projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: don't set_capacity before array is active.
[linux-2.6-block.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
2a48fc0a 39#include <linux/mutex.h>
1da177e4 40#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 41#include <linux/poll.h>
16f17b39 42#include <linux/ctype.h>
e7d2860b 43#include <linux/string.h>
fb4d8c76
N		 44#include <linux/hdreg.h>
45#include <linux/proc_fs.h>
46#include <linux/random.h>
47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
1da177e4
LT		 56
57#define DEBUG 0
58#define dprintk(x...) ((void)(DEBUG && printk(x)))
59
1da177e4 60#ifndef MODULE
d710e138 61static void autostart_arrays(int part);
1da177e4
LT		 62#endif
63
2604b703 64static LIST_HEAD(pers_list);
1da177e4
LT		 65static DEFINE_SPINLOCK(pers_lock);
66
5e56341d
AB		 67static void md_print_devices(void);
68
90b08710 69static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 70static struct workqueue_struct *md_wq;
71static struct workqueue_struct *md_misc_wq;
90b08710 72
5e56341d
AB		 73#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
74
1e50915f
RB		 75/*
76 * Default number of read corrections we'll attempt on an rdev
77 * before ejecting it from the array. We divide the read error
78 * count by 2 for every hour elapsed between read errors.
79 */
80#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 81/*
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 85 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
88 * idle IO detection.
89 *
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 91 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 92 */
93
94static int sysctl_speed_limit_min = 1000;
95static int sysctl_speed_limit_max = 200000;
88202a0c
N		 96static inline int speed_min(mddev_t *mddev)
97{
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
100}
101
102static inline int speed_max(mddev_t *mddev)
103{
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
106}
1da177e4
LT		 107
108static struct ctl_table_header *raid_table_header;
109
110static ctl_table raid_table[] = {
111 {
1da177e4
LT		 112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
80ca3a44 115 .mode = S_IRUGO|S_IWUSR,
6d456111 116 .proc_handler = proc_dointvec,
1da177e4
LT		 117 },
118 {
1da177e4
LT		 119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
80ca3a44 122 .mode = S_IRUGO|S_IWUSR,
6d456111 123 .proc_handler = proc_dointvec,
1da177e4 124 },
894d2491 125 { }
1da177e4
LT		 126};
127
128static ctl_table raid_dir_table[] = {
129 {
1da177e4
LT		 130 .procname = "raid",
131 .maxlen = 0,
80ca3a44 132 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 133 .child = raid_table,
134 },
894d2491 135 { }
1da177e4
LT		 136};
137
138static ctl_table raid_root_table[] = {
139 {
1da177e4
LT		 140 .procname = "dev",
141 .maxlen = 0,
142 .mode = 0555,
143 .child = raid_dir_table,
144 },
894d2491 145 { }
1da177e4
LT		 146};
147
83d5cde4 148static const struct block_device_operations md_fops;
1da177e4 149
f91de92e
N		 150static int start_readonly;
151
a167f663
N		 152/* bio_clone_mddev
153 * like bio_clone, but with a local bio set
154 */
155
156static void mddev_bio_destructor(struct bio *bio)
157{
158 mddev_t *mddev, **mddevp;
159
160 mddevp = (void*)bio;
161 mddev = mddevp[-1];
162
163 bio_free(bio, mddev->bio_set);
164}
165
166struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
167 mddev_t *mddev)
168{
169 struct bio *b;
170 mddev_t **mddevp;
171
172 if (!mddev || !mddev->bio_set)
173 return bio_alloc(gfp_mask, nr_iovecs);
174
175 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
176 mddev->bio_set);
177 if (!b)
178 return NULL;
179 mddevp = (void*)b;
180 mddevp[-1] = mddev;
181 b->bi_destructor = mddev_bio_destructor;
182 return b;
183}
184EXPORT_SYMBOL_GPL(bio_alloc_mddev);
185
186struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
187 mddev_t *mddev)
188{
189 struct bio *b;
190 mddev_t **mddevp;
191
192 if (!mddev || !mddev->bio_set)
193 return bio_clone(bio, gfp_mask);
194
195 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
196 mddev->bio_set);
197 if (!b)
198 return NULL;
199 mddevp = (void*)b;
200 mddevp[-1] = mddev;
201 b->bi_destructor = mddev_bio_destructor;
202 __bio_clone(b, bio);
203 if (bio_integrity(bio)) {
204 int ret;
205
206 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
207
208 if (ret < 0) {
209 bio_put(b);
210 return NULL;
211 }
212 }
213
214 return b;
215}
216EXPORT_SYMBOL_GPL(bio_clone_mddev);
217
d7603b7e
N		 218/*
219 * We have a system wide 'event count' that is incremented
220 * on any 'interesting' event, and readers of /proc/mdstat
221 * can use 'poll' or 'select' to find out when the event
222 * count increases.
223 *
224 * Events are:
225 * start array, stop array, error, add device, remove device,
226 * start build, activate spare
227 */
2989ddbd 228static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 229static atomic_t md_event_count;
29269553 230void md_new_event(mddev_t *mddev)
d7603b7e
N		 231{
232 atomic_inc(&md_event_count);
233 wake_up(&md_event_waiters);
234}
29269553 235EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 236
c331eb04
N		 237/* Alternate version that can be called from interrupts
238 * when calling sysfs_notify isn't needed.
239 */
05381954 240static void md_new_event_inintr(mddev_t *mddev)
c331eb04
N		 241{
242 atomic_inc(&md_event_count);
243 wake_up(&md_event_waiters);
244}
245
1da177e4
LT		 246/*
247 * Enables to iterate over all existing md arrays
248 * all_mddevs_lock protects this list.
249 */
250static LIST_HEAD(all_mddevs);
251static DEFINE_SPINLOCK(all_mddevs_lock);
252
253
254/*
255 * iterates through all used mddevs in the system.
256 * We take care to grab the all_mddevs_lock whenever navigating
257 * the list, and to always hold a refcount when unlocked.
258 * Any code which breaks out of this loop while own
259 * a reference to the current mddev and must mddev_put it.
260 */
29ac4aa3 261#define for_each_mddev(mddev,tmp) \
1da177e4
LT		 262 \
263 for (({ spin_lock(&all_mddevs_lock); \
264 tmp = all_mddevs.next; \
265 mddev = NULL;}); \
266 ({ if (tmp != &all_mddevs) \
267 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
268 spin_unlock(&all_mddevs_lock); \
269 if (mddev) mddev_put(mddev); \
270 mddev = list_entry(tmp, mddev_t, all_mddevs); \
271 tmp != &all_mddevs;}); \
272 ({ spin_lock(&all_mddevs_lock); \
273 tmp = tmp->next;}) \
274 )
275
276
409c57f3
N		 277/* Rather than calling directly into the personality make_request function,
278 * IO requests come here first so that we can check if the device is
279 * being suspended pending a reconfiguration.
280 * We hold a refcount over the call to ->make_request. By the time that
281 * call has finished, the bio has been linked into some internal structure
282 * and so is visible to ->quiesce(), so we don't need the refcount any more.
283 */
284static int md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 285{
49077326 286 const int rw = bio_data_dir(bio);
409c57f3
N		 287 mddev_t *mddev = q->queuedata;
288 int rv;
49077326 289 int cpu;
e91ece55 290 unsigned int sectors;
49077326 291
0ca69886
N		 292 if (mddev == NULL || mddev->pers == NULL
293 || !mddev->ready) {
409c57f3
N		 294 bio_io_error(bio);
295 return 0;
296 }
0ca69886 297 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 298 rcu_read_lock();
e9c7469b 299 if (mddev->suspended) {
409c57f3
N		 300 DEFINE_WAIT(__wait);
301 for (;;) {
302 prepare_to_wait(&mddev->sb_wait, &__wait,
303 TASK_UNINTERRUPTIBLE);
e9c7469b 304 if (!mddev->suspended)
409c57f3
N		 305 break;
306 rcu_read_unlock();
307 schedule();
308 rcu_read_lock();
309 }
310 finish_wait(&mddev->sb_wait, &__wait);
311 }
312 atomic_inc(&mddev->active_io);
313 rcu_read_unlock();
49077326 314
e91ece55
CM		 315 /*
316 * save the sectors now since our bio can
317 * go away inside make_request
318 */
319 sectors = bio_sectors(bio);
21a52c6d 320 rv = mddev->pers->make_request(mddev, bio);
49077326
N		 321
322 cpu = part_stat_lock();
323 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 324 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 325 part_stat_unlock();
326
409c57f3
N		 327 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
328 wake_up(&mddev->sb_wait);
329
330 return rv;
331}
332
9e35b99c
N		 333/* mddev_suspend makes sure no new requests are submitted
334 * to the device, and that any requests that have been submitted
335 * are completely handled.
336 * Once ->stop is called and completes, the module will be completely
337 * unused.
338 */
390ee602 339void mddev_suspend(mddev_t *mddev)
409c57f3
N		 340{
341 BUG_ON(mddev->suspended);
342 mddev->suspended = 1;
343 synchronize_rcu();
344 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
345 mddev->pers->quiesce(mddev, 1);
409c57f3 346}
390ee602 347EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 348
390ee602 349void mddev_resume(mddev_t *mddev)
409c57f3
N		 350{
351 mddev->suspended = 0;
352 wake_up(&mddev->sb_wait);
353 mddev->pers->quiesce(mddev, 0);
1da177e4 354}
390ee602 355EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 356
3fa841d7
N		 357int mddev_congested(mddev_t *mddev, int bits)
358{
359 return mddev->suspended;
360}
361EXPORT_SYMBOL(mddev_congested);
362
a2826aa9 363/*
e9c7469b 364 * Generic flush handling for md
a2826aa9
N		 365 */
366
e9c7469b 367static void md_end_flush(struct bio *bio, int err)
a2826aa9
N		 368{
369 mdk_rdev_t *rdev = bio->bi_private;
370 mddev_t *mddev = rdev->mddev;
a2826aa9
N		 371
372 rdev_dec_pending(rdev, mddev);
373
374 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 375 /* The pre-request flush has finished */
e804ac78 376 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 377 }
378 bio_put(bio);
379}
380
a7a07e69
N		 381static void md_submit_flush_data(struct work_struct *ws);
382
a035fc3e 383static void submit_flushes(struct work_struct *ws)
a2826aa9 384{
a035fc3e 385 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
a2826aa9
N		 386 mdk_rdev_t *rdev;
387
a7a07e69
N		 388 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
389 atomic_set(&mddev->flush_pending, 1);
a2826aa9
N		 390 rcu_read_lock();
391 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
392 if (rdev->raid_disk >= 0 &&
393 !test_bit(Faulty, &rdev->flags)) {
394 /* Take two references, one is dropped
395 * when request finishes, one after
396 * we reclaim rcu_read_lock
397 */
398 struct bio *bi;
399 atomic_inc(&rdev->nr_pending);
400 atomic_inc(&rdev->nr_pending);
401 rcu_read_unlock();
a167f663 402 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 403 bi->bi_end_io = md_end_flush;
a2826aa9
N		 404 bi->bi_private = rdev;
405 bi->bi_bdev = rdev->bdev;
406 atomic_inc(&mddev->flush_pending);
e9c7469b 407 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 408 rcu_read_lock();
409 rdev_dec_pending(rdev, mddev);
410 }
411 rcu_read_unlock();
a7a07e69
N		 412 if (atomic_dec_and_test(&mddev->flush_pending))
413 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 414}
415
e9c7469b 416static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 417{
e9c7469b
TH		 418 mddev_t *mddev = container_of(ws, mddev_t, flush_work);
419 struct bio *bio = mddev->flush_bio;
a2826aa9 420
e9c7469b 421 if (bio->bi_size == 0)
a2826aa9
N		 422 /* an empty barrier - all done */
423 bio_endio(bio, 0);
424 else {
e9c7469b 425 bio->bi_rw &= ~REQ_FLUSH;
21a52c6d 426 if (mddev->pers->make_request(mddev, bio))
a2826aa9 427 generic_make_request(bio);
a2826aa9 428 }
2b74e12e
N		 429
430 mddev->flush_bio = NULL;
431 wake_up(&mddev->sb_wait);
a2826aa9
N		 432}
433
e9c7469b 434void md_flush_request(mddev_t *mddev, struct bio *bio)
a2826aa9
N		 435{
436 spin_lock_irq(&mddev->write_lock);
437 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 438 !mddev->flush_bio,
a2826aa9 439 mddev->write_lock, /*nothing*/);
e9c7469b 440 mddev->flush_bio = bio;
a2826aa9
N		 441 spin_unlock_irq(&mddev->write_lock);
442
a035fc3e
N		 443 INIT_WORK(&mddev->flush_work, submit_flushes);
444 queue_work(md_wq, &mddev->flush_work);
a2826aa9 445}
e9c7469b 446EXPORT_SYMBOL(md_flush_request);
409c57f3 447
2ac87401
N		 448/* Support for plugging.
449 * This mirrors the plugging support in request_queue, but does not
450 * require having a whole queue
451 */
452static void plugger_work(struct work_struct *work)
453{
454 struct plug_handle *plug =
455 container_of(work, struct plug_handle, unplug_work);
456 plug->unplug_fn(plug);
457}
458static void plugger_timeout(unsigned long data)
459{
460 struct plug_handle *plug = (void *)data;
461 kblockd_schedule_work(NULL, &plug->unplug_work);
462}
463void plugger_init(struct plug_handle *plug,
464 void (*unplug_fn)(struct plug_handle *))
465{
466 plug->unplug_flag = 0;
467 plug->unplug_fn = unplug_fn;
468 init_timer(&plug->unplug_timer);
469 plug->unplug_timer.function = plugger_timeout;
470 plug->unplug_timer.data = (unsigned long)plug;
471 INIT_WORK(&plug->unplug_work, plugger_work);
472}
473EXPORT_SYMBOL_GPL(plugger_init);
474
475void plugger_set_plug(struct plug_handle *plug)
476{
477 if (!test_and_set_bit(PLUGGED_FLAG, &plug->unplug_flag))
478 mod_timer(&plug->unplug_timer, jiffies + msecs_to_jiffies(3)+1);
479}
480EXPORT_SYMBOL_GPL(plugger_set_plug);
481
482int plugger_remove_plug(struct plug_handle *plug)
483{
484 if (test_and_clear_bit(PLUGGED_FLAG, &plug->unplug_flag)) {
485 del_timer(&plug->unplug_timer);
486 return 1;
487 } else
488 return 0;
489}
490EXPORT_SYMBOL_GPL(plugger_remove_plug);
491
492
1da177e4
LT		 493static inline mddev_t *mddev_get(mddev_t *mddev)
494{
495 atomic_inc(&mddev->active);
496 return mddev;
497}
498
5fd3a17e 499static void mddev_delayed_delete(struct work_struct *ws);
d3374825 500
1da177e4
LT		 501static void mddev_put(mddev_t *mddev)
502{
a167f663
N		 503 struct bio_set *bs = NULL;
504
1da177e4
LT		 505 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
506 return;
d3374825 507 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 508 mddev->ctime == 0 && !mddev->hold_active) {
509 /* Array is not configured at all, and not held active,
510 * so destroy it */
1da177e4 511 list_del(&mddev->all_mddevs);
a167f663
N		 512 bs = mddev->bio_set;
513 mddev->bio_set = NULL;
d3374825 514 if (mddev->gendisk) {
e804ac78
TH		 515 /* We did a probe so need to clean up. Call
516 * queue_work inside the spinlock so that
517 * flush_workqueue() after mddev_find will
518 * succeed in waiting for the work to be done.
d3374825
N		 519 */
520 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 521 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 522 } else
523 kfree(mddev);
524 }
525 spin_unlock(&all_mddevs_lock);
a167f663
N		 526 if (bs)
527 bioset_free(bs);
1da177e4
LT		 528}
529
390ee602 530void mddev_init(mddev_t *mddev)
fafd7fb0
N		 531{
532 mutex_init(&mddev->open_mutex);
533 mutex_init(&mddev->reconfig_mutex);
534 mutex_init(&mddev->bitmap_info.mutex);
535 INIT_LIST_HEAD(&mddev->disks);
536 INIT_LIST_HEAD(&mddev->all_mddevs);
537 init_timer(&mddev->safemode_timer);
538 atomic_set(&mddev->active, 1);
539 atomic_set(&mddev->openers, 0);
540 atomic_set(&mddev->active_io, 0);
541 spin_lock_init(&mddev->write_lock);
542 atomic_set(&mddev->flush_pending, 0);
543 init_waitqueue_head(&mddev->sb_wait);
544 init_waitqueue_head(&mddev->recovery_wait);
545 mddev->reshape_position = MaxSector;
546 mddev->resync_min = 0;
547 mddev->resync_max = MaxSector;
548 mddev->level = LEVEL_NONE;
549}
390ee602 550EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 551
1da177e4
LT		 552static mddev_t * mddev_find(dev_t unit)
553{
554 mddev_t *mddev, *new = NULL;
555
556 retry:
557 spin_lock(&all_mddevs_lock);
efeb53c0
N		 558
559 if (unit) {
560 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
561 if (mddev->unit == unit) {
562 mddev_get(mddev);
563 spin_unlock(&all_mddevs_lock);
564 kfree(new);
565 return mddev;
566 }
567
568 if (new) {
569 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 570 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 571 new->hold_active = UNTIL_IOCTL;
572 return new;
1da177e4 573 }
efeb53c0
N		 574 } else if (new) {
575 /* find an unused unit number */
576 static int next_minor = 512;
577 int start = next_minor;
578 int is_free = 0;
579 int dev = 0;
580 while (!is_free) {
581 dev = MKDEV(MD_MAJOR, next_minor);
582 next_minor++;
583 if (next_minor > MINORMASK)
584 next_minor = 0;
585 if (next_minor == start) {
586 /* Oh dear, all in use. */
587 spin_unlock(&all_mddevs_lock);
588 kfree(new);
589 return NULL;
590 }
591
592 is_free = 1;
593 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
594 if (mddev->unit == dev) {
595 is_free = 0;
596 break;
597 }
598 }
599 new->unit = dev;
600 new->md_minor = MINOR(dev);
601 new->hold_active = UNTIL_STOP;
1da177e4
LT		 602 list_add(&new->all_mddevs, &all_mddevs);
603 spin_unlock(&all_mddevs_lock);
604 return new;
605 }
606 spin_unlock(&all_mddevs_lock);
607
9ffae0cf 608 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 609 if (!new)
610 return NULL;
611
1da177e4
LT		 612 new->unit = unit;
613 if (MAJOR(unit) == MD_MAJOR)
614 new->md_minor = MINOR(unit);
615 else
616 new->md_minor = MINOR(unit) >> MdpMinorShift;
617
fafd7fb0 618 mddev_init(new);
1da177e4 619
1da177e4
LT		 620 goto retry;
621}
622
623static inline int mddev_lock(mddev_t * mddev)
624{
df5b89b3 625 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 626}
627
b522adcd
DW		 628static inline int mddev_is_locked(mddev_t *mddev)
629{
630 return mutex_is_locked(&mddev->reconfig_mutex);
631}
632
1da177e4
LT		 633static inline int mddev_trylock(mddev_t * mddev)
634{
df5b89b3 635 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 636}
637
b6eb127d
N		 638static struct attribute_group md_redundancy_group;
639
a64c876f 640static void mddev_unlock(mddev_t * mddev)
1da177e4 641{
a64c876f 642 if (mddev->to_remove) {
b6eb127d
N		 643 /* These cannot be removed under reconfig_mutex as
644 * an access to the files will try to take reconfig_mutex
645 * while holding the file unremovable, which leads to
646 * a deadlock.
bb4f1e9d
N		 647 * So hold set sysfs_active while the remove in happeing,
648 * and anything else which might set ->to_remove or my
649 * otherwise change the sysfs namespace will fail with
650 * -EBUSY if sysfs_active is still set.
651 * We set sysfs_active under reconfig_mutex and elsewhere
652 * test it under the same mutex to ensure its correct value
653 * is seen.
b6eb127d 654 */
a64c876f
N		 655 struct attribute_group *to_remove = mddev->to_remove;
656 mddev->to_remove = NULL;
bb4f1e9d 657 mddev->sysfs_active = 1;
b6eb127d
N		 658 mutex_unlock(&mddev->reconfig_mutex);
659
00bcb4ac
N		 660 if (mddev->kobj.sd) {
661 if (to_remove != &md_redundancy_group)
662 sysfs_remove_group(&mddev->kobj, to_remove);
663 if (mddev->pers == NULL ||
664 mddev->pers->sync_request == NULL) {
665 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
666 if (mddev->sysfs_action)
667 sysfs_put(mddev->sysfs_action);
668 mddev->sysfs_action = NULL;
669 }
a64c876f 670 }
bb4f1e9d 671 mddev->sysfs_active = 0;
b6eb127d
N		 672 } else
673 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 674
005eca5e 675 md_wakeup_thread(mddev->thread);
1da177e4
LT		 676}
677
2989ddbd 678static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
1da177e4 679{
159ec1fc 680 mdk_rdev_t *rdev;
1da177e4 681
159ec1fc 682 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 683 if (rdev->desc_nr == nr)
684 return rdev;
159ec1fc 685
1da177e4
LT		 686 return NULL;
687}
688
689static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
690{
1da177e4
LT		 691 mdk_rdev_t *rdev;
692
159ec1fc 693 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 694 if (rdev->bdev->bd_dev == dev)
695 return rdev;
159ec1fc 696
1da177e4
LT		 697 return NULL;
698}
699
d9d166c2 700static struct mdk_personality *find_pers(int level, char *clevel)
2604b703
N		 701{
702 struct mdk_personality *pers;
d9d166c2
N		 703 list_for_each_entry(pers, &pers_list, list) {
704 if (level != LEVEL_NONE && pers->level == level)
2604b703 705 return pers;
d9d166c2
N		 706 if (strcmp(pers->name, clevel)==0)
707 return pers;
708 }
2604b703
N		 709 return NULL;
710}
711
b73df2d3 712/* return the offset of the super block in 512byte sectors */
57b2caa3 713static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
1da177e4 714{
57b2caa3 715 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 716 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 717}
718
1da177e4
LT		 719static int alloc_disk_sb(mdk_rdev_t * rdev)
720{
721 if (rdev->sb_page)
722 MD_BUG();
723
724 rdev->sb_page = alloc_page(GFP_KERNEL);
725 if (!rdev->sb_page) {
726 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 727 return -ENOMEM;
1da177e4
LT		 728 }
729
730 return 0;
731}
732
733static void free_disk_sb(mdk_rdev_t * rdev)
734{
735 if (rdev->sb_page) {
2d1f3b5d 736 put_page(rdev->sb_page);
1da177e4
LT		 737 rdev->sb_loaded = 0;
738 rdev->sb_page = NULL;
0f420358 739 rdev->sb_start = 0;
dd8ac336 740 rdev->sectors = 0;
1da177e4
LT		 741 }
742}
743
744
6712ecf8 745static void super_written(struct bio *bio, int error)
7bfa19f2
N		 746{
747 mdk_rdev_t *rdev = bio->bi_private;
a9701a30 748 mddev_t *mddev = rdev->mddev;
7bfa19f2 749
3a0f5bbb
N		 750 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
751 printk("md: super_written gets error=%d, uptodate=%d\n",
752 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
753 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 754 md_error(mddev, rdev);
3a0f5bbb 755 }
7bfa19f2 756
a9701a30
N		 757 if (atomic_dec_and_test(&mddev->pending_writes))
758 wake_up(&mddev->sb_wait);
f8b58edf 759 bio_put(bio);
7bfa19f2
N		 760}
761
762void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
763 sector_t sector, int size, struct page *page)
764{
765 /* write first size bytes of page to sector of rdev
766 * Increment mddev->pending_writes before returning
767 * and decrement it on completion, waking up sb_wait
768 * if zero is reached.
769 * If an error occurred, call md_error
770 */
a167f663 771 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 772
a6ff7e08 773 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 774 bio->bi_sector = sector;
775 bio_add_page(bio, page, size, 0);
776 bio->bi_private = rdev;
777 bio->bi_end_io = super_written;
a9701a30 778
7bfa19f2 779 atomic_inc(&mddev->pending_writes);
e9c7469b
TH		 780 submit_bio(REQ_WRITE | REQ_SYNC | REQ_UNPLUG | REQ_FLUSH | REQ_FUA,
781 bio);
a9701a30
N		 782}
783
784void md_super_wait(mddev_t *mddev)
785{
e9c7469b 786 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 787 DEFINE_WAIT(wq);
788 for(;;) {
789 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
790 if (atomic_read(&mddev->pending_writes)==0)
791 break;
a9701a30
N		 792 schedule();
793 }
794 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 795}
796
6712ecf8 797static void bi_complete(struct bio *bio, int error)
1da177e4 798{
1da177e4 799 complete((struct completion*)bio->bi_private);
1da177e4
LT		 800}
801
2b193363 802int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
ccebd4c4 803 struct page *page, int rw, bool metadata_op)
1da177e4 804{
a167f663 805 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 806 struct completion event;
807 int ret;
808
7b6d91da 809 rw |= REQ_SYNC | REQ_UNPLUG;
1da177e4 810
a6ff7e08
JB		 811 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
812 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 813 if (metadata_op)
814 bio->bi_sector = sector + rdev->sb_start;
815 else
816 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 817 bio_add_page(bio, page, size, 0);
818 init_completion(&event);
819 bio->bi_private = &event;
820 bio->bi_end_io = bi_complete;
821 submit_bio(rw, bio);
822 wait_for_completion(&event);
823
824 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
825 bio_put(bio);
826 return ret;
827}
a8745db2 828EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 829
0002b271 830static int read_disk_sb(mdk_rdev_t * rdev, int size)
1da177e4
LT		 831{
832 char b[BDEVNAME_SIZE];
833 if (!rdev->sb_page) {
834 MD_BUG();
835 return -EINVAL;
836 }
837 if (rdev->sb_loaded)
838 return 0;
839
840
ccebd4c4 841 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 842 goto fail;
843 rdev->sb_loaded = 1;
844 return 0;
845
846fail:
847 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
848 bdevname(rdev->bdev,b));
849 return -EINVAL;
850}
851
852static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
853{
05710466
AN		 854 return sb1->set_uuid0 == sb2->set_uuid0 &&
855 sb1->set_uuid1 == sb2->set_uuid1 &&
856 sb1->set_uuid2 == sb2->set_uuid2 &&
857 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 858}
859
1da177e4
LT		 860static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
861{
862 int ret;
863 mdp_super_t *tmp1, *tmp2;
864
865 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
866 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
867
868 if (!tmp1 || !tmp2) {
869 ret = 0;
35020f1a 870 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 871 goto abort;
872 }
873
874 *tmp1 = *sb1;
875 *tmp2 = *sb2;
876
877 /*
878 * nr_disks is not constant
879 */
880 tmp1->nr_disks = 0;
881 tmp2->nr_disks = 0;
882
ce0c8e05 883 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 884abort:
990a8baf
JJ		 885 kfree(tmp1);
886 kfree(tmp2);
1da177e4
LT		 887 return ret;
888}
889
4d167f09
N		 890
891static u32 md_csum_fold(u32 csum)
892{
893 csum = (csum & 0xffff) + (csum >> 16);
894 return (csum & 0xffff) + (csum >> 16);
895}
896
1da177e4
LT		 897static unsigned int calc_sb_csum(mdp_super_t * sb)
898{
4d167f09
N		 899 u64 newcsum = 0;
900 u32 *sb32 = (u32*)sb;
901 int i;
1da177e4
LT		 902 unsigned int disk_csum, csum;
903
904 disk_csum = sb->sb_csum;
905 sb->sb_csum = 0;
4d167f09
N		 906
907 for (i = 0; i < MD_SB_BYTES/4 ; i++)
908 newcsum += sb32[i];
909 csum = (newcsum & 0xffffffff) + (newcsum>>32);
910
911
912#ifdef CONFIG_ALPHA
913 /* This used to use csum_partial, which was wrong for several
914 * reasons including that different results are returned on
915 * different architectures. It isn't critical that we get exactly
916 * the same return value as before (we always csum_fold before
917 * testing, and that removes any differences). However as we
918 * know that csum_partial always returned a 16bit value on
919 * alphas, do a fold to maximise conformity to previous behaviour.
920 */
921 sb->sb_csum = md_csum_fold(disk_csum);
922#else
1da177e4 923 sb->sb_csum = disk_csum;
4d167f09 924#endif
1da177e4
LT		 925 return csum;
926}
927
928
929/*
930 * Handle superblock details.
931 * We want to be able to handle multiple superblock formats
932 * so we have a common interface to them all, and an array of
933 * different handlers.
934 * We rely on user-space to write the initial superblock, and support
935 * reading and updating of superblocks.
936 * Interface methods are:
937 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
938 * loads and validates a superblock on dev.
939 * if refdev != NULL, compare superblocks on both devices
940 * Return:
941 * 0 - dev has a superblock that is compatible with refdev
942 * 1 - dev has a superblock that is compatible and newer than refdev
943 * so dev should be used as the refdev in future
944 * -EINVAL superblock incompatible or invalid
945 * -othererror e.g. -EIO
946 *
947 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
948 * Verify that dev is acceptable into mddev.
949 * The first time, mddev->raid_disks will be 0, and data from
950 * dev should be merged in. Subsequent calls check that dev
951 * is new enough. Return 0 or -EINVAL
952 *
953 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
954 * Update the superblock for rdev with data in mddev
955 * This does not write to disc.
956 *
957 */
958
959struct super_type {
0cd17fec
CW		 960 char *name;
961 struct module *owner;
962 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
963 int minor_version);
964 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
965 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
966 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
15f4a5fd 967 sector_t num_sectors);
1da177e4
LT		 968};
969
0894cc30
AN		 970/*
971 * Check that the given mddev has no bitmap.
972 *
973 * This function is called from the run method of all personalities that do not
974 * support bitmaps. It prints an error message and returns non-zero if mddev
975 * has a bitmap. Otherwise, it returns 0.
976 *
977 */
978int md_check_no_bitmap(mddev_t *mddev)
979{
c3d9714e 980 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 981 return 0;
982 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
983 mdname(mddev), mddev->pers->name);
984 return 1;
985}
986EXPORT_SYMBOL(md_check_no_bitmap);
987
1da177e4
LT		 988/*
989 * load_super for 0.90.0
990 */
991static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
992{
993 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
994 mdp_super_t *sb;
995 int ret;
1da177e4
LT		 996
997 /*
0f420358 998 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 999 * it's at the end of the disk.
1000 *
1001 * It also happens to be a multiple of 4Kb.
1002 */
57b2caa3 1003 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1004
0002b271 1005 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1006 if (ret) return ret;
1007
1008 ret = -EINVAL;
1009
1010 bdevname(rdev->bdev, b);
1011 sb = (mdp_super_t*)page_address(rdev->sb_page);
1012
1013 if (sb->md_magic != MD_SB_MAGIC) {
1014 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1015 b);
1016 goto abort;
1017 }
1018
1019 if (sb->major_version != 0 ||
f6705578
N		 1020 sb->minor_version < 90 ||
1021 sb->minor_version > 91) {
1da177e4
LT		 1022 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1023 sb->major_version, sb->minor_version,
1024 b);
1025 goto abort;
1026 }
1027
1028 if (sb->raid_disks <= 0)
1029 goto abort;
1030
4d167f09 1031 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1032 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1033 b);
1034 goto abort;
1035 }
1036
1037 rdev->preferred_minor = sb->md_minor;
1038 rdev->data_offset = 0;
0002b271 1039 rdev->sb_size = MD_SB_BYTES;
1da177e4
LT		 1040
1041 if (sb->level == LEVEL_MULTIPATH)
1042 rdev->desc_nr = -1;
1043 else
1044 rdev->desc_nr = sb->this_disk.number;
1045
9a7b2b0f 1046 if (!refdev) {
1da177e4 1047 ret = 1;
9a7b2b0f 1048 } else {
1da177e4
LT		 1049 __u64 ev1, ev2;
1050 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
1051 if (!uuid_equal(refsb, sb)) {
1052 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1053 b, bdevname(refdev->bdev,b2));
1054 goto abort;
1055 }
1056 if (!sb_equal(refsb, sb)) {
1057 printk(KERN_WARNING "md: %s has same UUID"
1058 " but different superblock to %s\n",
1059 b, bdevname(refdev->bdev, b2));
1060 goto abort;
1061 }
1062 ev1 = md_event(sb);
1063 ev2 = md_event(refsb);
1064 if (ev1 > ev2)
1065 ret = 1;
1066 else
1067 ret = 0;
1068 }
8190e754 1069 rdev->sectors = rdev->sb_start;
1da177e4 1070
dd8ac336 1071 if (rdev->sectors < sb->size * 2 && sb->level > 1)
2bf071bf
N		 1072 /* "this cannot possibly happen" ... */
1073 ret = -EINVAL;
1074
1da177e4
LT		 1075 abort:
1076 return ret;
1077}
1078
1079/*
1080 * validate_super for 0.90.0
1081 */
1082static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1083{
1084 mdp_disk_t *desc;
1085 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
07d84d10 1086 __u64 ev1 = md_event(sb);
1da177e4 1087
41158c7e 1088 rdev->raid_disk = -1;
c5d79adb
N		 1089 clear_bit(Faulty, &rdev->flags);
1090 clear_bit(In_sync, &rdev->flags);
1091 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1092
1da177e4
LT		 1093 if (mddev->raid_disks == 0) {
1094 mddev->major_version = 0;
1095 mddev->minor_version = sb->minor_version;
1096 mddev->patch_version = sb->patch_version;
e691063a 1097 mddev->external = 0;
9d8f0363 1098 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1099 mddev->ctime = sb->ctime;
1100 mddev->utime = sb->utime;
1101 mddev->level = sb->level;
d9d166c2 1102 mddev->clevel[0] = 0;
1da177e4
LT		 1103 mddev->layout = sb->layout;
1104 mddev->raid_disks = sb->raid_disks;
58c0fed4 1105 mddev->dev_sectors = sb->size * 2;
07d84d10 1106 mddev->events = ev1;
c3d9714e
N		 1107 mddev->bitmap_info.offset = 0;
1108 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1109
f6705578
N		 1110 if (mddev->minor_version >= 91) {
1111 mddev->reshape_position = sb->reshape_position;
1112 mddev->delta_disks = sb->delta_disks;
1113 mddev->new_level = sb->new_level;
1114 mddev->new_layout = sb->new_layout;
664e7c41 1115 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1116 } else {
1117 mddev->reshape_position = MaxSector;
1118 mddev->delta_disks = 0;
1119 mddev->new_level = mddev->level;
1120 mddev->new_layout = mddev->layout;
664e7c41 1121 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1122 }
1123
1da177e4
LT		 1124 if (sb->state & (1<<MD_SB_CLEAN))
1125 mddev->recovery_cp = MaxSector;
1126 else {
1127 if (sb->events_hi == sb->cp_events_hi &&
1128 sb->events_lo == sb->cp_events_lo) {
1129 mddev->recovery_cp = sb->recovery_cp;
1130 } else
1131 mddev->recovery_cp = 0;
1132 }
1133
1134 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1135 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1136 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1137 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1138
1139 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1140
1141 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1142 mddev->bitmap_info.file == NULL)
1143 mddev->bitmap_info.offset =
1144 mddev->bitmap_info.default_offset;
a654b9d8 1145
41158c7e 1146 } else if (mddev->pers == NULL) {
be6800a7
N		 1147 /* Insist on good event counter while assembling, except
1148 * for spares (which don't need an event count) */
1da177e4 1149 ++ev1;
be6800a7
N		 1150 if (sb->disks[rdev->desc_nr].state & (
1151 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1152 if (ev1 < mddev->events)
1153 return -EINVAL;
41158c7e
N		 1154 } else if (mddev->bitmap) {
1155 /* if adding to array with a bitmap, then we can accept an
1156 * older device ... but not too old.
1157 */
41158c7e
N		 1158 if (ev1 < mddev->bitmap->events_cleared)
1159 return 0;
07d84d10
N		 1160 } else {
1161 if (ev1 < mddev->events)
1162 /* just a hot-add of a new device, leave raid_disk at -1 */
1163 return 0;
1164 }
41158c7e 1165
1da177e4 1166 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1167 desc = sb->disks + rdev->desc_nr;
1168
1169 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1170 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1171 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1172 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1173 set_bit(In_sync, &rdev->flags);
1da177e4 1174 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1175 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1176 /* active but not in sync implies recovery up to
1177 * reshape position. We don't know exactly where
1178 * that is, so set to zero for now */
1179 if (mddev->minor_version >= 91) {
1180 rdev->recovery_offset = 0;
1181 rdev->raid_disk = desc->raid_disk;
1182 }
1da177e4 1183 }
8ddf9efe
N		 1184 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1185 set_bit(WriteMostly, &rdev->flags);
41158c7e 1186 } else /* MULTIPATH are always insync */
b2d444d7 1187 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1188 return 0;
1189}
1190
1191/*
1192 * sync_super for 0.90.0
1193 */
1194static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1195{
1196 mdp_super_t *sb;
1da177e4
LT		 1197 mdk_rdev_t *rdev2;
1198 int next_spare = mddev->raid_disks;
19133a42 1199
1da177e4
LT		 1200
1201 /* make rdev->sb match mddev data..
1202 *
1203 * 1/ zero out disks
1204 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1205 * 3/ any empty disks < next_spare become removed
1206 *
1207 * disks[0] gets initialised to REMOVED because
1208 * we cannot be sure from other fields if it has
1209 * been initialised or not.
1210 */
1211 int i;
1212 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1213
61181565
N		 1214 rdev->sb_size = MD_SB_BYTES;
1215
1da177e4
LT		 1216 sb = (mdp_super_t*)page_address(rdev->sb_page);
1217
1218 memset(sb, 0, sizeof(*sb));
1219
1220 sb->md_magic = MD_SB_MAGIC;
1221 sb->major_version = mddev->major_version;
1da177e4
LT		 1222 sb->patch_version = mddev->patch_version;
1223 sb->gvalid_words = 0; /* ignored */
1224 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1225 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1226 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1227 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1228
1229 sb->ctime = mddev->ctime;
1230 sb->level = mddev->level;
58c0fed4 1231 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1232 sb->raid_disks = mddev->raid_disks;
1233 sb->md_minor = mddev->md_minor;
e691063a 1234 sb->not_persistent = 0;
1da177e4
LT		 1235 sb->utime = mddev->utime;
1236 sb->state = 0;
1237 sb->events_hi = (mddev->events>>32);
1238 sb->events_lo = (u32)mddev->events;
1239
f6705578
N		 1240 if (mddev->reshape_position == MaxSector)
1241 sb->minor_version = 90;
1242 else {
1243 sb->minor_version = 91;
1244 sb->reshape_position = mddev->reshape_position;
1245 sb->new_level = mddev->new_level;
1246 sb->delta_disks = mddev->delta_disks;
1247 sb->new_layout = mddev->new_layout;
664e7c41 1248 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1249 }
1250 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1251 if (mddev->in_sync)
1252 {
1253 sb->recovery_cp = mddev->recovery_cp;
1254 sb->cp_events_hi = (mddev->events>>32);
1255 sb->cp_events_lo = (u32)mddev->events;
1256 if (mddev->recovery_cp == MaxSector)
1257 sb->state = (1<< MD_SB_CLEAN);
1258 } else
1259 sb->recovery_cp = 0;
1260
1261 sb->layout = mddev->layout;
9d8f0363 1262 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1263
c3d9714e 1264 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1265 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1266
1da177e4 1267 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1268 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1269 mdp_disk_t *d;
86e6ffdd 1270 int desc_nr;
0261cd9f
N		 1271 int is_active = test_bit(In_sync, &rdev2->flags);
1272
1273 if (rdev2->raid_disk >= 0 &&
1274 sb->minor_version >= 91)
1275 /* we have nowhere to store the recovery_offset,
1276 * but if it is not below the reshape_position,
1277 * we can piggy-back on that.
1278 */
1279 is_active = 1;
1280 if (rdev2->raid_disk < 0 ||
1281 test_bit(Faulty, &rdev2->flags))
1282 is_active = 0;
1283 if (is_active)
86e6ffdd 1284 desc_nr = rdev2->raid_disk;
1da177e4 1285 else
86e6ffdd 1286 desc_nr = next_spare++;
19133a42 1287 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1288 d = &sb->disks[rdev2->desc_nr];
1289 nr_disks++;
1290 d->number = rdev2->desc_nr;
1291 d->major = MAJOR(rdev2->bdev->bd_dev);
1292 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1293 if (is_active)
1da177e4
LT		 1294 d->raid_disk = rdev2->raid_disk;
1295 else
1296 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1297 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1298 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1299 else if (is_active) {
1da177e4 1300 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1301 if (test_bit(In_sync, &rdev2->flags))
1302 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1303 active++;
1304 working++;
1305 } else {
1306 d->state = 0;
1307 spare++;
1308 working++;
1309 }
8ddf9efe
N		 1310 if (test_bit(WriteMostly, &rdev2->flags))
1311 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1312 }
1da177e4
LT		 1313 /* now set the "removed" and "faulty" bits on any missing devices */
1314 for (i=0 ; i < mddev->raid_disks ; i++) {
1315 mdp_disk_t *d = &sb->disks[i];
1316 if (d->state == 0 && d->number == 0) {
1317 d->number = i;
1318 d->raid_disk = i;
1319 d->state = (1<<MD_DISK_REMOVED);
1320 d->state |= (1<<MD_DISK_FAULTY);
1321 failed++;
1322 }
1323 }
1324 sb->nr_disks = nr_disks;
1325 sb->active_disks = active;
1326 sb->working_disks = working;
1327 sb->failed_disks = failed;
1328 sb->spare_disks = spare;
1329
1330 sb->this_disk = sb->disks[rdev->desc_nr];
1331 sb->sb_csum = calc_sb_csum(sb);
1332}
1333
0cd17fec
CW		 1334/*
1335 * rdev_size_change for 0.90.0
1336 */
1337static unsigned long long
15f4a5fd 1338super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec 1339{
58c0fed4 1340 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1341 return 0; /* component must fit device */
c3d9714e 1342 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1343 return 0; /* can't move bitmap */
57b2caa3 1344 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1345 if (!num_sectors || num_sectors > rdev->sb_start)
1346 num_sectors = rdev->sb_start;
0f420358 1347 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1348 rdev->sb_page);
1349 md_super_wait(rdev->mddev);
c26a44ed 1350 return num_sectors;
0cd17fec
CW		 1351}
1352
1353
1da177e4
LT		 1354/*
1355 * version 1 superblock
1356 */
1357
1c05b4bc 1358static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1359{
1c05b4bc
N		 1360 __le32 disk_csum;
1361 u32 csum;
1da177e4
LT		 1362 unsigned long long newcsum;
1363 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1364 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1365 int i;
1366
1367 disk_csum = sb->sb_csum;
1368 sb->sb_csum = 0;
1369 newcsum = 0;
1370 for (i=0; size>=4; size -= 4 )
1371 newcsum += le32_to_cpu(*isuper++);
1372
1373 if (size == 2)
1c05b4bc 1374 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1375
1376 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1377 sb->sb_csum = disk_csum;
1378 return cpu_to_le32(csum);
1379}
1380
1381static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1382{
1383 struct mdp_superblock_1 *sb;
1384 int ret;
0f420358 1385 sector_t sb_start;
1da177e4 1386 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1387 int bmask;
1da177e4
LT		 1388
1389 /*
0f420358 1390 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1391 * It is always aligned to a 4K boundary and
1392 * depeding on minor_version, it can be:
1393 * 0: At least 8K, but less than 12K, from end of device
1394 * 1: At start of device
1395 * 2: 4K from start of device.
1396 */
1397 switch(minor_version) {
1398 case 0:
77304d2a 1399 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1400 sb_start -= 8*2;
1401 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1402 break;
1403 case 1:
0f420358 1404 sb_start = 0;
1da177e4
LT		 1405 break;
1406 case 2:
0f420358 1407 sb_start = 8;
1da177e4
LT		 1408 break;
1409 default:
1410 return -EINVAL;
1411 }
0f420358 1412 rdev->sb_start = sb_start;
1da177e4 1413
0002b271
N		 1414 /* superblock is rarely larger than 1K, but it can be larger,
1415 * and it is safe to read 4k, so we do that
1416 */
1417 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1418 if (ret) return ret;
1419
1420
1421 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1422
1423 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1424 sb->major_version != cpu_to_le32(1) ||
1425 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1426 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1427 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1428 return -EINVAL;
1429
1430 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1431 printk("md: invalid superblock checksum on %s\n",
1432 bdevname(rdev->bdev,b));
1433 return -EINVAL;
1434 }
1435 if (le64_to_cpu(sb->data_size) < 10) {
1436 printk("md: data_size too small on %s\n",
1437 bdevname(rdev->bdev,b));
1438 return -EINVAL;
1439 }
e11e93fa 1440
1da177e4
LT		 1441 rdev->preferred_minor = 0xffff;
1442 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1443 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1444
0002b271 1445 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1446 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1447 if (rdev->sb_size & bmask)
a1801f85
N		 1448 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1449
1450 if (minor_version
0f420358 1451 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1452 return -EINVAL;
0002b271 1453
31b65a0d
N		 1454 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1455 rdev->desc_nr = -1;
1456 else
1457 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1458
9a7b2b0f 1459 if (!refdev) {
8ed75463 1460 ret = 1;
9a7b2b0f 1461 } else {
1da177e4
LT		 1462 __u64 ev1, ev2;
1463 struct mdp_superblock_1 *refsb =
1464 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1465
1466 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1467 sb->level != refsb->level ||
1468 sb->layout != refsb->layout ||
1469 sb->chunksize != refsb->chunksize) {
1470 printk(KERN_WARNING "md: %s has strangely different"
1471 " superblock to %s\n",
1472 bdevname(rdev->bdev,b),
1473 bdevname(refdev->bdev,b2));
1474 return -EINVAL;
1475 }
1476 ev1 = le64_to_cpu(sb->events);
1477 ev2 = le64_to_cpu(refsb->events);
1478
1479 if (ev1 > ev2)
8ed75463
N		 1480 ret = 1;
1481 else
1482 ret = 0;
1da177e4 1483 }
a1801f85 1484 if (minor_version)
77304d2a 1485 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1486 le64_to_cpu(sb->data_offset);
1da177e4 1487 else
dd8ac336
AN		 1488 rdev->sectors = rdev->sb_start;
1489 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1490 return -EINVAL;
dd8ac336 1491 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1492 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1493 return -EINVAL;
8ed75463 1494 return ret;
1da177e4
LT		 1495}
1496
1497static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1498{
1499 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
07d84d10 1500 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1501
41158c7e 1502 rdev->raid_disk = -1;
c5d79adb
N		 1503 clear_bit(Faulty, &rdev->flags);
1504 clear_bit(In_sync, &rdev->flags);
1505 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1506
1da177e4
LT		 1507 if (mddev->raid_disks == 0) {
1508 mddev->major_version = 1;
1509 mddev->patch_version = 0;
e691063a 1510 mddev->external = 0;
9d8f0363 1511 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1512 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1513 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1514 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1515 mddev->clevel[0] = 0;
1da177e4
LT		 1516 mddev->layout = le32_to_cpu(sb->layout);
1517 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1518 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1519 mddev->events = ev1;
c3d9714e
N		 1520 mddev->bitmap_info.offset = 0;
1521 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1522
1523 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1524 memcpy(mddev->uuid, sb->set_uuid, 16);
1525
1526 mddev->max_disks = (4096-256)/2;
a654b9d8 1527
71c0805c 1528 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1529 mddev->bitmap_info.file == NULL )
1530 mddev->bitmap_info.offset =
1531 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1532
f6705578
N		 1533 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1534 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1535 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1536 mddev->new_level = le32_to_cpu(sb->new_level);
1537 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1538 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1539 } else {
1540 mddev->reshape_position = MaxSector;
1541 mddev->delta_disks = 0;
1542 mddev->new_level = mddev->level;
1543 mddev->new_layout = mddev->layout;
664e7c41 1544 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1545 }
1546
41158c7e 1547 } else if (mddev->pers == NULL) {
be6800a7
N		 1548 /* Insist of good event counter while assembling, except for
1549 * spares (which don't need an event count) */
1da177e4 1550 ++ev1;
be6800a7
N		 1551 if (rdev->desc_nr >= 0 &&
1552 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1553 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1554 if (ev1 < mddev->events)
1555 return -EINVAL;
41158c7e
N		 1556 } else if (mddev->bitmap) {
1557 /* If adding to array with a bitmap, then we can accept an
1558 * older device, but not too old.
1559 */
41158c7e
N		 1560 if (ev1 < mddev->bitmap->events_cleared)
1561 return 0;
07d84d10
N		 1562 } else {
1563 if (ev1 < mddev->events)
1564 /* just a hot-add of a new device, leave raid_disk at -1 */
1565 return 0;
1566 }
1da177e4
LT		 1567 if (mddev->level != LEVEL_MULTIPATH) {
1568 int role;
3673f305
N		 1569 if (rdev->desc_nr < 0 ||
1570 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1571 role = 0xffff;
1572 rdev->desc_nr = -1;
1573 } else
1574 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1575 switch(role) {
1576 case 0xffff: /* spare */
1da177e4
LT		 1577 break;
1578 case 0xfffe: /* faulty */
b2d444d7 1579 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1580 break;
1581 default:
5fd6c1dc
N		 1582 if ((le32_to_cpu(sb->feature_map) &
1583 MD_FEATURE_RECOVERY_OFFSET))
1584 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1585 else
1586 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1587 rdev->raid_disk = role;
1588 break;
1589 }
8ddf9efe
N		 1590 if (sb->devflags & WriteMostly1)
1591 set_bit(WriteMostly, &rdev->flags);
41158c7e 1592 } else /* MULTIPATH are always insync */
b2d444d7 1593 set_bit(In_sync, &rdev->flags);
41158c7e 1594
1da177e4
LT		 1595 return 0;
1596}
1597
1598static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1599{
1600 struct mdp_superblock_1 *sb;
1da177e4
LT		 1601 mdk_rdev_t *rdev2;
1602 int max_dev, i;
1603 /* make rdev->sb match mddev and rdev data. */
1604
1605 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1606
1607 sb->feature_map = 0;
1608 sb->pad0 = 0;
5fd6c1dc 1609 sb->recovery_offset = cpu_to_le64(0);
1da177e4
LT		 1610 memset(sb->pad1, 0, sizeof(sb->pad1));
1611 memset(sb->pad2, 0, sizeof(sb->pad2));
1612 memset(sb->pad3, 0, sizeof(sb->pad3));
1613
1614 sb->utime = cpu_to_le64((__u64)mddev->utime);
1615 sb->events = cpu_to_le64(mddev->events);
1616 if (mddev->in_sync)
1617 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1618 else
1619 sb->resync_offset = cpu_to_le64(0);
1620
1c05b4bc 1621 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1622
f0ca340c 1623 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1624 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1625 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1626 sb->level = cpu_to_le32(mddev->level);
1627 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1628
c3d9714e
N		 1629 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1630 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1631 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1632 }
5fd6c1dc
N		 1633
1634 if (rdev->raid_disk >= 0 &&
97e4f42d 1635 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1636 sb->feature_map |=
1637 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1638 sb->recovery_offset =
1639 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1640 }
1641
f6705578
N		 1642 if (mddev->reshape_position != MaxSector) {
1643 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1644 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1645 sb->new_layout = cpu_to_le32(mddev->new_layout);
1646 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1647 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1648 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1649 }
a654b9d8 1650
1da177e4 1651 max_dev = 0;
159ec1fc 1652 list_for_each_entry(rdev2, &mddev->disks, same_set)
1da177e4
LT		 1653 if (rdev2->desc_nr+1 > max_dev)
1654 max_dev = rdev2->desc_nr+1;
a778b73f 1655
70471daf
N		 1656 if (max_dev > le32_to_cpu(sb->max_dev)) {
1657 int bmask;
a778b73f 1658 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1659 rdev->sb_size = max_dev * 2 + 256;
1660 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1661 if (rdev->sb_size & bmask)
1662 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1663 } else
1664 max_dev = le32_to_cpu(sb->max_dev);
1665
1da177e4
LT		 1666 for (i=0; i<max_dev;i++)
1667 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1668
159ec1fc 1669 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1670 i = rdev2->desc_nr;
b2d444d7 1671 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1672 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1673 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1674 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1675 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1676 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1677 else
1678 sb->dev_roles[i] = cpu_to_le16(0xffff);
1679 }
1680
1da177e4
LT		 1681 sb->sb_csum = calc_sb_1_csum(sb);
1682}
1683
0cd17fec 1684static unsigned long long
15f4a5fd 1685super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
0cd17fec
CW		 1686{
1687 struct mdp_superblock_1 *sb;
15f4a5fd 1688 sector_t max_sectors;
58c0fed4 1689 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1690 return 0; /* component must fit device */
0f420358 1691 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1692 /* minor versions 1 and 2; superblock before data */
77304d2a 1693 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1694 max_sectors -= rdev->data_offset;
1695 if (!num_sectors || num_sectors > max_sectors)
1696 num_sectors = max_sectors;
c3d9714e 1697 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1698 /* minor version 0 with bitmap we can't move */
1699 return 0;
1700 } else {
1701 /* minor version 0; superblock after data */
0f420358 1702 sector_t sb_start;
77304d2a 1703 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1704 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1705 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1706 if (!num_sectors || num_sectors > max_sectors)
1707 num_sectors = max_sectors;
0f420358 1708 rdev->sb_start = sb_start;
0cd17fec
CW		 1709 }
1710 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
15f4a5fd 1711 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1712 sb->super_offset = rdev->sb_start;
0cd17fec 1713 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1714 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1715 rdev->sb_page);
1716 md_super_wait(rdev->mddev);
c26a44ed 1717 return num_sectors;
0cd17fec 1718}
1da177e4 1719
75c96f85 1720static struct super_type super_types[] = {
1da177e4
LT		 1721 [0] = {
1722 .name = "0.90.0",
1723 .owner = THIS_MODULE,
0cd17fec
CW		 1724 .load_super = super_90_load,
1725 .validate_super = super_90_validate,
1726 .sync_super = super_90_sync,
1727 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1728 },
1729 [1] = {
1730 .name = "md-1",
1731 .owner = THIS_MODULE,
0cd17fec
CW		 1732 .load_super = super_1_load,
1733 .validate_super = super_1_validate,
1734 .sync_super = super_1_sync,
1735 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1736 },
1737};
1da177e4
LT		 1738
1739static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1740{
7dd5e7c3 1741 mdk_rdev_t *rdev, *rdev2;
1da177e4 1742
4b80991c
N		 1743 rcu_read_lock();
1744 rdev_for_each_rcu(rdev, mddev1)
1745 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1746 if (rdev->bdev->bd_contains ==
4b80991c
N		 1747 rdev2->bdev->bd_contains) {
1748 rcu_read_unlock();
7dd5e7c3 1749 return 1;
4b80991c
N		 1750 }
1751 rcu_read_unlock();
1da177e4
LT		 1752 return 0;
1753}
1754
1755static LIST_HEAD(pending_raid_disks);
1756
ac5e7113
AN		 1757/*
1758 * Try to register data integrity profile for an mddev
1759 *
1760 * This is called when an array is started and after a disk has been kicked
1761 * from the array. It only succeeds if all working and active component devices
1762 * are integrity capable with matching profiles.
1763 */
1764int md_integrity_register(mddev_t *mddev)
1765{
1766 mdk_rdev_t *rdev, *reference = NULL;
1767
1768 if (list_empty(&mddev->disks))
1769 return 0; /* nothing to do */
1770 if (blk_get_integrity(mddev->gendisk))
1771 return 0; /* already registered */
1772 list_for_each_entry(rdev, &mddev->disks, same_set) {
1773 /* skip spares and non-functional disks */
1774 if (test_bit(Faulty, &rdev->flags))
1775 continue;
1776 if (rdev->raid_disk < 0)
1777 continue;
1778 /*
1779 * If at least one rdev is not integrity capable, we can not
1780 * enable data integrity for the md device.
1781 */
1782 if (!bdev_get_integrity(rdev->bdev))
1783 return -EINVAL;
1784 if (!reference) {
1785 /* Use the first rdev as the reference */
1786 reference = rdev;
1787 continue;
1788 }
1789 /* does this rdev's profile match the reference profile? */
1790 if (blk_integrity_compare(reference->bdev->bd_disk,
1791 rdev->bdev->bd_disk) < 0)
1792 return -EINVAL;
1793 }
1794 /*
1795 * All component devices are integrity capable and have matching
1796 * profiles, register the common profile for the md device.
1797 */
1798 if (blk_integrity_register(mddev->gendisk,
1799 bdev_get_integrity(reference->bdev)) != 0) {
1800 printk(KERN_ERR "md: failed to register integrity for %s\n",
1801 mdname(mddev));
1802 return -EINVAL;
1803 }
1804 printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1805 mdname(mddev));
1806 return 0;
1807}
1808EXPORT_SYMBOL(md_integrity_register);
1809
1810/* Disable data integrity if non-capable/non-matching disk is being added */
1811void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
3f9d99c1 1812{
3f9d99c1 1813 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1814 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1815
ac5e7113 1816 if (!bi_mddev) /* nothing to do */
3f9d99c1 1817 return;
ac5e7113 1818 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1819 return;
ac5e7113
AN		 1820 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1821 rdev->bdev->bd_disk) >= 0)
1822 return;
1823 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1824 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 1825}
ac5e7113 1826EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 1827
1da177e4
LT		 1828static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1829{
7dd5e7c3 1830 char b[BDEVNAME_SIZE];
f637b9f9 1831 struct kobject *ko;
1edf80d3 1832 char *s;
5e55e2f5 1833 int err;
1da177e4
LT		 1834
1835 if (rdev->mddev) {
1836 MD_BUG();
1837 return -EINVAL;
1838 }
11e2ede0
DW		 1839
1840 /* prevent duplicates */
1841 if (find_rdev(mddev, rdev->bdev->bd_dev))
1842 return -EEXIST;
1843
dd8ac336
AN		 1844 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1845 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1846 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 1847 if (mddev->pers) {
1848 /* Cannot change size, so fail
1849 * If mddev->level <= 0, then we don't care
1850 * about aligning sizes (e.g. linear)
1851 */
1852 if (mddev->level > 0)
1853 return -ENOSPC;
1854 } else
dd8ac336 1855 mddev->dev_sectors = rdev->sectors;
2bf071bf 1856 }
1da177e4
LT		 1857
1858 /* Verify rdev->desc_nr is unique.
1859 * If it is -1, assign a free number, else
1860 * check number is not in use
1861 */
1862 if (rdev->desc_nr < 0) {
1863 int choice = 0;
1864 if (mddev->pers) choice = mddev->raid_disks;
1865 while (find_rdev_nr(mddev, choice))
1866 choice++;
1867 rdev->desc_nr = choice;
1868 } else {
1869 if (find_rdev_nr(mddev, rdev->desc_nr))
1870 return -EBUSY;
1871 }
de01dfad
N		 1872 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1873 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1874 mdname(mddev), mddev->max_disks);
1875 return -EBUSY;
1876 }
19133a42 1877 bdevname(rdev->bdev,b);
649316b2 1878 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 1879 *s = '!';
649316b2 1880
1da177e4 1881 rdev->mddev = mddev;
19133a42 1882 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 1883
b2d6db58 1884 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 1885 goto fail;
86e6ffdd 1886
0762b8bd 1887 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 1888 if (sysfs_create_link(&rdev->kobj, ko, "block"))
1889 /* failure here is OK */;
1890 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 1891
4b80991c 1892 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 1893 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 1894
1895 /* May as well allow recovery to be retried once */
1896 mddev->recovery_disabled = 0;
3f9d99c1 1897
1da177e4 1898 return 0;
5e55e2f5
N		 1899
1900 fail:
1901 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1902 b, mdname(mddev));
1903 return err;
1da177e4
LT		 1904}
1905
177a99b2 1906static void md_delayed_delete(struct work_struct *ws)
5792a285
N		 1907{
1908 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1909 kobject_del(&rdev->kobj);
177a99b2 1910 kobject_put(&rdev->kobj);
5792a285
N		 1911}
1912
1da177e4
LT		 1913static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1914{
1915 char b[BDEVNAME_SIZE];
1916 if (!rdev->mddev) {
1917 MD_BUG();
1918 return;
1919 }
49731baa 1920 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 1921 list_del_rcu(&rdev->same_set);
1da177e4
LT		 1922 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1923 rdev->mddev = NULL;
86e6ffdd 1924 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 1925 sysfs_put(rdev->sysfs_state);
1926 rdev->sysfs_state = NULL;
5792a285 1927 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 1928 * writing to 'remove' to "dev/state". We also need
1929 * to delay it due to rcu usage.
5792a285 1930 */
4b80991c 1931 synchronize_rcu();
177a99b2
N		 1932 INIT_WORK(&rdev->del_work, md_delayed_delete);
1933 kobject_get(&rdev->kobj);
e804ac78 1934 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 1935}
1936
1937/*
1938 * prevent the device from being mounted, repartitioned or
1939 * otherwise reused by a RAID array (or any other kernel
1940 * subsystem), by bd_claiming the device.
1941 */
c5d79adb 1942static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1da177e4
LT		 1943{
1944 int err = 0;
1945 struct block_device *bdev;
1946 char b[BDEVNAME_SIZE];
1947
d4d77629
TH		 1948 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
1949 shared ? (mdk_rdev_t *)lock_rdev : rdev);
1da177e4
LT		 1950 if (IS_ERR(bdev)) {
1951 printk(KERN_ERR "md: could not open %s.\n",
1952 __bdevname(dev, b));
1953 return PTR_ERR(bdev);
1954 }
1da177e4
LT		 1955 rdev->bdev = bdev;
1956 return err;
1957}
1958
1959static void unlock_rdev(mdk_rdev_t *rdev)
1960{
1961 struct block_device *bdev = rdev->bdev;
1962 rdev->bdev = NULL;
1963 if (!bdev)
1964 MD_BUG();
e525fd89 1965 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 1966}
1967
1968void md_autodetect_dev(dev_t dev);
1969
1970static void export_rdev(mdk_rdev_t * rdev)
1971{
1972 char b[BDEVNAME_SIZE];
1973 printk(KERN_INFO "md: export_rdev(%s)\n",
1974 bdevname(rdev->bdev,b));
1975 if (rdev->mddev)
1976 MD_BUG();
1977 free_disk_sb(rdev);
1da177e4 1978#ifndef MODULE
d0fae18f
N		 1979 if (test_bit(AutoDetected, &rdev->flags))
1980 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 1981#endif
1982 unlock_rdev(rdev);
86e6ffdd 1983 kobject_put(&rdev->kobj);
1da177e4
LT		 1984}
1985
1986static void kick_rdev_from_array(mdk_rdev_t * rdev)
1987{
1988 unbind_rdev_from_array(rdev);
1989 export_rdev(rdev);
1990}
1991
1992static void export_array(mddev_t *mddev)
1993{
159ec1fc 1994 mdk_rdev_t *rdev, *tmp;
1da177e4 1995
d089c6af 1996 rdev_for_each(rdev, tmp, mddev) {
1da177e4
LT		 1997 if (!rdev->mddev) {
1998 MD_BUG();
1999 continue;
2000 }
2001 kick_rdev_from_array(rdev);
2002 }
2003 if (!list_empty(&mddev->disks))
2004 MD_BUG();
2005 mddev->raid_disks = 0;
2006 mddev->major_version = 0;
2007}
2008
2009static void print_desc(mdp_disk_t *desc)
2010{
2011 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2012 desc->major,desc->minor,desc->raid_disk,desc->state);
2013}
2014
cd2ac932 2015static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2016{
2017 int i;
2018
2019 printk(KERN_INFO
2020 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2021 sb->major_version, sb->minor_version, sb->patch_version,
2022 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2023 sb->ctime);
2024 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2025 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2026 sb->md_minor, sb->layout, sb->chunk_size);
2027 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2028 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2029 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2030 sb->failed_disks, sb->spare_disks,
2031 sb->sb_csum, (unsigned long)sb->events_lo);
2032
2033 printk(KERN_INFO);
2034 for (i = 0; i < MD_SB_DISKS; i++) {
2035 mdp_disk_t *desc;
2036
2037 desc = sb->disks + i;
2038 if (desc->number || desc->major || desc->minor ||
2039 desc->raid_disk || (desc->state && (desc->state != 4))) {
2040 printk(" D %2d: ", i);
2041 print_desc(desc);
2042 }
2043 }
2044 printk(KERN_INFO "md: THIS: ");
2045 print_desc(&sb->this_disk);
cd2ac932 2046}
1da177e4 2047
cd2ac932
CR		 2048static void print_sb_1(struct mdp_superblock_1 *sb)
2049{
2050 __u8 *uuid;
2051
2052 uuid = sb->set_uuid;
ad361c98 2053 printk(KERN_INFO
7b75c2f8 2054 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2055 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2056 le32_to_cpu(sb->major_version),
2057 le32_to_cpu(sb->feature_map),
7b75c2f8 2058 uuid,
cd2ac932
CR		 2059 sb->set_name,
2060 (unsigned long long)le64_to_cpu(sb->ctime)
2061 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2062
2063 uuid = sb->device_uuid;
ad361c98
JP		 2064 printk(KERN_INFO
2065 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2066 " RO:%llu\n"
7b75c2f8 2067 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2068 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2069 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2070 le32_to_cpu(sb->level),
2071 (unsigned long long)le64_to_cpu(sb->size),
2072 le32_to_cpu(sb->raid_disks),
2073 le32_to_cpu(sb->layout),
2074 le32_to_cpu(sb->chunksize),
2075 (unsigned long long)le64_to_cpu(sb->data_offset),
2076 (unsigned long long)le64_to_cpu(sb->data_size),
2077 (unsigned long long)le64_to_cpu(sb->super_offset),
2078 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2079 le32_to_cpu(sb->dev_number),
7b75c2f8 2080 uuid,
cd2ac932
CR		 2081 sb->devflags,
2082 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2083 (unsigned long long)le64_to_cpu(sb->events),
2084 (unsigned long long)le64_to_cpu(sb->resync_offset),
2085 le32_to_cpu(sb->sb_csum),
2086 le32_to_cpu(sb->max_dev)
2087 );
1da177e4
LT		 2088}
2089
cd2ac932 2090static void print_rdev(mdk_rdev_t *rdev, int major_version)
1da177e4
LT		 2091{
2092 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2093 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2094 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2095 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2096 rdev->desc_nr);
1da177e4 2097 if (rdev->sb_loaded) {
cd2ac932
CR		 2098 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2099 switch (major_version) {
2100 case 0:
2101 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
2102 break;
2103 case 1:
2104 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
2105 break;
2106 }
1da177e4
LT		 2107 } else
2108 printk(KERN_INFO "md: no rdev superblock!\n");
2109}
2110
5e56341d 2111static void md_print_devices(void)
1da177e4 2112{
159ec1fc 2113 struct list_head *tmp;
1da177e4
LT		 2114 mdk_rdev_t *rdev;
2115 mddev_t *mddev;
2116 char b[BDEVNAME_SIZE];
2117
2118 printk("\n");
2119 printk("md: **********************************\n");
2120 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2121 printk("md: **********************************\n");
29ac4aa3 2122 for_each_mddev(mddev, tmp) {
1da177e4 2123
32a7627c
N		 2124 if (mddev->bitmap)
2125 bitmap_print_sb(mddev->bitmap);
2126 else
2127 printk("%s: ", mdname(mddev));
159ec1fc 2128 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2129 printk("<%s>", bdevname(rdev->bdev,b));
2130 printk("\n");
2131
159ec1fc 2132 list_for_each_entry(rdev, &mddev->disks, same_set)
cd2ac932 2133 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2134 }
2135 printk("md: **********************************\n");
2136 printk("\n");
2137}
2138
2139
42543769 2140static void sync_sbs(mddev_t * mddev, int nospares)
1da177e4 2141{
42543769
N		 2142 /* Update each superblock (in-memory image), but
2143 * if we are allowed to, skip spares which already
2144 * have the right event counter, or have one earlier
2145 * (which would mean they aren't being marked as dirty
2146 * with the rest of the array)
2147 */
1da177e4 2148 mdk_rdev_t *rdev;
159ec1fc 2149 list_for_each_entry(rdev, &mddev->disks, same_set) {
42543769
N		 2150 if (rdev->sb_events == mddev->events ||
2151 (nospares &&
2152 rdev->raid_disk < 0 &&
42543769
N		 2153 rdev->sb_events+1 == mddev->events)) {
2154 /* Don't update this superblock */
2155 rdev->sb_loaded = 2;
2156 } else {
2157 super_types[mddev->major_version].
2158 sync_super(mddev, rdev);
2159 rdev->sb_loaded = 1;
2160 }
1da177e4
LT		 2161 }
2162}
2163
850b2b42 2164static void md_update_sb(mddev_t * mddev, int force_change)
1da177e4 2165{
1da177e4 2166 mdk_rdev_t *rdev;
06d91a5f 2167 int sync_req;
42543769 2168 int nospares = 0;
1da177e4 2169
1da177e4 2170repeat:
3a3a5ddb
N		 2171 /* First make sure individual recovery_offsets are correct */
2172 list_for_each_entry(rdev, &mddev->disks, same_set) {
2173 if (rdev->raid_disk >= 0 &&
2174 mddev->delta_disks >= 0 &&
2175 !test_bit(In_sync, &rdev->flags) &&
2176 mddev->curr_resync_completed > rdev->recovery_offset)
2177 rdev->recovery_offset = mddev->curr_resync_completed;
2178
2179 }
bd52b746 2180 if (!mddev->persistent) {
070dc6dd 2181 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2182 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
d97a41dc
N		 2183 if (!mddev->external)
2184 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3a3a5ddb
N		 2185 wake_up(&mddev->sb_wait);
2186 return;
2187 }
2188
a9701a30 2189 spin_lock_irq(&mddev->write_lock);
84692195 2190
3a3a5ddb
N		 2191 mddev->utime = get_seconds();
2192
850b2b42
N		 2193 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2194 force_change = 1;
2195 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2196 /* just a clean<-> dirty transition, possibly leave spares alone,
2197 * though if events isn't the right even/odd, we will have to do
2198 * spares after all
2199 */
2200 nospares = 1;
2201 if (force_change)
2202 nospares = 0;
2203 if (mddev->degraded)
84692195
N		 2204 /* If the array is degraded, then skipping spares is both
2205 * dangerous and fairly pointless.
2206 * Dangerous because a device that was removed from the array
2207 * might have a event_count that still looks up-to-date,
2208 * so it can be re-added without a resync.
2209 * Pointless because if there are any spares to skip,
2210 * then a recovery will happen and soon that array won't
2211 * be degraded any more and the spare can go back to sleep then.
2212 */
850b2b42 2213 nospares = 0;
84692195 2214
06d91a5f 2215 sync_req = mddev->in_sync;
42543769
N		 2216
2217 /* If this is just a dirty<->clean transition, and the array is clean
2218 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2219 if (nospares
42543769 2220 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2221 && mddev->can_decrease_events
2222 && mddev->events != 1) {
42543769 2223 mddev->events--;
a8707c08
N		 2224 mddev->can_decrease_events = 0;
2225 } else {
42543769
N		 2226 /* otherwise we have to go forward and ... */
2227 mddev->events ++;
a8707c08 2228 mddev->can_decrease_events = nospares;
42543769 2229 }
1da177e4
LT		 2230
2231 if (!mddev->events) {
2232 /*
2233 * oops, this 64-bit counter should never wrap.
2234 * Either we are in around ~1 trillion A.C., assuming
2235 * 1 reboot per second, or we have a bug:
2236 */
2237 MD_BUG();
2238 mddev->events --;
2239 }
e691063a 2240 sync_sbs(mddev, nospares);
a9701a30 2241 spin_unlock_irq(&mddev->write_lock);
1da177e4
LT		 2242
2243 dprintk(KERN_INFO
2244 "md: updating %s RAID superblock on device (in sync %d)\n",
2245 mdname(mddev),mddev->in_sync);
2246
4ad13663 2247 bitmap_update_sb(mddev->bitmap);
159ec1fc 2248 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 2249 char b[BDEVNAME_SIZE];
2250 dprintk(KERN_INFO "md: ");
42543769
N		 2251 if (rdev->sb_loaded != 1)
2252 continue; /* no noise on spare devices */
b2d444d7 2253 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 2254 dprintk("(skipping faulty ");
2255
2256 dprintk("%s ", bdevname(rdev->bdev,b));
b2d444d7 2257 if (!test_bit(Faulty, &rdev->flags)) {
7bfa19f2 2258 md_super_write(mddev,rdev,
0f420358 2259 rdev->sb_start, rdev->sb_size,
7bfa19f2
N		 2260 rdev->sb_page);
2261 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2262 bdevname(rdev->bdev,b),
0f420358 2263 (unsigned long long)rdev->sb_start);
42543769 2264 rdev->sb_events = mddev->events;
7bfa19f2 2265
1da177e4
LT		 2266 } else
2267 dprintk(")\n");
7bfa19f2 2268 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2269 /* only need to write one superblock... */
2270 break;
2271 }
a9701a30 2272 md_super_wait(mddev);
850b2b42 2273 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2274
a9701a30 2275 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2276 if (mddev->in_sync != sync_req ||
2277 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2278 /* have to write it out again */
a9701a30 2279 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2280 goto repeat;
2281 }
850b2b42 2282 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2283 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2284 wake_up(&mddev->sb_wait);
acb180b0
N		 2285 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2286 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2287
1da177e4
LT		 2288}
2289
7f6ce769 2290/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2291 * We want to accept with case. For this we use cmd_match.
2292 */
2293static int cmd_match(const char *cmd, const char *str)
2294{
2295 /* See if cmd, written into a sysfs file, matches
2296 * str. They must either be the same, or cmd can
2297 * have a trailing newline
2298 */
2299 while (*cmd && *str && *cmd == *str) {
2300 cmd++;
2301 str++;
2302 }
2303 if (*cmd == '\n')
2304 cmd++;
2305 if (*str || *cmd)
2306 return 0;
2307 return 1;
2308}
2309
86e6ffdd
N		 2310struct rdev_sysfs_entry {
2311 struct attribute attr;
2312 ssize_t (*show)(mdk_rdev_t *, char *);
2313 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2314};
2315
2316static ssize_t
96de1e66 2317state_show(mdk_rdev_t *rdev, char *page)
86e6ffdd
N		 2318{
2319 char *sep = "";
20a49ff6 2320 size_t len = 0;
86e6ffdd 2321
b2d444d7 2322 if (test_bit(Faulty, &rdev->flags)) {
86e6ffdd
N		 2323 len+= sprintf(page+len, "%sfaulty",sep);
2324 sep = ",";
2325 }
b2d444d7 2326 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2327 len += sprintf(page+len, "%sin_sync",sep);
2328 sep = ",";
2329 }
f655675b
N		 2330 if (test_bit(WriteMostly, &rdev->flags)) {
2331 len += sprintf(page+len, "%swrite_mostly",sep);
2332 sep = ",";
2333 }
6bfe0b49
DW		 2334 if (test_bit(Blocked, &rdev->flags)) {
2335 len += sprintf(page+len, "%sblocked", sep);
2336 sep = ",";
2337 }
b2d444d7
N		 2338 if (!test_bit(Faulty, &rdev->flags) &&
2339 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2340 len += sprintf(page+len, "%sspare", sep);
2341 sep = ",";
2342 }
2343 return len+sprintf(page+len, "\n");
2344}
2345
45dc2de1
N		 2346static ssize_t
2347state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2348{
2349 /* can write
2350 * faulty - simulates and error
2351 * remove - disconnects the device
f655675b
N		 2352 * writemostly - sets write_mostly
2353 * -writemostly - clears write_mostly
6bfe0b49
DW		 2354 * blocked - sets the Blocked flag
2355 * -blocked - clears the Blocked flag
6d56e278 2356 * insync - sets Insync providing device isn't active
45dc2de1
N		 2357 */
2358 int err = -EINVAL;
2359 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2360 md_error(rdev->mddev, rdev);
2361 err = 0;
2362 } else if (cmd_match(buf, "remove")) {
2363 if (rdev->raid_disk >= 0)
2364 err = -EBUSY;
2365 else {
2366 mddev_t *mddev = rdev->mddev;
2367 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2368 if (mddev->pers)
2369 md_update_sb(mddev, 1);
45dc2de1
N		 2370 md_new_event(mddev);
2371 err = 0;
2372 }
f655675b
N		 2373 } else if (cmd_match(buf, "writemostly")) {
2374 set_bit(WriteMostly, &rdev->flags);
2375 err = 0;
2376 } else if (cmd_match(buf, "-writemostly")) {
2377 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2378 err = 0;
2379 } else if (cmd_match(buf, "blocked")) {
2380 set_bit(Blocked, &rdev->flags);
2381 err = 0;
2382 } else if (cmd_match(buf, "-blocked")) {
2383 clear_bit(Blocked, &rdev->flags);
2384 wake_up(&rdev->blocked_wait);
2385 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2386 md_wakeup_thread(rdev->mddev->thread);
2387
6d56e278
N		 2388 err = 0;
2389 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2390 set_bit(In_sync, &rdev->flags);
f655675b 2391 err = 0;
45dc2de1 2392 }
00bcb4ac
N		 2393 if (!err)
2394 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2395 return err ? err : len;
2396}
80ca3a44
N		 2397static struct rdev_sysfs_entry rdev_state =
2398__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2399
4dbcdc75
N		 2400static ssize_t
2401errors_show(mdk_rdev_t *rdev, char *page)
2402{
2403 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2404}
2405
2406static ssize_t
2407errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2408{
2409 char *e;
2410 unsigned long n = simple_strtoul(buf, &e, 10);
2411 if (*buf && (*e == 0 || *e == '\n')) {
2412 atomic_set(&rdev->corrected_errors, n);
2413 return len;
2414 }
2415 return -EINVAL;
2416}
2417static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2418__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2419
014236d2
N		 2420static ssize_t
2421slot_show(mdk_rdev_t *rdev, char *page)
2422{
2423 if (rdev->raid_disk < 0)
2424 return sprintf(page, "none\n");
2425 else
2426 return sprintf(page, "%d\n", rdev->raid_disk);
2427}
2428
2429static ssize_t
2430slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2431{
2432 char *e;
c303da6d
N		 2433 int err;
2434 char nm[20];
014236d2
N		 2435 int slot = simple_strtoul(buf, &e, 10);
2436 if (strncmp(buf, "none", 4)==0)
2437 slot = -1;
2438 else if (e==buf || (*e && *e!= '\n'))
2439 return -EINVAL;
6c2fce2e 2440 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2441 /* Setting 'slot' on an active array requires also
2442 * updating the 'rd%d' link, and communicating
2443 * with the personality with ->hot_*_disk.
2444 * For now we only support removing
2445 * failed/spare devices. This normally happens automatically,
2446 * but not when the metadata is externally managed.
2447 */
c303da6d
N		 2448 if (rdev->raid_disk == -1)
2449 return -EEXIST;
2450 /* personality does all needed checks */
2451 if (rdev->mddev->pers->hot_add_disk == NULL)
2452 return -EINVAL;
2453 err = rdev->mddev->pers->
2454 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2455 if (err)
2456 return err;
2457 sprintf(nm, "rd%d", rdev->raid_disk);
2458 sysfs_remove_link(&rdev->mddev->kobj, nm);
b7103107 2459 rdev->raid_disk = -1;
c303da6d
N		 2460 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2461 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e
NB		 2462 } else if (rdev->mddev->pers) {
2463 mdk_rdev_t *rdev2;
6c2fce2e 2464 /* Activating a spare .. or possibly reactivating
6d56e278 2465 * if we ever get bitmaps working here.
6c2fce2e
NB		 2466 */
2467
2468 if (rdev->raid_disk != -1)
2469 return -EBUSY;
2470
c6751b2b
N		 2471 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2472 return -EBUSY;
2473
6c2fce2e
NB		 2474 if (rdev->mddev->pers->hot_add_disk == NULL)
2475 return -EINVAL;
2476
159ec1fc 2477 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2478 if (rdev2->raid_disk == slot)
2479 return -EEXIST;
2480
ba1b41b6
N		 2481 if (slot >= rdev->mddev->raid_disks &&
2482 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2483 return -ENOSPC;
2484
6c2fce2e
NB		 2485 rdev->raid_disk = slot;
2486 if (test_bit(In_sync, &rdev->flags))
2487 rdev->saved_raid_disk = slot;
2488 else
2489 rdev->saved_raid_disk = -1;
2490 err = rdev->mddev->pers->
2491 hot_add_disk(rdev->mddev, rdev);
199050ea 2492 if (err) {
6c2fce2e 2493 rdev->raid_disk = -1;
6c2fce2e 2494 return err;
52664732 2495 } else
00bcb4ac 2496 sysfs_notify_dirent_safe(rdev->sysfs_state);
6c2fce2e
NB		 2497 sprintf(nm, "rd%d", rdev->raid_disk);
2498 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
00bcb4ac 2499 /* failure here is OK */;
6c2fce2e 2500 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2501 } else {
ba1b41b6
N		 2502 if (slot >= rdev->mddev->raid_disks &&
2503 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2504 return -ENOSPC;
2505 rdev->raid_disk = slot;
2506 /* assume it is working */
c5d79adb
N		 2507 clear_bit(Faulty, &rdev->flags);
2508 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2509 set_bit(In_sync, &rdev->flags);
00bcb4ac 2510 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2511 }
014236d2
N		 2512 return len;
2513}
2514
2515
2516static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2517__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2518
93c8cad0
N		 2519static ssize_t
2520offset_show(mdk_rdev_t *rdev, char *page)
2521{
6961ece4 2522 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2523}
2524
2525static ssize_t
2526offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2527{
2528 char *e;
2529 unsigned long long offset = simple_strtoull(buf, &e, 10);
2530 if (e==buf || (*e && *e != '\n'))
2531 return -EINVAL;
8ed0a521 2532 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2533 return -EBUSY;
dd8ac336 2534 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2535 /* Must set offset before size, so overlap checks
2536 * can be sane */
2537 return -EBUSY;
93c8cad0
N		 2538 rdev->data_offset = offset;
2539 return len;
2540}
2541
2542static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2543__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2544
83303b61
N		 2545static ssize_t
2546rdev_size_show(mdk_rdev_t *rdev, char *page)
2547{
dd8ac336 2548 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2549}
2550
c5d79adb
N		 2551static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2552{
2553 /* check if two start/length pairs overlap */
2554 if (s1+l1 <= s2)
2555 return 0;
2556 if (s2+l2 <= s1)
2557 return 0;
2558 return 1;
2559}
2560
b522adcd
DW		 2561static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2562{
2563 unsigned long long blocks;
2564 sector_t new;
2565
2566 if (strict_strtoull(buf, 10, &blocks) < 0)
2567 return -EINVAL;
2568
2569 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2570 return -EINVAL; /* sector conversion overflow */
2571
2572 new = blocks * 2;
2573 if (new != blocks * 2)
2574 return -EINVAL; /* unsigned long long to sector_t overflow */
2575
2576 *sectors = new;
2577 return 0;
2578}
2579
83303b61
N		 2580static ssize_t
2581rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2582{
27c529bb 2583 mddev_t *my_mddev = rdev->mddev;
dd8ac336 2584 sector_t oldsectors = rdev->sectors;
b522adcd 2585 sector_t sectors;
27c529bb 2586
b522adcd 2587 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2588 return -EINVAL;
0cd17fec 2589 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2590 if (my_mddev->persistent) {
dd8ac336
AN		 2591 sectors = super_types[my_mddev->major_version].
2592 rdev_size_change(rdev, sectors);
2593 if (!sectors)
0cd17fec 2594 return -EBUSY;
dd8ac336 2595 } else if (!sectors)
77304d2a 2596 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2597 rdev->data_offset;
0cd17fec 2598 }
dd8ac336 2599 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2600 return -EINVAL; /* component must fit device */
0cd17fec 2601
dd8ac336
AN		 2602 rdev->sectors = sectors;
2603 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2604 /* need to check that all other rdevs with the same ->bdev
2605 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2606 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2607 * we have to change it back, we will have the lock again.
2608 */
2609 mddev_t *mddev;
2610 int overlap = 0;
159ec1fc 2611 struct list_head *tmp;
c5d79adb 2612
27c529bb 2613 mddev_unlock(my_mddev);
29ac4aa3 2614 for_each_mddev(mddev, tmp) {
c5d79adb
N		 2615 mdk_rdev_t *rdev2;
2616
2617 mddev_lock(mddev);
159ec1fc 2618 list_for_each_entry(rdev2, &mddev->disks, same_set)
f21e9ff7
N		 2619 if (rdev->bdev == rdev2->bdev &&
2620 rdev != rdev2 &&
2621 overlaps(rdev->data_offset, rdev->sectors,
2622 rdev2->data_offset,
2623 rdev2->sectors)) {
c5d79adb
N		 2624 overlap = 1;
2625 break;
2626 }
2627 mddev_unlock(mddev);
2628 if (overlap) {
2629 mddev_put(mddev);
2630 break;
2631 }
2632 }
27c529bb 2633 mddev_lock(my_mddev);
c5d79adb
N		 2634 if (overlap) {
2635 /* Someone else could have slipped in a size
2636 * change here, but doing so is just silly.
dd8ac336 2637 * We put oldsectors back because we *know* it is
c5d79adb
N		 2638 * safe, and trust userspace not to race with
2639 * itself
2640 */
dd8ac336 2641 rdev->sectors = oldsectors;
c5d79adb
N		 2642 return -EBUSY;
2643 }
2644 }
83303b61
N		 2645 return len;
2646}
2647
2648static struct rdev_sysfs_entry rdev_size =
80ca3a44 2649__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2650
06e3c817
DW		 2651
2652static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2653{
2654 unsigned long long recovery_start = rdev->recovery_offset;
2655
2656 if (test_bit(In_sync, &rdev->flags) ||
2657 recovery_start == MaxSector)
2658 return sprintf(page, "none\n");
2659
2660 return sprintf(page, "%llu\n", recovery_start);
2661}
2662
2663static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2664{
2665 unsigned long long recovery_start;
2666
2667 if (cmd_match(buf, "none"))
2668 recovery_start = MaxSector;
2669 else if (strict_strtoull(buf, 10, &recovery_start))
2670 return -EINVAL;
2671
2672 if (rdev->mddev->pers &&
2673 rdev->raid_disk >= 0)
2674 return -EBUSY;
2675
2676 rdev->recovery_offset = recovery_start;
2677 if (recovery_start == MaxSector)
2678 set_bit(In_sync, &rdev->flags);
2679 else
2680 clear_bit(In_sync, &rdev->flags);
2681 return len;
2682}
2683
2684static struct rdev_sysfs_entry rdev_recovery_start =
2685__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2686
86e6ffdd
N		 2687static struct attribute *rdev_default_attrs[] = {
2688 &rdev_state.attr,
4dbcdc75 2689 &rdev_errors.attr,
014236d2 2690 &rdev_slot.attr,
93c8cad0 2691 &rdev_offset.attr,
83303b61 2692 &rdev_size.attr,
06e3c817 2693 &rdev_recovery_start.attr,
86e6ffdd
N		 2694 NULL,
2695};
2696static ssize_t
2697rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2698{
2699 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2700 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2701 mddev_t *mddev = rdev->mddev;
2702 ssize_t rv;
86e6ffdd
N		 2703
2704 if (!entry->show)
2705 return -EIO;
27c529bb
N		 2706
2707 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2708 if (!rv) {
2709 if (rdev->mddev == NULL)
2710 rv = -EBUSY;
2711 else
2712 rv = entry->show(rdev, page);
2713 mddev_unlock(mddev);
2714 }
2715 return rv;
86e6ffdd
N		 2716}
2717
2718static ssize_t
2719rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2720 const char *page, size_t length)
2721{
2722 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2723 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
27c529bb
N		 2724 ssize_t rv;
2725 mddev_t *mddev = rdev->mddev;
86e6ffdd
N		 2726
2727 if (!entry->store)
2728 return -EIO;
67463acb
N		 2729 if (!capable(CAP_SYS_ADMIN))
2730 return -EACCES;
27c529bb 2731 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 2732 if (!rv) {
27c529bb
N		 2733 if (rdev->mddev == NULL)
2734 rv = -EBUSY;
2735 else
2736 rv = entry->store(rdev, page, length);
6a51830e 2737 mddev_unlock(mddev);
ca388059
N		 2738 }
2739 return rv;
86e6ffdd
N		 2740}
2741
2742static void rdev_free(struct kobject *ko)
2743{
2744 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2745 kfree(rdev);
2746}
52cf25d0 2747static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 2748 .show = rdev_attr_show,
2749 .store = rdev_attr_store,
2750};
2751static struct kobj_type rdev_ktype = {
2752 .release = rdev_free,
2753 .sysfs_ops = &rdev_sysfs_ops,
2754 .default_attrs = rdev_default_attrs,
2755};
2756
e8bb9a83
N		 2757void md_rdev_init(mdk_rdev_t *rdev)
2758{
2759 rdev->desc_nr = -1;
2760 rdev->saved_raid_disk = -1;
2761 rdev->raid_disk = -1;
2762 rdev->flags = 0;
2763 rdev->data_offset = 0;
2764 rdev->sb_events = 0;
2765 rdev->last_read_error.tv_sec = 0;
2766 rdev->last_read_error.tv_nsec = 0;
2767 atomic_set(&rdev->nr_pending, 0);
2768 atomic_set(&rdev->read_errors, 0);
2769 atomic_set(&rdev->corrected_errors, 0);
2770
2771 INIT_LIST_HEAD(&rdev->same_set);
2772 init_waitqueue_head(&rdev->blocked_wait);
2773}
2774EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 2775/*
2776 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2777 *
2778 * mark the device faulty if:
2779 *
2780 * - the device is nonexistent (zero size)
2781 * - the device has no valid superblock
2782 *
2783 * a faulty rdev _never_ has rdev->sb set.
2784 */
2785static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2786{
2787 char b[BDEVNAME_SIZE];
2788 int err;
2789 mdk_rdev_t *rdev;
2790 sector_t size;
2791
9ffae0cf 2792 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 2793 if (!rdev) {
2794 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2795 return ERR_PTR(-ENOMEM);
2796 }
1da177e4 2797
e8bb9a83 2798 md_rdev_init(rdev);
1da177e4
LT		 2799 if ((err = alloc_disk_sb(rdev)))
2800 goto abort_free;
2801
c5d79adb 2802 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 2803 if (err)
2804 goto abort_free;
2805
f9cb074b 2806 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 2807
77304d2a 2808 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 2809 if (!size) {
2810 printk(KERN_WARNING
2811 "md: %s has zero or unknown size, marking faulty!\n",
2812 bdevname(rdev->bdev,b));
2813 err = -EINVAL;
2814 goto abort_free;
2815 }
2816
2817 if (super_format >= 0) {
2818 err = super_types[super_format].
2819 load_super(rdev, NULL, super_minor);
2820 if (err == -EINVAL) {
df968c4e
N		 2821 printk(KERN_WARNING
2822 "md: %s does not have a valid v%d.%d "
2823 "superblock, not importing!\n",
2824 bdevname(rdev->bdev,b),
2825 super_format, super_minor);
1da177e4
LT		 2826 goto abort_free;
2827 }
2828 if (err < 0) {
2829 printk(KERN_WARNING
2830 "md: could not read %s's sb, not importing!\n",
2831 bdevname(rdev->bdev,b));
2832 goto abort_free;
2833 }
2834 }
6bfe0b49 2835
1da177e4
LT		 2836 return rdev;
2837
2838abort_free:
2839 if (rdev->sb_page) {
2840 if (rdev->bdev)
2841 unlock_rdev(rdev);
2842 free_disk_sb(rdev);
2843 }
2844 kfree(rdev);
2845 return ERR_PTR(err);
2846}
2847
2848/*
2849 * Check a full RAID array for plausibility
2850 */
2851
2852
a757e64c 2853static void analyze_sbs(mddev_t * mddev)
1da177e4
LT		 2854{
2855 int i;
159ec1fc 2856 mdk_rdev_t *rdev, *freshest, *tmp;
1da177e4
LT		 2857 char b[BDEVNAME_SIZE];
2858
2859 freshest = NULL;
d089c6af 2860 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 2861 switch (super_types[mddev->major_version].
2862 load_super(rdev, freshest, mddev->minor_version)) {
2863 case 1:
2864 freshest = rdev;
2865 break;
2866 case 0:
2867 break;
2868 default:
2869 printk( KERN_ERR \
2870 "md: fatal superblock inconsistency in %s"
2871 " -- removing from array\n",
2872 bdevname(rdev->bdev,b));
2873 kick_rdev_from_array(rdev);
2874 }
2875
2876
2877 super_types[mddev->major_version].
2878 validate_super(mddev, freshest);
2879
2880 i = 0;
d089c6af 2881 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 2882 if (mddev->max_disks &&
2883 (rdev->desc_nr >= mddev->max_disks ||
2884 i > mddev->max_disks)) {
de01dfad
N		 2885 printk(KERN_WARNING
2886 "md: %s: %s: only %d devices permitted\n",
2887 mdname(mddev), bdevname(rdev->bdev, b),
2888 mddev->max_disks);
2889 kick_rdev_from_array(rdev);
2890 continue;
2891 }
1da177e4
LT		 2892 if (rdev != freshest)
2893 if (super_types[mddev->major_version].
2894 validate_super(mddev, rdev)) {
2895 printk(KERN_WARNING "md: kicking non-fresh %s"
2896 " from array!\n",
2897 bdevname(rdev->bdev,b));
2898 kick_rdev_from_array(rdev);
2899 continue;
2900 }
2901 if (mddev->level == LEVEL_MULTIPATH) {
2902 rdev->desc_nr = i++;
2903 rdev->raid_disk = rdev->desc_nr;
b2d444d7 2904 set_bit(In_sync, &rdev->flags);
5e5e3e78 2905 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 2906 rdev->raid_disk = -1;
2907 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 2908 }
2909 }
1da177e4
LT		 2910}
2911
72e02075
N		 2912/* Read a fixed-point number.
2913 * Numbers in sysfs attributes should be in "standard" units where
2914 * possible, so time should be in seconds.
2915 * However we internally use a a much smaller unit such as
2916 * milliseconds or jiffies.
2917 * This function takes a decimal number with a possible fractional
2918 * component, and produces an integer which is the result of
2919 * multiplying that number by 10^'scale'.
2920 * all without any floating-point arithmetic.
2921 */
2922int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
2923{
2924 unsigned long result = 0;
2925 long decimals = -1;
2926 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
2927 if (*cp == '.')
2928 decimals = 0;
2929 else if (decimals < scale) {
2930 unsigned int value;
2931 value = *cp - '0';
2932 result = result * 10 + value;
2933 if (decimals >= 0)
2934 decimals++;
2935 }
2936 cp++;
2937 }
2938 if (*cp == '\n')
2939 cp++;
2940 if (*cp)
2941 return -EINVAL;
2942 if (decimals < 0)
2943 decimals = 0;
2944 while (decimals < scale) {
2945 result *= 10;
2946 decimals ++;
2947 }
2948 *res = result;
2949 return 0;
2950}
2951
2952
19052c0e
N		 2953static void md_safemode_timeout(unsigned long data);
2954
16f17b39
N		 2955static ssize_t
2956safe_delay_show(mddev_t *mddev, char *page)
2957{
2958 int msec = (mddev->safemode_delay*1000)/HZ;
2959 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2960}
2961static ssize_t
2962safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2963{
16f17b39 2964 unsigned long msec;
97ce0a7f 2965
72e02075 2966 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 2967 return -EINVAL;
16f17b39
N		 2968 if (msec == 0)
2969 mddev->safemode_delay = 0;
2970 else {
19052c0e 2971 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 2972 mddev->safemode_delay = (msec*HZ)/1000;
2973 if (mddev->safemode_delay == 0)
2974 mddev->safemode_delay = 1;
19052c0e
N		 2975 if (mddev->safemode_delay < old_delay)
2976 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 2977 }
2978 return len;
2979}
2980static struct md_sysfs_entry md_safe_delay =
80ca3a44 2981__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 2982
eae1701f 2983static ssize_t
96de1e66 2984level_show(mddev_t *mddev, char *page)
eae1701f 2985{
2604b703 2986 struct mdk_personality *p = mddev->pers;
d9d166c2 2987 if (p)
eae1701f 2988 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 2989 else if (mddev->clevel[0])
2990 return sprintf(page, "%s\n", mddev->clevel);
2991 else if (mddev->level != LEVEL_NONE)
2992 return sprintf(page, "%d\n", mddev->level);
2993 else
2994 return 0;
eae1701f
N		 2995}
2996
d9d166c2
N		 2997static ssize_t
2998level_store(mddev_t *mddev, const char *buf, size_t len)
2999{
f2859af6 3000 char clevel[16];
20a49ff6 3001 ssize_t rv = len;
245f46c2 3002 struct mdk_personality *pers;
f2859af6 3003 long level;
245f46c2 3004 void *priv;
3a981b03 3005 mdk_rdev_t *rdev;
245f46c2
N		 3006
3007 if (mddev->pers == NULL) {
3008 if (len == 0)
3009 return 0;
3010 if (len >= sizeof(mddev->clevel))
3011 return -ENOSPC;
3012 strncpy(mddev->clevel, buf, len);
3013 if (mddev->clevel[len-1] == '\n')
3014 len--;
3015 mddev->clevel[len] = 0;
3016 mddev->level = LEVEL_NONE;
3017 return rv;
3018 }
3019
3020 /* request to change the personality. Need to ensure:
3021 * - array is not engaged in resync/recovery/reshape
3022 * - old personality can be suspended
3023 * - new personality will access other array.
3024 */
3025
bb4f1e9d
N		 3026 if (mddev->sync_thread ||
3027 mddev->reshape_position != MaxSector ||
3028 mddev->sysfs_active)
d9d166c2 3029 return -EBUSY;
245f46c2
N		 3030
3031 if (!mddev->pers->quiesce) {
3032 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3033 mdname(mddev), mddev->pers->name);
3034 return -EINVAL;
3035 }
3036
3037 /* Now find the new personality */
f2859af6 3038 if (len == 0 || len >= sizeof(clevel))
245f46c2 3039 return -EINVAL;
f2859af6
DW		 3040 strncpy(clevel, buf, len);
3041 if (clevel[len-1] == '\n')
d9d166c2 3042 len--;
f2859af6
DW		 3043 clevel[len] = 0;
3044 if (strict_strtol(clevel, 10, &level))
3045 level = LEVEL_NONE;
245f46c2 3046
f2859af6
DW		 3047 if (request_module("md-%s", clevel) != 0)
3048 request_module("md-level-%s", clevel);
245f46c2 3049 spin_lock(&pers_lock);
f2859af6 3050 pers = find_pers(level, clevel);
245f46c2
N		 3051 if (!pers || !try_module_get(pers->owner)) {
3052 spin_unlock(&pers_lock);
f2859af6 3053 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3054 return -EINVAL;
3055 }
3056 spin_unlock(&pers_lock);
3057
3058 if (pers == mddev->pers) {
3059 /* Nothing to do! */
3060 module_put(pers->owner);
3061 return rv;
3062 }
3063 if (!pers->takeover) {
3064 module_put(pers->owner);
3065 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3066 mdname(mddev), clevel);
245f46c2
N		 3067 return -EINVAL;
3068 }
3069
e93f68a1
N		 3070 list_for_each_entry(rdev, &mddev->disks, same_set)
3071 rdev->new_raid_disk = rdev->raid_disk;
3072
245f46c2
N		 3073 /* ->takeover must set new_* and/or delta_disks
3074 * if it succeeds, and may set them when it fails.
3075 */
3076 priv = pers->takeover(mddev);
3077 if (IS_ERR(priv)) {
3078 mddev->new_level = mddev->level;
3079 mddev->new_layout = mddev->layout;
664e7c41 3080 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3081 mddev->raid_disks -= mddev->delta_disks;
3082 mddev->delta_disks = 0;
3083 module_put(pers->owner);
3084 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3085 mdname(mddev), clevel);
245f46c2
N		 3086 return PTR_ERR(priv);
3087 }
3088
3089 /* Looks like we have a winner */
3090 mddev_suspend(mddev);
3091 mddev->pers->stop(mddev);
a64c876f
N		 3092
3093 if (mddev->pers->sync_request == NULL &&
3094 pers->sync_request != NULL) {
3095 /* need to add the md_redundancy_group */
3096 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3097 printk(KERN_WARNING
3098 "md: cannot register extra attributes for %s\n",
3099 mdname(mddev));
19fdb9ee 3100 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3101 }
3102 if (mddev->pers->sync_request != NULL &&
3103 pers->sync_request == NULL) {
3104 /* need to remove the md_redundancy_group */
3105 if (mddev->to_remove == NULL)
3106 mddev->to_remove = &md_redundancy_group;
3107 }
3108
54071b38
TM		 3109 if (mddev->pers->sync_request == NULL &&
3110 mddev->external) {
3111 /* We are converting from a no-redundancy array
3112 * to a redundancy array and metadata is managed
3113 * externally so we need to be sure that writes
3114 * won't block due to a need to transition
3115 * clean->dirty
3116 * until external management is started.
3117 */
3118 mddev->in_sync = 0;
3119 mddev->safemode_delay = 0;
3120 mddev->safemode = 0;
3121 }
3122
e93f68a1
N		 3123 list_for_each_entry(rdev, &mddev->disks, same_set) {
3124 char nm[20];
3125 if (rdev->raid_disk < 0)
3126 continue;
bf2cb0da 3127 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3128 rdev->new_raid_disk = -1;
3129 if (rdev->new_raid_disk == rdev->raid_disk)
3130 continue;
3131 sprintf(nm, "rd%d", rdev->raid_disk);
3132 sysfs_remove_link(&mddev->kobj, nm);
3133 }
3134 list_for_each_entry(rdev, &mddev->disks, same_set) {
3135 if (rdev->raid_disk < 0)
3136 continue;
3137 if (rdev->new_raid_disk == rdev->raid_disk)
3138 continue;
3139 rdev->raid_disk = rdev->new_raid_disk;
3140 if (rdev->raid_disk < 0)
3a981b03 3141 clear_bit(In_sync, &rdev->flags);
e93f68a1
N		 3142 else {
3143 char nm[20];
3144 sprintf(nm, "rd%d", rdev->raid_disk);
3145 if(sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3146 printk("md: cannot register %s for %s after level change\n",
3147 nm, mdname(mddev));
3a981b03 3148 }
e93f68a1
N		 3149 }
3150
3151 module_put(mddev->pers->owner);
245f46c2
N		 3152 mddev->pers = pers;
3153 mddev->private = priv;
3154 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3155 mddev->level = mddev->new_level;
3156 mddev->layout = mddev->new_layout;
664e7c41 3157 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3158 mddev->delta_disks = 0;
9af204cf
TM		 3159 if (mddev->pers->sync_request == NULL) {
3160 /* this is now an array without redundancy, so
3161 * it must always be in_sync
3162 */
3163 mddev->in_sync = 1;
3164 del_timer_sync(&mddev->safemode_timer);
3165 }
245f46c2
N		 3166 pers->run(mddev);
3167 mddev_resume(mddev);
3168 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3169 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3170 md_wakeup_thread(mddev->thread);
5cac7861 3171 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3172 md_new_event(mddev);
d9d166c2
N		 3173 return rv;
3174}
3175
3176static struct md_sysfs_entry md_level =
80ca3a44 3177__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3178
d4dbd025
N		 3179
3180static ssize_t
3181layout_show(mddev_t *mddev, char *page)
3182{
3183 /* just a number, not meaningful for all levels */
08a02ecd
N		 3184 if (mddev->reshape_position != MaxSector &&
3185 mddev->layout != mddev->new_layout)
3186 return sprintf(page, "%d (%d)\n",
3187 mddev->new_layout, mddev->layout);
d4dbd025
N		 3188 return sprintf(page, "%d\n", mddev->layout);
3189}
3190
3191static ssize_t
3192layout_store(mddev_t *mddev, const char *buf, size_t len)
3193{
3194 char *e;
3195 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3196
3197 if (!*buf || (*e && *e != '\n'))
3198 return -EINVAL;
3199
b3546035
N		 3200 if (mddev->pers) {
3201 int err;
50ac168a 3202 if (mddev->pers->check_reshape == NULL)
b3546035 3203 return -EBUSY;
597a711b 3204 mddev->new_layout = n;
50ac168a 3205 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3206 if (err) {
3207 mddev->new_layout = mddev->layout;
b3546035 3208 return err;
597a711b 3209 }
b3546035 3210 } else {
08a02ecd 3211 mddev->new_layout = n;
b3546035
N		 3212 if (mddev->reshape_position == MaxSector)
3213 mddev->layout = n;
3214 }
d4dbd025
N		 3215 return len;
3216}
3217static struct md_sysfs_entry md_layout =
80ca3a44 3218__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3219
3220
eae1701f 3221static ssize_t
96de1e66 3222raid_disks_show(mddev_t *mddev, char *page)
eae1701f 3223{
bb636547
N		 3224 if (mddev->raid_disks == 0)
3225 return 0;
08a02ecd
N		 3226 if (mddev->reshape_position != MaxSector &&
3227 mddev->delta_disks != 0)
3228 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3229 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3230 return sprintf(page, "%d\n", mddev->raid_disks);
3231}
3232
da943b99
N		 3233static int update_raid_disks(mddev_t *mddev, int raid_disks);
3234
3235static ssize_t
3236raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3237{
da943b99
N		 3238 char *e;
3239 int rv = 0;
3240 unsigned long n = simple_strtoul(buf, &e, 10);
3241
3242 if (!*buf || (*e && *e != '\n'))
3243 return -EINVAL;
3244
3245 if (mddev->pers)
3246 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3247 else if (mddev->reshape_position != MaxSector) {
3248 int olddisks = mddev->raid_disks - mddev->delta_disks;
3249 mddev->delta_disks = n - olddisks;
3250 mddev->raid_disks = n;
3251 } else
da943b99
N		 3252 mddev->raid_disks = n;
3253 return rv ? rv : len;
3254}
3255static struct md_sysfs_entry md_raid_disks =
80ca3a44 3256__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3257
3b34380a
N		 3258static ssize_t
3259chunk_size_show(mddev_t *mddev, char *page)
3260{
08a02ecd 3261 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3262 mddev->chunk_sectors != mddev->new_chunk_sectors)
3263 return sprintf(page, "%d (%d)\n",
3264 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3265 mddev->chunk_sectors << 9);
3266 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3267}
3268
3269static ssize_t
3270chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3271{
3b34380a
N		 3272 char *e;
3273 unsigned long n = simple_strtoul(buf, &e, 10);
3274
3b34380a
N		 3275 if (!*buf || (*e && *e != '\n'))
3276 return -EINVAL;
3277
b3546035
N		 3278 if (mddev->pers) {
3279 int err;
50ac168a 3280 if (mddev->pers->check_reshape == NULL)
b3546035 3281 return -EBUSY;
597a711b 3282 mddev->new_chunk_sectors = n >> 9;
50ac168a 3283 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3284 if (err) {
3285 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3286 return err;
597a711b 3287 }
b3546035 3288 } else {
664e7c41 3289 mddev->new_chunk_sectors = n >> 9;
b3546035 3290 if (mddev->reshape_position == MaxSector)
9d8f0363 3291 mddev->chunk_sectors = n >> 9;
b3546035 3292 }
3b34380a
N		 3293 return len;
3294}
3295static struct md_sysfs_entry md_chunk_size =
80ca3a44 3296__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3297
a94213b1
N		 3298static ssize_t
3299resync_start_show(mddev_t *mddev, char *page)
3300{
d1a7c503
N		 3301 if (mddev->recovery_cp == MaxSector)
3302 return sprintf(page, "none\n");
a94213b1
N		 3303 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3304}
3305
3306static ssize_t
3307resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3308{
a94213b1
N		 3309 char *e;
3310 unsigned long long n = simple_strtoull(buf, &e, 10);
3311
3312 if (mddev->pers)
3313 return -EBUSY;
06e3c817
DW		 3314 if (cmd_match(buf, "none"))
3315 n = MaxSector;
3316 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3317 return -EINVAL;
3318
3319 mddev->recovery_cp = n;
3320 return len;
3321}
3322static struct md_sysfs_entry md_resync_start =
80ca3a44 3323__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3324
9e653b63
N		 3325/*
3326 * The array state can be:
3327 *
3328 * clear
3329 * No devices, no size, no level
3330 * Equivalent to STOP_ARRAY ioctl
3331 * inactive
3332 * May have some settings, but array is not active
3333 * all IO results in error
3334 * When written, doesn't tear down array, but just stops it
3335 * suspended (not supported yet)
3336 * All IO requests will block. The array can be reconfigured.
910d8cb3 3337 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3338 * readonly
3339 * no resync can happen. no superblocks get written.
3340 * write requests fail
3341 * read-auto
3342 * like readonly, but behaves like 'clean' on a write request.
3343 *
3344 * clean - no pending writes, but otherwise active.
3345 * When written to inactive array, starts without resync
3346 * If a write request arrives then
3347 * if metadata is known, mark 'dirty' and switch to 'active'.
3348 * if not known, block and switch to write-pending
3349 * If written to an active array that has pending writes, then fails.
3350 * active
3351 * fully active: IO and resync can be happening.
3352 * When written to inactive array, starts with resync
3353 *
3354 * write-pending
3355 * clean, but writes are blocked waiting for 'active' to be written.
3356 *
3357 * active-idle
3358 * like active, but no writes have been seen for a while (100msec).
3359 *
3360 */
3361enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3362 write_pending, active_idle, bad_word};
05381954 3363static char *array_states[] = {
9e653b63
N		 3364 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3365 "write-pending", "active-idle", NULL };
3366
3367static int match_word(const char *word, char **list)
3368{
3369 int n;
3370 for (n=0; list[n]; n++)
3371 if (cmd_match(word, list[n]))
3372 break;
3373 return n;
3374}
3375
3376static ssize_t
3377array_state_show(mddev_t *mddev, char *page)
3378{
3379 enum array_state st = inactive;
3380
3381 if (mddev->pers)
3382 switch(mddev->ro) {
3383 case 1:
3384 st = readonly;
3385 break;
3386 case 2:
3387 st = read_auto;
3388 break;
3389 case 0:
3390 if (mddev->in_sync)
3391 st = clean;
070dc6dd 3392 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3393 st = write_pending;
9e653b63
N		 3394 else if (mddev->safemode)
3395 st = active_idle;
3396 else
3397 st = active;
3398 }
3399 else {
3400 if (list_empty(&mddev->disks) &&
3401 mddev->raid_disks == 0 &&
58c0fed4 3402 mddev->dev_sectors == 0)
9e653b63
N		 3403 st = clear;
3404 else
3405 st = inactive;
3406 }
3407 return sprintf(page, "%s\n", array_states[st]);
3408}
3409
df5b20cf 3410static int do_md_stop(mddev_t * mddev, int ro, int is_open);
a4bd82d0 3411static int md_set_readonly(mddev_t * mddev, int is_open);
9e653b63
N		 3412static int do_md_run(mddev_t * mddev);
3413static int restart_array(mddev_t *mddev);
3414
3415static ssize_t
3416array_state_store(mddev_t *mddev, const char *buf, size_t len)
3417{
3418 int err = -EINVAL;
3419 enum array_state st = match_word(buf, array_states);
3420 switch(st) {
3421 case bad_word:
3422 break;
3423 case clear:
3424 /* stopping an active array */
f2ea68cf 3425 if (atomic_read(&mddev->openers) > 0)
e691063a 3426 return -EBUSY;
df5b20cf 3427 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3428 break;
3429 case inactive:
3430 /* stopping an active array */
3431 if (mddev->pers) {
f2ea68cf 3432 if (atomic_read(&mddev->openers) > 0)
9e653b63 3433 return -EBUSY;
df5b20cf 3434 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3435 } else
3436 err = 0; /* already inactive */
9e653b63
N		 3437 break;
3438 case suspended:
3439 break; /* not supported yet */
3440 case readonly:
3441 if (mddev->pers)
a4bd82d0 3442 err = md_set_readonly(mddev, 0);
9e653b63
N		 3443 else {
3444 mddev->ro = 1;
648b629e 3445 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3446 err = do_md_run(mddev);
3447 }
3448 break;
3449 case read_auto:
9e653b63 3450 if (mddev->pers) {
80268ee9 3451 if (mddev->ro == 0)
a4bd82d0 3452 err = md_set_readonly(mddev, 0);
80268ee9 3453 else if (mddev->ro == 1)
648b629e
N		 3454 err = restart_array(mddev);
3455 if (err == 0) {
3456 mddev->ro = 2;
3457 set_disk_ro(mddev->gendisk, 0);
3458 }
9e653b63
N		 3459 } else {
3460 mddev->ro = 2;
3461 err = do_md_run(mddev);
3462 }
3463 break;
3464 case clean:
3465 if (mddev->pers) {
3466 restart_array(mddev);
3467 spin_lock_irq(&mddev->write_lock);
3468 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3469 if (mddev->in_sync == 0) {
3470 mddev->in_sync = 1;
31a59e34
N		 3471 if (mddev->safemode == 1)
3472 mddev->safemode = 0;
070dc6dd 3473 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3474 }
3475 err = 0;
3476 } else
3477 err = -EBUSY;
9e653b63 3478 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3479 } else
3480 err = -EINVAL;
9e653b63
N		 3481 break;
3482 case active:
3483 if (mddev->pers) {
3484 restart_array(mddev);
070dc6dd 3485 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3486 wake_up(&mddev->sb_wait);
3487 err = 0;
3488 } else {
3489 mddev->ro = 0;
648b629e 3490 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3491 err = do_md_run(mddev);
3492 }
3493 break;
3494 case write_pending:
3495 case active_idle:
3496 /* these cannot be set */
3497 break;
3498 }
3499 if (err)
3500 return err;
0fd62b86 3501 else {
00bcb4ac 3502 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3503 return len;
0fd62b86 3504 }
9e653b63 3505}
80ca3a44
N		 3506static struct md_sysfs_entry md_array_state =
3507__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3508
1e50915f
RB		 3509static ssize_t
3510max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3511 return sprintf(page, "%d\n",
3512 atomic_read(&mddev->max_corr_read_errors));
3513}
3514
3515static ssize_t
3516max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3517{
3518 char *e;
3519 unsigned long n = simple_strtoul(buf, &e, 10);
3520
3521 if (*buf && (*e == 0 || *e == '\n')) {
3522 atomic_set(&mddev->max_corr_read_errors, n);
3523 return len;
3524 }
3525 return -EINVAL;
3526}
3527
3528static struct md_sysfs_entry max_corr_read_errors =
3529__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3530 max_corrected_read_errors_store);
3531
6d7ff738
N		 3532static ssize_t
3533null_show(mddev_t *mddev, char *page)
3534{
3535 return -EINVAL;
3536}
3537
3538static ssize_t
3539new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3540{
3541 /* buf must be %d:%d\n? giving major and minor numbers */
3542 /* The new device is added to the array.
3543 * If the array has a persistent superblock, we read the
3544 * superblock to initialise info and check validity.
3545 * Otherwise, only checking done is that in bind_rdev_to_array,
3546 * which mainly checks size.
3547 */
3548 char *e;
3549 int major = simple_strtoul(buf, &e, 10);
3550 int minor;
3551 dev_t dev;
3552 mdk_rdev_t *rdev;
3553 int err;
3554
3555 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3556 return -EINVAL;
3557 minor = simple_strtoul(e+1, &e, 10);
3558 if (*e && *e != '\n')
3559 return -EINVAL;
3560 dev = MKDEV(major, minor);
3561 if (major != MAJOR(dev) ||
3562 minor != MINOR(dev))
3563 return -EOVERFLOW;
3564
3565
3566 if (mddev->persistent) {
3567 rdev = md_import_device(dev, mddev->major_version,
3568 mddev->minor_version);
3569 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3570 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3571 mdk_rdev_t, same_set);
3572 err = super_types[mddev->major_version]
3573 .load_super(rdev, rdev0, mddev->minor_version);
3574 if (err < 0)
3575 goto out;
3576 }
c5d79adb
N		 3577 } else if (mddev->external)
3578 rdev = md_import_device(dev, -2, -1);
3579 else
6d7ff738
N		 3580 rdev = md_import_device(dev, -1, -1);
3581
3582 if (IS_ERR(rdev))
3583 return PTR_ERR(rdev);
3584 err = bind_rdev_to_array(rdev, mddev);
3585 out:
3586 if (err)
3587 export_rdev(rdev);
3588 return err ? err : len;
3589}
3590
3591static struct md_sysfs_entry md_new_device =
80ca3a44 3592__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3593
9b1d1dac
PC		 3594static ssize_t
3595bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3596{
3597 char *end;
3598 unsigned long chunk, end_chunk;
3599
3600 if (!mddev->bitmap)
3601 goto out;
3602 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3603 while (*buf) {
3604 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3605 if (buf == end) break;
3606 if (*end == '-') { /* range */
3607 buf = end + 1;
3608 end_chunk = simple_strtoul(buf, &end, 0);
3609 if (buf == end) break;
3610 }
3611 if (*end && !isspace(*end)) break;
3612 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3613 buf = skip_spaces(end);
9b1d1dac
PC		 3614 }
3615 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3616out:
3617 return len;
3618}
3619
3620static struct md_sysfs_entry md_bitmap =
3621__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3622
a35b0d69
N		 3623static ssize_t
3624size_show(mddev_t *mddev, char *page)
3625{
58c0fed4
AN		 3626 return sprintf(page, "%llu\n",
3627 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3628}
3629
d71f9f88 3630static int update_size(mddev_t *mddev, sector_t num_sectors);
a35b0d69
N		 3631
3632static ssize_t
3633size_store(mddev_t *mddev, const char *buf, size_t len)
3634{
3635 /* If array is inactive, we can reduce the component size, but
3636 * not increase it (except from 0).
3637 * If array is active, we can try an on-line resize
3638 */
b522adcd
DW		 3639 sector_t sectors;
3640 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3641
58c0fed4
AN		 3642 if (err < 0)
3643 return err;
a35b0d69 3644 if (mddev->pers) {
58c0fed4 3645 err = update_size(mddev, sectors);
850b2b42 3646 md_update_sb(mddev, 1);
a35b0d69 3647 } else {
58c0fed4
AN		 3648 if (mddev->dev_sectors == 0 ||
3649 mddev->dev_sectors > sectors)
3650 mddev->dev_sectors = sectors;
a35b0d69
N		 3651 else
3652 err = -ENOSPC;
3653 }
3654 return err ? err : len;
3655}
3656
3657static struct md_sysfs_entry md_size =
80ca3a44 3658__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3659
8bb93aac
N		 3660
3661/* Metdata version.
e691063a
N		 3662 * This is one of
3663 * 'none' for arrays with no metadata (good luck...)
3664 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3665 * or N.M for internally known formats
3666 */
3667static ssize_t
3668metadata_show(mddev_t *mddev, char *page)
3669{
3670 if (mddev->persistent)
3671 return sprintf(page, "%d.%d\n",
3672 mddev->major_version, mddev->minor_version);
e691063a
N		 3673 else if (mddev->external)
3674 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3675 else
3676 return sprintf(page, "none\n");
3677}
3678
3679static ssize_t
3680metadata_store(mddev_t *mddev, const char *buf, size_t len)
3681{
3682 int major, minor;
3683 char *e;
ea43ddd8
N		 3684 /* Changing the details of 'external' metadata is
3685 * always permitted. Otherwise there must be
3686 * no devices attached to the array.
3687 */
3688 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3689 ;
3690 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3691 return -EBUSY;
3692
3693 if (cmd_match(buf, "none")) {
3694 mddev->persistent = 0;
e691063a
N		 3695 mddev->external = 0;
3696 mddev->major_version = 0;
3697 mddev->minor_version = 90;
3698 return len;
3699 }
3700 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3701 size_t namelen = len-9;
e691063a
N		 3702 if (namelen >= sizeof(mddev->metadata_type))
3703 namelen = sizeof(mddev->metadata_type)-1;
3704 strncpy(mddev->metadata_type, buf+9, namelen);
3705 mddev->metadata_type[namelen] = 0;
3706 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3707 mddev->metadata_type[--namelen] = 0;
3708 mddev->persistent = 0;
3709 mddev->external = 1;
8bb93aac
N		 3710 mddev->major_version = 0;
3711 mddev->minor_version = 90;
3712 return len;
3713 }
3714 major = simple_strtoul(buf, &e, 10);
3715 if (e==buf || *e != '.')
3716 return -EINVAL;
3717 buf = e+1;
3718 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 3719 if (e==buf || (*e && *e != '\n') )
8bb93aac 3720 return -EINVAL;
50511da3 3721 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 3722 return -ENOENT;
3723 mddev->major_version = major;
3724 mddev->minor_version = minor;
3725 mddev->persistent = 1;
e691063a 3726 mddev->external = 0;
8bb93aac
N		 3727 return len;
3728}
3729
3730static struct md_sysfs_entry md_metadata =
80ca3a44 3731__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 3732
24dd469d 3733static ssize_t
7eec314d 3734action_show(mddev_t *mddev, char *page)
24dd469d 3735{
7eec314d 3736 char *type = "idle";
b6a9ce68
N		 3737 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3738 type = "frozen";
3739 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 3740 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 3741 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3742 type = "reshape";
3743 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 3744 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3745 type = "resync";
3746 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3747 type = "check";
3748 else
3749 type = "repair";
72a23c21 3750 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 3751 type = "recover";
3752 }
3753 return sprintf(page, "%s\n", type);
3754}
3755
7ebc0be7
N		 3756static void reap_sync_thread(mddev_t *mddev);
3757
24dd469d 3758static ssize_t
7eec314d 3759action_store(mddev_t *mddev, const char *page, size_t len)
24dd469d 3760{
7eec314d
N		 3761 if (!mddev->pers || !mddev->pers->sync_request)
3762 return -EINVAL;
3763
b6a9ce68
N		 3764 if (cmd_match(page, "frozen"))
3765 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3766 else
3767 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3768
3769 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 3770 if (mddev->sync_thread) {
3771 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 3772 reap_sync_thread(mddev);
7eec314d 3773 }
03c902e1
N		 3774 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3775 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 3776 return -EBUSY;
72a23c21
NB		 3777 else if (cmd_match(page, "resync"))
3778 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3779 else if (cmd_match(page, "recover")) {
3780 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 3781 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 3782 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 3783 int err;
3784 if (mddev->pers->start_reshape == NULL)
3785 return -EINVAL;
3786 err = mddev->pers->start_reshape(mddev);
3787 if (err)
3788 return err;
a99ac971 3789 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 3790 } else {
bce74dac 3791 if (cmd_match(page, "check"))
7eec314d 3792 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 3793 else if (!cmd_match(page, "repair"))
7eec314d
N		 3794 return -EINVAL;
3795 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3796 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 3797 }
03c902e1 3798 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 3799 md_wakeup_thread(mddev->thread);
00bcb4ac 3800 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 3801 return len;
3802}
3803
9d88883e 3804static ssize_t
96de1e66 3805mismatch_cnt_show(mddev_t *mddev, char *page)
9d88883e
N		 3806{
3807 return sprintf(page, "%llu\n",
3808 (unsigned long long) mddev->resync_mismatches);
3809}
3810
80ca3a44
N		 3811static struct md_sysfs_entry md_scan_mode =
3812__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 3813
96de1e66 3814
80ca3a44 3815static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 3816
88202a0c
N		 3817static ssize_t
3818sync_min_show(mddev_t *mddev, char *page)
3819{
3820 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3821 mddev->sync_speed_min ? "local": "system");
3822}
3823
3824static ssize_t
3825sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3826{
3827 int min;
3828 char *e;
3829 if (strncmp(buf, "system", 6)==0) {
3830 mddev->sync_speed_min = 0;
3831 return len;
3832 }
3833 min = simple_strtoul(buf, &e, 10);
3834 if (buf == e || (*e && *e != '\n') || min <= 0)
3835 return -EINVAL;
3836 mddev->sync_speed_min = min;
3837 return len;
3838}
3839
3840static struct md_sysfs_entry md_sync_min =
3841__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3842
3843static ssize_t
3844sync_max_show(mddev_t *mddev, char *page)
3845{
3846 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3847 mddev->sync_speed_max ? "local": "system");
3848}
3849
3850static ssize_t
3851sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3852{
3853 int max;
3854 char *e;
3855 if (strncmp(buf, "system", 6)==0) {
3856 mddev->sync_speed_max = 0;
3857 return len;
3858 }
3859 max = simple_strtoul(buf, &e, 10);
3860 if (buf == e || (*e && *e != '\n') || max <= 0)
3861 return -EINVAL;
3862 mddev->sync_speed_max = max;
3863 return len;
3864}
3865
3866static struct md_sysfs_entry md_sync_max =
3867__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3868
d7f3d291
IP		 3869static ssize_t
3870degraded_show(mddev_t *mddev, char *page)
3871{
3872 return sprintf(page, "%d\n", mddev->degraded);
3873}
3874static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 3875
90b08710
BS		 3876static ssize_t
3877sync_force_parallel_show(mddev_t *mddev, char *page)
3878{
3879 return sprintf(page, "%d\n", mddev->parallel_resync);
3880}
3881
3882static ssize_t
3883sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3884{
3885 long n;
3886
3887 if (strict_strtol(buf, 10, &n))
3888 return -EINVAL;
3889
3890 if (n != 0 && n != 1)
3891 return -EINVAL;
3892
3893 mddev->parallel_resync = n;
3894
3895 if (mddev->sync_thread)
3896 wake_up(&resync_wait);
3897
3898 return len;
3899}
3900
3901/* force parallel resync, even with shared block devices */
3902static struct md_sysfs_entry md_sync_force_parallel =
3903__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3904 sync_force_parallel_show, sync_force_parallel_store);
3905
88202a0c
N		 3906static ssize_t
3907sync_speed_show(mddev_t *mddev, char *page)
3908{
3909 unsigned long resync, dt, db;
d1a7c503
N		 3910 if (mddev->curr_resync == 0)
3911 return sprintf(page, "none\n");
9687a60c
AN		 3912 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3913 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 3914 if (!dt) dt++;
9687a60c
AN		 3915 db = resync - mddev->resync_mark_cnt;
3916 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 3917}
3918
80ca3a44 3919static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 3920
3921static ssize_t
3922sync_completed_show(mddev_t *mddev, char *page)
3923{
13ae864b 3924 unsigned long long max_sectors, resync;
88202a0c 3925
acb180b0
N		 3926 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3927 return sprintf(page, "none\n");
3928
88202a0c 3929 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 3930 max_sectors = mddev->resync_max_sectors;
88202a0c 3931 else
58c0fed4 3932 max_sectors = mddev->dev_sectors;
88202a0c 3933
acb180b0 3934 resync = mddev->curr_resync_completed;
13ae864b 3935 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 3936}
3937
80ca3a44 3938static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 3939
5e96ee65
NB		 3940static ssize_t
3941min_sync_show(mddev_t *mddev, char *page)
3942{
3943 return sprintf(page, "%llu\n",
3944 (unsigned long long)mddev->resync_min);
3945}
3946static ssize_t
3947min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3948{
3949 unsigned long long min;
3950 if (strict_strtoull(buf, 10, &min))
3951 return -EINVAL;
3952 if (min > mddev->resync_max)
3953 return -EINVAL;
3954 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3955 return -EBUSY;
3956
3957 /* Must be a multiple of chunk_size */
9d8f0363 3958 if (mddev->chunk_sectors) {
2ac06c33 3959 sector_t temp = min;
9d8f0363 3960 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 3961 return -EINVAL;
3962 }
3963 mddev->resync_min = min;
3964
3965 return len;
3966}
3967
3968static struct md_sysfs_entry md_min_sync =
3969__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3970
c6207277
N		 3971static ssize_t
3972max_sync_show(mddev_t *mddev, char *page)
3973{
3974 if (mddev->resync_max == MaxSector)
3975 return sprintf(page, "max\n");
3976 else
3977 return sprintf(page, "%llu\n",
3978 (unsigned long long)mddev->resync_max);
3979}
3980static ssize_t
3981max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3982{
3983 if (strncmp(buf, "max", 3) == 0)
3984 mddev->resync_max = MaxSector;
3985 else {
5e96ee65
NB		 3986 unsigned long long max;
3987 if (strict_strtoull(buf, 10, &max))
3988 return -EINVAL;
3989 if (max < mddev->resync_min)
c6207277
N		 3990 return -EINVAL;
3991 if (max < mddev->resync_max &&
4d484a4a 3992 mddev->ro == 0 &&
c6207277
N		 3993 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3994 return -EBUSY;
3995
3996 /* Must be a multiple of chunk_size */
9d8f0363 3997 if (mddev->chunk_sectors) {
2ac06c33 3998 sector_t temp = max;
9d8f0363 3999 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4000 return -EINVAL;
4001 }
4002 mddev->resync_max = max;
4003 }
4004 wake_up(&mddev->recovery_wait);
4005 return len;
4006}
4007
4008static struct md_sysfs_entry md_max_sync =
4009__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4010
e464eafd
N		 4011static ssize_t
4012suspend_lo_show(mddev_t *mddev, char *page)
4013{
4014 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4015}
4016
4017static ssize_t
4018suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4019{
4020 char *e;
4021 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4022 unsigned long long old = mddev->suspend_lo;
e464eafd 4023
b8d966ef
N		 4024 if (mddev->pers == NULL ||
4025 mddev->pers->quiesce == NULL)
e464eafd
N		 4026 return -EINVAL;
4027 if (buf == e || (*e && *e != '\n'))
4028 return -EINVAL;
23ddff37
N		 4029
4030 mddev->suspend_lo = new;
4031 if (new >= old)
4032 /* Shrinking suspended region */
e464eafd 4033 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4034 else {
4035 /* Expanding suspended region - need to wait */
4036 mddev->pers->quiesce(mddev, 1);
4037 mddev->pers->quiesce(mddev, 0);
4038 }
4039 return len;
e464eafd
N		 4040}
4041static struct md_sysfs_entry md_suspend_lo =
4042__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4043
4044
4045static ssize_t
4046suspend_hi_show(mddev_t *mddev, char *page)
4047{
4048 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4049}
4050
4051static ssize_t
4052suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4053{
4054 char *e;
4055 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4056 unsigned long long old = mddev->suspend_hi;
e464eafd 4057
b8d966ef
N		 4058 if (mddev->pers == NULL ||
4059 mddev->pers->quiesce == NULL)
e464eafd
N		 4060 return -EINVAL;
4061 if (buf == e || (*e && *e != '\n'))
4062 return -EINVAL;
23ddff37
N		 4063
4064 mddev->suspend_hi = new;
4065 if (new <= old)
4066 /* Shrinking suspended region */
4067 mddev->pers->quiesce(mddev, 2);
4068 else {
4069 /* Expanding suspended region - need to wait */
e464eafd
N		 4070 mddev->pers->quiesce(mddev, 1);
4071 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4072 }
4073 return len;
e464eafd
N		 4074}
4075static struct md_sysfs_entry md_suspend_hi =
4076__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4077
08a02ecd
N		 4078static ssize_t
4079reshape_position_show(mddev_t *mddev, char *page)
4080{
4081 if (mddev->reshape_position != MaxSector)
4082 return sprintf(page, "%llu\n",
4083 (unsigned long long)mddev->reshape_position);
4084 strcpy(page, "none\n");
4085 return 5;
4086}
4087
4088static ssize_t
4089reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4090{
4091 char *e;
4092 unsigned long long new = simple_strtoull(buf, &e, 10);
4093 if (mddev->pers)
4094 return -EBUSY;
4095 if (buf == e || (*e && *e != '\n'))
4096 return -EINVAL;
4097 mddev->reshape_position = new;
4098 mddev->delta_disks = 0;
4099 mddev->new_level = mddev->level;
4100 mddev->new_layout = mddev->layout;
664e7c41 4101 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4102 return len;
4103}
4104
4105static struct md_sysfs_entry md_reshape_position =
4106__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4107 reshape_position_store);
4108
b522adcd
DW		 4109static ssize_t
4110array_size_show(mddev_t *mddev, char *page)
4111{
4112 if (mddev->external_size)
4113 return sprintf(page, "%llu\n",
4114 (unsigned long long)mddev->array_sectors/2);
4115 else
4116 return sprintf(page, "default\n");
4117}
4118
4119static ssize_t
4120array_size_store(mddev_t *mddev, const char *buf, size_t len)
4121{
4122 sector_t sectors;
4123
4124 if (strncmp(buf, "default", 7) == 0) {
4125 if (mddev->pers)
4126 sectors = mddev->pers->size(mddev, 0, 0);
4127 else
4128 sectors = mddev->array_sectors;
4129
4130 mddev->external_size = 0;
4131 } else {
4132 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4133 return -EINVAL;
4134 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4135 return -E2BIG;
b522adcd
DW		 4136
4137 mddev->external_size = 1;
4138 }
4139
4140 mddev->array_sectors = sectors;
cbe6ef1d
N		 4141 if (mddev->pers) {
4142 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4143 revalidate_disk(mddev->gendisk);
cbe6ef1d 4144 }
b522adcd
DW		 4145 return len;
4146}
4147
4148static struct md_sysfs_entry md_array_size =
4149__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4150 array_size_store);
e464eafd 4151
eae1701f
N		 4152static struct attribute *md_default_attrs[] = {
4153 &md_level.attr,
d4dbd025 4154 &md_layout.attr,
eae1701f 4155 &md_raid_disks.attr,
3b34380a 4156 &md_chunk_size.attr,
a35b0d69 4157 &md_size.attr,
a94213b1 4158 &md_resync_start.attr,
8bb93aac 4159 &md_metadata.attr,
6d7ff738 4160 &md_new_device.attr,
16f17b39 4161 &md_safe_delay.attr,
9e653b63 4162 &md_array_state.attr,
08a02ecd 4163 &md_reshape_position.attr,
b522adcd 4164 &md_array_size.attr,
1e50915f 4165 &max_corr_read_errors.attr,
411036fa
N		 4166 NULL,
4167};
4168
4169static struct attribute *md_redundancy_attrs[] = {
24dd469d 4170 &md_scan_mode.attr,
9d88883e 4171 &md_mismatches.attr,
88202a0c
N		 4172 &md_sync_min.attr,
4173 &md_sync_max.attr,
4174 &md_sync_speed.attr,
90b08710 4175 &md_sync_force_parallel.attr,
88202a0c 4176 &md_sync_completed.attr,
5e96ee65 4177 &md_min_sync.attr,
c6207277 4178 &md_max_sync.attr,
e464eafd
N		 4179 &md_suspend_lo.attr,
4180 &md_suspend_hi.attr,
9b1d1dac 4181 &md_bitmap.attr,
d7f3d291 4182 &md_degraded.attr,
eae1701f
N		 4183 NULL,
4184};
411036fa
N		 4185static struct attribute_group md_redundancy_group = {
4186 .name = NULL,
4187 .attrs = md_redundancy_attrs,
4188};
4189
eae1701f
N		 4190
4191static ssize_t
4192md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4193{
4194 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4195 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4196 ssize_t rv;
eae1701f
N		 4197
4198 if (!entry->show)
4199 return -EIO;
5dc5cf7d
IM		 4200 rv = mddev_lock(mddev);
4201 if (!rv) {
4202 rv = entry->show(mddev, page);
4203 mddev_unlock(mddev);
4204 }
96de1e66 4205 return rv;
eae1701f
N		 4206}
4207
4208static ssize_t
4209md_attr_store(struct kobject *kobj, struct attribute *attr,
4210 const char *page, size_t length)
4211{
4212 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4213 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
96de1e66 4214 ssize_t rv;
eae1701f
N		 4215
4216 if (!entry->store)
4217 return -EIO;
67463acb
N		 4218 if (!capable(CAP_SYS_ADMIN))
4219 return -EACCES;
5dc5cf7d 4220 rv = mddev_lock(mddev);
d3374825
N		 4221 if (mddev->hold_active == UNTIL_IOCTL)
4222 mddev->hold_active = 0;
5dc5cf7d
IM		 4223 if (!rv) {
4224 rv = entry->store(mddev, page, length);
4225 mddev_unlock(mddev);
4226 }
96de1e66 4227 return rv;
eae1701f
N		 4228}
4229
4230static void md_free(struct kobject *ko)
4231{
4232 mddev_t *mddev = container_of(ko, mddev_t, kobj);
a21d1504
N		 4233
4234 if (mddev->sysfs_state)
4235 sysfs_put(mddev->sysfs_state);
4236
4237 if (mddev->gendisk) {
4238 del_gendisk(mddev->gendisk);
4239 put_disk(mddev->gendisk);
4240 }
4241 if (mddev->queue)
4242 blk_cleanup_queue(mddev->queue);
4243
eae1701f
N		 4244 kfree(mddev);
4245}
4246
52cf25d0 4247static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4248 .show = md_attr_show,
4249 .store = md_attr_store,
4250};
4251static struct kobj_type md_ktype = {
4252 .release = md_free,
4253 .sysfs_ops = &md_sysfs_ops,
4254 .default_attrs = md_default_attrs,
4255};
4256
1da177e4
LT		 4257int mdp_major = 0;
4258
5fd3a17e
DW		 4259static void mddev_delayed_delete(struct work_struct *ws)
4260{
4261 mddev_t *mddev = container_of(ws, mddev_t, del_work);
4262
43a70507 4263 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4264 kobject_del(&mddev->kobj);
4265 kobject_put(&mddev->kobj);
4266}
4267
efeb53c0 4268static int md_alloc(dev_t dev, char *name)
1da177e4 4269{
48c9c27b 4270 static DEFINE_MUTEX(disks_mutex);
1da177e4
LT		 4271 mddev_t *mddev = mddev_find(dev);
4272 struct gendisk *disk;
efeb53c0
N		 4273 int partitioned;
4274 int shift;
4275 int unit;
3830c62f 4276 int error;
1da177e4
LT		 4277
4278 if (!mddev)
efeb53c0
N		 4279 return -ENODEV;
4280
4281 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4282 shift = partitioned ? MdpMinorShift : 0;
4283 unit = MINOR(mddev->unit) >> shift;
1da177e4 4284
e804ac78
TH		 4285 /* wait for any previous instance of this device to be
4286 * completely removed (mddev_delayed_delete).
d3374825 4287 */
e804ac78 4288 flush_workqueue(md_misc_wq);
d3374825 4289
48c9c27b 4290 mutex_lock(&disks_mutex);
0909dc44
N		 4291 error = -EEXIST;
4292 if (mddev->gendisk)
4293 goto abort;
efeb53c0
N		 4294
4295 if (name) {
4296 /* Need to ensure that 'name' is not a duplicate.
4297 */
4298 mddev_t *mddev2;
4299 spin_lock(&all_mddevs_lock);
4300
4301 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4302 if (mddev2->gendisk &&
4303 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4304 spin_unlock(&all_mddevs_lock);
0909dc44 4305 goto abort;
efeb53c0
N		 4306 }
4307 spin_unlock(&all_mddevs_lock);
1da177e4 4308 }
8b765398 4309
0909dc44 4310 error = -ENOMEM;
8b765398 4311 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4312 if (!mddev->queue)
4313 goto abort;
409c57f3
N		 4314 mddev->queue->queuedata = mddev;
4315
409c57f3 4316 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4317
1da177e4
LT		 4318 disk = alloc_disk(1 << shift);
4319 if (!disk) {
8b765398
N		 4320 blk_cleanup_queue(mddev->queue);
4321 mddev->queue = NULL;
0909dc44 4322 goto abort;
1da177e4 4323 }
efeb53c0 4324 disk->major = MAJOR(mddev->unit);
1da177e4 4325 disk->first_minor = unit << shift;
efeb53c0
N		 4326 if (name)
4327 strcpy(disk->disk_name, name);
4328 else if (partitioned)
1da177e4 4329 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4330 else
1da177e4 4331 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4332 disk->fops = &md_fops;
4333 disk->private_data = mddev;
4334 disk->queue = mddev->queue;
92850bbd 4335 /* Allow extended partitions. This makes the
d3374825 4336 * 'mdp' device redundant, but we can't really
92850bbd
N		 4337 * remove it now.
4338 */
4339 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4
LT		 4340 add_disk(disk);
4341 mddev->gendisk = disk;
ed9e1982
TH		 4342 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4343 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4344 if (error) {
4345 /* This isn't possible, but as kobject_init_and_add is marked
4346 * __must_check, we must do something with the result
4347 */
5e55e2f5
N		 4348 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4349 disk->disk_name);
0909dc44
N		 4350 error = 0;
4351 }
00bcb4ac
N		 4352 if (mddev->kobj.sd &&
4353 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4354 printk(KERN_DEBUG "pointless warning\n");
be20e6c6
DW		 4355
4356 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
0909dc44
N		 4357 abort:
4358 mutex_unlock(&disks_mutex);
00bcb4ac 4359 if (!error && mddev->kobj.sd) {
3830c62f 4360 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4361 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4362 }
d3374825 4363 mddev_put(mddev);
0909dc44 4364 return error;
efeb53c0
N		 4365}
4366
4367static struct kobject *md_probe(dev_t dev, int *part, void *data)
4368{
4369 md_alloc(dev, NULL);
1da177e4
LT		 4370 return NULL;
4371}
4372
efeb53c0
N		 4373static int add_named_array(const char *val, struct kernel_param *kp)
4374{
4375 /* val must be "md_*" where * is not all digits.
4376 * We allocate an array with a large free minor number, and
4377 * set the name to val. val must not already be an active name.
4378 */
4379 int len = strlen(val);
4380 char buf[DISK_NAME_LEN];
4381
4382 while (len && val[len-1] == '\n')
4383 len--;
4384 if (len >= DISK_NAME_LEN)
4385 return -E2BIG;
4386 strlcpy(buf, val, len+1);
4387 if (strncmp(buf, "md_", 3) != 0)
4388 return -EINVAL;
4389 return md_alloc(0, buf);
4390}
4391
1da177e4
LT		 4392static void md_safemode_timeout(unsigned long data)
4393{
4394 mddev_t *mddev = (mddev_t *) data;
4395
0fd62b86
NB		 4396 if (!atomic_read(&mddev->writes_pending)) {
4397 mddev->safemode = 1;
4398 if (mddev->external)
00bcb4ac 4399 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4400 }
1da177e4
LT		 4401 md_wakeup_thread(mddev->thread);
4402}
4403
6ff8d8ec 4404static int start_dirty_degraded;
1da177e4 4405
390ee602 4406int md_run(mddev_t *mddev)
1da177e4 4407{
2604b703 4408 int err;
1da177e4 4409 mdk_rdev_t *rdev;
2604b703 4410 struct mdk_personality *pers;
1da177e4 4411
a757e64c
N		 4412 if (list_empty(&mddev->disks))
4413 /* cannot run an array with no devices.. */
1da177e4 4414 return -EINVAL;
1da177e4
LT		 4415
4416 if (mddev->pers)
4417 return -EBUSY;
bb4f1e9d
N		 4418 /* Cannot run until previous stop completes properly */
4419 if (mddev->sysfs_active)
4420 return -EBUSY;
b6eb127d 4421
1da177e4
LT		 4422 /*
4423 * Analyze all RAID superblock(s)
4424 */
1ec4a939
N		 4425 if (!mddev->raid_disks) {
4426 if (!mddev->persistent)
4427 return -EINVAL;
a757e64c 4428 analyze_sbs(mddev);
1ec4a939 4429 }
1da177e4 4430
d9d166c2
N		 4431 if (mddev->level != LEVEL_NONE)
4432 request_module("md-level-%d", mddev->level);
4433 else if (mddev->clevel[0])
4434 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4435
4436 /*
4437 * Drop all container device buffers, from now on
4438 * the only valid external interface is through the md
4439 * device.
1da177e4 4440 */
159ec1fc 4441 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4442 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4443 continue;
4444 sync_blockdev(rdev->bdev);
f98393a6 4445 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4446
4447 /* perform some consistency tests on the device.
4448 * We don't want the data to overlap the metadata,
58c0fed4 4449 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4450 */
a6ff7e08
JB		 4451 if (rdev->meta_bdev) {
4452 /* Nothing to check */;
4453 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4454 if (mddev->dev_sectors &&
4455 rdev->data_offset + mddev->dev_sectors
0f420358 4456 > rdev->sb_start) {
f0d76d70
N		 4457 printk("md: %s: data overlaps metadata\n",
4458 mdname(mddev));
4459 return -EINVAL;
4460 }
4461 } else {
0f420358 4462 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4463 > rdev->data_offset) {
4464 printk("md: %s: metadata overlaps data\n",
4465 mdname(mddev));
4466 return -EINVAL;
4467 }
4468 }
00bcb4ac 4469 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4470 }
4471
a167f663
N		 4472 if (mddev->bio_set == NULL)
4473 mddev->bio_set = bioset_create(BIO_POOL_SIZE, sizeof(mddev));
4474
1da177e4 4475 spin_lock(&pers_lock);
d9d166c2 4476 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4477 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4478 spin_unlock(&pers_lock);
d9d166c2
N		 4479 if (mddev->level != LEVEL_NONE)
4480 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4481 mddev->level);
4482 else
4483 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4484 mddev->clevel);
1da177e4
LT		 4485 return -EINVAL;
4486 }
2604b703 4487 mddev->pers = pers;
1da177e4 4488 spin_unlock(&pers_lock);
34817e8c
N		 4489 if (mddev->level != pers->level) {
4490 mddev->level = pers->level;
4491 mddev->new_level = pers->level;
4492 }
d9d166c2 4493 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4494
f6705578 4495 if (mddev->reshape_position != MaxSector &&
63c70c4f 4496 pers->start_reshape == NULL) {
f6705578
N		 4497 /* This personality cannot handle reshaping... */
4498 mddev->pers = NULL;
4499 module_put(pers->owner);
4500 return -EINVAL;
4501 }
4502
7dd5e7c3
N		 4503 if (pers->sync_request) {
4504 /* Warn if this is a potentially silly
4505 * configuration.
4506 */
4507 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4508 mdk_rdev_t *rdev2;
7dd5e7c3 4509 int warned = 0;
159ec1fc
CR		 4510
4511 list_for_each_entry(rdev, &mddev->disks, same_set)
4512 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4513 if (rdev < rdev2 &&
4514 rdev->bdev->bd_contains ==
4515 rdev2->bdev->bd_contains) {
4516 printk(KERN_WARNING
4517 "%s: WARNING: %s appears to be"
4518 " on the same physical disk as"
4519 " %s.\n",
4520 mdname(mddev),
4521 bdevname(rdev->bdev,b),
4522 bdevname(rdev2->bdev,b2));
4523 warned = 1;
4524 }
4525 }
159ec1fc 4526
7dd5e7c3
N		 4527 if (warned)
4528 printk(KERN_WARNING
4529 "True protection against single-disk"
4530 " failure might be compromised.\n");
4531 }
4532
657390d2 4533 mddev->recovery = 0;
58c0fed4
AN		 4534 /* may be over-ridden by personality */
4535 mddev->resync_max_sectors = mddev->dev_sectors;
4536
6ff8d8ec 4537 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4538
0f9552b5 4539 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4540 mddev->ro = 2; /* read-only, but switch on first write */
4541
b15c2e57 4542 err = mddev->pers->run(mddev);
13e53df3
AN		 4543 if (err)
4544 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4545 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4546 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4547 " but 'external_size' not in effect?\n", __func__);
4548 printk(KERN_ERR
4549 "md: invalid array_size %llu > default size %llu\n",
4550 (unsigned long long)mddev->array_sectors / 2,
4551 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4552 err = -EINVAL;
4553 mddev->pers->stop(mddev);
4554 }
4555 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4556 err = bitmap_create(mddev);
4557 if (err) {
4558 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4559 mdname(mddev), err);
4560 mddev->pers->stop(mddev);
4561 }
4562 }
1da177e4 4563 if (err) {
1da177e4
LT		 4564 module_put(mddev->pers->owner);
4565 mddev->pers = NULL;
32a7627c
N		 4566 bitmap_destroy(mddev);
4567 return err;
1da177e4 4568 }
5e55e2f5 4569 if (mddev->pers->sync_request) {
00bcb4ac
N		 4570 if (mddev->kobj.sd &&
4571 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4572 printk(KERN_WARNING
4573 "md: cannot register extra attributes for %s\n",
4574 mdname(mddev));
00bcb4ac 4575 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4576 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4577 mddev->ro = 0;
4578
1da177e4 4579 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4580 atomic_set(&mddev->max_corr_read_errors,
4581 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4582 mddev->safemode = 0;
4583 mddev->safemode_timer.function = md_safemode_timeout;
4584 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4585 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4586 mddev->in_sync = 1;
0ca69886
N		 4587 smp_wmb();
4588 mddev->ready = 1;
159ec1fc 4589 list_for_each_entry(rdev, &mddev->disks, same_set)
86e6ffdd
N		 4590 if (rdev->raid_disk >= 0) {
4591 char nm[20];
4592 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5 4593 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
00bcb4ac 4594 /* failure here is OK */;
86e6ffdd 4595 }
1da177e4
LT		 4596
4597 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4598
850b2b42
N		 4599 if (mddev->flags)
4600 md_update_sb(mddev, 0);
1da177e4 4601
0b8c9de0
N		 4602 md_wakeup_thread(mddev->thread);
4603 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5fd6c1dc 4604
d7603b7e 4605 md_new_event(mddev);
00bcb4ac
N		 4606 sysfs_notify_dirent_safe(mddev->sysfs_state);
4607 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4608 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4609 return 0;
4610}
390ee602 4611EXPORT_SYMBOL_GPL(md_run);
1da177e4 4612
fe60b014
N		 4613static int do_md_run(mddev_t *mddev)
4614{
4615 int err;
4616
4617 err = md_run(mddev);
4618 if (err)
4619 goto out;
69e51b44
N		 4620 err = bitmap_load(mddev);
4621 if (err) {
4622 bitmap_destroy(mddev);
4623 goto out;
4624 }
fe60b014
N		 4625 set_capacity(mddev->gendisk, mddev->array_sectors);
4626 revalidate_disk(mddev->gendisk);
4627 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4628out:
4629 return err;
4630}
4631
1da177e4
LT		 4632static int restart_array(mddev_t *mddev)
4633{
4634 struct gendisk *disk = mddev->gendisk;
1da177e4 4635
80fab1d7 4636 /* Complain if it has no devices */
1da177e4 4637 if (list_empty(&mddev->disks))
80fab1d7
AN		 4638 return -ENXIO;
4639 if (!mddev->pers)
4640 return -EINVAL;
4641 if (!mddev->ro)
4642 return -EBUSY;
4643 mddev->safemode = 0;
4644 mddev->ro = 0;
4645 set_disk_ro(disk, 0);
4646 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4647 mdname(mddev));
4648 /* Kick recovery or resync if necessary */
4649 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4650 md_wakeup_thread(mddev->thread);
4651 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4652 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4653 return 0;
1da177e4
LT		 4654}
4655
acc55e22
N		 4656/* similar to deny_write_access, but accounts for our holding a reference
4657 * to the file ourselves */
4658static int deny_bitmap_write_access(struct file * file)
4659{
4660 struct inode *inode = file->f_mapping->host;
4661
4662 spin_lock(&inode->i_lock);
4663 if (atomic_read(&inode->i_writecount) > 1) {
4664 spin_unlock(&inode->i_lock);
4665 return -ETXTBSY;
4666 }
4667 atomic_set(&inode->i_writecount, -1);
4668 spin_unlock(&inode->i_lock);
4669
4670 return 0;
4671}
4672
43a70507 4673void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4674{
4675 struct inode *inode = file->f_mapping->host;
4676
4677 spin_lock(&inode->i_lock);
4678 atomic_set(&inode->i_writecount, 1);
4679 spin_unlock(&inode->i_lock);
4680}
4681
6177b472
N		 4682static void md_clean(mddev_t *mddev)
4683{
4684 mddev->array_sectors = 0;
4685 mddev->external_size = 0;
4686 mddev->dev_sectors = 0;
4687 mddev->raid_disks = 0;
4688 mddev->recovery_cp = 0;
4689 mddev->resync_min = 0;
4690 mddev->resync_max = MaxSector;
4691 mddev->reshape_position = MaxSector;
4692 mddev->external = 0;
4693 mddev->persistent = 0;
4694 mddev->level = LEVEL_NONE;
4695 mddev->clevel[0] = 0;
4696 mddev->flags = 0;
4697 mddev->ro = 0;
4698 mddev->metadata_type[0] = 0;
4699 mddev->chunk_sectors = 0;
4700 mddev->ctime = mddev->utime = 0;
4701 mddev->layout = 0;
4702 mddev->max_disks = 0;
4703 mddev->events = 0;
a8707c08 4704 mddev->can_decrease_events = 0;
6177b472
N		 4705 mddev->delta_disks = 0;
4706 mddev->new_level = LEVEL_NONE;
4707 mddev->new_layout = 0;
4708 mddev->new_chunk_sectors = 0;
4709 mddev->curr_resync = 0;
4710 mddev->resync_mismatches = 0;
4711 mddev->suspend_lo = mddev->suspend_hi = 0;
4712 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4713 mddev->recovery = 0;
4714 mddev->in_sync = 0;
4715 mddev->degraded = 0;
6177b472
N		 4716 mddev->safemode = 0;
4717 mddev->bitmap_info.offset = 0;
4718 mddev->bitmap_info.default_offset = 0;
4719 mddev->bitmap_info.chunksize = 0;
4720 mddev->bitmap_info.daemon_sleep = 0;
4721 mddev->bitmap_info.max_write_behind = 0;
252ac522 4722 mddev->plug = NULL;
6177b472
N		 4723}
4724
defad61a 4725static void __md_stop_writes(mddev_t *mddev)
a047e125
N		 4726{
4727 if (mddev->sync_thread) {
4728 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4729 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4730 reap_sync_thread(mddev);
a047e125
N		 4731 }
4732
4733 del_timer_sync(&mddev->safemode_timer);
4734
4735 bitmap_flush(mddev);
4736 md_super_wait(mddev);
4737
4738 if (!mddev->in_sync || mddev->flags) {
4739 /* mark array as shutdown cleanly */
4740 mddev->in_sync = 1;
4741 md_update_sb(mddev, 1);
4742 }
4743}
defad61a
N		 4744
4745void md_stop_writes(mddev_t *mddev)
4746{
4747 mddev_lock(mddev);
4748 __md_stop_writes(mddev);
4749 mddev_unlock(mddev);
4750}
390ee602 4751EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 4752
390ee602 4753void md_stop(mddev_t *mddev)
6177b472 4754{
0ca69886 4755 mddev->ready = 0;
6177b472
N		 4756 mddev->pers->stop(mddev);
4757 if (mddev->pers->sync_request && mddev->to_remove == NULL)
4758 mddev->to_remove = &md_redundancy_group;
4759 module_put(mddev->pers->owner);
4760 mddev->pers = NULL;
cca9cf90 4761 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 4762}
390ee602 4763EXPORT_SYMBOL_GPL(md_stop);
6177b472 4764
a4bd82d0
N		 4765static int md_set_readonly(mddev_t *mddev, int is_open)
4766{
4767 int err = 0;
4768 mutex_lock(&mddev->open_mutex);
4769 if (atomic_read(&mddev->openers) > is_open) {
4770 printk("md: %s still in use.\n",mdname(mddev));
4771 err = -EBUSY;
4772 goto out;
4773 }
4774 if (mddev->pers) {
defad61a 4775 __md_stop_writes(mddev);
a4bd82d0
N		 4776
4777 err = -ENXIO;
4778 if (mddev->ro==1)
4779 goto out;
4780 mddev->ro = 1;
4781 set_disk_ro(mddev->gendisk, 1);
4782 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 4783 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 4784 err = 0;
4785 }
4786out:
4787 mutex_unlock(&mddev->open_mutex);
4788 return err;
4789}
4790
9e653b63
N		 4791/* mode:
4792 * 0 - completely stop and dis-assemble array
9e653b63
N		 4793 * 2 - stop but do not disassemble array
4794 */
df5b20cf 4795static int do_md_stop(mddev_t * mddev, int mode, int is_open)
1da177e4 4796{
1da177e4 4797 struct gendisk *disk = mddev->gendisk;
c4647292 4798 mdk_rdev_t *rdev;
1da177e4 4799
c8c00a69 4800 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 4801 if (atomic_read(&mddev->openers) > is_open ||
4802 mddev->sysfs_active) {
df5b20cf 4803 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 4804 mutex_unlock(&mddev->open_mutex);
4805 return -EBUSY;
4806 }
1da177e4 4807
6e17b027 4808 if (mddev->pers) {
a4bd82d0
N		 4809 if (mddev->ro)
4810 set_disk_ro(disk, 0);
409c57f3 4811
defad61a 4812 __md_stop_writes(mddev);
a4bd82d0
N		 4813 md_stop(mddev);
4814 mddev->queue->merge_bvec_fn = NULL;
4815 mddev->queue->unplug_fn = NULL;
4816 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 4817
a4bd82d0 4818 /* tell userspace to handle 'inactive' */
00bcb4ac 4819 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 4820
a4bd82d0
N		 4821 list_for_each_entry(rdev, &mddev->disks, same_set)
4822 if (rdev->raid_disk >= 0) {
4823 char nm[20];
4824 sprintf(nm, "rd%d", rdev->raid_disk);
4825 sysfs_remove_link(&mddev->kobj, nm);
4826 }
c4647292 4827
a4bd82d0 4828 set_capacity(disk, 0);
6e17b027 4829 mutex_unlock(&mddev->open_mutex);
a4bd82d0 4830 revalidate_disk(disk);
0d4ca600 4831
a4bd82d0
N		 4832 if (mddev->ro)
4833 mddev->ro = 0;
6e17b027
N		 4834 } else
4835 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 4836 /*
4837 * Free resources if final stop
4838 */
9e653b63 4839 if (mode == 0) {
1da177e4
LT		 4840 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4841
978f946b 4842 bitmap_destroy(mddev);
c3d9714e
N		 4843 if (mddev->bitmap_info.file) {
4844 restore_bitmap_write_access(mddev->bitmap_info.file);
4845 fput(mddev->bitmap_info.file);
4846 mddev->bitmap_info.file = NULL;
978f946b 4847 }
c3d9714e 4848 mddev->bitmap_info.offset = 0;
978f946b 4849
1da177e4
LT		 4850 export_array(mddev);
4851
6177b472 4852 md_clean(mddev);
934d9c23 4853 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 4854 if (mddev->hold_active == UNTIL_STOP)
4855 mddev->hold_active = 0;
a4bd82d0 4856 }
3f9d99c1 4857 blk_integrity_unregister(disk);
d7603b7e 4858 md_new_event(mddev);
00bcb4ac 4859 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 4860 return 0;
1da177e4
LT		 4861}
4862
fdee8ae4 4863#ifndef MODULE
1da177e4
LT		 4864static void autorun_array(mddev_t *mddev)
4865{
4866 mdk_rdev_t *rdev;
1da177e4
LT		 4867 int err;
4868
a757e64c 4869 if (list_empty(&mddev->disks))
1da177e4 4870 return;
1da177e4
LT		 4871
4872 printk(KERN_INFO "md: running: ");
4873
159ec1fc 4874 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 4875 char b[BDEVNAME_SIZE];
4876 printk("<%s>", bdevname(rdev->bdev,b));
4877 }
4878 printk("\n");
4879
d710e138 4880 err = do_md_run(mddev);
1da177e4
LT		 4881 if (err) {
4882 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 4883 do_md_stop(mddev, 0, 0);
1da177e4
LT		 4884 }
4885}
4886
4887/*
4888 * lets try to run arrays based on all disks that have arrived
4889 * until now. (those are in pending_raid_disks)
4890 *
4891 * the method: pick the first pending disk, collect all disks with
4892 * the same UUID, remove all from the pending list and put them into
4893 * the 'same_array' list. Then order this list based on superblock
4894 * update time (freshest comes first), kick out 'old' disks and
4895 * compare superblocks. If everything's fine then run it.
4896 *
4897 * If "unit" is allocated, then bump its reference count
4898 */
4899static void autorun_devices(int part)
4900{
159ec1fc 4901 mdk_rdev_t *rdev0, *rdev, *tmp;
1da177e4
LT		 4902 mddev_t *mddev;
4903 char b[BDEVNAME_SIZE];
4904
4905 printk(KERN_INFO "md: autorun ...\n");
4906 while (!list_empty(&pending_raid_disks)) {
e8703fe1 4907 int unit;
1da177e4 4908 dev_t dev;
ad01c9e3 4909 LIST_HEAD(candidates);
1da177e4
LT		 4910 rdev0 = list_entry(pending_raid_disks.next,
4911 mdk_rdev_t, same_set);
4912
4913 printk(KERN_INFO "md: considering %s ...\n",
4914 bdevname(rdev0->bdev,b));
4915 INIT_LIST_HEAD(&candidates);
159ec1fc 4916 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 4917 if (super_90_load(rdev, rdev0, 0) >= 0) {
4918 printk(KERN_INFO "md: adding %s ...\n",
4919 bdevname(rdev->bdev,b));
4920 list_move(&rdev->same_set, &candidates);
4921 }
4922 /*
4923 * now we have a set of devices, with all of them having
4924 * mostly sane superblocks. It's time to allocate the
4925 * mddev.
4926 */
e8703fe1
N		 4927 if (part) {
4928 dev = MKDEV(mdp_major,
4929 rdev0->preferred_minor << MdpMinorShift);
4930 unit = MINOR(dev) >> MdpMinorShift;
4931 } else {
4932 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4933 unit = MINOR(dev);
4934 }
4935 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 4936 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4937 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4938 break;
4939 }
1da177e4
LT		 4940
4941 md_probe(dev, NULL, NULL);
4942 mddev = mddev_find(dev);
9bbbca3a
NB		 4943 if (!mddev || !mddev->gendisk) {
4944 if (mddev)
4945 mddev_put(mddev);
4946 printk(KERN_ERR
1da177e4
LT		 4947 "md: cannot allocate memory for md drive.\n");
4948 break;
4949 }
4950 if (mddev_lock(mddev))
4951 printk(KERN_WARNING "md: %s locked, cannot run\n",
4952 mdname(mddev));
4953 else if (mddev->raid_disks || mddev->major_version
4954 || !list_empty(&mddev->disks)) {
4955 printk(KERN_WARNING
4956 "md: %s already running, cannot run %s\n",
4957 mdname(mddev), bdevname(rdev0->bdev,b));
4958 mddev_unlock(mddev);
4959 } else {
4960 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 4961 mddev->persistent = 1;
159ec1fc 4962 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 4963 list_del_init(&rdev->same_set);
4964 if (bind_rdev_to_array(rdev, mddev))
4965 export_rdev(rdev);
4966 }
4967 autorun_array(mddev);
4968 mddev_unlock(mddev);
4969 }
4970 /* on success, candidates will be empty, on error
4971 * it won't...
4972 */
159ec1fc 4973 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 4974 list_del_init(&rdev->same_set);
1da177e4 4975 export_rdev(rdev);
4b80991c 4976 }
1da177e4
LT		 4977 mddev_put(mddev);
4978 }
4979 printk(KERN_INFO "md: ... autorun DONE.\n");
4980}
fdee8ae4 4981#endif /* !MODULE */
1da177e4 4982
1da177e4
LT		 4983static int get_version(void __user * arg)
4984{
4985 mdu_version_t ver;
4986
4987 ver.major = MD_MAJOR_VERSION;
4988 ver.minor = MD_MINOR_VERSION;
4989 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4990
4991 if (copy_to_user(arg, &ver, sizeof(ver)))
4992 return -EFAULT;
4993
4994 return 0;
4995}
4996
4997static int get_array_info(mddev_t * mddev, void __user * arg)
4998{
4999 mdu_array_info_t info;
a9f326eb 5000 int nr,working,insync,failed,spare;
1da177e4 5001 mdk_rdev_t *rdev;
1da177e4 5002
a9f326eb 5003 nr=working=insync=failed=spare=0;
159ec1fc 5004 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5005 nr++;
b2d444d7 5006 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5007 failed++;
5008 else {
5009 working++;
b2d444d7 5010 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5011 insync++;
1da177e4
LT		 5012 else
5013 spare++;
5014 }
5015 }
5016
5017 info.major_version = mddev->major_version;
5018 info.minor_version = mddev->minor_version;
5019 info.patch_version = MD_PATCHLEVEL_VERSION;
5020 info.ctime = mddev->ctime;
5021 info.level = mddev->level;
58c0fed4
AN		 5022 info.size = mddev->dev_sectors / 2;
5023 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5024 info.size = -1;
1da177e4
LT		 5025 info.nr_disks = nr;
5026 info.raid_disks = mddev->raid_disks;
5027 info.md_minor = mddev->md_minor;
5028 info.not_persistent= !mddev->persistent;
5029
5030 info.utime = mddev->utime;
5031 info.state = 0;
5032 if (mddev->in_sync)
5033 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5034 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5035 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5036 info.active_disks = insync;
1da177e4
LT		 5037 info.working_disks = working;
5038 info.failed_disks = failed;
5039 info.spare_disks = spare;
5040
5041 info.layout = mddev->layout;
9d8f0363 5042 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5043
5044 if (copy_to_user(arg, &info, sizeof(info)))
5045 return -EFAULT;
5046
5047 return 0;
5048}
5049
87162a28 5050static int get_bitmap_file(mddev_t * mddev, void __user * arg)
32a7627c
N		 5051{
5052 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5053 char *ptr, *buf = NULL;
5054 int err = -ENOMEM;
5055
b5470dc5
DW		 5056 if (md_allow_write(mddev))
5057 file = kmalloc(sizeof(*file), GFP_NOIO);
5058 else
5059 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5060
32a7627c
N		 5061 if (!file)
5062 goto out;
5063
5064 /* bitmap disabled, zero the first byte and copy out */
5065 if (!mddev->bitmap || !mddev->bitmap->file) {
5066 file->pathname[0] = '\0';
5067 goto copy_out;
5068 }
5069
5070 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5071 if (!buf)
5072 goto out;
5073
6bcfd601
CH		 5074 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5075 if (IS_ERR(ptr))
32a7627c
N		 5076 goto out;
5077
5078 strcpy(file->pathname, ptr);
5079
5080copy_out:
5081 err = 0;
5082 if (copy_to_user(arg, file, sizeof(*file)))
5083 err = -EFAULT;
5084out:
5085 kfree(buf);
5086 kfree(file);
5087 return err;
5088}
5089
1da177e4
LT		 5090static int get_disk_info(mddev_t * mddev, void __user * arg)
5091{
5092 mdu_disk_info_t info;
1da177e4
LT		 5093 mdk_rdev_t *rdev;
5094
5095 if (copy_from_user(&info, arg, sizeof(info)))
5096 return -EFAULT;
5097
26ef379f 5098 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5099 if (rdev) {
5100 info.major = MAJOR(rdev->bdev->bd_dev);
5101 info.minor = MINOR(rdev->bdev->bd_dev);
5102 info.raid_disk = rdev->raid_disk;
5103 info.state = 0;
b2d444d7 5104 if (test_bit(Faulty, &rdev->flags))
1da177e4 5105 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5106 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5107 info.state |= (1<<MD_DISK_ACTIVE);
5108 info.state |= (1<<MD_DISK_SYNC);
5109 }
8ddf9efe
N		 5110 if (test_bit(WriteMostly, &rdev->flags))
5111 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5112 } else {
5113 info.major = info.minor = 0;
5114 info.raid_disk = -1;
5115 info.state = (1<<MD_DISK_REMOVED);
5116 }
5117
5118 if (copy_to_user(arg, &info, sizeof(info)))
5119 return -EFAULT;
5120
5121 return 0;
5122}
5123
5124static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5125{
5126 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5127 mdk_rdev_t *rdev;
5128 dev_t dev = MKDEV(info->major,info->minor);
5129
5130 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5131 return -EOVERFLOW;
5132
5133 if (!mddev->raid_disks) {
5134 int err;
5135 /* expecting a device which has a superblock */
5136 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5137 if (IS_ERR(rdev)) {
5138 printk(KERN_WARNING
5139 "md: md_import_device returned %ld\n",
5140 PTR_ERR(rdev));
5141 return PTR_ERR(rdev);
5142 }
5143 if (!list_empty(&mddev->disks)) {
5144 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5145 mdk_rdev_t, same_set);
a9f326eb 5146 err = super_types[mddev->major_version]
1da177e4
LT		 5147 .load_super(rdev, rdev0, mddev->minor_version);
5148 if (err < 0) {
5149 printk(KERN_WARNING
5150 "md: %s has different UUID to %s\n",
5151 bdevname(rdev->bdev,b),
5152 bdevname(rdev0->bdev,b2));
5153 export_rdev(rdev);
5154 return -EINVAL;
5155 }
5156 }
5157 err = bind_rdev_to_array(rdev, mddev);
5158 if (err)
5159 export_rdev(rdev);
5160 return err;
5161 }
5162
5163 /*
5164 * add_new_disk can be used once the array is assembled
5165 * to add "hot spares". They must already have a superblock
5166 * written
5167 */
5168 if (mddev->pers) {
5169 int err;
5170 if (!mddev->pers->hot_add_disk) {
5171 printk(KERN_WARNING
5172 "%s: personality does not support diskops!\n",
5173 mdname(mddev));
5174 return -EINVAL;
5175 }
7b1e35f6
N		 5176 if (mddev->persistent)
5177 rdev = md_import_device(dev, mddev->major_version,
5178 mddev->minor_version);
5179 else
5180 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5181 if (IS_ERR(rdev)) {
5182 printk(KERN_WARNING
5183 "md: md_import_device returned %ld\n",
5184 PTR_ERR(rdev));
5185 return PTR_ERR(rdev);
5186 }
1a855a06 5187 /* set saved_raid_disk if appropriate */
41158c7e
N		 5188 if (!mddev->persistent) {
5189 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5190 info->raid_disk < mddev->raid_disks) {
41158c7e 5191 rdev->raid_disk = info->raid_disk;
bf572541
N		 5192 set_bit(In_sync, &rdev->flags);
5193 } else
41158c7e
N		 5194 rdev->raid_disk = -1;
5195 } else
5196 super_types[mddev->major_version].
5197 validate_super(mddev, rdev);
1a855a06
N		 5198 if (test_bit(In_sync, &rdev->flags))
5199 rdev->saved_raid_disk = rdev->raid_disk;
5200 else
5201 rdev->saved_raid_disk = -1;
41158c7e 5202
b2d444d7 5203 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5204 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5205 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5206 else
5207 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5208
1da177e4
LT		 5209 rdev->raid_disk = -1;
5210 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5211 if (!err && !mddev->pers->hot_remove_disk) {
5212 /* If there is hot_add_disk but no hot_remove_disk
5213 * then added disks for geometry changes,
5214 * and should be added immediately.
5215 */
5216 super_types[mddev->major_version].
5217 validate_super(mddev, rdev);
5218 err = mddev->pers->hot_add_disk(mddev, rdev);
5219 if (err)
5220 unbind_rdev_from_array(rdev);
5221 }
1da177e4
LT		 5222 if (err)
5223 export_rdev(rdev);
52664732 5224 else
00bcb4ac 5225 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5226
17571284 5227 md_update_sb(mddev, 1);
72a23c21
NB		 5228 if (mddev->degraded)
5229 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5230 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
005eca5e 5231 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5232 return err;
5233 }
5234
5235 /* otherwise, add_new_disk is only allowed
5236 * for major_version==0 superblocks
5237 */
5238 if (mddev->major_version != 0) {
5239 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5240 mdname(mddev));
5241 return -EINVAL;
5242 }
5243
5244 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5245 int err;
d710e138 5246 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5247 if (IS_ERR(rdev)) {
5248 printk(KERN_WARNING
5249 "md: error, md_import_device() returned %ld\n",
5250 PTR_ERR(rdev));
5251 return PTR_ERR(rdev);
5252 }
5253 rdev->desc_nr = info->number;
5254 if (info->raid_disk < mddev->raid_disks)
5255 rdev->raid_disk = info->raid_disk;
5256 else
5257 rdev->raid_disk = -1;
5258
1da177e4 5259 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5260 if (info->state & (1<<MD_DISK_SYNC))
5261 set_bit(In_sync, &rdev->flags);
1da177e4 5262
8ddf9efe
N		 5263 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5264 set_bit(WriteMostly, &rdev->flags);
5265
1da177e4
LT		 5266 if (!mddev->persistent) {
5267 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5268 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5269 } else
57b2caa3 5270 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5271 rdev->sectors = rdev->sb_start;
1da177e4 5272
2bf071bf
N		 5273 err = bind_rdev_to_array(rdev, mddev);
5274 if (err) {
5275 export_rdev(rdev);
5276 return err;
5277 }
1da177e4
LT		 5278 }
5279
5280 return 0;
5281}
5282
5283static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5284{
5285 char b[BDEVNAME_SIZE];
5286 mdk_rdev_t *rdev;
5287
1da177e4
LT		 5288 rdev = find_rdev(mddev, dev);
5289 if (!rdev)
5290 return -ENXIO;
5291
5292 if (rdev->raid_disk >= 0)
5293 goto busy;
5294
5295 kick_rdev_from_array(rdev);
850b2b42 5296 md_update_sb(mddev, 1);
d7603b7e 5297 md_new_event(mddev);
1da177e4
LT		 5298
5299 return 0;
5300busy:
fdefa4d8 5301 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5302 bdevname(rdev->bdev,b), mdname(mddev));
5303 return -EBUSY;
5304}
5305
5306static int hot_add_disk(mddev_t * mddev, dev_t dev)
5307{
5308 char b[BDEVNAME_SIZE];
5309 int err;
1da177e4
LT		 5310 mdk_rdev_t *rdev;
5311
5312 if (!mddev->pers)
5313 return -ENODEV;
5314
5315 if (mddev->major_version != 0) {
5316 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5317 " version-0 superblocks.\n",
5318 mdname(mddev));
5319 return -EINVAL;
5320 }
5321 if (!mddev->pers->hot_add_disk) {
5322 printk(KERN_WARNING
5323 "%s: personality does not support diskops!\n",
5324 mdname(mddev));
5325 return -EINVAL;
5326 }
5327
d710e138 5328 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5329 if (IS_ERR(rdev)) {
5330 printk(KERN_WARNING
5331 "md: error, md_import_device() returned %ld\n",
5332 PTR_ERR(rdev));
5333 return -EINVAL;
5334 }
5335
5336 if (mddev->persistent)
57b2caa3 5337 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5338 else
77304d2a 5339 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5340
8190e754 5341 rdev->sectors = rdev->sb_start;
1da177e4 5342
b2d444d7 5343 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5344 printk(KERN_WARNING
5345 "md: can not hot-add faulty %s disk to %s!\n",
5346 bdevname(rdev->bdev,b), mdname(mddev));
5347 err = -EINVAL;
5348 goto abort_export;
5349 }
b2d444d7 5350 clear_bit(In_sync, &rdev->flags);
1da177e4 5351 rdev->desc_nr = -1;
5842730d 5352 rdev->saved_raid_disk = -1;
2bf071bf
N		 5353 err = bind_rdev_to_array(rdev, mddev);
5354 if (err)
5355 goto abort_export;
1da177e4
LT		 5356
5357 /*
5358 * The rest should better be atomic, we can have disk failures
5359 * noticed in interrupt contexts ...
5360 */
5361
1da177e4
LT		 5362 rdev->raid_disk = -1;
5363
850b2b42 5364 md_update_sb(mddev, 1);
1da177e4
LT		 5365
5366 /*
5367 * Kick recovery, maybe this spare has to be added to the
5368 * array immediately.
5369 */
5370 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5371 md_wakeup_thread(mddev->thread);
d7603b7e 5372 md_new_event(mddev);
1da177e4
LT		 5373 return 0;
5374
1da177e4
LT		 5375abort_export:
5376 export_rdev(rdev);
5377 return err;
5378}
5379
32a7627c
N		 5380static int set_bitmap_file(mddev_t *mddev, int fd)
5381{
5382 int err;
5383
36fa3063
N		 5384 if (mddev->pers) {
5385 if (!mddev->pers->quiesce)
5386 return -EBUSY;
5387 if (mddev->recovery || mddev->sync_thread)
5388 return -EBUSY;
5389 /* we should be able to change the bitmap.. */
5390 }
32a7627c 5391
32a7627c 5392
36fa3063
N		 5393 if (fd >= 0) {
5394 if (mddev->bitmap)
5395 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5396 mddev->bitmap_info.file = fget(fd);
32a7627c 5397
c3d9714e 5398 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5399 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5400 mdname(mddev));
5401 return -EBADF;
5402 }
5403
c3d9714e 5404 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5405 if (err) {
5406 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5407 mdname(mddev));
c3d9714e
N		 5408 fput(mddev->bitmap_info.file);
5409 mddev->bitmap_info.file = NULL;
36fa3063
N		 5410 return err;
5411 }
c3d9714e 5412 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5413 } else if (mddev->bitmap == NULL)
5414 return -ENOENT; /* cannot remove what isn't there */
5415 err = 0;
5416 if (mddev->pers) {
5417 mddev->pers->quiesce(mddev, 1);
69e51b44 5418 if (fd >= 0) {
36fa3063 5419 err = bitmap_create(mddev);
69e51b44
N		 5420 if (!err)
5421 err = bitmap_load(mddev);
5422 }
d7375ab3 5423 if (fd < 0 || err) {
36fa3063 5424 bitmap_destroy(mddev);
d7375ab3
N		 5425 fd = -1; /* make sure to put the file */
5426 }
36fa3063 5427 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5428 }
5429 if (fd < 0) {
c3d9714e
N		 5430 if (mddev->bitmap_info.file) {
5431 restore_bitmap_write_access(mddev->bitmap_info.file);
5432 fput(mddev->bitmap_info.file);
acc55e22 5433 }
c3d9714e 5434 mddev->bitmap_info.file = NULL;
36fa3063
N		 5435 }
5436
32a7627c
N		 5437 return err;
5438}
5439
1da177e4
LT		 5440/*
5441 * set_array_info is used two different ways
5442 * The original usage is when creating a new array.
5443 * In this usage, raid_disks is > 0 and it together with
5444 * level, size, not_persistent,layout,chunksize determine the
5445 * shape of the array.
5446 * This will always create an array with a type-0.90.0 superblock.
5447 * The newer usage is when assembling an array.
5448 * In this case raid_disks will be 0, and the major_version field is
5449 * use to determine which style super-blocks are to be found on the devices.
5450 * The minor and patch _version numbers are also kept incase the
5451 * super_block handler wishes to interpret them.
5452 */
5453static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5454{
5455
5456 if (info->raid_disks == 0) {
5457 /* just setting version number for superblock loading */
5458 if (info->major_version < 0 ||
50511da3 5459 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5460 super_types[info->major_version].name == NULL) {
5461 /* maybe try to auto-load a module? */
5462 printk(KERN_INFO
5463 "md: superblock version %d not known\n",
5464 info->major_version);
5465 return -EINVAL;
5466 }
5467 mddev->major_version = info->major_version;
5468 mddev->minor_version = info->minor_version;
5469 mddev->patch_version = info->patch_version;
3f9d7b0d 5470 mddev->persistent = !info->not_persistent;
cbd19983
N		 5471 /* ensure mddev_put doesn't delete this now that there
5472 * is some minimal configuration.
5473 */
5474 mddev->ctime = get_seconds();
1da177e4
LT		 5475 return 0;
5476 }
5477 mddev->major_version = MD_MAJOR_VERSION;
5478 mddev->minor_version = MD_MINOR_VERSION;
5479 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5480 mddev->ctime = get_seconds();
5481
5482 mddev->level = info->level;
17115e03 5483 mddev->clevel[0] = 0;
58c0fed4 5484 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5485 mddev->raid_disks = info->raid_disks;
5486 /* don't set md_minor, it is determined by which /dev/md* was
5487 * openned
5488 */
5489 if (info->state & (1<<MD_SB_CLEAN))
5490 mddev->recovery_cp = MaxSector;
5491 else
5492 mddev->recovery_cp = 0;
5493 mddev->persistent = ! info->not_persistent;
e691063a 5494 mddev->external = 0;
1da177e4
LT		 5495
5496 mddev->layout = info->layout;
9d8f0363 5497 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5498
5499 mddev->max_disks = MD_SB_DISKS;
5500
e691063a
N		 5501 if (mddev->persistent)
5502 mddev->flags = 0;
850b2b42 5503 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5504
c3d9714e
N		 5505 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5506 mddev->bitmap_info.offset = 0;
b2a2703c 5507
f6705578
N		 5508 mddev->reshape_position = MaxSector;
5509
1da177e4
LT		 5510 /*
5511 * Generate a 128 bit UUID
5512 */
5513 get_random_bytes(mddev->uuid, 16);
5514
f6705578 5515 mddev->new_level = mddev->level;
664e7c41 5516 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5517 mddev->new_layout = mddev->layout;
5518 mddev->delta_disks = 0;
5519
1da177e4
LT		 5520 return 0;
5521}
5522
1f403624
DW		 5523void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5524{
b522adcd
DW		 5525 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5526
5527 if (mddev->external_size)
5528 return;
5529
1f403624
DW		 5530 mddev->array_sectors = array_sectors;
5531}
5532EXPORT_SYMBOL(md_set_array_sectors);
5533
d71f9f88 5534static int update_size(mddev_t *mddev, sector_t num_sectors)
a35b0d69 5535{
159ec1fc 5536 mdk_rdev_t *rdev;
a35b0d69 5537 int rv;
d71f9f88 5538 int fit = (num_sectors == 0);
a35b0d69
N		 5539
5540 if (mddev->pers->resize == NULL)
5541 return -EINVAL;
d71f9f88
AN		 5542 /* The "num_sectors" is the number of sectors of each device that
5543 * is used. This can only make sense for arrays with redundancy.
5544 * linear and raid0 always use whatever space is available. We can only
5545 * consider changing this number if no resync or reconstruction is
5546 * happening, and if the new size is acceptable. It must fit before the
0f420358 5547 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5548 * of each device. If num_sectors is zero, we find the largest size
5549 * that fits.
a35b0d69
N		 5550 */
5551 if (mddev->sync_thread)
5552 return -EBUSY;
dba034ee
N		 5553 if (mddev->bitmap)
5554 /* Sorry, cannot grow a bitmap yet, just remove it,
5555 * grow, and re-add.
5556 */
5557 return -EBUSY;
159ec1fc 5558 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5559 sector_t avail = rdev->sectors;
01ab5662 5560
d71f9f88
AN		 5561 if (fit && (num_sectors == 0 || num_sectors > avail))
5562 num_sectors = avail;
5563 if (avail < num_sectors)
a35b0d69
N		 5564 return -ENOSPC;
5565 }
d71f9f88 5566 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5567 if (!rv)
5568 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5569 return rv;
5570}
5571
da943b99
N		 5572static int update_raid_disks(mddev_t *mddev, int raid_disks)
5573{
5574 int rv;
5575 /* change the number of raid disks */
63c70c4f 5576 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5577 return -EINVAL;
5578 if (raid_disks <= 0 ||
233fca36 5579 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5580 return -EINVAL;
63c70c4f 5581 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5582 return -EBUSY;
63c70c4f
N		 5583 mddev->delta_disks = raid_disks - mddev->raid_disks;
5584
5585 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5586 if (rv < 0)
5587 mddev->delta_disks = 0;
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
LT		 6994 mddev->curr_resync = 0;
6995 wake_up(&resync_wait);
6996 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6997 md_wakeup_thread(mddev->thread);
c6207277
N		 6998 return;
6999
7000 interrupted:
7001 /*
7002 * got a signal, exit.
7003 */
7004 printk(KERN_INFO
7005 "md: md_do_sync() got signal ... exiting\n");
7006 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7007 goto out;
7008
1da177e4 7009}
29269553 7010EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4
LT		 7011
7012
b4c4c7b8
N		 7013static int remove_and_add_spares(mddev_t *mddev)
7014{
7015 mdk_rdev_t *rdev;
b4c4c7b8
N		 7016 int spares = 0;
7017
97e4f42d
N		 7018 mddev->curr_resync_completed = 0;
7019
159ec1fc 7020 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7021 if (rdev->raid_disk >= 0 &&
6bfe0b49 7022 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7023 (test_bit(Faulty, &rdev->flags) ||
7024 ! test_bit(In_sync, &rdev->flags)) &&
7025 atomic_read(&rdev->nr_pending)==0) {
7026 if (mddev->pers->hot_remove_disk(
7027 mddev, rdev->raid_disk)==0) {
7028 char nm[20];
7029 sprintf(nm,"rd%d", rdev->raid_disk);
7030 sysfs_remove_link(&mddev->kobj, nm);
7031 rdev->raid_disk = -1;
7032 }
7033 }
7034
a8c42c7f 7035 if (mddev->degraded && !mddev->recovery_disabled) {
159ec1fc 7036 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7037 if (rdev->raid_disk >= 0 &&
e5427135
DW		 7038 !test_bit(In_sync, &rdev->flags) &&
7039 !test_bit(Blocked, &rdev->flags))
dfc70645 7040 spares++;
b4c4c7b8
N		 7041 if (rdev->raid_disk < 0
7042 && !test_bit(Faulty, &rdev->flags)) {
7043 rdev->recovery_offset = 0;
199050ea
NB		 7044 if (mddev->pers->
7045 hot_add_disk(mddev, rdev) == 0) {
b4c4c7b8
N		 7046 char nm[20];
7047 sprintf(nm, "rd%d", rdev->raid_disk);
5e55e2f5
N		 7048 if (sysfs_create_link(&mddev->kobj,
7049 &rdev->kobj, nm))
00bcb4ac 7050 /* failure here is OK */;
b4c4c7b8
N		 7051 spares++;
7052 md_new_event(mddev);
93be75ff 7053 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7054 } else
7055 break;
7056 }
dfc70645 7057 }
b4c4c7b8
N		 7058 }
7059 return spares;
7060}
7ebc0be7
N		 7061
7062static void reap_sync_thread(mddev_t *mddev)
7063{
7064 mdk_rdev_t *rdev;
7065
7066 /* resync has finished, collect result */
7067 md_unregister_thread(mddev->sync_thread);
7068 mddev->sync_thread = NULL;
7069 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7070 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7071 /* success...*/
7072 /* activate any spares */
7073 if (mddev->pers->spare_active(mddev))
7074 sysfs_notify(&mddev->kobj, NULL,
7075 "degraded");
7076 }
7077 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7078 mddev->pers->finish_reshape)
7079 mddev->pers->finish_reshape(mddev);
7080 md_update_sb(mddev, 1);
7081
7082 /* if array is no-longer degraded, then any saved_raid_disk
7083 * information must be scrapped
7084 */
7085 if (!mddev->degraded)
7086 list_for_each_entry(rdev, &mddev->disks, same_set)
7087 rdev->saved_raid_disk = -1;
7088
7089 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7090 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7091 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7092 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7093 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7094 /* flag recovery needed just to double check */
7095 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7096 sysfs_notify_dirent_safe(mddev->sysfs_action);
7097 md_new_event(mddev);
7098}
7099
1da177e4
LT		 7100/*
7101 * This routine is regularly called by all per-raid-array threads to
7102 * deal with generic issues like resync and super-block update.
7103 * Raid personalities that don't have a thread (linear/raid0) do not
7104 * need this as they never do any recovery or update the superblock.
7105 *
7106 * It does not do any resync itself, but rather "forks" off other threads
7107 * to do that as needed.
7108 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7109 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7110 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7111 * and wakeups up this thread which will reap the thread and finish up.
7112 * This thread also removes any faulty devices (with nr_pending == 0).
7113 *
7114 * The overall approach is:
7115 * 1/ if the superblock needs updating, update it.
7116 * 2/ If a recovery thread is running, don't do anything else.
7117 * 3/ If recovery has finished, clean up, possibly marking spares active.
7118 * 4/ If there are any faulty devices, remove them.
7119 * 5/ If array is degraded, try to add spares devices
7120 * 6/ If array has spares or is not in-sync, start a resync thread.
7121 */
7122void md_check_recovery(mddev_t *mddev)
7123{
5f40402d 7124 if (mddev->bitmap)
aa5cbd10 7125 bitmap_daemon_work(mddev);
1da177e4
LT		 7126
7127 if (mddev->ro)
7128 return;
fca4d848
N		 7129
7130 if (signal_pending(current)) {
31a59e34 7131 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7132 printk(KERN_INFO "md: %s in immediate safe mode\n",
7133 mdname(mddev));
7134 mddev->safemode = 2;
7135 }
7136 flush_signals(current);
7137 }
7138
c89a8eee
N		 7139 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7140 return;
1da177e4 7141 if ( ! (
126925c0 7142 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7143 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7144 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7145 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7146 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7147 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7148 ))
7149 return;
fca4d848 7150
df5b89b3 7151 if (mddev_trylock(mddev)) {
b4c4c7b8 7152 int spares = 0;
fca4d848 7153
c89a8eee
N		 7154 if (mddev->ro) {
7155 /* Only thing we do on a ro array is remove
7156 * failed devices.
7157 */
a8c42c7f
N		 7158 mdk_rdev_t *rdev;
7159 list_for_each_entry(rdev, &mddev->disks, same_set)
7160 if (rdev->raid_disk >= 0 &&
7161 !test_bit(Blocked, &rdev->flags) &&
7162 test_bit(Faulty, &rdev->flags) &&
7163 atomic_read(&rdev->nr_pending)==0) {
7164 if (mddev->pers->hot_remove_disk(
7165 mddev, rdev->raid_disk)==0) {
7166 char nm[20];
7167 sprintf(nm,"rd%d", rdev->raid_disk);
7168 sysfs_remove_link(&mddev->kobj, nm);
7169 rdev->raid_disk = -1;
7170 }
7171 }
c89a8eee
N		 7172 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7173 goto unlock;
7174 }
7175
31a59e34 7176 if (!mddev->external) {
0fd62b86 7177 int did_change = 0;
31a59e34
N		 7178 spin_lock_irq(&mddev->write_lock);
7179 if (mddev->safemode &&
7180 !atomic_read(&mddev->writes_pending) &&
7181 !mddev->in_sync &&
7182 mddev->recovery_cp == MaxSector) {
7183 mddev->in_sync = 1;
0fd62b86 7184 did_change = 1;
070dc6dd 7185 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7186 }
7187 if (mddev->safemode == 1)
7188 mddev->safemode = 0;
7189 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7190 if (did_change)
00bcb4ac 7191 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7192 }
fca4d848 7193
850b2b42
N		 7194 if (mddev->flags)
7195 md_update_sb(mddev, 0);
06d91a5f 7196
1da177e4
LT		 7197 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7198 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7199 /* resync/recovery still happening */
7200 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7201 goto unlock;
7202 }
7203 if (mddev->sync_thread) {
7ebc0be7 7204 reap_sync_thread(mddev);
1da177e4
LT		 7205 goto unlock;
7206 }
72a23c21
NB		 7207 /* Set RUNNING before clearing NEEDED to avoid
7208 * any transients in the value of "sync_action".
7209 */
7210 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7211 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7212 /* Clear some bits that don't mean anything, but
7213 * might be left set
7214 */
24dd469d
N		 7215 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7216 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7217
5fd6c1dc
N		 7218 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7219 goto unlock;
1da177e4
LT		 7220 /* no recovery is running.
7221 * remove any failed drives, then
7222 * add spares if possible.
7223 * Spare are also removed and re-added, to allow
7224 * the personality to fail the re-add.
7225 */
1da177e4 7226
b4c4c7b8 7227 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7228 if (mddev->pers->check_reshape == NULL ||
7229 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7230 /* Cannot proceed */
7231 goto unlock;
7232 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7233 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7234 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7235 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7236 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7237 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7238 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7239 } else if (mddev->recovery_cp < MaxSector) {
7240 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7241 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7242 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7243 /* nothing to be done ... */
1da177e4 7244 goto unlock;
24dd469d 7245
1da177e4 7246 if (mddev->pers->sync_request) {
a654b9d8
N		 7247 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7248 /* We are adding a device or devices to an array
7249 * which has the bitmap stored on all devices.
7250 * So make sure all bitmap pages get written
7251 */
7252 bitmap_write_all(mddev->bitmap);
7253 }
1da177e4
LT		 7254 mddev->sync_thread = md_register_thread(md_do_sync,
7255 mddev,
0da3c619 7256 "resync");
1da177e4
LT		 7257 if (!mddev->sync_thread) {
7258 printk(KERN_ERR "%s: could not start resync"
7259 " thread...\n",
7260 mdname(mddev));
7261 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7262 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7263 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7264 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7265 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7266 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7267 } else
1da177e4 7268 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7269 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7270 md_new_event(mddev);
1da177e4
LT		 7271 }
7272 unlock:
72a23c21
NB		 7273 if (!mddev->sync_thread) {
7274 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7275 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7276 &mddev->recovery))
0c3573f1 7277 if (mddev->sysfs_action)
00bcb4ac 7278 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7279 }
1da177e4
LT		 7280 mddev_unlock(mddev);
7281 }
7282}
7283
6bfe0b49
DW		 7284void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7285{
00bcb4ac 7286 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49
DW		 7287 wait_event_timeout(rdev->blocked_wait,
7288 !test_bit(Blocked, &rdev->flags),
7289 msecs_to_jiffies(5000));
7290 rdev_dec_pending(rdev, mddev);
7291}
7292EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7293
75c96f85
AB		 7294static int md_notify_reboot(struct notifier_block *this,
7295 unsigned long code, void *x)
1da177e4
LT		 7296{
7297 struct list_head *tmp;
7298 mddev_t *mddev;
7299
7300 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
7301
7302 printk(KERN_INFO "md: stopping all md devices.\n");
7303
29ac4aa3 7304 for_each_mddev(mddev, tmp)
c71d4887 7305 if (mddev_trylock(mddev)) {
2b25000b
N		 7306 /* Force a switch to readonly even array
7307 * appears to still be in use. Hence
7308 * the '100'.
7309 */
a4bd82d0 7310 md_set_readonly(mddev, 100);
c71d4887
NB		 7311 mddev_unlock(mddev);
7312 }
1da177e4
LT		 7313 /*
7314 * certain more exotic SCSI devices are known to be
7315 * volatile wrt too early system reboots. While the
7316 * right place to handle this issue is the given
7317 * driver, we do want to have a safe RAID driver ...
7318 */
7319 mdelay(1000*1);
7320 }
7321 return NOTIFY_DONE;
7322}
7323
75c96f85 7324static struct notifier_block md_notifier = {
1da177e4
LT		 7325 .notifier_call = md_notify_reboot,
7326 .next = NULL,
7327 .priority = INT_MAX, /* before any real devices */
7328};
7329
7330static void md_geninit(void)
7331{
1da177e4
LT		 7332 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
7333
c7705f34 7334 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 7335}
7336
75c96f85 7337static int __init md_init(void)
1da177e4 7338{
e804ac78
TH		 7339 int ret = -ENOMEM;
7340
7341 md_wq = alloc_workqueue("md", WQ_RESCUER, 0);
7342 if (!md_wq)
7343 goto err_wq;
7344
7345 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
7346 if (!md_misc_wq)
7347 goto err_misc_wq;
7348
7349 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
7350 goto err_md;
7351
7352 if ((ret = register_blkdev(0, "mdp")) < 0)
7353 goto err_mdp;
7354 mdp_major = ret;
7355
3dbd8c2e 7356 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 7357 md_probe, NULL, NULL);
7358 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 7359 md_probe, NULL, NULL);
7360
1da177e4 7361 register_reboot_notifier(&md_notifier);
0b4d4147 7362 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 7363
7364 md_geninit();
d710e138 7365 return 0;
1da177e4 7366
e804ac78
TH		 7367err_mdp:
7368 unregister_blkdev(MD_MAJOR, "md");
7369err_md:
7370 destroy_workqueue(md_misc_wq);
7371err_misc_wq:
7372 destroy_workqueue(md_wq);
7373err_wq:
7374 return ret;
7375}
1da177e4
LT		 7376
7377#ifndef MODULE
7378
7379/*
7380 * Searches all registered partitions for autorun RAID arrays
7381 * at boot time.
7382 */
4d936ec1
ME		 7383
7384static LIST_HEAD(all_detected_devices);
7385struct detected_devices_node {
7386 struct list_head list;
7387 dev_t dev;
7388};
1da177e4
LT		 7389
7390void md_autodetect_dev(dev_t dev)
7391{
4d936ec1
ME		 7392 struct detected_devices_node *node_detected_dev;
7393
7394 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
7395 if (node_detected_dev) {
7396 node_detected_dev->dev = dev;
7397 list_add_tail(&node_detected_dev->list, &all_detected_devices);
7398 } else {
7399 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
7400 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
7401 }
1da177e4
LT		 7402}
7403
7404
7405static void autostart_arrays(int part)
7406{
7407 mdk_rdev_t *rdev;
4d936ec1
ME		 7408 struct detected_devices_node *node_detected_dev;
7409 dev_t dev;
7410 int i_scanned, i_passed;
1da177e4 7411
4d936ec1
ME		 7412 i_scanned = 0;
7413 i_passed = 0;
1da177e4 7414
4d936ec1 7415 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 7416
4d936ec1
ME		 7417 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
7418 i_scanned++;
7419 node_detected_dev = list_entry(all_detected_devices.next,
7420 struct detected_devices_node, list);
7421 list_del(&node_detected_dev->list);
7422 dev = node_detected_dev->dev;
7423 kfree(node_detected_dev);
df968c4e 7424 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 7425 if (IS_ERR(rdev))
7426 continue;
7427
b2d444d7 7428 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 7429 MD_BUG();
7430 continue;
7431 }
d0fae18f 7432 set_bit(AutoDetected, &rdev->flags);
1da177e4 7433 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 7434 i_passed++;
1da177e4 7435 }
4d936ec1
ME		 7436
7437 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
7438 i_scanned, i_passed);
1da177e4
LT		 7439
7440 autorun_devices(part);
7441}
7442
fdee8ae4 7443#endif /* !MODULE */
1da177e4
LT		 7444
7445static __exit void md_exit(void)
7446{
7447 mddev_t *mddev;
7448 struct list_head *tmp;
8ab5e4c1 7449
3dbd8c2e 7450 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 7451 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 7452
3dbd8c2e 7453 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 7454 unregister_blkdev(mdp_major, "mdp");
7455 unregister_reboot_notifier(&md_notifier);
7456 unregister_sysctl_table(raid_table_header);
7457 remove_proc_entry("mdstat", NULL);
29ac4aa3 7458 for_each_mddev(mddev, tmp) {
1da177e4 7459 export_array(mddev);
d3374825 7460 mddev->hold_active = 0;
1da177e4 7461 }
e804ac78
TH		 7462 destroy_workqueue(md_misc_wq);
7463 destroy_workqueue(md_wq);
1da177e4
LT		 7464}
7465
685784aa 7466subsys_initcall(md_init);
1da177e4
LT		 7467module_exit(md_exit)
7468
f91de92e
N		 7469static int get_ro(char *buffer, struct kernel_param *kp)
7470{
7471 return sprintf(buffer, "%d", start_readonly);
7472}
7473static int set_ro(const char *val, struct kernel_param *kp)
7474{
7475 char *e;
7476 int num = simple_strtoul(val, &e, 10);
7477 if (*val && (*e == '\0' || *e == '\n')) {
7478 start_readonly = num;
4dbcdc75 7479 return 0;
f91de92e
N		 7480 }
7481 return -EINVAL;
7482}
7483
80ca3a44
N		 7484module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
7485module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 7486
efeb53c0 7487module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 7488
1da177e4
LT		 7489EXPORT_SYMBOL(register_md_personality);
7490EXPORT_SYMBOL(unregister_md_personality);
7491EXPORT_SYMBOL(md_error);
7492EXPORT_SYMBOL(md_done_sync);
7493EXPORT_SYMBOL(md_write_start);
7494EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 7495EXPORT_SYMBOL(md_register_thread);
7496EXPORT_SYMBOL(md_unregister_thread);
7497EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 7498EXPORT_SYMBOL(md_check_recovery);
7499MODULE_LICENSE("GPL");
0efb9e61 7500MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 7501MODULE_ALIAS("md");
72008652 7502MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux 6.x block layer and io_uring tree(s)