projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: range check slot number when manually adding a spare.
[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 */
57b2caa3 708static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
1da177e4 709{
57b2caa3 710 sector_t num_sectors = i_size_read(rdev->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 767
a6ff7e08 768 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 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 797int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
ccebd4c4 798 struct page *page, int rw, bool metadata_op)
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
a6ff7e08
JB		 806 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
807 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 808 if (metadata_op)
809 bio->bi_sector = sector + rdev->sb_start;
810 else
811 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 812 bio_add_page(bio, page, size, 0);
813 init_completion(&event);
814 bio->bi_private = &event;
815 bio->bi_end_io = bi_complete;
816 submit_bio(rw, bio);
817 wait_for_completion(&event);
818
819 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
820 bio_put(bio);
821 return ret;
822}
a8745db2 823EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 824
0002b271 825static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 826{
827 char b[BDEVNAME_SIZE];
828 if (!rdev->sb_page) {
829 MD_BUG();
830 return -EINVAL;
831 }
832 if (rdev->sb_loaded)
833 return 0;
834
835
ccebd4c4 836 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 837 goto fail;
838 rdev->sb_loaded = 1;
839 return 0;
840
841fail:
842 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
843 bdevname(rdev->bdev,b));
844 return -EINVAL;
845}
846
847static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
848{
05710466
AN		 849 return sb1->set_uuid0 == sb2->set_uuid0 &&
850 sb1->set_uuid1 == sb2->set_uuid1 &&
851 sb1->set_uuid2 == sb2->set_uuid2 &&
852 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 853}
854
1da177e4
LT		 855static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
856{
857 int ret;
858 mdp_super_t *tmp1, *tmp2;
859
860 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
861 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
862
863 if (!tmp1 || !tmp2) {
864 ret = 0;
35020f1a 865 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 866 goto abort;
867 }
868
869 *tmp1 = *sb1;
870 *tmp2 = *sb2;
871
872 /*
873 * nr_disks is not constant
874 */
875 tmp1->nr_disks = 0;
876 tmp2->nr_disks = 0;
877
ce0c8e05 878 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 879abort:
990a8baf
JJ		 880 kfree(tmp1);
881 kfree(tmp2);
1da177e4
LT		 882 return ret;
883}
884
4d167f09
N		 885
886static u32 md_csum_fold(u32 csum)
887{
888 csum = (csum & 0xffff) + (csum >> 16);
889 return (csum & 0xffff) + (csum >> 16);
890}
891
1da177e4
LT		 892static unsigned int calc_sb_csum(mdp_super_t * sb)
893{
4d167f09
N		 894 u64 newcsum = 0;
895 u32 *sb32 = (u32*)sb;
896 int i;
1da177e4
LT		 897 unsigned int disk_csum, csum;
898
899 disk_csum = sb->sb_csum;
900 sb->sb_csum = 0;
4d167f09
N		 901
902 for (i = 0; i < MD_SB_BYTES/4 ; i++)
903 newcsum += sb32[i];
904 csum = (newcsum & 0xffffffff) + (newcsum>>32);
905
906
907#ifdef CONFIG_ALPHA
908 /* This used to use csum_partial, which was wrong for several
909 * reasons including that different results are returned on
910 * different architectures. It isn't critical that we get exactly
911 * the same return value as before (we always csum_fold before
912 * testing, and that removes any differences). However as we
913 * know that csum_partial always returned a 16bit value on
914 * alphas, do a fold to maximise conformity to previous behaviour.
915 */
916 sb->sb_csum = md_csum_fold(disk_csum);
917#else
1da177e4 918 sb->sb_csum = disk_csum;
4d167f09 919#endif
1da177e4
LT		 920 return csum;
921}
922
923
924/*
925 * Handle superblock details.
926 * We want to be able to handle multiple superblock formats
927 * so we have a common interface to them all, and an array of
928 * different handlers.
929 * We rely on user-space to write the initial superblock, and support
930 * reading and updating of superblocks.
931 * Interface methods are:
932 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
933 * loads and validates a superblock on dev.
934 * if refdev != NULL, compare superblocks on both devices
935 * Return:
936 * 0 - dev has a superblock that is compatible with refdev
937 * 1 - dev has a superblock that is compatible and newer than refdev
938 * so dev should be used as the refdev in future
939 * -EINVAL superblock incompatible or invalid
940 * -othererror e.g. -EIO
941 *
942 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
943 * Verify that dev is acceptable into mddev.
944 * The first time, mddev->raid_disks will be 0, and data from
945 * dev should be merged in. Subsequent calls check that dev
946 * is new enough. Return 0 or -EINVAL
947 *
948 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
949 * Update the superblock for rdev with data in mddev
950 * This does not write to disc.
951 *
952 */
953
954struct super_type {
0cd17fec
CW		 955 char *name;
956 struct module *owner;
957 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
958 int minor_version);
959 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
960 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
961 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
15f4a5fd 962 sector_t num_sectors);
1da177e4
LT		 963};
964
0894cc30
AN		 965/*
966 * Check that the given mddev has no bitmap.
967 *
968 * This function is called from the run method of all personalities that do not
969 * support bitmaps. It prints an error message and returns non-zero if mddev
970 * has a bitmap. Otherwise, it returns 0.
971 *
972 */
973int md_check_no_bitmap(mddev_t *mddev)
974{
c3d9714e 975 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 976 return 0;
977 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
978 mdname(mddev), mddev->pers->name);
979 return 1;
980}
981EXPORT_SYMBOL(md_check_no_bitmap);
982
1da177e4
LT		 983/*
984 * load_super for 0.90.0
985 */
986static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
987{
988 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
989 mdp_super_t *sb;
990 int ret;
1da177e4
LT		 991
992 /*
0f420358 993 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 994 * it's at the end of the disk.
995 *
996 * It also happens to be a multiple of 4Kb.
997 */
57b2caa3 998 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 999
0002b271 1000 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1001 if (ret) return ret;
1002
1003 ret = -EINVAL;
1004
1005 bdevname(rdev->bdev, b);
1006 sb = (mdp_super_t*)page_address(rdev->sb_page);
1007
1008 if (sb->md_magic != MD_SB_MAGIC) {
1009 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1010 b);
1011 goto abort;
1012 }
1013
1014 if (sb->major_version != 0 ||
f6705578
N		 1015 sb->minor_version < 90 ||
1016 sb->minor_version > 91) {
1da177e4
LT		 1017 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1018 sb->major_version, sb->minor_version,
1019 b);
1020 goto abort;
1021 }
1022
1023 if (sb->raid_disks <= 0)
1024 goto abort;
1025
4d167f09 1026 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1027 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1028 b);
1029 goto abort;
1030 }
1031
1032 rdev->preferred_minor = sb->md_minor;
1033 rdev->data_offset = 0;
0002b271 1034 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 1035
1036 if (sb->level == LEVEL_MULTIPATH)
1037 rdev->desc_nr = -1;
1038 else
1039 rdev->desc_nr = sb->this_disk.number;
1040
9a7b2b0f 1041 if (!refdev) {
1da177e4 1042 ret = 1;
9a7b2b0f 1043 } else {
1da177e4
LT		 1044 __u64 ev1, ev2;
1045 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1046 if (!uuid_equal(refsb, sb)) {
1047 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1048 b, bdevname(refdev->bdev,b2));
1049 goto abort;
1050 }
1051 if (!sb_equal(refsb, sb)) {
1052 printk(KERN_WARNING "md: %s has same UUID"
1053 " but different superblock to %s\n",
1054 b, bdevname(refdev->bdev, b2));
1055 goto abort;
1056 }
1057 ev1 = md_event(sb);
1058 ev2 = md_event(refsb);
1059 if (ev1 > ev2)
1060 ret = 1;
1061 else
1062 ret = 0;
1063 }
8190e754 1064 rdev->sectors = rdev->sb_start;
1da177e4 1065
dd8ac336 1066 if (rdev->sectors < sb->size * 2 && sb->level > 1)
2bf071bf
N		 1067 /* "this cannot possibly happen" ... */
1068 ret = -EINVAL;
1069
1da177e4
LT		 1070 abort:
1071 return ret;
1072}
1073
1074/*
1075 * validate_super for 0.90.0
1076 */
1077static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1078{
1079 mdp_disk_t *desc;
1080 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 1081 __u64 ev1 = md_event(sb);
1da177e4 1082
41158c7e 1083 rdev->raid_disk = -1;
c5d79adb
N		 1084 clear_bit(Faulty, &rdev->flags);
1085 clear_bit(In_sync, &rdev->flags);
1086 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1087
1da177e4
LT		 1088 if (mddev->raid_disks == 0) {
1089 mddev->major_version = 0;
1090 mddev->minor_version = sb->minor_version;
1091 mddev->patch_version = sb->patch_version;
e691063a 1092 mddev->external = 0;
9d8f0363 1093 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1094 mddev->ctime = sb->ctime;
1095 mddev->utime = sb->utime;
1096 mddev->level = sb->level;
d9d166c2 1097 mddev->clevel[0] = 0;
1da177e4
LT		 1098 mddev->layout = sb->layout;
1099 mddev->raid_disks = sb->raid_disks;
58c0fed4 1100 mddev->dev_sectors = sb->size * 2;
07d84d10 1101 mddev->events = ev1;
c3d9714e
N		 1102 mddev->bitmap_info.offset = 0;
1103 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1104
f6705578
N		 1105 if (mddev->minor_version >= 91) {
1106 mddev->reshape_position = sb->reshape_position;
1107 mddev->delta_disks = sb->delta_disks;
1108 mddev->new_level = sb->new_level;
1109 mddev->new_layout = sb->new_layout;
664e7c41 1110 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1111 } else {
1112 mddev->reshape_position = MaxSector;
1113 mddev->delta_disks = 0;
1114 mddev->new_level = mddev->level;
1115 mddev->new_layout = mddev->layout;
664e7c41 1116 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1117 }
1118
1da177e4
LT		 1119 if (sb->state & (1<<MD_SB_CLEAN))
1120 mddev->recovery_cp = MaxSector;
1121 else {
1122 if (sb->events_hi == sb->cp_events_hi &&
1123 sb->events_lo == sb->cp_events_lo) {
1124 mddev->recovery_cp = sb->recovery_cp;
1125 } else
1126 mddev->recovery_cp = 0;
1127 }
1128
1129 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1130 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1131 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1132 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1133
1134 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1135
1136 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1137 mddev->bitmap_info.file == NULL)
1138 mddev->bitmap_info.offset =
1139 mddev->bitmap_info.default_offset;
a654b9d8 1140
41158c7e 1141 } else if (mddev->pers == NULL) {
be6800a7
N		 1142 /* Insist on good event counter while assembling, except
1143 * for spares (which don't need an event count) */
1da177e4 1144 ++ev1;
be6800a7
N		 1145 if (sb->disks[rdev->desc_nr].state & (
1146 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1147 if (ev1 < mddev->events)
1148 return -EINVAL;
41158c7e
N		 1149 } else if (mddev->bitmap) {
1150 /* if adding to array with a bitmap, then we can accept an
1151 * older device ... but not too old.
1152 */
41158c7e
N		 1153 if (ev1 < mddev->bitmap->events_cleared)
1154 return 0;
07d84d10
N		 1155 } else {
1156 if (ev1 < mddev->events)
1157 /* just a hot-add of a new device, leave raid_disk at -1 */
1158 return 0;
1159 }
41158c7e 1160
1da177e4 1161 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1162 desc = sb->disks + rdev->desc_nr;
1163
1164 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1165 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1166 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1167 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1168 set_bit(In_sync, &rdev->flags);
1da177e4 1169 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1170 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1171 /* active but not in sync implies recovery up to
1172 * reshape position. We don't know exactly where
1173 * that is, so set to zero for now */
1174 if (mddev->minor_version >= 91) {
1175 rdev->recovery_offset = 0;
1176 rdev->raid_disk = desc->raid_disk;
1177 }
1da177e4 1178 }
8ddf9efe
N		 1179 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1180 set_bit(WriteMostly, &rdev->flags);
41158c7e 1181 } else /* MULTIPATH are always insync */
b2d444d7 1182 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1183 return 0;
1184}
1185
1186/*
1187 * sync_super for 0.90.0
1188 */
1189static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1190{
1191 mdp_super_t *sb;
1da177e4
LT		 1192 mdk_rdev_t *rdev2;
1193 int next_spare = mddev->raid_disks;
19133a42 1194
1da177e4
LT		 1195
1196 /* make rdev->sb match mddev data..
1197 *
1198 * 1/ zero out disks
1199 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1200 * 3/ any empty disks < next_spare become removed
1201 *
1202 * disks[0] gets initialised to REMOVED because
1203 * we cannot be sure from other fields if it has
1204 * been initialised or not.
1205 */
1206 int i;
1207 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1208
61181565
N		 1209 rdev->sb_size = MD_SB_BYTES;
1210
1da177e4
LT		 1211 sb = (mdp_super_t*)page_address(rdev->sb_page);
1212
1213 memset(sb, 0, sizeof(*sb));
1214
1215 sb->md_magic = MD_SB_MAGIC;
1216 sb->major_version = mddev->major_version;
1da177e4
LT		 1217 sb->patch_version = mddev->patch_version;
1218 sb->gvalid_words = 0; /* ignored */
1219 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1220 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1221 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1222 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1223
1224 sb->ctime = mddev->ctime;
1225 sb->level = mddev->level;
58c0fed4 1226 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1227 sb->raid_disks = mddev->raid_disks;
1228 sb->md_minor = mddev->md_minor;
e691063a 1229 sb->not_persistent = 0;
1da177e4
LT		 1230 sb->utime = mddev->utime;
1231 sb->state = 0;
1232 sb->events_hi = (mddev->events>>32);
1233 sb->events_lo = (u32)mddev->events;
1234
f6705578
N		 1235 if (mddev->reshape_position == MaxSector)
1236 sb->minor_version = 90;
1237 else {
1238 sb->minor_version = 91;
1239 sb->reshape_position = mddev->reshape_position;
1240 sb->new_level = mddev->new_level;
1241 sb->delta_disks = mddev->delta_disks;
1242 sb->new_layout = mddev->new_layout;
664e7c41 1243 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1244 }
1245 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1246 if (mddev->in_sync)
1247 {
1248 sb->recovery_cp = mddev->recovery_cp;
1249 sb->cp_events_hi = (mddev->events>>32);
1250 sb->cp_events_lo = (u32)mddev->events;
1251 if (mddev->recovery_cp == MaxSector)
1252 sb->state = (1<< MD_SB_CLEAN);
1253 } else
1254 sb->recovery_cp = 0;
1255
1256 sb->layout = mddev->layout;
9d8f0363 1257 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1258
c3d9714e 1259 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1260 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1261
1da177e4 1262 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1263 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1264 mdp_disk_t *d;
86e6ffdd 1265 int desc_nr;
0261cd9f
N		 1266 int is_active = test_bit(In_sync, &rdev2->flags);
1267
1268 if (rdev2->raid_disk >= 0 &&
1269 sb->minor_version >= 91)
1270 /* we have nowhere to store the recovery_offset,
1271 * but if it is not below the reshape_position,
1272 * we can piggy-back on that.
1273 */
1274 is_active = 1;
1275 if (rdev2->raid_disk < 0 ||
1276 test_bit(Faulty, &rdev2->flags))
1277 is_active = 0;
1278 if (is_active)
86e6ffdd 1279 desc_nr = rdev2->raid_disk;
1da177e4 1280 else
86e6ffdd 1281 desc_nr = next_spare++;
19133a42 1282 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1283 d = &sb->disks[rdev2->desc_nr];
1284 nr_disks++;
1285 d->number = rdev2->desc_nr;
1286 d->major = MAJOR(rdev2->bdev->bd_dev);
1287 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1288 if (is_active)
1da177e4
LT		 1289 d->raid_disk = rdev2->raid_disk;
1290 else
1291 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1292 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1293 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1294 else if (is_active) {
1da177e4 1295 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1296 if (test_bit(In_sync, &rdev2->flags))
1297 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1298 active++;
1299 working++;
1300 } else {
1301 d->state = 0;
1302 spare++;
1303 working++;
1304 }
8ddf9efe
N		 1305 if (test_bit(WriteMostly, &rdev2->flags))
1306 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1307 }
1da177e4
LT		 1308 /* now set the "removed" and "faulty" bits on any missing devices */
1309 for (i=0 ; i < mddev->raid_disks ; i++) {
1310 mdp_disk_t *d = &sb->disks[i];
1311 if (d->state == 0 && d->number == 0) {
1312 d->number = i;
1313 d->raid_disk = i;
1314 d->state = (1<<MD_DISK_REMOVED);
1315 d->state |= (1<<MD_DISK_FAULTY);
1316 failed++;
1317 }
1318 }
1319 sb->nr_disks = nr_disks;
1320 sb->active_disks = active;
1321 sb->working_disks = working;
1322 sb->failed_disks = failed;
1323 sb->spare_disks = spare;
1324
1325 sb->this_disk = sb->disks[rdev->desc_nr];
1326 sb->sb_csum = calc_sb_csum(sb);
1327}
1328
0cd17fec
CW		 1329/*
1330 * rdev_size_change for 0.90.0
1331 */
1332static unsigned long long
15f4a5fd 1333super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec 1334{
58c0fed4 1335 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1336 return 0; /* component must fit device */
c3d9714e 1337 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1338 return 0; /* can't move bitmap */
57b2caa3 1339 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1340 if (!num_sectors || num_sectors > rdev->sb_start)
1341 num_sectors = rdev->sb_start;
0f420358 1342 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1343 rdev->sb_page);
1344 md_super_wait(rdev->mddev);
c26a44ed 1345 return num_sectors;
0cd17fec
CW		 1346}
1347
1348
1da177e4
LT		 1349/*
1350 * version 1 superblock
1351 */
1352
1c05b4bc 1353static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1354{
1c05b4bc
N		 1355 __le32 disk_csum;
1356 u32 csum;
1da177e4
LT		 1357 unsigned long long newcsum;
1358 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1359 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1360 int i;
1361
1362 disk_csum = sb->sb_csum;
1363 sb->sb_csum = 0;
1364 newcsum = 0;
1365 for (i=0; size>=4; size -= 4 )
1366 newcsum += le32_to_cpu(*isuper++);
1367
1368 if (size == 2)
1c05b4bc 1369 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1370
1371 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1372 sb->sb_csum = disk_csum;
1373 return cpu_to_le32(csum);
1374}
1375
1376static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1377{
1378 struct mdp_superblock_1 *sb;
1379 int ret;
0f420358 1380 sector_t sb_start;
1da177e4 1381 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1382 int bmask;
1da177e4
LT		 1383
1384 /*
0f420358 1385 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1386 * It is always aligned to a 4K boundary and
1387 * depeding on minor_version, it can be:
1388 * 0: At least 8K, but less than 12K, from end of device
1389 * 1: At start of device
1390 * 2: 4K from start of device.
1391 */
1392 switch(minor_version) {
1393 case 0:
77304d2a 1394 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1395 sb_start -= 8*2;
1396 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1397 break;
1398 case 1:
0f420358 1399 sb_start = 0;
1da177e4
LT		 1400 break;
1401 case 2:
0f420358 1402 sb_start = 8;
1da177e4
LT		 1403 break;
1404 default:
1405 return -EINVAL;
1406 }
0f420358 1407 rdev->sb_start = sb_start;
1da177e4 1408
0002b271
N		 1409 /* superblock is rarely larger than 1K, but it can be larger,
1410 * and it is safe to read 4k, so we do that
1411 */
1412 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1413 if (ret) return ret;
1414
1415
1416 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1417
1418 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1419 sb->major_version != cpu_to_le32(1) ||
1420 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1421 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1422 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1423 return -EINVAL;
1424
1425 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1426 printk("md: invalid superblock checksum on %s\n",
1427 bdevname(rdev->bdev,b));
1428 return -EINVAL;
1429 }
1430 if (le64_to_cpu(sb->data_size) < 10) {
1431 printk("md: data_size too small on %s\n",
1432 bdevname(rdev->bdev,b));
1433 return -EINVAL;
1434 }
e11e93fa 1435
1da177e4
LT		 1436 rdev->preferred_minor = 0xffff;
1437 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1438 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1439
0002b271 1440 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1441 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1442 if (rdev->sb_size & bmask)
a1801f85
N		 1443 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1444
1445 if (minor_version
0f420358 1446 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1447 return -EINVAL;
0002b271 1448
31b65a0d
N		 1449 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1450 rdev->desc_nr = -1;
1451 else
1452 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1453
9a7b2b0f 1454 if (!refdev) {
8ed75463 1455 ret = 1;
9a7b2b0f 1456 } else {
1da177e4
LT		 1457 __u64 ev1, ev2;
1458 struct mdp_superblock_1 *refsb =
1459 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1460
1461 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1462 sb->level != refsb->level ||
1463 sb->layout != refsb->layout ||
1464 sb->chunksize != refsb->chunksize) {
1465 printk(KERN_WARNING "md: %s has strangely different"
1466 " superblock to %s\n",
1467 bdevname(rdev->bdev,b),
1468 bdevname(refdev->bdev,b2));
1469 return -EINVAL;
1470 }
1471 ev1 = le64_to_cpu(sb->events);
1472 ev2 = le64_to_cpu(refsb->events);
1473
1474 if (ev1 > ev2)
8ed75463
N		 1475 ret = 1;
1476 else
1477 ret = 0;
1da177e4 1478 }
a1801f85 1479 if (minor_version)
77304d2a 1480 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1481 le64_to_cpu(sb->data_offset);
1da177e4 1482 else
dd8ac336
AN		 1483 rdev->sectors = rdev->sb_start;
1484 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1485 return -EINVAL;
dd8ac336 1486 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1487 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1488 return -EINVAL;
8ed75463 1489 return ret;
1da177e4
LT		 1490}
1491
1492static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1493{
1494 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1495 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1496
41158c7e 1497 rdev->raid_disk = -1;
c5d79adb
N		 1498 clear_bit(Faulty, &rdev->flags);
1499 clear_bit(In_sync, &rdev->flags);
1500 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1501
1da177e4
LT		 1502 if (mddev->raid_disks == 0) {
1503 mddev->major_version = 1;
1504 mddev->patch_version = 0;
e691063a 1505 mddev->external = 0;
9d8f0363 1506 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1507 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1508 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1509 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1510 mddev->clevel[0] = 0;
1da177e4
LT		 1511 mddev->layout = le32_to_cpu(sb->layout);
1512 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1513 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1514 mddev->events = ev1;
c3d9714e
N		 1515 mddev->bitmap_info.offset = 0;
1516 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1517
1518 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1519 memcpy(mddev->uuid, sb->set_uuid, 16);
1520
1521 mddev->max_disks = (4096-256)/2;
a654b9d8 1522
71c0805c 1523 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1524 mddev->bitmap_info.file == NULL )
1525 mddev->bitmap_info.offset =
1526 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1527
f6705578
N		 1528 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1529 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1530 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1531 mddev->new_level = le32_to_cpu(sb->new_level);
1532 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1533 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1534 } else {
1535 mddev->reshape_position = MaxSector;
1536 mddev->delta_disks = 0;
1537 mddev->new_level = mddev->level;
1538 mddev->new_layout = mddev->layout;
664e7c41 1539 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1540 }
1541
41158c7e 1542 } else if (mddev->pers == NULL) {
be6800a7
N		 1543 /* Insist of good event counter while assembling, except for
1544 * spares (which don't need an event count) */
1da177e4 1545 ++ev1;
be6800a7
N		 1546 if (rdev->desc_nr >= 0 &&
1547 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1548 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1549 if (ev1 < mddev->events)
1550 return -EINVAL;
41158c7e
N		 1551 } else if (mddev->bitmap) {
1552 /* If adding to array with a bitmap, then we can accept an
1553 * older device, but not too old.
1554 */
41158c7e
N		 1555 if (ev1 < mddev->bitmap->events_cleared)
1556 return 0;
07d84d10
N		 1557 } else {
1558 if (ev1 < mddev->events)
1559 /* just a hot-add of a new device, leave raid_disk at -1 */
1560 return 0;
1561 }
1da177e4
LT		 1562 if (mddev->level != LEVEL_MULTIPATH) {
1563 int role;
3673f305
N		 1564 if (rdev->desc_nr < 0 ||
1565 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1566 role = 0xffff;
1567 rdev->desc_nr = -1;
1568 } else
1569 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1570 switch(role) {
1571 case 0xffff: /* spare */
1da177e4
LT		 1572 break;
1573 case 0xfffe: /* faulty */
b2d444d7 1574 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1575 break;
1576 default:
5fd6c1dc
N		 1577 if ((le32_to_cpu(sb->feature_map) &
1578 MD_FEATURE_RECOVERY_OFFSET))
1579 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1580 else
1581 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1582 rdev->raid_disk = role;
1583 break;
1584 }
8ddf9efe
N		 1585 if (sb->devflags & WriteMostly1)
1586 set_bit(WriteMostly, &rdev->flags);
41158c7e 1587 } else /* MULTIPATH are always insync */
b2d444d7 1588 set_bit(In_sync, &rdev->flags);
41158c7e 1589
1da177e4
LT		 1590 return 0;
1591}
1592
1593static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1594{
1595 struct mdp_superblock_1 *sb;
1da177e4
LT		 1596 mdk_rdev_t *rdev2;
1597 int max_dev, i;
1598 /* make rdev->sb match mddev and rdev data. */
1599
1600 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1601
1602 sb->feature_map = 0;
1603 sb->pad0 = 0;
5fd6c1dc 1604 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1605 memset(sb->pad1, 0, sizeof(sb->pad1));
1606 memset(sb->pad2, 0, sizeof(sb->pad2));
1607 memset(sb->pad3, 0, sizeof(sb->pad3));
1608
1609 sb->utime = cpu_to_le64((__u64)mddev->utime);
1610 sb->events = cpu_to_le64(mddev->events);
1611 if (mddev->in_sync)
1612 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1613 else
1614 sb->resync_offset = cpu_to_le64(0);
1615
1c05b4bc 1616 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1617
f0ca340c 1618 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1619 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1620 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1621 sb->level = cpu_to_le32(mddev->level);
1622 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1623
c3d9714e
N		 1624 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1625 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1626 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1627 }
5fd6c1dc
N		 1628
1629 if (rdev->raid_disk >= 0 &&
97e4f42d 1630 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1631 sb->feature_map |=
1632 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1633 sb->recovery_offset =
1634 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1635 }
1636
f6705578
N		 1637 if (mddev->reshape_position != MaxSector) {
1638 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1639 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1640 sb->new_layout = cpu_to_le32(mddev->new_layout);
1641 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1642 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1643 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1644 }
a654b9d8 1645
1da177e4 1646 max_dev = 0;
159ec1fc 1647 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1648 if (rdev2->desc_nr+1 > max_dev)
1649 max_dev = rdev2->desc_nr+1;
a778b73f 1650
70471daf
N		 1651 if (max_dev > le32_to_cpu(sb->max_dev)) {
1652 int bmask;
a778b73f 1653 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1654 rdev->sb_size = max_dev * 2 + 256;
1655 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1656 if (rdev->sb_size & bmask)
1657 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1658 } else
1659 max_dev = le32_to_cpu(sb->max_dev);
1660
1da177e4
LT		 1661 for (i=0; i<max_dev;i++)
1662 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1663
159ec1fc 1664 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1665 i = rdev2->desc_nr;
b2d444d7 1666 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1667 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1668 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1669 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1670 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1671 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1672 else
1673 sb->dev_roles[i] = cpu_to_le16(0xffff);
1674 }
1675
1da177e4
LT		 1676 sb->sb_csum = calc_sb_1_csum(sb);
1677}
1678
0cd17fec 1679static unsigned long long
15f4a5fd 1680super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec
CW		 1681{
1682 struct mdp_superblock_1 *sb;
15f4a5fd 1683 sector_t max_sectors;
58c0fed4 1684 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1685 return 0; /* component must fit device */
0f420358 1686 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1687 /* minor versions 1 and 2; superblock before data */
77304d2a 1688 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1689 max_sectors -= rdev->data_offset;
1690 if (!num_sectors || num_sectors > max_sectors)
1691 num_sectors = max_sectors;
c3d9714e 1692 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1693 /* minor version 0 with bitmap we can't move */
1694 return 0;
1695 } else {
1696 /* minor version 0; superblock after data */
0f420358 1697 sector_t sb_start;
77304d2a 1698 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1699 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1700 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1701 if (!num_sectors || num_sectors > max_sectors)
1702 num_sectors = max_sectors;
0f420358 1703 rdev->sb_start = sb_start;
0cd17fec
CW		 1704 }
1705 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
15f4a5fd 1706 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1707 sb->super_offset = rdev->sb_start;
0cd17fec 1708 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1709 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1710 rdev->sb_page);
1711 md_super_wait(rdev->mddev);
c26a44ed 1712 return num_sectors;
0cd17fec 1713}
1da177e4 1714
75c96f85 1715static struct super_type super_types[] = {
1da177e4
LT		 1716 [0] = {
1717 .name = "0.90.0",
1718 .owner = THIS_MODULE,
0cd17fec
CW		 1719 .load_super = super_90_load,
1720 .validate_super = super_90_validate,
1721 .sync_super = super_90_sync,
1722 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1723 },
1724 [1] = {
1725 .name = "md-1",
1726 .owner = THIS_MODULE,
0cd17fec
CW		 1727 .load_super = super_1_load,
1728 .validate_super = super_1_validate,
1729 .sync_super = super_1_sync,
1730 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1731 },
1732};
1da177e4
LT		 1733
1734static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1735{
7dd5e7c3 1736 mdk_rdev_t *rdev, *rdev2;
1da177e4 1737
4b80991c
N		 1738 rcu_read_lock();
1739 rdev_for_each_rcu(rdev, mddev1)
1740 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1741 if (rdev->bdev->bd_contains ==
4b80991c
N		 1742 rdev2->bdev->bd_contains) {
1743 rcu_read_unlock();
7dd5e7c3 1744 return 1;
4b80991c
N		 1745 }
1746 rcu_read_unlock();
1da177e4
LT		 1747 return 0;
1748}
1749
1750static LIST_HEAD(pending_raid_disks);
1751
ac5e7113
AN		 1752/*
1753 * Try to register data integrity profile for an mddev
1754 *
1755 * This is called when an array is started and after a disk has been kicked
1756 * from the array. It only succeeds if all working and active component devices
1757 * are integrity capable with matching profiles.
1758 */
1759int md_integrity_register(mddev_t *mddev)
1760{
1761 mdk_rdev_t *rdev, *reference = NULL;
1762
1763 if (list_empty(&mddev->disks))
1764 return 0; /* nothing to do */
1765 if (blk_get_integrity(mddev->gendisk))
1766 return 0; /* already registered */
1767 list_for_each_entry(rdev, &mddev->disks, same_set) {
1768 /* skip spares and non-functional disks */
1769 if (test_bit(Faulty, &rdev->flags))
1770 continue;
1771 if (rdev->raid_disk < 0)
1772 continue;
1773 /*
1774 * If at least one rdev is not integrity capable, we can not
1775 * enable data integrity for the md device.
1776 */
1777 if (!bdev_get_integrity(rdev->bdev))
1778 return -EINVAL;
1779 if (!reference) {
1780 /* Use the first rdev as the reference */
1781 reference = rdev;
1782 continue;
1783 }
1784 /* does this rdev's profile match the reference profile? */
1785 if (blk_integrity_compare(reference->bdev->bd_disk,
1786 rdev->bdev->bd_disk) < 0)
1787 return -EINVAL;
1788 }
1789 /*
1790 * All component devices are integrity capable and have matching
1791 * profiles, register the common profile for the md device.
1792 */
1793 if (blk_integrity_register(mddev->gendisk,
1794 bdev_get_integrity(reference->bdev)) != 0) {
1795 printk(KERN_ERR "md: failed to register integrity for %s\n",
1796 mdname(mddev));
1797 return -EINVAL;
1798 }
1799 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1800 mdname(mddev));
1801 return 0;
1802}
1803EXPORT_SYMBOL(md_integrity_register);
1804
1805/* Disable data integrity if non-capable/non-matching disk is being added */
1806void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
3f9d99c1 1807{
3f9d99c1 1808 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1809 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1810
ac5e7113 1811 if (!bi_mddev) /* nothing to do */
3f9d99c1 1812 return;
ac5e7113 1813 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1814 return;
ac5e7113
AN		 1815 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1816 rdev->bdev->bd_disk) >= 0)
1817 return;
1818 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1819 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 1820}
ac5e7113 1821EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 1822
1da177e4
LT		 1823static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1824{
7dd5e7c3 1825 char b[BDEVNAME_SIZE];
f637b9f9 1826 struct kobject *ko;
1edf80d3 1827 char *s;
5e55e2f5 1828 int err;
1da177e4
LT		 1829
1830 if (rdev->mddev) {
1831 MD_BUG();
1832 return -EINVAL;
1833 }
11e2ede0
DW		 1834
1835 /* prevent duplicates */
1836 if (find_rdev(mddev, rdev->bdev->bd_dev))
1837 return -EEXIST;
1838
dd8ac336
AN		 1839 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1840 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1841 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 1842 if (mddev->pers) {
1843 /* Cannot change size, so fail
1844 * If mddev->level <= 0, then we don't care
1845 * about aligning sizes (e.g. linear)
1846 */
1847 if (mddev->level > 0)
1848 return -ENOSPC;
1849 } else
dd8ac336 1850 mddev->dev_sectors = rdev->sectors;
2bf071bf 1851 }
1da177e4
LT		 1852
1853 /* Verify rdev->desc_nr is unique.
1854 * If it is -1, assign a free number, else
1855 * check number is not in use
1856 */
1857 if (rdev->desc_nr < 0) {
1858 int choice = 0;
1859 if (mddev->pers) choice = mddev->raid_disks;
1860 while (find_rdev_nr(mddev, choice))
1861 choice++;
1862 rdev->desc_nr = choice;
1863 } else {
1864 if (find_rdev_nr(mddev, rdev->desc_nr))
1865 return -EBUSY;
1866 }
de01dfad
N		 1867 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1868 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1869 mdname(mddev), mddev->max_disks);
1870 return -EBUSY;
1871 }
19133a42 1872 bdevname(rdev->bdev,b);
649316b2 1873 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1874 *s = '!';
649316b2 1875
1da177e4 1876 rdev->mddev = mddev;
19133a42 1877 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1878
b2d6db58 1879 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1880 goto fail;
86e6ffdd 1881
0762b8bd 1882 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 1883 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1884 /* failure here is OK */;
1885 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 1886
4b80991c 1887 list_add_rcu(&rdev->same_set, &mddev->disks);
c5d79adb 1888 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
4044ba58
N		 1889
1890 /* May as well allow recovery to be retried once */
1891 mddev->recovery_disabled = 0;
3f9d99c1 1892
1da177e4 1893 return 0;
5e55e2f5
N		 1894
1895 fail:
1896 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1897 b, mdname(mddev));
1898 return err;
1da177e4
LT		 1899}
1900
177a99b2 1901static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1902{
1903 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1904 kobject_del(&rdev->kobj);
177a99b2 1905 kobject_put(&rdev->kobj);
5792a285
N		 1906}
1907
1da177e4
LT		 1908static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1909{
1910 char b[BDEVNAME_SIZE];
1911 if (!rdev->mddev) {
1912 MD_BUG();
1913 return;
1914 }
5463c790 1915 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
4b80991c 1916 list_del_rcu(&rdev->same_set);
1da177e4
LT		 1917 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1918 rdev->mddev = NULL;
86e6ffdd 1919 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 1920 sysfs_put(rdev->sysfs_state);
1921 rdev->sysfs_state = NULL;
5792a285 1922 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 1923 * writing to 'remove' to "dev/state". We also need
1924 * to delay it due to rcu usage.
5792a285 1925 */
4b80991c 1926 synchronize_rcu();
177a99b2
N		 1927 INIT_WORK(&rdev->del_work, md_delayed_delete);
1928 kobject_get(&rdev->kobj);
e804ac78 1929 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 1930}
1931
1932/*
1933 * prevent the device from being mounted, repartitioned or
1934 * otherwise reused by a RAID array (or any other kernel
1935 * subsystem), by bd_claiming the device.
1936 */
c5d79adb 1937static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1938{
1939 int err = 0;
1940 struct block_device *bdev;
1941 char b[BDEVNAME_SIZE];
1942
2e7b651d 1943 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1944 if (IS_ERR(bdev)) {
1945 printk(KERN_ERR "md: could not open %s.\n",
1946 __bdevname(dev, b));
1947 return PTR_ERR(bdev);
1948 }
c5d79adb 1949 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1950 if (err) {
1951 printk(KERN_ERR "md: could not bd_claim %s.\n",
1952 bdevname(bdev, b));
9a1c3542 1953 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1954 return err;
1955 }
c5d79adb
N		 1956 if (!shared)
1957 set_bit(AllReserved, &rdev->flags);
1da177e4
LT		 1958 rdev->bdev = bdev;
1959 return err;
1960}
1961
1962static void unlock_rdev(mdk_rdev_t *rdev)
1963{
1964 struct block_device *bdev = rdev->bdev;
1965 rdev->bdev = NULL;
1966 if (!bdev)
1967 MD_BUG();
1968 bd_release(bdev);
9a1c3542 1969 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1da177e4
LT		 1970}
1971
1972void md_autodetect_dev(dev_t dev);
1973
1974static void export_rdev(mdk_rdev_t * rdev)
1975{
1976 char b[BDEVNAME_SIZE];
1977 printk(KERN_INFO "md: export_rdev(%s)\n",
1978 bdevname(rdev->bdev,b));
1979 if (rdev->mddev)
1980 MD_BUG();
1981 free_disk_sb(rdev);
1da177e4 1982#ifndef MODULE
d0fae18f
N		 1983 if (test_bit(AutoDetected, &rdev->flags))
1984 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1985#endif
1986 unlock_rdev(rdev);
86e6ffdd 1987 kobject_put(&rdev->kobj);
1da177e4
LT		 1988}
1989
1990static void kick_rdev_from_array(mdk_rdev_t * rdev)
1991{
1992 unbind_rdev_from_array(rdev);
1993 export_rdev(rdev);
1994}
1995
1996static void export_array(mddev_t *mddev)
1997{
159ec1fc 1998 mdk_rdev_t *rdev, *tmp;
1da177e4 1999
d089c6af 2000 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 2001 if (!rdev->mddev) {
2002 MD_BUG();
2003 continue;
2004 }
2005 kick_rdev_from_array(rdev);
2006 }
2007 if (!list_empty(&mddev->disks))
2008 MD_BUG();
2009 mddev->raid_disks = 0;
2010 mddev->major_version = 0;
2011}
2012
2013static void print_desc(mdp_disk_t *desc)
2014{
2015 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2016 desc->major,desc->minor,desc->raid_disk,desc->state);
2017}
2018
cd2ac932 2019static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2020{
2021 int i;
2022
2023 printk(KERN_INFO
2024 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2025 sb->major_version, sb->minor_version, sb->patch_version,
2026 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2027 sb->ctime);
2028 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2029 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2030 sb->md_minor, sb->layout, sb->chunk_size);
2031 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2032 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2033 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2034 sb->failed_disks, sb->spare_disks,
2035 sb->sb_csum, (unsigned long)sb->events_lo);
2036
2037 printk(KERN_INFO);
2038 for (i = 0; i < MD_SB_DISKS; i++) {
2039 mdp_disk_t *desc;
2040
2041 desc = sb->disks + i;
2042 if (desc->number || desc->major || desc->minor ||
2043 desc->raid_disk || (desc->state && (desc->state != 4))) {
2044 printk(" D %2d: ", i);
2045 print_desc(desc);
2046 }
2047 }
2048 printk(KERN_INFO "md: THIS: ");
2049 print_desc(&sb->this_disk);
cd2ac932 2050}
1da177e4 2051
cd2ac932
CR		 2052static void print_sb_1(struct mdp_superblock_1 *sb)
2053{
2054 __u8 *uuid;
2055
2056 uuid = sb->set_uuid;
ad361c98 2057 printk(KERN_INFO
7b75c2f8 2058 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2059 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2060 le32_to_cpu(sb->major_version),
2061 le32_to_cpu(sb->feature_map),
7b75c2f8 2062 uuid,
cd2ac932
CR		 2063 sb->set_name,
2064 (unsigned long long)le64_to_cpu(sb->ctime)
2065 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2066
2067 uuid = sb->device_uuid;
ad361c98
JP		 2068 printk(KERN_INFO
2069 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2070 " RO:%llu\n"
7b75c2f8 2071 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2072 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2073 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2074 le32_to_cpu(sb->level),
2075 (unsigned long long)le64_to_cpu(sb->size),
2076 le32_to_cpu(sb->raid_disks),
2077 le32_to_cpu(sb->layout),
2078 le32_to_cpu(sb->chunksize),
2079 (unsigned long long)le64_to_cpu(sb->data_offset),
2080 (unsigned long long)le64_to_cpu(sb->data_size),
2081 (unsigned long long)le64_to_cpu(sb->super_offset),
2082 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2083 le32_to_cpu(sb->dev_number),
7b75c2f8 2084 uuid,
cd2ac932
CR		 2085 sb->devflags,
2086 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2087 (unsigned long long)le64_to_cpu(sb->events),
2088 (unsigned long long)le64_to_cpu(sb->resync_offset),
2089 le32_to_cpu(sb->sb_csum),
2090 le32_to_cpu(sb->max_dev)
2091 );
1da177e4
LT		 2092}
2093
cd2ac932 2094static void print_rdev(mdk_rdev_t *rdev, int major_version)
1da177e4
LT		 2095{
2096 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2097 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2098 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2099 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2100 rdev->desc_nr);
1da177e4 2101 if (rdev->sb_loaded) {
cd2ac932
CR		 2102 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2103 switch (major_version) {
2104 case 0:
2105 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2106 break;
2107 case 1:
2108 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2109 break;
2110 }
1da177e4
LT		 2111 } else
2112 printk(KERN_INFO "md: no rdev superblock!\n");
2113}
2114
5e56341d 2115static void md_print_devices(void)
1da177e4 2116{
159ec1fc 2117 struct list_head *tmp;
1da177e4
LT		 2118 mdk_rdev_t *rdev;
2119 mddev_t *mddev;
2120 char b[BDEVNAME_SIZE];
2121
2122 printk("\n");
2123 printk("md: **********************************\n");
2124 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2125 printk("md: **********************************\n");
29ac4aa3 2126 for_each_mddev(mddev, tmp) {
1da177e4 2127
32a7627c
N		 2128 if (mddev->bitmap)
2129 bitmap_print_sb(mddev->bitmap);
2130 else
2131 printk("%s: ", mdname(mddev));
159ec1fc 2132 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2133 printk("<%s>", bdevname(rdev->bdev,b));
2134 printk("\n");
2135
159ec1fc 2136 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2137 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2138 }
2139 printk("md: **********************************\n");
2140 printk("\n");
2141}
2142
2143
42543769 2144static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 2145{
42543769
N		 2146 /* Update each superblock (in-memory image), but
2147 * if we are allowed to, skip spares which already
2148 * have the right event counter, or have one earlier
2149 * (which would mean they aren't being marked as dirty
2150 * with the rest of the array)
2151 */
1da177e4 2152 mdk_rdev_t *rdev;
159ec1fc 2153 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2154 if (rdev->sb_events == mddev->events ||
2155 (nospares &&
2156 rdev->raid_disk < 0 &&
42543769
N		 2157 rdev->sb_events+1 == mddev->events)) {
2158 /* Don't update this superblock */
2159 rdev->sb_loaded = 2;
2160 } else {
2161 super_types[mddev->major_version].
2162 sync_super(mddev, rdev);
2163 rdev->sb_loaded = 1;
2164 }
1da177e4
LT		 2165 }
2166}
2167
850b2b42 2168static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 2169{
1da177e4 2170 mdk_rdev_t *rdev;
06d91a5f 2171 int sync_req;
42543769 2172 int nospares = 0;
1da177e4 2173
1da177e4 2174repeat:
3a3a5ddb
N		 2175 /* First make sure individual recovery_offsets are correct */
2176 list_for_each_entry(rdev, &mddev->disks, same_set) {
2177 if (rdev->raid_disk >= 0 &&
2178 mddev->delta_disks >= 0 &&
2179 !test_bit(In_sync, &rdev->flags) &&
2180 mddev->curr_resync_completed > rdev->recovery_offset)
2181 rdev->recovery_offset = mddev->curr_resync_completed;
2182
2183 }
bd52b746 2184 if (!mddev->persistent) {
070dc6dd 2185 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2186 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
d97a41dc
N		 2187 if (!mddev->external)
2188 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3a3a5ddb
N		 2189 wake_up(&mddev->sb_wait);
2190 return;
2191 }
2192
a9701a30 2193 spin_lock_irq(&mddev->write_lock);
84692195 2194
3a3a5ddb
N		 2195 mddev->utime = get_seconds();
2196
850b2b42
N		 2197 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2198 force_change = 1;
2199 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2200 /* just a clean<-> dirty transition, possibly leave spares alone,
2201 * though if events isn't the right even/odd, we will have to do
2202 * spares after all
2203 */
2204 nospares = 1;
2205 if (force_change)
2206 nospares = 0;
2207 if (mddev->degraded)
84692195
N		 2208 /* If the array is degraded, then skipping spares is both
2209 * dangerous and fairly pointless.
2210 * Dangerous because a device that was removed from the array
2211 * might have a event_count that still looks up-to-date,
2212 * so it can be re-added without a resync.
2213 * Pointless because if there are any spares to skip,
2214 * then a recovery will happen and soon that array won't
2215 * be degraded any more and the spare can go back to sleep then.
2216 */
850b2b42 2217 nospares = 0;
84692195 2218
06d91a5f 2219 sync_req = mddev->in_sync;
42543769
N		 2220
2221 /* If this is just a dirty<->clean transition, and the array is clean
2222 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2223 if (nospares
42543769 2224 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2225 && mddev->can_decrease_events
2226 && mddev->events != 1) {
42543769 2227 mddev->events--;
a8707c08
N		 2228 mddev->can_decrease_events = 0;
2229 } else {
42543769
N		 2230 /* otherwise we have to go forward and ... */
2231 mddev->events ++;
a8707c08 2232 mddev->can_decrease_events = nospares;
42543769 2233 }
1da177e4
LT		 2234
2235 if (!mddev->events) {
2236 /*
2237 * oops, this 64-bit counter should never wrap.
2238 * Either we are in around ~1 trillion A.C., assuming
2239 * 1 reboot per second, or we have a bug:
2240 */
2241 MD_BUG();
2242 mddev->events --;
2243 }
e691063a 2244 sync_sbs(mddev, nospares);
a9701a30 2245 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 2246
2247 dprintk(KERN_INFO
2248 "md: updating %s RAID superblock on device (in sync %d)\n",
2249 mdname(mddev),mddev->in_sync);
2250
4ad13663 2251 bitmap_update_sb(mddev->bitmap);
159ec1fc 2252 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 2253 char b[BDEVNAME_SIZE];
2254 dprintk(KERN_INFO "md: ");
42543769
N		 2255 if (rdev->sb_loaded != 1)
2256 continue; /* no noise on spare devices */
b2d444d7 2257 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2258 dprintk("(skipping faulty ");
2259
2260 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 2261 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2262 md_super_write(mddev,rdev,
0f420358 2263 rdev->sb_start, rdev->sb_size,
7bfa19f2
N		 2264 rdev->sb_page);
2265 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2266 bdevname(rdev->bdev,b),
0f420358 2267 (unsigned long long)rdev->sb_start);
42543769 2268 rdev->sb_events = mddev->events;
7bfa19f2 2269
1da177e4
LT		 2270 } else
2271 dprintk(")\n");
7bfa19f2 2272 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2273 /* only need to write one superblock... */
2274 break;
2275 }
a9701a30 2276 md_super_wait(mddev);
850b2b42 2277 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2278
a9701a30 2279 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2280 if (mddev->in_sync != sync_req ||
2281 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2282 /* have to write it out again */
a9701a30 2283 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2284 goto repeat;
2285 }
850b2b42 2286 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2287 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2288 wake_up(&mddev->sb_wait);
acb180b0
N		 2289 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2290 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2291
1da177e4
LT		 2292}
2293
7f6ce769 2294/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2295 * We want to accept with case. For this we use cmd_match.
2296 */
2297static int cmd_match(const char *cmd, const char *str)
2298{
2299 /* See if cmd, written into a sysfs file, matches
2300 * str. They must either be the same, or cmd can
2301 * have a trailing newline
2302 */
2303 while (*cmd && *str && *cmd == *str) {
2304 cmd++;
2305 str++;
2306 }
2307 if (*cmd == '\n')
2308 cmd++;
2309 if (*str || *cmd)
2310 return 0;
2311 return 1;
2312}
2313
86e6ffdd
N		 2314struct rdev_sysfs_entry {
2315 struct attribute attr;
2316 ssize_t (*show)(mdk_rdev_t *, char *);
2317 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2318};
2319
2320static ssize_t
96de1e66 2321state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 2322{
2323 char *sep = "";
20a49ff6 2324 size_t len = 0;
86e6ffdd 2325
b2d444d7 2326 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 2327 len+= sprintf(page+len, "%sfaulty",sep);
2328 sep = ",";
2329 }
b2d444d7 2330 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2331 len += sprintf(page+len, "%sin_sync",sep);
2332 sep = ",";
2333 }
f655675b
N		 2334 if (test_bit(WriteMostly, &rdev->flags)) {
2335 len += sprintf(page+len, "%swrite_mostly",sep);
2336 sep = ",";
2337 }
6bfe0b49
DW		 2338 if (test_bit(Blocked, &rdev->flags)) {
2339 len += sprintf(page+len, "%sblocked", sep);
2340 sep = ",";
2341 }
b2d444d7
N		 2342 if (!test_bit(Faulty, &rdev->flags) &&
2343 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2344 len += sprintf(page+len, "%sspare", sep);
2345 sep = ",";
2346 }
2347 return len+sprintf(page+len, "\n");
2348}
2349
45dc2de1
N		 2350static ssize_t
2351state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2352{
2353 /* can write
2354 * faulty - simulates and error
2355 * remove - disconnects the device
f655675b
N		 2356 * writemostly - sets write_mostly
2357 * -writemostly - clears write_mostly
6bfe0b49
DW		 2358 * blocked - sets the Blocked flag
2359 * -blocked - clears the Blocked flag
6d56e278 2360 * insync - sets Insync providing device isn't active
45dc2de1
N		 2361 */
2362 int err = -EINVAL;
2363 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2364 md_error(rdev->mddev, rdev);
2365 err = 0;
2366 } else if (cmd_match(buf, "remove")) {
2367 if (rdev->raid_disk >= 0)
2368 err = -EBUSY;
2369 else {
2370 mddev_t *mddev = rdev->mddev;
2371 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2372 if (mddev->pers)
2373 md_update_sb(mddev, 1);
45dc2de1
N		 2374 md_new_event(mddev);
2375 err = 0;
2376 }
f655675b
N		 2377 } else if (cmd_match(buf, "writemostly")) {
2378 set_bit(WriteMostly, &rdev->flags);
2379 err = 0;
2380 } else if (cmd_match(buf, "-writemostly")) {
2381 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2382 err = 0;
2383 } else if (cmd_match(buf, "blocked")) {
2384 set_bit(Blocked, &rdev->flags);
2385 err = 0;
2386 } else if (cmd_match(buf, "-blocked")) {
2387 clear_bit(Blocked, &rdev->flags);
2388 wake_up(&rdev->blocked_wait);
2389 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2390 md_wakeup_thread(rdev->mddev->thread);
2391
6d56e278
N		 2392 err = 0;
2393 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2394 set_bit(In_sync, &rdev->flags);
f655675b 2395 err = 0;
45dc2de1 2396 }
00bcb4ac
N		 2397 if (!err)
2398 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2399 return err ? err : len;
2400}
80ca3a44
N		 2401static struct rdev_sysfs_entry rdev_state =
2402__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2403
4dbcdc75
N		 2404static ssize_t
2405errors_show(mdk_rdev_t *rdev, char *page)
2406{
2407 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2408}
2409
2410static ssize_t
2411errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2412{
2413 char *e;
2414 unsigned long n = simple_strtoul(buf, &e, 10);
2415 if (*buf && (*e == 0 || *e == '\n')) {
2416 atomic_set(&rdev->corrected_errors, n);
2417 return len;
2418 }
2419 return -EINVAL;
2420}
2421static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2422__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2423
014236d2
N		 2424static ssize_t
2425slot_show(mdk_rdev_t *rdev, char *page)
2426{
2427 if (rdev->raid_disk < 0)
2428 return sprintf(page, "none\n");
2429 else
2430 return sprintf(page, "%d\n", rdev->raid_disk);
2431}
2432
2433static ssize_t
2434slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2435{
2436 char *e;
c303da6d
N		 2437 int err;
2438 char nm[20];
014236d2
N		 2439 int slot = simple_strtoul(buf, &e, 10);
2440 if (strncmp(buf, "none", 4)==0)
2441 slot = -1;
2442 else if (e==buf || (*e && *e!= '\n'))
2443 return -EINVAL;
6c2fce2e 2444 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2445 /* Setting 'slot' on an active array requires also
2446 * updating the 'rd%d' link, and communicating
2447 * with the personality with ->hot_*_disk.
2448 * For now we only support removing
2449 * failed/spare devices. This normally happens automatically,
2450 * but not when the metadata is externally managed.
2451 */
c303da6d
N		 2452 if (rdev->raid_disk == -1)
2453 return -EEXIST;
2454 /* personality does all needed checks */
2455 if (rdev->mddev->pers->hot_add_disk == NULL)
2456 return -EINVAL;
2457 err = rdev->mddev->pers->
2458 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2459 if (err)
2460 return err;
2461 sprintf(nm, "rd%d", rdev->raid_disk);
2462 sysfs_remove_link(&rdev->mddev->kobj, nm);
b7103107 2463 rdev->raid_disk = -1;
c303da6d
N		 2464 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2465 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e
NB		 2466 } else if (rdev->mddev->pers) {
2467 mdk_rdev_t *rdev2;
6c2fce2e 2468 /* Activating a spare .. or possibly reactivating
6d56e278 2469 * if we ever get bitmaps working here.
6c2fce2e
NB		 2470 */
2471
2472 if (rdev->raid_disk != -1)
2473 return -EBUSY;
2474
2475 if (rdev->mddev->pers->hot_add_disk == NULL)
2476 return -EINVAL;
2477
159ec1fc 2478 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2479 if (rdev2->raid_disk == slot)
2480 return -EEXIST;
2481
ba1b41b6
N		 2482 if (slot >= rdev->mddev->raid_disks &&
2483 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2484 return -ENOSPC;
2485
6c2fce2e
NB		 2486 rdev->raid_disk = slot;
2487 if (test_bit(In_sync, &rdev->flags))
2488 rdev->saved_raid_disk = slot;
2489 else
2490 rdev->saved_raid_disk = -1;
2491 err = rdev->mddev->pers->
2492 hot_add_disk(rdev->mddev, rdev);
199050ea 2493 if (err) {
6c2fce2e 2494 rdev->raid_disk = -1;
6c2fce2e 2495 return err;
52664732 2496 } else
00bcb4ac 2497 sysfs_notify_dirent_safe(rdev->sysfs_state);
6c2fce2e
NB		 2498 sprintf(nm, "rd%d", rdev->raid_disk);
2499 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
00bcb4ac 2500 /* failure here is OK */;
6c2fce2e 2501 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2502 } else {
ba1b41b6
N		 2503 if (slot >= rdev->mddev->raid_disks &&
2504 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2505 return -ENOSPC;
2506 rdev->raid_disk = slot;
2507 /* assume it is working */
c5d79adb
N		 2508 clear_bit(Faulty, &rdev->flags);
2509 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2510 set_bit(In_sync, &rdev->flags);
00bcb4ac 2511 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2512 }
014236d2
N		 2513 return len;
2514}
2515
2516
2517static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2518__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2519
93c8cad0
N		 2520static ssize_t
2521offset_show(mdk_rdev_t *rdev, char *page)
2522{
6961ece4 2523 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2524}
2525
2526static ssize_t
2527offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2528{
2529 char *e;
2530 unsigned long long offset = simple_strtoull(buf, &e, 10);
2531 if (e==buf || (*e && *e != '\n'))
2532 return -EINVAL;
8ed0a521 2533 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2534 return -EBUSY;
dd8ac336 2535 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2536 /* Must set offset before size, so overlap checks
2537 * can be sane */
2538 return -EBUSY;
93c8cad0
N		 2539 rdev->data_offset = offset;
2540 return len;
2541}
2542
2543static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2544__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2545
83303b61
N		 2546static ssize_t
2547rdev_size_show(mdk_rdev_t *rdev, char *page)
2548{
dd8ac336 2549 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2550}
2551
c5d79adb
N		 2552static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2553{
2554 /* check if two start/length pairs overlap */
2555 if (s1+l1 <= s2)
2556 return 0;
2557 if (s2+l2 <= s1)
2558 return 0;
2559 return 1;
2560}
2561
b522adcd
DW		 2562static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2563{
2564 unsigned long long blocks;
2565 sector_t new;
2566
2567 if (strict_strtoull(buf, 10, &blocks) < 0)
2568 return -EINVAL;
2569
2570 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2571 return -EINVAL; /* sector conversion overflow */
2572
2573 new = blocks * 2;
2574 if (new != blocks * 2)
2575 return -EINVAL; /* unsigned long long to sector_t overflow */
2576
2577 *sectors = new;
2578 return 0;
2579}
2580
83303b61
N		 2581static ssize_t
2582rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2583{
27c529bb 2584 mddev_t *my_mddev = rdev->mddev;
dd8ac336 2585 sector_t oldsectors = rdev->sectors;
b522adcd 2586 sector_t sectors;
27c529bb 2587
b522adcd 2588 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2589 return -EINVAL;
0cd17fec 2590 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2591 if (my_mddev->persistent) {
dd8ac336
AN		 2592 sectors = super_types[my_mddev->major_version].
2593 rdev_size_change(rdev, sectors);
2594 if (!sectors)
0cd17fec 2595 return -EBUSY;
dd8ac336 2596 } else if (!sectors)
77304d2a 2597 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2598 rdev->data_offset;
0cd17fec 2599 }
dd8ac336 2600 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2601 return -EINVAL; /* component must fit device */
0cd17fec 2602
dd8ac336
AN		 2603 rdev->sectors = sectors;
2604 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2605 /* need to check that all other rdevs with the same ->bdev
2606 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2607 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2608 * we have to change it back, we will have the lock again.
2609 */
2610 mddev_t *mddev;
2611 int overlap = 0;
159ec1fc 2612 struct list_head *tmp;
c5d79adb 2613
27c529bb 2614 mddev_unlock(my_mddev);
29ac4aa3 2615 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2616 mdk_rdev_t *rdev2;
2617
2618 mddev_lock(mddev);
159ec1fc 2619 list_for_each_entry(rdev2, &mddev->disks, same_set)
c5d79adb
N		 2620 if (test_bit(AllReserved, &rdev2->flags) ||
2621 (rdev->bdev == rdev2->bdev &&
2622 rdev != rdev2 &&
dd8ac336 2623 overlaps(rdev->data_offset, rdev->sectors,
d07bd3bc 2624 rdev2->data_offset,
dd8ac336 2625 rdev2->sectors))) {
c5d79adb
N		 2626 overlap = 1;
2627 break;
2628 }
2629 mddev_unlock(mddev);
2630 if (overlap) {
2631 mddev_put(mddev);
2632 break;
2633 }
2634 }
27c529bb 2635 mddev_lock(my_mddev);
c5d79adb
N		 2636 if (overlap) {
2637 /* Someone else could have slipped in a size
2638 * change here, but doing so is just silly.
dd8ac336 2639 * We put oldsectors back because we *know* it is
c5d79adb
N		 2640 * safe, and trust userspace not to race with
2641 * itself
2642 */
dd8ac336 2643 rdev->sectors = oldsectors;
c5d79adb
N		 2644 return -EBUSY;
2645 }
2646 }
83303b61
N		 2647 return len;
2648}
2649
2650static struct rdev_sysfs_entry rdev_size =
80ca3a44 2651__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2652
06e3c817
DW		 2653
2654static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2655{
2656 unsigned long long recovery_start = rdev->recovery_offset;
2657
2658 if (test_bit(In_sync, &rdev->flags) ||
2659 recovery_start == MaxSector)
2660 return sprintf(page, "none\n");
2661
2662 return sprintf(page, "%llu\n", recovery_start);
2663}
2664
2665static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2666{
2667 unsigned long long recovery_start;
2668
2669 if (cmd_match(buf, "none"))
2670 recovery_start = MaxSector;
2671 else if (strict_strtoull(buf, 10, &recovery_start))
2672 return -EINVAL;
2673
2674 if (rdev->mddev->pers &&
2675 rdev->raid_disk >= 0)
2676 return -EBUSY;
2677
2678 rdev->recovery_offset = recovery_start;
2679 if (recovery_start == MaxSector)
2680 set_bit(In_sync, &rdev->flags);
2681 else
2682 clear_bit(In_sync, &rdev->flags);
2683 return len;
2684}
2685
2686static struct rdev_sysfs_entry rdev_recovery_start =
2687__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2688
86e6ffdd
N		 2689static struct attribute *rdev_default_attrs[] = {
2690 &rdev_state.attr,
4dbcdc75 2691 &rdev_errors.attr,
014236d2 2692 &rdev_slot.attr,
93c8cad0 2693 &rdev_offset.attr,
83303b61 2694 &rdev_size.attr,
06e3c817 2695 &rdev_recovery_start.attr,
86e6ffdd
N		 2696 NULL,
2697};
2698static ssize_t
2699rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2700{
2701 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2702 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2703 mddev_t *mddev = rdev->mddev;
2704 ssize_t rv;
86e6ffdd
N		 2705
2706 if (!entry->show)
2707 return -EIO;
27c529bb
N		 2708
2709 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2710 if (!rv) {
2711 if (rdev->mddev == NULL)
2712 rv = -EBUSY;
2713 else
2714 rv = entry->show(rdev, page);
2715 mddev_unlock(mddev);
2716 }
2717 return rv;
86e6ffdd
N		 2718}
2719
2720static ssize_t
2721rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2722 const char *page, size_t length)
2723{
2724 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2725 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2726 ssize_t rv;
2727 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2728
2729 if (!entry->store)
2730 return -EIO;
67463acb
N		 2731 if (!capable(CAP_SYS_ADMIN))
2732 return -EACCES;
27c529bb 2733 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2734 if (!rv) {
27c529bb
N		 2735 if (rdev->mddev == NULL)
2736 rv = -EBUSY;
2737 else
2738 rv = entry->store(rdev, page, length);
6a51830e 2739 mddev_unlock(mddev);
ca388059
N		 2740 }
2741 return rv;
86e6ffdd
N		 2742}
2743
2744static void rdev_free(struct kobject *ko)
2745{
2746 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2747 kfree(rdev);
2748}
52cf25d0 2749static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 2750 .show = rdev_attr_show,
2751 .store = rdev_attr_store,
2752};
2753static struct kobj_type rdev_ktype = {
2754 .release = rdev_free,
2755 .sysfs_ops = &rdev_sysfs_ops,
2756 .default_attrs = rdev_default_attrs,
2757};
2758
e8bb9a83
N		 2759void md_rdev_init(mdk_rdev_t *rdev)
2760{
2761 rdev->desc_nr = -1;
2762 rdev->saved_raid_disk = -1;
2763 rdev->raid_disk = -1;
2764 rdev->flags = 0;
2765 rdev->data_offset = 0;
2766 rdev->sb_events = 0;
2767 rdev->last_read_error.tv_sec = 0;
2768 rdev->last_read_error.tv_nsec = 0;
2769 atomic_set(&rdev->nr_pending, 0);
2770 atomic_set(&rdev->read_errors, 0);
2771 atomic_set(&rdev->corrected_errors, 0);
2772
2773 INIT_LIST_HEAD(&rdev->same_set);
2774 init_waitqueue_head(&rdev->blocked_wait);
2775}
2776EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 2777/*
2778 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2779 *
2780 * mark the device faulty if:
2781 *
2782 * - the device is nonexistent (zero size)
2783 * - the device has no valid superblock
2784 *
2785 * a faulty rdev _never_ has rdev->sb set.
2786 */
2787static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2788{
2789 char b[BDEVNAME_SIZE];
2790 int err;
2791 mdk_rdev_t *rdev;
2792 sector_t size;
2793
9ffae0cf 2794 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2795 if (!rdev) {
2796 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2797 return ERR_PTR(-ENOMEM);
2798 }
1da177e4 2799
e8bb9a83 2800 md_rdev_init(rdev);
1da177e4
LT		 2801 if ((err = alloc_disk_sb(rdev)))
2802 goto abort_free;
2803
c5d79adb 2804 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2805 if (err)
2806 goto abort_free;
2807
f9cb074b 2808 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2809
77304d2a 2810 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 2811 if (!size) {
2812 printk(KERN_WARNING
2813 "md: %s has zero or unknown size, marking faulty!\n",
2814 bdevname(rdev->bdev,b));
2815 err = -EINVAL;
2816 goto abort_free;
2817 }
2818
2819 if (super_format >= 0) {
2820 err = super_types[super_format].
2821 load_super(rdev, NULL, super_minor);
2822 if (err == -EINVAL) {
df968c4e
N		 2823 printk(KERN_WARNING
2824 "md: %s does not have a valid v%d.%d "
2825 "superblock, not importing!\n",
2826 bdevname(rdev->bdev,b),
2827 super_format, super_minor);
1da177e4
LT		 2828 goto abort_free;
2829 }
2830 if (err < 0) {
2831 printk(KERN_WARNING
2832 "md: could not read %s's sb, not importing!\n",
2833 bdevname(rdev->bdev,b));
2834 goto abort_free;
2835 }
2836 }
6bfe0b49 2837
1da177e4
LT		 2838 return rdev;
2839
2840abort_free:
2841 if (rdev->sb_page) {
2842 if (rdev->bdev)
2843 unlock_rdev(rdev);
2844 free_disk_sb(rdev);
2845 }
2846 kfree(rdev);
2847 return ERR_PTR(err);
2848}
2849
2850/*
2851 * Check a full RAID array for plausibility
2852 */
2853
2854
a757e64c 2855static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2856{
2857 int i;
159ec1fc 2858 mdk_rdev_t *rdev, *freshest, *tmp;
1da177e4
LT		 2859 char b[BDEVNAME_SIZE];
2860
2861 freshest = NULL;
d089c6af 2862 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2863 switch (super_types[mddev->major_version].
2864 load_super(rdev, freshest, mddev->minor_version)) {
2865 case 1:
2866 freshest = rdev;
2867 break;
2868 case 0:
2869 break;
2870 default:
2871 printk( KERN_ERR \
2872 "md: fatal superblock inconsistency in %s"
2873 " -- removing from array\n",
2874 bdevname(rdev->bdev,b));
2875 kick_rdev_from_array(rdev);
2876 }
2877
2878
2879 super_types[mddev->major_version].
2880 validate_super(mddev, freshest);
2881
2882 i = 0;
d089c6af 2883 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 2884 if (mddev->max_disks &&
2885 (rdev->desc_nr >= mddev->max_disks ||
2886 i > mddev->max_disks)) {
de01dfad
N		 2887 printk(KERN_WARNING
2888 "md: %s: %s: only %d devices permitted\n",
2889 mdname(mddev), bdevname(rdev->bdev, b),
2890 mddev->max_disks);
2891 kick_rdev_from_array(rdev);
2892 continue;
2893 }
1da177e4
LT		 2894 if (rdev != freshest)
2895 if (super_types[mddev->major_version].
2896 validate_super(mddev, rdev)) {
2897 printk(KERN_WARNING "md: kicking non-fresh %s"
2898 " from array!\n",
2899 bdevname(rdev->bdev,b));
2900 kick_rdev_from_array(rdev);
2901 continue;
2902 }
2903 if (mddev->level == LEVEL_MULTIPATH) {
2904 rdev->desc_nr = i++;
2905 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2906 set_bit(In_sync, &rdev->flags);
5e5e3e78 2907 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 2908 rdev->raid_disk = -1;
2909 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2910 }
2911 }
1da177e4
LT		 2912}
2913
72e02075
N		 2914/* Read a fixed-point number.
2915 * Numbers in sysfs attributes should be in "standard" units where
2916 * possible, so time should be in seconds.
2917 * However we internally use a a much smaller unit such as
2918 * milliseconds or jiffies.
2919 * This function takes a decimal number with a possible fractional
2920 * component, and produces an integer which is the result of
2921 * multiplying that number by 10^'scale'.
2922 * all without any floating-point arithmetic.
2923 */
2924int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2925{
2926 unsigned long result = 0;
2927 long decimals = -1;
2928 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2929 if (*cp == '.')
2930 decimals = 0;
2931 else if (decimals < scale) {
2932 unsigned int value;
2933 value = *cp - '0';
2934 result = result * 10 + value;
2935 if (decimals >= 0)
2936 decimals++;
2937 }
2938 cp++;
2939 }
2940 if (*cp == '\n')
2941 cp++;
2942 if (*cp)
2943 return -EINVAL;
2944 if (decimals < 0)
2945 decimals = 0;
2946 while (decimals < scale) {
2947 result *= 10;
2948 decimals ++;
2949 }
2950 *res = result;
2951 return 0;
2952}
2953
2954
19052c0e
N		 2955static void md_safemode_timeout(unsigned long data);
2956
16f17b39
N		 2957static ssize_t
2958safe_delay_show(mddev_t *mddev, char *page)
2959{
2960 int msec = (mddev->safemode_delay*1000)/HZ;
2961 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2962}
2963static ssize_t
2964safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2965{
16f17b39 2966 unsigned long msec;
97ce0a7f 2967
72e02075 2968 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 2969 return -EINVAL;
16f17b39
N		 2970 if (msec == 0)
2971 mddev->safemode_delay = 0;
2972 else {
19052c0e 2973 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 2974 mddev->safemode_delay = (msec*HZ)/1000;
2975 if (mddev->safemode_delay == 0)
2976 mddev->safemode_delay = 1;
19052c0e
N		 2977 if (mddev->safemode_delay < old_delay)
2978 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 2979 }
2980 return len;
2981}
2982static struct md_sysfs_entry md_safe_delay =
80ca3a44 2983__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2984
eae1701f 2985static ssize_t
96de1e66 2986level_show(mddev_t *mddev, char *page)
eae1701f 2987{
2604b703 2988 struct mdk_personality *p = mddev->pers;
d9d166c2 2989 if (p)
eae1701f 2990 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2991 else if (mddev->clevel[0])
2992 return sprintf(page, "%s\n", mddev->clevel);
2993 else if (mddev->level != LEVEL_NONE)
2994 return sprintf(page, "%d\n", mddev->level);
2995 else
2996 return 0;
eae1701f
N		 2997}
2998
d9d166c2
N		 2999static ssize_t
3000level_store(mddev_t *mddev, const char *buf, size_t len)
3001{
f2859af6 3002 char clevel[16];
20a49ff6 3003 ssize_t rv = len;
245f46c2 3004 struct mdk_personality *pers;
f2859af6 3005 long level;
245f46c2 3006 void *priv;
3a981b03 3007 mdk_rdev_t *rdev;
245f46c2
N		 3008
3009 if (mddev->pers == NULL) {
3010 if (len == 0)
3011 return 0;
3012 if (len >= sizeof(mddev->clevel))
3013 return -ENOSPC;
3014 strncpy(mddev->clevel, buf, len);
3015 if (mddev->clevel[len-1] == '\n')
3016 len--;
3017 mddev->clevel[len] = 0;
3018 mddev->level = LEVEL_NONE;
3019 return rv;
3020 }
3021
3022 /* request to change the personality. Need to ensure:
3023 * - array is not engaged in resync/recovery/reshape
3024 * - old personality can be suspended
3025 * - new personality will access other array.
3026 */
3027
bb4f1e9d
N		 3028 if (mddev->sync_thread ||
3029 mddev->reshape_position != MaxSector ||
3030 mddev->sysfs_active)
d9d166c2 3031 return -EBUSY;
245f46c2
N		 3032
3033 if (!mddev->pers->quiesce) {
3034 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3035 mdname(mddev), mddev->pers->name);
3036 return -EINVAL;
3037 }
3038
3039 /* Now find the new personality */
f2859af6 3040 if (len == 0 || len >= sizeof(clevel))
245f46c2 3041 return -EINVAL;
f2859af6
DW		 3042 strncpy(clevel, buf, len);
3043 if (clevel[len-1] == '\n')
d9d166c2 3044 len--;
f2859af6
DW		 3045 clevel[len] = 0;
3046 if (strict_strtol(clevel, 10, &level))
3047 level = LEVEL_NONE;
245f46c2 3048
f2859af6
DW		 3049 if (request_module("md-%s", clevel) != 0)
3050 request_module("md-level-%s", clevel);
245f46c2 3051 spin_lock(&pers_lock);
f2859af6 3052 pers = find_pers(level, clevel);
245f46c2
N		 3053 if (!pers || !try_module_get(pers->owner)) {
3054 spin_unlock(&pers_lock);
f2859af6 3055 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3056 return -EINVAL;
3057 }
3058 spin_unlock(&pers_lock);
3059
3060 if (pers == mddev->pers) {
3061 /* Nothing to do! */
3062 module_put(pers->owner);
3063 return rv;
3064 }
3065 if (!pers->takeover) {
3066 module_put(pers->owner);
3067 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3068 mdname(mddev), clevel);
245f46c2
N		 3069 return -EINVAL;
3070 }
3071
e93f68a1
N		 3072 list_for_each_entry(rdev, &mddev->disks, same_set)
3073 rdev->new_raid_disk = rdev->raid_disk;
3074
245f46c2
N		 3075 /* ->takeover must set new_* and/or delta_disks
3076 * if it succeeds, and may set them when it fails.
3077 */
3078 priv = pers->takeover(mddev);
3079 if (IS_ERR(priv)) {
3080 mddev->new_level = mddev->level;
3081 mddev->new_layout = mddev->layout;
664e7c41 3082 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3083 mddev->raid_disks -= mddev->delta_disks;
3084 mddev->delta_disks = 0;
3085 module_put(pers->owner);
3086 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3087 mdname(mddev), clevel);
245f46c2
N		 3088 return PTR_ERR(priv);
3089 }
3090
3091 /* Looks like we have a winner */
3092 mddev_suspend(mddev);
3093 mddev->pers->stop(mddev);
a64c876f
N		 3094
3095 if (mddev->pers->sync_request == NULL &&
3096 pers->sync_request != NULL) {
3097 /* need to add the md_redundancy_group */
3098 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3099 printk(KERN_WARNING
3100 "md: cannot register extra attributes for %s\n",
3101 mdname(mddev));
19fdb9ee 3102 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3103 }
3104 if (mddev->pers->sync_request != NULL &&
3105 pers->sync_request == NULL) {
3106 /* need to remove the md_redundancy_group */
3107 if (mddev->to_remove == NULL)
3108 mddev->to_remove = &md_redundancy_group;
3109 }
3110
54071b38
TM		 3111 if (mddev->pers->sync_request == NULL &&
3112 mddev->external) {
3113 /* We are converting from a no-redundancy array
3114 * to a redundancy array and metadata is managed
3115 * externally so we need to be sure that writes
3116 * won't block due to a need to transition
3117 * clean->dirty
3118 * until external management is started.
3119 */
3120 mddev->in_sync = 0;
3121 mddev->safemode_delay = 0;
3122 mddev->safemode = 0;
3123 }
3124
e93f68a1
N		 3125 list_for_each_entry(rdev, &mddev->disks, same_set) {
3126 char nm[20];
3127 if (rdev->raid_disk < 0)
3128 continue;
3129 if (rdev->new_raid_disk > mddev->raid_disks)
3130 rdev->new_raid_disk = -1;
3131 if (rdev->new_raid_disk == rdev->raid_disk)
3132 continue;
3133 sprintf(nm, "rd%d", rdev->raid_disk);
3134 sysfs_remove_link(&mddev->kobj, nm);
3135 }
3136 list_for_each_entry(rdev, &mddev->disks, same_set) {
3137 if (rdev->raid_disk < 0)
3138 continue;
3139 if (rdev->new_raid_disk == rdev->raid_disk)
3140 continue;
3141 rdev->raid_disk = rdev->new_raid_disk;
3142 if (rdev->raid_disk < 0)
3a981b03 3143 clear_bit(In_sync, &rdev->flags);
e93f68a1
N		 3144 else {
3145 char nm[20];
3146 sprintf(nm, "rd%d", rdev->raid_disk);
3147 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3148 printk("md: cannot register %s for %s after level change\n",
3149 nm, mdname(mddev));
3a981b03 3150 }
e93f68a1
N		 3151 }
3152
3153 module_put(mddev->pers->owner);
245f46c2
N		 3154 mddev->pers = pers;
3155 mddev->private = priv;
3156 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3157 mddev->level = mddev->new_level;
3158 mddev->layout = mddev->new_layout;
664e7c41 3159 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3160 mddev->delta_disks = 0;
9af204cf
TM		 3161 if (mddev->pers->sync_request == NULL) {
3162 /* this is now an array without redundancy, so
3163 * it must always be in_sync
3164 */
3165 mddev->in_sync = 1;
3166 del_timer_sync(&mddev->safemode_timer);
3167 }
245f46c2
N		 3168 pers->run(mddev);
3169 mddev_resume(mddev);
3170 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3171 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3172 md_wakeup_thread(mddev->thread);
5cac7861 3173 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3174 md_new_event(mddev);
d9d166c2
N		 3175 return rv;
3176}
3177
3178static struct md_sysfs_entry md_level =
80ca3a44 3179__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3180
d4dbd025
N		 3181
3182static ssize_t
3183layout_show(mddev_t *mddev, char *page)
3184{
3185 /* just a number, not meaningful for all levels */
08a02ecd
N		 3186 if (mddev->reshape_position != MaxSector &&
3187 mddev->layout != mddev->new_layout)
3188 return sprintf(page, "%d (%d)\n",
3189 mddev->new_layout, mddev->layout);
d4dbd025
N		 3190 return sprintf(page, "%d\n", mddev->layout);
3191}
3192
3193static ssize_t
3194layout_store(mddev_t *mddev, const char *buf, size_t len)
3195{
3196 char *e;
3197 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3198
3199 if (!*buf || (*e && *e != '\n'))
3200 return -EINVAL;
3201
b3546035
N		 3202 if (mddev->pers) {
3203 int err;
50ac168a 3204 if (mddev->pers->check_reshape == NULL)
b3546035 3205 return -EBUSY;
597a711b 3206 mddev->new_layout = n;
50ac168a 3207 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3208 if (err) {
3209 mddev->new_layout = mddev->layout;
b3546035 3210 return err;
597a711b 3211 }
b3546035 3212 } else {
08a02ecd 3213 mddev->new_layout = n;
b3546035
N		 3214 if (mddev->reshape_position == MaxSector)
3215 mddev->layout = n;
3216 }
d4dbd025
N		 3217 return len;
3218}
3219static struct md_sysfs_entry md_layout =
80ca3a44 3220__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3221
3222
eae1701f 3223static ssize_t
96de1e66 3224raid_disks_show(mddev_t *mddev, char *page)
eae1701f 3225{
bb636547
N		 3226 if (mddev->raid_disks == 0)
3227 return 0;
08a02ecd
N		 3228 if (mddev->reshape_position != MaxSector &&
3229 mddev->delta_disks != 0)
3230 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3231 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3232 return sprintf(page, "%d\n", mddev->raid_disks);
3233}
3234
da943b99
N		 3235static int update_raid_disks(mddev_t *mddev, int raid_disks);
3236
3237static ssize_t
3238raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3239{
da943b99
N		 3240 char *e;
3241 int rv = 0;
3242 unsigned long n = simple_strtoul(buf, &e, 10);
3243
3244 if (!*buf || (*e && *e != '\n'))
3245 return -EINVAL;
3246
3247 if (mddev->pers)
3248 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3249 else if (mddev->reshape_position != MaxSector) {
3250 int olddisks = mddev->raid_disks - mddev->delta_disks;
3251 mddev->delta_disks = n - olddisks;
3252 mddev->raid_disks = n;
3253 } else
da943b99
N		 3254 mddev->raid_disks = n;
3255 return rv ? rv : len;
3256}
3257static struct md_sysfs_entry md_raid_disks =
80ca3a44 3258__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3259
3b34380a
N		 3260static ssize_t
3261chunk_size_show(mddev_t *mddev, char *page)
3262{
08a02ecd 3263 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3264 mddev->chunk_sectors != mddev->new_chunk_sectors)
3265 return sprintf(page, "%d (%d)\n",
3266 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3267 mddev->chunk_sectors << 9);
3268 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3269}
3270
3271static ssize_t
3272chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3273{
3b34380a
N		 3274 char *e;
3275 unsigned long n = simple_strtoul(buf, &e, 10);
3276
3b34380a
N		 3277 if (!*buf || (*e && *e != '\n'))
3278 return -EINVAL;
3279
b3546035
N		 3280 if (mddev->pers) {
3281 int err;
50ac168a 3282 if (mddev->pers->check_reshape == NULL)
b3546035 3283 return -EBUSY;
597a711b 3284 mddev->new_chunk_sectors = n >> 9;
50ac168a 3285 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3286 if (err) {
3287 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3288 return err;
597a711b 3289 }
b3546035 3290 } else {
664e7c41 3291 mddev->new_chunk_sectors = n >> 9;
b3546035 3292 if (mddev->reshape_position == MaxSector)
9d8f0363 3293 mddev->chunk_sectors = n >> 9;
b3546035 3294 }
3b34380a
N		 3295 return len;
3296}
3297static struct md_sysfs_entry md_chunk_size =
80ca3a44 3298__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3299
a94213b1
N		 3300static ssize_t
3301resync_start_show(mddev_t *mddev, char *page)
3302{
d1a7c503
N		 3303 if (mddev->recovery_cp == MaxSector)
3304 return sprintf(page, "none\n");
a94213b1
N		 3305 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3306}
3307
3308static ssize_t
3309resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3310{
a94213b1
N		 3311 char *e;
3312 unsigned long long n = simple_strtoull(buf, &e, 10);
3313
3314 if (mddev->pers)
3315 return -EBUSY;
06e3c817
DW		 3316 if (cmd_match(buf, "none"))
3317 n = MaxSector;
3318 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3319 return -EINVAL;
3320
3321 mddev->recovery_cp = n;
3322 return len;
3323}
3324static struct md_sysfs_entry md_resync_start =
80ca3a44 3325__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3326
9e653b63
N		 3327/*
3328 * The array state can be:
3329 *
3330 * clear
3331 * No devices, no size, no level
3332 * Equivalent to STOP_ARRAY ioctl
3333 * inactive
3334 * May have some settings, but array is not active
3335 * all IO results in error
3336 * When written, doesn't tear down array, but just stops it
3337 * suspended (not supported yet)
3338 * All IO requests will block. The array can be reconfigured.
910d8cb3 3339 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3340 * readonly
3341 * no resync can happen. no superblocks get written.
3342 * write requests fail
3343 * read-auto
3344 * like readonly, but behaves like 'clean' on a write request.
3345 *
3346 * clean - no pending writes, but otherwise active.
3347 * When written to inactive array, starts without resync
3348 * If a write request arrives then
3349 * if metadata is known, mark 'dirty' and switch to 'active'.
3350 * if not known, block and switch to write-pending
3351 * If written to an active array that has pending writes, then fails.
3352 * active
3353 * fully active: IO and resync can be happening.
3354 * When written to inactive array, starts with resync
3355 *
3356 * write-pending
3357 * clean, but writes are blocked waiting for 'active' to be written.
3358 *
3359 * active-idle
3360 * like active, but no writes have been seen for a while (100msec).
3361 *
3362 */
3363enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3364 write_pending, active_idle, bad_word};
05381954 3365static char *array_states[] = {
9e653b63
N		 3366 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3367 "write-pending", "active-idle", NULL };
3368
3369static int match_word(const char *word, char **list)
3370{
3371 int n;
3372 for (n=0; list[n]; n++)
3373 if (cmd_match(word, list[n]))
3374 break;
3375 return n;
3376}
3377
3378static ssize_t
3379array_state_show(mddev_t *mddev, char *page)
3380{
3381 enum array_state st = inactive;
3382
3383 if (mddev->pers)
3384 switch(mddev->ro) {
3385 case 1:
3386 st = readonly;
3387 break;
3388 case 2:
3389 st = read_auto;
3390 break;
3391 case 0:
3392 if (mddev->in_sync)
3393 st = clean;
070dc6dd 3394 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3395 st = write_pending;
9e653b63
N		 3396 else if (mddev->safemode)
3397 st = active_idle;
3398 else
3399 st = active;
3400 }
3401 else {
3402 if (list_empty(&mddev->disks) &&
3403 mddev->raid_disks == 0 &&
58c0fed4 3404 mddev->dev_sectors == 0)
9e653b63
N		 3405 st = clear;
3406 else
3407 st = inactive;
3408 }
3409 return sprintf(page, "%s\n", array_states[st]);
3410}
3411
df5b20cf 3412static int do_md_stop(mddev_t * mddev, int ro, int is_open);
a4bd82d0 3413static int md_set_readonly(mddev_t * mddev, int is_open);
9e653b63
N		 3414static int do_md_run(mddev_t * mddev);
3415static int restart_array(mddev_t *mddev);
3416
3417static ssize_t
3418array_state_store(mddev_t *mddev, const char *buf, size_t len)
3419{
3420 int err = -EINVAL;
3421 enum array_state st = match_word(buf, array_states);
3422 switch(st) {
3423 case bad_word:
3424 break;
3425 case clear:
3426 /* stopping an active array */
f2ea68cf 3427 if (atomic_read(&mddev->openers) > 0)
e691063a 3428 return -EBUSY;
df5b20cf 3429 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3430 break;
3431 case inactive:
3432 /* stopping an active array */
3433 if (mddev->pers) {
f2ea68cf 3434 if (atomic_read(&mddev->openers) > 0)
9e653b63 3435 return -EBUSY;
df5b20cf 3436 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3437 } else
3438 err = 0; /* already inactive */
9e653b63
N		 3439 break;
3440 case suspended:
3441 break; /* not supported yet */
3442 case readonly:
3443 if (mddev->pers)
a4bd82d0 3444 err = md_set_readonly(mddev, 0);
9e653b63
N		 3445 else {
3446 mddev->ro = 1;
648b629e 3447 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3448 err = do_md_run(mddev);
3449 }
3450 break;
3451 case read_auto:
9e653b63 3452 if (mddev->pers) {
80268ee9 3453 if (mddev->ro == 0)
a4bd82d0 3454 err = md_set_readonly(mddev, 0);
80268ee9 3455 else if (mddev->ro == 1)
648b629e
N		 3456 err = restart_array(mddev);
3457 if (err == 0) {
3458 mddev->ro = 2;
3459 set_disk_ro(mddev->gendisk, 0);
3460 }
9e653b63
N		 3461 } else {
3462 mddev->ro = 2;
3463 err = do_md_run(mddev);
3464 }
3465 break;
3466 case clean:
3467 if (mddev->pers) {
3468 restart_array(mddev);
3469 spin_lock_irq(&mddev->write_lock);
3470 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3471 if (mddev->in_sync == 0) {
3472 mddev->in_sync = 1;
31a59e34
N		 3473 if (mddev->safemode == 1)
3474 mddev->safemode = 0;
070dc6dd 3475 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3476 }
3477 err = 0;
3478 } else
3479 err = -EBUSY;
9e653b63 3480 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3481 } else
3482 err = -EINVAL;
9e653b63
N		 3483 break;
3484 case active:
3485 if (mddev->pers) {
3486 restart_array(mddev);
070dc6dd 3487 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3488 wake_up(&mddev->sb_wait);
3489 err = 0;
3490 } else {
3491 mddev->ro = 0;
648b629e 3492 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3493 err = do_md_run(mddev);
3494 }
3495 break;
3496 case write_pending:
3497 case active_idle:
3498 /* these cannot be set */
3499 break;
3500 }
3501 if (err)
3502 return err;
0fd62b86 3503 else {
00bcb4ac 3504 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3505 return len;
0fd62b86 3506 }
9e653b63 3507}
80ca3a44
N		 3508static struct md_sysfs_entry md_array_state =
3509__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3510
1e50915f
RB		 3511static ssize_t
3512max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3513 return sprintf(page, "%d\n",
3514 atomic_read(&mddev->max_corr_read_errors));
3515}
3516
3517static ssize_t
3518max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3519{
3520 char *e;
3521 unsigned long n = simple_strtoul(buf, &e, 10);
3522
3523 if (*buf && (*e == 0 || *e == '\n')) {
3524 atomic_set(&mddev->max_corr_read_errors, n);
3525 return len;
3526 }
3527 return -EINVAL;
3528}
3529
3530static struct md_sysfs_entry max_corr_read_errors =
3531__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3532 max_corrected_read_errors_store);
3533
6d7ff738
N		 3534static ssize_t
3535null_show(mddev_t *mddev, char *page)
3536{
3537 return -EINVAL;
3538}
3539
3540static ssize_t
3541new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3542{
3543 /* buf must be %d:%d\n? giving major and minor numbers */
3544 /* The new device is added to the array.
3545 * If the array has a persistent superblock, we read the
3546 * superblock to initialise info and check validity.
3547 * Otherwise, only checking done is that in bind_rdev_to_array,
3548 * which mainly checks size.
3549 */
3550 char *e;
3551 int major = simple_strtoul(buf, &e, 10);
3552 int minor;
3553 dev_t dev;
3554 mdk_rdev_t *rdev;
3555 int err;
3556
3557 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3558 return -EINVAL;
3559 minor = simple_strtoul(e+1, &e, 10);
3560 if (*e && *e != '\n')
3561 return -EINVAL;
3562 dev = MKDEV(major, minor);
3563 if (major != MAJOR(dev) ||
3564 minor != MINOR(dev))
3565 return -EOVERFLOW;
3566
3567
3568 if (mddev->persistent) {
3569 rdev = md_import_device(dev, mddev->major_version,
3570 mddev->minor_version);
3571 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3572 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3573 mdk_rdev_t, same_set);
3574 err = super_types[mddev->major_version]
3575 .load_super(rdev, rdev0, mddev->minor_version);
3576 if (err < 0)
3577 goto out;
3578 }
c5d79adb
N		 3579 } else if (mddev->external)
3580 rdev = md_import_device(dev, -2, -1);
3581 else
6d7ff738
N		 3582 rdev = md_import_device(dev, -1, -1);
3583
3584 if (IS_ERR(rdev))
3585 return PTR_ERR(rdev);
3586 err = bind_rdev_to_array(rdev, mddev);
3587 out:
3588 if (err)
3589 export_rdev(rdev);
3590 return err ? err : len;
3591}
3592
3593static struct md_sysfs_entry md_new_device =
80ca3a44 3594__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3595
9b1d1dac
PC		 3596static ssize_t
3597bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3598{
3599 char *end;
3600 unsigned long chunk, end_chunk;
3601
3602 if (!mddev->bitmap)
3603 goto out;
3604 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3605 while (*buf) {
3606 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3607 if (buf == end) break;
3608 if (*end == '-') { /* range */
3609 buf = end + 1;
3610 end_chunk = simple_strtoul(buf, &end, 0);
3611 if (buf == end) break;
3612 }
3613 if (*end && !isspace(*end)) break;
3614 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3615 buf = skip_spaces(end);
9b1d1dac
PC		 3616 }
3617 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3618out:
3619 return len;
3620}
3621
3622static struct md_sysfs_entry md_bitmap =
3623__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3624
a35b0d69
N		 3625static ssize_t
3626size_show(mddev_t *mddev, char *page)
3627{
58c0fed4
AN		 3628 return sprintf(page, "%llu\n",
3629 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3630}
3631
d71f9f88 3632static int update_size(mddev_t *mddev, sector_t num_sectors);
a35b0d69
N		 3633
3634static ssize_t
3635size_store(mddev_t *mddev, const char *buf, size_t len)
3636{
3637 /* If array is inactive, we can reduce the component size, but
3638 * not increase it (except from 0).
3639 * If array is active, we can try an on-line resize
3640 */
b522adcd
DW		 3641 sector_t sectors;
3642 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3643
58c0fed4
AN		 3644 if (err < 0)
3645 return err;
a35b0d69 3646 if (mddev->pers) {
58c0fed4 3647 err = update_size(mddev, sectors);
850b2b42 3648 md_update_sb(mddev, 1);
a35b0d69 3649 } else {
58c0fed4
AN		 3650 if (mddev->dev_sectors == 0 ||
3651 mddev->dev_sectors > sectors)
3652 mddev->dev_sectors = sectors;
a35b0d69
N		 3653 else
3654 err = -ENOSPC;
3655 }
3656 return err ? err : len;
3657}
3658
3659static struct md_sysfs_entry md_size =
80ca3a44 3660__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3661
8bb93aac
N		 3662
3663/* Metdata version.
e691063a
N		 3664 * This is one of
3665 * 'none' for arrays with no metadata (good luck...)
3666 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3667 * or N.M for internally known formats
3668 */
3669static ssize_t
3670metadata_show(mddev_t *mddev, char *page)
3671{
3672 if (mddev->persistent)
3673 return sprintf(page, "%d.%d\n",
3674 mddev->major_version, mddev->minor_version);
e691063a
N		 3675 else if (mddev->external)
3676 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3677 else
3678 return sprintf(page, "none\n");
3679}
3680
3681static ssize_t
3682metadata_store(mddev_t *mddev, const char *buf, size_t len)
3683{
3684 int major, minor;
3685 char *e;
ea43ddd8
N		 3686 /* Changing the details of 'external' metadata is
3687 * always permitted. Otherwise there must be
3688 * no devices attached to the array.
3689 */
3690 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3691 ;
3692 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3693 return -EBUSY;
3694
3695 if (cmd_match(buf, "none")) {
3696 mddev->persistent = 0;
e691063a
N		 3697 mddev->external = 0;
3698 mddev->major_version = 0;
3699 mddev->minor_version = 90;
3700 return len;
3701 }
3702 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3703 size_t namelen = len-9;
e691063a
N		 3704 if (namelen >= sizeof(mddev->metadata_type))
3705 namelen = sizeof(mddev->metadata_type)-1;
3706 strncpy(mddev->metadata_type, buf+9, namelen);
3707 mddev->metadata_type[namelen] = 0;
3708 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3709 mddev->metadata_type[--namelen] = 0;
3710 mddev->persistent = 0;
3711 mddev->external = 1;
8bb93aac
N		 3712 mddev->major_version = 0;
3713 mddev->minor_version = 90;
3714 return len;
3715 }
3716 major = simple_strtoul(buf, &e, 10);
3717 if (e==buf || *e != '.')
3718 return -EINVAL;
3719 buf = e+1;
3720 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 3721 if (e==buf || (*e && *e != '\n') )
8bb93aac 3722 return -EINVAL;
50511da3 3723 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 3724 return -ENOENT;
3725 mddev->major_version = major;
3726 mddev->minor_version = minor;
3727 mddev->persistent = 1;
e691063a 3728 mddev->external = 0;
8bb93aac
N		 3729 return len;
3730}
3731
3732static struct md_sysfs_entry md_metadata =
80ca3a44 3733__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 3734
24dd469d 3735static ssize_t
7eec314d 3736action_show(mddev_t *mddev, char *page)
24dd469d 3737{
7eec314d 3738 char *type = "idle";
b6a9ce68
N		 3739 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3740 type = "frozen";
3741 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 3742 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 3743 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3744 type = "reshape";
3745 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 3746 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3747 type = "resync";
3748 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3749 type = "check";
3750 else
3751 type = "repair";
72a23c21 3752 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 3753 type = "recover";
3754 }
3755 return sprintf(page, "%s\n", type);
3756}
3757
7ebc0be7
N		 3758static void reap_sync_thread(mddev_t *mddev);
3759
24dd469d 3760static ssize_t
7eec314d 3761action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 3762{
7eec314d
N		 3763 if (!mddev->pers || !mddev->pers->sync_request)
3764 return -EINVAL;
3765
b6a9ce68
N		 3766 if (cmd_match(page, "frozen"))
3767 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3768 else
3769 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3770
3771 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 3772 if (mddev->sync_thread) {
3773 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 3774 reap_sync_thread(mddev);
7eec314d 3775 }
03c902e1
N		 3776 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3777 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 3778 return -EBUSY;
72a23c21
NB		 3779 else if (cmd_match(page, "resync"))
3780 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3781 else if (cmd_match(page, "recover")) {
3782 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 3783 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 3784 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 3785 int err;
3786 if (mddev->pers->start_reshape == NULL)
3787 return -EINVAL;
3788 err = mddev->pers->start_reshape(mddev);
3789 if (err)
3790 return err;
a99ac971 3791 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 3792 } else {
bce74dac 3793 if (cmd_match(page, "check"))
7eec314d 3794 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 3795 else if (!cmd_match(page, "repair"))
7eec314d
N		 3796 return -EINVAL;
3797 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3798 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 3799 }
03c902e1 3800 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 3801 md_wakeup_thread(mddev->thread);
00bcb4ac 3802 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 3803 return len;
3804}
3805
9d88883e 3806static ssize_t
96de1e66 3807mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 3808{
3809 return sprintf(page, "%llu\n",
3810 (unsigned long long) mddev->resync_mismatches);
3811}
3812
80ca3a44
N		 3813static struct md_sysfs_entry md_scan_mode =
3814__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 3815
96de1e66 3816
80ca3a44 3817static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 3818
88202a0c
N		 3819static ssize_t
3820sync_min_show(mddev_t *mddev, char *page)
3821{
3822 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3823 mddev->sync_speed_min ? "local": "system");
3824}
3825
3826static ssize_t
3827sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3828{
3829 int min;
3830 char *e;
3831 if (strncmp(buf, "system", 6)==0) {
3832 mddev->sync_speed_min = 0;
3833 return len;
3834 }
3835 min = simple_strtoul(buf, &e, 10);
3836 if (buf == e || (*e && *e != '\n') || min <= 0)
3837 return -EINVAL;
3838 mddev->sync_speed_min = min;
3839 return len;
3840}
3841
3842static struct md_sysfs_entry md_sync_min =
3843__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3844
3845static ssize_t
3846sync_max_show(mddev_t *mddev, char *page)
3847{
3848 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3849 mddev->sync_speed_max ? "local": "system");
3850}
3851
3852static ssize_t
3853sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3854{
3855 int max;
3856 char *e;
3857 if (strncmp(buf, "system", 6)==0) {
3858 mddev->sync_speed_max = 0;
3859 return len;
3860 }
3861 max = simple_strtoul(buf, &e, 10);
3862 if (buf == e || (*e && *e != '\n') || max <= 0)
3863 return -EINVAL;
3864 mddev->sync_speed_max = max;
3865 return len;
3866}
3867
3868static struct md_sysfs_entry md_sync_max =
3869__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3870
d7f3d291
IP		 3871static ssize_t
3872degraded_show(mddev_t *mddev, char *page)
3873{
3874 return sprintf(page, "%d\n", mddev->degraded);
3875}
3876static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 3877
90b08710
BS		 3878static ssize_t
3879sync_force_parallel_show(mddev_t *mddev, char *page)
3880{
3881 return sprintf(page, "%d\n", mddev->parallel_resync);
3882}
3883
3884static ssize_t
3885sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3886{
3887 long n;
3888
3889 if (strict_strtol(buf, 10, &n))
3890 return -EINVAL;
3891
3892 if (n != 0 && n != 1)
3893 return -EINVAL;
3894
3895 mddev->parallel_resync = n;
3896
3897 if (mddev->sync_thread)
3898 wake_up(&resync_wait);
3899
3900 return len;
3901}
3902
3903/* force parallel resync, even with shared block devices */
3904static struct md_sysfs_entry md_sync_force_parallel =
3905__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3906 sync_force_parallel_show, sync_force_parallel_store);
3907
88202a0c
N		 3908static ssize_t
3909sync_speed_show(mddev_t *mddev, char *page)
3910{
3911 unsigned long resync, dt, db;
d1a7c503
N		 3912 if (mddev->curr_resync == 0)
3913 return sprintf(page, "none\n");
9687a60c
AN		 3914 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3915 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 3916 if (!dt) dt++;
9687a60c
AN		 3917 db = resync - mddev->resync_mark_cnt;
3918 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 3919}
3920
80ca3a44 3921static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 3922
3923static ssize_t
3924sync_completed_show(mddev_t *mddev, char *page)
3925{
13ae864b 3926 unsigned long long max_sectors, resync;
88202a0c 3927
acb180b0
N		 3928 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3929 return sprintf(page, "none\n");
3930
88202a0c 3931 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 3932 max_sectors = mddev->resync_max_sectors;
88202a0c 3933 else
58c0fed4 3934 max_sectors = mddev->dev_sectors;
88202a0c 3935
acb180b0 3936 resync = mddev->curr_resync_completed;
13ae864b 3937 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 3938}
3939
80ca3a44 3940static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3941
5e96ee65
NB		 3942static ssize_t
3943min_sync_show(mddev_t *mddev, char *page)
3944{
3945 return sprintf(page, "%llu\n",
3946 (unsigned long long)mddev->resync_min);
3947}
3948static ssize_t
3949min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3950{
3951 unsigned long long min;
3952 if (strict_strtoull(buf, 10, &min))
3953 return -EINVAL;
3954 if (min > mddev->resync_max)
3955 return -EINVAL;
3956 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3957 return -EBUSY;
3958
3959 /* Must be a multiple of chunk_size */
9d8f0363 3960 if (mddev->chunk_sectors) {
2ac06c33 3961 sector_t temp = min;
9d8f0363 3962 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 3963 return -EINVAL;
3964 }
3965 mddev->resync_min = min;
3966
3967 return len;
3968}
3969
3970static struct md_sysfs_entry md_min_sync =
3971__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3972
c6207277
N		 3973static ssize_t
3974max_sync_show(mddev_t *mddev, char *page)
3975{
3976 if (mddev->resync_max == MaxSector)
3977 return sprintf(page, "max\n");
3978 else
3979 return sprintf(page, "%llu\n",
3980 (unsigned long long)mddev->resync_max);
3981}
3982static ssize_t
3983max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3984{
3985 if (strncmp(buf, "max", 3) == 0)
3986 mddev->resync_max = MaxSector;
3987 else {
5e96ee65
NB		 3988 unsigned long long max;
3989 if (strict_strtoull(buf, 10, &max))
3990 return -EINVAL;
3991 if (max < mddev->resync_min)
c6207277
N		 3992 return -EINVAL;
3993 if (max < mddev->resync_max &&
4d484a4a 3994 mddev->ro == 0 &&
c6207277
N		 3995 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3996 return -EBUSY;
3997
3998 /* Must be a multiple of chunk_size */
9d8f0363 3999 if (mddev->chunk_sectors) {
2ac06c33 4000 sector_t temp = max;
9d8f0363 4001 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4002 return -EINVAL;
4003 }
4004 mddev->resync_max = max;
4005 }
4006 wake_up(&mddev->recovery_wait);
4007 return len;
4008}
4009
4010static struct md_sysfs_entry md_max_sync =
4011__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4012
e464eafd
N		 4013static ssize_t
4014suspend_lo_show(mddev_t *mddev, char *page)
4015{
4016 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4017}
4018
4019static ssize_t
4020suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4021{
4022 char *e;
4023 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4024 unsigned long long old = mddev->suspend_lo;
e464eafd 4025
b8d966ef
N		 4026 if (mddev->pers == NULL ||
4027 mddev->pers->quiesce == NULL)
e464eafd
N		 4028 return -EINVAL;
4029 if (buf == e || (*e && *e != '\n'))
4030 return -EINVAL;
23ddff37
N		 4031
4032 mddev->suspend_lo = new;
4033 if (new >= old)
4034 /* Shrinking suspended region */
e464eafd 4035 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4036 else {
4037 /* Expanding suspended region - need to wait */
4038 mddev->pers->quiesce(mddev, 1);
4039 mddev->pers->quiesce(mddev, 0);
4040 }
4041 return len;
e464eafd
N		 4042}
4043static struct md_sysfs_entry md_suspend_lo =
4044__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4045
4046
4047static ssize_t
4048suspend_hi_show(mddev_t *mddev, char *page)
4049{
4050 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4051}
4052
4053static ssize_t
4054suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4055{
4056 char *e;
4057 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4058 unsigned long long old = mddev->suspend_hi;
e464eafd 4059
b8d966ef
N		 4060 if (mddev->pers == NULL ||
4061 mddev->pers->quiesce == NULL)
e464eafd
N		 4062 return -EINVAL;
4063 if (buf == e || (*e && *e != '\n'))
4064 return -EINVAL;
23ddff37
N		 4065
4066 mddev->suspend_hi = new;
4067 if (new <= old)
4068 /* Shrinking suspended region */
4069 mddev->pers->quiesce(mddev, 2);
4070 else {
4071 /* Expanding suspended region - need to wait */
e464eafd
N		 4072 mddev->pers->quiesce(mddev, 1);
4073 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4074 }
4075 return len;
e464eafd
N		 4076}
4077static struct md_sysfs_entry md_suspend_hi =
4078__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4079
08a02ecd
N		 4080static ssize_t
4081reshape_position_show(mddev_t *mddev, char *page)
4082{
4083 if (mddev->reshape_position != MaxSector)
4084 return sprintf(page, "%llu\n",
4085 (unsigned long long)mddev->reshape_position);
4086 strcpy(page, "none\n");
4087 return 5;
4088}
4089
4090static ssize_t
4091reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4092{
4093 char *e;
4094 unsigned long long new = simple_strtoull(buf, &e, 10);
4095 if (mddev->pers)
4096 return -EBUSY;
4097 if (buf == e || (*e && *e != '\n'))
4098 return -EINVAL;
4099 mddev->reshape_position = new;
4100 mddev->delta_disks = 0;
4101 mddev->new_level = mddev->level;
4102 mddev->new_layout = mddev->layout;
664e7c41 4103 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4104 return len;
4105}
4106
4107static struct md_sysfs_entry md_reshape_position =
4108__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4109 reshape_position_store);
4110
b522adcd
DW		 4111static ssize_t
4112array_size_show(mddev_t *mddev, char *page)
4113{
4114 if (mddev->external_size)
4115 return sprintf(page, "%llu\n",
4116 (unsigned long long)mddev->array_sectors/2);
4117 else
4118 return sprintf(page, "default\n");
4119}
4120
4121static ssize_t
4122array_size_store(mddev_t *mddev, const char *buf, size_t len)
4123{
4124 sector_t sectors;
4125
4126 if (strncmp(buf, "default", 7) == 0) {
4127 if (mddev->pers)
4128 sectors = mddev->pers->size(mddev, 0, 0);
4129 else
4130 sectors = mddev->array_sectors;
4131
4132 mddev->external_size = 0;
4133 } else {
4134 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4135 return -EINVAL;
4136 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4137 return -E2BIG;
b522adcd
DW		 4138
4139 mddev->external_size = 1;
4140 }
4141
4142 mddev->array_sectors = sectors;
4143 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e
N		 4144 if (mddev->pers)
4145 revalidate_disk(mddev->gendisk);
b522adcd
DW		 4146
4147 return len;
4148}
4149
4150static struct md_sysfs_entry md_array_size =
4151__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4152 array_size_store);
e464eafd 4153
eae1701f
N		 4154static struct attribute *md_default_attrs[] = {
4155 &md_level.attr,
d4dbd025 4156 &md_layout.attr,
eae1701f 4157 &md_raid_disks.attr,
3b34380a 4158 &md_chunk_size.attr,
a35b0d69 4159 &md_size.attr,
a94213b1 4160 &md_resync_start.attr,
8bb93aac 4161 &md_metadata.attr,
6d7ff738 4162 &md_new_device.attr,
16f17b39 4163 &md_safe_delay.attr,
9e653b63 4164 &md_array_state.attr,
08a02ecd 4165 &md_reshape_position.attr,
b522adcd 4166 &md_array_size.attr,
1e50915f 4167 &max_corr_read_errors.attr,
411036fa
N		 4168 NULL,
4169};
4170
4171static struct attribute *md_redundancy_attrs[] = {
24dd469d 4172 &md_scan_mode.attr,
9d88883e 4173 &md_mismatches.attr,
88202a0c
N		 4174 &md_sync_min.attr,
4175 &md_sync_max.attr,
4176 &md_sync_speed.attr,
90b08710 4177 &md_sync_force_parallel.attr,
88202a0c 4178 &md_sync_completed.attr,
5e96ee65 4179 &md_min_sync.attr,
c6207277 4180 &md_max_sync.attr,
e464eafd
N		 4181 &md_suspend_lo.attr,
4182 &md_suspend_hi.attr,
9b1d1dac 4183 &md_bitmap.attr,
d7f3d291 4184 &md_degraded.attr,
eae1701f
N		 4185 NULL,
4186};
411036fa
N		 4187static struct attribute_group md_redundancy_group = {
4188 .name = NULL,
4189 .attrs = md_redundancy_attrs,
4190};
4191
eae1701f
N		 4192
4193static ssize_t
4194md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
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->show)
4201 return -EIO;
5dc5cf7d
IM		 4202 rv = mddev_lock(mddev);
4203 if (!rv) {
4204 rv = entry->show(mddev, page);
4205 mddev_unlock(mddev);
4206 }
96de1e66 4207 return rv;
eae1701f
N		 4208}
4209
4210static ssize_t
4211md_attr_store(struct kobject *kobj, struct attribute *attr,
4212 const char *page, size_t length)
4213{
4214 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4215 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4216 ssize_t rv;
eae1701f
N		 4217
4218 if (!entry->store)
4219 return -EIO;
67463acb
N		 4220 if (!capable(CAP_SYS_ADMIN))
4221 return -EACCES;
5dc5cf7d 4222 rv = mddev_lock(mddev);
d3374825
N		 4223 if (mddev->hold_active == UNTIL_IOCTL)
4224 mddev->hold_active = 0;
5dc5cf7d
IM		 4225 if (!rv) {
4226 rv = entry->store(mddev, page, length);
4227 mddev_unlock(mddev);
4228 }
96de1e66 4229 return rv;
eae1701f
N		 4230}
4231
4232static void md_free(struct kobject *ko)
4233{
4234 mddev_t *mddev = container_of(ko, mddev_t, kobj);
a21d1504
N		 4235
4236 if (mddev->sysfs_state)
4237 sysfs_put(mddev->sysfs_state);
4238
4239 if (mddev->gendisk) {
4240 del_gendisk(mddev->gendisk);
4241 put_disk(mddev->gendisk);
4242 }
4243 if (mddev->queue)
4244 blk_cleanup_queue(mddev->queue);
4245
eae1701f
N		 4246 kfree(mddev);
4247}
4248
52cf25d0 4249static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4250 .show = md_attr_show,
4251 .store = md_attr_store,
4252};
4253static struct kobj_type md_ktype = {
4254 .release = md_free,
4255 .sysfs_ops = &md_sysfs_ops,
4256 .default_attrs = md_default_attrs,
4257};
4258
1da177e4
LT		 4259int mdp_major = 0;
4260
5fd3a17e
DW		 4261static void mddev_delayed_delete(struct work_struct *ws)
4262{
4263 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4264
43a70507 4265 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4266 kobject_del(&mddev->kobj);
4267 kobject_put(&mddev->kobj);
4268}
4269
efeb53c0 4270static int md_alloc(dev_t dev, char *name)
1da177e4 4271{
48c9c27b 4272 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 4273 mddev_t *mddev = mddev_find(dev);
4274 struct gendisk *disk;
efeb53c0
N		 4275 int partitioned;
4276 int shift;
4277 int unit;
3830c62f 4278 int error;
1da177e4
LT		 4279
4280 if (!mddev)
efeb53c0
N		 4281 return -ENODEV;
4282
4283 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4284 shift = partitioned ? MdpMinorShift : 0;
4285 unit = MINOR(mddev->unit) >> shift;
1da177e4 4286
e804ac78
TH		 4287 /* wait for any previous instance of this device to be
4288 * completely removed (mddev_delayed_delete).
d3374825 4289 */
e804ac78 4290 flush_workqueue(md_misc_wq);
d3374825 4291
48c9c27b 4292 mutex_lock(&disks_mutex);
0909dc44
N		 4293 error = -EEXIST;
4294 if (mddev->gendisk)
4295 goto abort;
efeb53c0
N		 4296
4297 if (name) {
4298 /* Need to ensure that 'name' is not a duplicate.
4299 */
4300 mddev_t *mddev2;
4301 spin_lock(&all_mddevs_lock);
4302
4303 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4304 if (mddev2->gendisk &&
4305 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4306 spin_unlock(&all_mddevs_lock);
0909dc44 4307 goto abort;
efeb53c0
N		 4308 }
4309 spin_unlock(&all_mddevs_lock);
1da177e4 4310 }
8b765398 4311
0909dc44 4312 error = -ENOMEM;
8b765398 4313 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4314 if (!mddev->queue)
4315 goto abort;
409c57f3
N		 4316 mddev->queue->queuedata = mddev;
4317
409c57f3 4318 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4319
1da177e4
LT		 4320 disk = alloc_disk(1 << shift);
4321 if (!disk) {
8b765398
N		 4322 blk_cleanup_queue(mddev->queue);
4323 mddev->queue = NULL;
0909dc44 4324 goto abort;
1da177e4 4325 }
efeb53c0 4326 disk->major = MAJOR(mddev->unit);
1da177e4 4327 disk->first_minor = unit << shift;
efeb53c0
N		 4328 if (name)
4329 strcpy(disk->disk_name, name);
4330 else if (partitioned)
1da177e4 4331 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4332 else
1da177e4 4333 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4334 disk->fops = &md_fops;
4335 disk->private_data = mddev;
4336 disk->queue = mddev->queue;
92850bbd 4337 /* Allow extended partitions. This makes the
d3374825 4338 * 'mdp' device redundant, but we can't really
92850bbd
N		 4339 * remove it now.
4340 */
4341 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4
LT		 4342 add_disk(disk);
4343 mddev->gendisk = disk;
ed9e1982
TH		 4344 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4345 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4346 if (error) {
4347 /* This isn't possible, but as kobject_init_and_add is marked
4348 * __must_check, we must do something with the result
4349 */
5e55e2f5
N		 4350 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4351 disk->disk_name);
0909dc44
N		 4352 error = 0;
4353 }
00bcb4ac
N		 4354 if (mddev->kobj.sd &&
4355 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4356 printk(KERN_DEBUG "pointless warning\n");
be20e6c6
DW		 4357
4358 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
0909dc44
N		 4359 abort:
4360 mutex_unlock(&disks_mutex);
00bcb4ac 4361 if (!error && mddev->kobj.sd) {
3830c62f 4362 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4363 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4364 }
d3374825 4365 mddev_put(mddev);
0909dc44 4366 return error;
efeb53c0
N		 4367}
4368
4369static struct kobject *md_probe(dev_t dev, int *part, void *data)
4370{
4371 md_alloc(dev, NULL);
1da177e4
LT		 4372 return NULL;
4373}
4374
efeb53c0
N		 4375static int add_named_array(const char *val, struct kernel_param *kp)
4376{
4377 /* val must be "md_*" where * is not all digits.
4378 * We allocate an array with a large free minor number, and
4379 * set the name to val. val must not already be an active name.
4380 */
4381 int len = strlen(val);
4382 char buf[DISK_NAME_LEN];
4383
4384 while (len && val[len-1] == '\n')
4385 len--;
4386 if (len >= DISK_NAME_LEN)
4387 return -E2BIG;
4388 strlcpy(buf, val, len+1);
4389 if (strncmp(buf, "md_", 3) != 0)
4390 return -EINVAL;
4391 return md_alloc(0, buf);
4392}
4393
1da177e4
LT		 4394static void md_safemode_timeout(unsigned long data)
4395{
4396 mddev_t *mddev = (mddev_t *) data;
4397
0fd62b86
NB		 4398 if (!atomic_read(&mddev->writes_pending)) {
4399 mddev->safemode = 1;
4400 if (mddev->external)
00bcb4ac 4401 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4402 }
1da177e4
LT		 4403 md_wakeup_thread(mddev->thread);
4404}
4405
6ff8d8ec 4406static int start_dirty_degraded;
1da177e4 4407
390ee602 4408int md_run(mddev_t *mddev)
1da177e4 4409{
2604b703 4410 int err;
1da177e4 4411 mdk_rdev_t *rdev;
2604b703 4412 struct mdk_personality *pers;
1da177e4 4413
a757e64c
N		 4414 if (list_empty(&mddev->disks))
4415 /* cannot run an array with no devices.. */
1da177e4 4416 return -EINVAL;
1da177e4
LT		 4417
4418 if (mddev->pers)
4419 return -EBUSY;
bb4f1e9d
N		 4420 /* Cannot run until previous stop completes properly */
4421 if (mddev->sysfs_active)
4422 return -EBUSY;
b6eb127d 4423
1da177e4
LT		 4424 /*
4425 * Analyze all RAID superblock(s)
4426 */
1ec4a939
N		 4427 if (!mddev->raid_disks) {
4428 if (!mddev->persistent)
4429 return -EINVAL;
a757e64c 4430 analyze_sbs(mddev);
1ec4a939 4431 }
1da177e4 4432
d9d166c2
N		 4433 if (mddev->level != LEVEL_NONE)
4434 request_module("md-level-%d", mddev->level);
4435 else if (mddev->clevel[0])
4436 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4437
4438 /*
4439 * Drop all container device buffers, from now on
4440 * the only valid external interface is through the md
4441 * device.
1da177e4 4442 */
159ec1fc 4443 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4444 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4445 continue;
4446 sync_blockdev(rdev->bdev);
f98393a6 4447 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4448
4449 /* perform some consistency tests on the device.
4450 * We don't want the data to overlap the metadata,
58c0fed4 4451 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4452 */
a6ff7e08
JB		 4453 if (rdev->meta_bdev) {
4454 /* Nothing to check */;
4455 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4456 if (mddev->dev_sectors &&
4457 rdev->data_offset + mddev->dev_sectors
0f420358 4458 > rdev->sb_start) {
f0d76d70
N		 4459 printk("md: %s: data overlaps metadata\n",
4460 mdname(mddev));
4461 return -EINVAL;
4462 }
4463 } else {
0f420358 4464 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4465 > rdev->data_offset) {
4466 printk("md: %s: metadata overlaps data\n",
4467 mdname(mddev));
4468 return -EINVAL;
4469 }
4470 }
00bcb4ac 4471 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4472 }
4473
a167f663
N		 4474 if (mddev->bio_set == NULL)
4475 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4476
1da177e4 4477 spin_lock(&pers_lock);
d9d166c2 4478 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4479 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4480 spin_unlock(&pers_lock);
d9d166c2
N		 4481 if (mddev->level != LEVEL_NONE)
4482 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4483 mddev->level);
4484 else
4485 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4486 mddev->clevel);
1da177e4
LT		 4487 return -EINVAL;
4488 }
2604b703 4489 mddev->pers = pers;
1da177e4 4490 spin_unlock(&pers_lock);
34817e8c
N		 4491 if (mddev->level != pers->level) {
4492 mddev->level = pers->level;
4493 mddev->new_level = pers->level;
4494 }
d9d166c2 4495 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4496
f6705578 4497 if (mddev->reshape_position != MaxSector &&
63c70c4f 4498 pers->start_reshape == NULL) {
f6705578
N		 4499 /* This personality cannot handle reshaping... */
4500 mddev->pers = NULL;
4501 module_put(pers->owner);
4502 return -EINVAL;
4503 }
4504
7dd5e7c3
N		 4505 if (pers->sync_request) {
4506 /* Warn if this is a potentially silly
4507 * configuration.
4508 */
4509 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4510 mdk_rdev_t *rdev2;
7dd5e7c3 4511 int warned = 0;
159ec1fc
CR		 4512
4513 list_for_each_entry(rdev, &mddev->disks, same_set)
4514 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4515 if (rdev < rdev2 &&
4516 rdev->bdev->bd_contains ==
4517 rdev2->bdev->bd_contains) {
4518 printk(KERN_WARNING
4519 "%s: WARNING: %s appears to be"
4520 " on the same physical disk as"
4521 " %s.\n",
4522 mdname(mddev),
4523 bdevname(rdev->bdev,b),
4524 bdevname(rdev2->bdev,b2));
4525 warned = 1;
4526 }
4527 }
159ec1fc 4528
7dd5e7c3
N		 4529 if (warned)
4530 printk(KERN_WARNING
4531 "True protection against single-disk"
4532 " failure might be compromised.\n");
4533 }
4534
657390d2 4535 mddev->recovery = 0;
58c0fed4
AN		 4536 /* may be over-ridden by personality */
4537 mddev->resync_max_sectors = mddev->dev_sectors;
4538
6ff8d8ec 4539 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4540
0f9552b5 4541 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4542 mddev->ro = 2; /* read-only, but switch on first write */
4543
b15c2e57 4544 err = mddev->pers->run(mddev);
13e53df3
AN		 4545 if (err)
4546 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4547 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4548 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4549 " but 'external_size' not in effect?\n", __func__);
4550 printk(KERN_ERR
4551 "md: invalid array_size %llu > default size %llu\n",
4552 (unsigned long long)mddev->array_sectors / 2,
4553 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4554 err = -EINVAL;
4555 mddev->pers->stop(mddev);
4556 }
4557 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4558 err = bitmap_create(mddev);
4559 if (err) {
4560 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4561 mdname(mddev), err);
4562 mddev->pers->stop(mddev);
4563 }
4564 }
1da177e4 4565 if (err) {
1da177e4
LT		 4566 module_put(mddev->pers->owner);
4567 mddev->pers = NULL;
32a7627c
N		 4568 bitmap_destroy(mddev);
4569 return err;
1da177e4 4570 }
5e55e2f5 4571 if (mddev->pers->sync_request) {
00bcb4ac
N		 4572 if (mddev->kobj.sd &&
4573 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4574 printk(KERN_WARNING
4575 "md: cannot register extra attributes for %s\n",
4576 mdname(mddev));
00bcb4ac 4577 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4578 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4579 mddev->ro = 0;
4580
1da177e4 4581 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4582 atomic_set(&mddev->max_corr_read_errors,
4583 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4584 mddev->safemode = 0;
4585 mddev->safemode_timer.function = md_safemode_timeout;
4586 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4587 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4588 mddev->in_sync = 1;
0ca69886
N		 4589 smp_wmb();
4590 mddev->ready = 1;
159ec1fc 4591 list_for_each_entry(rdev, &mddev->disks, same_set)
86e6ffdd
N		 4592 if (rdev->raid_disk >= 0) {
4593 char nm[20];
4594 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5 4595 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
00bcb4ac 4596 /* failure here is OK */;
86e6ffdd 4597 }
1da177e4
LT		 4598
4599 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4600
850b2b42
N		 4601 if (mddev->flags)
4602 md_update_sb(mddev, 0);
1da177e4 4603
0b8c9de0
N		 4604 md_wakeup_thread(mddev->thread);
4605 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 4606
d7603b7e 4607 md_new_event(mddev);
00bcb4ac
N		 4608 sysfs_notify_dirent_safe(mddev->sysfs_state);
4609 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4610 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4611 return 0;
4612}
390ee602 4613EXPORT_SYMBOL_GPL(md_run);
1da177e4 4614
fe60b014
N		 4615static int do_md_run(mddev_t *mddev)
4616{
4617 int err;
4618
4619 err = md_run(mddev);
4620 if (err)
4621 goto out;
69e51b44
N		 4622 err = bitmap_load(mddev);
4623 if (err) {
4624 bitmap_destroy(mddev);
4625 goto out;
4626 }
fe60b014
N		 4627 set_capacity(mddev->gendisk, mddev->array_sectors);
4628 revalidate_disk(mddev->gendisk);
4629 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4630out:
4631 return err;
4632}
4633
1da177e4
LT		 4634static int restart_array(mddev_t *mddev)
4635{
4636 struct gendisk *disk = mddev->gendisk;
1da177e4 4637
80fab1d7 4638 /* Complain if it has no devices */
1da177e4 4639 if (list_empty(&mddev->disks))
80fab1d7
AN		 4640 return -ENXIO;
4641 if (!mddev->pers)
4642 return -EINVAL;
4643 if (!mddev->ro)
4644 return -EBUSY;
4645 mddev->safemode = 0;
4646 mddev->ro = 0;
4647 set_disk_ro(disk, 0);
4648 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4649 mdname(mddev));
4650 /* Kick recovery or resync if necessary */
4651 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4652 md_wakeup_thread(mddev->thread);
4653 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4654 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4655 return 0;
1da177e4
LT		 4656}
4657
acc55e22
N		 4658/* similar to deny_write_access, but accounts for our holding a reference
4659 * to the file ourselves */
4660static int deny_bitmap_write_access(struct file * file)
4661{
4662 struct inode *inode = file->f_mapping->host;
4663
4664 spin_lock(&inode->i_lock);
4665 if (atomic_read(&inode->i_writecount) > 1) {
4666 spin_unlock(&inode->i_lock);
4667 return -ETXTBSY;
4668 }
4669 atomic_set(&inode->i_writecount, -1);
4670 spin_unlock(&inode->i_lock);
4671
4672 return 0;
4673}
4674
43a70507 4675void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4676{
4677 struct inode *inode = file->f_mapping->host;
4678
4679 spin_lock(&inode->i_lock);
4680 atomic_set(&inode->i_writecount, 1);
4681 spin_unlock(&inode->i_lock);
4682}
4683
6177b472
N		 4684static void md_clean(mddev_t *mddev)
4685{
4686 mddev->array_sectors = 0;
4687 mddev->external_size = 0;
4688 mddev->dev_sectors = 0;
4689 mddev->raid_disks = 0;
4690 mddev->recovery_cp = 0;
4691 mddev->resync_min = 0;
4692 mddev->resync_max = MaxSector;
4693 mddev->reshape_position = MaxSector;
4694 mddev->external = 0;
4695 mddev->persistent = 0;
4696 mddev->level = LEVEL_NONE;
4697 mddev->clevel[0] = 0;
4698 mddev->flags = 0;
4699 mddev->ro = 0;
4700 mddev->metadata_type[0] = 0;
4701 mddev->chunk_sectors = 0;
4702 mddev->ctime = mddev->utime = 0;
4703 mddev->layout = 0;
4704 mddev->max_disks = 0;
4705 mddev->events = 0;
a8707c08 4706 mddev->can_decrease_events = 0;
6177b472
N		 4707 mddev->delta_disks = 0;
4708 mddev->new_level = LEVEL_NONE;
4709 mddev->new_layout = 0;
4710 mddev->new_chunk_sectors = 0;
4711 mddev->curr_resync = 0;
4712 mddev->resync_mismatches = 0;
4713 mddev->suspend_lo = mddev->suspend_hi = 0;
4714 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4715 mddev->recovery = 0;
4716 mddev->in_sync = 0;
4717 mddev->degraded = 0;
6177b472
N		 4718 mddev->safemode = 0;
4719 mddev->bitmap_info.offset = 0;
4720 mddev->bitmap_info.default_offset = 0;
4721 mddev->bitmap_info.chunksize = 0;
4722 mddev->bitmap_info.daemon_sleep = 0;
4723 mddev->bitmap_info.max_write_behind = 0;
252ac522 4724 mddev->plug = NULL;
6177b472
N		 4725}
4726
defad61a 4727static void __md_stop_writes(mddev_t *mddev)
a047e125
N		 4728{
4729 if (mddev->sync_thread) {
4730 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4731 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4732 reap_sync_thread(mddev);
a047e125
N		 4733 }
4734
4735 del_timer_sync(&mddev->safemode_timer);
4736
4737 bitmap_flush(mddev);
4738 md_super_wait(mddev);
4739
4740 if (!mddev->in_sync || mddev->flags) {
4741 /* mark array as shutdown cleanly */
4742 mddev->in_sync = 1;
4743 md_update_sb(mddev, 1);
4744 }
4745}
defad61a
N		 4746
4747void md_stop_writes(mddev_t *mddev)
4748{
4749 mddev_lock(mddev);
4750 __md_stop_writes(mddev);
4751 mddev_unlock(mddev);
4752}
390ee602 4753EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 4754
390ee602 4755void md_stop(mddev_t *mddev)
6177b472 4756{
0ca69886 4757 mddev->ready = 0;
6177b472
N		 4758 mddev->pers->stop(mddev);
4759 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4760 mddev->to_remove = &md_redundancy_group;
4761 module_put(mddev->pers->owner);
4762 mddev->pers = NULL;
cca9cf90 4763 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 4764}
390ee602 4765EXPORT_SYMBOL_GPL(md_stop);
6177b472 4766
a4bd82d0
N		 4767static int md_set_readonly(mddev_t *mddev, int is_open)
4768{
4769 int err = 0;
4770 mutex_lock(&mddev->open_mutex);
4771 if (atomic_read(&mddev->openers) > is_open) {
4772 printk("md: %s still in use.\n",mdname(mddev));
4773 err = -EBUSY;
4774 goto out;
4775 }
4776 if (mddev->pers) {
defad61a 4777 __md_stop_writes(mddev);
a4bd82d0
N		 4778
4779 err = -ENXIO;
4780 if (mddev->ro==1)
4781 goto out;
4782 mddev->ro = 1;
4783 set_disk_ro(mddev->gendisk, 1);
4784 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 4785 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 4786 err = 0;
4787 }
4788out:
4789 mutex_unlock(&mddev->open_mutex);
4790 return err;
4791}
4792
9e653b63
N		 4793/* mode:
4794 * 0 - completely stop and dis-assemble array
9e653b63
N		 4795 * 2 - stop but do not disassemble array
4796 */
df5b20cf 4797static int do_md_stop(mddev_t * mddev, int mode, int is_open)
1da177e4 4798{
1da177e4 4799 struct gendisk *disk = mddev->gendisk;
c4647292 4800 mdk_rdev_t *rdev;
1da177e4 4801
c8c00a69 4802 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 4803 if (atomic_read(&mddev->openers) > is_open ||
4804 mddev->sysfs_active) {
df5b20cf 4805 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 4806 mutex_unlock(&mddev->open_mutex);
4807 return -EBUSY;
4808 }
1da177e4 4809
6e17b027 4810 if (mddev->pers) {
a4bd82d0
N		 4811 if (mddev->ro)
4812 set_disk_ro(disk, 0);
409c57f3 4813
defad61a 4814 __md_stop_writes(mddev);
a4bd82d0
N		 4815 md_stop(mddev);
4816 mddev->queue->merge_bvec_fn = NULL;
4817 mddev->queue->unplug_fn = NULL;
4818 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 4819
a4bd82d0 4820 /* tell userspace to handle 'inactive' */
00bcb4ac 4821 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 4822
a4bd82d0
N		 4823 list_for_each_entry(rdev, &mddev->disks, same_set)
4824 if (rdev->raid_disk >= 0) {
4825 char nm[20];
4826 sprintf(nm, "rd%d", rdev->raid_disk);
4827 sysfs_remove_link(&mddev->kobj, nm);
4828 }
c4647292 4829
a4bd82d0 4830 set_capacity(disk, 0);
6e17b027 4831 mutex_unlock(&mddev->open_mutex);
a4bd82d0 4832 revalidate_disk(disk);
0d4ca600 4833
a4bd82d0
N		 4834 if (mddev->ro)
4835 mddev->ro = 0;
6e17b027
N		 4836 } else
4837 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 4838 /*
4839 * Free resources if final stop
4840 */
9e653b63 4841 if (mode == 0) {
1da177e4
LT		 4842 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4843
978f946b 4844 bitmap_destroy(mddev);
c3d9714e
N		 4845 if (mddev->bitmap_info.file) {
4846 restore_bitmap_write_access(mddev->bitmap_info.file);
4847 fput(mddev->bitmap_info.file);
4848 mddev->bitmap_info.file = NULL;
978f946b 4849 }
c3d9714e 4850 mddev->bitmap_info.offset = 0;
978f946b 4851
1da177e4
LT		 4852 export_array(mddev);
4853
6177b472 4854 md_clean(mddev);
934d9c23 4855 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 4856 if (mddev->hold_active == UNTIL_STOP)
4857 mddev->hold_active = 0;
a4bd82d0 4858 }
3f9d99c1 4859 blk_integrity_unregister(disk);
d7603b7e 4860 md_new_event(mddev);
00bcb4ac 4861 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 4862 return 0;
1da177e4
LT		 4863}
4864
fdee8ae4 4865#ifndef MODULE
1da177e4
LT		 4866static void autorun_array(mddev_t *mddev)
4867{
4868 mdk_rdev_t *rdev;
1da177e4
LT		 4869 int err;
4870
a757e64c 4871 if (list_empty(&mddev->disks))
1da177e4 4872 return;
1da177e4
LT		 4873
4874 printk(KERN_INFO "md: running: ");
4875
159ec1fc 4876 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 4877 char b[BDEVNAME_SIZE];
4878 printk("<%s>", bdevname(rdev->bdev,b));
4879 }
4880 printk("\n");
4881
d710e138 4882 err = do_md_run(mddev);
1da177e4
LT		 4883 if (err) {
4884 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 4885 do_md_stop(mddev, 0, 0);
1da177e4
LT		 4886 }
4887}
4888
4889/*
4890 * lets try to run arrays based on all disks that have arrived
4891 * until now. (those are in pending_raid_disks)
4892 *
4893 * the method: pick the first pending disk, collect all disks with
4894 * the same UUID, remove all from the pending list and put them into
4895 * the 'same_array' list. Then order this list based on superblock
4896 * update time (freshest comes first), kick out 'old' disks and
4897 * compare superblocks. If everything's fine then run it.
4898 *
4899 * If "unit" is allocated, then bump its reference count
4900 */
4901static void autorun_devices(int part)
4902{
159ec1fc 4903 mdk_rdev_t *rdev0, *rdev, *tmp;
1da177e4
LT		 4904 mddev_t *mddev;
4905 char b[BDEVNAME_SIZE];
4906
4907 printk(KERN_INFO "md: autorun ...\n");
4908 while (!list_empty(&pending_raid_disks)) {
e8703fe1 4909 int unit;
1da177e4 4910 dev_t dev;
ad01c9e3 4911 LIST_HEAD(candidates);
1da177e4
LT		 4912 rdev0 = list_entry(pending_raid_disks.next,
4913 mdk_rdev_t, same_set);
4914
4915 printk(KERN_INFO "md: considering %s ...\n",
4916 bdevname(rdev0->bdev,b));
4917 INIT_LIST_HEAD(&candidates);
159ec1fc 4918 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 4919 if (super_90_load(rdev, rdev0, 0) >= 0) {
4920 printk(KERN_INFO "md: adding %s ...\n",
4921 bdevname(rdev->bdev,b));
4922 list_move(&rdev->same_set, &candidates);
4923 }
4924 /*
4925 * now we have a set of devices, with all of them having
4926 * mostly sane superblocks. It's time to allocate the
4927 * mddev.
4928 */
e8703fe1
N		 4929 if (part) {
4930 dev = MKDEV(mdp_major,
4931 rdev0->preferred_minor << MdpMinorShift);
4932 unit = MINOR(dev) >> MdpMinorShift;
4933 } else {
4934 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4935 unit = MINOR(dev);
4936 }
4937 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 4938 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4939 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4940 break;
4941 }
1da177e4
LT		 4942
4943 md_probe(dev, NULL, NULL);
4944 mddev = mddev_find(dev);
9bbbca3a
NB		 4945 if (!mddev || !mddev->gendisk) {
4946 if (mddev)
4947 mddev_put(mddev);
4948 printk(KERN_ERR
1da177e4
LT		 4949 "md: cannot allocate memory for md drive.\n");
4950 break;
4951 }
4952 if (mddev_lock(mddev))
4953 printk(KERN_WARNING "md: %s locked, cannot run\n",
4954 mdname(mddev));
4955 else if (mddev->raid_disks || mddev->major_version
4956 || !list_empty(&mddev->disks)) {
4957 printk(KERN_WARNING
4958 "md: %s already running, cannot run %s\n",
4959 mdname(mddev), bdevname(rdev0->bdev,b));
4960 mddev_unlock(mddev);
4961 } else {
4962 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 4963 mddev->persistent = 1;
159ec1fc 4964 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 4965 list_del_init(&rdev->same_set);
4966 if (bind_rdev_to_array(rdev, mddev))
4967 export_rdev(rdev);
4968 }
4969 autorun_array(mddev);
4970 mddev_unlock(mddev);
4971 }
4972 /* on success, candidates will be empty, on error
4973 * it won't...
4974 */
159ec1fc 4975 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 4976 list_del_init(&rdev->same_set);
1da177e4 4977 export_rdev(rdev);
4b80991c 4978 }
1da177e4
LT		 4979 mddev_put(mddev);
4980 }
4981 printk(KERN_INFO "md: ... autorun DONE.\n");
4982}
fdee8ae4 4983#endif /* !MODULE */
1da177e4 4984
1da177e4
LT		 4985static int get_version(void __user * arg)
4986{
4987 mdu_version_t ver;
4988
4989 ver.major = MD_MAJOR_VERSION;
4990 ver.minor = MD_MINOR_VERSION;
4991 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4992
4993 if (copy_to_user(arg, &ver, sizeof(ver)))
4994 return -EFAULT;
4995
4996 return 0;
4997}
4998
4999static int get_array_info(mddev_t * mddev, void __user * arg)
5000{
5001 mdu_array_info_t info;
a9f326eb 5002 int nr,working,insync,failed,spare;
1da177e4 5003 mdk_rdev_t *rdev;
1da177e4 5004
a9f326eb 5005 nr=working=insync=failed=spare=0;
159ec1fc 5006 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5007 nr++;
b2d444d7 5008 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5009 failed++;
5010 else {
5011 working++;
b2d444d7 5012 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5013 insync++;
1da177e4
LT		 5014 else
5015 spare++;
5016 }
5017 }
5018
5019 info.major_version = mddev->major_version;
5020 info.minor_version = mddev->minor_version;
5021 info.patch_version = MD_PATCHLEVEL_VERSION;
5022 info.ctime = mddev->ctime;
5023 info.level = mddev->level;
58c0fed4
AN		 5024 info.size = mddev->dev_sectors / 2;
5025 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5026 info.size = -1;
1da177e4
LT		 5027 info.nr_disks = nr;
5028 info.raid_disks = mddev->raid_disks;
5029 info.md_minor = mddev->md_minor;
5030 info.not_persistent= !mddev->persistent;
5031
5032 info.utime = mddev->utime;
5033 info.state = 0;
5034 if (mddev->in_sync)
5035 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5036 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5037 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5038 info.active_disks = insync;
1da177e4
LT		 5039 info.working_disks = working;
5040 info.failed_disks = failed;
5041 info.spare_disks = spare;
5042
5043 info.layout = mddev->layout;
9d8f0363 5044 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5045
5046 if (copy_to_user(arg, &info, sizeof(info)))
5047 return -EFAULT;
5048
5049 return 0;
5050}
5051
87162a28 5052static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 5053{
5054 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5055 char *ptr, *buf = NULL;
5056 int err = -ENOMEM;
5057
b5470dc5
DW		 5058 if (md_allow_write(mddev))
5059 file = kmalloc(sizeof(*file), GFP_NOIO);
5060 else
5061 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5062
32a7627c
N		 5063 if (!file)
5064 goto out;
5065
5066 /* bitmap disabled, zero the first byte and copy out */
5067 if (!mddev->bitmap || !mddev->bitmap->file) {
5068 file->pathname[0] = '\0';
5069 goto copy_out;
5070 }
5071
5072 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5073 if (!buf)
5074 goto out;
5075
6bcfd601
CH		 5076 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5077 if (IS_ERR(ptr))
32a7627c
N		 5078 goto out;
5079
5080 strcpy(file->pathname, ptr);
5081
5082copy_out:
5083 err = 0;
5084 if (copy_to_user(arg, file, sizeof(*file)))
5085 err = -EFAULT;
5086out:
5087 kfree(buf);
5088 kfree(file);
5089 return err;
5090}
5091
1da177e4
LT		 5092static int get_disk_info(mddev_t * mddev, void __user * arg)
5093{
5094 mdu_disk_info_t info;
1da177e4
LT		 5095 mdk_rdev_t *rdev;
5096
5097 if (copy_from_user(&info, arg, sizeof(info)))
5098 return -EFAULT;
5099
26ef379f 5100 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5101 if (rdev) {
5102 info.major = MAJOR(rdev->bdev->bd_dev);
5103 info.minor = MINOR(rdev->bdev->bd_dev);
5104 info.raid_disk = rdev->raid_disk;
5105 info.state = 0;
b2d444d7 5106 if (test_bit(Faulty, &rdev->flags))
1da177e4 5107 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5108 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5109 info.state |= (1<<MD_DISK_ACTIVE);
5110 info.state |= (1<<MD_DISK_SYNC);
5111 }
8ddf9efe
N		 5112 if (test_bit(WriteMostly, &rdev->flags))
5113 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5114 } else {
5115 info.major = info.minor = 0;
5116 info.raid_disk = -1;
5117 info.state = (1<<MD_DISK_REMOVED);
5118 }
5119
5120 if (copy_to_user(arg, &info, sizeof(info)))
5121 return -EFAULT;
5122
5123 return 0;
5124}
5125
5126static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5127{
5128 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5129 mdk_rdev_t *rdev;
5130 dev_t dev = MKDEV(info->major,info->minor);
5131
5132 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5133 return -EOVERFLOW;
5134
5135 if (!mddev->raid_disks) {
5136 int err;
5137 /* expecting a device which has a superblock */
5138 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5139 if (IS_ERR(rdev)) {
5140 printk(KERN_WARNING
5141 "md: md_import_device returned %ld\n",
5142 PTR_ERR(rdev));
5143 return PTR_ERR(rdev);
5144 }
5145 if (!list_empty(&mddev->disks)) {
5146 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5147 mdk_rdev_t, same_set);
a9f326eb 5148 err = super_types[mddev->major_version]
1da177e4
LT		 5149 .load_super(rdev, rdev0, mddev->minor_version);
5150 if (err < 0) {
5151 printk(KERN_WARNING
5152 "md: %s has different UUID to %s\n",
5153 bdevname(rdev->bdev,b),
5154 bdevname(rdev0->bdev,b2));
5155 export_rdev(rdev);
5156 return -EINVAL;
5157 }
5158 }
5159 err = bind_rdev_to_array(rdev, mddev);
5160 if (err)
5161 export_rdev(rdev);
5162 return err;
5163 }
5164
5165 /*
5166 * add_new_disk can be used once the array is assembled
5167 * to add "hot spares". They must already have a superblock
5168 * written
5169 */
5170 if (mddev->pers) {
5171 int err;
5172 if (!mddev->pers->hot_add_disk) {
5173 printk(KERN_WARNING
5174 "%s: personality does not support diskops!\n",
5175 mdname(mddev));
5176 return -EINVAL;
5177 }
7b1e35f6
N		 5178 if (mddev->persistent)
5179 rdev = md_import_device(dev, mddev->major_version,
5180 mddev->minor_version);
5181 else
5182 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5183 if (IS_ERR(rdev)) {
5184 printk(KERN_WARNING
5185 "md: md_import_device returned %ld\n",
5186 PTR_ERR(rdev));
5187 return PTR_ERR(rdev);
5188 }
1a855a06 5189 /* set saved_raid_disk if appropriate */
41158c7e
N		 5190 if (!mddev->persistent) {
5191 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5192 info->raid_disk < mddev->raid_disks) {
41158c7e 5193 rdev->raid_disk = info->raid_disk;
bf572541
N		 5194 set_bit(In_sync, &rdev->flags);
5195 } else
41158c7e
N		 5196 rdev->raid_disk = -1;
5197 } else
5198 super_types[mddev->major_version].
5199 validate_super(mddev, rdev);
1a855a06
N		 5200 if (test_bit(In_sync, &rdev->flags))
5201 rdev->saved_raid_disk = rdev->raid_disk;
5202 else
5203 rdev->saved_raid_disk = -1;
41158c7e 5204
b2d444d7 5205 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5206 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5207 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5208 else
5209 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5210
1da177e4
LT		 5211 rdev->raid_disk = -1;
5212 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5213 if (!err && !mddev->pers->hot_remove_disk) {
5214 /* If there is hot_add_disk but no hot_remove_disk
5215 * then added disks for geometry changes,
5216 * and should be added immediately.
5217 */
5218 super_types[mddev->major_version].
5219 validate_super(mddev, rdev);
5220 err = mddev->pers->hot_add_disk(mddev, rdev);
5221 if (err)
5222 unbind_rdev_from_array(rdev);
5223 }
1da177e4
LT		 5224 if (err)
5225 export_rdev(rdev);
52664732 5226 else
00bcb4ac 5227 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5228
17571284 5229 md_update_sb(mddev, 1);
72a23c21
NB		 5230 if (mddev->degraded)
5231 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5232 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 5233 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5234 return err;
5235 }
5236
5237 /* otherwise, add_new_disk is only allowed
5238 * for major_version==0 superblocks
5239 */
5240 if (mddev->major_version != 0) {
5241 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5242 mdname(mddev));
5243 return -EINVAL;
5244 }
5245
5246 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5247 int err;
d710e138 5248 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5249 if (IS_ERR(rdev)) {
5250 printk(KERN_WARNING
5251 "md: error, md_import_device() returned %ld\n",
5252 PTR_ERR(rdev));
5253 return PTR_ERR(rdev);
5254 }
5255 rdev->desc_nr = info->number;
5256 if (info->raid_disk < mddev->raid_disks)
5257 rdev->raid_disk = info->raid_disk;
5258 else
5259 rdev->raid_disk = -1;
5260
1da177e4 5261 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5262 if (info->state & (1<<MD_DISK_SYNC))
5263 set_bit(In_sync, &rdev->flags);
1da177e4 5264
8ddf9efe
N		 5265 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5266 set_bit(WriteMostly, &rdev->flags);
5267
1da177e4
LT		 5268 if (!mddev->persistent) {
5269 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5270 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5271 } else
57b2caa3 5272 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5273 rdev->sectors = rdev->sb_start;
1da177e4 5274
2bf071bf
N		 5275 err = bind_rdev_to_array(rdev, mddev);
5276 if (err) {
5277 export_rdev(rdev);
5278 return err;
5279 }
1da177e4
LT		 5280 }
5281
5282 return 0;
5283}
5284
5285static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5286{
5287 char b[BDEVNAME_SIZE];
5288 mdk_rdev_t *rdev;
5289
1da177e4
LT		 5290 rdev = find_rdev(mddev, dev);
5291 if (!rdev)
5292 return -ENXIO;
5293
5294 if (rdev->raid_disk >= 0)
5295 goto busy;
5296
5297 kick_rdev_from_array(rdev);
850b2b42 5298 md_update_sb(mddev, 1);
d7603b7e 5299 md_new_event(mddev);
1da177e4
LT		 5300
5301 return 0;
5302busy:
fdefa4d8 5303 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5304 bdevname(rdev->bdev,b), mdname(mddev));
5305 return -EBUSY;
5306}
5307
5308static int hot_add_disk(mddev_t * mddev, dev_t dev)
5309{
5310 char b[BDEVNAME_SIZE];
5311 int err;
1da177e4
LT		 5312 mdk_rdev_t *rdev;
5313
5314 if (!mddev->pers)
5315 return -ENODEV;
5316
5317 if (mddev->major_version != 0) {
5318 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5319 " version-0 superblocks.\n",
5320 mdname(mddev));
5321 return -EINVAL;
5322 }
5323 if (!mddev->pers->hot_add_disk) {
5324 printk(KERN_WARNING
5325 "%s: personality does not support diskops!\n",
5326 mdname(mddev));
5327 return -EINVAL;
5328 }
5329
d710e138 5330 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5331 if (IS_ERR(rdev)) {
5332 printk(KERN_WARNING
5333 "md: error, md_import_device() returned %ld\n",
5334 PTR_ERR(rdev));
5335 return -EINVAL;
5336 }
5337
5338 if (mddev->persistent)
57b2caa3 5339 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5340 else
77304d2a 5341 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5342
8190e754 5343 rdev->sectors = rdev->sb_start;
1da177e4 5344
b2d444d7 5345 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5346 printk(KERN_WARNING
5347 "md: can not hot-add faulty %s disk to %s!\n",
5348 bdevname(rdev->bdev,b), mdname(mddev));
5349 err = -EINVAL;
5350 goto abort_export;
5351 }
b2d444d7 5352 clear_bit(In_sync, &rdev->flags);
1da177e4 5353 rdev->desc_nr = -1;
5842730d 5354 rdev->saved_raid_disk = -1;
2bf071bf
N		 5355 err = bind_rdev_to_array(rdev, mddev);
5356 if (err)
5357 goto abort_export;
1da177e4
LT		 5358
5359 /*
5360 * The rest should better be atomic, we can have disk failures
5361 * noticed in interrupt contexts ...
5362 */
5363
1da177e4
LT		 5364 rdev->raid_disk = -1;
5365
850b2b42 5366 md_update_sb(mddev, 1);
1da177e4
LT		 5367
5368 /*
5369 * Kick recovery, maybe this spare has to be added to the
5370 * array immediately.
5371 */
5372 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5373 md_wakeup_thread(mddev->thread);
d7603b7e 5374 md_new_event(mddev);
1da177e4
LT		 5375 return 0;
5376
1da177e4
LT		 5377abort_export:
5378 export_rdev(rdev);
5379 return err;
5380}
5381
32a7627c
N		 5382static int set_bitmap_file(mddev_t *mddev, int fd)
5383{
5384 int err;
5385
36fa3063
N		 5386 if (mddev->pers) {
5387 if (!mddev->pers->quiesce)
5388 return -EBUSY;
5389 if (mddev->recovery || mddev->sync_thread)
5390 return -EBUSY;
5391 /* we should be able to change the bitmap.. */
5392 }
32a7627c 5393
32a7627c 5394
36fa3063
N		 5395 if (fd >= 0) {
5396 if (mddev->bitmap)
5397 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5398 mddev->bitmap_info.file = fget(fd);
32a7627c 5399
c3d9714e 5400 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5401 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5402 mdname(mddev));
5403 return -EBADF;
5404 }
5405
c3d9714e 5406 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5407 if (err) {
5408 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5409 mdname(mddev));
c3d9714e
N		 5410 fput(mddev->bitmap_info.file);
5411 mddev->bitmap_info.file = NULL;
36fa3063
N		 5412 return err;
5413 }
c3d9714e 5414 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5415 } else if (mddev->bitmap == NULL)
5416 return -ENOENT; /* cannot remove what isn't there */
5417 err = 0;
5418 if (mddev->pers) {
5419 mddev->pers->quiesce(mddev, 1);
69e51b44 5420 if (fd >= 0) {
36fa3063 5421 err = bitmap_create(mddev);
69e51b44
N		 5422 if (!err)
5423 err = bitmap_load(mddev);
5424 }
d7375ab3 5425 if (fd < 0 || err) {
36fa3063 5426 bitmap_destroy(mddev);
d7375ab3
N		 5427 fd = -1; /* make sure to put the file */
5428 }
36fa3063 5429 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5430 }
5431 if (fd < 0) {
c3d9714e
N		 5432 if (mddev->bitmap_info.file) {
5433 restore_bitmap_write_access(mddev->bitmap_info.file);
5434 fput(mddev->bitmap_info.file);
acc55e22 5435 }
c3d9714e 5436 mddev->bitmap_info.file = NULL;
36fa3063
N		 5437 }
5438
32a7627c
N		 5439 return err;
5440}
5441
1da177e4
LT		 5442/*
5443 * set_array_info is used two different ways
5444 * The original usage is when creating a new array.
5445 * In this usage, raid_disks is > 0 and it together with
5446 * level, size, not_persistent,layout,chunksize determine the
5447 * shape of the array.
5448 * This will always create an array with a type-0.90.0 superblock.
5449 * The newer usage is when assembling an array.
5450 * In this case raid_disks will be 0, and the major_version field is
5451 * use to determine which style super-blocks are to be found on the devices.
5452 * The minor and patch _version numbers are also kept incase the
5453 * super_block handler wishes to interpret them.
5454 */
5455static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5456{
5457
5458 if (info->raid_disks == 0) {
5459 /* just setting version number for superblock loading */
5460 if (info->major_version < 0 ||
50511da3 5461 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5462 super_types[info->major_version].name == NULL) {
5463 /* maybe try to auto-load a module? */
5464 printk(KERN_INFO
5465 "md: superblock version %d not known\n",
5466 info->major_version);
5467 return -EINVAL;
5468 }
5469 mddev->major_version = info->major_version;
5470 mddev->minor_version = info->minor_version;
5471 mddev->patch_version = info->patch_version;
3f9d7b0d 5472 mddev->persistent = !info->not_persistent;
cbd19983
N		 5473 /* ensure mddev_put doesn't delete this now that there
5474 * is some minimal configuration.
5475 */
5476 mddev->ctime = get_seconds();
1da177e4
LT		 5477 return 0;
5478 }
5479 mddev->major_version = MD_MAJOR_VERSION;
5480 mddev->minor_version = MD_MINOR_VERSION;
5481 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5482 mddev->ctime = get_seconds();
5483
5484 mddev->level = info->level;
17115e03 5485 mddev->clevel[0] = 0;
58c0fed4 5486 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5487 mddev->raid_disks = info->raid_disks;
5488 /* don't set md_minor, it is determined by which /dev/md* was
5489 * openned
5490 */
5491 if (info->state & (1<<MD_SB_CLEAN))
5492 mddev->recovery_cp = MaxSector;
5493 else
5494 mddev->recovery_cp = 0;
5495 mddev->persistent = ! info->not_persistent;
e691063a 5496 mddev->external = 0;
1da177e4
LT		 5497
5498 mddev->layout = info->layout;
9d8f0363 5499 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5500
5501 mddev->max_disks = MD_SB_DISKS;
5502
e691063a
N		 5503 if (mddev->persistent)
5504 mddev->flags = 0;
850b2b42 5505 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5506
c3d9714e
N		 5507 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5508 mddev->bitmap_info.offset = 0;
b2a2703c 5509
f6705578
N		 5510 mddev->reshape_position = MaxSector;
5511
1da177e4
LT		 5512 /*
5513 * Generate a 128 bit UUID
5514 */
5515 get_random_bytes(mddev->uuid, 16);
5516
f6705578 5517 mddev->new_level = mddev->level;
664e7c41 5518 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5519 mddev->new_layout = mddev->layout;
5520 mddev->delta_disks = 0;
5521
1da177e4
LT		 5522 return 0;
5523}
5524
1f403624
DW		 5525void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5526{
b522adcd
DW		 5527 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5528
5529 if (mddev->external_size)
5530 return;
5531
1f403624
DW		 5532 mddev->array_sectors = array_sectors;
5533}
5534EXPORT_SYMBOL(md_set_array_sectors);
5535
d71f9f88 5536static int update_size(mddev_t *mddev, sector_t num_sectors)
a35b0d69 5537{
159ec1fc 5538 mdk_rdev_t *rdev;
a35b0d69 5539 int rv;
d71f9f88 5540 int fit = (num_sectors == 0);
a35b0d69
N		 5541
5542 if (mddev->pers->resize == NULL)
5543 return -EINVAL;
d71f9f88
AN		 5544 /* The "num_sectors" is the number of sectors of each device that
5545 * is used. This can only make sense for arrays with redundancy.
5546 * linear and raid0 always use whatever space is available. We can only
5547 * consider changing this number if no resync or reconstruction is
5548 * happening, and if the new size is acceptable. It must fit before the
0f420358 5549 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5550 * of each device. If num_sectors is zero, we find the largest size
5551 * that fits.
a35b0d69
N		 5552 */
5553 if (mddev->sync_thread)
5554 return -EBUSY;
dba034ee
N		 5555 if (mddev->bitmap)
5556 /* Sorry, cannot grow a bitmap yet, just remove it,
5557 * grow, and re-add.
5558 */
5559 return -EBUSY;
159ec1fc 5560 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5561 sector_t avail = rdev->sectors;
01ab5662 5562
d71f9f88
AN		 5563 if (fit && (num_sectors == 0 || num_sectors > avail))
5564 num_sectors = avail;
5565 if (avail < num_sectors)
a35b0d69
N		 5566 return -ENOSPC;
5567 }
d71f9f88 5568 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5569 if (!rv)
5570 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5571 return rv;
5572}
5573
da943b99
N		 5574static int update_raid_disks(mddev_t *mddev, int raid_disks)
5575{
5576 int rv;
5577 /* change the number of raid disks */
63c70c4f 5578 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5579 return -EINVAL;
5580 if (raid_disks <= 0 ||
233fca36 5581 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5582 return -EINVAL;
63c70c4f 5583 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5584 return -EBUSY;
63c70c4f
N		 5585 mddev->delta_disks = raid_disks - mddev->raid_disks;
5586
5587 rv = mddev->pers->check_reshape(mddev);
da943b99
N		 5588 return rv;
5589}
5590
5591
1da177e4
LT		 5592/*
5593 * update_array_info is used to change the configuration of an
5594 * on-line array.
5595 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5596 * fields in the info are checked against the array.
5597 * Any differences that cannot be handled will cause an error.
5598 * Normally, only one change can be managed at a time.
5599 */
5600static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5601{
5602 int rv = 0;
5603 int cnt = 0;
36fa3063
N		 5604 int state = 0;
5605
5606 /* calculate expected state,ignoring low bits */
c3d9714e 5607 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5608 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5609
5610 if (mddev->major_version != info->major_version ||
5611 mddev->minor_version != info->minor_version ||
5612/* mddev->patch_version != info->patch_version || */
5613 mddev->ctime != info->ctime ||
5614 mddev->level != info->level ||
5615/* mddev->layout != info->layout || */
5616 !mddev->persistent != info->not_persistent||
9d8f0363 5617 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5618 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5619 ((state^info->state) & 0xfffffe00)
5620 )
1da177e4
LT		 5621 return -EINVAL;
5622 /* Check there is only one change */
58c0fed4
AN		 5623 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5624 cnt++;
5625 if (mddev->raid_disks != info->raid_disks)
5626 cnt++;
5627 if (mddev->layout != info->layout)
5628 cnt++;
5629 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5630 cnt++;
5631 if (cnt == 0)
5632 return 0;
5633 if (cnt > 1)
5634 return -EINVAL;
1da177e4
LT		 5635
5636 if (mddev->layout != info->layout) {
5637 /* Change layout
5638 * we don't need to do anything at the md level, the
5639 * personality will take care of it all.
5640 */
50ac168a 5641 if (mddev->pers->check_reshape == NULL)
1da177e4 5642 return -EINVAL;
597a711b
N		 5643 else {
5644 mddev->new_layout = info->layout;
50ac168a 5645 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5646 if (rv)
5647 mddev->new_layout = mddev->layout;
5648 return rv;
5649 }
1da177e4 5650 }
58c0fed4 5651 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5652 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5653
da943b99
N		 5654 if (mddev->raid_disks != info->raid_disks)
5655 rv = update_raid_disks(mddev, info->raid_disks);
5656
36fa3063
N		 5657 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5658 if (mddev->pers->quiesce == NULL)
5659 return -EINVAL;
5660 if (mddev->recovery || mddev->sync_thread)
5661 return -EBUSY;
5662 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5663 /* add the bitmap */
5664 if (mddev->bitmap)
5665 return -EEXIST;
c3d9714e 5666 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5667 return -EINVAL;
c3d9714e
N		 5668 mddev->bitmap_info.offset =
5669 mddev->bitmap_info.default_offset;
36fa3063
N		 5670 mddev->pers->quiesce(mddev, 1);
5671 rv = bitmap_create(mddev);
69e51b44
N		 5672 if (!rv)
5673 rv = bitmap_load(mddev);
36fa3063
N		 5674 if (rv)
5675 bitmap_destroy(mddev);
5676 mddev->pers->quiesce(mddev, 0);
5677 } else {
5678 /* remove the bitmap */
5679 if (!mddev->bitmap)
5680 return -ENOENT;
5681 if (mddev->bitmap->file)
5682 return -EINVAL;
5683 mddev->pers->quiesce(mddev, 1);
5684 bitmap_destroy(mddev);
5685 mddev->pers->quiesce(mddev, 0);
c3d9714e 5686 mddev->bitmap_info.offset = 0;
36fa3063
N		 5687 }
5688 }
850b2b42 5689 md_update_sb(mddev, 1);
1da177e4
LT		 5690 return rv;
5691}
5692
5693static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5694{
5695 mdk_rdev_t *rdev;
5696
5697 if (mddev->pers == NULL)
5698 return -ENODEV;
5699
5700 rdev = find_rdev(mddev, dev);
5701 if (!rdev)
5702 return -ENODEV;
5703
5704 md_error(mddev, rdev);
5705 return 0;
5706}
5707
2f9618ce
AN		 5708/*
5709 * We have a problem here : there is no easy way to give a CHS
5710 * virtual geometry. We currently pretend that we have a 2 heads
5711 * 4 sectors (with a BIG number of cylinders...). This drives
5712 * dosfs just mad... ;-)
5713 */
a885c8c4
CH		 5714static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5715{
5716 mddev_t *mddev = bdev->bd_disk->private_data;
5717
5718 geo->heads = 2;
5719 geo->sectors = 4;
49ce6cea 5720 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 5721 return 0;
5722}
5723
a39907fa 5724static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 5725 unsigned int cmd, unsigned long arg)
5726{
5727 int err = 0;
5728 void __user *argp = (void __user *)arg;
1da177e4 5729 mddev_t *mddev = NULL;
e2218350 5730 int ro;
1da177e4
LT		 5731
5732 if (!capable(CAP_SYS_ADMIN))
5733 return -EACCES;
5734
5735 /*
5736 * Commands dealing with the RAID driver but not any
5737 * particular array:
5738 */
5739 switch (cmd)
5740 {
5741 case RAID_VERSION:
5742 err = get_version(argp);
5743 goto done;
5744
5745 case PRINT_RAID_DEBUG:
5746 err = 0;
5747 md_print_devices();
5748 goto done;
5749
5750#ifndef MODULE
5751 case RAID_AUTORUN:
5752 err = 0;
5753 autostart_arrays(arg);
5754 goto done;
5755#endif
5756 default:;
5757 }
5758
5759 /*
5760 * Commands creating/starting a new array:
5761 */
5762
a39907fa 5763 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 5764
5765 if (!mddev) {
5766 BUG();
5767 goto abort;
5768 }
5769
1da177e4
LT		 5770 err = mddev_lock(mddev);
5771 if (err) {
5772 printk(KERN_INFO
5773 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5774 err, cmd);
5775 goto abort;
5776 }
5777
5778 switch (cmd)
5779 {
5780 case SET_ARRAY_INFO:
5781 {
5782 mdu_array_info_t info;
5783 if (!arg)
5784 memset(&info, 0, sizeof(info));
5785 else if (copy_from_user(&info, argp, sizeof(info))) {
5786 err = -EFAULT;
5787 goto abort_unlock;
5788 }
5789 if (mddev->pers) {
5790 err = update_array_info(mddev, &info);
5791 if (err) {
5792 printk(KERN_WARNING "md: couldn't update"
5793 " array info. %d\n", err);
5794 goto abort_unlock;
5795 }
5796 goto done_unlock;
5797 }
5798 if (!list_empty(&mddev->disks)) {
5799 printk(KERN_WARNING
5800 "md: array %s already has disks!\n",
5801 mdname(mddev));
5802 err = -EBUSY;
5803 goto abort_unlock;
5804 }
5805 if (mddev->raid_disks) {
5806 printk(KERN_WARNING
5807 "md: array %s already initialised!\n",
5808 mdname(mddev));
5809 err = -EBUSY;
5810 goto abort_unlock;
5811 }
5812 err = set_array_info(mddev, &info);
5813 if (err) {
5814 printk(KERN_WARNING "md: couldn't set"
5815 " array info. %d\n", err);
5816 goto abort_unlock;
5817 }
5818 }
5819 goto done_unlock;
5820
5821 default:;
5822 }
5823
5824 /*
5825 * Commands querying/configuring an existing array:
5826 */
32a7627c 5827 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 5828 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 5829 if ((!mddev->raid_disks && !mddev->external)
5830 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5831 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5832 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 5833 err = -ENODEV;
5834 goto abort_unlock;
5835 }
5836
5837 /*
5838 * Commands even a read-only array can execute:
5839 */
5840 switch (cmd)
5841 {
5842 case GET_ARRAY_INFO:
5843 err = get_array_info(mddev, argp);
5844 goto done_unlock;
5845
32a7627c 5846 case GET_BITMAP_FILE:
87162a28 5847 err = get_bitmap_file(mddev, argp);
32a7627c
N		 5848 goto done_unlock;
5849
1da177e4
LT		 5850 case GET_DISK_INFO:
5851 err = get_disk_info(mddev, argp);
5852 goto done_unlock;
5853
5854 case RESTART_ARRAY_RW:
5855 err = restart_array(mddev);
5856 goto done_unlock;
5857
5858 case STOP_ARRAY:
d710e138 5859 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 5860 goto done_unlock;
5861
5862 case STOP_ARRAY_RO:
a4bd82d0 5863 err = md_set_readonly(mddev, 1);
1da177e4
LT		 5864 goto done_unlock;
5865
e2218350
DW		 5866 case BLKROSET:
5867 if (get_user(ro, (int __user *)(arg))) {
5868 err = -EFAULT;
5869 goto done_unlock;
5870 }
5871 err = -EINVAL;
5872
5873 /* if the bdev is going readonly the value of mddev->ro
5874 * does not matter, no writes are coming
5875 */
5876 if (ro)
5877 goto done_unlock;
5878
5879 /* are we are already prepared for writes? */
5880 if (mddev->ro != 1)
5881 goto done_unlock;
5882
5883 /* transitioning to readauto need only happen for
5884 * arrays that call md_write_start
5885 */
5886 if (mddev->pers) {
5887 err = restart_array(mddev);
5888 if (err == 0) {
5889 mddev->ro = 2;
5890 set_disk_ro(mddev->gendisk, 0);
5891 }
5892 }
5893 goto done_unlock;
1da177e4
LT		 5894 }
5895
5896 /*
5897 * The remaining ioctls are changing the state of the
f91de92e
N		 5898 * superblock, so we do not allow them on read-only arrays.
5899 * However non-MD ioctls (e.g. get-size) will still come through
5900 * here and hit the 'default' below, so only disallow
5901 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 5902 */
bb57fc64 5903 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 5904 if (mddev->ro == 2) {
5905 mddev->ro = 0;
00bcb4ac 5906 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 5907 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5908 md_wakeup_thread(mddev->thread);
f91de92e
N		 5909 } else {
5910 err = -EROFS;
5911 goto abort_unlock;
5912 }
1da177e4
LT		 5913 }
5914
5915 switch (cmd)
5916 {
5917 case ADD_NEW_DISK:
5918 {
5919 mdu_disk_info_t info;
5920 if (copy_from_user(&info, argp, sizeof(info)))
5921 err = -EFAULT;
5922 else
5923 err = add_new_disk(mddev, &info);
5924 goto done_unlock;
5925 }
5926
5927 case HOT_REMOVE_DISK:
5928 err = hot_remove_disk(mddev, new_decode_dev(arg));
5929 goto done_unlock;
5930
5931 case HOT_ADD_DISK:
5932 err = hot_add_disk(mddev, new_decode_dev(arg));
5933 goto done_unlock;
5934
5935 case SET_DISK_FAULTY:
5936 err = set_disk_faulty(mddev, new_decode_dev(arg));
5937 goto done_unlock;
5938
5939 case RUN_ARRAY:
d710e138 5940 err = do_md_run(mddev);
1da177e4
LT		 5941 goto done_unlock;
5942
32a7627c
N		 5943 case SET_BITMAP_FILE:
5944 err = set_bitmap_file(mddev, (int)arg);
5945 goto done_unlock;
5946
1da177e4 5947 default:
1da177e4
LT		 5948 err = -EINVAL;
5949 goto abort_unlock;
5950 }
5951
5952done_unlock:
5953abort_unlock:
d3374825
N		 5954 if (mddev->hold_active == UNTIL_IOCTL &&
5955 err != -EINVAL)
5956 mddev->hold_active = 0;
1da177e4
LT		 5957 mddev_unlock(mddev);
5958
5959 return err;
5960done:
5961 if (err)
5962 MD_BUG();
5963abort:
5964 return err;
5965}
aa98aa31
AB		 5966#ifdef CONFIG_COMPAT
5967static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
5968 unsigned int cmd, unsigned long arg)
5969{
5970 switch (cmd) {
5971 case HOT_REMOVE_DISK:
5972 case HOT_ADD_DISK:
5973 case SET_DISK_FAULTY:
5974 case SET_BITMAP_FILE:
5975 /* These take in integer arg, do not convert */
5976 break;
5977 default:
5978 arg = (unsigned long)compat_ptr(arg);
5979 break;
5980 }
5981
5982 return md_ioctl(bdev, mode, cmd, arg);
5983}
5984#endif /* CONFIG_COMPAT */
1da177e4 5985
a39907fa 5986static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 5987{
5988 /*
5989 * Succeed if we can lock the mddev, which confirms that
5990 * it isn't being stopped right now.
5991 */
d3374825 5992 mddev_t *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 5993 int err;
5994
d3374825
N		 5995 if (mddev->gendisk != bdev->bd_disk) {
5996 /* we are racing with mddev_put which is discarding this
5997 * bd_disk.
5998 */
5999 mddev_put(mddev);
6000 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6001 flush_workqueue(md_misc_wq);
d3374825
N		 6002 /* Then retry the open from the top */
6003 return -ERESTARTSYS;
6004 }
6005 BUG_ON(mddev != bdev->bd_disk->private_data);
6006
c8c00a69 6007 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6008 goto out;
6009
6010 err = 0;
f2ea68cf 6011 atomic_inc(&mddev->openers);
c8c00a69 6012 mutex_unlock(&mddev->open_mutex);
1da177e4 6013
f3b99be1 6014 check_disk_size_change(mddev->gendisk, bdev);
1da177e4
LT		 6015 out:
6016 return err;
6017}
6018
a39907fa 6019static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6020{
a39907fa 6021 mddev_t *mddev = disk->private_data;
1da177e4 6022
52e5f9d1 6023 BUG_ON(!mddev);
f2ea68cf 6024 atomic_dec(&mddev->openers);
1da177e4
LT		 6025 mddev_put(mddev);
6026
6027 return 0;
6028}
83d5cde4 6029static const struct block_device_operations md_fops =
1da177e4
LT		 6030{
6031 .owner = THIS_MODULE,
a39907fa
AV		 6032 .open = md_open,
6033 .release = md_release,
b492b852 6034 .ioctl = md_ioctl,
aa98aa31
AB		 6035#ifdef CONFIG_COMPAT
6036 .compat_ioctl = md_compat_ioctl,
6037#endif
a885c8c4 6038 .getgeo = md_getgeo,
1da177e4
LT		 6039};
6040
75c96f85 6041static int md_thread(void * arg)
1da177e4
LT		 6042{
6043 mdk_thread_t *thread = arg;
6044
1da177e4
LT		 6045 /*
6046 * md_thread is a 'system-thread', it's priority should be very
6047 * high. We avoid resource deadlocks individually in each
6048 * raid personality. (RAID5 does preallocation) We also use RR and
6049 * the very same RT priority as kswapd, thus we will never get
6050 * into a priority inversion deadlock.
6051 *
6052 * we definitely have to have equal or higher priority than
6053 * bdflush, otherwise bdflush will deadlock if there are too
6054 * many dirty RAID5 blocks.
6055 */
1da177e4 6056
6985c43f 6057 allow_signal(SIGKILL);
a6fb0934 6058 while (!kthread_should_stop()) {
1da177e4 6059
93588e22
N		 6060 /* We need to wait INTERRUPTIBLE so that
6061 * we don't add to the load-average.
6062 * That means we need to be sure no signals are
6063 * pending
6064 */
6065 if (signal_pending(current))
6066 flush_signals(current);
6067
6068 wait_event_interruptible_timeout
6069 (thread->wqueue,
6070 test_bit(THREAD_WAKEUP, &thread->flags)
6071 || kthread_should_stop(),
6072 thread->timeout);
1da177e4 6073
6c987910
N		 6074 clear_bit(THREAD_WAKEUP, &thread->flags);
6075 if (!kthread_should_stop())
589a594b 6076 thread->run(thread->mddev);
1da177e4 6077 }
a6fb0934 6078
1da177e4
LT		 6079 return 0;
6080}
6081
6082void md_wakeup_thread(mdk_thread_t *thread)
6083{
6084 if (thread) {
6085 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6086 set_bit(THREAD_WAKEUP, &thread->flags);
6087 wake_up(&thread->wqueue);
6088 }
6089}
6090
6091mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6092 const char *name)
6093{
6094 mdk_thread_t *thread;
1da177e4 6095
9ffae0cf 6096 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
1da177e4
LT		 6097 if (!thread)
6098 return NULL;
6099
1da177e4
LT		 6100 init_waitqueue_head(&thread->wqueue);
6101
1da177e4
LT		 6102 thread->run = run;
6103 thread->mddev = mddev;
32a7627c 6104 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6105 thread->tsk = kthread_run(md_thread, thread,
6106 "%s_%s",
6107 mdname(thread->mddev),
6108 name ?: mddev->pers->name);
a6fb0934 6109 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6110 kfree(thread);
6111 return NULL;
6112 }
1da177e4
LT		 6113 return thread;
6114}
6115
1da177e4
LT		 6116void md_unregister_thread(mdk_thread_t *thread)
6117{
e0cf8f04
N		 6118 if (!thread)
6119 return;
ba25f9dc 6120 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
a6fb0934
N		 6121
6122 kthread_stop(thread->tsk);
1da177e4
LT		 6123 kfree(thread);
6124}
6125
6126void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6127{
6128 if (!mddev) {
6129 MD_BUG();
6130 return;
6131 }
6132
b2d444d7 6133 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6134 return;
6bfe0b49
DW		 6135
6136 if (mddev->external)
6137 set_bit(Blocked, &rdev->flags);
32a7627c 6138/*
1da177e4
LT		 6139 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
6140 mdname(mddev),
6141 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
6142 __builtin_return_address(0),__builtin_return_address(1),
6143 __builtin_return_address(2),__builtin_return_address(3));
32a7627c 6144*/
d0a0a5ee
AM		 6145 if (!mddev->pers)
6146 return;
1da177e4
LT		 6147 if (!mddev->pers->error_handler)
6148 return;
6149 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6150 if (mddev->degraded)
6151 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6152 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6153 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6155 md_wakeup_thread(mddev->thread);
768a418d 6156 if (mddev->event_work.func)
e804ac78 6157 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6158 md_new_event_inintr(mddev);
1da177e4
LT		 6159}
6160
6161/* seq_file implementation /proc/mdstat */
6162
6163static void status_unused(struct seq_file *seq)
6164{
6165 int i = 0;
6166 mdk_rdev_t *rdev;
1da177e4
LT		 6167
6168 seq_printf(seq, "unused devices: ");
6169
159ec1fc 6170 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6171 char b[BDEVNAME_SIZE];
6172 i++;
6173 seq_printf(seq, "%s ",
6174 bdevname(rdev->bdev,b));
6175 }
6176 if (!i)
6177 seq_printf(seq, "<none>");
6178
6179 seq_printf(seq, "\n");
6180}
6181
6182
6183static void status_resync(struct seq_file *seq, mddev_t * mddev)
6184{
dd71cf6b
N		 6185 sector_t max_sectors, resync, res;
6186 unsigned long dt, db;
6187 sector_t rt;
4588b42e
N		 6188 int scale;
6189 unsigned int per_milli;
1da177e4 6190
dd71cf6b 6191 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6192
6193 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6194 max_sectors = mddev->resync_max_sectors;
1da177e4 6195 else
dd71cf6b 6196 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6197
6198 /*
6199 * Should not happen.
6200 */
dd71cf6b 6201 if (!max_sectors) {
1da177e4
LT		 6202 MD_BUG();
6203 return;
6204 }
4588b42e 6205 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6206 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6207 * u32, as those are the requirements for sector_div.
6208 * Thus 'scale' must be at least 10
6209 */
6210 scale = 10;
6211 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6212 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6213 scale++;
6214 }
6215 res = (resync>>scale)*1000;
dd71cf6b 6216 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6217
6218 per_milli = res;
1da177e4 6219 {
4588b42e 6220 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6221 seq_printf(seq, "[");
6222 for (i = 0; i < x; i++)
6223 seq_printf(seq, "=");
6224 seq_printf(seq, ">");
6225 for (i = 0; i < y; i++)
6226 seq_printf(seq, ".");
6227 seq_printf(seq, "] ");
6228 }
4588b42e 6229 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6230 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6231 "reshape" :
61df9d91
N		 6232 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6233 "check" :
6234 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6235 "resync" : "recovery"))),
6236 per_milli/10, per_milli % 10,
dd71cf6b
N		 6237 (unsigned long long) resync/2,
6238 (unsigned long long) max_sectors/2);
1da177e4
LT		 6239
6240 /*
1da177e4
LT		 6241 * dt: time from mark until now
6242 * db: blocks written from mark until now
6243 * rt: remaining time
dd71cf6b
N		 6244 *
6245 * rt is a sector_t, so could be 32bit or 64bit.
6246 * So we divide before multiply in case it is 32bit and close
6247 * to the limit.
6248 * We scale the divisor (db) by 32 to avoid loosing precision
6249 * near the end of resync when the number of remaining sectors
6250 * is close to 'db'.
6251 * We then divide rt by 32 after multiplying by db to compensate.
6252 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6253 */
6254 dt = ((jiffies - mddev->resync_mark) / HZ);
6255 if (!dt) dt++;
ff4e8d9a
N		 6256 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6257 - mddev->resync_mark_cnt;
1da177e4 6258
dd71cf6b
N		 6259 rt = max_sectors - resync; /* number of remaining sectors */
6260 sector_div(rt, db/32+1);
6261 rt *= dt;
6262 rt >>= 5;
6263
6264 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6265 ((unsigned long)rt % 60)/6);
1da177e4 6266
ff4e8d9a 6267 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6268}
6269
6270static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6271{
6272 struct list_head *tmp;
6273 loff_t l = *pos;
6274 mddev_t *mddev;
6275
6276 if (l >= 0x10000)
6277 return NULL;
6278 if (!l--)
6279 /* header */
6280 return (void*)1;
6281
6282 spin_lock(&all_mddevs_lock);
6283 list_for_each(tmp,&all_mddevs)
6284 if (!l--) {
6285 mddev = list_entry(tmp, mddev_t, all_mddevs);
6286 mddev_get(mddev);
6287 spin_unlock(&all_mddevs_lock);
6288 return mddev;
6289 }
6290 spin_unlock(&all_mddevs_lock);
6291 if (!l--)
6292 return (void*)2;/* tail */
6293 return NULL;
6294}
6295
6296static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6297{
6298 struct list_head *tmp;
6299 mddev_t *next_mddev, *mddev = v;
6300
6301 ++*pos;
6302 if (v == (void*)2)
6303 return NULL;
6304
6305 spin_lock(&all_mddevs_lock);
6306 if (v == (void*)1)
6307 tmp = all_mddevs.next;
6308 else
6309 tmp = mddev->all_mddevs.next;
6310 if (tmp != &all_mddevs)
6311 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6312 else {
6313 next_mddev = (void*)2;
6314 *pos = 0x10000;
6315 }
6316 spin_unlock(&all_mddevs_lock);
6317
6318 if (v != (void*)1)
6319 mddev_put(mddev);
6320 return next_mddev;
6321
6322}
6323
6324static void md_seq_stop(struct seq_file *seq, void *v)
6325{
6326 mddev_t *mddev = v;
6327
6328 if (mddev && v != (void*)1 && v != (void*)2)
6329 mddev_put(mddev);
6330}
6331
d7603b7e
N		 6332struct mdstat_info {
6333 int event;
6334};
6335
1da177e4
LT		 6336static int md_seq_show(struct seq_file *seq, void *v)
6337{
6338 mddev_t *mddev = v;
dd8ac336 6339 sector_t sectors;
1da177e4 6340 mdk_rdev_t *rdev;
d7603b7e 6341 struct mdstat_info *mi = seq->private;
32a7627c 6342 struct bitmap *bitmap;
1da177e4
LT		 6343
6344 if (v == (void*)1) {
2604b703 6345 struct mdk_personality *pers;
1da177e4
LT		 6346 seq_printf(seq, "Personalities : ");
6347 spin_lock(&pers_lock);
2604b703
N		 6348 list_for_each_entry(pers, &pers_list, list)
6349 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6350
6351 spin_unlock(&pers_lock);
6352 seq_printf(seq, "\n");
d7603b7e 6353 mi->event = atomic_read(&md_event_count);
1da177e4
LT		 6354 return 0;
6355 }
6356 if (v == (void*)2) {
6357 status_unused(seq);
6358 return 0;
6359 }
6360
5dc5cf7d 6361 if (mddev_lock(mddev) < 0)
1da177e4 6362 return -EINTR;
5dc5cf7d 6363
1da177e4
LT		 6364 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6365 seq_printf(seq, "%s : %sactive", mdname(mddev),
6366 mddev->pers ? "" : "in");
6367 if (mddev->pers) {
f91de92e 6368 if (mddev->ro==1)
1da177e4 6369 seq_printf(seq, " (read-only)");
f91de92e 6370 if (mddev->ro==2)
52720ae7 6371 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6372 seq_printf(seq, " %s", mddev->pers->name);
6373 }
6374
dd8ac336 6375 sectors = 0;
159ec1fc 6376 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6377 char b[BDEVNAME_SIZE];
6378 seq_printf(seq, " %s[%d]",
6379 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6380 if (test_bit(WriteMostly, &rdev->flags))
6381 seq_printf(seq, "(W)");
b2d444d7 6382 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6383 seq_printf(seq, "(F)");
6384 continue;
b325a32e
N		 6385 } else if (rdev->raid_disk < 0)
6386 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6387 sectors += rdev->sectors;
1da177e4
LT		 6388 }
6389
6390 if (!list_empty(&mddev->disks)) {
6391 if (mddev->pers)
6392 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6393 (unsigned long long)
6394 mddev->array_sectors / 2);
1da177e4
LT		 6395 else
6396 seq_printf(seq, "\n %llu blocks",
dd8ac336 6397 (unsigned long long)sectors / 2);
1da177e4 6398 }
1cd6bf19
N		 6399 if (mddev->persistent) {
6400 if (mddev->major_version != 0 ||
6401 mddev->minor_version != 90) {
6402 seq_printf(seq," super %d.%d",
6403 mddev->major_version,
6404 mddev->minor_version);
6405 }
e691063a
N		 6406 } else if (mddev->external)
6407 seq_printf(seq, " super external:%s",
6408 mddev->metadata_type);
6409 else
1cd6bf19 6410 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6411
6412 if (mddev->pers) {
d710e138 6413 mddev->pers->status(seq, mddev);
1da177e4 6414 seq_printf(seq, "\n ");
8e1b39d6
N		 6415 if (mddev->pers->sync_request) {
6416 if (mddev->curr_resync > 2) {
d710e138 6417 status_resync(seq, mddev);
8e1b39d6
N		 6418 seq_printf(seq, "\n ");
6419 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6420 seq_printf(seq, "\tresync=DELAYED\n ");
6421 else if (mddev->recovery_cp < MaxSector)
6422 seq_printf(seq, "\tresync=PENDING\n ");
6423 }
32a7627c
N		 6424 } else
6425 seq_printf(seq, "\n ");
6426
6427 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6428 unsigned long chunk_kb;
6429 unsigned long flags;
32a7627c 6430 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6431 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6432 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6433 "%lu%s chunk",
6434 bitmap->pages - bitmap->missing_pages,
6435 bitmap->pages,
6436 (bitmap->pages - bitmap->missing_pages)
6437 << (PAGE_SHIFT - 10),
42a04b50 6438 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6439 chunk_kb ? "KB" : "B");
78d742d8
N		 6440 if (bitmap->file) {
6441 seq_printf(seq, ", file: ");
c32c2f63 6442 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6443 }
78d742d8 6444
32a7627c
N		 6445 seq_printf(seq, "\n");
6446 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6447 }
6448
6449 seq_printf(seq, "\n");
6450 }
6451 mddev_unlock(mddev);
6452
6453 return 0;
6454}
6455
110518bc 6456static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6457 .start = md_seq_start,
6458 .next = md_seq_next,
6459 .stop = md_seq_stop,
6460 .show = md_seq_show,
6461};
6462
6463static int md_seq_open(struct inode *inode, struct file *file)
6464{
6465 int error;
d7603b7e
N		 6466 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
6467 if (mi == NULL)
6468 return -ENOMEM;
1da177e4
LT		 6469
6470 error = seq_open(file, &md_seq_ops);
d7603b7e
N		 6471 if (error)
6472 kfree(mi);
6473 else {
6474 struct seq_file *p = file->private_data;
6475 p->private = mi;
6476 mi->event = atomic_read(&md_event_count);
6477 }
1da177e4
LT		 6478 return error;
6479}
6480
d7603b7e
N		 6481static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6482{
6483 struct seq_file *m = filp->private_data;
6484 struct mdstat_info *mi = m->private;
6485 int mask;
6486
6487 poll_wait(filp, &md_event_waiters, wait);
6488
6489 /* always allow read */
6490 mask = POLLIN | POLLRDNORM;
6491
6492 if (mi->event != atomic_read(&md_event_count))
6493 mask |= POLLERR | POLLPRI;
6494 return mask;
6495}
6496
fa027c2a 6497static const struct file_operations md_seq_fops = {
e24650c2 6498 .owner = THIS_MODULE,
1da177e4
LT		 6499 .open = md_seq_open,
6500 .read = seq_read,
6501 .llseek = seq_lseek,
c3f94b40 6502 .release = seq_release_private,
d7603b7e 6503 .poll = mdstat_poll,
1da177e4
LT		 6504};
6505
2604b703 6506int register_md_personality(struct mdk_personality *p)
1da177e4 6507{
1da177e4 6508 spin_lock(&pers_lock);
2604b703
N		 6509 list_add_tail(&p->list, &pers_list);
6510 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6511 spin_unlock(&pers_lock);
6512 return 0;
6513}
6514
2604b703 6515int unregister_md_personality(struct mdk_personality *p)
1da177e4 6516{
2604b703 6517 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6518 spin_lock(&pers_lock);
2604b703 6519 list_del_init(&p->list);
1da177e4
LT		 6520 spin_unlock(&pers_lock);
6521 return 0;
6522}
6523
eea1bf38 6524static int is_mddev_idle(mddev_t *mddev, int init)
1da177e4
LT		 6525{
6526 mdk_rdev_t * rdev;
1da177e4 6527 int idle;
eea1bf38 6528 int curr_events;
1da177e4
LT		 6529
6530 idle = 1;
4b80991c
N		 6531 rcu_read_lock();
6532 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6533 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6534 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6535 (int)part_stat_read(&disk->part0, sectors[1]) -
6536 atomic_read(&disk->sync_io);
713f6ab1
N		 6537 /* sync IO will cause sync_io to increase before the disk_stats
6538 * as sync_io is counted when a request starts, and
6539 * disk_stats is counted when it completes.
6540 * So resync activity will cause curr_events to be smaller than
6541 * when there was no such activity.
6542 * non-sync IO will cause disk_stat to increase without
6543 * increasing sync_io so curr_events will (eventually)
6544 * be larger than it was before. Once it becomes
6545 * substantially larger, the test below will cause
6546 * the array to appear non-idle, and resync will slow
6547 * down.
6548 * If there is a lot of outstanding resync activity when
6549 * we set last_event to curr_events, then all that activity
6550 * completing might cause the array to appear non-idle
6551 * and resync will be slowed down even though there might
6552 * not have been non-resync activity. This will only
6553 * happen once though. 'last_events' will soon reflect
6554 * the state where there is little or no outstanding
6555 * resync requests, and further resync activity will
6556 * always make curr_events less than last_events.
c0e48521 6557 *
1da177e4 6558 */
eea1bf38 6559 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6560 rdev->last_events = curr_events;
6561 idle = 0;
6562 }
6563 }
4b80991c 6564 rcu_read_unlock();
1da177e4
LT		 6565 return idle;
6566}
6567
6568void md_done_sync(mddev_t *mddev, int blocks, int ok)
6569{
6570 /* another "blocks" (512byte) blocks have been synced */
6571 atomic_sub(blocks, &mddev->recovery_active);
6572 wake_up(&mddev->recovery_wait);
6573 if (!ok) {
dfc70645 6574 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6575 md_wakeup_thread(mddev->thread);
6576 // stop recovery, signal do_sync ....
6577 }
6578}
6579
6580
06d91a5f
N		 6581/* md_write_start(mddev, bi)
6582 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6583 * in superblock) before writing, schedule a superblock update
6584 * and wait for it to complete.
06d91a5f 6585 */
3d310eb7 6586void md_write_start(mddev_t *mddev, struct bio *bi)
1da177e4 6587{
0fd62b86 6588 int did_change = 0;
06d91a5f 6589 if (bio_data_dir(bi) != WRITE)
3d310eb7 6590 return;
06d91a5f 6591
f91de92e
N		 6592 BUG_ON(mddev->ro == 1);
6593 if (mddev->ro == 2) {
6594 /* need to switch to read/write */
6595 mddev->ro = 0;
6596 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6597 md_wakeup_thread(mddev->thread);
25156198 6598 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6599 did_change = 1;
f91de92e 6600 }
06d91a5f 6601 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6602 if (mddev->safemode == 1)
6603 mddev->safemode = 0;
06d91a5f 6604 if (mddev->in_sync) {
a9701a30 6605 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6606 if (mddev->in_sync) {
6607 mddev->in_sync = 0;
850b2b42 6608 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6609 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6610 md_wakeup_thread(mddev->thread);
0fd62b86 6611 did_change = 1;
3d310eb7 6612 }
a9701a30 6613 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6614 }
0fd62b86 6615 if (did_change)
00bcb4ac 6616 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6617 wait_event(mddev->sb_wait,
09a44cc1 6618 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6619}
6620
6621void md_write_end(mddev_t *mddev)
6622{
6623 if (atomic_dec_and_test(&mddev->writes_pending)) {
6624 if (mddev->safemode == 2)
6625 md_wakeup_thread(mddev->thread);
16f17b39 6626 else if (mddev->safemode_delay)
1da177e4
LT		 6627 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6628 }
6629}
6630
2a2275d6
N		 6631/* md_allow_write(mddev)
6632 * Calling this ensures that the array is marked 'active' so that writes
6633 * may proceed without blocking. It is important to call this before
6634 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6635 * Must be called with mddev_lock held.
b5470dc5
DW		 6636 *
6637 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6638 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6639 */
b5470dc5 6640int md_allow_write(mddev_t *mddev)
2a2275d6
N		 6641{
6642 if (!mddev->pers)
b5470dc5 6643 return 0;
2a2275d6 6644 if (mddev->ro)
b5470dc5 6645 return 0;
1a0fd497 6646 if (!mddev->pers->sync_request)
b5470dc5 6647 return 0;
2a2275d6
N		 6648
6649 spin_lock_irq(&mddev->write_lock);
6650 if (mddev->in_sync) {
6651 mddev->in_sync = 0;
6652 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6653 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6654 if (mddev->safemode_delay &&
6655 mddev->safemode == 0)
6656 mddev->safemode = 1;
6657 spin_unlock_irq(&mddev->write_lock);
6658 md_update_sb(mddev, 0);
00bcb4ac 6659 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6660 } else
6661 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6662
070dc6dd 6663 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6664 return -EAGAIN;
6665 else
6666 return 0;
2a2275d6
N		 6667}
6668EXPORT_SYMBOL_GPL(md_allow_write);
6669
b63d7c2e 6670void md_unplug(mddev_t *mddev)
252ac522
N		 6671{
6672 if (mddev->queue)
6673 blk_unplug(mddev->queue);
6674 if (mddev->plug)
6675 mddev->plug->unplug_fn(mddev->plug);
6676}
6677
1da177e4
LT		 6678#define SYNC_MARKS 10
6679#define SYNC_MARK_STEP (3*HZ)
29269553 6680void md_do_sync(mddev_t *mddev)
1da177e4
LT		 6681{
6682 mddev_t *mddev2;
6683 unsigned int currspeed = 0,
6684 window;
57afd89f 6685 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 6686 unsigned long mark[SYNC_MARKS];
6687 sector_t mark_cnt[SYNC_MARKS];
6688 int last_mark,m;
6689 struct list_head *tmp;
6690 sector_t last_check;
57afd89f 6691 int skipped = 0;
5fd6c1dc 6692 mdk_rdev_t *rdev;
61df9d91 6693 char *desc;
1da177e4
LT		 6694
6695 /* just incase thread restarts... */
6696 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6697 return;
5fd6c1dc
N		 6698 if (mddev->ro) /* never try to sync a read-only array */
6699 return;
1da177e4 6700
61df9d91
N		 6701 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6702 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6703 desc = "data-check";
6704 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6705 desc = "requested-resync";
6706 else
6707 desc = "resync";
6708 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6709 desc = "reshape";
6710 else
6711 desc = "recovery";
6712
1da177e4
LT		 6713 /* we overload curr_resync somewhat here.
6714 * 0 == not engaged in resync at all
6715 * 2 == checking that there is no conflict with another sync
6716 * 1 == like 2, but have yielded to allow conflicting resync to
6717 * commense
6718 * other == active in resync - this many blocks
6719 *
6720 * Before starting a resync we must have set curr_resync to
6721 * 2, and then checked that every "conflicting" array has curr_resync
6722 * less than ours. When we find one that is the same or higher
6723 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6724 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6725 * This will mean we have to start checking from the beginning again.
6726 *
6727 */
6728
6729 do {
6730 mddev->curr_resync = 2;
6731
6732 try_again:
404e4b43 6733 if (kthread_should_stop())
6985c43f 6734 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 6735
6736 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 6737 goto skip;
29ac4aa3 6738 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 6739 if (mddev2 == mddev)
6740 continue;
90b08710
BS		 6741 if (!mddev->parallel_resync
6742 && mddev2->curr_resync
6743 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 6744 DEFINE_WAIT(wq);
6745 if (mddev < mddev2 && mddev->curr_resync == 2) {
6746 /* arbitrarily yield */
6747 mddev->curr_resync = 1;
6748 wake_up(&resync_wait);
6749 }
6750 if (mddev > mddev2 && mddev->curr_resync == 1)
6751 /* no need to wait here, we can wait the next
6752 * time 'round when curr_resync == 2
6753 */
6754 continue;
9744197c
N		 6755 /* We need to wait 'interruptible' so as not to
6756 * contribute to the load average, and not to
6757 * be caught by 'softlockup'
6758 */
6759 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 6760 if (!kthread_should_stop() &&
8712e553 6761 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 6762 printk(KERN_INFO "md: delaying %s of %s"
6763 " until %s has finished (they"
1da177e4 6764 " share one or more physical units)\n",
61df9d91 6765 desc, mdname(mddev), mdname(mddev2));
1da177e4 6766 mddev_put(mddev2);
9744197c
N		 6767 if (signal_pending(current))
6768 flush_signals(current);
1da177e4
LT		 6769 schedule();
6770 finish_wait(&resync_wait, &wq);
6771 goto try_again;
6772 }
6773 finish_wait(&resync_wait, &wq);
6774 }
6775 }
6776 } while (mddev->curr_resync < 2);
6777
5fd6c1dc 6778 j = 0;
9d88883e 6779 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 6780 /* resync follows the size requested by the personality,
57afd89f 6781 * which defaults to physical size, but can be virtual size
1da177e4
LT		 6782 */
6783 max_sectors = mddev->resync_max_sectors;
9d88883e 6784 mddev->resync_mismatches = 0;
5fd6c1dc 6785 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 6786 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6787 j = mddev->resync_min;
6788 else if (!mddev->bitmap)
5fd6c1dc 6789 j = mddev->recovery_cp;
5e96ee65 6790
ccfcc3c1 6791 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 6792 max_sectors = mddev->dev_sectors;
5fd6c1dc 6793 else {
1da177e4 6794 /* recovery follows the physical size of devices */
58c0fed4 6795 max_sectors = mddev->dev_sectors;
5fd6c1dc 6796 j = MaxSector;
4e59ca7d
DW		 6797 rcu_read_lock();
6798 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 6799 if (rdev->raid_disk >= 0 &&
6800 !test_bit(Faulty, &rdev->flags) &&
6801 !test_bit(In_sync, &rdev->flags) &&
6802 rdev->recovery_offset < j)
6803 j = rdev->recovery_offset;
4e59ca7d 6804 rcu_read_unlock();
5fd6c1dc 6805 }
1da177e4 6806
61df9d91
N		 6807 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6808 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6809 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 6810 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 6811 "(but not more than %d KB/sec) for %s.\n",
6812 speed_max(mddev), desc);
1da177e4 6813
eea1bf38 6814 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 6815
57afd89f 6816 io_sectors = 0;
1da177e4
LT		 6817 for (m = 0; m < SYNC_MARKS; m++) {
6818 mark[m] = jiffies;
57afd89f 6819 mark_cnt[m] = io_sectors;
1da177e4
LT		 6820 }
6821 last_mark = 0;
6822 mddev->resync_mark = mark[last_mark];
6823 mddev->resync_mark_cnt = mark_cnt[last_mark];
6824
6825 /*
6826 * Tune reconstruction:
6827 */
6828 window = 32*(PAGE_SIZE/512);
6829 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6830 window/2,(unsigned long long) max_sectors/2);
6831
6832 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 6833 last_check = 0;
6834
6835 if (j>2) {
6836 printk(KERN_INFO
61df9d91
N		 6837 "md: resuming %s of %s from checkpoint.\n",
6838 desc, mdname(mddev));
1da177e4
LT		 6839 mddev->curr_resync = j;
6840 }
75d3da43 6841 mddev->curr_resync_completed = j;
1da177e4
LT		 6842
6843 while (j < max_sectors) {
57afd89f 6844 sector_t sectors;
1da177e4 6845
57afd89f 6846 skipped = 0;
97e4f42d 6847
7a91ee1f
N		 6848 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6849 ((mddev->curr_resync > mddev->curr_resync_completed &&
6850 (mddev->curr_resync - mddev->curr_resync_completed)
6851 > (max_sectors >> 4)) ||
6852 (j - mddev->curr_resync_completed)*2
6853 >= mddev->resync_max - mddev->curr_resync_completed
6854 )) {
97e4f42d 6855 /* time to update curr_resync_completed */
252ac522 6856 md_unplug(mddev);
97e4f42d
N		 6857 wait_event(mddev->recovery_wait,
6858 atomic_read(&mddev->recovery_active) == 0);
75d3da43 6859 mddev->curr_resync_completed = j;
070dc6dd 6860 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 6861 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 6862 }
acb180b0 6863
e62e58a5
N		 6864 while (j >= mddev->resync_max && !kthread_should_stop()) {
6865 /* As this condition is controlled by user-space,
6866 * we can block indefinitely, so use '_interruptible'
6867 * to avoid triggering warnings.
6868 */
6869 flush_signals(current); /* just in case */
6870 wait_event_interruptible(mddev->recovery_wait,
6871 mddev->resync_max > j
6872 || kthread_should_stop());
6873 }
acb180b0
N		 6874
6875 if (kthread_should_stop())
6876 goto interrupted;
6877
57afd89f 6878 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 6879 currspeed < speed_min(mddev));
57afd89f 6880 if (sectors == 0) {
dfc70645 6881 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6882 goto out;
6883 }
57afd89f
N		 6884
6885 if (!skipped) { /* actual IO requested */
6886 io_sectors += sectors;
6887 atomic_add(sectors, &mddev->recovery_active);
6888 }
6889
1da177e4
LT		 6890 j += sectors;
6891 if (j>1) mddev->curr_resync = j;
ff4e8d9a 6892 mddev->curr_mark_cnt = io_sectors;
d7603b7e
N		 6893 if (last_check == 0)
6894 /* this is the earliers that rebuilt will be
6895 * visible in /proc/mdstat
6896 */
6897 md_new_event(mddev);
57afd89f
N		 6898
6899 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 6900 continue;
6901
57afd89f 6902 last_check = io_sectors;
1da177e4 6903
dfc70645 6904 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4
LT		 6905 break;
6906
6907 repeat:
6908 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6909 /* step marks */
6910 int next = (last_mark+1) % SYNC_MARKS;
6911
6912 mddev->resync_mark = mark[next];
6913 mddev->resync_mark_cnt = mark_cnt[next];
6914 mark[next] = jiffies;
57afd89f 6915 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6916 last_mark = next;
6917 }
6918
6919
c6207277
N		 6920 if (kthread_should_stop())
6921 goto interrupted;
6922
1da177e4
LT		 6923
6924 /*
6925 * this loop exits only if either when we are slower than
6926 * the 'hard' speed limit, or the system was IO-idle for
6927 * a jiffy.
6928 * the system might be non-idle CPU-wise, but we only care
6929 * about not overloading the IO subsystem. (things like an
6930 * e2fsck being done on the RAID array should execute fast)
6931 */
252ac522 6932 md_unplug(mddev);
1da177e4
LT		 6933 cond_resched();
6934
57afd89f
N		 6935 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6936 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 6937
88202a0c
N		 6938 if (currspeed > speed_min(mddev)) {
6939 if ((currspeed > speed_max(mddev)) ||
eea1bf38 6940 !is_mddev_idle(mddev, 0)) {
c0e48521 6941 msleep(500);
1da177e4
LT		 6942 goto repeat;
6943 }
6944 }
6945 }
61df9d91 6946 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 6947 /*
6948 * this also signals 'finished resyncing' to md_stop
6949 */
6950 out:
252ac522 6951 md_unplug(mddev);
1da177e4
LT		 6952
6953 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6954
6955 /* tell personality that we are finished */
57afd89f 6956 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 6957
dfc70645 6958 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 6959 mddev->curr_resync > 2) {
6960 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6961 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6962 if (mddev->curr_resync >= mddev->recovery_cp) {
6963 printk(KERN_INFO
61df9d91
N		 6964 "md: checkpointing %s of %s.\n",
6965 desc, mdname(mddev));
5fd6c1dc
N		 6966 mddev->recovery_cp = mddev->curr_resync;
6967 }
6968 } else
6969 mddev->recovery_cp = MaxSector;
6970 } else {
6971 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6972 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 6973 rcu_read_lock();
6974 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 6975 if (rdev->raid_disk >= 0 &&
70fffd0b 6976 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 6977 !test_bit(Faulty, &rdev->flags) &&
6978 !test_bit(In_sync, &rdev->flags) &&
6979 rdev->recovery_offset < mddev->curr_resync)
6980 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 6981 rcu_read_unlock();
5fd6c1dc 6982 }
1da177e4 6983 }
17571284 6984 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 6985
1da177e4 6986 skip:
c07b70ad
N		 6987 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6988 /* We completed so min/max setting can be forgotten if used. */
6989 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6990 mddev->resync_min = 0;
6991 mddev->resync_max = MaxSector;
6992 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6993 mddev->resync_min = mddev->curr_resync_completed;
1da177e4 6994 mddev->curr_resync = 0;
efa59339
N		 6995 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6996 mddev->curr_resync_completed = 0;
c6207277 6997 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
1da177e4
LT		 6998 wake_up(&resync_wait);
6999 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7000 md_wakeup_thread(mddev->thread);
c6207277
N		 7001 return;
7002
7003 interrupted:
7004 /*
7005 * got a signal, exit.
7006 */
7007 printk(KERN_INFO
7008 "md: md_do_sync() got signal ... exiting\n");
7009 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7010 goto out;
7011
1da177e4 7012}
29269553 7013EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 7014
7015
b4c4c7b8
N		 7016static int remove_and_add_spares(mddev_t *mddev)
7017{
7018 mdk_rdev_t *rdev;
b4c4c7b8
N		 7019 int spares = 0;
7020
97e4f42d
N		 7021 mddev->curr_resync_completed = 0;
7022
159ec1fc 7023 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7024 if (rdev->raid_disk >= 0 &&
6bfe0b49 7025 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7026 (test_bit(Faulty, &rdev->flags) ||
7027 ! test_bit(In_sync, &rdev->flags)) &&
7028 atomic_read(&rdev->nr_pending)==0) {
7029 if (mddev->pers->hot_remove_disk(
7030 mddev, rdev->raid_disk)==0) {
7031 char nm[20];
7032 sprintf(nm,"rd%d", rdev->raid_disk);
7033 sysfs_remove_link(&mddev->kobj, nm);
7034 rdev->raid_disk = -1;
7035 }
7036 }
7037
4044ba58 7038 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
159ec1fc 7039 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7040 if (rdev->raid_disk >= 0 &&
e5427135
DW		 7041 !test_bit(In_sync, &rdev->flags) &&
7042 !test_bit(Blocked, &rdev->flags))
dfc70645 7043 spares++;
b4c4c7b8
N		 7044 if (rdev->raid_disk < 0
7045 && !test_bit(Faulty, &rdev->flags)) {
7046 rdev->recovery_offset = 0;
199050ea
NB		 7047 if (mddev->pers->
7048 hot_add_disk(mddev, rdev) == 0) {
b4c4c7b8
N		 7049 char nm[20];
7050 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 7051 if (sysfs_create_link(&mddev->kobj,
7052 &rdev->kobj, nm))
00bcb4ac 7053 /* failure here is OK */;
b4c4c7b8
N		 7054 spares++;
7055 md_new_event(mddev);
93be75ff 7056 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7057 } else
7058 break;
7059 }
dfc70645 7060 }
b4c4c7b8
N		 7061 }
7062 return spares;
7063}
7ebc0be7
N		 7064
7065static void reap_sync_thread(mddev_t *mddev)
7066{
7067 mdk_rdev_t *rdev;
7068
7069 /* resync has finished, collect result */
7070 md_unregister_thread(mddev->sync_thread);
7071 mddev->sync_thread = NULL;
7072 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7073 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7074 /* success...*/
7075 /* activate any spares */
7076 if (mddev->pers->spare_active(mddev))
7077 sysfs_notify(&mddev->kobj, NULL,
7078 "degraded");
7079 }
7080 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7081 mddev->pers->finish_reshape)
7082 mddev->pers->finish_reshape(mddev);
7083 md_update_sb(mddev, 1);
7084
7085 /* if array is no-longer degraded, then any saved_raid_disk
7086 * information must be scrapped
7087 */
7088 if (!mddev->degraded)
7089 list_for_each_entry(rdev, &mddev->disks, same_set)
7090 rdev->saved_raid_disk = -1;
7091
7092 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7093 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7094 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7095 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7096 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7097 /* flag recovery needed just to double check */
7098 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7099 sysfs_notify_dirent_safe(mddev->sysfs_action);
7100 md_new_event(mddev);
7101}
7102
1da177e4
LT		 7103/*
7104 * This routine is regularly called by all per-raid-array threads to
7105 * deal with generic issues like resync and super-block update.
7106 * Raid personalities that don't have a thread (linear/raid0) do not
7107 * need this as they never do any recovery or update the superblock.
7108 *
7109 * It does not do any resync itself, but rather "forks" off other threads
7110 * to do that as needed.
7111 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7112 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7113 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7114 * and wakeups up this thread which will reap the thread and finish up.
7115 * This thread also removes any faulty devices (with nr_pending == 0).
7116 *
7117 * The overall approach is:
7118 * 1/ if the superblock needs updating, update it.
7119 * 2/ If a recovery thread is running, don't do anything else.
7120 * 3/ If recovery has finished, clean up, possibly marking spares active.
7121 * 4/ If there are any faulty devices, remove them.
7122 * 5/ If array is degraded, try to add spares devices
7123 * 6/ If array has spares or is not in-sync, start a resync thread.
7124 */
7125void md_check_recovery(mddev_t *mddev)
7126{
5f40402d 7127 if (mddev->bitmap)
aa5cbd10 7128 bitmap_daemon_work(mddev);
1da177e4
LT		 7129
7130 if (mddev->ro)
7131 return;
fca4d848
N		 7132
7133 if (signal_pending(current)) {
31a59e34 7134 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7135 printk(KERN_INFO "md: %s in immediate safe mode\n",
7136 mdname(mddev));
7137 mddev->safemode = 2;
7138 }
7139 flush_signals(current);
7140 }
7141
c89a8eee
N		 7142 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7143 return;
1da177e4 7144 if ( ! (
126925c0 7145 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7146 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7147 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7148 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7149 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7150 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7151 ))
7152 return;
fca4d848 7153
df5b89b3 7154 if (mddev_trylock(mddev)) {
b4c4c7b8 7155 int spares = 0;
fca4d848 7156
c89a8eee
N		 7157 if (mddev->ro) {
7158 /* Only thing we do on a ro array is remove
7159 * failed devices.
7160 */
7161 remove_and_add_spares(mddev);
7162 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7163 goto unlock;
7164 }
7165
31a59e34 7166 if (!mddev->external) {
0fd62b86 7167 int did_change = 0;
31a59e34
N		 7168 spin_lock_irq(&mddev->write_lock);
7169 if (mddev->safemode &&
7170 !atomic_read(&mddev->writes_pending) &&
7171 !mddev->in_sync &&
7172 mddev->recovery_cp == MaxSector) {
7173 mddev->in_sync = 1;
0fd62b86 7174 did_change = 1;
070dc6dd 7175 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7176 }
7177 if (mddev->safemode == 1)
7178 mddev->safemode = 0;
7179 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7180 if (did_change)
00bcb4ac 7181 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7182 }
fca4d848 7183
850b2b42
N		 7184 if (mddev->flags)
7185 md_update_sb(mddev, 0);
06d91a5f 7186
1da177e4
LT		 7187 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7188 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7189 /* resync/recovery still happening */
7190 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7191 goto unlock;
7192 }
7193 if (mddev->sync_thread) {
7ebc0be7 7194 reap_sync_thread(mddev);
1da177e4
LT		 7195 goto unlock;
7196 }
72a23c21
NB		 7197 /* Set RUNNING before clearing NEEDED to avoid
7198 * any transients in the value of "sync_action".
7199 */
7200 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7201 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7202 /* Clear some bits that don't mean anything, but
7203 * might be left set
7204 */
24dd469d
N		 7205 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7206 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7207
5fd6c1dc
N		 7208 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7209 goto unlock;
1da177e4
LT		 7210 /* no recovery is running.
7211 * remove any failed drives, then
7212 * add spares if possible.
7213 * Spare are also removed and re-added, to allow
7214 * the personality to fail the re-add.
7215 */
1da177e4 7216
b4c4c7b8 7217 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7218 if (mddev->pers->check_reshape == NULL ||
7219 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7220 /* Cannot proceed */
7221 goto unlock;
7222 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7223 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7224 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7225 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7226 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7227 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7228 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7229 } else if (mddev->recovery_cp < MaxSector) {
7230 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7231 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7232 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7233 /* nothing to be done ... */
1da177e4 7234 goto unlock;
24dd469d 7235
1da177e4 7236 if (mddev->pers->sync_request) {
a654b9d8
N		 7237 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7238 /* We are adding a device or devices to an array
7239 * which has the bitmap stored on all devices.
7240 * So make sure all bitmap pages get written
7241 */
7242 bitmap_write_all(mddev->bitmap);
7243 }
1da177e4
LT		 7244 mddev->sync_thread = md_register_thread(md_do_sync,
7245 mddev,
0da3c619 7246 "resync");
1da177e4
LT		 7247 if (!mddev->sync_thread) {
7248 printk(KERN_ERR "%s: could not start resync"
7249 " thread...\n",
7250 mdname(mddev));
7251 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7252 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7253 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7254 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7255 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7256 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7257 } else
1da177e4 7258 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7259 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7260 md_new_event(mddev);
1da177e4
LT		 7261 }
7262 unlock:
72a23c21
NB		 7263 if (!mddev->sync_thread) {
7264 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7265 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7266 &mddev->recovery))
0c3573f1 7267 if (mddev->sysfs_action)
00bcb4ac 7268 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7269 }
1da177e4
LT		 7270 mddev_unlock(mddev);
7271 }
7272}
7273
6bfe0b49
DW		 7274void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7275{
00bcb4ac 7276 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49
DW		 7277 wait_event_timeout(rdev->blocked_wait,
7278 !test_bit(Blocked, &rdev->flags),
7279 msecs_to_jiffies(5000));
7280 rdev_dec_pending(rdev, mddev);
7281}
7282EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7283
75c96f85
AB		 7284static int md_notify_reboot(struct notifier_block *this,
7285 unsigned long code, void *x)
1da177e4
LT		 7286{
7287 struct list_head *tmp;
7288 mddev_t *mddev;
7289
7290 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7291
7292 printk(KERN_INFO "md: stopping all md devices.\n");
7293
29ac4aa3 7294 for_each_mddev(mddev, tmp)
c71d4887 7295 if (mddev_trylock(mddev)) {
2b25000b
N		 7296 /* Force a switch to readonly even array
7297 * appears to still be in use. Hence
7298 * the '100'.
7299 */
a4bd82d0 7300 md_set_readonly(mddev, 100);
c71d4887
NB		 7301 mddev_unlock(mddev);
7302 }
1da177e4
LT		 7303 /*
7304 * certain more exotic SCSI devices are known to be
7305 * volatile wrt too early system reboots. While the
7306 * right place to handle this issue is the given
7307 * driver, we do want to have a safe RAID driver ...
7308 */
7309 mdelay(1000*1);
7310 }
7311 return NOTIFY_DONE;
7312}
7313
75c96f85 7314static struct notifier_block md_notifier = {
1da177e4
LT		 7315 .notifier_call = md_notify_reboot,
7316 .next = NULL,
7317 .priority = INT_MAX, /* before any real devices */
7318};
7319
7320static void md_geninit(void)
7321{
1da177e4
LT		 7322 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7323
c7705f34 7324 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 7325}
7326
75c96f85 7327static int __init md_init(void)
1da177e4 7328{
e804ac78
TH		 7329 int ret = -ENOMEM;
7330
7331 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7332 if (!md_wq)
7333 goto err_wq;
7334
7335 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7336 if (!md_misc_wq)
7337 goto err_misc_wq;
7338
7339 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7340 goto err_md;
7341
7342 if ((ret = register_blkdev(0, "mdp")) < 0)
7343 goto err_mdp;
7344 mdp_major = ret;
7345
3dbd8c2e 7346 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 7347 md_probe, NULL, NULL);
7348 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 7349 md_probe, NULL, NULL);
7350
1da177e4 7351 register_reboot_notifier(&md_notifier);
0b4d4147 7352 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 7353
7354 md_geninit();
d710e138 7355 return 0;
1da177e4 7356
e804ac78
TH		 7357err_mdp:
7358 unregister_blkdev(MD_MAJOR, "md");
7359err_md:
7360 destroy_workqueue(md_misc_wq);
7361err_misc_wq:
7362 destroy_workqueue(md_wq);
7363err_wq:
7364 return ret;
7365}
1da177e4
LT		 7366
7367#ifndef MODULE
7368
7369/*
7370 * Searches all registered partitions for autorun RAID arrays
7371 * at boot time.
7372 */
4d936ec1
ME		 7373
7374static LIST_HEAD(all_detected_devices);
7375struct detected_devices_node {
7376 struct list_head list;
7377 dev_t dev;
7378};
1da177e4
LT		 7379
7380void md_autodetect_dev(dev_t dev)
7381{
4d936ec1
ME		 7382 struct detected_devices_node *node_detected_dev;
7383
7384 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7385 if (node_detected_dev) {
7386 node_detected_dev->dev = dev;
7387 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7388 } else {
7389 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7390 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7391 }
1da177e4
LT		 7392}
7393
7394
7395static void autostart_arrays(int part)
7396{
7397 mdk_rdev_t *rdev;
4d936ec1
ME		 7398 struct detected_devices_node *node_detected_dev;
7399 dev_t dev;
7400 int i_scanned, i_passed;
1da177e4 7401
4d936ec1
ME		 7402 i_scanned = 0;
7403 i_passed = 0;
1da177e4 7404
4d936ec1 7405 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 7406
4d936ec1
ME		 7407 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7408 i_scanned++;
7409 node_detected_dev = list_entry(all_detected_devices.next,
7410 struct detected_devices_node, list);
7411 list_del(&node_detected_dev->list);
7412 dev = node_detected_dev->dev;
7413 kfree(node_detected_dev);
df968c4e 7414 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 7415 if (IS_ERR(rdev))
7416 continue;
7417
b2d444d7 7418 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7419 MD_BUG();
7420 continue;
7421 }
d0fae18f 7422 set_bit(AutoDetected, &rdev->flags);
1da177e4 7423 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 7424 i_passed++;
1da177e4 7425 }
4d936ec1
ME		 7426
7427 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7428 i_scanned, i_passed);
1da177e4
LT		 7429
7430 autorun_devices(part);
7431}
7432
fdee8ae4 7433#endif /* !MODULE */
1da177e4
LT		 7434
7435static __exit void md_exit(void)
7436{
7437 mddev_t *mddev;
7438 struct list_head *tmp;
8ab5e4c1 7439
3dbd8c2e 7440 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 7441 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 7442
3dbd8c2e 7443 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 7444 unregister_blkdev(mdp_major, "mdp");
7445 unregister_reboot_notifier(&md_notifier);
7446 unregister_sysctl_table(raid_table_header);
7447 remove_proc_entry("mdstat", NULL);
29ac4aa3 7448 for_each_mddev(mddev, tmp) {
1da177e4 7449 export_array(mddev);
d3374825 7450 mddev->hold_active = 0;
1da177e4 7451 }
e804ac78
TH		 7452 destroy_workqueue(md_misc_wq);
7453 destroy_workqueue(md_wq);
1da177e4
LT		 7454}
7455
685784aa 7456subsys_initcall(md_init);
1da177e4
LT		 7457module_exit(md_exit)
7458
f91de92e
N		 7459static int get_ro(char *buffer, struct kernel_param *kp)
7460{
7461 return sprintf(buffer, "%d", start_readonly);
7462}
7463static int set_ro(const char *val, struct kernel_param *kp)
7464{
7465 char *e;
7466 int num = simple_strtoul(val, &e, 10);
7467 if (*val && (*e == '\0' || *e == '\n')) {
7468 start_readonly = num;
4dbcdc75 7469 return 0;
f91de92e
N		 7470 }
7471 return -EINVAL;
7472}
7473
80ca3a44
N		 7474module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7475module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 7476
efeb53c0 7477module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 7478
1da177e4
LT		 7479EXPORT_SYMBOL(register_md_personality);
7480EXPORT_SYMBOL(unregister_md_personality);
7481EXPORT_SYMBOL(md_error);
7482EXPORT_SYMBOL(md_done_sync);
7483EXPORT_SYMBOL(md_write_start);
7484EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 7485EXPORT_SYMBOL(md_register_thread);
7486EXPORT_SYMBOL(md_unregister_thread);
7487EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 7488EXPORT_SYMBOL(md_check_recovery);
7489MODULE_LICENSE("GPL");
0efb9e61 7490MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 7491MODULE_ALIAS("md");
72008652 7492MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)