest comes first), kick out 'old' disks and
4877 * compare superblocks. If everything's fine then run it.
4878 *
4879 * If "unit" is allocated, then bump its reference count
4880 */
4881static void autorun_devices(int part)
4882{
159ec1fc 4883 mdk_rdev_t *rdev0, *rdev, *tmp;
1da177e4
LT		 4884 mddev_t *mddev;
4885 char b[BDEVNAME_SIZE];
4886
4887 printk(KERN_INFO "md: autorun ...\n");
4888 while (!list_empty(&pending_raid_disks)) {
e8703fe1 4889 int unit;
1da177e4 4890 dev_t dev;
ad01c9e3 4891 LIST_HEAD(candidates);
1da177e4
LT		 4892 rdev0 = list_entry(pending_raid_disks.next,
4893 mdk_rdev_t, same_set);
4894
4895 printk(KERN_INFO "md: considering %s ...\n",
4896 bdevname(rdev0->bdev,b));
4897 INIT_LIST_HEAD(&candidates);
159ec1fc 4898 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 4899 if (super_90_load(rdev, rdev0, 0) >= 0) {
4900 printk(KERN_INFO "md: adding %s ...\n",
4901 bdevname(rdev->bdev,b));
4902 list_move(&rdev->same_set, &candidates);
4903 }
4904 /*
4905 * now we have a set of devices, with all of them having
4906 * mostly sane superblocks. It's time to allocate the
4907 * mddev.
4908 */
e8703fe1
N		 4909 if (part) {
4910 dev = MKDEV(mdp_major,
4911 rdev0->preferred_minor << MdpMinorShift);
4912 unit = MINOR(dev) >> MdpMinorShift;
4913 } else {
4914 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4915 unit = MINOR(dev);
4916 }
4917 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 4918 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4919 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4920 break;
4921 }
1da177e4
LT		 4922
4923 md_probe(dev, NULL, NULL);
4924 mddev = mddev_find(dev);
9bbbca3a
NB		 4925 if (!mddev || !mddev->gendisk) {
4926 if (mddev)
4927 mddev_put(mddev);
4928 printk(KERN_ERR
1da177e4
LT		 4929 "md: cannot allocate memory for md drive.\n");
4930 break;
4931 }
4932 if (mddev_lock(mddev))
4933 printk(KERN_WARNING "md: %s locked, cannot run\n",
4934 mdname(mddev));
4935 else if (mddev->raid_disks || mddev->major_version
4936 || !list_empty(&mddev->disks)) {
4937 printk(KERN_WARNING
4938 "md: %s already running, cannot run %s\n",
4939 mdname(mddev), bdevname(rdev0->bdev,b));
4940 mddev_unlock(mddev);
4941 } else {
4942 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 4943 mddev->persistent = 1;
159ec1fc 4944 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 4945 list_del_init(&rdev->same_set);
4946 if (bind_rdev_to_array(rdev, mddev))
4947 export_rdev(rdev);
4948 }
4949 autorun_array(mddev);
4950 mddev_unlock(mddev);
4951 }
4952 /* on success, candidates will be empty, on error
4953 * it won't...
4954 */
159ec1fc 4955 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 4956 list_del_init(&rdev->same_set);
1da177e4 4957 export_rdev(rdev);
4b80991c 4958 }
1da177e4
LT		 4959 mddev_put(mddev);
4960 }
4961 printk(KERN_INFO "md: ... autorun DONE.\n");
4962}
fdee8ae4 4963#endif /* !MODULE */
1da177e4 4964
1da177e4
LT		 4965static int get_version(void __user * arg)
4966{
4967 mdu_version_t ver;
4968
4969 ver.major = MD_MAJOR_VERSION;
4970 ver.minor = MD_MINOR_VERSION;
4971 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4972
4973 if (copy_to_user(arg, &ver, sizeof(ver)))
4974 return -EFAULT;
4975
4976 return 0;
4977}
4978
4979static int get_array_info(mddev_t * mddev, void __user * arg)
4980{
4981 mdu_array_info_t info;
a9f326eb 4982 int nr,working,insync,failed,spare;
1da177e4 4983 mdk_rdev_t *rdev;
1da177e4 4984
a9f326eb 4985 nr=working=insync=failed=spare=0;
159ec1fc 4986 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 4987 nr++;
b2d444d7 4988 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4989 failed++;
4990 else {
4991 working++;
b2d444d7 4992 if (test_bit(In_sync, &rdev->flags))
a9f326eb 4993 insync++;
1da177e4
LT		 4994 else
4995 spare++;
4996 }
4997 }
4998
4999 info.major_version = mddev->major_version;
5000 info.minor_version = mddev->minor_version;
5001 info.patch_version = MD_PATCHLEVEL_VERSION;
5002 info.ctime = mddev->ctime;
5003 info.level = mddev->level;
58c0fed4
AN		 5004 info.size = mddev->dev_sectors / 2;
5005 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5006 info.size = -1;
1da177e4
LT		 5007 info.nr_disks = nr;
5008 info.raid_disks = mddev->raid_disks;
5009 info.md_minor = mddev->md_minor;
5010 info.not_persistent= !mddev->persistent;
5011
5012 info.utime = mddev->utime;
5013 info.state = 0;
5014 if (mddev->in_sync)
5015 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5016 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5017 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5018 info.active_disks = insync;
1da177e4
LT		 5019 info.working_disks = working;
5020 info.failed_disks = failed;
5021 info.spare_disks = spare;
5022
5023 info.layout = mddev->layout;
9d8f0363 5024 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5025
5026 if (copy_to_user(arg, &info, sizeof(info)))
5027 return -EFAULT;
5028
5029 return 0;
5030}
5031
87162a28 5032static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 5033{
5034 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5035 char *ptr, *buf = NULL;
5036 int err = -ENOMEM;
5037
b5470dc5
DW		 5038 if (md_allow_write(mddev))
5039 file = kmalloc(sizeof(*file), GFP_NOIO);
5040 else
5041 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5042
32a7627c
N		 5043 if (!file)
5044 goto out;
5045
5046 /* bitmap disabled, zero the first byte and copy out */
5047 if (!mddev->bitmap || !mddev->bitmap->file) {
5048 file->pathname[0] = '\0';
5049 goto copy_out;
5050 }
5051
5052 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5053 if (!buf)
5054 goto out;
5055
6bcfd601
CH		 5056 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5057 if (IS_ERR(ptr))
32a7627c
N		 5058 goto out;
5059
5060 strcpy(file->pathname, ptr);
5061
5062copy_out:
5063 err = 0;
5064 if (copy_to_user(arg, file, sizeof(*file)))
5065 err = -EFAULT;
5066out:
5067 kfree(buf);
5068 kfree(file);
5069 return err;
5070}
5071
1da177e4
LT		 5072static int get_disk_info(mddev_t * mddev, void __user * arg)
5073{
5074 mdu_disk_info_t info;
1da177e4
LT		 5075 mdk_rdev_t *rdev;
5076
5077 if (copy_from_user(&info, arg, sizeof(info)))
5078 return -EFAULT;
5079
26ef379f 5080 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5081 if (rdev) {
5082 info.major = MAJOR(rdev->bdev->bd_dev);
5083 info.minor = MINOR(rdev->bdev->bd_dev);
5084 info.raid_disk = rdev->raid_disk;
5085 info.state = 0;
b2d444d7 5086 if (test_bit(Faulty, &rdev->flags))
1da177e4 5087 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5088 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5089 info.state |= (1<<MD_DISK_ACTIVE);
5090 info.state |= (1<<MD_DISK_SYNC);
5091 }
8ddf9efe
N		 5092 if (test_bit(WriteMostly, &rdev->flags))
5093 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5094 } else {
5095 info.major = info.minor = 0;
5096 info.raid_disk = -1;
5097 info.state = (1<<MD_DISK_REMOVED);
5098 }
5099
5100 if (copy_to_user(arg, &info, sizeof(info)))
5101 return -EFAULT;
5102
5103 return 0;
5104}
5105
5106static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5107{
5108 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5109 mdk_rdev_t *rdev;
5110 dev_t dev = MKDEV(info->major,info->minor);
5111
5112 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5113 return -EOVERFLOW;
5114
5115 if (!mddev->raid_disks) {
5116 int err;
5117 /* expecting a device which has a superblock */
5118 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5119 if (IS_ERR(rdev)) {
5120 printk(KERN_WARNING
5121 "md: md_import_device returned %ld\n",
5122 PTR_ERR(rdev));
5123 return PTR_ERR(rdev);
5124 }
5125 if (!list_empty(&mddev->disks)) {
5126 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5127 mdk_rdev_t, same_set);
a9f326eb 5128 err = super_types[mddev->major_version]
1da177e4
LT		 5129 .load_super(rdev, rdev0, mddev->minor_version);
5130 if (err < 0) {
5131 printk(KERN_WARNING
5132 "md: %s has different UUID to %s\n",
5133 bdevname(rdev->bdev,b),
5134 bdevname(rdev0->bdev,b2));
5135 export_rdev(rdev);
5136 return -EINVAL;
5137 }
5138 }
5139 err = bind_rdev_to_array(rdev, mddev);
5140 if (err)
5141 export_rdev(rdev);
5142 return err;
5143 }
5144
5145 /*
5146 * add_new_disk can be used once the array is assembled
5147 * to add "hot spares". They must already have a superblock
5148 * written
5149 */
5150 if (mddev->pers) {
5151 int err;
5152 if (!mddev->pers->hot_add_disk) {
5153 printk(KERN_WARNING
5154 "%s: personality does not support diskops!\n",
5155 mdname(mddev));
5156 return -EINVAL;
5157 }
7b1e35f6
N		 5158 if (mddev->persistent)
5159 rdev = md_import_device(dev, mddev->major_version,
5160 mddev->minor_version);
5161 else
5162 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5163 if (IS_ERR(rdev)) {
5164 printk(KERN_WARNING
5165 "md: md_import_device returned %ld\n",
5166 PTR_ERR(rdev));
5167 return PTR_ERR(rdev);
5168 }
1a855a06 5169 /* set saved_raid_disk if appropriate */
41158c7e
N		 5170 if (!mddev->persistent) {
5171 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5172 info->raid_disk < mddev->raid_disks) {
41158c7e 5173 rdev->raid_disk = info->raid_disk;
bf572541
N		 5174 set_bit(In_sync, &rdev->flags);
5175 } else
41158c7e
N		 5176 rdev->raid_disk = -1;
5177 } else
5178 super_types[mddev->major_version].
5179 validate_super(mddev, rdev);
1a855a06
N		 5180 if (test_bit(In_sync, &rdev->flags))
5181 rdev->saved_raid_disk = rdev->raid_disk;
5182 else
5183 rdev->saved_raid_disk = -1;
41158c7e 5184
b2d444d7 5185 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5186 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5187 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5188 else
5189 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5190
1da177e4
LT		 5191 rdev->raid_disk = -1;
5192 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5193 if (!err && !mddev->pers->hot_remove_disk) {
5194 /* If there is hot_add_disk but no hot_remove_disk
5195 * then added disks for geometry changes,
5196 * and should be added immediately.
5197 */
5198 super_types[mddev->major_version].
5199 validate_super(mddev, rdev);
5200 err = mddev->pers->hot_add_disk(mddev, rdev);
5201 if (err)
5202 unbind_rdev_from_array(rdev);
5203 }
1da177e4
LT		 5204 if (err)
5205 export_rdev(rdev);
52664732 5206 else
00bcb4ac 5207 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5208
17571284 5209 md_update_sb(mddev, 1);
72a23c21
NB		 5210 if (mddev->degraded)
5211 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5212 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 5213 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5214 return err;
5215 }
5216
5217 /* otherwise, add_new_disk is only allowed
5218 * for major_version==0 superblocks
5219 */
5220 if (mddev->major_version != 0) {
5221 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5222 mdname(mddev));
5223 return -EINVAL;
5224 }
5225
5226 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5227 int err;
d710e138 5228 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5229 if (IS_ERR(rdev)) {
5230 printk(KERN_WARNING
5231 "md: error, md_import_device() returned %ld\n",
5232 PTR_ERR(rdev));
5233 return PTR_ERR(rdev);
5234 }
5235 rdev->desc_nr = info->number;
5236 if (info->raid_disk < mddev->raid_disks)
5237 rdev->raid_disk = info->raid_disk;
5238 else
5239 rdev->raid_disk = -1;
5240
1da177e4 5241 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5242 if (info->state & (1<<MD_DISK_SYNC))
5243 set_bit(In_sync, &rdev->flags);
1da177e4 5244
8ddf9efe
N		 5245 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5246 set_bit(WriteMostly, &rdev->flags);
5247
1da177e4
LT		 5248 if (!mddev->persistent) {
5249 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5250 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5251 } else
0f420358 5252 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
8190e754 5253 rdev->sectors = rdev->sb_start;
1da177e4 5254
2bf071bf
N		 5255 err = bind_rdev_to_array(rdev, mddev);
5256 if (err) {
5257 export_rdev(rdev);
5258 return err;
5259 }
1da177e4
LT		 5260 }
5261
5262 return 0;
5263}
5264
5265static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5266{
5267 char b[BDEVNAME_SIZE];
5268 mdk_rdev_t *rdev;
5269
1da177e4
LT		 5270 rdev = find_rdev(mddev, dev);
5271 if (!rdev)
5272 return -ENXIO;
5273
5274 if (rdev->raid_disk >= 0)
5275 goto busy;
5276
5277 kick_rdev_from_array(rdev);
850b2b42 5278 md_update_sb(mddev, 1);
d7603b7e 5279 md_new_event(mddev);
1da177e4
LT		 5280
5281 return 0;
5282busy:
fdefa4d8 5283 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5284 bdevname(rdev->bdev,b), mdname(mddev));
5285 return -EBUSY;
5286}
5287
5288static int hot_add_disk(mddev_t * mddev, dev_t dev)
5289{
5290 char b[BDEVNAME_SIZE];
5291 int err;
1da177e4
LT		 5292 mdk_rdev_t *rdev;
5293
5294 if (!mddev->pers)
5295 return -ENODEV;
5296
5297 if (mddev->major_version != 0) {
5298 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5299 " version-0 superblocks.\n",
5300 mdname(mddev));
5301 return -EINVAL;
5302 }
5303 if (!mddev->pers->hot_add_disk) {
5304 printk(KERN_WARNING
5305 "%s: personality does not support diskops!\n",
5306 mdname(mddev));
5307 return -EINVAL;
5308 }
5309
d710e138 5310 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5311 if (IS_ERR(rdev)) {
5312 printk(KERN_WARNING
5313 "md: error, md_import_device() returned %ld\n",
5314 PTR_ERR(rdev));
5315 return -EINVAL;
5316 }
5317
5318 if (mddev->persistent)
0f420358 5319 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1da177e4 5320 else
77304d2a 5321 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5322
8190e754 5323 rdev->sectors = rdev->sb_start;
1da177e4 5324
b2d444d7 5325 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5326 printk(KERN_WARNING
5327 "md: can not hot-add faulty %s disk to %s!\n",
5328 bdevname(rdev->bdev,b), mdname(mddev));
5329 err = -EINVAL;
5330 goto abort_export;
5331 }
b2d444d7 5332 clear_bit(In_sync, &rdev->flags);
1da177e4 5333 rdev->desc_nr = -1;
5842730d 5334 rdev->saved_raid_disk = -1;
2bf071bf
N		 5335 err = bind_rdev_to_array(rdev, mddev);
5336 if (err)
5337 goto abort_export;
1da177e4
LT		 5338
5339 /*
5340 * The rest should better be atomic, we can have disk failures
5341 * noticed in interrupt contexts ...
5342 */
5343
1da177e4
LT		 5344 rdev->raid_disk = -1;
5345
850b2b42 5346 md_update_sb(mddev, 1);
1da177e4
LT		 5347
5348 /*
5349 * Kick recovery, maybe this spare has to be added to the
5350 * array immediately.
5351 */
5352 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5353 md_wakeup_thread(mddev->thread);
d7603b7e 5354 md_new_event(mddev);
1da177e4
LT		 5355 return 0;
5356
1da177e4
LT		 5357abort_export:
5358 export_rdev(rdev);
5359 return err;
5360}
5361
32a7627c
N		 5362static int set_bitmap_file(mddev_t *mddev, int fd)
5363{
5364 int err;
5365
36fa3063
N		 5366 if (mddev->pers) {
5367 if (!mddev->pers->quiesce)
5368 return -EBUSY;
5369 if (mddev->recovery || mddev->sync_thread)
5370 return -EBUSY;
5371 /* we should be able to change the bitmap.. */
5372 }
32a7627c 5373
32a7627c 5374
36fa3063
N		 5375 if (fd >= 0) {
5376 if (mddev->bitmap)
5377 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5378 mddev->bitmap_info.file = fget(fd);
32a7627c 5379
c3d9714e 5380 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5381 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5382 mdname(mddev));
5383 return -EBADF;
5384 }
5385
c3d9714e 5386 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5387 if (err) {
5388 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5389 mdname(mddev));
c3d9714e
N		 5390 fput(mddev->bitmap_info.file);
5391 mddev->bitmap_info.file = NULL;
36fa3063
N		 5392 return err;
5393 }
c3d9714e 5394 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5395 } else if (mddev->bitmap == NULL)
5396 return -ENOENT; /* cannot remove what isn't there */
5397 err = 0;
5398 if (mddev->pers) {
5399 mddev->pers->quiesce(mddev, 1);
69e51b44 5400 if (fd >= 0) {
36fa3063 5401 err = bitmap_create(mddev);
69e51b44
N		 5402 if (!err)
5403 err = bitmap_load(mddev);
5404 }
d7375ab3 5405 if (fd < 0 || err) {
36fa3063 5406 bitmap_destroy(mddev);
d7375ab3
N		 5407 fd = -1; /* make sure to put the file */
5408 }
36fa3063 5409 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5410 }
5411 if (fd < 0) {
c3d9714e
N		 5412 if (mddev->bitmap_info.file) {
5413 restore_bitmap_write_access(mddev->bitmap_info.file);
5414 fput(mddev->bitmap_info.file);
acc55e22 5415 }
c3d9714e 5416 mddev->bitmap_info.file = NULL;
36fa3063
N		 5417 }
5418
32a7627c
N		 5419 return err;
5420}
5421
1da177e4
LT		 5422/*
5423 * set_array_info is used two different ways
5424 * The original usage is when creating a new array.
5425 * In this usage, raid_disks is > 0 and it together with
5426 * level, size, not_persistent,layout,chunksize determine the
5427 * shape of the array.
5428 * This will always create an array with a type-0.90.0 superblock.
5429 * The newer usage is when assembling an array.
5430 * In this case raid_disks will be 0, and the major_version field is
5431 * use to determine which style super-blocks are to be found on the devices.
5432 * The minor and patch _version numbers are also kept incase the
5433 * super_block handler wishes to interpret them.
5434 */
5435static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5436{
5437
5438 if (info->raid_disks == 0) {
5439 /* just setting version number for superblock loading */
5440 if (info->major_version < 0 ||
50511da3 5441 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5442 super_types[info->major_version].name == NULL) {
5443 /* maybe try to auto-load a module? */
5444 printk(KERN_INFO
5445 "md: superblock version %d not known\n",
5446 info->major_version);
5447 return -EINVAL;
5448 }
5449 mddev->major_version = info->major_version;
5450 mddev->minor_version = info->minor_version;
5451 mddev->patch_version = info->patch_version;
3f9d7b0d 5452 mddev->persistent = !info->not_persistent;
cbd19983
N		 5453 /* ensure mddev_put doesn't delete this now that there
5454 * is some minimal configuration.
5455 */
5456 mddev->ctime = get_seconds();
1da177e4
LT		 5457 return 0;
5458 }
5459 mddev->major_version = MD_MAJOR_VERSION;
5460 mddev->minor_version = MD_MINOR_VERSION;
5461 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5462 mddev->ctime = get_seconds();
5463
5464 mddev->level = info->level;
17115e03 5465 mddev->clevel[0] = 0;
58c0fed4 5466 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5467 mddev->raid_disks = info->raid_disks;
5468 /* don't set md_minor, it is determined by which /dev/md* was
5469 * openned
5470 */
5471 if (info->state & (1<<MD_SB_CLEAN))
5472 mddev->recovery_cp = MaxSector;
5473 else
5474 mddev->recovery_cp = 0;
5475 mddev->persistent = ! info->not_persistent;
e691063a 5476 mddev->external = 0;
1da177e4
LT		 5477
5478 mddev->layout = info->layout;
9d8f0363 5479 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5480
5481 mddev->max_disks = MD_SB_DISKS;
5482
e691063a
N		 5483 if (mddev->persistent)
5484 mddev->flags = 0;
850b2b42 5485 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5486
c3d9714e
N		 5487 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5488 mddev->bitmap_info.offset = 0;
b2a2703c 5489
f6705578
N		 5490 mddev->reshape_position = MaxSector;
5491
1da177e4
LT		 5492 /*
5493 * Generate a 128 bit UUID
5494 */
5495 get_random_bytes(mddev->uuid, 16);
5496
f6705578 5497 mddev->new_level = mddev->level;
664e7c41 5498 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5499 mddev->new_layout = mddev->layout;
5500 mddev->delta_disks = 0;
5501
1da177e4
LT		 5502 return 0;
5503}
5504
1f403624
DW		 5505void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5506{
b522adcd
DW		 5507 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5508
5509 if (mddev->external_size)
5510 return;
5511
1f403624
DW		 5512 mddev->array_sectors = array_sectors;
5513}
5514EXPORT_SYMBOL(md_set_array_sectors);
5515
d71f9f88 5516static int update_size(mddev_t *mddev, sector_t num_sectors)
a35b0d69 5517{
159ec1fc 5518 mdk_rdev_t *rdev;
a35b0d69 5519 int rv;
d71f9f88 5520 int fit = (num_sectors == 0);
a35b0d69
N		 5521
5522 if (mddev->pers->resize == NULL)
5523 return -EINVAL;
d71f9f88
AN		 5524 /* The "num_sectors" is the number of sectors of each device that
5525 * is used. This can only make sense for arrays with redundancy.
5526 * linear and raid0 always use whatever space is available. We can only
5527 * consider changing this number if no resync or reconstruction is
5528 * happening, and if the new size is acceptable. It must fit before the
0f420358 5529 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5530 * of each device. If num_sectors is zero, we find the largest size
5531 * that fits.
5532
a35b0d69
N		 5533 */
5534 if (mddev->sync_thread)
5535 return -EBUSY;
dba034ee
N		 5536 if (mddev->bitmap)
5537 /* Sorry, cannot grow a bitmap yet, just remove it,
5538 * grow, and re-add.
5539 */
5540 return -EBUSY;
159ec1fc 5541 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5542 sector_t avail = rdev->sectors;
01ab5662 5543
d71f9f88
AN		 5544 if (fit && (num_sectors == 0 || num_sectors > avail))
5545 num_sectors = avail;
5546 if (avail < num_sectors)
a35b0d69
N		 5547 return -ENOSPC;
5548 }
d71f9f88 5549 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5550 if (!rv)
5551 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5552 return rv;
5553}
5554
da943b99
N		 5555static int update_raid_disks(mddev_t *mddev, int raid_disks)
5556{
5557 int rv;
5558 /* change the number of raid disks */
63c70c4f 5559 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5560 return -EINVAL;
5561 if (raid_disks <= 0 ||
233fca36 5562 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5563 return -EINVAL;
63c70c4f 5564 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5565 return -EBUSY;
63c70c4f
N		 5566 mddev->delta_disks = raid_disks - mddev->raid_disks;
5567
5568 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 5569 return rv;
5570}
5571
5572
1da177e4
LT		 5573/*
5574 * update_array_info is used to change the configuration of an
5575 * on-line array.
5576 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5577 * fields in the info are checked against the array.
5578 * Any differences that cannot be handled will cause an error.
5579 * Normally, only one change can be managed at a time.
5580 */
5581static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5582{
5583 int rv = 0;
5584 int cnt = 0;
36fa3063
N		 5585 int state = 0;
5586
5587 /* calculate expected state,ignoring low bits */
c3d9714e 5588 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5589 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5590
5591 if (mddev->major_version != info->major_version ||
5592 mddev->minor_version != info->minor_version ||
5593/* mddev->patch_version != info->patch_version || */
5594 mddev->ctime != info->ctime ||
5595 mddev->level != info->level ||
5596/* mddev->layout != info->layout || */
5597 !mddev->persistent != info->not_persistent||
9d8f0363 5598 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5599 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5600 ((state^info->state) & 0xfffffe00)
5601 )
1da177e4
LT		 5602 return -EINVAL;
5603 /* Check there is only one change */
58c0fed4
AN		 5604 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5605 cnt++;
5606 if (mddev->raid_disks != info->raid_disks)
5607 cnt++;
5608 if (mddev->layout != info->layout)
5609 cnt++;
5610 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5611 cnt++;
5612 if (cnt == 0)
5613 return 0;
5614 if (cnt > 1)
5615 return -EINVAL;
1da177e4
LT		 5616
5617 if (mddev->layout != info->layout) {
5618 /* Change layout
5619 * we don't need to do anything at the md level, the
5620 * personality will take care of it all.
5621 */
50ac168a 5622 if (mddev->pers->check_reshape == NULL)
1da177e4 5623 return -EINVAL;
597a711b
N		 5624 else {
5625 mddev->new_layout = info->layout;
50ac168a 5626 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5627 if (rv)
5628 mddev->new_layout = mddev->layout;
5629 return rv;
5630 }
1da177e4 5631 }
58c0fed4 5632 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5633 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5634
da943b99
N		 5635 if (mddev->raid_disks != info->raid_disks)
5636 rv = update_raid_disks(mddev, info->raid_disks);
5637
36fa3063
N		 5638 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5639 if (mddev->pers->quiesce == NULL)
5640 return -EINVAL;
5641 if (mddev->recovery || mddev->sync_thread)
5642 return -EBUSY;
5643 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5644 /* add the bitmap */
5645 if (mddev->bitmap)
5646 return -EEXIST;
c3d9714e 5647 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5648 return -EINVAL;
c3d9714e
N		 5649 mddev->bitmap_info.offset =
5650 mddev->bitmap_info.default_offset;
36fa3063
N		 5651 mddev->pers->quiesce(mddev, 1);
5652 rv = bitmap_create(mddev);
69e51b44
N		 5653 if (!rv)
5654 rv = bitmap_load(mddev);
36fa3063
N		 5655 if (rv)
5656 bitmap_destroy(mddev);
5657 mddev->pers->quiesce(mddev, 0);
5658 } else {
5659 /* remove the bitmap */
5660 if (!mddev->bitmap)
5661 return -ENOENT;
5662 if (mddev->bitmap->file)
5663 return -EINVAL;
5664 mddev->pers->quiesce(mddev, 1);
5665 bitmap_destroy(mddev);
5666 mddev->pers->quiesce(mddev, 0);
c3d9714e 5667 mddev->bitmap_info.offset = 0;
36fa3063
N		 5668 }
5669 }
850b2b42 5670 md_update_sb(mddev, 1);
1da177e4
LT		 5671 return rv;
5672}
5673
5674static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5675{
5676 mdk_rdev_t *rdev;
5677
5678 if (mddev->pers == NULL)
5679 return -ENODEV;
5680
5681 rdev = find_rdev(mddev, dev);
5682 if (!rdev)
5683 return -ENODEV;
5684
5685 md_error(mddev, rdev);
5686 return 0;
5687}
5688
2f9618ce
AN		 5689/*
5690 * We have a problem here : there is no easy way to give a CHS
5691 * virtual geometry. We currently pretend that we have a 2 heads
5692 * 4 sectors (with a BIG number of cylinders...). This drives
5693 * dosfs just mad... ;-)
5694 */
a885c8c4
CH		 5695static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5696{
5697 mddev_t *mddev = bdev->bd_disk->private_data;
5698
5699 geo->heads = 2;
5700 geo->sectors = 4;
49ce6cea 5701 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 5702 return 0;
5703}
5704
a39907fa 5705static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 5706 unsigned int cmd, unsigned long arg)
5707{
5708 int err = 0;
5709 void __user *argp = (void __user *)arg;
1da177e4 5710 mddev_t *mddev = NULL;
e2218350 5711 int ro;
1da177e4
LT		 5712
5713 if (!capable(CAP_SYS_ADMIN))
5714 return -EACCES;
5715
5716 /*
5717 * Commands dealing with the RAID driver but not any
5718 * particular array:
5719 */
5720 switch (cmd)
5721 {
5722 case RAID_VERSION:
5723 err = get_version(argp);
5724 goto done;
5725
5726 case PRINT_RAID_DEBUG:
5727 err = 0;
5728 md_print_devices();
5729 goto done;
5730
5731#ifndef MODULE
5732 case RAID_AUTORUN:
5733 err = 0;
5734 autostart_arrays(arg);
5735 goto done;
5736#endif
5737 default:;
5738 }
5739
5740 /*
5741 * Commands creating/starting a new array:
5742 */
5743
a39907fa 5744 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 5745
5746 if (!mddev) {
5747 BUG();
5748 goto abort;
5749 }
5750
1da177e4
LT		 5751 err = mddev_lock(mddev);
5752 if (err) {
5753 printk(KERN_INFO
5754 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5755 err, cmd);
5756 goto abort;
5757 }
5758
5759 switch (cmd)
5760 {
5761 case SET_ARRAY_INFO:
5762 {
5763 mdu_array_info_t info;
5764 if (!arg)
5765 memset(&info, 0, sizeof(info));
5766 else if (copy_from_user(&info, argp, sizeof(info))) {
5767 err = -EFAULT;
5768 goto abort_unlock;
5769 }
5770 if (mddev->pers) {
5771 err = update_array_info(mddev, &info);
5772 if (err) {
5773 printk(KERN_WARNING "md: couldn't update"
5774 " array info. %d\n", err);
5775 goto abort_unlock;
5776 }
5777 goto done_unlock;
5778 }
5779 if (!list_empty(&mddev->disks)) {
5780 printk(KERN_WARNING
5781 "md: array %s already has disks!\n",
5782 mdname(mddev));
5783 err = -EBUSY;
5784 goto abort_unlock;
5785 }
5786 if (mddev->raid_disks) {
5787 printk(KERN_WARNING
5788 "md: array %s already initialised!\n",
5789 mdname(mddev));
5790 err = -EBUSY;
5791 goto abort_unlock;
5792 }
5793 err = set_array_info(mddev, &info);
5794 if (err) {
5795 printk(KERN_WARNING "md: couldn't set"
5796 " array info. %d\n", err);
5797 goto abort_unlock;
5798 }
5799 }
5800 goto done_unlock;
5801
5802 default:;
5803 }
5804
5805 /*
5806 * Commands querying/configuring an existing array:
5807 */
32a7627c 5808 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 5809 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 5810 if ((!mddev->raid_disks && !mddev->external)
5811 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5812 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5813 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 5814 err = -ENODEV;
5815 goto abort_unlock;
5816 }
5817
5818 /*
5819 * Commands even a read-only array can execute:
5820 */
5821 switch (cmd)
5822 {
5823 case GET_ARRAY_INFO:
5824 err = get_array_info(mddev, argp);
5825 goto done_unlock;
5826
32a7627c 5827 case GET_BITMAP_FILE:
87162a28 5828 err = get_bitmap_file(mddev, argp);
32a7627c
N		 5829 goto done_unlock;
5830
1da177e4
LT		 5831 case GET_DISK_INFO:
5832 err = get_disk_info(mddev, argp);
5833 goto done_unlock;
5834
5835 case RESTART_ARRAY_RW:
5836 err = restart_array(mddev);
5837 goto done_unlock;
5838
5839 case STOP_ARRAY:
d710e138 5840 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 5841 goto done_unlock;
5842
5843 case STOP_ARRAY_RO:
a4bd82d0 5844 err = md_set_readonly(mddev, 1);
1da177e4
LT		 5845 goto done_unlock;
5846
e2218350
DW		 5847 case BLKROSET:
5848 if (get_user(ro, (int __user *)(arg))) {
5849 err = -EFAULT;
5850 goto done_unlock;
5851 }
5852 err = -EINVAL;
5853
5854 /* if the bdev is going readonly the value of mddev->ro
5855 * does not matter, no writes are coming
5856 */
5857 if (ro)
5858 goto done_unlock;
5859
5860 /* are we are already prepared for writes? */
5861 if (mddev->ro != 1)
5862 goto done_unlock;
5863
5864 /* transitioning to readauto need only happen for
5865 * arrays that call md_write_start
5866 */
5867 if (mddev->pers) {
5868 err = restart_array(mddev);
5869 if (err == 0) {
5870 mddev->ro = 2;
5871 set_disk_ro(mddev->gendisk, 0);
5872 }
5873 }
5874 goto done_unlock;
1da177e4
LT		 5875 }
5876
5877 /*
5878 * The remaining ioctls are changing the state of the
f91de92e
N		 5879 * superblock, so we do not allow them on read-only arrays.
5880 * However non-MD ioctls (e.g. get-size) will still come through
5881 * here and hit the 'default' below, so only disallow
5882 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 5883 */
bb57fc64 5884 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 5885 if (mddev->ro == 2) {
5886 mddev->ro = 0;
00bcb4ac 5887 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 5888 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5889 md_wakeup_thread(mddev->thread);
f91de92e
N		 5890 } else {
5891 err = -EROFS;
5892 goto abort_unlock;
5893 }
1da177e4
LT		 5894 }
5895
5896 switch (cmd)
5897 {
5898 case ADD_NEW_DISK:
5899 {
5900 mdu_disk_info_t info;
5901 if (copy_from_user(&info, argp, sizeof(info)))
5902 err = -EFAULT;
5903 else
5904 err = add_new_disk(mddev, &info);
5905 goto done_unlock;
5906 }
5907
5908 case HOT_REMOVE_DISK:
5909 err = hot_remove_disk(mddev, new_decode_dev(arg));
5910 goto done_unlock;
5911
5912 case HOT_ADD_DISK:
5913 err = hot_add_disk(mddev, new_decode_dev(arg));
5914 goto done_unlock;
5915
5916 case SET_DISK_FAULTY:
5917 err = set_disk_faulty(mddev, new_decode_dev(arg));
5918 goto done_unlock;
5919
5920 case RUN_ARRAY:
d710e138 5921 err = do_md_run(mddev);
1da177e4
LT		 5922 goto done_unlock;
5923
32a7627c
N		 5924 case SET_BITMAP_FILE:
5925 err = set_bitmap_file(mddev, (int)arg);
5926 goto done_unlock;
5927
1da177e4 5928 default:
1da177e4
LT		 5929 err = -EINVAL;
5930 goto abort_unlock;
5931 }
5932
5933done_unlock:
5934abort_unlock:
d3374825
N		 5935 if (mddev->hold_active == UNTIL_IOCTL &&
5936 err != -EINVAL)
5937 mddev->hold_active = 0;
1da177e4
LT		 5938 mddev_unlock(mddev);
5939
5940 return err;
5941done:
5942 if (err)
5943 MD_BUG();
5944abort:
5945 return err;
5946}
aa98aa31
AB		 5947#ifdef CONFIG_COMPAT
5948static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5949 unsigned int cmd, unsigned long arg)
5950{
5951 switch (cmd) {
5952 case HOT_REMOVE_DISK:
5953 case HOT_ADD_DISK:
5954 case SET_DISK_FAULTY:
5955 case SET_BITMAP_FILE:
5956 /* These take in integer arg, do not convert */
5957 break;
5958 default:
5959 arg = (unsigned long)compat_ptr(arg);
5960 break;
5961 }
5962
5963 return md_ioctl(bdev, mode, cmd, arg);
5964}
5965#endif /* CONFIG_COMPAT */
1da177e4 5966
a39907fa 5967static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 5968{
5969 /*
5970 * Succeed if we can lock the mddev, which confirms that
5971 * it isn't being stopped right now.
5972 */
d3374825 5973 mddev_t *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 5974 int err;
5975
d3374825
N		 5976 if (mddev->gendisk != bdev->bd_disk) {
5977 /* we are racing with mddev_put which is discarding this
5978 * bd_disk.
5979 */
5980 mddev_put(mddev);
5981 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 5982 flush_workqueue(md_misc_wq);
d3374825
N		 5983 /* Then retry the open from the top */
5984 return -ERESTARTSYS;
5985 }
5986 BUG_ON(mddev != bdev->bd_disk->private_data);
5987
c8c00a69 5988 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 5989 goto out;
5990
5991 err = 0;
f2ea68cf 5992 atomic_inc(&mddev->openers);
c8c00a69 5993 mutex_unlock(&mddev->open_mutex);
1da177e4 5994
f3b99be1 5995 check_disk_size_change(mddev->gendisk, bdev);
1da177e4
LT		 5996 out:
5997 return err;
5998}
5999
a39907fa 6000static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6001{
a39907fa 6002 mddev_t *mddev = disk->private_data;
1da177e4 6003
52e5f9d1 6004 BUG_ON(!mddev);
f2ea68cf 6005 atomic_dec(&mddev->openers);
1da177e4
LT		 6006 mddev_put(mddev);
6007
6008 return 0;
6009}
83d5cde4 6010static const struct block_device_operations md_fops =
1da177e4
LT		 6011{
6012 .owner = THIS_MODULE,
a39907fa
AV		 6013 .open = md_open,
6014 .release = md_release,
b492b852 6015 .ioctl = md_ioctl,
aa98aa31
AB		 6016#ifdef CONFIG_COMPAT
6017 .compat_ioctl = md_compat_ioctl,
6018#endif
a885c8c4 6019 .getgeo = md_getgeo,
1da177e4
LT		 6020};
6021
75c96f85 6022static int md_thread(void * arg)
1da177e4
LT		 6023{
6024 mdk_thread_t *thread = arg;
6025
1da177e4
LT		 6026 /*
6027 * md_thread is a 'system-thread', it's priority should be very
6028 * high. We avoid resource deadlocks individually in each
6029 * raid personality. (RAID5 does preallocation) We also use RR and
6030 * the very same RT priority as kswapd, thus we will never get
6031 * into a priority inversion deadlock.
6032 *
6033 * we definitely have to have equal or higher priority than
6034 * bdflush, otherwise bdflush will deadlock if there are too
6035 * many dirty RAID5 blocks.
6036 */
1da177e4 6037
6985c43f 6038 allow_signal(SIGKILL);
a6fb0934 6039 while (!kthread_should_stop()) {
1da177e4 6040
93588e22
N		 6041 /* We need to wait INTERRUPTIBLE so that
6042 * we don't add to the load-average.
6043 * That means we need to be sure no signals are
6044 * pending
6045 */
6046 if (signal_pending(current))
6047 flush_signals(current);
6048
6049 wait_event_interruptible_timeout
6050 (thread->wqueue,
6051 test_bit(THREAD_WAKEUP, &thread->flags)
6052 || kthread_should_stop(),
6053 thread->timeout);
1da177e4 6054
6c987910
N		 6055 clear_bit(THREAD_WAKEUP, &thread->flags);
6056 if (!kthread_should_stop())
589a594b 6057 thread->run(thread->mddev);
1da177e4 6058 }
a6fb0934 6059
1da177e4
LT		 6060 return 0;
6061}
6062
6063void md_wakeup_thread(mdk_thread_t *thread)
6064{
6065 if (thread) {
6066 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6067 set_bit(THREAD_WAKEUP, &thread->flags);
6068 wake_up(&thread->wqueue);
6069 }
6070}
6071
6072mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6073 const char *name)
6074{
6075 mdk_thread_t *thread;
1da177e4 6076
9ffae0cf 6077 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 6078 if (!thread)
6079 return NULL;
6080
1da177e4
LT		 6081 init_waitqueue_head(&thread->wqueue);
6082
1da177e4
LT		 6083 thread->run = run;
6084 thread->mddev = mddev;
32a7627c 6085 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6086 thread->tsk = kthread_run(md_thread, thread,
6087 "%s_%s",
6088 mdname(thread->mddev),
6089 name ?: mddev->pers->name);
a6fb0934 6090 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6091 kfree(thread);
6092 return NULL;
6093 }
1da177e4
LT		 6094 return thread;
6095}
6096
1da177e4
LT		 6097void md_unregister_thread(mdk_thread_t *thread)
6098{
e0cf8f04
N		 6099 if (!thread)
6100 return;
ba25f9dc 6101 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 6102
6103 kthread_stop(thread->tsk);
1da177e4
LT		 6104 kfree(thread);
6105}
6106
6107void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6108{
6109 if (!mddev) {
6110 MD_BUG();
6111 return;
6112 }
6113
b2d444d7 6114 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6115 return;
6bfe0b49
DW		 6116
6117 if (mddev->external)
6118 set_bit(Blocked, &rdev->flags);
32a7627c 6119/*
1da177e4
LT		 6120 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6121 mdname(mddev),
6122 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6123 __builtin_return_address(0),__builtin_return_address(1),
6124 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 6125*/
d0a0a5ee
AM		 6126 if (!mddev->pers)
6127 return;
1da177e4
LT		 6128 if (!mddev->pers->error_handler)
6129 return;
6130 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6131 if (mddev->degraded)
6132 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6133 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6134 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6135 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6136 md_wakeup_thread(mddev->thread);
768a418d 6137 if (mddev->event_work.func)
e804ac78 6138 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6139 md_new_event_inintr(mddev);
1da177e4
LT		 6140}
6141
6142/* seq_file implementation /proc/mdstat */
6143
6144static void status_unused(struct seq_file *seq)
6145{
6146 int i = 0;
6147 mdk_rdev_t *rdev;
1da177e4
LT		 6148
6149 seq_printf(seq, "unused devices: ");
6150
159ec1fc 6151 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6152 char b[BDEVNAME_SIZE];
6153 i++;
6154 seq_printf(seq, "%s ",
6155 bdevname(rdev->bdev,b));
6156 }
6157 if (!i)
6158 seq_printf(seq, "<none>");
6159
6160 seq_printf(seq, "\n");
6161}
6162
6163
6164static void status_resync(struct seq_file *seq, mddev_t * mddev)
6165{
dd71cf6b
N		 6166 sector_t max_sectors, resync, res;
6167 unsigned long dt, db;
6168 sector_t rt;
4588b42e
N		 6169 int scale;
6170 unsigned int per_milli;
1da177e4 6171
dd71cf6b 6172 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6173
6174 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6175 max_sectors = mddev->resync_max_sectors;
1da177e4 6176 else
dd71cf6b 6177 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6178
6179 /*
6180 * Should not happen.
6181 */
dd71cf6b 6182 if (!max_sectors) {
1da177e4
LT		 6183 MD_BUG();
6184 return;
6185 }
4588b42e 6186 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6187 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6188 * u32, as those are the requirements for sector_div.
6189 * Thus 'scale' must be at least 10
6190 */
6191 scale = 10;
6192 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6193 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6194 scale++;
6195 }
6196 res = (resync>>scale)*1000;
dd71cf6b 6197 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6198
6199 per_milli = res;
1da177e4 6200 {
4588b42e 6201 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6202 seq_printf(seq, "[");
6203 for (i = 0; i < x; i++)
6204 seq_printf(seq, "=");
6205 seq_printf(seq, ">");
6206 for (i = 0; i < y; i++)
6207 seq_printf(seq, ".");
6208 seq_printf(seq, "] ");
6209 }
4588b42e 6210 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6211 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6212 "reshape" :
61df9d91
N		 6213 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6214 "check" :
6215 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6216 "resync" : "recovery"))),
6217 per_milli/10, per_milli % 10,
dd71cf6b
N		 6218 (unsigned long long) resync/2,
6219 (unsigned long long) max_sectors/2);
1da177e4
LT		 6220
6221 /*
1da177e4
LT		 6222 * dt: time from mark until now
6223 * db: blocks written from mark until now
6224 * rt: remaining time
dd71cf6b
N		 6225 *
6226 * rt is a sector_t, so could be 32bit or 64bit.
6227 * So we divide before multiply in case it is 32bit and close
6228 * to the limit.
6229 * We scale the divisor (db) by 32 to avoid loosing precision
6230 * near the end of resync when the number of remaining sectors
6231 * is close to 'db'.
6232 * We then divide rt by 32 after multiplying by db to compensate.
6233 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6234 */
6235 dt = ((jiffies - mddev->resync_mark) / HZ);
6236 if (!dt) dt++;
ff4e8d9a
N		 6237 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6238 - mddev->resync_mark_cnt;
1da177e4 6239
dd71cf6b
N		 6240 rt = max_sectors - resync; /* number of remaining sectors */
6241 sector_div(rt, db/32+1);
6242 rt *= dt;
6243 rt >>= 5;
6244
6245 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6246 ((unsigned long)rt % 60)/6);
1da177e4 6247
ff4e8d9a 6248 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6249}
6250
6251static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6252{
6253 struct list_head *tmp;
6254 loff_t l = *pos;
6255 mddev_t *mddev;
6256
6257 if (l >= 0x10000)
6258 return NULL;
6259 if (!l--)
6260 /* header */
6261 return (void*)1;
6262
6263 spin_lock(&all_mddevs_lock);
6264 list_for_each(tmp,&all_mddevs)
6265 if (!l--) {
6266 mddev = list_entry(tmp, mddev_t, all_mddevs);
6267 mddev_get(mddev);
6268 spin_unlock(&all_mddevs_lock);
6269 return mddev;
6270 }
6271 spin_unlock(&all_mddevs_lock);
6272 if (!l--)
6273 return (void*)2;/* tail */
6274 return NULL;
6275}
6276
6277static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6278{
6279 struct list_head *tmp;
6280 mddev_t *next_mddev, *mddev = v;
6281
6282 ++*pos;
6283 if (v == (void*)2)
6284 return NULL;
6285
6286 spin_lock(&all_mddevs_lock);
6287 if (v == (void*)1)
6288 tmp = all_mddevs.next;
6289 else
6290 tmp = mddev->all_mddevs.next;
6291 if (tmp != &all_mddevs)
6292 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6293 else {
6294 next_mddev = (void*)2;
6295 *pos = 0x10000;
6296 }
6297 spin_unlock(&all_mddevs_lock);
6298
6299 if (v != (void*)1)
6300 mddev_put(mddev);
6301 return next_mddev;
6302
6303}
6304
6305static void md_seq_stop(struct seq_file *seq, void *v)
6306{
6307 mddev_t *mddev = v;
6308
6309 if (mddev && v != (void*)1 && v != (void*)2)
6310 mddev_put(mddev);
6311}
6312
d7603b7e
N		 6313struct mdstat_info {
6314 int event;
6315};
6316
1da177e4
LT		 6317static int md_seq_show(struct seq_file *seq, void *v)
6318{
6319 mddev_t *mddev = v;
dd8ac336 6320 sector_t sectors;
1da177e4 6321 mdk_rdev_t *rdev;
d7603b7e 6322 struct mdstat_info *mi = seq->private;
32a7627c 6323 struct bitmap *bitmap;
1da177e4
LT		 6324
6325 if (v == (void*)1) {
2604b703 6326 struct mdk_personality *pers;
1da177e4
LT		 6327 seq_printf(seq, "Personalities : ");
6328 spin_lock(&pers_lock);
2604b703
N		 6329 list_for_each_entry(pers, &pers_list, list)
6330 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6331
6332 spin_unlock(&pers_lock);
6333 seq_printf(seq, "\n");
d7603b7e 6334 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 6335 return 0;
6336 }
6337 if (v == (void*)2) {
6338 status_unused(seq);
6339 return 0;
6340 }
6341
5dc5cf7d 6342 if (mddev_lock(mddev) < 0)
1da177e4 6343 return -EINTR;
5dc5cf7d 6344
1da177e4
LT		 6345 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6346 seq_printf(seq, "%s : %sactive", mdname(mddev),
6347 mddev->pers ? "" : "in");
6348 if (mddev->pers) {
f91de92e 6349 if (mddev->ro==1)
1da177e4 6350 seq_printf(seq, " (read-only)");
f91de92e 6351 if (mddev->ro==2)
52720ae7 6352 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6353 seq_printf(seq, " %s", mddev->pers->name);
6354 }
6355
dd8ac336 6356 sectors = 0;
159ec1fc 6357 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6358 char b[BDEVNAME_SIZE];
6359 seq_printf(seq, " %s[%d]",
6360 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6361 if (test_bit(WriteMostly, &rdev->flags))
6362 seq_printf(seq, "(W)");
b2d444d7 6363 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6364 seq_printf(seq, "(F)");
6365 continue;
b325a32e
N		 6366 } else if (rdev->raid_disk < 0)
6367 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6368 sectors += rdev->sectors;
1da177e4
LT		 6369 }
6370
6371 if (!list_empty(&mddev->disks)) {
6372 if (mddev->pers)
6373 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6374 (unsigned long long)
6375 mddev->array_sectors / 2);
1da177e4
LT		 6376 else
6377 seq_printf(seq, "\n %llu blocks",
dd8ac336 6378 (unsigned long long)sectors / 2);
1da177e4 6379 }
1cd6bf19
N		 6380 if (mddev->persistent) {
6381 if (mddev->major_version != 0 ||
6382 mddev->minor_version != 90) {
6383 seq_printf(seq," super %d.%d",
6384 mddev->major_version,
6385 mddev->minor_version);
6386 }
e691063a
N		 6387 } else if (mddev->external)
6388 seq_printf(seq, " super external:%s",
6389 mddev->metadata_type);
6390 else
1cd6bf19 6391 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6392
6393 if (mddev->pers) {
d710e138 6394 mddev->pers->status(seq, mddev);
1da177e4 6395 seq_printf(seq, "\n ");
8e1b39d6
N		 6396 if (mddev->pers->sync_request) {
6397 if (mddev->curr_resync > 2) {
d710e138 6398 status_resync(seq, mddev);
8e1b39d6
N		 6399 seq_printf(seq, "\n ");
6400 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6401 seq_printf(seq, "\tresync=DELAYED\n ");
6402 else if (mddev->recovery_cp < MaxSector)
6403 seq_printf(seq, "\tresync=PENDING\n ");
6404 }
32a7627c
N		 6405 } else
6406 seq_printf(seq, "\n ");
6407
6408 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6409 unsigned long chunk_kb;
6410 unsigned long flags;
32a7627c 6411 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6412 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6413 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6414 "%lu%s chunk",
6415 bitmap->pages - bitmap->missing_pages,
6416 bitmap->pages,
6417 (bitmap->pages - bitmap->missing_pages)
6418 << (PAGE_SHIFT - 10),
42a04b50 6419 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6420 chunk_kb ? "KB" : "B");
78d742d8
N		 6421 if (bitmap->file) {
6422 seq_printf(seq, ", file: ");
c32c2f63 6423 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6424 }
78d742d8 6425
32a7627c
N		 6426 seq_printf(seq, "\n");
6427 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6428 }
6429
6430 seq_printf(seq, "\n");
6431 }
6432 mddev_unlock(mddev);
6433
6434 return 0;
6435}
6436
110518bc 6437static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6438 .start = md_seq_start,
6439 .next = md_seq_next,
6440 .stop = md_seq_stop,
6441 .show = md_seq_show,
6442};
6443
6444static int md_seq_open(struct inode *inode, struct file *file)
6445{
6446 int error;
d7603b7e
N		 6447 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6448 if (mi == NULL)
6449 return -ENOMEM;
1da177e4
LT		 6450
6451 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 6452 if (error)
6453 kfree(mi);
6454 else {
6455 struct seq_file *p = file->private_data;
6456 p->private = mi;
6457 mi->event = atomic_read(&md_event_count);
6458 }
1da177e4
LT		 6459 return error;
6460}
6461
d7603b7e
N		 6462static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6463{
6464 struct seq_file *m = filp->private_data;
6465 struct mdstat_info *mi = m->private;
6466 int mask;
6467
6468 poll_wait(filp, &md_event_waiters, wait);
6469
6470 /* always allow read */
6471 mask = POLLIN | POLLRDNORM;
6472
6473 if (mi->event != atomic_read(&md_event_count))
6474 mask |= POLLERR | POLLPRI;
6475 return mask;
6476}
6477
fa027c2a 6478static const struct file_operations md_seq_fops = {
e24650c2 6479 .owner = THIS_MODULE,
1da177e4
LT		 6480 .open = md_seq_open,
6481 .read = seq_read,
6482 .llseek = seq_lseek,
c3f94b40 6483 .release = seq_release_private,
d7603b7e 6484 .poll = mdstat_poll,
1da177e4
LT		 6485};
6486
2604b703 6487int register_md_personality(struct mdk_personality *p)
1da177e4 6488{
1da177e4 6489 spin_lock(&pers_lock);
2604b703
N		 6490 list_add_tail(&p->list, &pers_list);
6491 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6492 spin_unlock(&pers_lock);
6493 return 0;
6494}
6495
2604b703 6496int unregister_md_personality(struct mdk_personality *p)
1da177e4 6497{
2604b703 6498 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6499 spin_lock(&pers_lock);
2604b703 6500 list_del_init(&p->list);
1da177e4
LT		 6501 spin_unlock(&pers_lock);
6502 return 0;
6503}
6504
eea1bf38 6505static int is_mddev_idle(mddev_t *mddev, int init)
1da177e4
LT		 6506{
6507 mdk_rdev_t * rdev;
1da177e4 6508 int idle;
eea1bf38 6509 int curr_events;
1da177e4
LT		 6510
6511 idle = 1;
4b80991c
N		 6512 rcu_read_lock();
6513 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6514 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6515 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6516 (int)part_stat_read(&disk->part0, sectors[1]) -
6517 atomic_read(&disk->sync_io);
713f6ab1
N		 6518 /* sync IO will cause sync_io to increase before the disk_stats
6519 * as sync_io is counted when a request starts, and
6520 * disk_stats is counted when it completes.
6521 * So resync activity will cause curr_events to be smaller than
6522 * when there was no such activity.
6523 * non-sync IO will cause disk_stat to increase without
6524 * increasing sync_io so curr_events will (eventually)
6525 * be larger than it was before. Once it becomes
6526 * substantially larger, the test below will cause
6527 * the array to appear non-idle, and resync will slow
6528 * down.
6529 * If there is a lot of outstanding resync activity when
6530 * we set last_event to curr_events, then all that activity
6531 * completing might cause the array to appear non-idle
6532 * and resync will be slowed down even though there might
6533 * not have been non-resync activity. This will only
6534 * happen once though. 'last_events' will soon reflect
6535 * the state where there is little or no outstanding
6536 * resync requests, and further resync activity will
6537 * always make curr_events less than last_events.
c0e48521 6538 *
1da177e4 6539 */
eea1bf38 6540 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6541 rdev->last_events = curr_events;
6542 idle = 0;
6543 }
6544 }
4b80991c 6545 rcu_read_unlock();
1da177e4
LT		 6546 return idle;
6547}
6548
6549void md_done_sync(mddev_t *mddev, int blocks, int ok)
6550{
6551 /* another "blocks" (512byte) blocks have been synced */
6552 atomic_sub(blocks, &mddev->recovery_active);
6553 wake_up(&mddev->recovery_wait);
6554 if (!ok) {
dfc70645 6555 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6556 md_wakeup_thread(mddev->thread);
6557 // stop recovery, signal do_sync ....
6558 }
6559}
6560
6561
06d91a5f
N		 6562/* md_write_start(mddev, bi)
6563 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6564 * in superblock) before writing, schedule a superblock update
6565 * and wait for it to complete.
06d91a5f 6566 */
3d310eb7 6567void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 6568{
0fd62b86 6569 int did_change = 0;
06d91a5f 6570 if (bio_data_dir(bi) != WRITE)
3d310eb7 6571 return;
06d91a5f 6572
f91de92e
N		 6573 BUG_ON(mddev->ro == 1);
6574 if (mddev->ro == 2) {
6575 /* need to switch to read/write */
6576 mddev->ro = 0;
6577 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6578 md_wakeup_thread(mddev->thread);
25156198 6579 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6580 did_change = 1;
f91de92e 6581 }
06d91a5f 6582 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6583 if (mddev->safemode == 1)
6584 mddev->safemode = 0;
06d91a5f 6585 if (mddev->in_sync) {
a9701a30 6586 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6587 if (mddev->in_sync) {
6588 mddev->in_sync = 0;
850b2b42 6589 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6590 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6591 md_wakeup_thread(mddev->thread);
0fd62b86 6592 did_change = 1;
3d310eb7 6593 }
a9701a30 6594 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6595 }
0fd62b86 6596 if (did_change)
00bcb4ac 6597 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6598 wait_event(mddev->sb_wait,
09a44cc1 6599 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6600}
6601
6602void md_write_end(mddev_t *mddev)
6603{
6604 if (atomic_dec_and_test(&mddev->writes_pending)) {
6605 if (mddev->safemode == 2)
6606 md_wakeup_thread(mddev->thread);
16f17b39 6607 else if (mddev->safemode_delay)
1da177e4
LT		 6608 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6609 }
6610}
6611
2a2275d6
N		 6612/* md_allow_write(mddev)
6613 * Calling this ensures that the array is marked 'active' so that writes
6614 * may proceed without blocking. It is important to call this before
6615 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6616 * Must be called with mddev_lock held.
b5470dc5
DW		 6617 *
6618 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6619 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6620 */
b5470dc5 6621int md_allow_write(mddev_t *mddev)
2a2275d6
N		 6622{
6623 if (!mddev->pers)
b5470dc5 6624 return 0;
2a2275d6 6625 if (mddev->ro)
b5470dc5 6626 return 0;
1a0fd497 6627 if (!mddev->pers->sync_request)
b5470dc5 6628 return 0;
2a2275d6
N		 6629
6630 spin_lock_irq(&mddev->write_lock);
6631 if (mddev->in_sync) {
6632 mddev->in_sync = 0;
6633 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6634 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6635 if (mddev->safemode_delay &&
6636 mddev->safemode == 0)
6637 mddev->safemode = 1;
6638 spin_unlock_irq(&mddev->write_lock);
6639 md_update_sb(mddev, 0);
00bcb4ac 6640 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6641 } else
6642 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6643
070dc6dd 6644 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6645 return -EAGAIN;
6646 else
6647 return 0;
2a2275d6
N		 6648}
6649EXPORT_SYMBOL_GPL(md_allow_write);
6650
b63d7c2e 6651void md_unplug(mddev_t *mddev)
252ac522
N		 6652{
6653 if (mddev->queue)
6654 blk_unplug(mddev->queue);
6655 if (mddev->plug)
6656 mddev->plug->unplug_fn(mddev->plug);
6657}
6658
1da177e4
LT		 6659#define SYNC_MARKS 10
6660#define SYNC_MARK_STEP (3*HZ)
29269553 6661void md_do_sync(mddev_t *mddev)
1da177e4
LT		 6662{
6663 mddev_t *mddev2;
6664 unsigned int currspeed = 0,
6665 window;
57afd89f 6666 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 6667 unsigned long mark[SYNC_MARKS];
6668 sector_t mark_cnt[SYNC_MARKS];
6669 int last_mark,m;
6670 struct list_head *tmp;
6671 sector_t last_check;
57afd89f 6672 int skipped = 0;
5fd6c1dc 6673 mdk_rdev_t *rdev;
61df9d91 6674 char *desc;
1da177e4
LT		 6675
6676 /* just incase thread restarts... */
6677 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6678 return;
5fd6c1dc
N		 6679 if (mddev->ro) /* never try to sync a read-only array */
6680 return;
1da177e4 6681
61df9d91
N		 6682 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6683 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6684 desc = "data-check";
6685 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6686 desc = "requested-resync";
6687 else
6688 desc = "resync";
6689 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6690 desc = "reshape";
6691 else
6692 desc = "recovery";
6693
1da177e4
LT		 6694 /* we overload curr_resync somewhat here.
6695 * 0 == not engaged in resync at all
6696 * 2 == checking that there is no conflict with another sync
6697 * 1 == like 2, but have yielded to allow conflicting resync to
6698 * commense
6699 * other == active in resync - this many blocks
6700 *
6701 * Before starting a resync we must have set curr_resync to
6702 * 2, and then checked that every "conflicting" array has curr_resync
6703 * less than ours. When we find one that is the same or higher
6704 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6705 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6706 * This will mean we have to start checking from the beginning again.
6707 *
6708 */
6709
6710 do {
6711 mddev->curr_resync = 2;
6712
6713 try_again:
404e4b43 6714 if (kthread_should_stop())
6985c43f 6715 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 6716
6717 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 6718 goto skip;
29ac4aa3 6719 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 6720 if (mddev2 == mddev)
6721 continue;
90b08710
BS		 6722 if (!mddev->parallel_resync
6723 && mddev2->curr_resync
6724 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 6725 DEFINE_WAIT(wq);
6726 if (mddev < mddev2 && mddev->curr_resync == 2) {
6727 /* arbitrarily yield */
6728 mddev->curr_resync = 1;
6729 wake_up(&resync_wait);
6730 }
6731 if (mddev > mddev2 && mddev->curr_resync == 1)
6732 /* no need to wait here, we can wait the next
6733 * time 'round when curr_resync == 2
6734 */
6735 continue;
9744197c
N		 6736 /* We need to wait 'interruptible' so as not to
6737 * contribute to the load average, and not to
6738 * be caught by 'softlockup'
6739 */
6740 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 6741 if (!kthread_should_stop() &&
8712e553 6742 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 6743 printk(KERN_INFO "md: delaying %s of %s"
6744 " until %s has finished (they"
1da177e4 6745 " share one or more physical units)\n",
61df9d91 6746 desc, mdname(mddev), mdname(mddev2));
1da177e4 6747 mddev_put(mddev2);
9744197c
N		 6748 if (signal_pending(current))
6749 flush_signals(current);
1da177e4
LT		 6750 schedule();
6751 finish_wait(&resync_wait, &wq);
6752 goto try_again;
6753 }
6754 finish_wait(&resync_wait, &wq);
6755 }
6756 }
6757 } while (mddev->curr_resync < 2);
6758
5fd6c1dc 6759 j = 0;
9d88883e 6760 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 6761 /* resync follows the size requested by the personality,
57afd89f 6762 * which defaults to physical size, but can be virtual size
1da177e4
LT		 6763 */
6764 max_sectors = mddev->resync_max_sectors;
9d88883e 6765 mddev->resync_mismatches = 0;
5fd6c1dc 6766 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 6767 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6768 j = mddev->resync_min;
6769 else if (!mddev->bitmap)
5fd6c1dc 6770 j = mddev->recovery_cp;
5e96ee65 6771
ccfcc3c1 6772 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 6773 max_sectors = mddev->dev_sectors;
5fd6c1dc 6774 else {
1da177e4 6775 /* recovery follows the physical size of devices */
58c0fed4 6776 max_sectors = mddev->dev_sectors;
5fd6c1dc 6777 j = MaxSector;
4e59ca7d
DW		 6778 rcu_read_lock();
6779 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 6780 if (rdev->raid_disk >= 0 &&
6781 !test_bit(Faulty, &rdev->flags) &&
6782 !test_bit(In_sync, &rdev->flags) &&
6783 rdev->recovery_offset < j)
6784 j = rdev->recovery_offset;
4e59ca7d 6785 rcu_read_unlock();
5fd6c1dc 6786 }
1da177e4 6787
61df9d91
N		 6788 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6789 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6790 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 6791 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 6792 "(but not more than %d KB/sec) for %s.\n",
6793 speed_max(mddev), desc);
1da177e4 6794
eea1bf38 6795 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 6796
57afd89f 6797 io_sectors = 0;
1da177e4
LT		 6798 for (m = 0; m < SYNC_MARKS; m++) {
6799 mark[m] = jiffies;
57afd89f 6800 mark_cnt[m] = io_sectors;
1da177e4
LT		 6801 }
6802 last_mark = 0;
6803 mddev->resync_mark = mark[last_mark];
6804 mddev->resync_mark_cnt = mark_cnt[last_mark];
6805
6806 /*
6807 * Tune reconstruction:
6808 */
6809 window = 32*(PAGE_SIZE/512);
6810 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6811 window/2,(unsigned long long) max_sectors/2);
6812
6813 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 6814 last_check = 0;
6815
6816 if (j>2) {
6817 printk(KERN_INFO
61df9d91
N		 6818 "md: resuming %s of %s from checkpoint.\n",
6819 desc, mdname(mddev));
1da177e4
LT		 6820 mddev->curr_resync = j;
6821 }
efa59339 6822 mddev->curr_resync_completed = mddev->curr_resync;
1da177e4
LT		 6823
6824 while (j < max_sectors) {
57afd89f 6825 sector_t sectors;
1da177e4 6826
57afd89f 6827 skipped = 0;
97e4f42d 6828
7a91ee1f
N		 6829 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6830 ((mddev->curr_resync > mddev->curr_resync_completed &&
6831 (mddev->curr_resync - mddev->curr_resync_completed)
6832 > (max_sectors >> 4)) ||
6833 (j - mddev->curr_resync_completed)*2
6834 >= mddev->resync_max - mddev->curr_resync_completed
6835 )) {
97e4f42d 6836 /* time to update curr_resync_completed */
252ac522 6837 md_unplug(mddev);
97e4f42d
N		 6838 wait_event(mddev->recovery_wait,
6839 atomic_read(&mddev->recovery_active) == 0);
6840 mddev->curr_resync_completed =
6841 mddev->curr_resync;
070dc6dd 6842 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 6843 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 6844 }
acb180b0 6845
e62e58a5
N		 6846 while (j >= mddev->resync_max && !kthread_should_stop()) {
6847 /* As this condition is controlled by user-space,
6848 * we can block indefinitely, so use '_interruptible'
6849 * to avoid triggering warnings.
6850 */
6851 flush_signals(current); /* just in case */
6852 wait_event_interruptible(mddev->recovery_wait,
6853 mddev->resync_max > j
6854 || kthread_should_stop());
6855 }
acb180b0
N		 6856
6857 if (kthread_should_stop())
6858 goto interrupted;
6859
57afd89f 6860 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 6861 currspeed < speed_min(mddev));
57afd89f 6862 if (sectors == 0) {
dfc70645 6863 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6864 goto out;
6865 }
57afd89f
N		 6866
6867 if (!skipped) { /* actual IO requested */
6868 io_sectors += sectors;
6869 atomic_add(sectors, &mddev->recovery_active);
6870 }
6871
1da177e4
LT		 6872 j += sectors;
6873 if (j>1) mddev->curr_resync = j;
ff4e8d9a 6874 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 6875 if (last_check == 0)
6876 /* this is the earliers that rebuilt will be
6877 * visible in /proc/mdstat
6878 */
6879 md_new_event(mddev);
57afd89f
N		 6880
6881 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 6882 continue;
6883
57afd89f 6884 last_check = io_sectors;
1da177e4 6885
dfc70645 6886 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4
LT		 6887 break;
6888
6889 repeat:
6890 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6891 /* step marks */
6892 int next = (last_mark+1) % SYNC_MARKS;
6893
6894 mddev->resync_mark = mark[next];
6895 mddev->resync_mark_cnt = mark_cnt[next];
6896 mark[next] = jiffies;
57afd89f 6897 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6898 last_mark = next;
6899 }
6900
6901
c6207277
N		 6902 if (kthread_should_stop())
6903 goto interrupted;
6904
1da177e4
LT		 6905
6906 /*
6907 * this loop exits only if either when we are slower than
6908 * the 'hard' speed limit, or the system was IO-idle for
6909 * a jiffy.
6910 * the system might be non-idle CPU-wise, but we only care
6911 * about not overloading the IO subsystem. (things like an
6912 * e2fsck being done on the RAID array should execute fast)
6913 */
252ac522 6914 md_unplug(mddev);
1da177e4
LT		 6915 cond_resched();
6916
57afd89f
N		 6917 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6918 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 6919
88202a0c
N		 6920 if (currspeed > speed_min(mddev)) {
6921 if ((currspeed > speed_max(mddev)) ||
eea1bf38 6922 !is_mddev_idle(mddev, 0)) {
c0e48521 6923 msleep(500);
1da177e4
LT		 6924 goto repeat;
6925 }
6926 }
6927 }
61df9d91 6928 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 6929 /*
6930 * this also signals 'finished resyncing' to md_stop
6931 */
6932 out:
252ac522 6933 md_unplug(mddev);
1da177e4
LT		 6934
6935 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6936
6937 /* tell personality that we are finished */
57afd89f 6938 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 6939
dfc70645 6940 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 6941 mddev->curr_resync > 2) {
6942 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6943 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6944 if (mddev->curr_resync >= mddev->recovery_cp) {
6945 printk(KERN_INFO
61df9d91
N		 6946 "md: checkpointing %s of %s.\n",
6947 desc, mdname(mddev));
5fd6c1dc
N		 6948 mddev->recovery_cp = mddev->curr_resync;
6949 }
6950 } else
6951 mddev->recovery_cp = MaxSector;
6952 } else {
6953 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6954 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 6955 rcu_read_lock();
6956 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 6957 if (rdev->raid_disk >= 0 &&
70fffd0b 6958 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 6959 !test_bit(Faulty, &rdev->flags) &&
6960 !test_bit(In_sync, &rdev->flags) &&
6961 rdev->recovery_offset < mddev->curr_resync)
6962 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 6963 rcu_read_unlock();
5fd6c1dc 6964 }
1da177e4 6965 }
17571284 6966 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6967
1da177e4 6968 skip:
c07b70ad
N		 6969 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6970 /* We completed so min/max setting can be forgotten if used. */
6971 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6972 mddev->resync_min = 0;
6973 mddev->resync_max = MaxSector;
6974 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6975 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 6976 mddev->curr_resync = 0;
efa59339
N		 6977 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6978 mddev->curr_resync_completed = 0;
c6207277 6979 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 6980 wake_up(&resync_wait);
6981 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6982 md_wakeup_thread(mddev->thread);
c6207277
N		 6983 return;
6984
6985 interrupted:
6986 /*
6987 * got a signal, exit.
6988 */
6989 printk(KERN_INFO
6990 "md: md_do_sync() got signal ... exiting\n");
6991 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6992 goto out;
6993
1da177e4 6994}
29269553 6995EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 6996
6997
b4c4c7b8
N		 6998static int remove_and_add_spares(mddev_t *mddev)
6999{
7000 mdk_rdev_t *rdev;
b4c4c7b8
N		 7001 int spares = 0;
7002
97e4f42d
N		 7003 mddev->curr_resync_completed = 0;
7004
159ec1fc 7005 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7006 if (rdev->raid_disk >= 0 &&
6bfe0b49 7007 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7008 (test_bit(Faulty, &rdev->flags) ||
7009 ! test_bit(In_sync, &rdev->flags)) &&
7010 atomic_read(&rdev->nr_pending)==0) {
7011 if (mddev->pers->hot_remove_disk(
7012 mddev, rdev->raid_disk)==0) {
7013 char nm[20];
7014 sprintf(nm,"rd%d", rdev->raid_disk);
7015 sysfs_remove_link(&mddev->kobj, nm);
7016 rdev->raid_disk = -1;
7017 }
7018 }
7019
4044ba58 7020 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
159ec1fc 7021 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7022 if (rdev->raid_disk >= 0 &&
e5427135
DW		 7023 !test_bit(In_sync, &rdev->flags) &&
7024 !test_bit(Blocked, &rdev->flags))
dfc70645 7025 spares++;
b4c4c7b8
N		 7026 if (rdev->raid_disk < 0
7027 && !test_bit(Faulty, &rdev->flags)) {
7028 rdev->recovery_offset = 0;
199050ea
NB		 7029 if (mddev->pers->
7030 hot_add_disk(mddev, rdev) == 0) {
b4c4c7b8
N		 7031 char nm[20];
7032 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 7033 if (sysfs_create_link(&mddev->kobj,
7034 &rdev->kobj, nm))
00bcb4ac 7035 /* failure here is OK */;
b4c4c7b8
N		 7036 spares++;
7037 md_new_event(mddev);
93be75ff 7038 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7039 } else
7040 break;
7041 }
dfc70645 7042 }
b4c4c7b8
N		 7043 }
7044 return spares;
7045}
7ebc0be7
N		 7046
7047static void reap_sync_thread(mddev_t *mddev)
7048{
7049 mdk_rdev_t *rdev;
7050
7051 /* resync has finished, collect result */
7052 md_unregister_thread(mddev->sync_thread);
7053 mddev->sync_thread = NULL;
7054 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7055 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7056 /* success...*/
7057 /* activate any spares */
7058 if (mddev->pers->spare_active(mddev))
7059 sysfs_notify(&mddev->kobj, NULL,
7060 "degraded");
7061 }
7062 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7063 mddev->pers->finish_reshape)
7064 mddev->pers->finish_reshape(mddev);
7065 md_update_sb(mddev, 1);
7066
7067 /* if array is no-longer degraded, then any saved_raid_disk
7068 * information must be scrapped
7069 */
7070 if (!mddev->degraded)
7071 list_for_each_entry(rdev, &mddev->disks, same_set)
7072 rdev->saved_raid_disk = -1;
7073
7074 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7075 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7076 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7077 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7078 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7079 /* flag recovery needed just to double check */
7080 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7081 sysfs_notify_dirent_safe(mddev->sysfs_action);
7082 md_new_event(mddev);
7083}
7084
1da177e4
LT		 7085/*
7086 * This routine is regularly called by all per-raid-array threads to
7087 * deal with generic issues like resync and super-block update.
7088 * Raid personalities that don't have a thread (linear/raid0) do not
7089 * need this as they never do any recovery or update the superblock.
7090 *
7091 * It does not do any resync itself, but rather "forks" off other threads
7092 * to do that as needed.
7093 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7094 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7095 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7096 * and wakeups up this thread which will reap the thread and finish up.
7097 * This thread also removes any faulty devices (with nr_pending == 0).
7098 *
7099 * The overall approach is:
7100 * 1/ if the superblock needs updating, update it.
7101 * 2/ If a recovery thread is running, don't do anything else.
7102 * 3/ If recovery has finished, clean up, possibly marking spares active.
7103 * 4/ If there are any faulty devices, remove them.
7104 * 5/ If array is degraded, try to add spares devices
7105 * 6/ If array has spares or is not in-sync, start a resync thread.
7106 */
7107void md_check_recovery(mddev_t *mddev)
7108{
5f40402d 7109 if (mddev->bitmap)
aa5cbd10 7110 bitmap_daemon_work(mddev);
1da177e4
LT		 7111
7112 if (mddev->ro)
7113 return;
fca4d848
N		 7114
7115 if (signal_pending(current)) {
31a59e34 7116 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7117 printk(KERN_INFO "md: %s in immediate safe mode\n",
7118 mdname(mddev));
7119 mddev->safemode = 2;
7120 }
7121 flush_signals(current);
7122 }
7123
c89a8eee
N		 7124 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7125 return;
1da177e4 7126 if ( ! (
126925c0 7127 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7128 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7129 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7130 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7131 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7132 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7133 ))
7134 return;
fca4d848 7135
df5b89b3 7136 if (mddev_trylock(mddev)) {
b4c4c7b8 7137 int spares = 0;
fca4d848 7138
c89a8eee
N		 7139 if (mddev->ro) {
7140 /* Only thing we do on a ro array is remove
7141 * failed devices.
7142 */
7143 remove_and_add_spares(mddev);
7144 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7145 goto unlock;
7146 }
7147
31a59e34 7148 if (!mddev->external) {
0fd62b86 7149 int did_change = 0;
31a59e34
N		 7150 spin_lock_irq(&mddev->write_lock);
7151 if (mddev->safemode &&
7152 !atomic_read(&mddev->writes_pending) &&
7153 !mddev->in_sync &&
7154 mddev->recovery_cp == MaxSector) {
7155 mddev->in_sync = 1;
0fd62b86 7156 did_change = 1;
070dc6dd 7157 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7158 }
7159 if (mddev->safemode == 1)
7160 mddev->safemode = 0;
7161 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7162 if (did_change)
00bcb4ac 7163 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7164 }
fca4d848 7165
850b2b42
N		 7166 if (mddev->flags)
7167 md_update_sb(mddev, 0);
06d91a5f 7168
1da177e4
LT		 7169 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7170 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7171 /* resync/recovery still happening */
7172 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7173 goto unlock;
7174 }
7175 if (mddev->sync_thread) {
7ebc0be7 7176 reap_sync_thread(mddev);
1da177e4
LT		 7177 goto unlock;
7178 }
72a23c21
NB		 7179 /* Set RUNNING before clearing NEEDED to avoid
7180 * any transients in the value of "sync_action".
7181 */
7182 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7183 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7184 /* Clear some bits that don't mean anything, but
7185 * might be left set
7186 */
24dd469d
N		 7187 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7188 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7189
5fd6c1dc
N		 7190 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7191 goto unlock;
1da177e4
LT		 7192 /* no recovery is running.
7193 * remove any failed drives, then
7194 * add spares if possible.
7195 * Spare are also removed and re-added, to allow
7196 * the personality to fail the re-add.
7197 */
1da177e4 7198
b4c4c7b8 7199 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7200 if (mddev->pers->check_reshape == NULL ||
7201 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7202 /* Cannot proceed */
7203 goto unlock;
7204 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7205 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7206 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7207 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7208 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7209 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7210 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7211 } else if (mddev->recovery_cp < MaxSector) {
7212 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7213 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7214 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7215 /* nothing to be done ... */
1da177e4 7216 goto unlock;
24dd469d 7217
1da177e4 7218 if (mddev->pers->sync_request) {
a654b9d8
N		 7219 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7220 /* We are adding a device or devices to an array
7221 * which has the bitmap stored on all devices.
7222 * So make sure all bitmap pages get written
7223 */
7224 bitmap_write_all(mddev->bitmap);
7225 }
1da177e4
LT		 7226 mddev->sync_thread = md_register_thread(md_do_sync,
7227 mddev,
0da3c619 7228 "resync");
1da177e4
LT		 7229 if (!mddev->sync_thread) {
7230 printk(KERN_ERR "%s: could not start resync"
7231 " thread...\n",
7232 mdname(mddev));
7233 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7234 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7235 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7236 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7237 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7238 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7239 } else
1da177e4 7240 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7241 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7242 md_new_event(mddev);
1da177e4
LT		 7243 }
7244 unlock:
72a23c21
NB		 7245 if (!mddev->sync_thread) {
7246 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7247 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7248 &mddev->recovery))
0c3573f1 7249 if (mddev->sysfs_action)
00bcb4ac 7250 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7251 }
1da177e4
LT		 7252 mddev_unlock(mddev);
7253 }
7254}
7255
6bfe0b49
DW		 7256void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7257{
00bcb4ac 7258 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49
DW		 7259 wait_event_timeout(rdev->blocked_wait,
7260 !test_bit(Blocked, &rdev->flags),
7261 msecs_to_jiffies(5000));
7262 rdev_dec_pending(rdev, mddev);
7263}
7264EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7265
75c96f85
AB		 7266static int md_notify_reboot(struct notifier_block *this,
7267 unsigned long code, void *x)
1da177e4
LT		 7268{
7269 struct list_head *tmp;
7270 mddev_t *mddev;
7271
7272 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7273
7274 printk(KERN_INFO "md: stopping all md devices.\n");
7275
29ac4aa3 7276 for_each_mddev(mddev, tmp)
c71d4887 7277 if (mddev_trylock(mddev)) {
2b25000b
N		 7278 /* Force a switch to readonly even array
7279 * appears to still be in use. Hence
7280 * the '100'.
7281 */
a4bd82d0 7282 md_set_readonly(mddev, 100);
c71d4887
NB		 7283 mddev_unlock(mddev);
7284 }
1da177e4
LT		 7285 /*
7286 * certain more exotic SCSI devices are known to be
7287 * volatile wrt too early system reboots. While the
7288 * right place to handle this issue is the given
7289 * driver, we do want to have a safe RAID driver ...
7290 */
7291 mdelay(1000*1);
7292 }
7293 return NOTIFY_DONE;
7294}
7295
75c96f85 7296static struct notifier_block md_notifier = {
1da177e4
LT		 7297 .notifier_call = md_notify_reboot,
7298 .next = NULL,
7299 .priority = INT_MAX, /* before any real devices */
7300};
7301
7302static void md_geninit(void)
7303{
1da177e4
LT		 7304 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7305
c7705f34 7306 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 7307}
7308
75c96f85 7309static int __init md_init(void)
1da177e4 7310{
e804ac78
TH		 7311 int ret = -ENOMEM;
7312
7313 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7314 if (!md_wq)
7315 goto err_wq;
7316
7317 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7318 if (!md_misc_wq)
7319 goto err_misc_wq;
7320
7321 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7322 goto err_md;
7323
7324 if ((ret = register_blkdev(0, "mdp")) < 0)
7325 goto err_mdp;
7326 mdp_major = ret;
7327
3dbd8c2e 7328 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 7329 md_probe, NULL, NULL);
7330 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 7331 md_probe, NULL, NULL);
7332
1da177e4 7333 register_reboot_notifier(&md_notifier);
0b4d4147 7334 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 7335
7336 md_geninit();
d710e138 7337 return 0;
1da177e4 7338
e804ac78
TH		 7339err_mdp:
7340 unregister_blkdev(MD_MAJOR, "md");
7341err_md:
7342 destroy_workqueue(md_misc_wq);
7343err_misc_wq:
7344 destroy_workqueue(md_wq);
7345err_wq:
7346 return ret;
7347}
1da177e4
LT		 7348
7349#ifndef MODULE
7350
7351/*
7352 * Searches all registered partitions for autorun RAID arrays
7353 * at boot time.
7354 */
4d936ec1
ME		 7355
7356static LIST_HEAD(all_detected_devices);
7357struct detected_devices_node {
7358 struct list_head list;
7359 dev_t dev;
7360};
1da177e4
LT		 7361
7362void md_autodetect_dev(dev_t dev)
7363{
4d936ec1
ME		 7364 struct detected_devices_node *node_detected_dev;
7365
7366 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7367 if (node_detected_dev) {
7368 node_detected_dev->dev = dev;
7369 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7370 } else {
7371 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7372 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7373 }
1da177e4
LT		 7374}
7375
7376
7377static void autostart_arrays(int part)
7378{
7379 mdk_rdev_t *rdev;
4d936ec1
ME		 7380 struct detected_devices_node *node_detected_dev;
7381 dev_t dev;
7382 int i_scanned, i_passed;
1da177e4 7383
4d936ec1
ME		 7384 i_scanned = 0;
7385 i_passed = 0;
1da177e4 7386
4d936ec1 7387 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 7388
4d936ec1
ME		 7389 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7390 i_scanned++;
7391 node_detected_dev = list_entry(all_detected_devices.next,
7392 struct detected_devices_node, list);
7393 list_del(&node_detected_dev->list);
7394 dev = node_detected_dev->dev;
7395 kfree(node_detected_dev);
df968c4e 7396 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 7397 if (IS_ERR(rdev))
7398 continue;
7399
b2d444d7 7400 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7401 MD_BUG();
7402 continue;
7403 }
d0fae18f 7404 set_bit(AutoDetected, &rdev->flags);
1da177e4 7405 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 7406 i_passed++;
1da177e4 7407 }
4d936ec1
ME		 7408
7409 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7410 i_scanned, i_passed);
1da177e4
LT		 7411
7412 autorun_devices(part);
7413}
7414
fdee8ae4 7415#endif /* !MODULE */
1da177e4
LT		 7416
7417static __exit void md_exit(void)
7418{
7419 mddev_t *mddev;
7420 struct list_head *tmp;
8ab5e4c1 7421
3dbd8c2e 7422 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 7423 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 7424
3dbd8c2e 7425 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 7426 unregister_blkdev(mdp_major, "mdp");
7427 unregister_reboot_notifier(&md_notifier);
7428 unregister_sysctl_table(raid_table_header);
7429 remove_proc_entry("mdstat", NULL);
29ac4aa3 7430 for_each_mddev(mddev, tmp) {
1da177e4 7431 export_array(mddev);
d3374825 7432 mddev->hold_active = 0;
1da177e4 7433 }
e804ac78
TH		 7434 destroy_workqueue(md_misc_wq);
7435 destroy_workqueue(md_wq);
1da177e4
LT		 7436}
7437
685784aa 7438subsys_initcall(md_init);
1da177e4
LT		 7439module_exit(md_exit)
7440
f91de92e
N		 7441static int get_ro(char *buffer, struct kernel_param *kp)
7442{
7443 return sprintf(buffer, "%d", start_readonly);
7444}
7445static int set_ro(const char *val, struct kernel_param *kp)
7446{
7447 char *e;
7448 int num = simple_strtoul(val, &e, 10);
7449 if (*val && (*e == '\0' || *e == '\n')) {
7450 start_readonly = num;
4dbcdc75 7451 return 0;
f91de92e
N		 7452 }
7453 return -EINVAL;
7454}
7455
80ca3a44
N		 7456module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7457module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 7458
efeb53c0 7459module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 7460
1da177e4
LT		 7461EXPORT_SYMBOL(register_md_personality);
7462EXPORT_SYMBOL(unregister_md_personality);
7463EXPORT_SYMBOL(md_error);
7464EXPORT_SYMBOL(md_done_sync);
7465EXPORT_SYMBOL(md_write_start);
7466EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 7467EXPORT_SYMBOL(md_register_thread);
7468EXPORT_SYMBOL(md_unregister_thread);
7469EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 7470EXPORT_SYMBOL(md_check_recovery);
7471MODULE_LICENSE("GPL");
0efb9e61 7472MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 7473MODULE_ALIAS("md");
72008652 7474MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)