projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md/bitmap: remove some unused noise from bitmap.h
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
ff01bb48 39#include <linux/fs.h>
d7603b7e 40#include <linux/poll.h>
16f17b39 41#include <linux/ctype.h>
e7d2860b 42#include <linux/string.h>
fb4d8c76
N		 43#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
056075c7 46#include <linux/module.h>
fb4d8c76 47#include <linux/reboot.h>
32a7627c 48#include <linux/file.h>
aa98aa31 49#include <linux/compat.h>
25570727 50#include <linux/delay.h>
bff61975
N		 51#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
5a0e3ad6 53#include <linux/slab.h>
43b2e5d8 54#include "md.h"
ef740c37 55#include "bitmap.h"
1da177e4 56
1da177e4 57#ifndef MODULE
d710e138 58static void autostart_arrays(int part);
1da177e4
LT		 59#endif
60
01f96c0a
N		 61/* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
2604b703 66static LIST_HEAD(pers_list);
1da177e4
LT		 67static DEFINE_SPINLOCK(pers_lock);
68
5e56341d
AB		 69static void md_print_devices(void);
70
90b08710 71static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 72static struct workqueue_struct *md_wq;
73static struct workqueue_struct *md_misc_wq;
90b08710 74
5e56341d
AB		 75#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76
1e50915f
RB		 77/*
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
81 */
82#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 83/*
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 87 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
90 * idle IO detection.
91 *
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 93 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 94 */
95
96static int sysctl_speed_limit_min = 1000;
97static int sysctl_speed_limit_max = 200000;
fd01b88c 98static inline int speed_min(struct mddev *mddev)
88202a0c
N		 99{
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
102}
103
fd01b88c 104static inline int speed_max(struct mddev *mddev)
88202a0c
N		 105{
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
108}
1da177e4
LT		 109
110static struct ctl_table_header *raid_table_header;
111
112static ctl_table raid_table[] = {
113 {
1da177e4
LT		 114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
80ca3a44 117 .mode = S_IRUGO|S_IWUSR,
6d456111 118 .proc_handler = proc_dointvec,
1da177e4
LT		 119 },
120 {
1da177e4
LT		 121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
80ca3a44 124 .mode = S_IRUGO|S_IWUSR,
6d456111 125 .proc_handler = proc_dointvec,
1da177e4 126 },
894d2491 127 { }
1da177e4
LT		 128};
129
130static ctl_table raid_dir_table[] = {
131 {
1da177e4
LT		 132 .procname = "raid",
133 .maxlen = 0,
80ca3a44 134 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 135 .child = raid_table,
136 },
894d2491 137 { }
1da177e4
LT		 138};
139
140static ctl_table raid_root_table[] = {
141 {
1da177e4
LT		 142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
894d2491 147 { }
1da177e4
LT		 148};
149
83d5cde4 150static const struct block_device_operations md_fops;
1da177e4 151
f91de92e
N		 152static int start_readonly;
153
a167f663
N		 154/* bio_clone_mddev
155 * like bio_clone, but with a local bio set
156 */
157
158static void mddev_bio_destructor(struct bio *bio)
159{
fd01b88c 160 struct mddev *mddev, **mddevp;
a167f663
N		 161
162 mddevp = (void*)bio;
163 mddev = mddevp[-1];
164
165 bio_free(bio, mddev->bio_set);
166}
167
168struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 169 struct mddev *mddev)
a167f663
N		 170{
171 struct bio *b;
fd01b88c 172 struct mddev **mddevp;
a167f663
N		 173
174 if (!mddev || !mddev->bio_set)
175 return bio_alloc(gfp_mask, nr_iovecs);
176
177 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
178 mddev->bio_set);
179 if (!b)
180 return NULL;
181 mddevp = (void*)b;
182 mddevp[-1] = mddev;
183 b->bi_destructor = mddev_bio_destructor;
184 return b;
185}
186EXPORT_SYMBOL_GPL(bio_alloc_mddev);
187
188struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 189 struct mddev *mddev)
a167f663
N		 190{
191 struct bio *b;
fd01b88c 192 struct mddev **mddevp;
a167f663
N		 193
194 if (!mddev || !mddev->bio_set)
195 return bio_clone(bio, gfp_mask);
196
197 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
198 mddev->bio_set);
199 if (!b)
200 return NULL;
201 mddevp = (void*)b;
202 mddevp[-1] = mddev;
203 b->bi_destructor = mddev_bio_destructor;
204 __bio_clone(b, bio);
205 if (bio_integrity(bio)) {
206 int ret;
207
208 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
209
210 if (ret < 0) {
211 bio_put(b);
212 return NULL;
213 }
214 }
215
216 return b;
217}
218EXPORT_SYMBOL_GPL(bio_clone_mddev);
219
d2eb35ac
N		 220void md_trim_bio(struct bio *bio, int offset, int size)
221{
222 /* 'bio' is a cloned bio which we need to trim to match
223 * the given offset and size.
224 * This requires adjusting bi_sector, bi_size, and bi_io_vec
225 */
226 int i;
227 struct bio_vec *bvec;
228 int sofar = 0;
229
230 size <<= 9;
231 if (offset == 0 && size == bio->bi_size)
232 return;
233
234 bio->bi_sector += offset;
235 bio->bi_size = size;
236 offset <<= 9;
237 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
238
239 while (bio->bi_idx < bio->bi_vcnt &&
240 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
241 /* remove this whole bio_vec */
242 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
243 bio->bi_idx++;
244 }
245 if (bio->bi_idx < bio->bi_vcnt) {
246 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
247 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
248 }
249 /* avoid any complications with bi_idx being non-zero*/
250 if (bio->bi_idx) {
251 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
252 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
253 bio->bi_vcnt -= bio->bi_idx;
254 bio->bi_idx = 0;
255 }
256 /* Make sure vcnt and last bv are not too big */
257 bio_for_each_segment(bvec, bio, i) {
258 if (sofar + bvec->bv_len > size)
259 bvec->bv_len = size - sofar;
260 if (bvec->bv_len == 0) {
261 bio->bi_vcnt = i;
262 break;
263 }
264 sofar += bvec->bv_len;
265 }
266}
267EXPORT_SYMBOL_GPL(md_trim_bio);
268
d7603b7e
N		 269/*
270 * We have a system wide 'event count' that is incremented
271 * on any 'interesting' event, and readers of /proc/mdstat
272 * can use 'poll' or 'select' to find out when the event
273 * count increases.
274 *
275 * Events are:
276 * start array, stop array, error, add device, remove device,
277 * start build, activate spare
278 */
2989ddbd 279static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 280static atomic_t md_event_count;
fd01b88c 281void md_new_event(struct mddev *mddev)
d7603b7e
N		 282{
283 atomic_inc(&md_event_count);
284 wake_up(&md_event_waiters);
285}
29269553 286EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 287
c331eb04
N		 288/* Alternate version that can be called from interrupts
289 * when calling sysfs_notify isn't needed.
290 */
fd01b88c 291static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 292{
293 atomic_inc(&md_event_count);
294 wake_up(&md_event_waiters);
295}
296
1da177e4
LT		 297/*
298 * Enables to iterate over all existing md arrays
299 * all_mddevs_lock protects this list.
300 */
301static LIST_HEAD(all_mddevs);
302static DEFINE_SPINLOCK(all_mddevs_lock);
303
304
305/*
306 * iterates through all used mddevs in the system.
307 * We take care to grab the all_mddevs_lock whenever navigating
308 * the list, and to always hold a refcount when unlocked.
309 * Any code which breaks out of this loop while own
310 * a reference to the current mddev and must mddev_put it.
311 */
fd01b88c 312#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 313 \
314 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 315 _tmp = all_mddevs.next; \
316 _mddev = NULL;}); \
317 ({ if (_tmp != &all_mddevs) \
318 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 319 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 320 if (_mddev) mddev_put(_mddev); \
321 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
322 _tmp != &all_mddevs;}); \
1da177e4 323 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 324 _tmp = _tmp->next;}) \
1da177e4
LT		 325 )
326
327
409c57f3
N		 328/* Rather than calling directly into the personality make_request function,
329 * IO requests come here first so that we can check if the device is
330 * being suspended pending a reconfiguration.
331 * We hold a refcount over the call to ->make_request. By the time that
332 * call has finished, the bio has been linked into some internal structure
333 * and so is visible to ->quiesce(), so we don't need the refcount any more.
334 */
5a7bbad2 335static void md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 336{
49077326 337 const int rw = bio_data_dir(bio);
fd01b88c 338 struct mddev *mddev = q->queuedata;
49077326 339 int cpu;
e91ece55 340 unsigned int sectors;
49077326 341
0ca69886
N		 342 if (mddev == NULL || mddev->pers == NULL
343 || !mddev->ready) {
409c57f3 344 bio_io_error(bio);
5a7bbad2 345 return;
409c57f3 346 }
0ca69886 347 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 348 rcu_read_lock();
e9c7469b 349 if (mddev->suspended) {
409c57f3
N		 350 DEFINE_WAIT(__wait);
351 for (;;) {
352 prepare_to_wait(&mddev->sb_wait, &__wait,
353 TASK_UNINTERRUPTIBLE);
e9c7469b 354 if (!mddev->suspended)
409c57f3
N		 355 break;
356 rcu_read_unlock();
357 schedule();
358 rcu_read_lock();
359 }
360 finish_wait(&mddev->sb_wait, &__wait);
361 }
362 atomic_inc(&mddev->active_io);
363 rcu_read_unlock();
49077326 364
e91ece55
CM		 365 /*
366 * save the sectors now since our bio can
367 * go away inside make_request
368 */
369 sectors = bio_sectors(bio);
5a7bbad2 370 mddev->pers->make_request(mddev, bio);
49077326
N		 371
372 cpu = part_stat_lock();
373 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 374 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 375 part_stat_unlock();
376
409c57f3
N		 377 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
378 wake_up(&mddev->sb_wait);
409c57f3
N		 379}
380
9e35b99c
N		 381/* mddev_suspend makes sure no new requests are submitted
382 * to the device, and that any requests that have been submitted
383 * are completely handled.
384 * Once ->stop is called and completes, the module will be completely
385 * unused.
386 */
fd01b88c 387void mddev_suspend(struct mddev *mddev)
409c57f3
N		 388{
389 BUG_ON(mddev->suspended);
390 mddev->suspended = 1;
391 synchronize_rcu();
392 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
393 mddev->pers->quiesce(mddev, 1);
409c57f3 394}
390ee602 395EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 396
fd01b88c 397void mddev_resume(struct mddev *mddev)
409c57f3
N		 398{
399 mddev->suspended = 0;
400 wake_up(&mddev->sb_wait);
401 mddev->pers->quiesce(mddev, 0);
0fd018af
JB		 402
403 md_wakeup_thread(mddev->thread);
404 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 405}
390ee602 406EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 407
fd01b88c 408int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 409{
410 return mddev->suspended;
411}
412EXPORT_SYMBOL(mddev_congested);
413
a2826aa9 414/*
e9c7469b 415 * Generic flush handling for md
a2826aa9
N		 416 */
417
e9c7469b 418static void md_end_flush(struct bio *bio, int err)
a2826aa9 419{
3cb03002 420 struct md_rdev *rdev = bio->bi_private;
fd01b88c 421 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 422
423 rdev_dec_pending(rdev, mddev);
424
425 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 426 /* The pre-request flush has finished */
e804ac78 427 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 428 }
429 bio_put(bio);
430}
431
a7a07e69
N		 432static void md_submit_flush_data(struct work_struct *ws);
433
a035fc3e 434static void submit_flushes(struct work_struct *ws)
a2826aa9 435{
fd01b88c 436 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 437 struct md_rdev *rdev;
a2826aa9 438
a7a07e69
N		 439 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
440 atomic_set(&mddev->flush_pending, 1);
a2826aa9 441 rcu_read_lock();
dafb20fa 442 rdev_for_each_rcu(rdev, mddev)
a2826aa9
N		 443 if (rdev->raid_disk >= 0 &&
444 !test_bit(Faulty, &rdev->flags)) {
445 /* Take two references, one is dropped
446 * when request finishes, one after
447 * we reclaim rcu_read_lock
448 */
449 struct bio *bi;
450 atomic_inc(&rdev->nr_pending);
451 atomic_inc(&rdev->nr_pending);
452 rcu_read_unlock();
a167f663 453 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 454 bi->bi_end_io = md_end_flush;
a2826aa9
N		 455 bi->bi_private = rdev;
456 bi->bi_bdev = rdev->bdev;
457 atomic_inc(&mddev->flush_pending);
e9c7469b 458 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 459 rcu_read_lock();
460 rdev_dec_pending(rdev, mddev);
461 }
462 rcu_read_unlock();
a7a07e69
N		 463 if (atomic_dec_and_test(&mddev->flush_pending))
464 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 465}
466
e9c7469b 467static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 468{
fd01b88c 469 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 470 struct bio *bio = mddev->flush_bio;
a2826aa9 471
e9c7469b 472 if (bio->bi_size == 0)
a2826aa9
N		 473 /* an empty barrier - all done */
474 bio_endio(bio, 0);
475 else {
e9c7469b 476 bio->bi_rw &= ~REQ_FLUSH;
5a7bbad2 477 mddev->pers->make_request(mddev, bio);
a2826aa9 478 }
2b74e12e
N		 479
480 mddev->flush_bio = NULL;
481 wake_up(&mddev->sb_wait);
a2826aa9
N		 482}
483
fd01b88c 484void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 485{
486 spin_lock_irq(&mddev->write_lock);
487 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 488 !mddev->flush_bio,
a2826aa9 489 mddev->write_lock, /*nothing*/);
e9c7469b 490 mddev->flush_bio = bio;
a2826aa9
N		 491 spin_unlock_irq(&mddev->write_lock);
492
a035fc3e
N		 493 INIT_WORK(&mddev->flush_work, submit_flushes);
494 queue_work(md_wq, &mddev->flush_work);
a2826aa9 495}
e9c7469b 496EXPORT_SYMBOL(md_flush_request);
409c57f3 497
97658cdd
N		 498/* Support for plugging.
499 * This mirrors the plugging support in request_queue, but does not
500 * require having a whole queue or request structures.
501 * We allocate an md_plug_cb for each md device and each thread it gets
502 * plugged on. This links tot the private plug_handle structure in the
503 * personality data where we keep a count of the number of outstanding
504 * plugs so other code can see if a plug is active.
505 */
506struct md_plug_cb {
507 struct blk_plug_cb cb;
fd01b88c 508 struct mddev *mddev;
97658cdd
N		 509};
510
511static void plugger_unplug(struct blk_plug_cb *cb)
512{
513 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
514 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
515 md_wakeup_thread(mdcb->mddev->thread);
516 kfree(mdcb);
517}
518
519/* Check that an unplug wakeup will come shortly.
520 * If not, wakeup the md thread immediately
521 */
fd01b88c 522int mddev_check_plugged(struct mddev *mddev)
97658cdd
N		 523{
524 struct blk_plug *plug = current->plug;
525 struct md_plug_cb *mdcb;
526
527 if (!plug)
528 return 0;
529
530 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
531 if (mdcb->cb.callback == plugger_unplug &&
532 mdcb->mddev == mddev) {
533 /* Already on the list, move to top */
534 if (mdcb != list_first_entry(&plug->cb_list,
535 struct md_plug_cb,
536 cb.list))
537 list_move(&mdcb->cb.list, &plug->cb_list);
538 return 1;
539 }
540 }
541 /* Not currently on the callback list */
542 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
543 if (!mdcb)
544 return 0;
545
546 mdcb->mddev = mddev;
547 mdcb->cb.callback = plugger_unplug;
548 atomic_inc(&mddev->plug_cnt);
549 list_add(&mdcb->cb.list, &plug->cb_list);
550 return 1;
551}
552EXPORT_SYMBOL_GPL(mddev_check_plugged);
2ac87401 553
fd01b88c 554static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 555{
556 atomic_inc(&mddev->active);
557 return mddev;
558}
559
5fd3a17e 560static void mddev_delayed_delete(struct work_struct *ws);
d3374825 561
fd01b88c 562static void mddev_put(struct mddev *mddev)
1da177e4 563{
a167f663
N		 564 struct bio_set *bs = NULL;
565
1da177e4
LT		 566 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
567 return;
d3374825 568 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 569 mddev->ctime == 0 && !mddev->hold_active) {
570 /* Array is not configured at all, and not held active,
571 * so destroy it */
af8a2434 572 list_del_init(&mddev->all_mddevs);
a167f663
N		 573 bs = mddev->bio_set;
574 mddev->bio_set = NULL;
d3374825 575 if (mddev->gendisk) {
e804ac78
TH		 576 /* We did a probe so need to clean up. Call
577 * queue_work inside the spinlock so that
578 * flush_workqueue() after mddev_find will
579 * succeed in waiting for the work to be done.
d3374825
N		 580 */
581 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 582 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 583 } else
584 kfree(mddev);
585 }
586 spin_unlock(&all_mddevs_lock);
a167f663
N		 587 if (bs)
588 bioset_free(bs);
1da177e4
LT		 589}
590
fd01b88c 591void mddev_init(struct mddev *mddev)
fafd7fb0
N		 592{
593 mutex_init(&mddev->open_mutex);
594 mutex_init(&mddev->reconfig_mutex);
595 mutex_init(&mddev->bitmap_info.mutex);
596 INIT_LIST_HEAD(&mddev->disks);
597 INIT_LIST_HEAD(&mddev->all_mddevs);
598 init_timer(&mddev->safemode_timer);
599 atomic_set(&mddev->active, 1);
600 atomic_set(&mddev->openers, 0);
601 atomic_set(&mddev->active_io, 0);
97658cdd 602 atomic_set(&mddev->plug_cnt, 0);
fafd7fb0
N		 603 spin_lock_init(&mddev->write_lock);
604 atomic_set(&mddev->flush_pending, 0);
605 init_waitqueue_head(&mddev->sb_wait);
606 init_waitqueue_head(&mddev->recovery_wait);
607 mddev->reshape_position = MaxSector;
608 mddev->resync_min = 0;
609 mddev->resync_max = MaxSector;
610 mddev->level = LEVEL_NONE;
611}
390ee602 612EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 613
fd01b88c 614static struct mddev * mddev_find(dev_t unit)
1da177e4 615{
fd01b88c 616 struct mddev *mddev, *new = NULL;
1da177e4 617
8f5f02c4
N		 618 if (unit && MAJOR(unit) != MD_MAJOR)
619 unit &= ~((1<<MdpMinorShift)-1);
620
1da177e4
LT		 621 retry:
622 spin_lock(&all_mddevs_lock);
efeb53c0
N		 623
624 if (unit) {
625 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
626 if (mddev->unit == unit) {
627 mddev_get(mddev);
628 spin_unlock(&all_mddevs_lock);
629 kfree(new);
630 return mddev;
631 }
632
633 if (new) {
634 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 635 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 636 new->hold_active = UNTIL_IOCTL;
637 return new;
1da177e4 638 }
efeb53c0
N		 639 } else if (new) {
640 /* find an unused unit number */
641 static int next_minor = 512;
642 int start = next_minor;
643 int is_free = 0;
644 int dev = 0;
645 while (!is_free) {
646 dev = MKDEV(MD_MAJOR, next_minor);
647 next_minor++;
648 if (next_minor > MINORMASK)
649 next_minor = 0;
650 if (next_minor == start) {
651 /* Oh dear, all in use. */
652 spin_unlock(&all_mddevs_lock);
653 kfree(new);
654 return NULL;
655 }
656
657 is_free = 1;
658 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
659 if (mddev->unit == dev) {
660 is_free = 0;
661 break;
662 }
663 }
664 new->unit = dev;
665 new->md_minor = MINOR(dev);
666 new->hold_active = UNTIL_STOP;
1da177e4
LT		 667 list_add(&new->all_mddevs, &all_mddevs);
668 spin_unlock(&all_mddevs_lock);
669 return new;
670 }
671 spin_unlock(&all_mddevs_lock);
672
9ffae0cf 673 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 674 if (!new)
675 return NULL;
676
1da177e4
LT		 677 new->unit = unit;
678 if (MAJOR(unit) == MD_MAJOR)
679 new->md_minor = MINOR(unit);
680 else
681 new->md_minor = MINOR(unit) >> MdpMinorShift;
682
fafd7fb0 683 mddev_init(new);
1da177e4 684
1da177e4
LT		 685 goto retry;
686}
687
fd01b88c 688static inline int mddev_lock(struct mddev * mddev)
1da177e4 689{
df5b89b3 690 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 691}
692
fd01b88c 693static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 694{
695 return mutex_is_locked(&mddev->reconfig_mutex);
696}
697
fd01b88c 698static inline int mddev_trylock(struct mddev * mddev)
1da177e4 699{
df5b89b3 700 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 701}
702
b6eb127d
N		 703static struct attribute_group md_redundancy_group;
704
fd01b88c 705static void mddev_unlock(struct mddev * mddev)
1da177e4 706{
a64c876f 707 if (mddev->to_remove) {
b6eb127d
N		 708 /* These cannot be removed under reconfig_mutex as
709 * an access to the files will try to take reconfig_mutex
710 * while holding the file unremovable, which leads to
711 * a deadlock.
bb4f1e9d
N		 712 * So hold set sysfs_active while the remove in happeing,
713 * and anything else which might set ->to_remove or my
714 * otherwise change the sysfs namespace will fail with
715 * -EBUSY if sysfs_active is still set.
716 * We set sysfs_active under reconfig_mutex and elsewhere
717 * test it under the same mutex to ensure its correct value
718 * is seen.
b6eb127d 719 */
a64c876f
N		 720 struct attribute_group *to_remove = mddev->to_remove;
721 mddev->to_remove = NULL;
bb4f1e9d 722 mddev->sysfs_active = 1;
b6eb127d
N		 723 mutex_unlock(&mddev->reconfig_mutex);
724
00bcb4ac
N		 725 if (mddev->kobj.sd) {
726 if (to_remove != &md_redundancy_group)
727 sysfs_remove_group(&mddev->kobj, to_remove);
728 if (mddev->pers == NULL ||
729 mddev->pers->sync_request == NULL) {
730 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
731 if (mddev->sysfs_action)
732 sysfs_put(mddev->sysfs_action);
733 mddev->sysfs_action = NULL;
734 }
a64c876f 735 }
bb4f1e9d 736 mddev->sysfs_active = 0;
b6eb127d
N		 737 } else
738 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 739
751e67ca
CD		 740 /* As we've dropped the mutex we need a spinlock to
741 * make sure the thread doesn't disappear
01f96c0a
N		 742 */
743 spin_lock(&pers_lock);
005eca5e 744 md_wakeup_thread(mddev->thread);
01f96c0a 745 spin_unlock(&pers_lock);
1da177e4
LT		 746}
747
fd01b88c 748static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 749{
3cb03002 750 struct md_rdev *rdev;
1da177e4 751
dafb20fa 752 rdev_for_each(rdev, mddev)
1da177e4
LT		 753 if (rdev->desc_nr == nr)
754 return rdev;
159ec1fc 755
1da177e4
LT		 756 return NULL;
757}
758
fd01b88c 759static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
1da177e4 760{
3cb03002 761 struct md_rdev *rdev;
1da177e4 762
dafb20fa 763 rdev_for_each(rdev, mddev)
1da177e4
LT		 764 if (rdev->bdev->bd_dev == dev)
765 return rdev;
159ec1fc 766
1da177e4
LT		 767 return NULL;
768}
769
84fc4b56 770static struct md_personality *find_pers(int level, char *clevel)
2604b703 771{
84fc4b56 772 struct md_personality *pers;
d9d166c2
N		 773 list_for_each_entry(pers, &pers_list, list) {
774 if (level != LEVEL_NONE && pers->level == level)
2604b703 775 return pers;
d9d166c2
N		 776 if (strcmp(pers->name, clevel)==0)
777 return pers;
778 }
2604b703
N		 779 return NULL;
780}
781
b73df2d3 782/* return the offset of the super block in 512byte sectors */
3cb03002 783static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 784{
57b2caa3 785 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 786 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 787}
788
3cb03002 789static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 790{
791 if (rdev->sb_page)
792 MD_BUG();
793
794 rdev->sb_page = alloc_page(GFP_KERNEL);
795 if (!rdev->sb_page) {
796 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 797 return -ENOMEM;
1da177e4
LT		 798 }
799
800 return 0;
801}
802
3cb03002 803static void free_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 804{
805 if (rdev->sb_page) {
2d1f3b5d 806 put_page(rdev->sb_page);
1da177e4
LT		 807 rdev->sb_loaded = 0;
808 rdev->sb_page = NULL;
0f420358 809 rdev->sb_start = 0;
dd8ac336 810 rdev->sectors = 0;
1da177e4 811 }
2699b672
N		 812 if (rdev->bb_page) {
813 put_page(rdev->bb_page);
814 rdev->bb_page = NULL;
815 }
1da177e4
LT		 816}
817
818
6712ecf8 819static void super_written(struct bio *bio, int error)
7bfa19f2 820{
3cb03002 821 struct md_rdev *rdev = bio->bi_private;
fd01b88c 822 struct mddev *mddev = rdev->mddev;
7bfa19f2 823
3a0f5bbb
N		 824 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
825 printk("md: super_written gets error=%d, uptodate=%d\n",
826 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
827 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 828 md_error(mddev, rdev);
3a0f5bbb 829 }
7bfa19f2 830
a9701a30
N		 831 if (atomic_dec_and_test(&mddev->pending_writes))
832 wake_up(&mddev->sb_wait);
f8b58edf 833 bio_put(bio);
7bfa19f2
N		 834}
835
fd01b88c 836void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 837 sector_t sector, int size, struct page *page)
838{
839 /* write first size bytes of page to sector of rdev
840 * Increment mddev->pending_writes before returning
841 * and decrement it on completion, waking up sb_wait
842 * if zero is reached.
843 * If an error occurred, call md_error
844 */
a167f663 845 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 846
a6ff7e08 847 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 848 bio->bi_sector = sector;
849 bio_add_page(bio, page, size, 0);
850 bio->bi_private = rdev;
851 bio->bi_end_io = super_written;
a9701a30 852
7bfa19f2 853 atomic_inc(&mddev->pending_writes);
a5bf4df0 854 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 855}
856
fd01b88c 857void md_super_wait(struct mddev *mddev)
a9701a30 858{
e9c7469b 859 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 860 DEFINE_WAIT(wq);
861 for(;;) {
862 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
863 if (atomic_read(&mddev->pending_writes)==0)
864 break;
a9701a30
N		 865 schedule();
866 }
867 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 868}
869
6712ecf8 870static void bi_complete(struct bio *bio, int error)
1da177e4 871{
1da177e4 872 complete((struct completion*)bio->bi_private);
1da177e4
LT		 873}
874
3cb03002 875int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 876 struct page *page, int rw, bool metadata_op)
1da177e4 877{
a167f663 878 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 879 struct completion event;
880 int ret;
881
721a9602 882 rw |= REQ_SYNC;
1da177e4 883
a6ff7e08
JB		 884 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
885 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 886 if (metadata_op)
887 bio->bi_sector = sector + rdev->sb_start;
888 else
889 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 890 bio_add_page(bio, page, size, 0);
891 init_completion(&event);
892 bio->bi_private = &event;
893 bio->bi_end_io = bi_complete;
894 submit_bio(rw, bio);
895 wait_for_completion(&event);
896
897 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
898 bio_put(bio);
899 return ret;
900}
a8745db2 901EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 902
3cb03002 903static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 904{
905 char b[BDEVNAME_SIZE];
906 if (!rdev->sb_page) {
907 MD_BUG();
908 return -EINVAL;
909 }
910 if (rdev->sb_loaded)
911 return 0;
912
913
ccebd4c4 914 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 915 goto fail;
916 rdev->sb_loaded = 1;
917 return 0;
918
919fail:
920 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
921 bdevname(rdev->bdev,b));
922 return -EINVAL;
923}
924
925static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
926{
05710466
AN		 927 return sb1->set_uuid0 == sb2->set_uuid0 &&
928 sb1->set_uuid1 == sb2->set_uuid1 &&
929 sb1->set_uuid2 == sb2->set_uuid2 &&
930 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 931}
932
1da177e4
LT		 933static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
934{
935 int ret;
936 mdp_super_t *tmp1, *tmp2;
937
938 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
939 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
940
941 if (!tmp1 || !tmp2) {
942 ret = 0;
35020f1a 943 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 944 goto abort;
945 }
946
947 *tmp1 = *sb1;
948 *tmp2 = *sb2;
949
950 /*
951 * nr_disks is not constant
952 */
953 tmp1->nr_disks = 0;
954 tmp2->nr_disks = 0;
955
ce0c8e05 956 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 957abort:
990a8baf
JJ		 958 kfree(tmp1);
959 kfree(tmp2);
1da177e4
LT		 960 return ret;
961}
962
4d167f09
N		 963
964static u32 md_csum_fold(u32 csum)
965{
966 csum = (csum & 0xffff) + (csum >> 16);
967 return (csum & 0xffff) + (csum >> 16);
968}
969
1da177e4
LT		 970static unsigned int calc_sb_csum(mdp_super_t * sb)
971{
4d167f09
N		 972 u64 newcsum = 0;
973 u32 *sb32 = (u32*)sb;
974 int i;
1da177e4
LT		 975 unsigned int disk_csum, csum;
976
977 disk_csum = sb->sb_csum;
978 sb->sb_csum = 0;
4d167f09
N		 979
980 for (i = 0; i < MD_SB_BYTES/4 ; i++)
981 newcsum += sb32[i];
982 csum = (newcsum & 0xffffffff) + (newcsum>>32);
983
984
985#ifdef CONFIG_ALPHA
986 /* This used to use csum_partial, which was wrong for several
987 * reasons including that different results are returned on
988 * different architectures. It isn't critical that we get exactly
989 * the same return value as before (we always csum_fold before
990 * testing, and that removes any differences). However as we
991 * know that csum_partial always returned a 16bit value on
992 * alphas, do a fold to maximise conformity to previous behaviour.
993 */
994 sb->sb_csum = md_csum_fold(disk_csum);
995#else
1da177e4 996 sb->sb_csum = disk_csum;
4d167f09 997#endif
1da177e4
LT		 998 return csum;
999}
1000
1001
1002/*
1003 * Handle superblock details.
1004 * We want to be able to handle multiple superblock formats
1005 * so we have a common interface to them all, and an array of
1006 * different handlers.
1007 * We rely on user-space to write the initial superblock, and support
1008 * reading and updating of superblocks.
1009 * Interface methods are:
3cb03002 1010 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1011 * loads and validates a superblock on dev.
1012 * if refdev != NULL, compare superblocks on both devices
1013 * Return:
1014 * 0 - dev has a superblock that is compatible with refdev
1015 * 1 - dev has a superblock that is compatible and newer than refdev
1016 * so dev should be used as the refdev in future
1017 * -EINVAL superblock incompatible or invalid
1018 * -othererror e.g. -EIO
1019 *
fd01b88c 1020 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1021 * Verify that dev is acceptable into mddev.
1022 * The first time, mddev->raid_disks will be 0, and data from
1023 * dev should be merged in. Subsequent calls check that dev
1024 * is new enough. Return 0 or -EINVAL
1025 *
fd01b88c 1026 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1027 * Update the superblock for rdev with data in mddev
1028 * This does not write to disc.
1029 *
1030 */
1031
1032struct super_type {
0cd17fec
CW		 1033 char *name;
1034 struct module *owner;
3cb03002 1035 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
0cd17fec 1036 int minor_version);
fd01b88c
N		 1037 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1038 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
3cb03002 1039 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1040 sector_t num_sectors);
1da177e4
LT		 1041};
1042
0894cc30
AN		 1043/*
1044 * Check that the given mddev has no bitmap.
1045 *
1046 * This function is called from the run method of all personalities that do not
1047 * support bitmaps. It prints an error message and returns non-zero if mddev
1048 * has a bitmap. Otherwise, it returns 0.
1049 *
1050 */
fd01b88c 1051int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1052{
c3d9714e 1053 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1054 return 0;
1055 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1056 mdname(mddev), mddev->pers->name);
1057 return 1;
1058}
1059EXPORT_SYMBOL(md_check_no_bitmap);
1060
1da177e4
LT		 1061/*
1062 * load_super for 0.90.0
1063 */
3cb03002 1064static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1065{
1066 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1067 mdp_super_t *sb;
1068 int ret;
1da177e4
LT		 1069
1070 /*
0f420358 1071 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1072 * it's at the end of the disk.
1073 *
1074 * It also happens to be a multiple of 4Kb.
1075 */
57b2caa3 1076 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1077
0002b271 1078 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1079 if (ret) return ret;
1080
1081 ret = -EINVAL;
1082
1083 bdevname(rdev->bdev, b);
65a06f06 1084 sb = page_address(rdev->sb_page);
1da177e4
LT		 1085
1086 if (sb->md_magic != MD_SB_MAGIC) {
1087 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1088 b);
1089 goto abort;
1090 }
1091
1092 if (sb->major_version != 0 ||
f6705578
N		 1093 sb->minor_version < 90 ||
1094 sb->minor_version > 91) {
1da177e4
LT		 1095 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1096 sb->major_version, sb->minor_version,
1097 b);
1098 goto abort;
1099 }
1100
1101 if (sb->raid_disks <= 0)
1102 goto abort;
1103
4d167f09 1104 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1105 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1106 b);
1107 goto abort;
1108 }
1109
1110 rdev->preferred_minor = sb->md_minor;
1111 rdev->data_offset = 0;
0002b271 1112 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1113 rdev->badblocks.shift = -1;
1da177e4
LT		 1114
1115 if (sb->level == LEVEL_MULTIPATH)
1116 rdev->desc_nr = -1;
1117 else
1118 rdev->desc_nr = sb->this_disk.number;
1119
9a7b2b0f 1120 if (!refdev) {
1da177e4 1121 ret = 1;
9a7b2b0f 1122 } else {
1da177e4 1123 __u64 ev1, ev2;
65a06f06 1124 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1125 if (!uuid_equal(refsb, sb)) {
1126 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1127 b, bdevname(refdev->bdev,b2));
1128 goto abort;
1129 }
1130 if (!sb_equal(refsb, sb)) {
1131 printk(KERN_WARNING "md: %s has same UUID"
1132 " but different superblock to %s\n",
1133 b, bdevname(refdev->bdev, b2));
1134 goto abort;
1135 }
1136 ev1 = md_event(sb);
1137 ev2 = md_event(refsb);
1138 if (ev1 > ev2)
1139 ret = 1;
1140 else
1141 ret = 0;
1142 }
8190e754 1143 rdev->sectors = rdev->sb_start;
27a7b260
N		 1144 /* Limit to 4TB as metadata cannot record more than that */
1145 if (rdev->sectors >= (2ULL << 32))
1146 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1147
27a7b260 1148 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1149 /* "this cannot possibly happen" ... */
1150 ret = -EINVAL;
1151
1da177e4
LT		 1152 abort:
1153 return ret;
1154}
1155
1156/*
1157 * validate_super for 0.90.0
1158 */
fd01b88c 1159static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1160{
1161 mdp_disk_t *desc;
65a06f06 1162 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1163 __u64 ev1 = md_event(sb);
1da177e4 1164
41158c7e 1165 rdev->raid_disk = -1;
c5d79adb
N		 1166 clear_bit(Faulty, &rdev->flags);
1167 clear_bit(In_sync, &rdev->flags);
1168 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1169
1da177e4
LT		 1170 if (mddev->raid_disks == 0) {
1171 mddev->major_version = 0;
1172 mddev->minor_version = sb->minor_version;
1173 mddev->patch_version = sb->patch_version;
e691063a 1174 mddev->external = 0;
9d8f0363 1175 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1176 mddev->ctime = sb->ctime;
1177 mddev->utime = sb->utime;
1178 mddev->level = sb->level;
d9d166c2 1179 mddev->clevel[0] = 0;
1da177e4
LT		 1180 mddev->layout = sb->layout;
1181 mddev->raid_disks = sb->raid_disks;
27a7b260 1182 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1183 mddev->events = ev1;
c3d9714e
N		 1184 mddev->bitmap_info.offset = 0;
1185 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1186
f6705578
N		 1187 if (mddev->minor_version >= 91) {
1188 mddev->reshape_position = sb->reshape_position;
1189 mddev->delta_disks = sb->delta_disks;
1190 mddev->new_level = sb->new_level;
1191 mddev->new_layout = sb->new_layout;
664e7c41 1192 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1193 } else {
1194 mddev->reshape_position = MaxSector;
1195 mddev->delta_disks = 0;
1196 mddev->new_level = mddev->level;
1197 mddev->new_layout = mddev->layout;
664e7c41 1198 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1199 }
1200
1da177e4
LT		 1201 if (sb->state & (1<<MD_SB_CLEAN))
1202 mddev->recovery_cp = MaxSector;
1203 else {
1204 if (sb->events_hi == sb->cp_events_hi &&
1205 sb->events_lo == sb->cp_events_lo) {
1206 mddev->recovery_cp = sb->recovery_cp;
1207 } else
1208 mddev->recovery_cp = 0;
1209 }
1210
1211 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1212 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1213 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1214 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1215
1216 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1217
1218 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1219 mddev->bitmap_info.file == NULL)
1220 mddev->bitmap_info.offset =
1221 mddev->bitmap_info.default_offset;
a654b9d8 1222
41158c7e 1223 } else if (mddev->pers == NULL) {
be6800a7
N		 1224 /* Insist on good event counter while assembling, except
1225 * for spares (which don't need an event count) */
1da177e4 1226 ++ev1;
be6800a7
N		 1227 if (sb->disks[rdev->desc_nr].state & (
1228 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1229 if (ev1 < mddev->events)
1230 return -EINVAL;
41158c7e
N		 1231 } else if (mddev->bitmap) {
1232 /* if adding to array with a bitmap, then we can accept an
1233 * older device ... but not too old.
1234 */
41158c7e
N		 1235 if (ev1 < mddev->bitmap->events_cleared)
1236 return 0;
07d84d10
N		 1237 } else {
1238 if (ev1 < mddev->events)
1239 /* just a hot-add of a new device, leave raid_disk at -1 */
1240 return 0;
1241 }
41158c7e 1242
1da177e4 1243 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1244 desc = sb->disks + rdev->desc_nr;
1245
1246 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1247 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1248 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1249 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1250 set_bit(In_sync, &rdev->flags);
1da177e4 1251 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1252 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1253 /* active but not in sync implies recovery up to
1254 * reshape position. We don't know exactly where
1255 * that is, so set to zero for now */
1256 if (mddev->minor_version >= 91) {
1257 rdev->recovery_offset = 0;
1258 rdev->raid_disk = desc->raid_disk;
1259 }
1da177e4 1260 }
8ddf9efe
N		 1261 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1262 set_bit(WriteMostly, &rdev->flags);
41158c7e 1263 } else /* MULTIPATH are always insync */
b2d444d7 1264 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1265 return 0;
1266}
1267
1268/*
1269 * sync_super for 0.90.0
1270 */
fd01b88c 1271static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1272{
1273 mdp_super_t *sb;
3cb03002 1274 struct md_rdev *rdev2;
1da177e4 1275 int next_spare = mddev->raid_disks;
19133a42 1276
1da177e4
LT		 1277
1278 /* make rdev->sb match mddev data..
1279 *
1280 * 1/ zero out disks
1281 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1282 * 3/ any empty disks < next_spare become removed
1283 *
1284 * disks[0] gets initialised to REMOVED because
1285 * we cannot be sure from other fields if it has
1286 * been initialised or not.
1287 */
1288 int i;
1289 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1290
61181565
N		 1291 rdev->sb_size = MD_SB_BYTES;
1292
65a06f06 1293 sb = page_address(rdev->sb_page);
1da177e4
LT		 1294
1295 memset(sb, 0, sizeof(*sb));
1296
1297 sb->md_magic = MD_SB_MAGIC;
1298 sb->major_version = mddev->major_version;
1da177e4
LT		 1299 sb->patch_version = mddev->patch_version;
1300 sb->gvalid_words = 0; /* ignored */
1301 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1302 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1303 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1304 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1305
1306 sb->ctime = mddev->ctime;
1307 sb->level = mddev->level;
58c0fed4 1308 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1309 sb->raid_disks = mddev->raid_disks;
1310 sb->md_minor = mddev->md_minor;
e691063a 1311 sb->not_persistent = 0;
1da177e4
LT		 1312 sb->utime = mddev->utime;
1313 sb->state = 0;
1314 sb->events_hi = (mddev->events>>32);
1315 sb->events_lo = (u32)mddev->events;
1316
f6705578
N		 1317 if (mddev->reshape_position == MaxSector)
1318 sb->minor_version = 90;
1319 else {
1320 sb->minor_version = 91;
1321 sb->reshape_position = mddev->reshape_position;
1322 sb->new_level = mddev->new_level;
1323 sb->delta_disks = mddev->delta_disks;
1324 sb->new_layout = mddev->new_layout;
664e7c41 1325 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1326 }
1327 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1328 if (mddev->in_sync)
1329 {
1330 sb->recovery_cp = mddev->recovery_cp;
1331 sb->cp_events_hi = (mddev->events>>32);
1332 sb->cp_events_lo = (u32)mddev->events;
1333 if (mddev->recovery_cp == MaxSector)
1334 sb->state = (1<< MD_SB_CLEAN);
1335 } else
1336 sb->recovery_cp = 0;
1337
1338 sb->layout = mddev->layout;
9d8f0363 1339 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1340
c3d9714e 1341 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1342 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1343
1da177e4 1344 sb->disks[0].state = (1<<MD_DISK_REMOVED);
dafb20fa 1345 rdev_for_each(rdev2, mddev) {
1da177e4 1346 mdp_disk_t *d;
86e6ffdd 1347 int desc_nr;
0261cd9f
N		 1348 int is_active = test_bit(In_sync, &rdev2->flags);
1349
1350 if (rdev2->raid_disk >= 0 &&
1351 sb->minor_version >= 91)
1352 /* we have nowhere to store the recovery_offset,
1353 * but if it is not below the reshape_position,
1354 * we can piggy-back on that.
1355 */
1356 is_active = 1;
1357 if (rdev2->raid_disk < 0 ||
1358 test_bit(Faulty, &rdev2->flags))
1359 is_active = 0;
1360 if (is_active)
86e6ffdd 1361 desc_nr = rdev2->raid_disk;
1da177e4 1362 else
86e6ffdd 1363 desc_nr = next_spare++;
19133a42 1364 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1365 d = &sb->disks[rdev2->desc_nr];
1366 nr_disks++;
1367 d->number = rdev2->desc_nr;
1368 d->major = MAJOR(rdev2->bdev->bd_dev);
1369 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1370 if (is_active)
1da177e4
LT		 1371 d->raid_disk = rdev2->raid_disk;
1372 else
1373 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1374 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1375 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1376 else if (is_active) {
1da177e4 1377 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1378 if (test_bit(In_sync, &rdev2->flags))
1379 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1380 active++;
1381 working++;
1382 } else {
1383 d->state = 0;
1384 spare++;
1385 working++;
1386 }
8ddf9efe
N		 1387 if (test_bit(WriteMostly, &rdev2->flags))
1388 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1389 }
1da177e4
LT		 1390 /* now set the "removed" and "faulty" bits on any missing devices */
1391 for (i=0 ; i < mddev->raid_disks ; i++) {
1392 mdp_disk_t *d = &sb->disks[i];
1393 if (d->state == 0 && d->number == 0) {
1394 d->number = i;
1395 d->raid_disk = i;
1396 d->state = (1<<MD_DISK_REMOVED);
1397 d->state |= (1<<MD_DISK_FAULTY);
1398 failed++;
1399 }
1400 }
1401 sb->nr_disks = nr_disks;
1402 sb->active_disks = active;
1403 sb->working_disks = working;
1404 sb->failed_disks = failed;
1405 sb->spare_disks = spare;
1406
1407 sb->this_disk = sb->disks[rdev->desc_nr];
1408 sb->sb_csum = calc_sb_csum(sb);
1409}
1410
0cd17fec
CW		 1411/*
1412 * rdev_size_change for 0.90.0
1413 */
1414static unsigned long long
3cb03002 1415super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1416{
58c0fed4 1417 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1418 return 0; /* component must fit device */
c3d9714e 1419 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1420 return 0; /* can't move bitmap */
57b2caa3 1421 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1422 if (!num_sectors || num_sectors > rdev->sb_start)
1423 num_sectors = rdev->sb_start;
27a7b260
N		 1424 /* Limit to 4TB as metadata cannot record more than that.
1425 * 4TB == 2^32 KB, or 2*2^32 sectors.
1426 */
1427 if (num_sectors >= (2ULL << 32))
1428 num_sectors = (2ULL << 32) - 2;
0f420358 1429 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1430 rdev->sb_page);
1431 md_super_wait(rdev->mddev);
c26a44ed 1432 return num_sectors;
0cd17fec
CW		 1433}
1434
1435
1da177e4
LT		 1436/*
1437 * version 1 superblock
1438 */
1439
1c05b4bc 1440static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1441{
1c05b4bc
N		 1442 __le32 disk_csum;
1443 u32 csum;
1da177e4
LT		 1444 unsigned long long newcsum;
1445 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1446 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1447 int i;
1448
1449 disk_csum = sb->sb_csum;
1450 sb->sb_csum = 0;
1451 newcsum = 0;
1452 for (i=0; size>=4; size -= 4 )
1453 newcsum += le32_to_cpu(*isuper++);
1454
1455 if (size == 2)
1c05b4bc 1456 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1457
1458 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1459 sb->sb_csum = disk_csum;
1460 return cpu_to_le32(csum);
1461}
1462
2699b672
N		 1463static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1464 int acknowledged);
3cb03002 1465static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1466{
1467 struct mdp_superblock_1 *sb;
1468 int ret;
0f420358 1469 sector_t sb_start;
1da177e4 1470 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1471 int bmask;
1da177e4
LT		 1472
1473 /*
0f420358 1474 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1475 * It is always aligned to a 4K boundary and
1476 * depeding on minor_version, it can be:
1477 * 0: At least 8K, but less than 12K, from end of device
1478 * 1: At start of device
1479 * 2: 4K from start of device.
1480 */
1481 switch(minor_version) {
1482 case 0:
77304d2a 1483 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1484 sb_start -= 8*2;
1485 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1486 break;
1487 case 1:
0f420358 1488 sb_start = 0;
1da177e4
LT		 1489 break;
1490 case 2:
0f420358 1491 sb_start = 8;
1da177e4
LT		 1492 break;
1493 default:
1494 return -EINVAL;
1495 }
0f420358 1496 rdev->sb_start = sb_start;
1da177e4 1497
0002b271
N		 1498 /* superblock is rarely larger than 1K, but it can be larger,
1499 * and it is safe to read 4k, so we do that
1500 */
1501 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1502 if (ret) return ret;
1503
1504
65a06f06 1505 sb = page_address(rdev->sb_page);
1da177e4
LT		 1506
1507 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1508 sb->major_version != cpu_to_le32(1) ||
1509 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1510 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1511 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1512 return -EINVAL;
1513
1514 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1515 printk("md: invalid superblock checksum on %s\n",
1516 bdevname(rdev->bdev,b));
1517 return -EINVAL;
1518 }
1519 if (le64_to_cpu(sb->data_size) < 10) {
1520 printk("md: data_size too small on %s\n",
1521 bdevname(rdev->bdev,b));
1522 return -EINVAL;
1523 }
e11e93fa 1524
1da177e4
LT		 1525 rdev->preferred_minor = 0xffff;
1526 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1527 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1528
0002b271 1529 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1530 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1531 if (rdev->sb_size & bmask)
a1801f85
N		 1532 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1533
1534 if (minor_version
0f420358 1535 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1536 return -EINVAL;
0002b271 1537
31b65a0d
N		 1538 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1539 rdev->desc_nr = -1;
1540 else
1541 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1542
2699b672
N		 1543 if (!rdev->bb_page) {
1544 rdev->bb_page = alloc_page(GFP_KERNEL);
1545 if (!rdev->bb_page)
1546 return -ENOMEM;
1547 }
1548 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1549 rdev->badblocks.count == 0) {
1550 /* need to load the bad block list.
1551 * Currently we limit it to one page.
1552 */
1553 s32 offset;
1554 sector_t bb_sector;
1555 u64 *bbp;
1556 int i;
1557 int sectors = le16_to_cpu(sb->bblog_size);
1558 if (sectors > (PAGE_SIZE / 512))
1559 return -EINVAL;
1560 offset = le32_to_cpu(sb->bblog_offset);
1561 if (offset == 0)
1562 return -EINVAL;
1563 bb_sector = (long long)offset;
1564 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1565 rdev->bb_page, READ, true))
1566 return -EIO;
1567 bbp = (u64 *)page_address(rdev->bb_page);
1568 rdev->badblocks.shift = sb->bblog_shift;
1569 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1570 u64 bb = le64_to_cpu(*bbp);
1571 int count = bb & (0x3ff);
1572 u64 sector = bb >> 10;
1573 sector <<= sb->bblog_shift;
1574 count <<= sb->bblog_shift;
1575 if (bb + 1 == 0)
1576 break;
1577 if (md_set_badblocks(&rdev->badblocks,
1578 sector, count, 1) == 0)
1579 return -EINVAL;
1580 }
1581 } else if (sb->bblog_offset == 0)
1582 rdev->badblocks.shift = -1;
1583
9a7b2b0f 1584 if (!refdev) {
8ed75463 1585 ret = 1;
9a7b2b0f 1586 } else {
1da177e4 1587 __u64 ev1, ev2;
65a06f06 1588 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1589
1590 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1591 sb->level != refsb->level ||
1592 sb->layout != refsb->layout ||
1593 sb->chunksize != refsb->chunksize) {
1594 printk(KERN_WARNING "md: %s has strangely different"
1595 " superblock to %s\n",
1596 bdevname(rdev->bdev,b),
1597 bdevname(refdev->bdev,b2));
1598 return -EINVAL;
1599 }
1600 ev1 = le64_to_cpu(sb->events);
1601 ev2 = le64_to_cpu(refsb->events);
1602
1603 if (ev1 > ev2)
8ed75463
N		 1604 ret = 1;
1605 else
1606 ret = 0;
1da177e4 1607 }
a1801f85 1608 if (minor_version)
77304d2a 1609 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1610 le64_to_cpu(sb->data_offset);
1da177e4 1611 else
dd8ac336
AN		 1612 rdev->sectors = rdev->sb_start;
1613 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1614 return -EINVAL;
dd8ac336 1615 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1616 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1617 return -EINVAL;
8ed75463 1618 return ret;
1da177e4
LT		 1619}
1620
fd01b88c 1621static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1622{
65a06f06 1623 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1624 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1625
41158c7e 1626 rdev->raid_disk = -1;
c5d79adb
N		 1627 clear_bit(Faulty, &rdev->flags);
1628 clear_bit(In_sync, &rdev->flags);
1629 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1630
1da177e4
LT		 1631 if (mddev->raid_disks == 0) {
1632 mddev->major_version = 1;
1633 mddev->patch_version = 0;
e691063a 1634 mddev->external = 0;
9d8f0363 1635 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1636 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1637 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1638 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1639 mddev->clevel[0] = 0;
1da177e4
LT		 1640 mddev->layout = le32_to_cpu(sb->layout);
1641 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1642 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1643 mddev->events = ev1;
c3d9714e
N		 1644 mddev->bitmap_info.offset = 0;
1645 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1646
1647 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1648 memcpy(mddev->uuid, sb->set_uuid, 16);
1649
1650 mddev->max_disks = (4096-256)/2;
a654b9d8 1651
71c0805c 1652 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1653 mddev->bitmap_info.file == NULL )
1654 mddev->bitmap_info.offset =
1655 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1656
f6705578
N		 1657 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1658 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1659 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1660 mddev->new_level = le32_to_cpu(sb->new_level);
1661 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1662 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1663 } else {
1664 mddev->reshape_position = MaxSector;
1665 mddev->delta_disks = 0;
1666 mddev->new_level = mddev->level;
1667 mddev->new_layout = mddev->layout;
664e7c41 1668 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1669 }
1670
41158c7e 1671 } else if (mddev->pers == NULL) {
be6800a7
N		 1672 /* Insist of good event counter while assembling, except for
1673 * spares (which don't need an event count) */
1da177e4 1674 ++ev1;
be6800a7
N		 1675 if (rdev->desc_nr >= 0 &&
1676 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1677 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1678 if (ev1 < mddev->events)
1679 return -EINVAL;
41158c7e
N		 1680 } else if (mddev->bitmap) {
1681 /* If adding to array with a bitmap, then we can accept an
1682 * older device, but not too old.
1683 */
41158c7e
N		 1684 if (ev1 < mddev->bitmap->events_cleared)
1685 return 0;
07d84d10
N		 1686 } else {
1687 if (ev1 < mddev->events)
1688 /* just a hot-add of a new device, leave raid_disk at -1 */
1689 return 0;
1690 }
1da177e4
LT		 1691 if (mddev->level != LEVEL_MULTIPATH) {
1692 int role;
3673f305
N		 1693 if (rdev->desc_nr < 0 ||
1694 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1695 role = 0xffff;
1696 rdev->desc_nr = -1;
1697 } else
1698 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1699 switch(role) {
1700 case 0xffff: /* spare */
1da177e4
LT		 1701 break;
1702 case 0xfffe: /* faulty */
b2d444d7 1703 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1704 break;
1705 default:
5fd6c1dc
N		 1706 if ((le32_to_cpu(sb->feature_map) &
1707 MD_FEATURE_RECOVERY_OFFSET))
1708 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1709 else
1710 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1711 rdev->raid_disk = role;
1712 break;
1713 }
8ddf9efe
N		 1714 if (sb->devflags & WriteMostly1)
1715 set_bit(WriteMostly, &rdev->flags);
2d78f8c4
N		 1716 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1717 set_bit(Replacement, &rdev->flags);
41158c7e 1718 } else /* MULTIPATH are always insync */
b2d444d7 1719 set_bit(In_sync, &rdev->flags);
41158c7e 1720
1da177e4
LT		 1721 return 0;
1722}
1723
fd01b88c 1724static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1725{
1726 struct mdp_superblock_1 *sb;
3cb03002 1727 struct md_rdev *rdev2;
1da177e4
LT		 1728 int max_dev, i;
1729 /* make rdev->sb match mddev and rdev data. */
1730
65a06f06 1731 sb = page_address(rdev->sb_page);
1da177e4
LT		 1732
1733 sb->feature_map = 0;
1734 sb->pad0 = 0;
5fd6c1dc 1735 sb->recovery_offset = cpu_to_le64(0);
1da177e4 1736 memset(sb->pad1, 0, sizeof(sb->pad1));
1da177e4
LT		 1737 memset(sb->pad3, 0, sizeof(sb->pad3));
1738
1739 sb->utime = cpu_to_le64((__u64)mddev->utime);
1740 sb->events = cpu_to_le64(mddev->events);
1741 if (mddev->in_sync)
1742 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1743 else
1744 sb->resync_offset = cpu_to_le64(0);
1745
1c05b4bc 1746 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1747
f0ca340c 1748 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1749 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1750 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1751 sb->level = cpu_to_le32(mddev->level);
1752 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1753
aeb9b211
N		 1754 if (test_bit(WriteMostly, &rdev->flags))
1755 sb->devflags |= WriteMostly1;
1756 else
1757 sb->devflags &= ~WriteMostly1;
1758
c3d9714e
N		 1759 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1760 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1761 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1762 }
5fd6c1dc
N		 1763
1764 if (rdev->raid_disk >= 0 &&
97e4f42d 1765 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1766 sb->feature_map |=
1767 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1768 sb->recovery_offset =
1769 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc 1770 }
2d78f8c4
N		 1771 if (test_bit(Replacement, &rdev->flags))
1772 sb->feature_map |=
1773 cpu_to_le32(MD_FEATURE_REPLACEMENT);
5fd6c1dc 1774
f6705578
N		 1775 if (mddev->reshape_position != MaxSector) {
1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1777 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1778 sb->new_layout = cpu_to_le32(mddev->new_layout);
1779 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1780 sb->new_level = cpu_to_le32(mddev->new_level);
664e7c41 1781 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
f6705578 1782 }
a654b9d8 1783
2699b672
N		 1784 if (rdev->badblocks.count == 0)
1785 /* Nothing to do for bad blocks*/ ;
1786 else if (sb->bblog_offset == 0)
1787 /* Cannot record bad blocks on this device */
1788 md_error(mddev, rdev);
1789 else {
1790 struct badblocks *bb = &rdev->badblocks;
1791 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1792 u64 *p = bb->page;
1793 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1794 if (bb->changed) {
1795 unsigned seq;
1796
1797retry:
1798 seq = read_seqbegin(&bb->lock);
1799
1800 memset(bbp, 0xff, PAGE_SIZE);
1801
1802 for (i = 0 ; i < bb->count ; i++) {
1803 u64 internal_bb = *p++;
1804 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1805 | BB_LEN(internal_bb));
1806 *bbp++ = cpu_to_le64(store_bb);
1807 }
1808 if (read_seqretry(&bb->lock, seq))
1809 goto retry;
1810
1811 bb->sector = (rdev->sb_start +
1812 (int)le32_to_cpu(sb->bblog_offset));
1813 bb->size = le16_to_cpu(sb->bblog_size);
1814 bb->changed = 0;
1815 }
1816 }
1817
1da177e4 1818 max_dev = 0;
dafb20fa 1819 rdev_for_each(rdev2, mddev)
1da177e4
LT		 1820 if (rdev2->desc_nr+1 > max_dev)
1821 max_dev = rdev2->desc_nr+1;
a778b73f 1822
70471daf
N		 1823 if (max_dev > le32_to_cpu(sb->max_dev)) {
1824 int bmask;
a778b73f 1825 sb->max_dev = cpu_to_le32(max_dev);
70471daf
N		 1826 rdev->sb_size = max_dev * 2 + 256;
1827 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1828 if (rdev->sb_size & bmask)
1829 rdev->sb_size = (rdev->sb_size | bmask) + 1;
ddcf3522
N		 1830 } else
1831 max_dev = le32_to_cpu(sb->max_dev);
1832
1da177e4
LT		 1833 for (i=0; i<max_dev;i++)
1834 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1835
dafb20fa 1836 rdev_for_each(rdev2, mddev) {
1da177e4 1837 i = rdev2->desc_nr;
b2d444d7 1838 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1839 sb->dev_roles[i] = cpu_to_le16(0xfffe);
b2d444d7 1840 else if (test_bit(In_sync, &rdev2->flags))
1da177e4 1841 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
93be75ff 1842 else if (rdev2->raid_disk >= 0)
5fd6c1dc 1843 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1da177e4
LT		 1844 else
1845 sb->dev_roles[i] = cpu_to_le16(0xffff);
1846 }
1847
1da177e4
LT		 1848 sb->sb_csum = calc_sb_1_csum(sb);
1849}
1850
0cd17fec 1851static unsigned long long
3cb03002 1852super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec
CW		 1853{
1854 struct mdp_superblock_1 *sb;
15f4a5fd 1855 sector_t max_sectors;
58c0fed4 1856 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1857 return 0; /* component must fit device */
0f420358 1858 if (rdev->sb_start < rdev->data_offset) {
0cd17fec 1859 /* minor versions 1 and 2; superblock before data */
77304d2a 1860 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
15f4a5fd
AN		 1861 max_sectors -= rdev->data_offset;
1862 if (!num_sectors || num_sectors > max_sectors)
1863 num_sectors = max_sectors;
c3d9714e 1864 } else if (rdev->mddev->bitmap_info.offset) {
0cd17fec
CW		 1865 /* minor version 0 with bitmap we can't move */
1866 return 0;
1867 } else {
1868 /* minor version 0; superblock after data */
0f420358 1869 sector_t sb_start;
77304d2a 1870 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
0f420358 1871 sb_start &= ~(sector_t)(4*2 - 1);
dd8ac336 1872 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
15f4a5fd
AN		 1873 if (!num_sectors || num_sectors > max_sectors)
1874 num_sectors = max_sectors;
0f420358 1875 rdev->sb_start = sb_start;
0cd17fec 1876 }
65a06f06 1877 sb = page_address(rdev->sb_page);
15f4a5fd 1878 sb->data_size = cpu_to_le64(num_sectors);
0f420358 1879 sb->super_offset = rdev->sb_start;
0cd17fec 1880 sb->sb_csum = calc_sb_1_csum(sb);
0f420358 1881 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1882 rdev->sb_page);
1883 md_super_wait(rdev->mddev);
c26a44ed 1884 return num_sectors;
0cd17fec 1885}
1da177e4 1886
75c96f85 1887static struct super_type super_types[] = {
1da177e4
LT		 1888 [0] = {
1889 .name = "0.90.0",
1890 .owner = THIS_MODULE,
0cd17fec
CW		 1891 .load_super = super_90_load,
1892 .validate_super = super_90_validate,
1893 .sync_super = super_90_sync,
1894 .rdev_size_change = super_90_rdev_size_change,
1da177e4
LT		 1895 },
1896 [1] = {
1897 .name = "md-1",
1898 .owner = THIS_MODULE,
0cd17fec
CW		 1899 .load_super = super_1_load,
1900 .validate_super = super_1_validate,
1901 .sync_super = super_1_sync,
1902 .rdev_size_change = super_1_rdev_size_change,
1da177e4
LT		 1903 },
1904};
1da177e4 1905
fd01b88c 1906static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
076f968b
JB		 1907{
1908 if (mddev->sync_super) {
1909 mddev->sync_super(mddev, rdev);
1910 return;
1911 }
1912
1913 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1914
1915 super_types[mddev->major_version].sync_super(mddev, rdev);
1916}
1917
fd01b88c 1918static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1da177e4 1919{
3cb03002 1920 struct md_rdev *rdev, *rdev2;
1da177e4 1921
4b80991c
N		 1922 rcu_read_lock();
1923 rdev_for_each_rcu(rdev, mddev1)
1924 rdev_for_each_rcu(rdev2, mddev2)
7dd5e7c3 1925 if (rdev->bdev->bd_contains ==
4b80991c
N		 1926 rdev2->bdev->bd_contains) {
1927 rcu_read_unlock();
7dd5e7c3 1928 return 1;
4b80991c
N		 1929 }
1930 rcu_read_unlock();
1da177e4
LT		 1931 return 0;
1932}
1933
1934static LIST_HEAD(pending_raid_disks);
1935
ac5e7113
AN		 1936/*
1937 * Try to register data integrity profile for an mddev
1938 *
1939 * This is called when an array is started and after a disk has been kicked
1940 * from the array. It only succeeds if all working and active component devices
1941 * are integrity capable with matching profiles.
1942 */
fd01b88c 1943int md_integrity_register(struct mddev *mddev)
ac5e7113 1944{
3cb03002 1945 struct md_rdev *rdev, *reference = NULL;
ac5e7113
AN		 1946
1947 if (list_empty(&mddev->disks))
1948 return 0; /* nothing to do */
629acb6a
JB		 1949 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1950 return 0; /* shouldn't register, or already is */
dafb20fa 1951 rdev_for_each(rdev, mddev) {
ac5e7113
AN		 1952 /* skip spares and non-functional disks */
1953 if (test_bit(Faulty, &rdev->flags))
1954 continue;
1955 if (rdev->raid_disk < 0)
1956 continue;
ac5e7113
AN		 1957 if (!reference) {
1958 /* Use the first rdev as the reference */
1959 reference = rdev;
1960 continue;
1961 }
1962 /* does this rdev's profile match the reference profile? */
1963 if (blk_integrity_compare(reference->bdev->bd_disk,
1964 rdev->bdev->bd_disk) < 0)
1965 return -EINVAL;
1966 }
89078d57
MP		 1967 if (!reference || !bdev_get_integrity(reference->bdev))
1968 return 0;
ac5e7113
AN		 1969 /*
1970 * All component devices are integrity capable and have matching
1971 * profiles, register the common profile for the md device.
1972 */
1973 if (blk_integrity_register(mddev->gendisk,
1974 bdev_get_integrity(reference->bdev)) != 0) {
1975 printk(KERN_ERR "md: failed to register integrity for %s\n",
1976 mdname(mddev));
1977 return -EINVAL;
1978 }
a91a2785
MP		 1979 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1980 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1981 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1982 mdname(mddev));
1983 return -EINVAL;
1984 }
ac5e7113
AN		 1985 return 0;
1986}
1987EXPORT_SYMBOL(md_integrity_register);
1988
1989/* Disable data integrity if non-capable/non-matching disk is being added */
fd01b88c 1990void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
3f9d99c1 1991{
3f9d99c1 1992 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
ac5e7113 1993 struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
3f9d99c1 1994
ac5e7113 1995 if (!bi_mddev) /* nothing to do */
3f9d99c1 1996 return;
ac5e7113 1997 if (rdev->raid_disk < 0) /* skip spares */
3f9d99c1 1998 return;
ac5e7113
AN		 1999 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2000 rdev->bdev->bd_disk) >= 0)
2001 return;
2002 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2003 blk_integrity_unregister(mddev->gendisk);
3f9d99c1 2004}
ac5e7113 2005EXPORT_SYMBOL(md_integrity_add_rdev);
3f9d99c1 2006
fd01b88c 2007static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
1da177e4 2008{
7dd5e7c3 2009 char b[BDEVNAME_SIZE];
f637b9f9 2010 struct kobject *ko;
1edf80d3 2011 char *s;
5e55e2f5 2012 int err;
1da177e4
LT		 2013
2014 if (rdev->mddev) {
2015 MD_BUG();
2016 return -EINVAL;
2017 }
11e2ede0
DW		 2018
2019 /* prevent duplicates */
2020 if (find_rdev(mddev, rdev->bdev->bd_dev))
2021 return -EEXIST;
2022
dd8ac336
AN		 2023 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2024 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2025 rdev->sectors < mddev->dev_sectors)) {
a778b73f
N		 2026 if (mddev->pers) {
2027 /* Cannot change size, so fail
2028 * If mddev->level <= 0, then we don't care
2029 * about aligning sizes (e.g. linear)
2030 */
2031 if (mddev->level > 0)
2032 return -ENOSPC;
2033 } else
dd8ac336 2034 mddev->dev_sectors = rdev->sectors;
2bf071bf 2035 }
1da177e4
LT		 2036
2037 /* Verify rdev->desc_nr is unique.
2038 * If it is -1, assign a free number, else
2039 * check number is not in use
2040 */
2041 if (rdev->desc_nr < 0) {
2042 int choice = 0;
2043 if (mddev->pers) choice = mddev->raid_disks;
2044 while (find_rdev_nr(mddev, choice))
2045 choice++;
2046 rdev->desc_nr = choice;
2047 } else {
2048 if (find_rdev_nr(mddev, rdev->desc_nr))
2049 return -EBUSY;
2050 }
de01dfad
N		 2051 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2052 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2053 mdname(mddev), mddev->max_disks);
2054 return -EBUSY;
2055 }
19133a42 2056 bdevname(rdev->bdev,b);
649316b2 2057 while ( (s=strchr(b, '/')) != NULL)
1edf80d3 2058 *s = '!';
649316b2 2059
1da177e4 2060 rdev->mddev = mddev;
19133a42 2061 printk(KERN_INFO "md: bind<%s>\n", b);
86e6ffdd 2062
b2d6db58 2063 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
5e55e2f5 2064 goto fail;
86e6ffdd 2065
0762b8bd 2066 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
00bcb4ac
N		 2067 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2068 /* failure here is OK */;
2069 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
3c0ee63a 2070
4b80991c 2071 list_add_rcu(&rdev->same_set, &mddev->disks);
e09b457b 2072 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
4044ba58
N		 2073
2074 /* May as well allow recovery to be retried once */
5389042f 2075 mddev->recovery_disabled++;
3f9d99c1 2076
1da177e4 2077 return 0;
5e55e2f5
N		 2078
2079 fail:
2080 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2081 b, mdname(mddev));
2082 return err;
1da177e4
LT		 2083}
2084
177a99b2 2085static void md_delayed_delete(struct work_struct *ws)
5792a285 2086{
3cb03002 2087 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
5792a285 2088 kobject_del(&rdev->kobj);
177a99b2 2089 kobject_put(&rdev->kobj);
5792a285
N		 2090}
2091
3cb03002 2092static void unbind_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2093{
2094 char b[BDEVNAME_SIZE];
2095 if (!rdev->mddev) {
2096 MD_BUG();
2097 return;
2098 }
49731baa 2099 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
4b80991c 2100 list_del_rcu(&rdev->same_set);
1da177e4
LT		 2101 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2102 rdev->mddev = NULL;
86e6ffdd 2103 sysfs_remove_link(&rdev->kobj, "block");
3c0ee63a
N		 2104 sysfs_put(rdev->sysfs_state);
2105 rdev->sysfs_state = NULL;
2230dfe4
N		 2106 kfree(rdev->badblocks.page);
2107 rdev->badblocks.count = 0;
2108 rdev->badblocks.page = NULL;
5792a285 2109 /* We need to delay this, otherwise we can deadlock when
4b80991c
N		 2110 * writing to 'remove' to "dev/state". We also need
2111 * to delay it due to rcu usage.
5792a285 2112 */
4b80991c 2113 synchronize_rcu();
177a99b2
N		 2114 INIT_WORK(&rdev->del_work, md_delayed_delete);
2115 kobject_get(&rdev->kobj);
e804ac78 2116 queue_work(md_misc_wq, &rdev->del_work);
1da177e4
LT		 2117}
2118
2119/*
2120 * prevent the device from being mounted, repartitioned or
2121 * otherwise reused by a RAID array (or any other kernel
2122 * subsystem), by bd_claiming the device.
2123 */
3cb03002 2124static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
1da177e4
LT		 2125{
2126 int err = 0;
2127 struct block_device *bdev;
2128 char b[BDEVNAME_SIZE];
2129
d4d77629 2130 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
3cb03002 2131 shared ? (struct md_rdev *)lock_rdev : rdev);
1da177e4
LT		 2132 if (IS_ERR(bdev)) {
2133 printk(KERN_ERR "md: could not open %s.\n",
2134 __bdevname(dev, b));
2135 return PTR_ERR(bdev);
2136 }
1da177e4
LT		 2137 rdev->bdev = bdev;
2138 return err;
2139}
2140
3cb03002 2141static void unlock_rdev(struct md_rdev *rdev)
1da177e4
LT		 2142{
2143 struct block_device *bdev = rdev->bdev;
2144 rdev->bdev = NULL;
2145 if (!bdev)
2146 MD_BUG();
e525fd89 2147 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
1da177e4
LT		 2148}
2149
2150void md_autodetect_dev(dev_t dev);
2151
3cb03002 2152static void export_rdev(struct md_rdev * rdev)
1da177e4
LT		 2153{
2154 char b[BDEVNAME_SIZE];
2155 printk(KERN_INFO "md: export_rdev(%s)\n",
2156 bdevname(rdev->bdev,b));
2157 if (rdev->mddev)
2158 MD_BUG();
2159 free_disk_sb(rdev);
1da177e4 2160#ifndef MODULE
d0fae18f
N		 2161 if (test_bit(AutoDetected, &rdev->flags))
2162 md_autodetect_dev(rdev->bdev->bd_dev);
1da177e4
LT		 2163#endif
2164 unlock_rdev(rdev);
86e6ffdd 2165 kobject_put(&rdev->kobj);
1da177e4
LT		 2166}
2167
3cb03002 2168static void kick_rdev_from_array(struct md_rdev * rdev)
1da177e4
LT		 2169{
2170 unbind_rdev_from_array(rdev);
2171 export_rdev(rdev);
2172}
2173
fd01b88c 2174static void export_array(struct mddev *mddev)
1da177e4 2175{
3cb03002 2176 struct md_rdev *rdev, *tmp;
1da177e4 2177
dafb20fa 2178 rdev_for_each_safe(rdev, tmp, mddev) {
1da177e4
LT		 2179 if (!rdev->mddev) {
2180 MD_BUG();
2181 continue;
2182 }
2183 kick_rdev_from_array(rdev);
2184 }
2185 if (!list_empty(&mddev->disks))
2186 MD_BUG();
2187 mddev->raid_disks = 0;
2188 mddev->major_version = 0;
2189}
2190
2191static void print_desc(mdp_disk_t *desc)
2192{
2193 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2194 desc->major,desc->minor,desc->raid_disk,desc->state);
2195}
2196
cd2ac932 2197static void print_sb_90(mdp_super_t *sb)
1da177e4
LT		 2198{
2199 int i;
2200
2201 printk(KERN_INFO
2202 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2203 sb->major_version, sb->minor_version, sb->patch_version,
2204 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2205 sb->ctime);
2206 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2207 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2208 sb->md_minor, sb->layout, sb->chunk_size);
2209 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2210 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2211 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2212 sb->failed_disks, sb->spare_disks,
2213 sb->sb_csum, (unsigned long)sb->events_lo);
2214
2215 printk(KERN_INFO);
2216 for (i = 0; i < MD_SB_DISKS; i++) {
2217 mdp_disk_t *desc;
2218
2219 desc = sb->disks + i;
2220 if (desc->number || desc->major || desc->minor ||
2221 desc->raid_disk || (desc->state && (desc->state != 4))) {
2222 printk(" D %2d: ", i);
2223 print_desc(desc);
2224 }
2225 }
2226 printk(KERN_INFO "md: THIS: ");
2227 print_desc(&sb->this_disk);
cd2ac932 2228}
1da177e4 2229
cd2ac932
CR		 2230static void print_sb_1(struct mdp_superblock_1 *sb)
2231{
2232 __u8 *uuid;
2233
2234 uuid = sb->set_uuid;
ad361c98 2235 printk(KERN_INFO
7b75c2f8 2236 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
ad361c98 2237 "md: Name: \"%s\" CT:%llu\n",
cd2ac932
CR		 2238 le32_to_cpu(sb->major_version),
2239 le32_to_cpu(sb->feature_map),
7b75c2f8 2240 uuid,
cd2ac932
CR		 2241 sb->set_name,
2242 (unsigned long long)le64_to_cpu(sb->ctime)
2243 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2244
2245 uuid = sb->device_uuid;
ad361c98
JP		 2246 printk(KERN_INFO
2247 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
cd2ac932 2248 " RO:%llu\n"
7b75c2f8 2249 "md: Dev:%08x UUID: %pU\n"
ad361c98
JP		 2250 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2251 "md: (MaxDev:%u) \n",
cd2ac932
CR		 2252 le32_to_cpu(sb->level),
2253 (unsigned long long)le64_to_cpu(sb->size),
2254 le32_to_cpu(sb->raid_disks),
2255 le32_to_cpu(sb->layout),
2256 le32_to_cpu(sb->chunksize),
2257 (unsigned long long)le64_to_cpu(sb->data_offset),
2258 (unsigned long long)le64_to_cpu(sb->data_size),
2259 (unsigned long long)le64_to_cpu(sb->super_offset),
2260 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2261 le32_to_cpu(sb->dev_number),
7b75c2f8 2262 uuid,
cd2ac932
CR		 2263 sb->devflags,
2264 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2265 (unsigned long long)le64_to_cpu(sb->events),
2266 (unsigned long long)le64_to_cpu(sb->resync_offset),
2267 le32_to_cpu(sb->sb_csum),
2268 le32_to_cpu(sb->max_dev)
2269 );
1da177e4
LT		 2270}
2271
3cb03002 2272static void print_rdev(struct md_rdev *rdev, int major_version)
1da177e4
LT		 2273{
2274 char b[BDEVNAME_SIZE];
dd8ac336
AN		 2275 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2276 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
b2d444d7
N		 2277 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2278 rdev->desc_nr);
1da177e4 2279 if (rdev->sb_loaded) {
cd2ac932
CR		 2280 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2281 switch (major_version) {
2282 case 0:
65a06f06 2283 print_sb_90(page_address(rdev->sb_page));
cd2ac932
CR		 2284 break;
2285 case 1:
65a06f06 2286 print_sb_1(page_address(rdev->sb_page));
cd2ac932
CR		 2287 break;
2288 }
1da177e4
LT		 2289 } else
2290 printk(KERN_INFO "md: no rdev superblock!\n");
2291}
2292
5e56341d 2293static void md_print_devices(void)
1da177e4 2294{
159ec1fc 2295 struct list_head *tmp;
3cb03002 2296 struct md_rdev *rdev;
fd01b88c 2297 struct mddev *mddev;
1da177e4
LT		 2298 char b[BDEVNAME_SIZE];
2299
2300 printk("\n");
2301 printk("md: **********************************\n");
2302 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2303 printk("md: **********************************\n");
29ac4aa3 2304 for_each_mddev(mddev, tmp) {
1da177e4 2305
32a7627c
N		 2306 if (mddev->bitmap)
2307 bitmap_print_sb(mddev->bitmap);
2308 else
2309 printk("%s: ", mdname(mddev));
dafb20fa 2310 rdev_for_each(rdev, mddev)
1da177e4
LT		 2311 printk("<%s>", bdevname(rdev->bdev,b));
2312 printk("\n");
2313
dafb20fa 2314 rdev_for_each(rdev, mddev)
cd2ac932 2315 print_rdev(rdev, mddev->major_version);
1da177e4
LT		 2316 }
2317 printk("md: **********************************\n");
2318 printk("\n");
2319}
2320
2321
fd01b88c 2322static void sync_sbs(struct mddev * mddev, int nospares)
1da177e4 2323{
42543769
N		 2324 /* Update each superblock (in-memory image), but
2325 * if we are allowed to, skip spares which already
2326 * have the right event counter, or have one earlier
2327 * (which would mean they aren't being marked as dirty
2328 * with the rest of the array)
2329 */
3cb03002 2330 struct md_rdev *rdev;
dafb20fa 2331 rdev_for_each(rdev, mddev) {
42543769
N		 2332 if (rdev->sb_events == mddev->events ||
2333 (nospares &&
2334 rdev->raid_disk < 0 &&
42543769
N		 2335 rdev->sb_events+1 == mddev->events)) {
2336 /* Don't update this superblock */
2337 rdev->sb_loaded = 2;
2338 } else {
076f968b 2339 sync_super(mddev, rdev);
42543769
N		 2340 rdev->sb_loaded = 1;
2341 }
1da177e4
LT		 2342 }
2343}
2344
fd01b88c 2345static void md_update_sb(struct mddev * mddev, int force_change)
1da177e4 2346{
3cb03002 2347 struct md_rdev *rdev;
06d91a5f 2348 int sync_req;
42543769 2349 int nospares = 0;
2699b672 2350 int any_badblocks_changed = 0;
1da177e4 2351
1da177e4 2352repeat:
3a3a5ddb 2353 /* First make sure individual recovery_offsets are correct */
dafb20fa 2354 rdev_for_each(rdev, mddev) {
3a3a5ddb
N		 2355 if (rdev->raid_disk >= 0 &&
2356 mddev->delta_disks >= 0 &&
2357 !test_bit(In_sync, &rdev->flags) &&
2358 mddev->curr_resync_completed > rdev->recovery_offset)
2359 rdev->recovery_offset = mddev->curr_resync_completed;
2360
2361 }
bd52b746 2362 if (!mddev->persistent) {
070dc6dd 2363 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3a3a5ddb 2364 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
de393cde 2365 if (!mddev->external) {
d97a41dc 2366 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
dafb20fa 2367 rdev_for_each(rdev, mddev) {
de393cde
N		 2368 if (rdev->badblocks.changed) {
2369 md_ack_all_badblocks(&rdev->badblocks);
2370 md_error(mddev, rdev);
2371 }
2372 clear_bit(Blocked, &rdev->flags);
2373 clear_bit(BlockedBadBlocks, &rdev->flags);
2374 wake_up(&rdev->blocked_wait);
2375 }
2376 }
3a3a5ddb
N		 2377 wake_up(&mddev->sb_wait);
2378 return;
2379 }
2380
a9701a30 2381 spin_lock_irq(&mddev->write_lock);
84692195 2382
3a3a5ddb
N		 2383 mddev->utime = get_seconds();
2384
850b2b42
N		 2385 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2386 force_change = 1;
2387 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2388 /* just a clean<-> dirty transition, possibly leave spares alone,
2389 * though if events isn't the right even/odd, we will have to do
2390 * spares after all
2391 */
2392 nospares = 1;
2393 if (force_change)
2394 nospares = 0;
2395 if (mddev->degraded)
84692195
N		 2396 /* If the array is degraded, then skipping spares is both
2397 * dangerous and fairly pointless.
2398 * Dangerous because a device that was removed from the array
2399 * might have a event_count that still looks up-to-date,
2400 * so it can be re-added without a resync.
2401 * Pointless because if there are any spares to skip,
2402 * then a recovery will happen and soon that array won't
2403 * be degraded any more and the spare can go back to sleep then.
2404 */
850b2b42 2405 nospares = 0;
84692195 2406
06d91a5f 2407 sync_req = mddev->in_sync;
42543769
N		 2408
2409 /* If this is just a dirty<->clean transition, and the array is clean
2410 * and 'events' is odd, we can roll back to the previous clean state */
850b2b42 2411 if (nospares
42543769 2412 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
a8707c08
N		 2413 && mddev->can_decrease_events
2414 && mddev->events != 1) {
42543769 2415 mddev->events--;
a8707c08
N		 2416 mddev->can_decrease_events = 0;
2417 } else {
42543769
N		 2418 /* otherwise we have to go forward and ... */
2419 mddev->events ++;
a8707c08 2420 mddev->can_decrease_events = nospares;
42543769 2421 }
1da177e4
LT		 2422
2423 if (!mddev->events) {
2424 /*
2425 * oops, this 64-bit counter should never wrap.
2426 * Either we are in around ~1 trillion A.C., assuming
2427 * 1 reboot per second, or we have a bug:
2428 */
2429 MD_BUG();
2430 mddev->events --;
2431 }
2699b672 2432
dafb20fa 2433 rdev_for_each(rdev, mddev) {
2699b672
N		 2434 if (rdev->badblocks.changed)
2435 any_badblocks_changed++;
de393cde
N		 2436 if (test_bit(Faulty, &rdev->flags))
2437 set_bit(FaultRecorded, &rdev->flags);
2438 }
2699b672 2439
e691063a 2440 sync_sbs(mddev, nospares);
a9701a30 2441 spin_unlock_irq(&mddev->write_lock);
1da177e4 2442
36a4e1fe
N		 2443 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2444 mdname(mddev), mddev->in_sync);
1da177e4 2445
4ad13663 2446 bitmap_update_sb(mddev->bitmap);
dafb20fa 2447 rdev_for_each(rdev, mddev) {
1da177e4 2448 char b[BDEVNAME_SIZE];
36a4e1fe 2449
42543769
N		 2450 if (rdev->sb_loaded != 1)
2451 continue; /* no noise on spare devices */
1da177e4 2452
d70ed2e4
AW		 2453 if (!test_bit(Faulty, &rdev->flags) &&
2454 rdev->saved_raid_disk == -1) {
7bfa19f2 2455 md_super_write(mddev,rdev,
0f420358 2456 rdev->sb_start, rdev->sb_size,
7bfa19f2 2457 rdev->sb_page);
36a4e1fe
N		 2458 pr_debug("md: (write) %s's sb offset: %llu\n",
2459 bdevname(rdev->bdev, b),
2460 (unsigned long long)rdev->sb_start);
42543769 2461 rdev->sb_events = mddev->events;
2699b672
N		 2462 if (rdev->badblocks.size) {
2463 md_super_write(mddev, rdev,
2464 rdev->badblocks.sector,
2465 rdev->badblocks.size << 9,
2466 rdev->bb_page);
2467 rdev->badblocks.size = 0;
2468 }
7bfa19f2 2469
d70ed2e4 2470 } else if (test_bit(Faulty, &rdev->flags))
36a4e1fe
N		 2471 pr_debug("md: %s (skipping faulty)\n",
2472 bdevname(rdev->bdev, b));
d70ed2e4
AW		 2473 else
2474 pr_debug("(skipping incremental s/r ");
2475
7bfa19f2 2476 if (mddev->level == LEVEL_MULTIPATH)
1da177e4
LT		 2477 /* only need to write one superblock... */
2478 break;
2479 }
a9701a30 2480 md_super_wait(mddev);
850b2b42 2481 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
7bfa19f2 2482
a9701a30 2483 spin_lock_irq(&mddev->write_lock);
850b2b42
N		 2484 if (mddev->in_sync != sync_req ||
2485 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
06d91a5f 2486 /* have to write it out again */
a9701a30 2487 spin_unlock_irq(&mddev->write_lock);
06d91a5f
N		 2488 goto repeat;
2489 }
850b2b42 2490 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
a9701a30 2491 spin_unlock_irq(&mddev->write_lock);
3d310eb7 2492 wake_up(&mddev->sb_wait);
acb180b0
N		 2493 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2494 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
06d91a5f 2495
dafb20fa 2496 rdev_for_each(rdev, mddev) {
de393cde
N		 2497 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2498 clear_bit(Blocked, &rdev->flags);
2499
2500 if (any_badblocks_changed)
2699b672 2501 md_ack_all_badblocks(&rdev->badblocks);
de393cde
N		 2502 clear_bit(BlockedBadBlocks, &rdev->flags);
2503 wake_up(&rdev->blocked_wait);
2504 }
1da177e4
LT		 2505}
2506
7f6ce769 2507/* words written to sysfs files may, or may not, be \n terminated.
bce74dac
N		 2508 * We want to accept with case. For this we use cmd_match.
2509 */
2510static int cmd_match(const char *cmd, const char *str)
2511{
2512 /* See if cmd, written into a sysfs file, matches
2513 * str. They must either be the same, or cmd can
2514 * have a trailing newline
2515 */
2516 while (*cmd && *str && *cmd == *str) {
2517 cmd++;
2518 str++;
2519 }
2520 if (*cmd == '\n')
2521 cmd++;
2522 if (*str || *cmd)
2523 return 0;
2524 return 1;
2525}
2526
86e6ffdd
N		 2527struct rdev_sysfs_entry {
2528 struct attribute attr;
3cb03002
N		 2529 ssize_t (*show)(struct md_rdev *, char *);
2530 ssize_t (*store)(struct md_rdev *, const char *, size_t);
86e6ffdd
N		 2531};
2532
2533static ssize_t
3cb03002 2534state_show(struct md_rdev *rdev, char *page)
86e6ffdd
N		 2535{
2536 char *sep = "";
20a49ff6 2537 size_t len = 0;
86e6ffdd 2538
de393cde
N		 2539 if (test_bit(Faulty, &rdev->flags) ||
2540 rdev->badblocks.unacked_exist) {
86e6ffdd
N		 2541 len+= sprintf(page+len, "%sfaulty",sep);
2542 sep = ",";
2543 }
b2d444d7 2544 if (test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2545 len += sprintf(page+len, "%sin_sync",sep);
2546 sep = ",";
2547 }
f655675b
N		 2548 if (test_bit(WriteMostly, &rdev->flags)) {
2549 len += sprintf(page+len, "%swrite_mostly",sep);
2550 sep = ",";
2551 }
de393cde 2552 if (test_bit(Blocked, &rdev->flags) ||
52c64152
N		 2553 (rdev->badblocks.unacked_exist
2554 && !test_bit(Faulty, &rdev->flags))) {
6bfe0b49
DW		 2555 len += sprintf(page+len, "%sblocked", sep);
2556 sep = ",";
2557 }
b2d444d7
N		 2558 if (!test_bit(Faulty, &rdev->flags) &&
2559 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2560 len += sprintf(page+len, "%sspare", sep);
2561 sep = ",";
2562 }
d7a9d443
N		 2563 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2564 len += sprintf(page+len, "%swrite_error", sep);
2565 sep = ",";
2566 }
2d78f8c4
N		 2567 if (test_bit(WantReplacement, &rdev->flags)) {
2568 len += sprintf(page+len, "%swant_replacement", sep);
2569 sep = ",";
2570 }
2571 if (test_bit(Replacement, &rdev->flags)) {
2572 len += sprintf(page+len, "%sreplacement", sep);
2573 sep = ",";
2574 }
2575
86e6ffdd
N		 2576 return len+sprintf(page+len, "\n");
2577}
2578
45dc2de1 2579static ssize_t
3cb03002 2580state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2581{
2582 /* can write
de393cde 2583 * faulty - simulates an error
45dc2de1 2584 * remove - disconnects the device
f655675b
N		 2585 * writemostly - sets write_mostly
2586 * -writemostly - clears write_mostly
de393cde
N		 2587 * blocked - sets the Blocked flags
2588 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2589 * insync - sets Insync providing device isn't active
d7a9d443
N		 2590 * write_error - sets WriteErrorSeen
2591 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2592 */
2593 int err = -EINVAL;
2594 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2595 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2596 if (test_bit(Faulty, &rdev->flags))
2597 err = 0;
2598 else
2599 err = -EBUSY;
45dc2de1
N		 2600 } else if (cmd_match(buf, "remove")) {
2601 if (rdev->raid_disk >= 0)
2602 err = -EBUSY;
2603 else {
fd01b88c 2604 struct mddev *mddev = rdev->mddev;
45dc2de1 2605 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2606 if (mddev->pers)
2607 md_update_sb(mddev, 1);
45dc2de1
N		 2608 md_new_event(mddev);
2609 err = 0;
2610 }
f655675b
N		 2611 } else if (cmd_match(buf, "writemostly")) {
2612 set_bit(WriteMostly, &rdev->flags);
2613 err = 0;
2614 } else if (cmd_match(buf, "-writemostly")) {
2615 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2616 err = 0;
2617 } else if (cmd_match(buf, "blocked")) {
2618 set_bit(Blocked, &rdev->flags);
2619 err = 0;
2620 } else if (cmd_match(buf, "-blocked")) {
de393cde 2621 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2622 rdev->badblocks.unacked_exist) {
de393cde
N		 2623 /* metadata handler doesn't understand badblocks,
2624 * so we need to fail the device
2625 */
2626 md_error(rdev->mddev, rdev);
2627 }
6bfe0b49 2628 clear_bit(Blocked, &rdev->flags);
de393cde 2629 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2630 wake_up(&rdev->blocked_wait);
2631 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2632 md_wakeup_thread(rdev->mddev->thread);
2633
6d56e278
N		 2634 err = 0;
2635 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2636 set_bit(In_sync, &rdev->flags);
f655675b 2637 err = 0;
d7a9d443
N		 2638 } else if (cmd_match(buf, "write_error")) {
2639 set_bit(WriteErrorSeen, &rdev->flags);
2640 err = 0;
2641 } else if (cmd_match(buf, "-write_error")) {
2642 clear_bit(WriteErrorSeen, &rdev->flags);
2643 err = 0;
2d78f8c4
N		 2644 } else if (cmd_match(buf, "want_replacement")) {
2645 /* Any non-spare device that is not a replacement can
2646 * become want_replacement at any time, but we then need to
2647 * check if recovery is needed.
2648 */
2649 if (rdev->raid_disk >= 0 &&
2650 !test_bit(Replacement, &rdev->flags))
2651 set_bit(WantReplacement, &rdev->flags);
2652 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2653 md_wakeup_thread(rdev->mddev->thread);
2654 err = 0;
2655 } else if (cmd_match(buf, "-want_replacement")) {
2656 /* Clearing 'want_replacement' is always allowed.
2657 * Once replacements starts it is too late though.
2658 */
2659 err = 0;
2660 clear_bit(WantReplacement, &rdev->flags);
2661 } else if (cmd_match(buf, "replacement")) {
2662 /* Can only set a device as a replacement when array has not
2663 * yet been started. Once running, replacement is automatic
2664 * from spares, or by assigning 'slot'.
2665 */
2666 if (rdev->mddev->pers)
2667 err = -EBUSY;
2668 else {
2669 set_bit(Replacement, &rdev->flags);
2670 err = 0;
2671 }
2672 } else if (cmd_match(buf, "-replacement")) {
2673 /* Similarly, can only clear Replacement before start */
2674 if (rdev->mddev->pers)
2675 err = -EBUSY;
2676 else {
2677 clear_bit(Replacement, &rdev->flags);
2678 err = 0;
2679 }
45dc2de1 2680 }
00bcb4ac
N		 2681 if (!err)
2682 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2683 return err ? err : len;
2684}
80ca3a44
N		 2685static struct rdev_sysfs_entry rdev_state =
2686__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2687
4dbcdc75 2688static ssize_t
3cb03002 2689errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2690{
2691 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2692}
2693
2694static ssize_t
3cb03002 2695errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2696{
2697 char *e;
2698 unsigned long n = simple_strtoul(buf, &e, 10);
2699 if (*buf && (*e == 0 || *e == '\n')) {
2700 atomic_set(&rdev->corrected_errors, n);
2701 return len;
2702 }
2703 return -EINVAL;
2704}
2705static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2706__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2707
014236d2 2708static ssize_t
3cb03002 2709slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2710{
2711 if (rdev->raid_disk < 0)
2712 return sprintf(page, "none\n");
2713 else
2714 return sprintf(page, "%d\n", rdev->raid_disk);
2715}
2716
2717static ssize_t
3cb03002 2718slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2719{
2720 char *e;
c303da6d 2721 int err;
014236d2
N		 2722 int slot = simple_strtoul(buf, &e, 10);
2723 if (strncmp(buf, "none", 4)==0)
2724 slot = -1;
2725 else if (e==buf || (*e && *e!= '\n'))
2726 return -EINVAL;
6c2fce2e 2727 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2728 /* Setting 'slot' on an active array requires also
2729 * updating the 'rd%d' link, and communicating
2730 * with the personality with ->hot_*_disk.
2731 * For now we only support removing
2732 * failed/spare devices. This normally happens automatically,
2733 * but not when the metadata is externally managed.
2734 */
c303da6d
N		 2735 if (rdev->raid_disk == -1)
2736 return -EEXIST;
2737 /* personality does all needed checks */
01393f3d 2738 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d
N		 2739 return -EINVAL;
2740 err = rdev->mddev->pers->
b8321b68 2741 hot_remove_disk(rdev->mddev, rdev);
c303da6d
N		 2742 if (err)
2743 return err;
36fad858 2744 sysfs_unlink_rdev(rdev->mddev, rdev);
b7103107 2745 rdev->raid_disk = -1;
c303da6d
N		 2746 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2747 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2748 } else if (rdev->mddev->pers) {
6c2fce2e 2749 /* Activating a spare .. or possibly reactivating
6d56e278 2750 * if we ever get bitmaps working here.
6c2fce2e
NB		 2751 */
2752
2753 if (rdev->raid_disk != -1)
2754 return -EBUSY;
2755
c6751b2b
N		 2756 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2757 return -EBUSY;
2758
6c2fce2e
NB		 2759 if (rdev->mddev->pers->hot_add_disk == NULL)
2760 return -EINVAL;
2761
ba1b41b6
N		 2762 if (slot >= rdev->mddev->raid_disks &&
2763 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2764 return -ENOSPC;
2765
6c2fce2e
NB		 2766 rdev->raid_disk = slot;
2767 if (test_bit(In_sync, &rdev->flags))
2768 rdev->saved_raid_disk = slot;
2769 else
2770 rdev->saved_raid_disk = -1;
d30519fc 2771 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2772 err = rdev->mddev->pers->
2773 hot_add_disk(rdev->mddev, rdev);
199050ea 2774 if (err) {
6c2fce2e 2775 rdev->raid_disk = -1;
6c2fce2e 2776 return err;
52664732 2777 } else
00bcb4ac 2778 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2779 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2780 /* failure here is OK */;
6c2fce2e 2781 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2782 } else {
ba1b41b6
N		 2783 if (slot >= rdev->mddev->raid_disks &&
2784 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2785 return -ENOSPC;
2786 rdev->raid_disk = slot;
2787 /* assume it is working */
c5d79adb
N		 2788 clear_bit(Faulty, &rdev->flags);
2789 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2790 set_bit(In_sync, &rdev->flags);
00bcb4ac 2791 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2792 }
014236d2
N		 2793 return len;
2794}
2795
2796
2797static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2798__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2799
93c8cad0 2800static ssize_t
3cb03002 2801offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2802{
6961ece4 2803 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2804}
2805
2806static ssize_t
3cb03002 2807offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0
N		 2808{
2809 char *e;
2810 unsigned long long offset = simple_strtoull(buf, &e, 10);
2811 if (e==buf || (*e && *e != '\n'))
2812 return -EINVAL;
8ed0a521 2813 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2814 return -EBUSY;
dd8ac336 2815 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2816 /* Must set offset before size, so overlap checks
2817 * can be sane */
2818 return -EBUSY;
93c8cad0
N		 2819 rdev->data_offset = offset;
2820 return len;
2821}
2822
2823static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2824__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2825
83303b61 2826static ssize_t
3cb03002 2827rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2828{
dd8ac336 2829 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2830}
2831
c5d79adb
N		 2832static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2833{
2834 /* check if two start/length pairs overlap */
2835 if (s1+l1 <= s2)
2836 return 0;
2837 if (s2+l2 <= s1)
2838 return 0;
2839 return 1;
2840}
2841
b522adcd
DW		 2842static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2843{
2844 unsigned long long blocks;
2845 sector_t new;
2846
2847 if (strict_strtoull(buf, 10, &blocks) < 0)
2848 return -EINVAL;
2849
2850 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2851 return -EINVAL; /* sector conversion overflow */
2852
2853 new = blocks * 2;
2854 if (new != blocks * 2)
2855 return -EINVAL; /* unsigned long long to sector_t overflow */
2856
2857 *sectors = new;
2858 return 0;
2859}
2860
83303b61 2861static ssize_t
3cb03002 2862rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2863{
fd01b88c 2864 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2865 sector_t oldsectors = rdev->sectors;
b522adcd 2866 sector_t sectors;
27c529bb 2867
b522adcd 2868 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2869 return -EINVAL;
0cd17fec 2870 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2871 if (my_mddev->persistent) {
dd8ac336
AN		 2872 sectors = super_types[my_mddev->major_version].
2873 rdev_size_change(rdev, sectors);
2874 if (!sectors)
0cd17fec 2875 return -EBUSY;
dd8ac336 2876 } else if (!sectors)
77304d2a 2877 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2878 rdev->data_offset;
0cd17fec 2879 }
dd8ac336 2880 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2881 return -EINVAL; /* component must fit device */
0cd17fec 2882
dd8ac336
AN		 2883 rdev->sectors = sectors;
2884 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2885 /* need to check that all other rdevs with the same ->bdev
2886 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2887 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2888 * we have to change it back, we will have the lock again.
2889 */
fd01b88c 2890 struct mddev *mddev;
c5d79adb 2891 int overlap = 0;
159ec1fc 2892 struct list_head *tmp;
c5d79adb 2893
27c529bb 2894 mddev_unlock(my_mddev);
29ac4aa3 2895 for_each_mddev(mddev, tmp) {
3cb03002 2896 struct md_rdev *rdev2;
c5d79adb
N		 2897
2898 mddev_lock(mddev);
dafb20fa 2899 rdev_for_each(rdev2, mddev)
f21e9ff7
N		 2900 if (rdev->bdev == rdev2->bdev &&
2901 rdev != rdev2 &&
2902 overlaps(rdev->data_offset, rdev->sectors,
2903 rdev2->data_offset,
2904 rdev2->sectors)) {
c5d79adb
N		 2905 overlap = 1;
2906 break;
2907 }
2908 mddev_unlock(mddev);
2909 if (overlap) {
2910 mddev_put(mddev);
2911 break;
2912 }
2913 }
27c529bb 2914 mddev_lock(my_mddev);
c5d79adb
N		 2915 if (overlap) {
2916 /* Someone else could have slipped in a size
2917 * change here, but doing so is just silly.
dd8ac336 2918 * We put oldsectors back because we *know* it is
c5d79adb
N		 2919 * safe, and trust userspace not to race with
2920 * itself
2921 */
dd8ac336 2922 rdev->sectors = oldsectors;
c5d79adb
N		 2923 return -EBUSY;
2924 }
2925 }
83303b61
N		 2926 return len;
2927}
2928
2929static struct rdev_sysfs_entry rdev_size =
80ca3a44 2930__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2931
06e3c817 2932
3cb03002 2933static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2934{
2935 unsigned long long recovery_start = rdev->recovery_offset;
2936
2937 if (test_bit(In_sync, &rdev->flags) ||
2938 recovery_start == MaxSector)
2939 return sprintf(page, "none\n");
2940
2941 return sprintf(page, "%llu\n", recovery_start);
2942}
2943
3cb03002 2944static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2945{
2946 unsigned long long recovery_start;
2947
2948 if (cmd_match(buf, "none"))
2949 recovery_start = MaxSector;
2950 else if (strict_strtoull(buf, 10, &recovery_start))
2951 return -EINVAL;
2952
2953 if (rdev->mddev->pers &&
2954 rdev->raid_disk >= 0)
2955 return -EBUSY;
2956
2957 rdev->recovery_offset = recovery_start;
2958 if (recovery_start == MaxSector)
2959 set_bit(In_sync, &rdev->flags);
2960 else
2961 clear_bit(In_sync, &rdev->flags);
2962 return len;
2963}
2964
2965static struct rdev_sysfs_entry rdev_recovery_start =
2966__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2967
16c791a5
N		 2968
2969static ssize_t
2970badblocks_show(struct badblocks *bb, char *page, int unack);
2971static ssize_t
2972badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2973
3cb03002 2974static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2975{
2976 return badblocks_show(&rdev->badblocks, page, 0);
2977}
3cb03002 2978static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2979{
de393cde
N		 2980 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2981 /* Maybe that ack was all we needed */
2982 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2983 wake_up(&rdev->blocked_wait);
2984 return rv;
16c791a5
N		 2985}
2986static struct rdev_sysfs_entry rdev_bad_blocks =
2987__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2988
2989
3cb03002 2990static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2991{
2992 return badblocks_show(&rdev->badblocks, page, 1);
2993}
3cb03002 2994static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2995{
2996 return badblocks_store(&rdev->badblocks, page, len, 1);
2997}
2998static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2999__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3000
86e6ffdd
N		 3001static struct attribute *rdev_default_attrs[] = {
3002 &rdev_state.attr,
4dbcdc75 3003 &rdev_errors.attr,
014236d2 3004 &rdev_slot.attr,
93c8cad0 3005 &rdev_offset.attr,
83303b61 3006 &rdev_size.attr,
06e3c817 3007 &rdev_recovery_start.attr,
16c791a5
N		 3008 &rdev_bad_blocks.attr,
3009 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 3010 NULL,
3011};
3012static ssize_t
3013rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3014{
3015 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3016 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 3017 struct mddev *mddev = rdev->mddev;
27c529bb 3018 ssize_t rv;
86e6ffdd
N		 3019
3020 if (!entry->show)
3021 return -EIO;
27c529bb
N		 3022
3023 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3024 if (!rv) {
3025 if (rdev->mddev == NULL)
3026 rv = -EBUSY;
3027 else
3028 rv = entry->show(rdev, page);
3029 mddev_unlock(mddev);
3030 }
3031 return rv;
86e6ffdd
N		 3032}
3033
3034static ssize_t
3035rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3036 const char *page, size_t length)
3037{
3038 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 3039 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 3040 ssize_t rv;
fd01b88c 3041 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3042
3043 if (!entry->store)
3044 return -EIO;
67463acb
N		 3045 if (!capable(CAP_SYS_ADMIN))
3046 return -EACCES;
27c529bb 3047 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3048 if (!rv) {
27c529bb
N		 3049 if (rdev->mddev == NULL)
3050 rv = -EBUSY;
3051 else
3052 rv = entry->store(rdev, page, length);
6a51830e 3053 mddev_unlock(mddev);
ca388059
N		 3054 }
3055 return rv;
86e6ffdd
N		 3056}
3057
3058static void rdev_free(struct kobject *ko)
3059{
3cb03002 3060 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3061 kfree(rdev);
3062}
52cf25d0 3063static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3064 .show = rdev_attr_show,
3065 .store = rdev_attr_store,
3066};
3067static struct kobj_type rdev_ktype = {
3068 .release = rdev_free,
3069 .sysfs_ops = &rdev_sysfs_ops,
3070 .default_attrs = rdev_default_attrs,
3071};
3072
3cb03002 3073int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3074{
3075 rdev->desc_nr = -1;
3076 rdev->saved_raid_disk = -1;
3077 rdev->raid_disk = -1;
3078 rdev->flags = 0;
3079 rdev->data_offset = 0;
3080 rdev->sb_events = 0;
3081 rdev->last_read_error.tv_sec = 0;
3082 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3083 rdev->sb_loaded = 0;
3084 rdev->bb_page = NULL;
e8bb9a83
N		 3085 atomic_set(&rdev->nr_pending, 0);
3086 atomic_set(&rdev->read_errors, 0);
3087 atomic_set(&rdev->corrected_errors, 0);
3088
3089 INIT_LIST_HEAD(&rdev->same_set);
3090 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3091
3092 /* Add space to store bad block list.
3093 * This reserves the space even on arrays where it cannot
3094 * be used - I wonder if that matters
3095 */
3096 rdev->badblocks.count = 0;
3097 rdev->badblocks.shift = 0;
3098 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3099 seqlock_init(&rdev->badblocks.lock);
3100 if (rdev->badblocks.page == NULL)
3101 return -ENOMEM;
3102
3103 return 0;
e8bb9a83
N		 3104}
3105EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3106/*
3107 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3108 *
3109 * mark the device faulty if:
3110 *
3111 * - the device is nonexistent (zero size)
3112 * - the device has no valid superblock
3113 *
3114 * a faulty rdev _never_ has rdev->sb set.
3115 */
3cb03002 3116static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3117{
3118 char b[BDEVNAME_SIZE];
3119 int err;
3cb03002 3120 struct md_rdev *rdev;
1da177e4
LT		 3121 sector_t size;
3122
9ffae0cf 3123 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3124 if (!rdev) {
3125 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3126 return ERR_PTR(-ENOMEM);
3127 }
1da177e4 3128
2230dfe4
N		 3129 err = md_rdev_init(rdev);
3130 if (err)
3131 goto abort_free;
3132 err = alloc_disk_sb(rdev);
3133 if (err)
1da177e4
LT		 3134 goto abort_free;
3135
c5d79adb 3136 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3137 if (err)
3138 goto abort_free;
3139
f9cb074b 3140 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3141
77304d2a 3142 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3143 if (!size) {
3144 printk(KERN_WARNING
3145 "md: %s has zero or unknown size, marking faulty!\n",
3146 bdevname(rdev->bdev,b));
3147 err = -EINVAL;
3148 goto abort_free;
3149 }
3150
3151 if (super_format >= 0) {
3152 err = super_types[super_format].
3153 load_super(rdev, NULL, super_minor);
3154 if (err == -EINVAL) {
df968c4e
N		 3155 printk(KERN_WARNING
3156 "md: %s does not have a valid v%d.%d "
3157 "superblock, not importing!\n",
3158 bdevname(rdev->bdev,b),
3159 super_format, super_minor);
1da177e4
LT		 3160 goto abort_free;
3161 }
3162 if (err < 0) {
3163 printk(KERN_WARNING
3164 "md: could not read %s's sb, not importing!\n",
3165 bdevname(rdev->bdev,b));
3166 goto abort_free;
3167 }
3168 }
9f2f3830
N		 3169 if (super_format == -1)
3170 /* hot-add for 0.90, or non-persistent: so no badblocks */
3171 rdev->badblocks.shift = -1;
6bfe0b49 3172
1da177e4
LT		 3173 return rdev;
3174
3175abort_free:
2699b672
N		 3176 if (rdev->bdev)
3177 unlock_rdev(rdev);
3178 free_disk_sb(rdev);
2230dfe4 3179 kfree(rdev->badblocks.page);
1da177e4
LT		 3180 kfree(rdev);
3181 return ERR_PTR(err);
3182}
3183
3184/*
3185 * Check a full RAID array for plausibility
3186 */
3187
3188
fd01b88c 3189static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3190{
3191 int i;
3cb03002 3192 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3193 char b[BDEVNAME_SIZE];
3194
3195 freshest = NULL;
dafb20fa 3196 rdev_for_each_safe(rdev, tmp, mddev)
1da177e4
LT		 3197 switch (super_types[mddev->major_version].
3198 load_super(rdev, freshest, mddev->minor_version)) {
3199 case 1:
3200 freshest = rdev;
3201 break;
3202 case 0:
3203 break;
3204 default:
3205 printk( KERN_ERR \
3206 "md: fatal superblock inconsistency in %s"
3207 " -- removing from array\n",
3208 bdevname(rdev->bdev,b));
3209 kick_rdev_from_array(rdev);
3210 }
3211
3212
3213 super_types[mddev->major_version].
3214 validate_super(mddev, freshest);
3215
3216 i = 0;
dafb20fa 3217 rdev_for_each_safe(rdev, tmp, mddev) {
233fca36
N		 3218 if (mddev->max_disks &&
3219 (rdev->desc_nr >= mddev->max_disks ||
3220 i > mddev->max_disks)) {
de01dfad
N		 3221 printk(KERN_WARNING
3222 "md: %s: %s: only %d devices permitted\n",
3223 mdname(mddev), bdevname(rdev->bdev, b),
3224 mddev->max_disks);
3225 kick_rdev_from_array(rdev);
3226 continue;
3227 }
1da177e4
LT		 3228 if (rdev != freshest)
3229 if (super_types[mddev->major_version].
3230 validate_super(mddev, rdev)) {
3231 printk(KERN_WARNING "md: kicking non-fresh %s"
3232 " from array!\n",
3233 bdevname(rdev->bdev,b));
3234 kick_rdev_from_array(rdev);
3235 continue;
3236 }
3237 if (mddev->level == LEVEL_MULTIPATH) {
3238 rdev->desc_nr = i++;
3239 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3240 set_bit(In_sync, &rdev->flags);
5e5e3e78 3241 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3242 rdev->raid_disk = -1;
3243 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3244 }
3245 }
1da177e4
LT		 3246}
3247
72e02075
N		 3248/* Read a fixed-point number.
3249 * Numbers in sysfs attributes should be in "standard" units where
3250 * possible, so time should be in seconds.
3251 * However we internally use a a much smaller unit such as
3252 * milliseconds or jiffies.
3253 * This function takes a decimal number with a possible fractional
3254 * component, and produces an integer which is the result of
3255 * multiplying that number by 10^'scale'.
3256 * all without any floating-point arithmetic.
3257 */
3258int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3259{
3260 unsigned long result = 0;
3261 long decimals = -1;
3262 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3263 if (*cp == '.')
3264 decimals = 0;
3265 else if (decimals < scale) {
3266 unsigned int value;
3267 value = *cp - '0';
3268 result = result * 10 + value;
3269 if (decimals >= 0)
3270 decimals++;
3271 }
3272 cp++;
3273 }
3274 if (*cp == '\n')
3275 cp++;
3276 if (*cp)
3277 return -EINVAL;
3278 if (decimals < 0)
3279 decimals = 0;
3280 while (decimals < scale) {
3281 result *= 10;
3282 decimals ++;
3283 }
3284 *res = result;
3285 return 0;
3286}
3287
3288
19052c0e
N		 3289static void md_safemode_timeout(unsigned long data);
3290
16f17b39 3291static ssize_t
fd01b88c 3292safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3293{
3294 int msec = (mddev->safemode_delay*1000)/HZ;
3295 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3296}
3297static ssize_t
fd01b88c 3298safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3299{
16f17b39 3300 unsigned long msec;
97ce0a7f 3301
72e02075 3302 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3303 return -EINVAL;
16f17b39
N		 3304 if (msec == 0)
3305 mddev->safemode_delay = 0;
3306 else {
19052c0e 3307 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3308 mddev->safemode_delay = (msec*HZ)/1000;
3309 if (mddev->safemode_delay == 0)
3310 mddev->safemode_delay = 1;
19052c0e
N		 3311 if (mddev->safemode_delay < old_delay)
3312 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3313 }
3314 return len;
3315}
3316static struct md_sysfs_entry md_safe_delay =
80ca3a44 3317__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3318
eae1701f 3319static ssize_t
fd01b88c 3320level_show(struct mddev *mddev, char *page)
eae1701f 3321{
84fc4b56 3322 struct md_personality *p = mddev->pers;
d9d166c2 3323 if (p)
eae1701f 3324 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3325 else if (mddev->clevel[0])
3326 return sprintf(page, "%s\n", mddev->clevel);
3327 else if (mddev->level != LEVEL_NONE)
3328 return sprintf(page, "%d\n", mddev->level);
3329 else
3330 return 0;
eae1701f
N		 3331}
3332
d9d166c2 3333static ssize_t
fd01b88c 3334level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3335{
f2859af6 3336 char clevel[16];
20a49ff6 3337 ssize_t rv = len;
84fc4b56 3338 struct md_personality *pers;
f2859af6 3339 long level;
245f46c2 3340 void *priv;
3cb03002 3341 struct md_rdev *rdev;
245f46c2
N		 3342
3343 if (mddev->pers == NULL) {
3344 if (len == 0)
3345 return 0;
3346 if (len >= sizeof(mddev->clevel))
3347 return -ENOSPC;
3348 strncpy(mddev->clevel, buf, len);
3349 if (mddev->clevel[len-1] == '\n')
3350 len--;
3351 mddev->clevel[len] = 0;
3352 mddev->level = LEVEL_NONE;
3353 return rv;
3354 }
3355
3356 /* request to change the personality. Need to ensure:
3357 * - array is not engaged in resync/recovery/reshape
3358 * - old personality can be suspended
3359 * - new personality will access other array.
3360 */
3361
bb4f1e9d
N		 3362 if (mddev->sync_thread ||
3363 mddev->reshape_position != MaxSector ||
3364 mddev->sysfs_active)
d9d166c2 3365 return -EBUSY;
245f46c2
N		 3366
3367 if (!mddev->pers->quiesce) {
3368 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3369 mdname(mddev), mddev->pers->name);
3370 return -EINVAL;
3371 }
3372
3373 /* Now find the new personality */
f2859af6 3374 if (len == 0 || len >= sizeof(clevel))
245f46c2 3375 return -EINVAL;
f2859af6
DW		 3376 strncpy(clevel, buf, len);
3377 if (clevel[len-1] == '\n')
d9d166c2 3378 len--;
f2859af6
DW		 3379 clevel[len] = 0;
3380 if (strict_strtol(clevel, 10, &level))
3381 level = LEVEL_NONE;
245f46c2 3382
f2859af6
DW		 3383 if (request_module("md-%s", clevel) != 0)
3384 request_module("md-level-%s", clevel);
245f46c2 3385 spin_lock(&pers_lock);
f2859af6 3386 pers = find_pers(level, clevel);
245f46c2
N		 3387 if (!pers || !try_module_get(pers->owner)) {
3388 spin_unlock(&pers_lock);
f2859af6 3389 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3390 return -EINVAL;
3391 }
3392 spin_unlock(&pers_lock);
3393
3394 if (pers == mddev->pers) {
3395 /* Nothing to do! */
3396 module_put(pers->owner);
3397 return rv;
3398 }
3399 if (!pers->takeover) {
3400 module_put(pers->owner);
3401 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3402 mdname(mddev), clevel);
245f46c2
N		 3403 return -EINVAL;
3404 }
3405
dafb20fa 3406 rdev_for_each(rdev, mddev)
e93f68a1
N		 3407 rdev->new_raid_disk = rdev->raid_disk;
3408
245f46c2
N		 3409 /* ->takeover must set new_* and/or delta_disks
3410 * if it succeeds, and may set them when it fails.
3411 */
3412 priv = pers->takeover(mddev);
3413 if (IS_ERR(priv)) {
3414 mddev->new_level = mddev->level;
3415 mddev->new_layout = mddev->layout;
664e7c41 3416 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3417 mddev->raid_disks -= mddev->delta_disks;
3418 mddev->delta_disks = 0;
3419 module_put(pers->owner);
3420 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3421 mdname(mddev), clevel);
245f46c2
N		 3422 return PTR_ERR(priv);
3423 }
3424
3425 /* Looks like we have a winner */
3426 mddev_suspend(mddev);
3427 mddev->pers->stop(mddev);
a64c876f
N		 3428
3429 if (mddev->pers->sync_request == NULL &&
3430 pers->sync_request != NULL) {
3431 /* need to add the md_redundancy_group */
3432 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3433 printk(KERN_WARNING
3434 "md: cannot register extra attributes for %s\n",
3435 mdname(mddev));
19fdb9ee 3436 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3437 }
3438 if (mddev->pers->sync_request != NULL &&
3439 pers->sync_request == NULL) {
3440 /* need to remove the md_redundancy_group */
3441 if (mddev->to_remove == NULL)
3442 mddev->to_remove = &md_redundancy_group;
3443 }
3444
54071b38
TM		 3445 if (mddev->pers->sync_request == NULL &&
3446 mddev->external) {
3447 /* We are converting from a no-redundancy array
3448 * to a redundancy array and metadata is managed
3449 * externally so we need to be sure that writes
3450 * won't block due to a need to transition
3451 * clean->dirty
3452 * until external management is started.
3453 */
3454 mddev->in_sync = 0;
3455 mddev->safemode_delay = 0;
3456 mddev->safemode = 0;
3457 }
3458
dafb20fa 3459 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3460 if (rdev->raid_disk < 0)
3461 continue;
bf2cb0da 3462 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3463 rdev->new_raid_disk = -1;
3464 if (rdev->new_raid_disk == rdev->raid_disk)
3465 continue;
36fad858 3466 sysfs_unlink_rdev(mddev, rdev);
e93f68a1 3467 }
dafb20fa 3468 rdev_for_each(rdev, mddev) {
e93f68a1
N		 3469 if (rdev->raid_disk < 0)
3470 continue;
3471 if (rdev->new_raid_disk == rdev->raid_disk)
3472 continue;
3473 rdev->raid_disk = rdev->new_raid_disk;
3474 if (rdev->raid_disk < 0)
3a981b03 3475 clear_bit(In_sync, &rdev->flags);
e93f68a1 3476 else {
36fad858
NK		 3477 if (sysfs_link_rdev(mddev, rdev))
3478 printk(KERN_WARNING "md: cannot register rd%d"
3479 " for %s after level change\n",
3480 rdev->raid_disk, mdname(mddev));
3a981b03 3481 }
e93f68a1
N		 3482 }
3483
3484 module_put(mddev->pers->owner);
245f46c2
N		 3485 mddev->pers = pers;
3486 mddev->private = priv;
3487 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3488 mddev->level = mddev->new_level;
3489 mddev->layout = mddev->new_layout;
664e7c41 3490 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3491 mddev->delta_disks = 0;
fee68723 3492 mddev->degraded = 0;
9af204cf
TM		 3493 if (mddev->pers->sync_request == NULL) {
3494 /* this is now an array without redundancy, so
3495 * it must always be in_sync
3496 */
3497 mddev->in_sync = 1;
3498 del_timer_sync(&mddev->safemode_timer);
3499 }
245f46c2
N		 3500 pers->run(mddev);
3501 mddev_resume(mddev);
3502 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3503 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3504 md_wakeup_thread(mddev->thread);
5cac7861 3505 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3506 md_new_event(mddev);
d9d166c2
N		 3507 return rv;
3508}
3509
3510static struct md_sysfs_entry md_level =
80ca3a44 3511__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3512
d4dbd025
N		 3513
3514static ssize_t
fd01b88c 3515layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3516{
3517 /* just a number, not meaningful for all levels */
08a02ecd
N		 3518 if (mddev->reshape_position != MaxSector &&
3519 mddev->layout != mddev->new_layout)
3520 return sprintf(page, "%d (%d)\n",
3521 mddev->new_layout, mddev->layout);
d4dbd025
N		 3522 return sprintf(page, "%d\n", mddev->layout);
3523}
3524
3525static ssize_t
fd01b88c 3526layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3527{
3528 char *e;
3529 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3530
3531 if (!*buf || (*e && *e != '\n'))
3532 return -EINVAL;
3533
b3546035
N		 3534 if (mddev->pers) {
3535 int err;
50ac168a 3536 if (mddev->pers->check_reshape == NULL)
b3546035 3537 return -EBUSY;
597a711b 3538 mddev->new_layout = n;
50ac168a 3539 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3540 if (err) {
3541 mddev->new_layout = mddev->layout;
b3546035 3542 return err;
597a711b 3543 }
b3546035 3544 } else {
08a02ecd 3545 mddev->new_layout = n;
b3546035
N		 3546 if (mddev->reshape_position == MaxSector)
3547 mddev->layout = n;
3548 }
d4dbd025
N		 3549 return len;
3550}
3551static struct md_sysfs_entry md_layout =
80ca3a44 3552__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3553
3554
eae1701f 3555static ssize_t
fd01b88c 3556raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3557{
bb636547
N		 3558 if (mddev->raid_disks == 0)
3559 return 0;
08a02ecd
N		 3560 if (mddev->reshape_position != MaxSector &&
3561 mddev->delta_disks != 0)
3562 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3563 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3564 return sprintf(page, "%d\n", mddev->raid_disks);
3565}
3566
fd01b88c 3567static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3568
3569static ssize_t
fd01b88c 3570raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3571{
da943b99
N		 3572 char *e;
3573 int rv = 0;
3574 unsigned long n = simple_strtoul(buf, &e, 10);
3575
3576 if (!*buf || (*e && *e != '\n'))
3577 return -EINVAL;
3578
3579 if (mddev->pers)
3580 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3581 else if (mddev->reshape_position != MaxSector) {
3582 int olddisks = mddev->raid_disks - mddev->delta_disks;
3583 mddev->delta_disks = n - olddisks;
3584 mddev->raid_disks = n;
3585 } else
da943b99
N		 3586 mddev->raid_disks = n;
3587 return rv ? rv : len;
3588}
3589static struct md_sysfs_entry md_raid_disks =
80ca3a44 3590__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3591
3b34380a 3592static ssize_t
fd01b88c 3593chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3594{
08a02ecd 3595 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3596 mddev->chunk_sectors != mddev->new_chunk_sectors)
3597 return sprintf(page, "%d (%d)\n",
3598 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3599 mddev->chunk_sectors << 9);
3600 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3601}
3602
3603static ssize_t
fd01b88c 3604chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3605{
3b34380a
N		 3606 char *e;
3607 unsigned long n = simple_strtoul(buf, &e, 10);
3608
3b34380a
N		 3609 if (!*buf || (*e && *e != '\n'))
3610 return -EINVAL;
3611
b3546035
N		 3612 if (mddev->pers) {
3613 int err;
50ac168a 3614 if (mddev->pers->check_reshape == NULL)
b3546035 3615 return -EBUSY;
597a711b 3616 mddev->new_chunk_sectors = n >> 9;
50ac168a 3617 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3618 if (err) {
3619 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3620 return err;
597a711b 3621 }
b3546035 3622 } else {
664e7c41 3623 mddev->new_chunk_sectors = n >> 9;
b3546035 3624 if (mddev->reshape_position == MaxSector)
9d8f0363 3625 mddev->chunk_sectors = n >> 9;
b3546035 3626 }
3b34380a
N		 3627 return len;
3628}
3629static struct md_sysfs_entry md_chunk_size =
80ca3a44 3630__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3631
a94213b1 3632static ssize_t
fd01b88c 3633resync_start_show(struct mddev *mddev, char *page)
a94213b1 3634{
d1a7c503
N		 3635 if (mddev->recovery_cp == MaxSector)
3636 return sprintf(page, "none\n");
a94213b1
N		 3637 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3638}
3639
3640static ssize_t
fd01b88c 3641resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3642{
a94213b1
N		 3643 char *e;
3644 unsigned long long n = simple_strtoull(buf, &e, 10);
3645
b098636c 3646 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3647 return -EBUSY;
06e3c817
DW		 3648 if (cmd_match(buf, "none"))
3649 n = MaxSector;
3650 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3651 return -EINVAL;
3652
3653 mddev->recovery_cp = n;
3654 return len;
3655}
3656static struct md_sysfs_entry md_resync_start =
80ca3a44 3657__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3658
9e653b63
N		 3659/*
3660 * The array state can be:
3661 *
3662 * clear
3663 * No devices, no size, no level
3664 * Equivalent to STOP_ARRAY ioctl
3665 * inactive
3666 * May have some settings, but array is not active
3667 * all IO results in error
3668 * When written, doesn't tear down array, but just stops it
3669 * suspended (not supported yet)
3670 * All IO requests will block. The array can be reconfigured.
910d8cb3 3671 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3672 * readonly
3673 * no resync can happen. no superblocks get written.
3674 * write requests fail
3675 * read-auto
3676 * like readonly, but behaves like 'clean' on a write request.
3677 *
3678 * clean - no pending writes, but otherwise active.
3679 * When written to inactive array, starts without resync
3680 * If a write request arrives then
3681 * if metadata is known, mark 'dirty' and switch to 'active'.
3682 * if not known, block and switch to write-pending
3683 * If written to an active array that has pending writes, then fails.
3684 * active
3685 * fully active: IO and resync can be happening.
3686 * When written to inactive array, starts with resync
3687 *
3688 * write-pending
3689 * clean, but writes are blocked waiting for 'active' to be written.
3690 *
3691 * active-idle
3692 * like active, but no writes have been seen for a while (100msec).
3693 *
3694 */
3695enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3696 write_pending, active_idle, bad_word};
05381954 3697static char *array_states[] = {
9e653b63
N		 3698 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3699 "write-pending", "active-idle", NULL };
3700
3701static int match_word(const char *word, char **list)
3702{
3703 int n;
3704 for (n=0; list[n]; n++)
3705 if (cmd_match(word, list[n]))
3706 break;
3707 return n;
3708}
3709
3710static ssize_t
fd01b88c 3711array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3712{
3713 enum array_state st = inactive;
3714
3715 if (mddev->pers)
3716 switch(mddev->ro) {
3717 case 1:
3718 st = readonly;
3719 break;
3720 case 2:
3721 st = read_auto;
3722 break;
3723 case 0:
3724 if (mddev->in_sync)
3725 st = clean;
070dc6dd 3726 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3727 st = write_pending;
9e653b63
N		 3728 else if (mddev->safemode)
3729 st = active_idle;
3730 else
3731 st = active;
3732 }
3733 else {
3734 if (list_empty(&mddev->disks) &&
3735 mddev->raid_disks == 0 &&
58c0fed4 3736 mddev->dev_sectors == 0)
9e653b63
N		 3737 st = clear;
3738 else
3739 st = inactive;
3740 }
3741 return sprintf(page, "%s\n", array_states[st]);
3742}
3743
fd01b88c
N		 3744static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3745static int md_set_readonly(struct mddev * mddev, int is_open);
3746static int do_md_run(struct mddev * mddev);
3747static int restart_array(struct mddev *mddev);
9e653b63
N		 3748
3749static ssize_t
fd01b88c 3750array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3751{
3752 int err = -EINVAL;
3753 enum array_state st = match_word(buf, array_states);
3754 switch(st) {
3755 case bad_word:
3756 break;
3757 case clear:
3758 /* stopping an active array */
f2ea68cf 3759 if (atomic_read(&mddev->openers) > 0)
e691063a 3760 return -EBUSY;
df5b20cf 3761 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3762 break;
3763 case inactive:
3764 /* stopping an active array */
3765 if (mddev->pers) {
f2ea68cf 3766 if (atomic_read(&mddev->openers) > 0)
9e653b63 3767 return -EBUSY;
df5b20cf 3768 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3769 } else
3770 err = 0; /* already inactive */
9e653b63
N		 3771 break;
3772 case suspended:
3773 break; /* not supported yet */
3774 case readonly:
3775 if (mddev->pers)
a4bd82d0 3776 err = md_set_readonly(mddev, 0);
9e653b63
N		 3777 else {
3778 mddev->ro = 1;
648b629e 3779 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3780 err = do_md_run(mddev);
3781 }
3782 break;
3783 case read_auto:
9e653b63 3784 if (mddev->pers) {
80268ee9 3785 if (mddev->ro == 0)
a4bd82d0 3786 err = md_set_readonly(mddev, 0);
80268ee9 3787 else if (mddev->ro == 1)
648b629e
N		 3788 err = restart_array(mddev);
3789 if (err == 0) {
3790 mddev->ro = 2;
3791 set_disk_ro(mddev->gendisk, 0);
3792 }
9e653b63
N		 3793 } else {
3794 mddev->ro = 2;
3795 err = do_md_run(mddev);
3796 }
3797 break;
3798 case clean:
3799 if (mddev->pers) {
3800 restart_array(mddev);
3801 spin_lock_irq(&mddev->write_lock);
3802 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3803 if (mddev->in_sync == 0) {
3804 mddev->in_sync = 1;
31a59e34
N		 3805 if (mddev->safemode == 1)
3806 mddev->safemode = 0;
070dc6dd 3807 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3808 }
3809 err = 0;
3810 } else
3811 err = -EBUSY;
9e653b63 3812 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3813 } else
3814 err = -EINVAL;
9e653b63
N		 3815 break;
3816 case active:
3817 if (mddev->pers) {
3818 restart_array(mddev);
070dc6dd 3819 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3820 wake_up(&mddev->sb_wait);
3821 err = 0;
3822 } else {
3823 mddev->ro = 0;
648b629e 3824 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3825 err = do_md_run(mddev);
3826 }
3827 break;
3828 case write_pending:
3829 case active_idle:
3830 /* these cannot be set */
3831 break;
3832 }
3833 if (err)
3834 return err;
0fd62b86 3835 else {
1d23f178
N		 3836 if (mddev->hold_active == UNTIL_IOCTL)
3837 mddev->hold_active = 0;
00bcb4ac 3838 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3839 return len;
0fd62b86 3840 }
9e653b63 3841}
80ca3a44
N		 3842static struct md_sysfs_entry md_array_state =
3843__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3844
1e50915f 3845static ssize_t
fd01b88c 3846max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3847 return sprintf(page, "%d\n",
3848 atomic_read(&mddev->max_corr_read_errors));
3849}
3850
3851static ssize_t
fd01b88c 3852max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3853{
3854 char *e;
3855 unsigned long n = simple_strtoul(buf, &e, 10);
3856
3857 if (*buf && (*e == 0 || *e == '\n')) {
3858 atomic_set(&mddev->max_corr_read_errors, n);
3859 return len;
3860 }
3861 return -EINVAL;
3862}
3863
3864static struct md_sysfs_entry max_corr_read_errors =
3865__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3866 max_corrected_read_errors_store);
3867
6d7ff738 3868static ssize_t
fd01b88c 3869null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3870{
3871 return -EINVAL;
3872}
3873
3874static ssize_t
fd01b88c 3875new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3876{
3877 /* buf must be %d:%d\n? giving major and minor numbers */
3878 /* The new device is added to the array.
3879 * If the array has a persistent superblock, we read the
3880 * superblock to initialise info and check validity.
3881 * Otherwise, only checking done is that in bind_rdev_to_array,
3882 * which mainly checks size.
3883 */
3884 char *e;
3885 int major = simple_strtoul(buf, &e, 10);
3886 int minor;
3887 dev_t dev;
3cb03002 3888 struct md_rdev *rdev;
6d7ff738
N		 3889 int err;
3890
3891 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3892 return -EINVAL;
3893 minor = simple_strtoul(e+1, &e, 10);
3894 if (*e && *e != '\n')
3895 return -EINVAL;
3896 dev = MKDEV(major, minor);
3897 if (major != MAJOR(dev) ||
3898 minor != MINOR(dev))
3899 return -EOVERFLOW;
3900
3901
3902 if (mddev->persistent) {
3903 rdev = md_import_device(dev, mddev->major_version,
3904 mddev->minor_version);
3905 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3906 struct md_rdev *rdev0
3907 = list_entry(mddev->disks.next,
3908 struct md_rdev, same_set);
6d7ff738
N		 3909 err = super_types[mddev->major_version]
3910 .load_super(rdev, rdev0, mddev->minor_version);
3911 if (err < 0)
3912 goto out;
3913 }
c5d79adb
N		 3914 } else if (mddev->external)
3915 rdev = md_import_device(dev, -2, -1);
3916 else
6d7ff738
N		 3917 rdev = md_import_device(dev, -1, -1);
3918
3919 if (IS_ERR(rdev))
3920 return PTR_ERR(rdev);
3921 err = bind_rdev_to_array(rdev, mddev);
3922 out:
3923 if (err)
3924 export_rdev(rdev);
3925 return err ? err : len;
3926}
3927
3928static struct md_sysfs_entry md_new_device =
80ca3a44 3929__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3930
9b1d1dac 3931static ssize_t
fd01b88c 3932bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3933{
3934 char *end;
3935 unsigned long chunk, end_chunk;
3936
3937 if (!mddev->bitmap)
3938 goto out;
3939 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3940 while (*buf) {
3941 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3942 if (buf == end) break;
3943 if (*end == '-') { /* range */
3944 buf = end + 1;
3945 end_chunk = simple_strtoul(buf, &end, 0);
3946 if (buf == end) break;
3947 }
3948 if (*end && !isspace(*end)) break;
3949 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3950 buf = skip_spaces(end);
9b1d1dac
PC		 3951 }
3952 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3953out:
3954 return len;
3955}
3956
3957static struct md_sysfs_entry md_bitmap =
3958__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3959
a35b0d69 3960static ssize_t
fd01b88c 3961size_show(struct mddev *mddev, char *page)
a35b0d69 3962{
58c0fed4
AN		 3963 return sprintf(page, "%llu\n",
3964 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3965}
3966
fd01b88c 3967static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 3968
3969static ssize_t
fd01b88c 3970size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 3971{
3972 /* If array is inactive, we can reduce the component size, but
3973 * not increase it (except from 0).
3974 * If array is active, we can try an on-line resize
3975 */
b522adcd
DW		 3976 sector_t sectors;
3977 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3978
58c0fed4
AN		 3979 if (err < 0)
3980 return err;
a35b0d69 3981 if (mddev->pers) {
58c0fed4 3982 err = update_size(mddev, sectors);
850b2b42 3983 md_update_sb(mddev, 1);
a35b0d69 3984 } else {
58c0fed4
AN		 3985 if (mddev->dev_sectors == 0 ||
3986 mddev->dev_sectors > sectors)
3987 mddev->dev_sectors = sectors;
a35b0d69
N		 3988 else
3989 err = -ENOSPC;
3990 }
3991 return err ? err : len;
3992}
3993
3994static struct md_sysfs_entry md_size =
80ca3a44 3995__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3996
8bb93aac
N		 3997
3998/* Metdata version.
e691063a
N		 3999 * This is one of
4000 * 'none' for arrays with no metadata (good luck...)
4001 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 4002 * or N.M for internally known formats
4003 */
4004static ssize_t
fd01b88c 4005metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 4006{
4007 if (mddev->persistent)
4008 return sprintf(page, "%d.%d\n",
4009 mddev->major_version, mddev->minor_version);
e691063a
N		 4010 else if (mddev->external)
4011 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 4012 else
4013 return sprintf(page, "none\n");
4014}
4015
4016static ssize_t
fd01b88c 4017metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 4018{
4019 int major, minor;
4020 char *e;
ea43ddd8
N		 4021 /* Changing the details of 'external' metadata is
4022 * always permitted. Otherwise there must be
4023 * no devices attached to the array.
4024 */
4025 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4026 ;
4027 else if (!list_empty(&mddev->disks))
8bb93aac
N		 4028 return -EBUSY;
4029
4030 if (cmd_match(buf, "none")) {
4031 mddev->persistent = 0;
e691063a
N		 4032 mddev->external = 0;
4033 mddev->major_version = 0;
4034 mddev->minor_version = 90;
4035 return len;
4036 }
4037 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 4038 size_t namelen = len-9;
e691063a
N		 4039 if (namelen >= sizeof(mddev->metadata_type))
4040 namelen = sizeof(mddev->metadata_type)-1;
4041 strncpy(mddev->metadata_type, buf+9, namelen);
4042 mddev->metadata_type[namelen] = 0;
4043 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4044 mddev->metadata_type[--namelen] = 0;
4045 mddev->persistent = 0;
4046 mddev->external = 1;
8bb93aac
N		 4047 mddev->major_version = 0;
4048 mddev->minor_version = 90;
4049 return len;
4050 }
4051 major = simple_strtoul(buf, &e, 10);
4052 if (e==buf || *e != '.')
4053 return -EINVAL;
4054 buf = e+1;
4055 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4056 if (e==buf || (*e && *e != '\n') )
8bb93aac 4057 return -EINVAL;
50511da3 4058 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4059 return -ENOENT;
4060 mddev->major_version = major;
4061 mddev->minor_version = minor;
4062 mddev->persistent = 1;
e691063a 4063 mddev->external = 0;
8bb93aac
N		 4064 return len;
4065}
4066
4067static struct md_sysfs_entry md_metadata =
80ca3a44 4068__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4069
24dd469d 4070static ssize_t
fd01b88c 4071action_show(struct mddev *mddev, char *page)
24dd469d 4072{
7eec314d 4073 char *type = "idle";
b6a9ce68
N		 4074 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4075 type = "frozen";
4076 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4077 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4078 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4079 type = "reshape";
4080 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4081 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4082 type = "resync";
4083 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4084 type = "check";
4085 else
4086 type = "repair";
72a23c21 4087 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4088 type = "recover";
4089 }
4090 return sprintf(page, "%s\n", type);
4091}
4092
fd01b88c 4093static void reap_sync_thread(struct mddev *mddev);
7ebc0be7 4094
24dd469d 4095static ssize_t
fd01b88c 4096action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4097{
7eec314d
N		 4098 if (!mddev->pers || !mddev->pers->sync_request)
4099 return -EINVAL;
4100
b6a9ce68
N		 4101 if (cmd_match(page, "frozen"))
4102 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4103 else
4104 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4105
4106 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4107 if (mddev->sync_thread) {
4108 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4109 reap_sync_thread(mddev);
7eec314d 4110 }
03c902e1
N		 4111 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4112 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4113 return -EBUSY;
72a23c21
NB		 4114 else if (cmd_match(page, "resync"))
4115 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4116 else if (cmd_match(page, "recover")) {
4117 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4118 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4119 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4120 int err;
4121 if (mddev->pers->start_reshape == NULL)
4122 return -EINVAL;
4123 err = mddev->pers->start_reshape(mddev);
4124 if (err)
4125 return err;
a99ac971 4126 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4127 } else {
bce74dac 4128 if (cmd_match(page, "check"))
7eec314d 4129 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4130 else if (!cmd_match(page, "repair"))
7eec314d
N		 4131 return -EINVAL;
4132 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4133 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4134 }
03c902e1 4135 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4136 md_wakeup_thread(mddev->thread);
00bcb4ac 4137 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4138 return len;
4139}
4140
9d88883e 4141static ssize_t
fd01b88c 4142mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4143{
4144 return sprintf(page, "%llu\n",
4145 (unsigned long long) mddev->resync_mismatches);
4146}
4147
80ca3a44
N		 4148static struct md_sysfs_entry md_scan_mode =
4149__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4150
96de1e66 4151
80ca3a44 4152static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4153
88202a0c 4154static ssize_t
fd01b88c 4155sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4156{
4157 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4158 mddev->sync_speed_min ? "local": "system");
4159}
4160
4161static ssize_t
fd01b88c 4162sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4163{
4164 int min;
4165 char *e;
4166 if (strncmp(buf, "system", 6)==0) {
4167 mddev->sync_speed_min = 0;
4168 return len;
4169 }
4170 min = simple_strtoul(buf, &e, 10);
4171 if (buf == e || (*e && *e != '\n') || min <= 0)
4172 return -EINVAL;
4173 mddev->sync_speed_min = min;
4174 return len;
4175}
4176
4177static struct md_sysfs_entry md_sync_min =
4178__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4179
4180static ssize_t
fd01b88c 4181sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4182{
4183 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4184 mddev->sync_speed_max ? "local": "system");
4185}
4186
4187static ssize_t
fd01b88c 4188sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4189{
4190 int max;
4191 char *e;
4192 if (strncmp(buf, "system", 6)==0) {
4193 mddev->sync_speed_max = 0;
4194 return len;
4195 }
4196 max = simple_strtoul(buf, &e, 10);
4197 if (buf == e || (*e && *e != '\n') || max <= 0)
4198 return -EINVAL;
4199 mddev->sync_speed_max = max;
4200 return len;
4201}
4202
4203static struct md_sysfs_entry md_sync_max =
4204__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4205
d7f3d291 4206static ssize_t
fd01b88c 4207degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4208{
4209 return sprintf(page, "%d\n", mddev->degraded);
4210}
4211static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4212
90b08710 4213static ssize_t
fd01b88c 4214sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4215{
4216 return sprintf(page, "%d\n", mddev->parallel_resync);
4217}
4218
4219static ssize_t
fd01b88c 4220sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4221{
4222 long n;
4223
4224 if (strict_strtol(buf, 10, &n))
4225 return -EINVAL;
4226
4227 if (n != 0 && n != 1)
4228 return -EINVAL;
4229
4230 mddev->parallel_resync = n;
4231
4232 if (mddev->sync_thread)
4233 wake_up(&resync_wait);
4234
4235 return len;
4236}
4237
4238/* force parallel resync, even with shared block devices */
4239static struct md_sysfs_entry md_sync_force_parallel =
4240__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4241 sync_force_parallel_show, sync_force_parallel_store);
4242
88202a0c 4243static ssize_t
fd01b88c 4244sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4245{
4246 unsigned long resync, dt, db;
d1a7c503
N		 4247 if (mddev->curr_resync == 0)
4248 return sprintf(page, "none\n");
9687a60c
AN		 4249 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4250 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4251 if (!dt) dt++;
9687a60c
AN		 4252 db = resync - mddev->resync_mark_cnt;
4253 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4254}
4255
80ca3a44 4256static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4257
4258static ssize_t
fd01b88c 4259sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4260{
13ae864b 4261 unsigned long long max_sectors, resync;
88202a0c 4262
acb180b0
N		 4263 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4264 return sprintf(page, "none\n");
4265
88202a0c 4266 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 4267 max_sectors = mddev->resync_max_sectors;
88202a0c 4268 else
58c0fed4 4269 max_sectors = mddev->dev_sectors;
88202a0c 4270
acb180b0 4271 resync = mddev->curr_resync_completed;
13ae864b 4272 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4273}
4274
80ca3a44 4275static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4276
5e96ee65 4277static ssize_t
fd01b88c 4278min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4279{
4280 return sprintf(page, "%llu\n",
4281 (unsigned long long)mddev->resync_min);
4282}
4283static ssize_t
fd01b88c 4284min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4285{
4286 unsigned long long min;
4287 if (strict_strtoull(buf, 10, &min))
4288 return -EINVAL;
4289 if (min > mddev->resync_max)
4290 return -EINVAL;
4291 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4292 return -EBUSY;
4293
4294 /* Must be a multiple of chunk_size */
9d8f0363 4295 if (mddev->chunk_sectors) {
2ac06c33 4296 sector_t temp = min;
9d8f0363 4297 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4298 return -EINVAL;
4299 }
4300 mddev->resync_min = min;
4301
4302 return len;
4303}
4304
4305static struct md_sysfs_entry md_min_sync =
4306__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4307
c6207277 4308static ssize_t
fd01b88c 4309max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4310{
4311 if (mddev->resync_max == MaxSector)
4312 return sprintf(page, "max\n");
4313 else
4314 return sprintf(page, "%llu\n",
4315 (unsigned long long)mddev->resync_max);
4316}
4317static ssize_t
fd01b88c 4318max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4319{
4320 if (strncmp(buf, "max", 3) == 0)
4321 mddev->resync_max = MaxSector;
4322 else {
5e96ee65
NB		 4323 unsigned long long max;
4324 if (strict_strtoull(buf, 10, &max))
4325 return -EINVAL;
4326 if (max < mddev->resync_min)
c6207277
N		 4327 return -EINVAL;
4328 if (max < mddev->resync_max &&
4d484a4a 4329 mddev->ro == 0 &&
c6207277
N		 4330 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4331 return -EBUSY;
4332
4333 /* Must be a multiple of chunk_size */
9d8f0363 4334 if (mddev->chunk_sectors) {
2ac06c33 4335 sector_t temp = max;
9d8f0363 4336 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4337 return -EINVAL;
4338 }
4339 mddev->resync_max = max;
4340 }
4341 wake_up(&mddev->recovery_wait);
4342 return len;
4343}
4344
4345static struct md_sysfs_entry md_max_sync =
4346__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4347
e464eafd 4348static ssize_t
fd01b88c 4349suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4350{
4351 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4352}
4353
4354static ssize_t
fd01b88c 4355suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4356{
4357 char *e;
4358 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4359 unsigned long long old = mddev->suspend_lo;
e464eafd 4360
b8d966ef
N		 4361 if (mddev->pers == NULL ||
4362 mddev->pers->quiesce == NULL)
e464eafd
N		 4363 return -EINVAL;
4364 if (buf == e || (*e && *e != '\n'))
4365 return -EINVAL;
23ddff37
N		 4366
4367 mddev->suspend_lo = new;
4368 if (new >= old)
4369 /* Shrinking suspended region */
e464eafd 4370 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4371 else {
4372 /* Expanding suspended region - need to wait */
4373 mddev->pers->quiesce(mddev, 1);
4374 mddev->pers->quiesce(mddev, 0);
4375 }
4376 return len;
e464eafd
N		 4377}
4378static struct md_sysfs_entry md_suspend_lo =
4379__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4380
4381
4382static ssize_t
fd01b88c 4383suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4384{
4385 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4386}
4387
4388static ssize_t
fd01b88c 4389suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4390{
4391 char *e;
4392 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4393 unsigned long long old = mddev->suspend_hi;
e464eafd 4394
b8d966ef
N		 4395 if (mddev->pers == NULL ||
4396 mddev->pers->quiesce == NULL)
e464eafd
N		 4397 return -EINVAL;
4398 if (buf == e || (*e && *e != '\n'))
4399 return -EINVAL;
23ddff37
N		 4400
4401 mddev->suspend_hi = new;
4402 if (new <= old)
4403 /* Shrinking suspended region */
4404 mddev->pers->quiesce(mddev, 2);
4405 else {
4406 /* Expanding suspended region - need to wait */
e464eafd
N		 4407 mddev->pers->quiesce(mddev, 1);
4408 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4409 }
4410 return len;
e464eafd
N		 4411}
4412static struct md_sysfs_entry md_suspend_hi =
4413__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4414
08a02ecd 4415static ssize_t
fd01b88c 4416reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4417{
4418 if (mddev->reshape_position != MaxSector)
4419 return sprintf(page, "%llu\n",
4420 (unsigned long long)mddev->reshape_position);
4421 strcpy(page, "none\n");
4422 return 5;
4423}
4424
4425static ssize_t
fd01b88c 4426reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd
N		 4427{
4428 char *e;
4429 unsigned long long new = simple_strtoull(buf, &e, 10);
4430 if (mddev->pers)
4431 return -EBUSY;
4432 if (buf == e || (*e && *e != '\n'))
4433 return -EINVAL;
4434 mddev->reshape_position = new;
4435 mddev->delta_disks = 0;
4436 mddev->new_level = mddev->level;
4437 mddev->new_layout = mddev->layout;
664e7c41 4438 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4439 return len;
4440}
4441
4442static struct md_sysfs_entry md_reshape_position =
4443__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4444 reshape_position_store);
4445
b522adcd 4446static ssize_t
fd01b88c 4447array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4448{
4449 if (mddev->external_size)
4450 return sprintf(page, "%llu\n",
4451 (unsigned long long)mddev->array_sectors/2);
4452 else
4453 return sprintf(page, "default\n");
4454}
4455
4456static ssize_t
fd01b88c 4457array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4458{
4459 sector_t sectors;
4460
4461 if (strncmp(buf, "default", 7) == 0) {
4462 if (mddev->pers)
4463 sectors = mddev->pers->size(mddev, 0, 0);
4464 else
4465 sectors = mddev->array_sectors;
4466
4467 mddev->external_size = 0;
4468 } else {
4469 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4470 return -EINVAL;
4471 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4472 return -E2BIG;
b522adcd
DW		 4473
4474 mddev->external_size = 1;
4475 }
4476
4477 mddev->array_sectors = sectors;
cbe6ef1d
N		 4478 if (mddev->pers) {
4479 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4480 revalidate_disk(mddev->gendisk);
cbe6ef1d 4481 }
b522adcd
DW		 4482 return len;
4483}
4484
4485static struct md_sysfs_entry md_array_size =
4486__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4487 array_size_store);
e464eafd 4488
eae1701f
N		 4489static struct attribute *md_default_attrs[] = {
4490 &md_level.attr,
d4dbd025 4491 &md_layout.attr,
eae1701f 4492 &md_raid_disks.attr,
3b34380a 4493 &md_chunk_size.attr,
a35b0d69 4494 &md_size.attr,
a94213b1 4495 &md_resync_start.attr,
8bb93aac 4496 &md_metadata.attr,
6d7ff738 4497 &md_new_device.attr,
16f17b39 4498 &md_safe_delay.attr,
9e653b63 4499 &md_array_state.attr,
08a02ecd 4500 &md_reshape_position.attr,
b522adcd 4501 &md_array_size.attr,
1e50915f 4502 &max_corr_read_errors.attr,
411036fa
N		 4503 NULL,
4504};
4505
4506static struct attribute *md_redundancy_attrs[] = {
24dd469d 4507 &md_scan_mode.attr,
9d88883e 4508 &md_mismatches.attr,
88202a0c
N		 4509 &md_sync_min.attr,
4510 &md_sync_max.attr,
4511 &md_sync_speed.attr,
90b08710 4512 &md_sync_force_parallel.attr,
88202a0c 4513 &md_sync_completed.attr,
5e96ee65 4514 &md_min_sync.attr,
c6207277 4515 &md_max_sync.attr,
e464eafd
N		 4516 &md_suspend_lo.attr,
4517 &md_suspend_hi.attr,
9b1d1dac 4518 &md_bitmap.attr,
d7f3d291 4519 &md_degraded.attr,
eae1701f
N		 4520 NULL,
4521};
411036fa
N		 4522static struct attribute_group md_redundancy_group = {
4523 .name = NULL,
4524 .attrs = md_redundancy_attrs,
4525};
4526
eae1701f
N		 4527
4528static ssize_t
4529md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4530{
4531 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4532 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4533 ssize_t rv;
eae1701f
N		 4534
4535 if (!entry->show)
4536 return -EIO;
af8a2434
N		 4537 spin_lock(&all_mddevs_lock);
4538 if (list_empty(&mddev->all_mddevs)) {
4539 spin_unlock(&all_mddevs_lock);
4540 return -EBUSY;
4541 }
4542 mddev_get(mddev);
4543 spin_unlock(&all_mddevs_lock);
4544
5dc5cf7d
IM		 4545 rv = mddev_lock(mddev);
4546 if (!rv) {
4547 rv = entry->show(mddev, page);
4548 mddev_unlock(mddev);
4549 }
af8a2434 4550 mddev_put(mddev);
96de1e66 4551 return rv;
eae1701f
N		 4552}
4553
4554static ssize_t
4555md_attr_store(struct kobject *kobj, struct attribute *attr,
4556 const char *page, size_t length)
4557{
4558 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4559 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4560 ssize_t rv;
eae1701f
N		 4561
4562 if (!entry->store)
4563 return -EIO;
67463acb
N		 4564 if (!capable(CAP_SYS_ADMIN))
4565 return -EACCES;
af8a2434
N		 4566 spin_lock(&all_mddevs_lock);
4567 if (list_empty(&mddev->all_mddevs)) {
4568 spin_unlock(&all_mddevs_lock);
4569 return -EBUSY;
4570 }
4571 mddev_get(mddev);
4572 spin_unlock(&all_mddevs_lock);
5dc5cf7d
IM		 4573 rv = mddev_lock(mddev);
4574 if (!rv) {
4575 rv = entry->store(mddev, page, length);
4576 mddev_unlock(mddev);
4577 }
af8a2434 4578 mddev_put(mddev);
96de1e66 4579 return rv;
eae1701f
N		 4580}
4581
4582static void md_free(struct kobject *ko)
4583{
fd01b88c 4584 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4585
4586 if (mddev->sysfs_state)
4587 sysfs_put(mddev->sysfs_state);
4588
4589 if (mddev->gendisk) {
4590 del_gendisk(mddev->gendisk);
4591 put_disk(mddev->gendisk);
4592 }
4593 if (mddev->queue)
4594 blk_cleanup_queue(mddev->queue);
4595
eae1701f
N		 4596 kfree(mddev);
4597}
4598
52cf25d0 4599static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4600 .show = md_attr_show,
4601 .store = md_attr_store,
4602};
4603static struct kobj_type md_ktype = {
4604 .release = md_free,
4605 .sysfs_ops = &md_sysfs_ops,
4606 .default_attrs = md_default_attrs,
4607};
4608
1da177e4
LT		 4609int mdp_major = 0;
4610
5fd3a17e
DW		 4611static void mddev_delayed_delete(struct work_struct *ws)
4612{
fd01b88c 4613 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4614
43a70507 4615 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4616 kobject_del(&mddev->kobj);
4617 kobject_put(&mddev->kobj);
4618}
4619
efeb53c0 4620static int md_alloc(dev_t dev, char *name)
1da177e4 4621{
48c9c27b 4622 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4623 struct mddev *mddev = mddev_find(dev);
1da177e4 4624 struct gendisk *disk;
efeb53c0
N		 4625 int partitioned;
4626 int shift;
4627 int unit;
3830c62f 4628 int error;
1da177e4
LT		 4629
4630 if (!mddev)
efeb53c0
N		 4631 return -ENODEV;
4632
4633 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4634 shift = partitioned ? MdpMinorShift : 0;
4635 unit = MINOR(mddev->unit) >> shift;
1da177e4 4636
e804ac78
TH		 4637 /* wait for any previous instance of this device to be
4638 * completely removed (mddev_delayed_delete).
d3374825 4639 */
e804ac78 4640 flush_workqueue(md_misc_wq);
d3374825 4641
48c9c27b 4642 mutex_lock(&disks_mutex);
0909dc44
N		 4643 error = -EEXIST;
4644 if (mddev->gendisk)
4645 goto abort;
efeb53c0
N		 4646
4647 if (name) {
4648 /* Need to ensure that 'name' is not a duplicate.
4649 */
fd01b88c 4650 struct mddev *mddev2;
efeb53c0
N		 4651 spin_lock(&all_mddevs_lock);
4652
4653 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4654 if (mddev2->gendisk &&
4655 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4656 spin_unlock(&all_mddevs_lock);
0909dc44 4657 goto abort;
efeb53c0
N		 4658 }
4659 spin_unlock(&all_mddevs_lock);
1da177e4 4660 }
8b765398 4661
0909dc44 4662 error = -ENOMEM;
8b765398 4663 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4664 if (!mddev->queue)
4665 goto abort;
409c57f3
N		 4666 mddev->queue->queuedata = mddev;
4667
409c57f3 4668 blk_queue_make_request(mddev->queue, md_make_request);
b1bd055d 4669 blk_set_stacking_limits(&mddev->queue->limits);
8b765398 4670
1da177e4
LT		 4671 disk = alloc_disk(1 << shift);
4672 if (!disk) {
8b765398
N		 4673 blk_cleanup_queue(mddev->queue);
4674 mddev->queue = NULL;
0909dc44 4675 goto abort;
1da177e4 4676 }
efeb53c0 4677 disk->major = MAJOR(mddev->unit);
1da177e4 4678 disk->first_minor = unit << shift;
efeb53c0
N		 4679 if (name)
4680 strcpy(disk->disk_name, name);
4681 else if (partitioned)
1da177e4 4682 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4683 else
1da177e4 4684 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4685 disk->fops = &md_fops;
4686 disk->private_data = mddev;
4687 disk->queue = mddev->queue;
b0140891 4688 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4689 /* Allow extended partitions. This makes the
d3374825 4690 * 'mdp' device redundant, but we can't really
92850bbd
N		 4691 * remove it now.
4692 */
4693 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4694 mddev->gendisk = disk;
b0140891
N		 4695 /* As soon as we call add_disk(), another thread could get
4696 * through to md_open, so make sure it doesn't get too far
4697 */
4698 mutex_lock(&mddev->open_mutex);
4699 add_disk(disk);
4700
ed9e1982
TH		 4701 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4702 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4703 if (error) {
4704 /* This isn't possible, but as kobject_init_and_add is marked
4705 * __must_check, we must do something with the result
4706 */
5e55e2f5
N		 4707 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4708 disk->disk_name);
0909dc44
N		 4709 error = 0;
4710 }
00bcb4ac
N		 4711 if (mddev->kobj.sd &&
4712 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4713 printk(KERN_DEBUG "pointless warning\n");
b0140891 4714 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4715 abort:
4716 mutex_unlock(&disks_mutex);
00bcb4ac 4717 if (!error && mddev->kobj.sd) {
3830c62f 4718 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4719 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4720 }
d3374825 4721 mddev_put(mddev);
0909dc44 4722 return error;
efeb53c0
N		 4723}
4724
4725static struct kobject *md_probe(dev_t dev, int *part, void *data)
4726{
4727 md_alloc(dev, NULL);
1da177e4
LT		 4728 return NULL;
4729}
4730
efeb53c0
N		 4731static int add_named_array(const char *val, struct kernel_param *kp)
4732{
4733 /* val must be "md_*" where * is not all digits.
4734 * We allocate an array with a large free minor number, and
4735 * set the name to val. val must not already be an active name.
4736 */
4737 int len = strlen(val);
4738 char buf[DISK_NAME_LEN];
4739
4740 while (len && val[len-1] == '\n')
4741 len--;
4742 if (len >= DISK_NAME_LEN)
4743 return -E2BIG;
4744 strlcpy(buf, val, len+1);
4745 if (strncmp(buf, "md_", 3) != 0)
4746 return -EINVAL;
4747 return md_alloc(0, buf);
4748}
4749
1da177e4
LT		 4750static void md_safemode_timeout(unsigned long data)
4751{
fd01b88c 4752 struct mddev *mddev = (struct mddev *) data;
1da177e4 4753
0fd62b86
NB		 4754 if (!atomic_read(&mddev->writes_pending)) {
4755 mddev->safemode = 1;
4756 if (mddev->external)
00bcb4ac 4757 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4758 }
1da177e4
LT		 4759 md_wakeup_thread(mddev->thread);
4760}
4761
6ff8d8ec 4762static int start_dirty_degraded;
1da177e4 4763
fd01b88c 4764int md_run(struct mddev *mddev)
1da177e4 4765{
2604b703 4766 int err;
3cb03002 4767 struct md_rdev *rdev;
84fc4b56 4768 struct md_personality *pers;
1da177e4 4769
a757e64c
N		 4770 if (list_empty(&mddev->disks))
4771 /* cannot run an array with no devices.. */
1da177e4 4772 return -EINVAL;
1da177e4
LT		 4773
4774 if (mddev->pers)
4775 return -EBUSY;
bb4f1e9d
N		 4776 /* Cannot run until previous stop completes properly */
4777 if (mddev->sysfs_active)
4778 return -EBUSY;
b6eb127d 4779
1da177e4
LT		 4780 /*
4781 * Analyze all RAID superblock(s)
4782 */
1ec4a939
N		 4783 if (!mddev->raid_disks) {
4784 if (!mddev->persistent)
4785 return -EINVAL;
a757e64c 4786 analyze_sbs(mddev);
1ec4a939 4787 }
1da177e4 4788
d9d166c2
N		 4789 if (mddev->level != LEVEL_NONE)
4790 request_module("md-level-%d", mddev->level);
4791 else if (mddev->clevel[0])
4792 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4793
4794 /*
4795 * Drop all container device buffers, from now on
4796 * the only valid external interface is through the md
4797 * device.
1da177e4 4798 */
dafb20fa 4799 rdev_for_each(rdev, mddev) {
b2d444d7 4800 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4801 continue;
4802 sync_blockdev(rdev->bdev);
f98393a6 4803 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4804
4805 /* perform some consistency tests on the device.
4806 * We don't want the data to overlap the metadata,
58c0fed4 4807 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4808 */
a6ff7e08
JB		 4809 if (rdev->meta_bdev) {
4810 /* Nothing to check */;
4811 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4812 if (mddev->dev_sectors &&
4813 rdev->data_offset + mddev->dev_sectors
0f420358 4814 > rdev->sb_start) {
f0d76d70
N		 4815 printk("md: %s: data overlaps metadata\n",
4816 mdname(mddev));
4817 return -EINVAL;
4818 }
4819 } else {
0f420358 4820 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4821 > rdev->data_offset) {
4822 printk("md: %s: metadata overlaps data\n",
4823 mdname(mddev));
4824 return -EINVAL;
4825 }
4826 }
00bcb4ac 4827 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4828 }
4829
a167f663 4830 if (mddev->bio_set == NULL)
a519b26d 4831 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 4832 sizeof(struct mddev *));
a167f663 4833
1da177e4 4834 spin_lock(&pers_lock);
d9d166c2 4835 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4836 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4837 spin_unlock(&pers_lock);
d9d166c2
N		 4838 if (mddev->level != LEVEL_NONE)
4839 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4840 mddev->level);
4841 else
4842 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4843 mddev->clevel);
1da177e4
LT		 4844 return -EINVAL;
4845 }
2604b703 4846 mddev->pers = pers;
1da177e4 4847 spin_unlock(&pers_lock);
34817e8c
N		 4848 if (mddev->level != pers->level) {
4849 mddev->level = pers->level;
4850 mddev->new_level = pers->level;
4851 }
d9d166c2 4852 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4853
f6705578 4854 if (mddev->reshape_position != MaxSector &&
63c70c4f 4855 pers->start_reshape == NULL) {
f6705578
N		 4856 /* This personality cannot handle reshaping... */
4857 mddev->pers = NULL;
4858 module_put(pers->owner);
4859 return -EINVAL;
4860 }
4861
7dd5e7c3
N		 4862 if (pers->sync_request) {
4863 /* Warn if this is a potentially silly
4864 * configuration.
4865 */
4866 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 4867 struct md_rdev *rdev2;
7dd5e7c3 4868 int warned = 0;
159ec1fc 4869
dafb20fa
N		 4870 rdev_for_each(rdev, mddev)
4871 rdev_for_each(rdev2, mddev) {
7dd5e7c3
N		 4872 if (rdev < rdev2 &&
4873 rdev->bdev->bd_contains ==
4874 rdev2->bdev->bd_contains) {
4875 printk(KERN_WARNING
4876 "%s: WARNING: %s appears to be"
4877 " on the same physical disk as"
4878 " %s.\n",
4879 mdname(mddev),
4880 bdevname(rdev->bdev,b),
4881 bdevname(rdev2->bdev,b2));
4882 warned = 1;
4883 }
4884 }
159ec1fc 4885
7dd5e7c3
N		 4886 if (warned)
4887 printk(KERN_WARNING
4888 "True protection against single-disk"
4889 " failure might be compromised.\n");
4890 }
4891
657390d2 4892 mddev->recovery = 0;
58c0fed4
AN		 4893 /* may be over-ridden by personality */
4894 mddev->resync_max_sectors = mddev->dev_sectors;
4895
6ff8d8ec 4896 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4897
0f9552b5 4898 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4899 mddev->ro = 2; /* read-only, but switch on first write */
4900
b15c2e57 4901 err = mddev->pers->run(mddev);
13e53df3
AN		 4902 if (err)
4903 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4904 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4905 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4906 " but 'external_size' not in effect?\n", __func__);
4907 printk(KERN_ERR
4908 "md: invalid array_size %llu > default size %llu\n",
4909 (unsigned long long)mddev->array_sectors / 2,
4910 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4911 err = -EINVAL;
4912 mddev->pers->stop(mddev);
4913 }
4914 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4915 err = bitmap_create(mddev);
4916 if (err) {
4917 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4918 mdname(mddev), err);
4919 mddev->pers->stop(mddev);
4920 }
4921 }
1da177e4 4922 if (err) {
1da177e4
LT		 4923 module_put(mddev->pers->owner);
4924 mddev->pers = NULL;
32a7627c
N		 4925 bitmap_destroy(mddev);
4926 return err;
1da177e4 4927 }
5e55e2f5 4928 if (mddev->pers->sync_request) {
00bcb4ac
N		 4929 if (mddev->kobj.sd &&
4930 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4931 printk(KERN_WARNING
4932 "md: cannot register extra attributes for %s\n",
4933 mdname(mddev));
00bcb4ac 4934 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4935 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4936 mddev->ro = 0;
4937
1da177e4 4938 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4939 atomic_set(&mddev->max_corr_read_errors,
4940 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4941 mddev->safemode = 0;
4942 mddev->safemode_timer.function = md_safemode_timeout;
4943 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4944 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4945 mddev->in_sync = 1;
0ca69886
N		 4946 smp_wmb();
4947 mddev->ready = 1;
dafb20fa 4948 rdev_for_each(rdev, mddev)
36fad858
NK		 4949 if (rdev->raid_disk >= 0)
4950 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 4951 /* failure here is OK */;
1da177e4
LT		 4952
4953 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4954
850b2b42
N		 4955 if (mddev->flags)
4956 md_update_sb(mddev, 0);
1da177e4 4957
d7603b7e 4958 md_new_event(mddev);
00bcb4ac
N		 4959 sysfs_notify_dirent_safe(mddev->sysfs_state);
4960 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4961 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4962 return 0;
4963}
390ee602 4964EXPORT_SYMBOL_GPL(md_run);
1da177e4 4965
fd01b88c 4966static int do_md_run(struct mddev *mddev)
fe60b014
N		 4967{
4968 int err;
4969
4970 err = md_run(mddev);
4971 if (err)
4972 goto out;
69e51b44
N		 4973 err = bitmap_load(mddev);
4974 if (err) {
4975 bitmap_destroy(mddev);
4976 goto out;
4977 }
0fd018af
JB		 4978
4979 md_wakeup_thread(mddev->thread);
4980 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4981
fe60b014
N		 4982 set_capacity(mddev->gendisk, mddev->array_sectors);
4983 revalidate_disk(mddev->gendisk);
f0b4f7e2 4984 mddev->changed = 1;
fe60b014
N		 4985 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4986out:
4987 return err;
4988}
4989
fd01b88c 4990static int restart_array(struct mddev *mddev)
1da177e4
LT		 4991{
4992 struct gendisk *disk = mddev->gendisk;
1da177e4 4993
80fab1d7 4994 /* Complain if it has no devices */
1da177e4 4995 if (list_empty(&mddev->disks))
80fab1d7
AN		 4996 return -ENXIO;
4997 if (!mddev->pers)
4998 return -EINVAL;
4999 if (!mddev->ro)
5000 return -EBUSY;
5001 mddev->safemode = 0;
5002 mddev->ro = 0;
5003 set_disk_ro(disk, 0);
5004 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5005 mdname(mddev));
5006 /* Kick recovery or resync if necessary */
5007 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5008 md_wakeup_thread(mddev->thread);
5009 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 5010 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 5011 return 0;
1da177e4
LT		 5012}
5013
acc55e22
N		 5014/* similar to deny_write_access, but accounts for our holding a reference
5015 * to the file ourselves */
5016static int deny_bitmap_write_access(struct file * file)
5017{
5018 struct inode *inode = file->f_mapping->host;
5019
5020 spin_lock(&inode->i_lock);
5021 if (atomic_read(&inode->i_writecount) > 1) {
5022 spin_unlock(&inode->i_lock);
5023 return -ETXTBSY;
5024 }
5025 atomic_set(&inode->i_writecount, -1);
5026 spin_unlock(&inode->i_lock);
5027
5028 return 0;
5029}
5030
43a70507 5031void restore_bitmap_write_access(struct file *file)
acc55e22
N		 5032{
5033 struct inode *inode = file->f_mapping->host;
5034
5035 spin_lock(&inode->i_lock);
5036 atomic_set(&inode->i_writecount, 1);
5037 spin_unlock(&inode->i_lock);
5038}
5039
fd01b88c 5040static void md_clean(struct mddev *mddev)
6177b472
N		 5041{
5042 mddev->array_sectors = 0;
5043 mddev->external_size = 0;
5044 mddev->dev_sectors = 0;
5045 mddev->raid_disks = 0;
5046 mddev->recovery_cp = 0;
5047 mddev->resync_min = 0;
5048 mddev->resync_max = MaxSector;
5049 mddev->reshape_position = MaxSector;
5050 mddev->external = 0;
5051 mddev->persistent = 0;
5052 mddev->level = LEVEL_NONE;
5053 mddev->clevel[0] = 0;
5054 mddev->flags = 0;
5055 mddev->ro = 0;
5056 mddev->metadata_type[0] = 0;
5057 mddev->chunk_sectors = 0;
5058 mddev->ctime = mddev->utime = 0;
5059 mddev->layout = 0;
5060 mddev->max_disks = 0;
5061 mddev->events = 0;
a8707c08 5062 mddev->can_decrease_events = 0;
6177b472
N		 5063 mddev->delta_disks = 0;
5064 mddev->new_level = LEVEL_NONE;
5065 mddev->new_layout = 0;
5066 mddev->new_chunk_sectors = 0;
5067 mddev->curr_resync = 0;
5068 mddev->resync_mismatches = 0;
5069 mddev->suspend_lo = mddev->suspend_hi = 0;
5070 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5071 mddev->recovery = 0;
5072 mddev->in_sync = 0;
f0b4f7e2 5073 mddev->changed = 0;
6177b472 5074 mddev->degraded = 0;
6177b472 5075 mddev->safemode = 0;
050b6615 5076 mddev->merge_check_needed = 0;
6177b472
N		 5077 mddev->bitmap_info.offset = 0;
5078 mddev->bitmap_info.default_offset = 0;
5079 mddev->bitmap_info.chunksize = 0;
5080 mddev->bitmap_info.daemon_sleep = 0;
5081 mddev->bitmap_info.max_write_behind = 0;
5082}
5083
fd01b88c 5084static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5085{
5086 if (mddev->sync_thread) {
5087 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5088 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5089 reap_sync_thread(mddev);
a047e125
N		 5090 }
5091
5092 del_timer_sync(&mddev->safemode_timer);
5093
5094 bitmap_flush(mddev);
5095 md_super_wait(mddev);
5096
5097 if (!mddev->in_sync || mddev->flags) {
5098 /* mark array as shutdown cleanly */
5099 mddev->in_sync = 1;
5100 md_update_sb(mddev, 1);
5101 }
5102}
defad61a 5103
fd01b88c 5104void md_stop_writes(struct mddev *mddev)
defad61a
N		 5105{
5106 mddev_lock(mddev);
5107 __md_stop_writes(mddev);
5108 mddev_unlock(mddev);
5109}
390ee602 5110EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5111
fd01b88c 5112void md_stop(struct mddev *mddev)
6177b472 5113{
0ca69886 5114 mddev->ready = 0;
6177b472
N		 5115 mddev->pers->stop(mddev);
5116 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5117 mddev->to_remove = &md_redundancy_group;
5118 module_put(mddev->pers->owner);
5119 mddev->pers = NULL;
cca9cf90 5120 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5121}
390ee602 5122EXPORT_SYMBOL_GPL(md_stop);
6177b472 5123
fd01b88c 5124static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N		 5125{
5126 int err = 0;
5127 mutex_lock(&mddev->open_mutex);
5128 if (atomic_read(&mddev->openers) > is_open) {
5129 printk("md: %s still in use.\n",mdname(mddev));
5130 err = -EBUSY;
5131 goto out;
5132 }
5133 if (mddev->pers) {
defad61a 5134 __md_stop_writes(mddev);
a4bd82d0
N		 5135
5136 err = -ENXIO;
5137 if (mddev->ro==1)
5138 goto out;
5139 mddev->ro = 1;
5140 set_disk_ro(mddev->gendisk, 1);
5141 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5142 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5143 err = 0;
5144 }
5145out:
5146 mutex_unlock(&mddev->open_mutex);
5147 return err;
5148}
5149
9e653b63
N		 5150/* mode:
5151 * 0 - completely stop and dis-assemble array
9e653b63
N		 5152 * 2 - stop but do not disassemble array
5153 */
fd01b88c 5154static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5155{
1da177e4 5156 struct gendisk *disk = mddev->gendisk;
3cb03002 5157 struct md_rdev *rdev;
1da177e4 5158
c8c00a69 5159 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 5160 if (atomic_read(&mddev->openers) > is_open ||
5161 mddev->sysfs_active) {
df5b20cf 5162 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5163 mutex_unlock(&mddev->open_mutex);
5164 return -EBUSY;
5165 }
1da177e4 5166
6e17b027 5167 if (mddev->pers) {
a4bd82d0
N		 5168 if (mddev->ro)
5169 set_disk_ro(disk, 0);
409c57f3 5170
defad61a 5171 __md_stop_writes(mddev);
a4bd82d0
N		 5172 md_stop(mddev);
5173 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5174 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5175
a4bd82d0 5176 /* tell userspace to handle 'inactive' */
00bcb4ac 5177 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5178
dafb20fa 5179 rdev_for_each(rdev, mddev)
36fad858
NK		 5180 if (rdev->raid_disk >= 0)
5181 sysfs_unlink_rdev(mddev, rdev);
c4647292 5182
a4bd82d0 5183 set_capacity(disk, 0);
6e17b027 5184 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5185 mddev->changed = 1;
a4bd82d0 5186 revalidate_disk(disk);
0d4ca600 5187
a4bd82d0
N		 5188 if (mddev->ro)
5189 mddev->ro = 0;
6e17b027
N		 5190 } else
5191 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5192 /*
5193 * Free resources if final stop
5194 */
9e653b63 5195 if (mode == 0) {
1da177e4
LT		 5196 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5197
978f946b 5198 bitmap_destroy(mddev);
c3d9714e
N		 5199 if (mddev->bitmap_info.file) {
5200 restore_bitmap_write_access(mddev->bitmap_info.file);
5201 fput(mddev->bitmap_info.file);
5202 mddev->bitmap_info.file = NULL;
978f946b 5203 }
c3d9714e 5204 mddev->bitmap_info.offset = 0;
978f946b 5205
1da177e4
LT		 5206 export_array(mddev);
5207
6177b472 5208 md_clean(mddev);
934d9c23 5209 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5210 if (mddev->hold_active == UNTIL_STOP)
5211 mddev->hold_active = 0;
a4bd82d0 5212 }
3f9d99c1 5213 blk_integrity_unregister(disk);
d7603b7e 5214 md_new_event(mddev);
00bcb4ac 5215 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5216 return 0;
1da177e4
LT		 5217}
5218
fdee8ae4 5219#ifndef MODULE
fd01b88c 5220static void autorun_array(struct mddev *mddev)
1da177e4 5221{
3cb03002 5222 struct md_rdev *rdev;
1da177e4
LT		 5223 int err;
5224
a757e64c 5225 if (list_empty(&mddev->disks))
1da177e4 5226 return;
1da177e4
LT		 5227
5228 printk(KERN_INFO "md: running: ");
5229
dafb20fa 5230 rdev_for_each(rdev, mddev) {
1da177e4
LT		 5231 char b[BDEVNAME_SIZE];
5232 printk("<%s>", bdevname(rdev->bdev,b));
5233 }
5234 printk("\n");
5235
d710e138 5236 err = do_md_run(mddev);
1da177e4
LT		 5237 if (err) {
5238 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5239 do_md_stop(mddev, 0, 0);
1da177e4
LT		 5240 }
5241}
5242
5243/*
5244 * lets try to run arrays based on all disks that have arrived
5245 * until now. (those are in pending_raid_disks)
5246 *
5247 * the method: pick the first pending disk, collect all disks with
5248 * the same UUID, remove all from the pending list and put them into
5249 * the 'same_array' list. Then order this list based on superblock
5250 * update time (freshest comes first), kick out 'old' disks and
5251 * compare superblocks. If everything's fine then run it.
5252 *
5253 * If "unit" is allocated, then bump its reference count
5254 */
5255static void autorun_devices(int part)
5256{
3cb03002 5257 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5258 struct mddev *mddev;
1da177e4
LT		 5259 char b[BDEVNAME_SIZE];
5260
5261 printk(KERN_INFO "md: autorun ...\n");
5262 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5263 int unit;
1da177e4 5264 dev_t dev;
ad01c9e3 5265 LIST_HEAD(candidates);
1da177e4 5266 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5267 struct md_rdev, same_set);
1da177e4
LT		 5268
5269 printk(KERN_INFO "md: considering %s ...\n",
5270 bdevname(rdev0->bdev,b));
5271 INIT_LIST_HEAD(&candidates);
159ec1fc 5272 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5273 if (super_90_load(rdev, rdev0, 0) >= 0) {
5274 printk(KERN_INFO "md: adding %s ...\n",
5275 bdevname(rdev->bdev,b));
5276 list_move(&rdev->same_set, &candidates);
5277 }
5278 /*
5279 * now we have a set of devices, with all of them having
5280 * mostly sane superblocks. It's time to allocate the
5281 * mddev.
5282 */
e8703fe1
N		 5283 if (part) {
5284 dev = MKDEV(mdp_major,
5285 rdev0->preferred_minor << MdpMinorShift);
5286 unit = MINOR(dev) >> MdpMinorShift;
5287 } else {
5288 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5289 unit = MINOR(dev);
5290 }
5291 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5292 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5293 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5294 break;
5295 }
1da177e4
LT		 5296
5297 md_probe(dev, NULL, NULL);
5298 mddev = mddev_find(dev);
9bbbca3a
NB		 5299 if (!mddev || !mddev->gendisk) {
5300 if (mddev)
5301 mddev_put(mddev);
5302 printk(KERN_ERR
1da177e4
LT		 5303 "md: cannot allocate memory for md drive.\n");
5304 break;
5305 }
5306 if (mddev_lock(mddev))
5307 printk(KERN_WARNING "md: %s locked, cannot run\n",
5308 mdname(mddev));
5309 else if (mddev->raid_disks || mddev->major_version
5310 || !list_empty(&mddev->disks)) {
5311 printk(KERN_WARNING
5312 "md: %s already running, cannot run %s\n",
5313 mdname(mddev), bdevname(rdev0->bdev,b));
5314 mddev_unlock(mddev);
5315 } else {
5316 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5317 mddev->persistent = 1;
159ec1fc 5318 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5319 list_del_init(&rdev->same_set);
5320 if (bind_rdev_to_array(rdev, mddev))
5321 export_rdev(rdev);
5322 }
5323 autorun_array(mddev);
5324 mddev_unlock(mddev);
5325 }
5326 /* on success, candidates will be empty, on error
5327 * it won't...
5328 */
159ec1fc 5329 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5330 list_del_init(&rdev->same_set);
1da177e4 5331 export_rdev(rdev);
4b80991c 5332 }
1da177e4
LT		 5333 mddev_put(mddev);
5334 }
5335 printk(KERN_INFO "md: ... autorun DONE.\n");
5336}
fdee8ae4 5337#endif /* !MODULE */
1da177e4 5338
1da177e4
LT		 5339static int get_version(void __user * arg)
5340{
5341 mdu_version_t ver;
5342
5343 ver.major = MD_MAJOR_VERSION;
5344 ver.minor = MD_MINOR_VERSION;
5345 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5346
5347 if (copy_to_user(arg, &ver, sizeof(ver)))
5348 return -EFAULT;
5349
5350 return 0;
5351}
5352
fd01b88c 5353static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5354{
5355 mdu_array_info_t info;
a9f326eb 5356 int nr,working,insync,failed,spare;
3cb03002 5357 struct md_rdev *rdev;
1da177e4 5358
a9f326eb 5359 nr=working=insync=failed=spare=0;
dafb20fa 5360 rdev_for_each(rdev, mddev) {
1da177e4 5361 nr++;
b2d444d7 5362 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5363 failed++;
5364 else {
5365 working++;
b2d444d7 5366 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5367 insync++;
1da177e4
LT		 5368 else
5369 spare++;
5370 }
5371 }
5372
5373 info.major_version = mddev->major_version;
5374 info.minor_version = mddev->minor_version;
5375 info.patch_version = MD_PATCHLEVEL_VERSION;
5376 info.ctime = mddev->ctime;
5377 info.level = mddev->level;
58c0fed4
AN		 5378 info.size = mddev->dev_sectors / 2;
5379 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5380 info.size = -1;
1da177e4
LT		 5381 info.nr_disks = nr;
5382 info.raid_disks = mddev->raid_disks;
5383 info.md_minor = mddev->md_minor;
5384 info.not_persistent= !mddev->persistent;
5385
5386 info.utime = mddev->utime;
5387 info.state = 0;
5388 if (mddev->in_sync)
5389 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5390 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5391 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5392 info.active_disks = insync;
1da177e4
LT		 5393 info.working_disks = working;
5394 info.failed_disks = failed;
5395 info.spare_disks = spare;
5396
5397 info.layout = mddev->layout;
9d8f0363 5398 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5399
5400 if (copy_to_user(arg, &info, sizeof(info)))
5401 return -EFAULT;
5402
5403 return 0;
5404}
5405
fd01b88c 5406static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5407{
5408 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5409 char *ptr, *buf = NULL;
5410 int err = -ENOMEM;
5411
b5470dc5
DW		 5412 if (md_allow_write(mddev))
5413 file = kmalloc(sizeof(*file), GFP_NOIO);
5414 else
5415 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5416
32a7627c
N		 5417 if (!file)
5418 goto out;
5419
5420 /* bitmap disabled, zero the first byte and copy out */
5421 if (!mddev->bitmap || !mddev->bitmap->file) {
5422 file->pathname[0] = '\0';
5423 goto copy_out;
5424 }
5425
5426 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5427 if (!buf)
5428 goto out;
5429
6bcfd601
CH		 5430 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5431 if (IS_ERR(ptr))
32a7627c
N		 5432 goto out;
5433
5434 strcpy(file->pathname, ptr);
5435
5436copy_out:
5437 err = 0;
5438 if (copy_to_user(arg, file, sizeof(*file)))
5439 err = -EFAULT;
5440out:
5441 kfree(buf);
5442 kfree(file);
5443 return err;
5444}
5445
fd01b88c 5446static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5447{
5448 mdu_disk_info_t info;
3cb03002 5449 struct md_rdev *rdev;
1da177e4
LT		 5450
5451 if (copy_from_user(&info, arg, sizeof(info)))
5452 return -EFAULT;
5453
26ef379f 5454 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5455 if (rdev) {
5456 info.major = MAJOR(rdev->bdev->bd_dev);
5457 info.minor = MINOR(rdev->bdev->bd_dev);
5458 info.raid_disk = rdev->raid_disk;
5459 info.state = 0;
b2d444d7 5460 if (test_bit(Faulty, &rdev->flags))
1da177e4 5461 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5462 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5463 info.state |= (1<<MD_DISK_ACTIVE);
5464 info.state |= (1<<MD_DISK_SYNC);
5465 }
8ddf9efe
N		 5466 if (test_bit(WriteMostly, &rdev->flags))
5467 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5468 } else {
5469 info.major = info.minor = 0;
5470 info.raid_disk = -1;
5471 info.state = (1<<MD_DISK_REMOVED);
5472 }
5473
5474 if (copy_to_user(arg, &info, sizeof(info)))
5475 return -EFAULT;
5476
5477 return 0;
5478}
5479
fd01b88c 5480static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5481{
5482 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5483 struct md_rdev *rdev;
1da177e4
LT		 5484 dev_t dev = MKDEV(info->major,info->minor);
5485
5486 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5487 return -EOVERFLOW;
5488
5489 if (!mddev->raid_disks) {
5490 int err;
5491 /* expecting a device which has a superblock */
5492 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5493 if (IS_ERR(rdev)) {
5494 printk(KERN_WARNING
5495 "md: md_import_device returned %ld\n",
5496 PTR_ERR(rdev));
5497 return PTR_ERR(rdev);
5498 }
5499 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5500 struct md_rdev *rdev0
5501 = list_entry(mddev->disks.next,
5502 struct md_rdev, same_set);
a9f326eb 5503 err = super_types[mddev->major_version]
1da177e4
LT		 5504 .load_super(rdev, rdev0, mddev->minor_version);
5505 if (err < 0) {
5506 printk(KERN_WARNING
5507 "md: %s has different UUID to %s\n",
5508 bdevname(rdev->bdev,b),
5509 bdevname(rdev0->bdev,b2));
5510 export_rdev(rdev);
5511 return -EINVAL;
5512 }
5513 }
5514 err = bind_rdev_to_array(rdev, mddev);
5515 if (err)
5516 export_rdev(rdev);
5517 return err;
5518 }
5519
5520 /*
5521 * add_new_disk can be used once the array is assembled
5522 * to add "hot spares". They must already have a superblock
5523 * written
5524 */
5525 if (mddev->pers) {
5526 int err;
5527 if (!mddev->pers->hot_add_disk) {
5528 printk(KERN_WARNING
5529 "%s: personality does not support diskops!\n",
5530 mdname(mddev));
5531 return -EINVAL;
5532 }
7b1e35f6
N		 5533 if (mddev->persistent)
5534 rdev = md_import_device(dev, mddev->major_version,
5535 mddev->minor_version);
5536 else
5537 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5538 if (IS_ERR(rdev)) {
5539 printk(KERN_WARNING
5540 "md: md_import_device returned %ld\n",
5541 PTR_ERR(rdev));
5542 return PTR_ERR(rdev);
5543 }
1a855a06 5544 /* set saved_raid_disk if appropriate */
41158c7e
N		 5545 if (!mddev->persistent) {
5546 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5547 info->raid_disk < mddev->raid_disks) {
41158c7e 5548 rdev->raid_disk = info->raid_disk;
bf572541
N		 5549 set_bit(In_sync, &rdev->flags);
5550 } else
41158c7e
N		 5551 rdev->raid_disk = -1;
5552 } else
5553 super_types[mddev->major_version].
5554 validate_super(mddev, rdev);
bedd86b7
N		 5555 if ((info->state & (1<<MD_DISK_SYNC)) &&
5556 (!test_bit(In_sync, &rdev->flags) ||
5557 rdev->raid_disk != info->raid_disk)) {
5558 /* This was a hot-add request, but events doesn't
5559 * match, so reject it.
5560 */
5561 export_rdev(rdev);
5562 return -EINVAL;
5563 }
5564
1a855a06
N		 5565 if (test_bit(In_sync, &rdev->flags))
5566 rdev->saved_raid_disk = rdev->raid_disk;
5567 else
5568 rdev->saved_raid_disk = -1;
41158c7e 5569
b2d444d7 5570 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5571 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5572 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5573 else
5574 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5575
1da177e4
LT		 5576 rdev->raid_disk = -1;
5577 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5578 if (!err && !mddev->pers->hot_remove_disk) {
5579 /* If there is hot_add_disk but no hot_remove_disk
5580 * then added disks for geometry changes,
5581 * and should be added immediately.
5582 */
5583 super_types[mddev->major_version].
5584 validate_super(mddev, rdev);
5585 err = mddev->pers->hot_add_disk(mddev, rdev);
5586 if (err)
5587 unbind_rdev_from_array(rdev);
5588 }
1da177e4
LT		 5589 if (err)
5590 export_rdev(rdev);
52664732 5591 else
00bcb4ac 5592 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5593
17571284 5594 md_update_sb(mddev, 1);
72a23c21
NB		 5595 if (mddev->degraded)
5596 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5597 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5598 if (!err)
5599 md_new_event(mddev);
005eca5e 5600 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5601 return err;
5602 }
5603
5604 /* otherwise, add_new_disk is only allowed
5605 * for major_version==0 superblocks
5606 */
5607 if (mddev->major_version != 0) {
5608 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5609 mdname(mddev));
5610 return -EINVAL;
5611 }
5612
5613 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5614 int err;
d710e138 5615 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5616 if (IS_ERR(rdev)) {
5617 printk(KERN_WARNING
5618 "md: error, md_import_device() returned %ld\n",
5619 PTR_ERR(rdev));
5620 return PTR_ERR(rdev);
5621 }
5622 rdev->desc_nr = info->number;
5623 if (info->raid_disk < mddev->raid_disks)
5624 rdev->raid_disk = info->raid_disk;
5625 else
5626 rdev->raid_disk = -1;
5627
1da177e4 5628 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5629 if (info->state & (1<<MD_DISK_SYNC))
5630 set_bit(In_sync, &rdev->flags);
1da177e4 5631
8ddf9efe
N		 5632 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5633 set_bit(WriteMostly, &rdev->flags);
5634
1da177e4
LT		 5635 if (!mddev->persistent) {
5636 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5637 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5638 } else
57b2caa3 5639 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5640 rdev->sectors = rdev->sb_start;
1da177e4 5641
2bf071bf
N		 5642 err = bind_rdev_to_array(rdev, mddev);
5643 if (err) {
5644 export_rdev(rdev);
5645 return err;
5646 }
1da177e4
LT		 5647 }
5648
5649 return 0;
5650}
5651
fd01b88c 5652static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5653{
5654 char b[BDEVNAME_SIZE];
3cb03002 5655 struct md_rdev *rdev;
1da177e4 5656
1da177e4
LT		 5657 rdev = find_rdev(mddev, dev);
5658 if (!rdev)
5659 return -ENXIO;
5660
5661 if (rdev->raid_disk >= 0)
5662 goto busy;
5663
5664 kick_rdev_from_array(rdev);
850b2b42 5665 md_update_sb(mddev, 1);
d7603b7e 5666 md_new_event(mddev);
1da177e4
LT		 5667
5668 return 0;
5669busy:
fdefa4d8 5670 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5671 bdevname(rdev->bdev,b), mdname(mddev));
5672 return -EBUSY;
5673}
5674
fd01b88c 5675static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5676{
5677 char b[BDEVNAME_SIZE];
5678 int err;
3cb03002 5679 struct md_rdev *rdev;
1da177e4
LT		 5680
5681 if (!mddev->pers)
5682 return -ENODEV;
5683
5684 if (mddev->major_version != 0) {
5685 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5686 " version-0 superblocks.\n",
5687 mdname(mddev));
5688 return -EINVAL;
5689 }
5690 if (!mddev->pers->hot_add_disk) {
5691 printk(KERN_WARNING
5692 "%s: personality does not support diskops!\n",
5693 mdname(mddev));
5694 return -EINVAL;
5695 }
5696
d710e138 5697 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5698 if (IS_ERR(rdev)) {
5699 printk(KERN_WARNING
5700 "md: error, md_import_device() returned %ld\n",
5701 PTR_ERR(rdev));
5702 return -EINVAL;
5703 }
5704
5705 if (mddev->persistent)
57b2caa3 5706 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5707 else
77304d2a 5708 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5709
8190e754 5710 rdev->sectors = rdev->sb_start;
1da177e4 5711
b2d444d7 5712 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5713 printk(KERN_WARNING
5714 "md: can not hot-add faulty %s disk to %s!\n",
5715 bdevname(rdev->bdev,b), mdname(mddev));
5716 err = -EINVAL;
5717 goto abort_export;
5718 }
b2d444d7 5719 clear_bit(In_sync, &rdev->flags);
1da177e4 5720 rdev->desc_nr = -1;
5842730d 5721 rdev->saved_raid_disk = -1;
2bf071bf
N		 5722 err = bind_rdev_to_array(rdev, mddev);
5723 if (err)
5724 goto abort_export;
1da177e4
LT		 5725
5726 /*
5727 * The rest should better be atomic, we can have disk failures
5728 * noticed in interrupt contexts ...
5729 */
5730
1da177e4
LT		 5731 rdev->raid_disk = -1;
5732
850b2b42 5733 md_update_sb(mddev, 1);
1da177e4
LT		 5734
5735 /*
5736 * Kick recovery, maybe this spare has to be added to the
5737 * array immediately.
5738 */
5739 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5740 md_wakeup_thread(mddev->thread);
d7603b7e 5741 md_new_event(mddev);
1da177e4
LT		 5742 return 0;
5743
1da177e4
LT		 5744abort_export:
5745 export_rdev(rdev);
5746 return err;
5747}
5748
fd01b88c 5749static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5750{
5751 int err;
5752
36fa3063
N		 5753 if (mddev->pers) {
5754 if (!mddev->pers->quiesce)
5755 return -EBUSY;
5756 if (mddev->recovery || mddev->sync_thread)
5757 return -EBUSY;
5758 /* we should be able to change the bitmap.. */
5759 }
32a7627c 5760
32a7627c 5761
36fa3063
N		 5762 if (fd >= 0) {
5763 if (mddev->bitmap)
5764 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5765 mddev->bitmap_info.file = fget(fd);
32a7627c 5766
c3d9714e 5767 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5768 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5769 mdname(mddev));
5770 return -EBADF;
5771 }
5772
c3d9714e 5773 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5774 if (err) {
5775 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5776 mdname(mddev));
c3d9714e
N		 5777 fput(mddev->bitmap_info.file);
5778 mddev->bitmap_info.file = NULL;
36fa3063
N		 5779 return err;
5780 }
c3d9714e 5781 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5782 } else if (mddev->bitmap == NULL)
5783 return -ENOENT; /* cannot remove what isn't there */
5784 err = 0;
5785 if (mddev->pers) {
5786 mddev->pers->quiesce(mddev, 1);
69e51b44 5787 if (fd >= 0) {
36fa3063 5788 err = bitmap_create(mddev);
69e51b44
N		 5789 if (!err)
5790 err = bitmap_load(mddev);
5791 }
d7375ab3 5792 if (fd < 0 || err) {
36fa3063 5793 bitmap_destroy(mddev);
d7375ab3
N		 5794 fd = -1; /* make sure to put the file */
5795 }
36fa3063 5796 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5797 }
5798 if (fd < 0) {
c3d9714e
N		 5799 if (mddev->bitmap_info.file) {
5800 restore_bitmap_write_access(mddev->bitmap_info.file);
5801 fput(mddev->bitmap_info.file);
acc55e22 5802 }
c3d9714e 5803 mddev->bitmap_info.file = NULL;
36fa3063
N		 5804 }
5805
32a7627c
N		 5806 return err;
5807}
5808
1da177e4
LT		 5809/*
5810 * set_array_info is used two different ways
5811 * The original usage is when creating a new array.
5812 * In this usage, raid_disks is > 0 and it together with
5813 * level, size, not_persistent,layout,chunksize determine the
5814 * shape of the array.
5815 * This will always create an array with a type-0.90.0 superblock.
5816 * The newer usage is when assembling an array.
5817 * In this case raid_disks will be 0, and the major_version field is
5818 * use to determine which style super-blocks are to be found on the devices.
5819 * The minor and patch _version numbers are also kept incase the
5820 * super_block handler wishes to interpret them.
5821 */
fd01b88c 5822static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 5823{
5824
5825 if (info->raid_disks == 0) {
5826 /* just setting version number for superblock loading */
5827 if (info->major_version < 0 ||
50511da3 5828 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5829 super_types[info->major_version].name == NULL) {
5830 /* maybe try to auto-load a module? */
5831 printk(KERN_INFO
5832 "md: superblock version %d not known\n",
5833 info->major_version);
5834 return -EINVAL;
5835 }
5836 mddev->major_version = info->major_version;
5837 mddev->minor_version = info->minor_version;
5838 mddev->patch_version = info->patch_version;
3f9d7b0d 5839 mddev->persistent = !info->not_persistent;
cbd19983
N		 5840 /* ensure mddev_put doesn't delete this now that there
5841 * is some minimal configuration.
5842 */
5843 mddev->ctime = get_seconds();
1da177e4
LT		 5844 return 0;
5845 }
5846 mddev->major_version = MD_MAJOR_VERSION;
5847 mddev->minor_version = MD_MINOR_VERSION;
5848 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5849 mddev->ctime = get_seconds();
5850
5851 mddev->level = info->level;
17115e03 5852 mddev->clevel[0] = 0;
58c0fed4 5853 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5854 mddev->raid_disks = info->raid_disks;
5855 /* don't set md_minor, it is determined by which /dev/md* was
5856 * openned
5857 */
5858 if (info->state & (1<<MD_SB_CLEAN))
5859 mddev->recovery_cp = MaxSector;
5860 else
5861 mddev->recovery_cp = 0;
5862 mddev->persistent = ! info->not_persistent;
e691063a 5863 mddev->external = 0;
1da177e4
LT		 5864
5865 mddev->layout = info->layout;
9d8f0363 5866 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5867
5868 mddev->max_disks = MD_SB_DISKS;
5869
e691063a
N		 5870 if (mddev->persistent)
5871 mddev->flags = 0;
850b2b42 5872 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5873
c3d9714e
N		 5874 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5875 mddev->bitmap_info.offset = 0;
b2a2703c 5876
f6705578
N		 5877 mddev->reshape_position = MaxSector;
5878
1da177e4
LT		 5879 /*
5880 * Generate a 128 bit UUID
5881 */
5882 get_random_bytes(mddev->uuid, 16);
5883
f6705578 5884 mddev->new_level = mddev->level;
664e7c41 5885 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5886 mddev->new_layout = mddev->layout;
5887 mddev->delta_disks = 0;
5888
1da177e4
LT		 5889 return 0;
5890}
5891
fd01b88c 5892void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 5893{
b522adcd
DW		 5894 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5895
5896 if (mddev->external_size)
5897 return;
5898
1f403624
DW		 5899 mddev->array_sectors = array_sectors;
5900}
5901EXPORT_SYMBOL(md_set_array_sectors);
5902
fd01b88c 5903static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 5904{
3cb03002 5905 struct md_rdev *rdev;
a35b0d69 5906 int rv;
d71f9f88 5907 int fit = (num_sectors == 0);
a35b0d69
N		 5908
5909 if (mddev->pers->resize == NULL)
5910 return -EINVAL;
d71f9f88
AN		 5911 /* The "num_sectors" is the number of sectors of each device that
5912 * is used. This can only make sense for arrays with redundancy.
5913 * linear and raid0 always use whatever space is available. We can only
5914 * consider changing this number if no resync or reconstruction is
5915 * happening, and if the new size is acceptable. It must fit before the
0f420358 5916 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5917 * of each device. If num_sectors is zero, we find the largest size
5918 * that fits.
a35b0d69
N		 5919 */
5920 if (mddev->sync_thread)
5921 return -EBUSY;
dba034ee
N		 5922 if (mddev->bitmap)
5923 /* Sorry, cannot grow a bitmap yet, just remove it,
5924 * grow, and re-add.
5925 */
5926 return -EBUSY;
dafb20fa 5927 rdev_for_each(rdev, mddev) {
dd8ac336 5928 sector_t avail = rdev->sectors;
01ab5662 5929
d71f9f88
AN		 5930 if (fit && (num_sectors == 0 || num_sectors > avail))
5931 num_sectors = avail;
5932 if (avail < num_sectors)
a35b0d69
N		 5933 return -ENOSPC;
5934 }
d71f9f88 5935 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5936 if (!rv)
5937 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5938 return rv;
5939}
5940
fd01b88c 5941static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 5942{
5943 int rv;
5944 /* change the number of raid disks */
63c70c4f 5945 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5946 return -EINVAL;
5947 if (raid_disks <= 0 ||
233fca36 5948 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5949 return -EINVAL;
63c70c4f 5950 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5951 return -EBUSY;
63c70c4f
N		 5952 mddev->delta_disks = raid_disks - mddev->raid_disks;
5953
5954 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5955 if (rv < 0)
5956 mddev->delta_disks = 0;
da943b99
N		 5957 return rv;
5958}
5959
5960
1da177e4
LT		 5961/*
5962 * update_array_info is used to change the configuration of an
5963 * on-line array.
5964 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5965 * fields in the info are checked against the array.
5966 * Any differences that cannot be handled will cause an error.
5967 * Normally, only one change can be managed at a time.
5968 */
fd01b88c 5969static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 5970{
5971 int rv = 0;
5972 int cnt = 0;
36fa3063
N		 5973 int state = 0;
5974
5975 /* calculate expected state,ignoring low bits */
c3d9714e 5976 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5977 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5978
5979 if (mddev->major_version != info->major_version ||
5980 mddev->minor_version != info->minor_version ||
5981/* mddev->patch_version != info->patch_version || */
5982 mddev->ctime != info->ctime ||
5983 mddev->level != info->level ||
5984/* mddev->layout != info->layout || */
5985 !mddev->persistent != info->not_persistent||
9d8f0363 5986 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5987 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5988 ((state^info->state) & 0xfffffe00)
5989 )
1da177e4
LT		 5990 return -EINVAL;
5991 /* Check there is only one change */
58c0fed4
AN		 5992 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5993 cnt++;
5994 if (mddev->raid_disks != info->raid_disks)
5995 cnt++;
5996 if (mddev->layout != info->layout)
5997 cnt++;
5998 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5999 cnt++;
6000 if (cnt == 0)
6001 return 0;
6002 if (cnt > 1)
6003 return -EINVAL;
1da177e4
LT		 6004
6005 if (mddev->layout != info->layout) {
6006 /* Change layout
6007 * we don't need to do anything at the md level, the
6008 * personality will take care of it all.
6009 */
50ac168a 6010 if (mddev->pers->check_reshape == NULL)
1da177e4 6011 return -EINVAL;
597a711b
N		 6012 else {
6013 mddev->new_layout = info->layout;
50ac168a 6014 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 6015 if (rv)
6016 mddev->new_layout = mddev->layout;
6017 return rv;
6018 }
1da177e4 6019 }
58c0fed4 6020 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 6021 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 6022
da943b99
N		 6023 if (mddev->raid_disks != info->raid_disks)
6024 rv = update_raid_disks(mddev, info->raid_disks);
6025
36fa3063
N		 6026 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6027 if (mddev->pers->quiesce == NULL)
6028 return -EINVAL;
6029 if (mddev->recovery || mddev->sync_thread)
6030 return -EBUSY;
6031 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6032 /* add the bitmap */
6033 if (mddev->bitmap)
6034 return -EEXIST;
c3d9714e 6035 if (mddev->bitmap_info.default_offset == 0)
36fa3063 6036 return -EINVAL;
c3d9714e
N		 6037 mddev->bitmap_info.offset =
6038 mddev->bitmap_info.default_offset;
36fa3063
N		 6039 mddev->pers->quiesce(mddev, 1);
6040 rv = bitmap_create(mddev);
69e51b44
N		 6041 if (!rv)
6042 rv = bitmap_load(mddev);
36fa3063
N		 6043 if (rv)
6044 bitmap_destroy(mddev);
6045 mddev->pers->quiesce(mddev, 0);
6046 } else {
6047 /* remove the bitmap */
6048 if (!mddev->bitmap)
6049 return -ENOENT;
6050 if (mddev->bitmap->file)
6051 return -EINVAL;
6052 mddev->pers->quiesce(mddev, 1);
6053 bitmap_destroy(mddev);
6054 mddev->pers->quiesce(mddev, 0);
c3d9714e 6055 mddev->bitmap_info.offset = 0;
36fa3063
N		 6056 }
6057 }
850b2b42 6058 md_update_sb(mddev, 1);
1da177e4
LT		 6059 return rv;
6060}
6061
fd01b88c 6062static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6063{
3cb03002 6064 struct md_rdev *rdev;
1da177e4
LT		 6065
6066 if (mddev->pers == NULL)
6067 return -ENODEV;
6068
6069 rdev = find_rdev(mddev, dev);
6070 if (!rdev)
6071 return -ENODEV;
6072
6073 md_error(mddev, rdev);
5ef56c8f
N		 6074 if (!test_bit(Faulty, &rdev->flags))
6075 return -EBUSY;
1da177e4
LT		 6076 return 0;
6077}
6078
2f9618ce
AN		 6079/*
6080 * We have a problem here : there is no easy way to give a CHS
6081 * virtual geometry. We currently pretend that we have a 2 heads
6082 * 4 sectors (with a BIG number of cylinders...). This drives
6083 * dosfs just mad... ;-)
6084 */
a885c8c4
CH		 6085static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6086{
fd01b88c 6087 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6088
6089 geo->heads = 2;
6090 geo->sectors = 4;
49ce6cea 6091 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6092 return 0;
6093}
6094
a39907fa 6095static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6096 unsigned int cmd, unsigned long arg)
6097{
6098 int err = 0;
6099 void __user *argp = (void __user *)arg;
fd01b88c 6100 struct mddev *mddev = NULL;
e2218350 6101 int ro;
1da177e4 6102
506c9e44
N		 6103 switch (cmd) {
6104 case RAID_VERSION:
6105 case GET_ARRAY_INFO:
6106 case GET_DISK_INFO:
6107 break;
6108 default:
6109 if (!capable(CAP_SYS_ADMIN))
6110 return -EACCES;
6111 }
1da177e4
LT		 6112
6113 /*
6114 * Commands dealing with the RAID driver but not any
6115 * particular array:
6116 */
6117 switch (cmd)
6118 {
6119 case RAID_VERSION:
6120 err = get_version(argp);
6121 goto done;
6122
6123 case PRINT_RAID_DEBUG:
6124 err = 0;
6125 md_print_devices();
6126 goto done;
6127
6128#ifndef MODULE
6129 case RAID_AUTORUN:
6130 err = 0;
6131 autostart_arrays(arg);
6132 goto done;
6133#endif
6134 default:;
6135 }
6136
6137 /*
6138 * Commands creating/starting a new array:
6139 */
6140
a39907fa 6141 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6142
6143 if (!mddev) {
6144 BUG();
6145 goto abort;
6146 }
6147
1da177e4
LT		 6148 err = mddev_lock(mddev);
6149 if (err) {
6150 printk(KERN_INFO
6151 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6152 err, cmd);
6153 goto abort;
6154 }
6155
6156 switch (cmd)
6157 {
6158 case SET_ARRAY_INFO:
6159 {
6160 mdu_array_info_t info;
6161 if (!arg)
6162 memset(&info, 0, sizeof(info));
6163 else if (copy_from_user(&info, argp, sizeof(info))) {
6164 err = -EFAULT;
6165 goto abort_unlock;
6166 }
6167 if (mddev->pers) {
6168 err = update_array_info(mddev, &info);
6169 if (err) {
6170 printk(KERN_WARNING "md: couldn't update"
6171 " array info. %d\n", err);
6172 goto abort_unlock;
6173 }
6174 goto done_unlock;
6175 }
6176 if (!list_empty(&mddev->disks)) {
6177 printk(KERN_WARNING
6178 "md: array %s already has disks!\n",
6179 mdname(mddev));
6180 err = -EBUSY;
6181 goto abort_unlock;
6182 }
6183 if (mddev->raid_disks) {
6184 printk(KERN_WARNING
6185 "md: array %s already initialised!\n",
6186 mdname(mddev));
6187 err = -EBUSY;
6188 goto abort_unlock;
6189 }
6190 err = set_array_info(mddev, &info);
6191 if (err) {
6192 printk(KERN_WARNING "md: couldn't set"
6193 " array info. %d\n", err);
6194 goto abort_unlock;
6195 }
6196 }
6197 goto done_unlock;
6198
6199 default:;
6200 }
6201
6202 /*
6203 * Commands querying/configuring an existing array:
6204 */
32a7627c 6205 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6206 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6207 if ((!mddev->raid_disks && !mddev->external)
6208 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6209 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6210 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6211 err = -ENODEV;
6212 goto abort_unlock;
6213 }
6214
6215 /*
6216 * Commands even a read-only array can execute:
6217 */
6218 switch (cmd)
6219 {
6220 case GET_ARRAY_INFO:
6221 err = get_array_info(mddev, argp);
6222 goto done_unlock;
6223
32a7627c 6224 case GET_BITMAP_FILE:
87162a28 6225 err = get_bitmap_file(mddev, argp);
32a7627c
N		 6226 goto done_unlock;
6227
1da177e4
LT		 6228 case GET_DISK_INFO:
6229 err = get_disk_info(mddev, argp);
6230 goto done_unlock;
6231
6232 case RESTART_ARRAY_RW:
6233 err = restart_array(mddev);
6234 goto done_unlock;
6235
6236 case STOP_ARRAY:
d710e138 6237 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 6238 goto done_unlock;
6239
6240 case STOP_ARRAY_RO:
a4bd82d0 6241 err = md_set_readonly(mddev, 1);
1da177e4
LT		 6242 goto done_unlock;
6243
e2218350
DW		 6244 case BLKROSET:
6245 if (get_user(ro, (int __user *)(arg))) {
6246 err = -EFAULT;
6247 goto done_unlock;
6248 }
6249 err = -EINVAL;
6250
6251 /* if the bdev is going readonly the value of mddev->ro
6252 * does not matter, no writes are coming
6253 */
6254 if (ro)
6255 goto done_unlock;
6256
6257 /* are we are already prepared for writes? */
6258 if (mddev->ro != 1)
6259 goto done_unlock;
6260
6261 /* transitioning to readauto need only happen for
6262 * arrays that call md_write_start
6263 */
6264 if (mddev->pers) {
6265 err = restart_array(mddev);
6266 if (err == 0) {
6267 mddev->ro = 2;
6268 set_disk_ro(mddev->gendisk, 0);
6269 }
6270 }
6271 goto done_unlock;
1da177e4
LT		 6272 }
6273
6274 /*
6275 * The remaining ioctls are changing the state of the
f91de92e
N		 6276 * superblock, so we do not allow them on read-only arrays.
6277 * However non-MD ioctls (e.g. get-size) will still come through
6278 * here and hit the 'default' below, so only disallow
6279 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6280 */
bb57fc64 6281 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6282 if (mddev->ro == 2) {
6283 mddev->ro = 0;
00bcb4ac 6284 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 6285 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6286 md_wakeup_thread(mddev->thread);
f91de92e
N		 6287 } else {
6288 err = -EROFS;
6289 goto abort_unlock;
6290 }
1da177e4
LT		 6291 }
6292
6293 switch (cmd)
6294 {
6295 case ADD_NEW_DISK:
6296 {
6297 mdu_disk_info_t info;
6298 if (copy_from_user(&info, argp, sizeof(info)))
6299 err = -EFAULT;
6300 else
6301 err = add_new_disk(mddev, &info);
6302 goto done_unlock;
6303 }
6304
6305 case HOT_REMOVE_DISK:
6306 err = hot_remove_disk(mddev, new_decode_dev(arg));
6307 goto done_unlock;
6308
6309 case HOT_ADD_DISK:
6310 err = hot_add_disk(mddev, new_decode_dev(arg));
6311 goto done_unlock;
6312
6313 case SET_DISK_FAULTY:
6314 err = set_disk_faulty(mddev, new_decode_dev(arg));
6315 goto done_unlock;
6316
6317 case RUN_ARRAY:
d710e138 6318 err = do_md_run(mddev);
1da177e4
LT		 6319 goto done_unlock;
6320
32a7627c
N		 6321 case SET_BITMAP_FILE:
6322 err = set_bitmap_file(mddev, (int)arg);
6323 goto done_unlock;
6324
1da177e4 6325 default:
1da177e4
LT		 6326 err = -EINVAL;
6327 goto abort_unlock;
6328 }
6329
6330done_unlock:
6331abort_unlock:
d3374825
N		 6332 if (mddev->hold_active == UNTIL_IOCTL &&
6333 err != -EINVAL)
6334 mddev->hold_active = 0;
1da177e4
LT		 6335 mddev_unlock(mddev);
6336
6337 return err;
6338done:
6339 if (err)
6340 MD_BUG();
6341abort:
6342 return err;
6343}
aa98aa31
AB		 6344#ifdef CONFIG_COMPAT
6345static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6346 unsigned int cmd, unsigned long arg)
6347{
6348 switch (cmd) {
6349 case HOT_REMOVE_DISK:
6350 case HOT_ADD_DISK:
6351 case SET_DISK_FAULTY:
6352 case SET_BITMAP_FILE:
6353 /* These take in integer arg, do not convert */
6354 break;
6355 default:
6356 arg = (unsigned long)compat_ptr(arg);
6357 break;
6358 }
6359
6360 return md_ioctl(bdev, mode, cmd, arg);
6361}
6362#endif /* CONFIG_COMPAT */
1da177e4 6363
a39907fa 6364static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6365{
6366 /*
6367 * Succeed if we can lock the mddev, which confirms that
6368 * it isn't being stopped right now.
6369 */
fd01b88c 6370 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6371 int err;
6372
d3374825
N		 6373 if (mddev->gendisk != bdev->bd_disk) {
6374 /* we are racing with mddev_put which is discarding this
6375 * bd_disk.
6376 */
6377 mddev_put(mddev);
6378 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6379 flush_workqueue(md_misc_wq);
d3374825
N		 6380 /* Then retry the open from the top */
6381 return -ERESTARTSYS;
6382 }
6383 BUG_ON(mddev != bdev->bd_disk->private_data);
6384
c8c00a69 6385 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6386 goto out;
6387
6388 err = 0;
f2ea68cf 6389 atomic_inc(&mddev->openers);
c8c00a69 6390 mutex_unlock(&mddev->open_mutex);
1da177e4 6391
f0b4f7e2 6392 check_disk_change(bdev);
1da177e4
LT		 6393 out:
6394 return err;
6395}
6396
a39907fa 6397static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6398{
fd01b88c 6399 struct mddev *mddev = disk->private_data;
1da177e4 6400
52e5f9d1 6401 BUG_ON(!mddev);
f2ea68cf 6402 atomic_dec(&mddev->openers);
1da177e4
LT		 6403 mddev_put(mddev);
6404
6405 return 0;
6406}
f0b4f7e2
N		 6407
6408static int md_media_changed(struct gendisk *disk)
6409{
fd01b88c 6410 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6411
6412 return mddev->changed;
6413}
6414
6415static int md_revalidate(struct gendisk *disk)
6416{
fd01b88c 6417 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6418
6419 mddev->changed = 0;
6420 return 0;
6421}
83d5cde4 6422static const struct block_device_operations md_fops =
1da177e4
LT		 6423{
6424 .owner = THIS_MODULE,
a39907fa
AV		 6425 .open = md_open,
6426 .release = md_release,
b492b852 6427 .ioctl = md_ioctl,
aa98aa31
AB		 6428#ifdef CONFIG_COMPAT
6429 .compat_ioctl = md_compat_ioctl,
6430#endif
a885c8c4 6431 .getgeo = md_getgeo,
f0b4f7e2
N		 6432 .media_changed = md_media_changed,
6433 .revalidate_disk= md_revalidate,
1da177e4
LT		 6434};
6435
75c96f85 6436static int md_thread(void * arg)
1da177e4 6437{
2b8bf345 6438 struct md_thread *thread = arg;
1da177e4 6439
1da177e4
LT		 6440 /*
6441 * md_thread is a 'system-thread', it's priority should be very
6442 * high. We avoid resource deadlocks individually in each
6443 * raid personality. (RAID5 does preallocation) We also use RR and
6444 * the very same RT priority as kswapd, thus we will never get
6445 * into a priority inversion deadlock.
6446 *
6447 * we definitely have to have equal or higher priority than
6448 * bdflush, otherwise bdflush will deadlock if there are too
6449 * many dirty RAID5 blocks.
6450 */
1da177e4 6451
6985c43f 6452 allow_signal(SIGKILL);
a6fb0934 6453 while (!kthread_should_stop()) {
1da177e4 6454
93588e22
N		 6455 /* We need to wait INTERRUPTIBLE so that
6456 * we don't add to the load-average.
6457 * That means we need to be sure no signals are
6458 * pending
6459 */
6460 if (signal_pending(current))
6461 flush_signals(current);
6462
6463 wait_event_interruptible_timeout
6464 (thread->wqueue,
6465 test_bit(THREAD_WAKEUP, &thread->flags)
6466 || kthread_should_stop(),
6467 thread->timeout);
1da177e4 6468
6c987910
N		 6469 clear_bit(THREAD_WAKEUP, &thread->flags);
6470 if (!kthread_should_stop())
589a594b 6471 thread->run(thread->mddev);
1da177e4 6472 }
a6fb0934 6473
1da177e4
LT		 6474 return 0;
6475}
6476
2b8bf345 6477void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6478{
6479 if (thread) {
36a4e1fe 6480 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6481 set_bit(THREAD_WAKEUP, &thread->flags);
6482 wake_up(&thread->wqueue);
6483 }
6484}
6485
2b8bf345 6486struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT		 6487 const char *name)
6488{
2b8bf345 6489 struct md_thread *thread;
1da177e4 6490
2b8bf345 6491 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6492 if (!thread)
6493 return NULL;
6494
1da177e4
LT		 6495 init_waitqueue_head(&thread->wqueue);
6496
1da177e4
LT		 6497 thread->run = run;
6498 thread->mddev = mddev;
32a7627c 6499 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6500 thread->tsk = kthread_run(md_thread, thread,
6501 "%s_%s",
6502 mdname(thread->mddev),
6503 name ?: mddev->pers->name);
a6fb0934 6504 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6505 kfree(thread);
6506 return NULL;
6507 }
1da177e4
LT		 6508 return thread;
6509}
6510
2b8bf345 6511void md_unregister_thread(struct md_thread **threadp)
1da177e4 6512{
2b8bf345 6513 struct md_thread *thread = *threadp;
e0cf8f04
N		 6514 if (!thread)
6515 return;
36a4e1fe 6516 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6517 /* Locking ensures that mddev_unlock does not wake_up a
6518 * non-existent thread
6519 */
6520 spin_lock(&pers_lock);
6521 *threadp = NULL;
6522 spin_unlock(&pers_lock);
a6fb0934
N		 6523
6524 kthread_stop(thread->tsk);
1da177e4
LT		 6525 kfree(thread);
6526}
6527
fd01b88c 6528void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6529{
6530 if (!mddev) {
6531 MD_BUG();
6532 return;
6533 }
6534
b2d444d7 6535 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6536 return;
6bfe0b49 6537
de393cde 6538 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6539 return;
6540 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6541 if (mddev->degraded)
6542 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6543 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6544 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6545 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6546 md_wakeup_thread(mddev->thread);
768a418d 6547 if (mddev->event_work.func)
e804ac78 6548 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6549 md_new_event_inintr(mddev);
1da177e4
LT		 6550}
6551
6552/* seq_file implementation /proc/mdstat */
6553
6554static void status_unused(struct seq_file *seq)
6555{
6556 int i = 0;
3cb03002 6557 struct md_rdev *rdev;
1da177e4
LT		 6558
6559 seq_printf(seq, "unused devices: ");
6560
159ec1fc 6561 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6562 char b[BDEVNAME_SIZE];
6563 i++;
6564 seq_printf(seq, "%s ",
6565 bdevname(rdev->bdev,b));
6566 }
6567 if (!i)
6568 seq_printf(seq, "<none>");
6569
6570 seq_printf(seq, "\n");
6571}
6572
6573
fd01b88c 6574static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6575{
dd71cf6b
N		 6576 sector_t max_sectors, resync, res;
6577 unsigned long dt, db;
6578 sector_t rt;
4588b42e
N		 6579 int scale;
6580 unsigned int per_milli;
1da177e4 6581
dd71cf6b 6582 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6583
6584 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6585 max_sectors = mddev->resync_max_sectors;
1da177e4 6586 else
dd71cf6b 6587 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6588
6589 /*
6590 * Should not happen.
6591 */
dd71cf6b 6592 if (!max_sectors) {
1da177e4
LT		 6593 MD_BUG();
6594 return;
6595 }
4588b42e 6596 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6597 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6598 * u32, as those are the requirements for sector_div.
6599 * Thus 'scale' must be at least 10
6600 */
6601 scale = 10;
6602 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6603 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6604 scale++;
6605 }
6606 res = (resync>>scale)*1000;
dd71cf6b 6607 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6608
6609 per_milli = res;
1da177e4 6610 {
4588b42e 6611 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6612 seq_printf(seq, "[");
6613 for (i = 0; i < x; i++)
6614 seq_printf(seq, "=");
6615 seq_printf(seq, ">");
6616 for (i = 0; i < y; i++)
6617 seq_printf(seq, ".");
6618 seq_printf(seq, "] ");
6619 }
4588b42e 6620 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6621 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6622 "reshape" :
61df9d91
N		 6623 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6624 "check" :
6625 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6626 "resync" : "recovery"))),
6627 per_milli/10, per_milli % 10,
dd71cf6b
N		 6628 (unsigned long long) resync/2,
6629 (unsigned long long) max_sectors/2);
1da177e4
LT		 6630
6631 /*
1da177e4
LT		 6632 * dt: time from mark until now
6633 * db: blocks written from mark until now
6634 * rt: remaining time
dd71cf6b
N		 6635 *
6636 * rt is a sector_t, so could be 32bit or 64bit.
6637 * So we divide before multiply in case it is 32bit and close
6638 * to the limit.
25985edc 6639 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6640 * near the end of resync when the number of remaining sectors
6641 * is close to 'db'.
6642 * We then divide rt by 32 after multiplying by db to compensate.
6643 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6644 */
6645 dt = ((jiffies - mddev->resync_mark) / HZ);
6646 if (!dt) dt++;
ff4e8d9a
N		 6647 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6648 - mddev->resync_mark_cnt;
1da177e4 6649
dd71cf6b
N		 6650 rt = max_sectors - resync; /* number of remaining sectors */
6651 sector_div(rt, db/32+1);
6652 rt *= dt;
6653 rt >>= 5;
6654
6655 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6656 ((unsigned long)rt % 60)/6);
1da177e4 6657
ff4e8d9a 6658 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6659}
6660
6661static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6662{
6663 struct list_head *tmp;
6664 loff_t l = *pos;
fd01b88c 6665 struct mddev *mddev;
1da177e4
LT		 6666
6667 if (l >= 0x10000)
6668 return NULL;
6669 if (!l--)
6670 /* header */
6671 return (void*)1;
6672
6673 spin_lock(&all_mddevs_lock);
6674 list_for_each(tmp,&all_mddevs)
6675 if (!l--) {
fd01b88c 6676 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6677 mddev_get(mddev);
6678 spin_unlock(&all_mddevs_lock);
6679 return mddev;
6680 }
6681 spin_unlock(&all_mddevs_lock);
6682 if (!l--)
6683 return (void*)2;/* tail */
6684 return NULL;
6685}
6686
6687static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6688{
6689 struct list_head *tmp;
fd01b88c 6690 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6691
6692 ++*pos;
6693 if (v == (void*)2)
6694 return NULL;
6695
6696 spin_lock(&all_mddevs_lock);
6697 if (v == (void*)1)
6698 tmp = all_mddevs.next;
6699 else
6700 tmp = mddev->all_mddevs.next;
6701 if (tmp != &all_mddevs)
fd01b88c 6702 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6703 else {
6704 next_mddev = (void*)2;
6705 *pos = 0x10000;
6706 }
6707 spin_unlock(&all_mddevs_lock);
6708
6709 if (v != (void*)1)
6710 mddev_put(mddev);
6711 return next_mddev;
6712
6713}
6714
6715static void md_seq_stop(struct seq_file *seq, void *v)
6716{
fd01b88c 6717 struct mddev *mddev = v;
1da177e4
LT		 6718
6719 if (mddev && v != (void*)1 && v != (void*)2)
6720 mddev_put(mddev);
6721}
6722
6723static int md_seq_show(struct seq_file *seq, void *v)
6724{
fd01b88c 6725 struct mddev *mddev = v;
dd8ac336 6726 sector_t sectors;
3cb03002 6727 struct md_rdev *rdev;
32a7627c 6728 struct bitmap *bitmap;
1da177e4
LT		 6729
6730 if (v == (void*)1) {
84fc4b56 6731 struct md_personality *pers;
1da177e4
LT		 6732 seq_printf(seq, "Personalities : ");
6733 spin_lock(&pers_lock);
2604b703
N		 6734 list_for_each_entry(pers, &pers_list, list)
6735 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6736
6737 spin_unlock(&pers_lock);
6738 seq_printf(seq, "\n");
f1514638 6739 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6740 return 0;
6741 }
6742 if (v == (void*)2) {
6743 status_unused(seq);
6744 return 0;
6745 }
6746
5dc5cf7d 6747 if (mddev_lock(mddev) < 0)
1da177e4 6748 return -EINTR;
5dc5cf7d 6749
1da177e4
LT		 6750 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6751 seq_printf(seq, "%s : %sactive", mdname(mddev),
6752 mddev->pers ? "" : "in");
6753 if (mddev->pers) {
f91de92e 6754 if (mddev->ro==1)
1da177e4 6755 seq_printf(seq, " (read-only)");
f91de92e 6756 if (mddev->ro==2)
52720ae7 6757 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6758 seq_printf(seq, " %s", mddev->pers->name);
6759 }
6760
dd8ac336 6761 sectors = 0;
dafb20fa 6762 rdev_for_each(rdev, mddev) {
1da177e4
LT		 6763 char b[BDEVNAME_SIZE];
6764 seq_printf(seq, " %s[%d]",
6765 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6766 if (test_bit(WriteMostly, &rdev->flags))
6767 seq_printf(seq, "(W)");
b2d444d7 6768 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6769 seq_printf(seq, "(F)");
6770 continue;
2d78f8c4
N		 6771 }
6772 if (rdev->raid_disk < 0)
b325a32e 6773 seq_printf(seq, "(S)"); /* spare */
2d78f8c4
N		 6774 if (test_bit(Replacement, &rdev->flags))
6775 seq_printf(seq, "(R)");
dd8ac336 6776 sectors += rdev->sectors;
1da177e4
LT		 6777 }
6778
6779 if (!list_empty(&mddev->disks)) {
6780 if (mddev->pers)
6781 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6782 (unsigned long long)
6783 mddev->array_sectors / 2);
1da177e4
LT		 6784 else
6785 seq_printf(seq, "\n %llu blocks",
dd8ac336 6786 (unsigned long long)sectors / 2);
1da177e4 6787 }
1cd6bf19
N		 6788 if (mddev->persistent) {
6789 if (mddev->major_version != 0 ||
6790 mddev->minor_version != 90) {
6791 seq_printf(seq," super %d.%d",
6792 mddev->major_version,
6793 mddev->minor_version);
6794 }
e691063a
N		 6795 } else if (mddev->external)
6796 seq_printf(seq, " super external:%s",
6797 mddev->metadata_type);
6798 else
1cd6bf19 6799 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6800
6801 if (mddev->pers) {
d710e138 6802 mddev->pers->status(seq, mddev);
1da177e4 6803 seq_printf(seq, "\n ");
8e1b39d6
N		 6804 if (mddev->pers->sync_request) {
6805 if (mddev->curr_resync > 2) {
d710e138 6806 status_resync(seq, mddev);
8e1b39d6
N		 6807 seq_printf(seq, "\n ");
6808 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6809 seq_printf(seq, "\tresync=DELAYED\n ");
6810 else if (mddev->recovery_cp < MaxSector)
6811 seq_printf(seq, "\tresync=PENDING\n ");
6812 }
32a7627c
N		 6813 } else
6814 seq_printf(seq, "\n ");
6815
6816 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6817 unsigned long chunk_kb;
6818 unsigned long flags;
32a7627c 6819 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6820 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6821 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6822 "%lu%s chunk",
6823 bitmap->pages - bitmap->missing_pages,
6824 bitmap->pages,
6825 (bitmap->pages - bitmap->missing_pages)
6826 << (PAGE_SHIFT - 10),
42a04b50 6827 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6828 chunk_kb ? "KB" : "B");
78d742d8
N		 6829 if (bitmap->file) {
6830 seq_printf(seq, ", file: ");
c32c2f63 6831 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6832 }
78d742d8 6833
32a7627c
N		 6834 seq_printf(seq, "\n");
6835 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6836 }
6837
6838 seq_printf(seq, "\n");
6839 }
6840 mddev_unlock(mddev);
6841
6842 return 0;
6843}
6844
110518bc 6845static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6846 .start = md_seq_start,
6847 .next = md_seq_next,
6848 .stop = md_seq_stop,
6849 .show = md_seq_show,
6850};
6851
6852static int md_seq_open(struct inode *inode, struct file *file)
6853{
f1514638 6854 struct seq_file *seq;
1da177e4
LT		 6855 int error;
6856
6857 error = seq_open(file, &md_seq_ops);
d7603b7e 6858 if (error)
f1514638
KS		 6859 return error;
6860
6861 seq = file->private_data;
6862 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6863 return error;
6864}
6865
d7603b7e
N		 6866static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6867{
f1514638 6868 struct seq_file *seq = filp->private_data;
d7603b7e
N		 6869 int mask;
6870
6871 poll_wait(filp, &md_event_waiters, wait);
6872
6873 /* always allow read */
6874 mask = POLLIN | POLLRDNORM;
6875
f1514638 6876 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 6877 mask |= POLLERR | POLLPRI;
6878 return mask;
6879}
6880
fa027c2a 6881static const struct file_operations md_seq_fops = {
e24650c2 6882 .owner = THIS_MODULE,
1da177e4
LT		 6883 .open = md_seq_open,
6884 .read = seq_read,
6885 .llseek = seq_lseek,
c3f94b40 6886 .release = seq_release_private,
d7603b7e 6887 .poll = mdstat_poll,
1da177e4
LT		 6888};
6889
84fc4b56 6890int register_md_personality(struct md_personality *p)
1da177e4 6891{
1da177e4 6892 spin_lock(&pers_lock);
2604b703
N		 6893 list_add_tail(&p->list, &pers_list);
6894 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6895 spin_unlock(&pers_lock);
6896 return 0;
6897}
6898
84fc4b56 6899int unregister_md_personality(struct md_personality *p)
1da177e4 6900{
2604b703 6901 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6902 spin_lock(&pers_lock);
2604b703 6903 list_del_init(&p->list);
1da177e4
LT		 6904 spin_unlock(&pers_lock);
6905 return 0;
6906}
6907
fd01b88c 6908static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 6909{
3cb03002 6910 struct md_rdev * rdev;
1da177e4 6911 int idle;
eea1bf38 6912 int curr_events;
1da177e4
LT		 6913
6914 idle = 1;
4b80991c
N		 6915 rcu_read_lock();
6916 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6917 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6918 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6919 (int)part_stat_read(&disk->part0, sectors[1]) -
6920 atomic_read(&disk->sync_io);
713f6ab1
N		 6921 /* sync IO will cause sync_io to increase before the disk_stats
6922 * as sync_io is counted when a request starts, and
6923 * disk_stats is counted when it completes.
6924 * So resync activity will cause curr_events to be smaller than
6925 * when there was no such activity.
6926 * non-sync IO will cause disk_stat to increase without
6927 * increasing sync_io so curr_events will (eventually)
6928 * be larger than it was before. Once it becomes
6929 * substantially larger, the test below will cause
6930 * the array to appear non-idle, and resync will slow
6931 * down.
6932 * If there is a lot of outstanding resync activity when
6933 * we set last_event to curr_events, then all that activity
6934 * completing might cause the array to appear non-idle
6935 * and resync will be slowed down even though there might
6936 * not have been non-resync activity. This will only
6937 * happen once though. 'last_events' will soon reflect
6938 * the state where there is little or no outstanding
6939 * resync requests, and further resync activity will
6940 * always make curr_events less than last_events.
c0e48521 6941 *
1da177e4 6942 */
eea1bf38 6943 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6944 rdev->last_events = curr_events;
6945 idle = 0;
6946 }
6947 }
4b80991c 6948 rcu_read_unlock();
1da177e4
LT		 6949 return idle;
6950}
6951
fd01b88c 6952void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 6953{
6954 /* another "blocks" (512byte) blocks have been synced */
6955 atomic_sub(blocks, &mddev->recovery_active);
6956 wake_up(&mddev->recovery_wait);
6957 if (!ok) {
dfc70645 6958 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6959 md_wakeup_thread(mddev->thread);
6960 // stop recovery, signal do_sync ....
6961 }
6962}
6963
6964
06d91a5f
N		 6965/* md_write_start(mddev, bi)
6966 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6967 * in superblock) before writing, schedule a superblock update
6968 * and wait for it to complete.
06d91a5f 6969 */
fd01b88c 6970void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 6971{
0fd62b86 6972 int did_change = 0;
06d91a5f 6973 if (bio_data_dir(bi) != WRITE)
3d310eb7 6974 return;
06d91a5f 6975
f91de92e
N		 6976 BUG_ON(mddev->ro == 1);
6977 if (mddev->ro == 2) {
6978 /* need to switch to read/write */
6979 mddev->ro = 0;
6980 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6981 md_wakeup_thread(mddev->thread);
25156198 6982 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6983 did_change = 1;
f91de92e 6984 }
06d91a5f 6985 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6986 if (mddev->safemode == 1)
6987 mddev->safemode = 0;
06d91a5f 6988 if (mddev->in_sync) {
a9701a30 6989 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6990 if (mddev->in_sync) {
6991 mddev->in_sync = 0;
850b2b42 6992 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6993 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6994 md_wakeup_thread(mddev->thread);
0fd62b86 6995 did_change = 1;
3d310eb7 6996 }
a9701a30 6997 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6998 }
0fd62b86 6999 if (did_change)
00bcb4ac 7000 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 7001 wait_event(mddev->sb_wait,
09a44cc1 7002 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 7003}
7004
fd01b88c 7005void md_write_end(struct mddev *mddev)
1da177e4
LT		 7006{
7007 if (atomic_dec_and_test(&mddev->writes_pending)) {
7008 if (mddev->safemode == 2)
7009 md_wakeup_thread(mddev->thread);
16f17b39 7010 else if (mddev->safemode_delay)
1da177e4
LT		 7011 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7012 }
7013}
7014
2a2275d6
N		 7015/* md_allow_write(mddev)
7016 * Calling this ensures that the array is marked 'active' so that writes
7017 * may proceed without blocking. It is important to call this before
7018 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7019 * Must be called with mddev_lock held.
b5470dc5
DW		 7020 *
7021 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7022 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 7023 */
fd01b88c 7024int md_allow_write(struct mddev *mddev)
2a2275d6
N		 7025{
7026 if (!mddev->pers)
b5470dc5 7027 return 0;
2a2275d6 7028 if (mddev->ro)
b5470dc5 7029 return 0;
1a0fd497 7030 if (!mddev->pers->sync_request)
b5470dc5 7031 return 0;
2a2275d6
N		 7032
7033 spin_lock_irq(&mddev->write_lock);
7034 if (mddev->in_sync) {
7035 mddev->in_sync = 0;
7036 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 7037 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 7038 if (mddev->safemode_delay &&
7039 mddev->safemode == 0)
7040 mddev->safemode = 1;
7041 spin_unlock_irq(&mddev->write_lock);
7042 md_update_sb(mddev, 0);
00bcb4ac 7043 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 7044 } else
7045 spin_unlock_irq(&mddev->write_lock);
b5470dc5 7046
070dc6dd 7047 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 7048 return -EAGAIN;
7049 else
7050 return 0;
2a2275d6
N		 7051}
7052EXPORT_SYMBOL_GPL(md_allow_write);
7053
1da177e4
LT		 7054#define SYNC_MARKS 10
7055#define SYNC_MARK_STEP (3*HZ)
fd01b88c 7056void md_do_sync(struct mddev *mddev)
1da177e4 7057{
fd01b88c 7058 struct mddev *mddev2;
1da177e4
LT		 7059 unsigned int currspeed = 0,
7060 window;
57afd89f 7061 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 7062 unsigned long mark[SYNC_MARKS];
7063 sector_t mark_cnt[SYNC_MARKS];
7064 int last_mark,m;
7065 struct list_head *tmp;
7066 sector_t last_check;
57afd89f 7067 int skipped = 0;
3cb03002 7068 struct md_rdev *rdev;
61df9d91 7069 char *desc;
1da177e4
LT		 7070
7071 /* just incase thread restarts... */
7072 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7073 return;
5fd6c1dc
N		 7074 if (mddev->ro) /* never try to sync a read-only array */
7075 return;
1da177e4 7076
61df9d91
N		 7077 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7078 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7079 desc = "data-check";
7080 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7081 desc = "requested-resync";
7082 else
7083 desc = "resync";
7084 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7085 desc = "reshape";
7086 else
7087 desc = "recovery";
7088
1da177e4
LT		 7089 /* we overload curr_resync somewhat here.
7090 * 0 == not engaged in resync at all
7091 * 2 == checking that there is no conflict with another sync
7092 * 1 == like 2, but have yielded to allow conflicting resync to
7093 * commense
7094 * other == active in resync - this many blocks
7095 *
7096 * Before starting a resync we must have set curr_resync to
7097 * 2, and then checked that every "conflicting" array has curr_resync
7098 * less than ours. When we find one that is the same or higher
7099 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7100 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7101 * This will mean we have to start checking from the beginning again.
7102 *
7103 */
7104
7105 do {
7106 mddev->curr_resync = 2;
7107
7108 try_again:
404e4b43 7109 if (kthread_should_stop())
6985c43f 7110 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7111
7112 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7113 goto skip;
29ac4aa3 7114 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7115 if (mddev2 == mddev)
7116 continue;
90b08710
BS		 7117 if (!mddev->parallel_resync
7118 && mddev2->curr_resync
7119 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7120 DEFINE_WAIT(wq);
7121 if (mddev < mddev2 && mddev->curr_resync == 2) {
7122 /* arbitrarily yield */
7123 mddev->curr_resync = 1;
7124 wake_up(&resync_wait);
7125 }
7126 if (mddev > mddev2 && mddev->curr_resync == 1)
7127 /* no need to wait here, we can wait the next
7128 * time 'round when curr_resync == 2
7129 */
7130 continue;
9744197c
N		 7131 /* We need to wait 'interruptible' so as not to
7132 * contribute to the load average, and not to
7133 * be caught by 'softlockup'
7134 */
7135 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7136 if (!kthread_should_stop() &&
8712e553 7137 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7138 printk(KERN_INFO "md: delaying %s of %s"
7139 " until %s has finished (they"
1da177e4 7140 " share one or more physical units)\n",
61df9d91 7141 desc, mdname(mddev), mdname(mddev2));
1da177e4 7142 mddev_put(mddev2);
9744197c
N		 7143 if (signal_pending(current))
7144 flush_signals(current);
1da177e4
LT		 7145 schedule();
7146 finish_wait(&resync_wait, &wq);
7147 goto try_again;
7148 }
7149 finish_wait(&resync_wait, &wq);
7150 }
7151 }
7152 } while (mddev->curr_resync < 2);
7153
5fd6c1dc 7154 j = 0;
9d88883e 7155 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7156 /* resync follows the size requested by the personality,
57afd89f 7157 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7158 */
7159 max_sectors = mddev->resync_max_sectors;
9d88883e 7160 mddev->resync_mismatches = 0;
5fd6c1dc 7161 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7162 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7163 j = mddev->resync_min;
7164 else if (!mddev->bitmap)
5fd6c1dc 7165 j = mddev->recovery_cp;
5e96ee65 7166
ccfcc3c1 7167 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 7168 max_sectors = mddev->dev_sectors;
5fd6c1dc 7169 else {
1da177e4 7170 /* recovery follows the physical size of devices */
58c0fed4 7171 max_sectors = mddev->dev_sectors;
5fd6c1dc 7172 j = MaxSector;
4e59ca7d 7173 rcu_read_lock();
dafb20fa 7174 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc
N		 7175 if (rdev->raid_disk >= 0 &&
7176 !test_bit(Faulty, &rdev->flags) &&
7177 !test_bit(In_sync, &rdev->flags) &&
7178 rdev->recovery_offset < j)
7179 j = rdev->recovery_offset;
4e59ca7d 7180 rcu_read_unlock();
5fd6c1dc 7181 }
1da177e4 7182
61df9d91
N		 7183 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7184 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7185 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7186 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7187 "(but not more than %d KB/sec) for %s.\n",
7188 speed_max(mddev), desc);
1da177e4 7189
eea1bf38 7190 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7191
57afd89f 7192 io_sectors = 0;
1da177e4
LT		 7193 for (m = 0; m < SYNC_MARKS; m++) {
7194 mark[m] = jiffies;
57afd89f 7195 mark_cnt[m] = io_sectors;
1da177e4
LT		 7196 }
7197 last_mark = 0;
7198 mddev->resync_mark = mark[last_mark];
7199 mddev->resync_mark_cnt = mark_cnt[last_mark];
7200
7201 /*
7202 * Tune reconstruction:
7203 */
7204 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7205 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7206 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7207
7208 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7209 last_check = 0;
7210
7211 if (j>2) {
7212 printk(KERN_INFO
61df9d91
N		 7213 "md: resuming %s of %s from checkpoint.\n",
7214 desc, mdname(mddev));
1da177e4
LT		 7215 mddev->curr_resync = j;
7216 }
75d3da43 7217 mddev->curr_resync_completed = j;
1da177e4
LT		 7218
7219 while (j < max_sectors) {
57afd89f 7220 sector_t sectors;
1da177e4 7221
57afd89f 7222 skipped = 0;
97e4f42d 7223
7a91ee1f
N		 7224 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7225 ((mddev->curr_resync > mddev->curr_resync_completed &&
7226 (mddev->curr_resync - mddev->curr_resync_completed)
7227 > (max_sectors >> 4)) ||
7228 (j - mddev->curr_resync_completed)*2
7229 >= mddev->resync_max - mddev->curr_resync_completed
7230 )) {
97e4f42d 7231 /* time to update curr_resync_completed */
97e4f42d
N		 7232 wait_event(mddev->recovery_wait,
7233 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7234 mddev->curr_resync_completed = j;
070dc6dd 7235 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7236 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7237 }
acb180b0 7238
e62e58a5
N		 7239 while (j >= mddev->resync_max && !kthread_should_stop()) {
7240 /* As this condition is controlled by user-space,
7241 * we can block indefinitely, so use '_interruptible'
7242 * to avoid triggering warnings.
7243 */
7244 flush_signals(current); /* just in case */
7245 wait_event_interruptible(mddev->recovery_wait,
7246 mddev->resync_max > j
7247 || kthread_should_stop());
7248 }
acb180b0
N		 7249
7250 if (kthread_should_stop())
7251 goto interrupted;
7252
57afd89f 7253 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7254 currspeed < speed_min(mddev));
57afd89f 7255 if (sectors == 0) {
dfc70645 7256 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7257 goto out;
7258 }
57afd89f
N		 7259
7260 if (!skipped) { /* actual IO requested */
7261 io_sectors += sectors;
7262 atomic_add(sectors, &mddev->recovery_active);
7263 }
7264
e875ecea
N		 7265 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7266 break;
7267
1da177e4
LT		 7268 j += sectors;
7269 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7270 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7271 if (last_check == 0)
e875ecea 7272 /* this is the earliest that rebuild will be
d7603b7e
N		 7273 * visible in /proc/mdstat
7274 */
7275 md_new_event(mddev);
57afd89f
N		 7276
7277 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7278 continue;
7279
57afd89f 7280 last_check = io_sectors;
1da177e4
LT		 7281 repeat:
7282 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7283 /* step marks */
7284 int next = (last_mark+1) % SYNC_MARKS;
7285
7286 mddev->resync_mark = mark[next];
7287 mddev->resync_mark_cnt = mark_cnt[next];
7288 mark[next] = jiffies;
57afd89f 7289 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7290 last_mark = next;
7291 }
7292
7293
c6207277
N		 7294 if (kthread_should_stop())
7295 goto interrupted;
7296
1da177e4
LT		 7297
7298 /*
7299 * this loop exits only if either when we are slower than
7300 * the 'hard' speed limit, or the system was IO-idle for
7301 * a jiffy.
7302 * the system might be non-idle CPU-wise, but we only care
7303 * about not overloading the IO subsystem. (things like an
7304 * e2fsck being done on the RAID array should execute fast)
7305 */
1da177e4
LT		 7306 cond_resched();
7307
57afd89f
N		 7308 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7309 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7310
88202a0c
N		 7311 if (currspeed > speed_min(mddev)) {
7312 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7313 !is_mddev_idle(mddev, 0)) {
c0e48521 7314 msleep(500);
1da177e4
LT		 7315 goto repeat;
7316 }
7317 }
7318 }
61df9d91 7319 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7320 /*
7321 * this also signals 'finished resyncing' to md_stop
7322 */
7323 out:
1da177e4
LT		 7324 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7325
7326 /* tell personality that we are finished */
57afd89f 7327 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7328
dfc70645 7329 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7330 mddev->curr_resync > 2) {
7331 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7332 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7333 if (mddev->curr_resync >= mddev->recovery_cp) {
7334 printk(KERN_INFO
61df9d91
N		 7335 "md: checkpointing %s of %s.\n",
7336 desc, mdname(mddev));
db91ff55
N		 7337 mddev->recovery_cp =
7338 mddev->curr_resync_completed;
5fd6c1dc
N		 7339 }
7340 } else
7341 mddev->recovery_cp = MaxSector;
7342 } else {
7343 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7344 mddev->curr_resync = MaxSector;
4e59ca7d 7345 rcu_read_lock();
dafb20fa 7346 rdev_for_each_rcu(rdev, mddev)
5fd6c1dc 7347 if (rdev->raid_disk >= 0 &&
70fffd0b 7348 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7349 !test_bit(Faulty, &rdev->flags) &&
7350 !test_bit(In_sync, &rdev->flags) &&
7351 rdev->recovery_offset < mddev->curr_resync)
7352 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7353 rcu_read_unlock();
5fd6c1dc 7354 }
1da177e4 7355 }
db91ff55 7356 skip:
17571284 7357 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7358
c07b70ad
N		 7359 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7360 /* We completed so min/max setting can be forgotten if used. */
7361 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7362 mddev->resync_min = 0;
7363 mddev->resync_max = MaxSector;
7364 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7365 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7366 mddev->curr_resync = 0;
7367 wake_up(&resync_wait);
7368 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7369 md_wakeup_thread(mddev->thread);
c6207277
N		 7370 return;
7371
7372 interrupted:
7373 /*
7374 * got a signal, exit.
7375 */
7376 printk(KERN_INFO
7377 "md: md_do_sync() got signal ... exiting\n");
7378 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7379 goto out;
7380
1da177e4 7381}
29269553 7382EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7383
fd01b88c 7384static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7385{
3cb03002 7386 struct md_rdev *rdev;
b4c4c7b8 7387 int spares = 0;
f2a371c5 7388 int removed = 0;
b4c4c7b8 7389
97e4f42d
N		 7390 mddev->curr_resync_completed = 0;
7391
dafb20fa 7392 rdev_for_each(rdev, mddev)
b4c4c7b8 7393 if (rdev->raid_disk >= 0 &&
6bfe0b49 7394 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7395 (test_bit(Faulty, &rdev->flags) ||
7396 ! test_bit(In_sync, &rdev->flags)) &&
7397 atomic_read(&rdev->nr_pending)==0) {
7398 if (mddev->pers->hot_remove_disk(
b8321b68 7399 mddev, rdev) == 0) {
36fad858 7400 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8 7401 rdev->raid_disk = -1;
f2a371c5 7402 removed++;
b4c4c7b8
N		 7403 }
7404 }
f2a371c5
N		 7405 if (removed)
7406 sysfs_notify(&mddev->kobj, NULL,
7407 "degraded");
7408
b4c4c7b8 7409
dafb20fa 7410 rdev_for_each(rdev, mddev) {
7bfec5f3
N		 7411 if (rdev->raid_disk >= 0 &&
7412 !test_bit(In_sync, &rdev->flags) &&
7413 !test_bit(Faulty, &rdev->flags))
7414 spares++;
7415 if (rdev->raid_disk < 0
7416 && !test_bit(Faulty, &rdev->flags)) {
7417 rdev->recovery_offset = 0;
7418 if (mddev->pers->
7419 hot_add_disk(mddev, rdev) == 0) {
7420 if (sysfs_link_rdev(mddev, rdev))
7421 /* failure here is OK */;
dfc70645 7422 spares++;
7bfec5f3
N		 7423 md_new_event(mddev);
7424 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8 7425 }
dfc70645 7426 }
b4c4c7b8
N		 7427 }
7428 return spares;
7429}
7ebc0be7 7430
fd01b88c 7431static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7432{
3cb03002 7433 struct md_rdev *rdev;
7ebc0be7
N		 7434
7435 /* resync has finished, collect result */
01f96c0a 7436 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N		 7437 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7438 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7439 /* success...*/
7440 /* activate any spares */
7441 if (mddev->pers->spare_active(mddev))
7442 sysfs_notify(&mddev->kobj, NULL,
7443 "degraded");
7444 }
7445 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7446 mddev->pers->finish_reshape)
7447 mddev->pers->finish_reshape(mddev);
7ebc0be7 7448
d70ed2e4
AW		 7449 /* If array is no-longer degraded, then any saved_raid_disk
7450 * information must be scrapped. Also if any device is now
7451 * In_sync we must scrape the saved_raid_disk for that device
7452 * do the superblock for an incrementally recovered device
7453 * written out.
7ebc0be7 7454 */
dafb20fa 7455 rdev_for_each(rdev, mddev)
d70ed2e4
AW		 7456 if (!mddev->degraded ||
7457 test_bit(In_sync, &rdev->flags))
7ebc0be7
N		 7458 rdev->saved_raid_disk = -1;
7459
d70ed2e4 7460 md_update_sb(mddev, 1);
7ebc0be7
N		 7461 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7462 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7463 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7464 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7465 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7466 /* flag recovery needed just to double check */
7467 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7468 sysfs_notify_dirent_safe(mddev->sysfs_action);
7469 md_new_event(mddev);
768e587e
JB		 7470 if (mddev->event_work.func)
7471 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N		 7472}
7473
1da177e4
LT		 7474/*
7475 * This routine is regularly called by all per-raid-array threads to
7476 * deal with generic issues like resync and super-block update.
7477 * Raid personalities that don't have a thread (linear/raid0) do not
7478 * need this as they never do any recovery or update the superblock.
7479 *
7480 * It does not do any resync itself, but rather "forks" off other threads
7481 * to do that as needed.
7482 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7483 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7484 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7485 * and wakeups up this thread which will reap the thread and finish up.
7486 * This thread also removes any faulty devices (with nr_pending == 0).
7487 *
7488 * The overall approach is:
7489 * 1/ if the superblock needs updating, update it.
7490 * 2/ If a recovery thread is running, don't do anything else.
7491 * 3/ If recovery has finished, clean up, possibly marking spares active.
7492 * 4/ If there are any faulty devices, remove them.
7493 * 5/ If array is degraded, try to add spares devices
7494 * 6/ If array has spares or is not in-sync, start a resync thread.
7495 */
fd01b88c 7496void md_check_recovery(struct mddev *mddev)
1da177e4 7497{
68866e42
JB		 7498 if (mddev->suspended)
7499 return;
7500
5f40402d 7501 if (mddev->bitmap)
aa5cbd10 7502 bitmap_daemon_work(mddev);
1da177e4 7503
fca4d848 7504 if (signal_pending(current)) {
31a59e34 7505 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7506 printk(KERN_INFO "md: %s in immediate safe mode\n",
7507 mdname(mddev));
7508 mddev->safemode = 2;
7509 }
7510 flush_signals(current);
7511 }
7512
c89a8eee
N		 7513 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7514 return;
1da177e4 7515 if ( ! (
126925c0 7516 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7517 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7518 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7519 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7520 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7521 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7522 ))
7523 return;
fca4d848 7524
df5b89b3 7525 if (mddev_trylock(mddev)) {
b4c4c7b8 7526 int spares = 0;
fca4d848 7527
c89a8eee
N		 7528 if (mddev->ro) {
7529 /* Only thing we do on a ro array is remove
7530 * failed devices.
7531 */
3cb03002 7532 struct md_rdev *rdev;
dafb20fa 7533 rdev_for_each(rdev, mddev)
a8c42c7f
N		 7534 if (rdev->raid_disk >= 0 &&
7535 !test_bit(Blocked, &rdev->flags) &&
7536 test_bit(Faulty, &rdev->flags) &&
7537 atomic_read(&rdev->nr_pending)==0) {
7538 if (mddev->pers->hot_remove_disk(
b8321b68 7539 mddev, rdev) == 0) {
36fad858 7540 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N		 7541 rdev->raid_disk = -1;
7542 }
7543 }
c89a8eee
N		 7544 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7545 goto unlock;
7546 }
7547
31a59e34 7548 if (!mddev->external) {
0fd62b86 7549 int did_change = 0;
31a59e34
N		 7550 spin_lock_irq(&mddev->write_lock);
7551 if (mddev->safemode &&
7552 !atomic_read(&mddev->writes_pending) &&
7553 !mddev->in_sync &&
7554 mddev->recovery_cp == MaxSector) {
7555 mddev->in_sync = 1;
0fd62b86 7556 did_change = 1;
070dc6dd 7557 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7558 }
7559 if (mddev->safemode == 1)
7560 mddev->safemode = 0;
7561 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7562 if (did_change)
00bcb4ac 7563 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7564 }
fca4d848 7565
850b2b42
N		 7566 if (mddev->flags)
7567 md_update_sb(mddev, 0);
06d91a5f 7568
1da177e4
LT		 7569 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7570 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7571 /* resync/recovery still happening */
7572 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7573 goto unlock;
7574 }
7575 if (mddev->sync_thread) {
7ebc0be7 7576 reap_sync_thread(mddev);
1da177e4
LT		 7577 goto unlock;
7578 }
72a23c21
NB		 7579 /* Set RUNNING before clearing NEEDED to avoid
7580 * any transients in the value of "sync_action".
7581 */
7582 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7583 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7584 /* Clear some bits that don't mean anything, but
7585 * might be left set
7586 */
24dd469d
N		 7587 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7588 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7589
5fd6c1dc
N		 7590 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7591 goto unlock;
1da177e4
LT		 7592 /* no recovery is running.
7593 * remove any failed drives, then
7594 * add spares if possible.
7595 * Spare are also removed and re-added, to allow
7596 * the personality to fail the re-add.
7597 */
1da177e4 7598
b4c4c7b8 7599 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7600 if (mddev->pers->check_reshape == NULL ||
7601 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7602 /* Cannot proceed */
7603 goto unlock;
7604 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7605 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7606 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7607 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7608 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7609 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7610 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7611 } else if (mddev->recovery_cp < MaxSector) {
7612 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7613 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7614 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7615 /* nothing to be done ... */
1da177e4 7616 goto unlock;
24dd469d 7617
1da177e4 7618 if (mddev->pers->sync_request) {
a654b9d8
N		 7619 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7620 /* We are adding a device or devices to an array
7621 * which has the bitmap stored on all devices.
7622 * So make sure all bitmap pages get written
7623 */
7624 bitmap_write_all(mddev->bitmap);
7625 }
1da177e4
LT		 7626 mddev->sync_thread = md_register_thread(md_do_sync,
7627 mddev,
0da3c619 7628 "resync");
1da177e4
LT		 7629 if (!mddev->sync_thread) {
7630 printk(KERN_ERR "%s: could not start resync"
7631 " thread...\n",
7632 mdname(mddev));
7633 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7634 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7635 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7636 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7637 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7638 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7639 } else
1da177e4 7640 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7641 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7642 md_new_event(mddev);
1da177e4
LT		 7643 }
7644 unlock:
72a23c21
NB		 7645 if (!mddev->sync_thread) {
7646 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7647 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7648 &mddev->recovery))
0c3573f1 7649 if (mddev->sysfs_action)
00bcb4ac 7650 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7651 }
1da177e4
LT		 7652 mddev_unlock(mddev);
7653 }
7654}
7655
fd01b88c 7656void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7657{
00bcb4ac 7658 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7659 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7660 !test_bit(Blocked, &rdev->flags) &&
7661 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7662 msecs_to_jiffies(5000));
7663 rdev_dec_pending(rdev, mddev);
7664}
7665EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7666
2230dfe4
N		 7667
7668/* Bad block management.
7669 * We can record which blocks on each device are 'bad' and so just
7670 * fail those blocks, or that stripe, rather than the whole device.
7671 * Entries in the bad-block table are 64bits wide. This comprises:
7672 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7673 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7674 * A 'shift' can be set so that larger blocks are tracked and
7675 * consequently larger devices can be covered.
7676 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7677 *
7678 * Locking of the bad-block table uses a seqlock so md_is_badblock
7679 * might need to retry if it is very unlucky.
7680 * We will sometimes want to check for bad blocks in a bi_end_io function,
7681 * so we use the write_seqlock_irq variant.
7682 *
7683 * When looking for a bad block we specify a range and want to
7684 * know if any block in the range is bad. So we binary-search
7685 * to the last range that starts at-or-before the given endpoint,
7686 * (or "before the sector after the target range")
7687 * then see if it ends after the given start.
7688 * We return
7689 * 0 if there are no known bad blocks in the range
7690 * 1 if there are known bad block which are all acknowledged
7691 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7692 * plus the start/length of the first bad section we overlap.
7693 */
7694int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7695 sector_t *first_bad, int *bad_sectors)
7696{
7697 int hi;
7698 int lo = 0;
7699 u64 *p = bb->page;
7700 int rv = 0;
7701 sector_t target = s + sectors;
7702 unsigned seq;
7703
7704 if (bb->shift > 0) {
7705 /* round the start down, and the end up */
7706 s >>= bb->shift;
7707 target += (1<<bb->shift) - 1;
7708 target >>= bb->shift;
7709 sectors = target - s;
7710 }
7711 /* 'target' is now the first block after the bad range */
7712
7713retry:
7714 seq = read_seqbegin(&bb->lock);
7715
7716 hi = bb->count;
7717
7718 /* Binary search between lo and hi for 'target'
7719 * i.e. for the last range that starts before 'target'
7720 */
7721 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7722 * are known not to be the last range before target.
7723 * VARIANT: hi-lo is the number of possible
7724 * ranges, and decreases until it reaches 1
7725 */
7726 while (hi - lo > 1) {
7727 int mid = (lo + hi) / 2;
7728 sector_t a = BB_OFFSET(p[mid]);
7729 if (a < target)
7730 /* This could still be the one, earlier ranges
7731 * could not. */
7732 lo = mid;
7733 else
7734 /* This and later ranges are definitely out. */
7735 hi = mid;
7736 }
7737 /* 'lo' might be the last that started before target, but 'hi' isn't */
7738 if (hi > lo) {
7739 /* need to check all range that end after 's' to see if
7740 * any are unacknowledged.
7741 */
7742 while (lo >= 0 &&
7743 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7744 if (BB_OFFSET(p[lo]) < target) {
7745 /* starts before the end, and finishes after
7746 * the start, so they must overlap
7747 */
7748 if (rv != -1 && BB_ACK(p[lo]))
7749 rv = 1;
7750 else
7751 rv = -1;
7752 *first_bad = BB_OFFSET(p[lo]);
7753 *bad_sectors = BB_LEN(p[lo]);
7754 }
7755 lo--;
7756 }
7757 }
7758
7759 if (read_seqretry(&bb->lock, seq))
7760 goto retry;
7761
7762 return rv;
7763}
7764EXPORT_SYMBOL_GPL(md_is_badblock);
7765
7766/*
7767 * Add a range of bad blocks to the table.
7768 * This might extend the table, or might contract it
7769 * if two adjacent ranges can be merged.
7770 * We binary-search to find the 'insertion' point, then
7771 * decide how best to handle it.
7772 */
7773static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7774 int acknowledged)
7775{
7776 u64 *p;
7777 int lo, hi;
7778 int rv = 1;
7779
7780 if (bb->shift < 0)
7781 /* badblocks are disabled */
7782 return 0;
7783
7784 if (bb->shift) {
7785 /* round the start down, and the end up */
7786 sector_t next = s + sectors;
7787 s >>= bb->shift;
7788 next += (1<<bb->shift) - 1;
7789 next >>= bb->shift;
7790 sectors = next - s;
7791 }
7792
7793 write_seqlock_irq(&bb->lock);
7794
7795 p = bb->page;
7796 lo = 0;
7797 hi = bb->count;
7798 /* Find the last range that starts at-or-before 's' */
7799 while (hi - lo > 1) {
7800 int mid = (lo + hi) / 2;
7801 sector_t a = BB_OFFSET(p[mid]);
7802 if (a <= s)
7803 lo = mid;
7804 else
7805 hi = mid;
7806 }
7807 if (hi > lo && BB_OFFSET(p[lo]) > s)
7808 hi = lo;
7809
7810 if (hi > lo) {
7811 /* we found a range that might merge with the start
7812 * of our new range
7813 */
7814 sector_t a = BB_OFFSET(p[lo]);
7815 sector_t e = a + BB_LEN(p[lo]);
7816 int ack = BB_ACK(p[lo]);
7817 if (e >= s) {
7818 /* Yes, we can merge with a previous range */
7819 if (s == a && s + sectors >= e)
7820 /* new range covers old */
7821 ack = acknowledged;
7822 else
7823 ack = ack && acknowledged;
7824
7825 if (e < s + sectors)
7826 e = s + sectors;
7827 if (e - a <= BB_MAX_LEN) {
7828 p[lo] = BB_MAKE(a, e-a, ack);
7829 s = e;
7830 } else {
7831 /* does not all fit in one range,
7832 * make p[lo] maximal
7833 */
7834 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7835 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7836 s = a + BB_MAX_LEN;
7837 }
7838 sectors = e - s;
7839 }
7840 }
7841 if (sectors && hi < bb->count) {
7842 /* 'hi' points to the first range that starts after 's'.
7843 * Maybe we can merge with the start of that range */
7844 sector_t a = BB_OFFSET(p[hi]);
7845 sector_t e = a + BB_LEN(p[hi]);
7846 int ack = BB_ACK(p[hi]);
7847 if (a <= s + sectors) {
7848 /* merging is possible */
7849 if (e <= s + sectors) {
7850 /* full overlap */
7851 e = s + sectors;
7852 ack = acknowledged;
7853 } else
7854 ack = ack && acknowledged;
7855
7856 a = s;
7857 if (e - a <= BB_MAX_LEN) {
7858 p[hi] = BB_MAKE(a, e-a, ack);
7859 s = e;
7860 } else {
7861 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7862 s = a + BB_MAX_LEN;
7863 }
7864 sectors = e - s;
7865 lo = hi;
7866 hi++;
7867 }
7868 }
7869 if (sectors == 0 && hi < bb->count) {
7870 /* we might be able to combine lo and hi */
7871 /* Note: 's' is at the end of 'lo' */
7872 sector_t a = BB_OFFSET(p[hi]);
7873 int lolen = BB_LEN(p[lo]);
7874 int hilen = BB_LEN(p[hi]);
7875 int newlen = lolen + hilen - (s - a);
7876 if (s >= a && newlen < BB_MAX_LEN) {
7877 /* yes, we can combine them */
7878 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7879 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7880 memmove(p + hi, p + hi + 1,
7881 (bb->count - hi - 1) * 8);
7882 bb->count--;
7883 }
7884 }
7885 while (sectors) {
7886 /* didn't merge (it all).
7887 * Need to add a range just before 'hi' */
7888 if (bb->count >= MD_MAX_BADBLOCKS) {
7889 /* No room for more */
7890 rv = 0;
7891 break;
7892 } else {
7893 int this_sectors = sectors;
7894 memmove(p + hi + 1, p + hi,
7895 (bb->count - hi) * 8);
7896 bb->count++;
7897
7898 if (this_sectors > BB_MAX_LEN)
7899 this_sectors = BB_MAX_LEN;
7900 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7901 sectors -= this_sectors;
7902 s += this_sectors;
7903 }
7904 }
7905
7906 bb->changed = 1;
de393cde
N		 7907 if (!acknowledged)
7908 bb->unacked_exist = 1;
2230dfe4
N		 7909 write_sequnlock_irq(&bb->lock);
7910
7911 return rv;
7912}
7913
3cb03002 7914int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
2230dfe4
N		 7915 int acknowledged)
7916{
7917 int rv = md_set_badblocks(&rdev->badblocks,
7918 s + rdev->data_offset, sectors, acknowledged);
7919 if (rv) {
7920 /* Make sure they get written out promptly */
8bd2f0a0 7921 sysfs_notify_dirent_safe(rdev->sysfs_state);
2230dfe4
N		 7922 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7923 md_wakeup_thread(rdev->mddev->thread);
7924 }
7925 return rv;
7926}
7927EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7928
7929/*
7930 * Remove a range of bad blocks from the table.
7931 * This may involve extending the table if we spilt a region,
7932 * but it must not fail. So if the table becomes full, we just
7933 * drop the remove request.
7934 */
7935static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7936{
7937 u64 *p;
7938 int lo, hi;
7939 sector_t target = s + sectors;
7940 int rv = 0;
7941
7942 if (bb->shift > 0) {
7943 /* When clearing we round the start up and the end down.
7944 * This should not matter as the shift should align with
7945 * the block size and no rounding should ever be needed.
7946 * However it is better the think a block is bad when it
7947 * isn't than to think a block is not bad when it is.
7948 */
7949 s += (1<<bb->shift) - 1;
7950 s >>= bb->shift;
7951 target >>= bb->shift;
7952 sectors = target - s;
7953 }
7954
7955 write_seqlock_irq(&bb->lock);
7956
7957 p = bb->page;
7958 lo = 0;
7959 hi = bb->count;
7960 /* Find the last range that starts before 'target' */
7961 while (hi - lo > 1) {
7962 int mid = (lo + hi) / 2;
7963 sector_t a = BB_OFFSET(p[mid]);
7964 if (a < target)
7965 lo = mid;
7966 else
7967 hi = mid;
7968 }
7969 if (hi > lo) {
7970 /* p[lo] is the last range that could overlap the
7971 * current range. Earlier ranges could also overlap,
7972 * but only this one can overlap the end of the range.
7973 */
7974 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7975 /* Partial overlap, leave the tail of this range */
7976 int ack = BB_ACK(p[lo]);
7977 sector_t a = BB_OFFSET(p[lo]);
7978 sector_t end = a + BB_LEN(p[lo]);
7979
7980 if (a < s) {
7981 /* we need to split this range */
7982 if (bb->count >= MD_MAX_BADBLOCKS) {
7983 rv = 0;
7984 goto out;
7985 }
7986 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7987 bb->count++;
7988 p[lo] = BB_MAKE(a, s-a, ack);
7989 lo++;
7990 }
7991 p[lo] = BB_MAKE(target, end - target, ack);
7992 /* there is no longer an overlap */
7993 hi = lo;
7994 lo--;
7995 }
7996 while (lo >= 0 &&
7997 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7998 /* This range does overlap */
7999 if (BB_OFFSET(p[lo]) < s) {
8000 /* Keep the early parts of this range. */
8001 int ack = BB_ACK(p[lo]);
8002 sector_t start = BB_OFFSET(p[lo]);
8003 p[lo] = BB_MAKE(start, s - start, ack);
8004 /* now low doesn't overlap, so.. */
8005 break;
8006 }
8007 lo--;
8008 }
8009 /* 'lo' is strictly before, 'hi' is strictly after,
8010 * anything between needs to be discarded
8011 */
8012 if (hi - lo > 1) {
8013 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8014 bb->count -= (hi - lo - 1);
8015 }
8016 }
8017
8018 bb->changed = 1;
8019out:
8020 write_sequnlock_irq(&bb->lock);
8021 return rv;
8022}
8023
3cb03002 8024int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
2230dfe4
N		 8025{
8026 return md_clear_badblocks(&rdev->badblocks,
8027 s + rdev->data_offset,
8028 sectors);
8029}
8030EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8031
8032/*
8033 * Acknowledge all bad blocks in a list.
8034 * This only succeeds if ->changed is clear. It is used by
8035 * in-kernel metadata updates
8036 */
8037void md_ack_all_badblocks(struct badblocks *bb)
8038{
8039 if (bb->page == NULL || bb->changed)
8040 /* no point even trying */
8041 return;
8042 write_seqlock_irq(&bb->lock);
8043
8044 if (bb->changed == 0) {
8045 u64 *p = bb->page;
8046 int i;
8047 for (i = 0; i < bb->count ; i++) {
8048 if (!BB_ACK(p[i])) {
8049 sector_t start = BB_OFFSET(p[i]);
8050 int len = BB_LEN(p[i]);
8051 p[i] = BB_MAKE(start, len, 1);
8052 }
8053 }
de393cde 8054 bb->unacked_exist = 0;
2230dfe4
N		 8055 }
8056 write_sequnlock_irq(&bb->lock);
8057}
8058EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8059
16c791a5
N		 8060/* sysfs access to bad-blocks list.
8061 * We present two files.
8062 * 'bad-blocks' lists sector numbers and lengths of ranges that
8063 * are recorded as bad. The list is truncated to fit within
8064 * the one-page limit of sysfs.
8065 * Writing "sector length" to this file adds an acknowledged
8066 * bad block list.
8067 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8068 * been acknowledged. Writing to this file adds bad blocks
8069 * without acknowledging them. This is largely for testing.
8070 */
8071
8072static ssize_t
8073badblocks_show(struct badblocks *bb, char *page, int unack)
8074{
8075 size_t len;
8076 int i;
8077 u64 *p = bb->page;
8078 unsigned seq;
8079
8080 if (bb->shift < 0)
8081 return 0;
8082
8083retry:
8084 seq = read_seqbegin(&bb->lock);
8085
8086 len = 0;
8087 i = 0;
8088
8089 while (len < PAGE_SIZE && i < bb->count) {
8090 sector_t s = BB_OFFSET(p[i]);
8091 unsigned int length = BB_LEN(p[i]);
8092 int ack = BB_ACK(p[i]);
8093 i++;
8094
8095 if (unack && ack)
8096 continue;
8097
8098 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8099 (unsigned long long)s << bb->shift,
8100 length << bb->shift);
8101 }
de393cde
N		 8102 if (unack && len == 0)
8103 bb->unacked_exist = 0;
16c791a5
N		 8104
8105 if (read_seqretry(&bb->lock, seq))
8106 goto retry;
8107
8108 return len;
8109}
8110
8111#define DO_DEBUG 1
8112
8113static ssize_t
8114badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8115{
8116 unsigned long long sector;
8117 int length;
8118 char newline;
8119#ifdef DO_DEBUG
8120 /* Allow clearing via sysfs *only* for testing/debugging.
8121 * Normally only a successful write may clear a badblock
8122 */
8123 int clear = 0;
8124 if (page[0] == '-') {
8125 clear = 1;
8126 page++;
8127 }
8128#endif /* DO_DEBUG */
8129
8130 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8131 case 3:
8132 if (newline != '\n')
8133 return -EINVAL;
8134 case 2:
8135 if (length <= 0)
8136 return -EINVAL;
8137 break;
8138 default:
8139 return -EINVAL;
8140 }
8141
8142#ifdef DO_DEBUG
8143 if (clear) {
8144 md_clear_badblocks(bb, sector, length);
8145 return len;
8146 }
8147#endif /* DO_DEBUG */
8148 if (md_set_badblocks(bb, sector, length, !unack))
8149 return len;
8150 else
8151 return -ENOSPC;
8152}
8153
75c96f85
AB		 8154static int md_notify_reboot(struct notifier_block *this,
8155 unsigned long code, void *x)
1da177e4
LT		 8156{
8157 struct list_head *tmp;
fd01b88c 8158 struct mddev *mddev;
2dba6a91 8159 int need_delay = 0;
1da177e4 8160
c744a65c
N		 8161 for_each_mddev(mddev, tmp) {
8162 if (mddev_trylock(mddev)) {
8163 __md_stop_writes(mddev);
8164 mddev->safemode = 2;
8165 mddev_unlock(mddev);
2dba6a91 8166 }
c744a65c 8167 need_delay = 1;
1da177e4 8168 }
c744a65c
N		 8169 /*
8170 * certain more exotic SCSI devices are known to be
8171 * volatile wrt too early system reboots. While the
8172 * right place to handle this issue is the given
8173 * driver, we do want to have a safe RAID driver ...
8174 */
8175 if (need_delay)
8176 mdelay(1000*1);
8177
1da177e4
LT		 8178 return NOTIFY_DONE;
8179}
8180
75c96f85 8181static struct notifier_block md_notifier = {
1da177e4
LT		 8182 .notifier_call = md_notify_reboot,
8183 .next = NULL,
8184 .priority = INT_MAX, /* before any real devices */
8185};
8186
8187static void md_geninit(void)
8188{
36a4e1fe 8189 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8190
c7705f34 8191 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8192}
8193
75c96f85 8194static int __init md_init(void)
1da177e4 8195{
e804ac78
TH		 8196 int ret = -ENOMEM;
8197
ada609ee 8198 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8199 if (!md_wq)
8200 goto err_wq;
8201
8202 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8203 if (!md_misc_wq)
8204 goto err_misc_wq;
8205
8206 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8207 goto err_md;
8208
8209 if ((ret = register_blkdev(0, "mdp")) < 0)
8210 goto err_mdp;
8211 mdp_major = ret;
8212
3dbd8c2e 8213 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8214 md_probe, NULL, NULL);
8215 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8216 md_probe, NULL, NULL);
8217
1da177e4 8218 register_reboot_notifier(&md_notifier);
0b4d4147 8219 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8220
8221 md_geninit();
d710e138 8222 return 0;
1da177e4 8223
e804ac78
TH		 8224err_mdp:
8225 unregister_blkdev(MD_MAJOR, "md");
8226err_md:
8227 destroy_workqueue(md_misc_wq);
8228err_misc_wq:
8229 destroy_workqueue(md_wq);
8230err_wq:
8231 return ret;
8232}
1da177e4
LT		 8233
8234#ifndef MODULE
8235
8236/*
8237 * Searches all registered partitions for autorun RAID arrays
8238 * at boot time.
8239 */
4d936ec1
ME		 8240
8241static LIST_HEAD(all_detected_devices);
8242struct detected_devices_node {
8243 struct list_head list;
8244 dev_t dev;
8245};
1da177e4
LT		 8246
8247void md_autodetect_dev(dev_t dev)
8248{
4d936ec1
ME		 8249 struct detected_devices_node *node_detected_dev;
8250
8251 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8252 if (node_detected_dev) {
8253 node_detected_dev->dev = dev;
8254 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8255 } else {
8256 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8257 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8258 }
1da177e4
LT		 8259}
8260
8261
8262static void autostart_arrays(int part)
8263{
3cb03002 8264 struct md_rdev *rdev;
4d936ec1
ME		 8265 struct detected_devices_node *node_detected_dev;
8266 dev_t dev;
8267 int i_scanned, i_passed;
1da177e4 8268
4d936ec1
ME		 8269 i_scanned = 0;
8270 i_passed = 0;
1da177e4 8271
4d936ec1 8272 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8273
4d936ec1
ME		 8274 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8275 i_scanned++;
8276 node_detected_dev = list_entry(all_detected_devices.next,
8277 struct detected_devices_node, list);
8278 list_del(&node_detected_dev->list);
8279 dev = node_detected_dev->dev;
8280 kfree(node_detected_dev);
df968c4e 8281 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8282 if (IS_ERR(rdev))
8283 continue;
8284
b2d444d7 8285 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8286 MD_BUG();
8287 continue;
8288 }
d0fae18f 8289 set_bit(AutoDetected, &rdev->flags);
1da177e4 8290 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8291 i_passed++;
1da177e4 8292 }
4d936ec1
ME		 8293
8294 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8295 i_scanned, i_passed);
1da177e4
LT		 8296
8297 autorun_devices(part);
8298}
8299
fdee8ae4 8300#endif /* !MODULE */
1da177e4
LT		 8301
8302static __exit void md_exit(void)
8303{
fd01b88c 8304 struct mddev *mddev;
1da177e4 8305 struct list_head *tmp;
8ab5e4c1 8306
3dbd8c2e 8307 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8308 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8309
3dbd8c2e 8310 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8311 unregister_blkdev(mdp_major, "mdp");
8312 unregister_reboot_notifier(&md_notifier);
8313 unregister_sysctl_table(raid_table_header);
8314 remove_proc_entry("mdstat", NULL);
29ac4aa3 8315 for_each_mddev(mddev, tmp) {
1da177e4 8316 export_array(mddev);
d3374825 8317 mddev->hold_active = 0;
1da177e4 8318 }
e804ac78
TH		 8319 destroy_workqueue(md_misc_wq);
8320 destroy_workqueue(md_wq);
1da177e4
LT		 8321}
8322
685784aa 8323subsys_initcall(md_init);
1da177e4
LT		 8324module_exit(md_exit)
8325
f91de92e
N		 8326static int get_ro(char *buffer, struct kernel_param *kp)
8327{
8328 return sprintf(buffer, "%d", start_readonly);
8329}
8330static int set_ro(const char *val, struct kernel_param *kp)
8331{
8332 char *e;
8333 int num = simple_strtoul(val, &e, 10);
8334 if (*val && (*e == '\0' || *e == '\n')) {
8335 start_readonly = num;
4dbcdc75 8336 return 0;
f91de92e
N		 8337 }
8338 return -EINVAL;
8339}
8340
80ca3a44
N		 8341module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8342module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8343
efeb53c0 8344module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8345
1da177e4
LT		 8346EXPORT_SYMBOL(register_md_personality);
8347EXPORT_SYMBOL(unregister_md_personality);
8348EXPORT_SYMBOL(md_error);
8349EXPORT_SYMBOL(md_done_sync);
8350EXPORT_SYMBOL(md_write_start);
8351EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8352EXPORT_SYMBOL(md_register_thread);
8353EXPORT_SYMBOL(md_unregister_thread);
8354EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 8355EXPORT_SYMBOL(md_check_recovery);
8356MODULE_LICENSE("GPL");
0efb9e61 8357MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8358MODULE_ALIAS("md");
72008652 8359MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
kernel source for telekom puls (alcatel/mediatek)