projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
md: Add module.h to all files using it implicitly
[deliverable/linux.git] / drivers / md / md.c
CommitLineData
1da177e4
LT		 1/*
2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5 completely rewritten, based on the MD driver code from Marc Zyngier
6
7 Changes:
8
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
19
20 Neil Brown <neilb@cse.unsw.edu.au>.
21
32a7627c
N		 22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
1da177e4
LT		 25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
28 any later version.
29
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33*/
34
a6fb0934 35#include <linux/kthread.h>
bff61975 36#include <linux/blkdev.h>
1da177e4 37#include <linux/sysctl.h>
bff61975 38#include <linux/seq_file.h>
2a48fc0a 39#include <linux/mutex.h>
1da177e4 40#include <linux/buffer_head.h> /* for invalidate_bdev */
d7603b7e 41#include <linux/poll.h>
16f17b39 42#include <linux/ctype.h>
e7d2860b 43#include <linux/string.h>
fb4d8c76
N		 44#include <linux/hdreg.h>
45#include <linux/proc_fs.h>
46#include <linux/random.h>
056075c7 47#include <linux/module.h>
fb4d8c76 48#include <linux/reboot.h>
32a7627c 49#include <linux/file.h>
aa98aa31 50#include <linux/compat.h>
25570727 51#include <linux/delay.h>
bff61975
N		 52#include <linux/raid/md_p.h>
53#include <linux/raid/md_u.h>
5a0e3ad6 54#include <linux/slab.h>
43b2e5d8 55#include "md.h"
ef740c37 56#include "bitmap.h"
1da177e4 57
1da177e4 58#ifndef MODULE
d710e138 59static void autostart_arrays(int part);
1da177e4
LT		 60#endif
61
01f96c0a
N		 62/* pers_list is a list of registered personalities protected
63 * by pers_lock.
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
66 */
2604b703 67static LIST_HEAD(pers_list);
1da177e4
LT		 68static DEFINE_SPINLOCK(pers_lock);
69
5e56341d
AB		 70static void md_print_devices(void);
71
90b08710 72static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
e804ac78
TH		 73static struct workqueue_struct *md_wq;
74static struct workqueue_struct *md_misc_wq;
90b08710 75
5e56341d
AB		 76#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
1e50915f
RB		 78/*
79 * Default number of read corrections we'll attempt on an rdev
80 * before ejecting it from the array. We divide the read error
81 * count by 2 for every hour elapsed between read errors.
82 */
83#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
1da177e4
LT		 84/*
85 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
86 * is 1000 KB/sec, so the extra system load does not show up that much.
87 * Increase it if you want to have more _guaranteed_ speed. Note that
338cec32 88 * the RAID driver will use the maximum available bandwidth if the IO
1da177e4
LT		 89 * subsystem is idle. There is also an 'absolute maximum' reconstruction
90 * speed limit - in case reconstruction slows down your system despite
91 * idle IO detection.
92 *
93 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88202a0c 94 * or /sys/block/mdX/md/sync_speed_{min,max}
1da177e4
LT		 95 */
96
97static int sysctl_speed_limit_min = 1000;
98static int sysctl_speed_limit_max = 200000;
fd01b88c 99static inline int speed_min(struct mddev *mddev)
88202a0c
N		 100{
101 return mddev->sync_speed_min ?
102 mddev->sync_speed_min : sysctl_speed_limit_min;
103}
104
fd01b88c 105static inline int speed_max(struct mddev *mddev)
88202a0c
N		 106{
107 return mddev->sync_speed_max ?
108 mddev->sync_speed_max : sysctl_speed_limit_max;
109}
1da177e4
LT		 110
111static struct ctl_table_header *raid_table_header;
112
113static ctl_table raid_table[] = {
114 {
1da177e4
LT		 115 .procname = "speed_limit_min",
116 .data = &sysctl_speed_limit_min,
117 .maxlen = sizeof(int),
80ca3a44 118 .mode = S_IRUGO|S_IWUSR,
6d456111 119 .proc_handler = proc_dointvec,
1da177e4
LT		 120 },
121 {
1da177e4
LT		 122 .procname = "speed_limit_max",
123 .data = &sysctl_speed_limit_max,
124 .maxlen = sizeof(int),
80ca3a44 125 .mode = S_IRUGO|S_IWUSR,
6d456111 126 .proc_handler = proc_dointvec,
1da177e4 127 },
894d2491 128 { }
1da177e4
LT		 129};
130
131static ctl_table raid_dir_table[] = {
132 {
1da177e4
LT		 133 .procname = "raid",
134 .maxlen = 0,
80ca3a44 135 .mode = S_IRUGO|S_IXUGO,
1da177e4
LT		 136 .child = raid_table,
137 },
894d2491 138 { }
1da177e4
LT		 139};
140
141static ctl_table raid_root_table[] = {
142 {
1da177e4
LT		 143 .procname = "dev",
144 .maxlen = 0,
145 .mode = 0555,
146 .child = raid_dir_table,
147 },
894d2491 148 { }
1da177e4
LT		 149};
150
83d5cde4 151static const struct block_device_operations md_fops;
1da177e4 152
f91de92e
N		 153static int start_readonly;
154
a167f663
N		 155/* bio_clone_mddev
156 * like bio_clone, but with a local bio set
157 */
158
159static void mddev_bio_destructor(struct bio *bio)
160{
fd01b88c 161 struct mddev *mddev, **mddevp;
a167f663
N		 162
163 mddevp = (void*)bio;
164 mddev = mddevp[-1];
165
166 bio_free(bio, mddev->bio_set);
167}
168
169struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
fd01b88c 170 struct mddev *mddev)
a167f663
N		 171{
172 struct bio *b;
fd01b88c 173 struct mddev **mddevp;
a167f663
N		 174
175 if (!mddev || !mddev->bio_set)
176 return bio_alloc(gfp_mask, nr_iovecs);
177
178 b = bio_alloc_bioset(gfp_mask, nr_iovecs,
179 mddev->bio_set);
180 if (!b)
181 return NULL;
182 mddevp = (void*)b;
183 mddevp[-1] = mddev;
184 b->bi_destructor = mddev_bio_destructor;
185 return b;
186}
187EXPORT_SYMBOL_GPL(bio_alloc_mddev);
188
189struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
fd01b88c 190 struct mddev *mddev)
a167f663
N		 191{
192 struct bio *b;
fd01b88c 193 struct mddev **mddevp;
a167f663
N		 194
195 if (!mddev || !mddev->bio_set)
196 return bio_clone(bio, gfp_mask);
197
198 b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
199 mddev->bio_set);
200 if (!b)
201 return NULL;
202 mddevp = (void*)b;
203 mddevp[-1] = mddev;
204 b->bi_destructor = mddev_bio_destructor;
205 __bio_clone(b, bio);
206 if (bio_integrity(bio)) {
207 int ret;
208
209 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
210
211 if (ret < 0) {
212 bio_put(b);
213 return NULL;
214 }
215 }
216
217 return b;
218}
219EXPORT_SYMBOL_GPL(bio_clone_mddev);
220
d2eb35ac
N		 221void md_trim_bio(struct bio *bio, int offset, int size)
222{
223 /* 'bio' is a cloned bio which we need to trim to match
224 * the given offset and size.
225 * This requires adjusting bi_sector, bi_size, and bi_io_vec
226 */
227 int i;
228 struct bio_vec *bvec;
229 int sofar = 0;
230
231 size <<= 9;
232 if (offset == 0 && size == bio->bi_size)
233 return;
234
235 bio->bi_sector += offset;
236 bio->bi_size = size;
237 offset <<= 9;
238 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
239
240 while (bio->bi_idx < bio->bi_vcnt &&
241 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
242 /* remove this whole bio_vec */
243 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
244 bio->bi_idx++;
245 }
246 if (bio->bi_idx < bio->bi_vcnt) {
247 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
248 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
249 }
250 /* avoid any complications with bi_idx being non-zero*/
251 if (bio->bi_idx) {
252 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
253 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
254 bio->bi_vcnt -= bio->bi_idx;
255 bio->bi_idx = 0;
256 }
257 /* Make sure vcnt and last bv are not too big */
258 bio_for_each_segment(bvec, bio, i) {
259 if (sofar + bvec->bv_len > size)
260 bvec->bv_len = size - sofar;
261 if (bvec->bv_len == 0) {
262 bio->bi_vcnt = i;
263 break;
264 }
265 sofar += bvec->bv_len;
266 }
267}
268EXPORT_SYMBOL_GPL(md_trim_bio);
269
d7603b7e
N		 270/*
271 * We have a system wide 'event count' that is incremented
272 * on any 'interesting' event, and readers of /proc/mdstat
273 * can use 'poll' or 'select' to find out when the event
274 * count increases.
275 *
276 * Events are:
277 * start array, stop array, error, add device, remove device,
278 * start build, activate spare
279 */
2989ddbd 280static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
d7603b7e 281static atomic_t md_event_count;
fd01b88c 282void md_new_event(struct mddev *mddev)
d7603b7e
N		 283{
284 atomic_inc(&md_event_count);
285 wake_up(&md_event_waiters);
286}
29269553 287EXPORT_SYMBOL_GPL(md_new_event);
d7603b7e 288
c331eb04
N		 289/* Alternate version that can be called from interrupts
290 * when calling sysfs_notify isn't needed.
291 */
fd01b88c 292static void md_new_event_inintr(struct mddev *mddev)
c331eb04
N		 293{
294 atomic_inc(&md_event_count);
295 wake_up(&md_event_waiters);
296}
297
1da177e4
LT		 298/*
299 * Enables to iterate over all existing md arrays
300 * all_mddevs_lock protects this list.
301 */
302static LIST_HEAD(all_mddevs);
303static DEFINE_SPINLOCK(all_mddevs_lock);
304
305
306/*
307 * iterates through all used mddevs in the system.
308 * We take care to grab the all_mddevs_lock whenever navigating
309 * the list, and to always hold a refcount when unlocked.
310 * Any code which breaks out of this loop while own
311 * a reference to the current mddev and must mddev_put it.
312 */
fd01b88c 313#define for_each_mddev(_mddev,_tmp) \
1da177e4
LT		 314 \
315 for (({ spin_lock(&all_mddevs_lock); \
fd01b88c
N		 316 _tmp = all_mddevs.next; \
317 _mddev = NULL;}); \
318 ({ if (_tmp != &all_mddevs) \
319 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
1da177e4 320 spin_unlock(&all_mddevs_lock); \
fd01b88c
N		 321 if (_mddev) mddev_put(_mddev); \
322 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
323 _tmp != &all_mddevs;}); \
1da177e4 324 ({ spin_lock(&all_mddevs_lock); \
fd01b88c 325 _tmp = _tmp->next;}) \
1da177e4
LT		 326 )
327
328
409c57f3
N		 329/* Rather than calling directly into the personality make_request function,
330 * IO requests come here first so that we can check if the device is
331 * being suspended pending a reconfiguration.
332 * We hold a refcount over the call to ->make_request. By the time that
333 * call has finished, the bio has been linked into some internal structure
334 * and so is visible to ->quiesce(), so we don't need the refcount any more.
335 */
336static int md_make_request(struct request_queue *q, struct bio *bio)
1da177e4 337{
49077326 338 const int rw = bio_data_dir(bio);
fd01b88c 339 struct mddev *mddev = q->queuedata;
409c57f3 340 int rv;
49077326 341 int cpu;
e91ece55 342 unsigned int sectors;
49077326 343
0ca69886
N		 344 if (mddev == NULL || mddev->pers == NULL
345 || !mddev->ready) {
409c57f3
N		 346 bio_io_error(bio);
347 return 0;
348 }
0ca69886 349 smp_rmb(); /* Ensure implications of 'active' are visible */
409c57f3 350 rcu_read_lock();
e9c7469b 351 if (mddev->suspended) {
409c57f3
N		 352 DEFINE_WAIT(__wait);
353 for (;;) {
354 prepare_to_wait(&mddev->sb_wait, &__wait,
355 TASK_UNINTERRUPTIBLE);
e9c7469b 356 if (!mddev->suspended)
409c57f3
N		 357 break;
358 rcu_read_unlock();
359 schedule();
360 rcu_read_lock();
361 }
362 finish_wait(&mddev->sb_wait, &__wait);
363 }
364 atomic_inc(&mddev->active_io);
365 rcu_read_unlock();
49077326 366
e91ece55
CM		 367 /*
368 * save the sectors now since our bio can
369 * go away inside make_request
370 */
371 sectors = bio_sectors(bio);
21a52c6d 372 rv = mddev->pers->make_request(mddev, bio);
49077326
N		 373
374 cpu = part_stat_lock();
375 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
e91ece55 376 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
49077326
N		 377 part_stat_unlock();
378
409c57f3
N		 379 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
380 wake_up(&mddev->sb_wait);
381
382 return rv;
383}
384
9e35b99c
N		 385/* mddev_suspend makes sure no new requests are submitted
386 * to the device, and that any requests that have been submitted
387 * are completely handled.
388 * Once ->stop is called and completes, the module will be completely
389 * unused.
390 */
fd01b88c 391void mddev_suspend(struct mddev *mddev)
409c57f3
N		 392{
393 BUG_ON(mddev->suspended);
394 mddev->suspended = 1;
395 synchronize_rcu();
396 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
397 mddev->pers->quiesce(mddev, 1);
409c57f3 398}
390ee602 399EXPORT_SYMBOL_GPL(mddev_suspend);
409c57f3 400
fd01b88c 401void mddev_resume(struct mddev *mddev)
409c57f3
N		 402{
403 mddev->suspended = 0;
404 wake_up(&mddev->sb_wait);
405 mddev->pers->quiesce(mddev, 0);
0fd018af
JB		 406
407 md_wakeup_thread(mddev->thread);
408 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
1da177e4 409}
390ee602 410EXPORT_SYMBOL_GPL(mddev_resume);
1da177e4 411
fd01b88c 412int mddev_congested(struct mddev *mddev, int bits)
3fa841d7
N		 413{
414 return mddev->suspended;
415}
416EXPORT_SYMBOL(mddev_congested);
417
a2826aa9 418/*
e9c7469b 419 * Generic flush handling for md
a2826aa9
N		 420 */
421
e9c7469b 422static void md_end_flush(struct bio *bio, int err)
a2826aa9 423{
3cb03002 424 struct md_rdev *rdev = bio->bi_private;
fd01b88c 425 struct mddev *mddev = rdev->mddev;
a2826aa9
N		 426
427 rdev_dec_pending(rdev, mddev);
428
429 if (atomic_dec_and_test(&mddev->flush_pending)) {
e9c7469b 430 /* The pre-request flush has finished */
e804ac78 431 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 432 }
433 bio_put(bio);
434}
435
a7a07e69
N		 436static void md_submit_flush_data(struct work_struct *ws);
437
a035fc3e 438static void submit_flushes(struct work_struct *ws)
a2826aa9 439{
fd01b88c 440 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
3cb03002 441 struct md_rdev *rdev;
a2826aa9 442
a7a07e69
N		 443 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
444 atomic_set(&mddev->flush_pending, 1);
a2826aa9
N		 445 rcu_read_lock();
446 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
447 if (rdev->raid_disk >= 0 &&
448 !test_bit(Faulty, &rdev->flags)) {
449 /* Take two references, one is dropped
450 * when request finishes, one after
451 * we reclaim rcu_read_lock
452 */
453 struct bio *bi;
454 atomic_inc(&rdev->nr_pending);
455 atomic_inc(&rdev->nr_pending);
456 rcu_read_unlock();
a167f663 457 bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
e9c7469b 458 bi->bi_end_io = md_end_flush;
a2826aa9
N		 459 bi->bi_private = rdev;
460 bi->bi_bdev = rdev->bdev;
461 atomic_inc(&mddev->flush_pending);
e9c7469b 462 submit_bio(WRITE_FLUSH, bi);
a2826aa9
N		 463 rcu_read_lock();
464 rdev_dec_pending(rdev, mddev);
465 }
466 rcu_read_unlock();
a7a07e69
N		 467 if (atomic_dec_and_test(&mddev->flush_pending))
468 queue_work(md_wq, &mddev->flush_work);
a2826aa9
N		 469}
470
e9c7469b 471static void md_submit_flush_data(struct work_struct *ws)
a2826aa9 472{
fd01b88c 473 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
e9c7469b 474 struct bio *bio = mddev->flush_bio;
a2826aa9 475
e9c7469b 476 if (bio->bi_size == 0)
a2826aa9
N		 477 /* an empty barrier - all done */
478 bio_endio(bio, 0);
479 else {
e9c7469b 480 bio->bi_rw &= ~REQ_FLUSH;
21a52c6d 481 if (mddev->pers->make_request(mddev, bio))
a2826aa9 482 generic_make_request(bio);
a2826aa9 483 }
2b74e12e
N		 484
485 mddev->flush_bio = NULL;
486 wake_up(&mddev->sb_wait);
a2826aa9
N		 487}
488
fd01b88c 489void md_flush_request(struct mddev *mddev, struct bio *bio)
a2826aa9
N		 490{
491 spin_lock_irq(&mddev->write_lock);
492 wait_event_lock_irq(mddev->sb_wait,
e9c7469b 493 !mddev->flush_bio,
a2826aa9 494 mddev->write_lock, /*nothing*/);
e9c7469b 495 mddev->flush_bio = bio;
a2826aa9
N		 496 spin_unlock_irq(&mddev->write_lock);
497
a035fc3e
N		 498 INIT_WORK(&mddev->flush_work, submit_flushes);
499 queue_work(md_wq, &mddev->flush_work);
a2826aa9 500}
e9c7469b 501EXPORT_SYMBOL(md_flush_request);
409c57f3 502
97658cdd
N		 503/* Support for plugging.
504 * This mirrors the plugging support in request_queue, but does not
505 * require having a whole queue or request structures.
506 * We allocate an md_plug_cb for each md device and each thread it gets
507 * plugged on. This links tot the private plug_handle structure in the
508 * personality data where we keep a count of the number of outstanding
509 * plugs so other code can see if a plug is active.
510 */
511struct md_plug_cb {
512 struct blk_plug_cb cb;
fd01b88c 513 struct mddev *mddev;
97658cdd
N		 514};
515
516static void plugger_unplug(struct blk_plug_cb *cb)
517{
518 struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
519 if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
520 md_wakeup_thread(mdcb->mddev->thread);
521 kfree(mdcb);
522}
523
524/* Check that an unplug wakeup will come shortly.
525 * If not, wakeup the md thread immediately
526 */
fd01b88c 527int mddev_check_plugged(struct mddev *mddev)
97658cdd
N		 528{
529 struct blk_plug *plug = current->plug;
530 struct md_plug_cb *mdcb;
531
532 if (!plug)
533 return 0;
534
535 list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
536 if (mdcb->cb.callback == plugger_unplug &&
537 mdcb->mddev == mddev) {
538 /* Already on the list, move to top */
539 if (mdcb != list_first_entry(&plug->cb_list,
540 struct md_plug_cb,
541 cb.list))
542 list_move(&mdcb->cb.list, &plug->cb_list);
543 return 1;
544 }
545 }
546 /* Not currently on the callback list */
547 mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
548 if (!mdcb)
549 return 0;
550
551 mdcb->mddev = mddev;
552 mdcb->cb.callback = plugger_unplug;
553 atomic_inc(&mddev->plug_cnt);
554 list_add(&mdcb->cb.list, &plug->cb_list);
555 return 1;
556}
557EXPORT_SYMBOL_GPL(mddev_check_plugged);
2ac87401 558
fd01b88c 559static inline struct mddev *mddev_get(struct mddev *mddev)
1da177e4
LT		 560{
561 atomic_inc(&mddev->active);
562 return mddev;
563}
564
5fd3a17e 565static void mddev_delayed_delete(struct work_struct *ws);
d3374825 566
fd01b88c 567static void mddev_put(struct mddev *mddev)
1da177e4 568{
a167f663
N		 569 struct bio_set *bs = NULL;
570
1da177e4
LT		 571 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
572 return;
d3374825 573 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
cbd19983
N		 574 mddev->ctime == 0 && !mddev->hold_active) {
575 /* Array is not configured at all, and not held active,
576 * so destroy it */
1da177e4 577 list_del(&mddev->all_mddevs);
a167f663
N		 578 bs = mddev->bio_set;
579 mddev->bio_set = NULL;
d3374825 580 if (mddev->gendisk) {
e804ac78
TH		 581 /* We did a probe so need to clean up. Call
582 * queue_work inside the spinlock so that
583 * flush_workqueue() after mddev_find will
584 * succeed in waiting for the work to be done.
d3374825
N		 585 */
586 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
e804ac78 587 queue_work(md_misc_wq, &mddev->del_work);
d3374825
N		 588 } else
589 kfree(mddev);
590 }
591 spin_unlock(&all_mddevs_lock);
a167f663
N		 592 if (bs)
593 bioset_free(bs);
1da177e4
LT		 594}
595
fd01b88c 596void mddev_init(struct mddev *mddev)
fafd7fb0
N		 597{
598 mutex_init(&mddev->open_mutex);
599 mutex_init(&mddev->reconfig_mutex);
600 mutex_init(&mddev->bitmap_info.mutex);
601 INIT_LIST_HEAD(&mddev->disks);
602 INIT_LIST_HEAD(&mddev->all_mddevs);
603 init_timer(&mddev->safemode_timer);
604 atomic_set(&mddev->active, 1);
605 atomic_set(&mddev->openers, 0);
606 atomic_set(&mddev->active_io, 0);
97658cdd 607 atomic_set(&mddev->plug_cnt, 0);
fafd7fb0
N		 608 spin_lock_init(&mddev->write_lock);
609 atomic_set(&mddev->flush_pending, 0);
610 init_waitqueue_head(&mddev->sb_wait);
611 init_waitqueue_head(&mddev->recovery_wait);
612 mddev->reshape_position = MaxSector;
613 mddev->resync_min = 0;
614 mddev->resync_max = MaxSector;
615 mddev->level = LEVEL_NONE;
616}
390ee602 617EXPORT_SYMBOL_GPL(mddev_init);
fafd7fb0 618
fd01b88c 619static struct mddev * mddev_find(dev_t unit)
1da177e4 620{
fd01b88c 621 struct mddev *mddev, *new = NULL;
1da177e4 622
8f5f02c4
N		 623 if (unit && MAJOR(unit) != MD_MAJOR)
624 unit &= ~((1<<MdpMinorShift)-1);
625
1da177e4
LT		 626 retry:
627 spin_lock(&all_mddevs_lock);
efeb53c0
N		 628
629 if (unit) {
630 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
631 if (mddev->unit == unit) {
632 mddev_get(mddev);
633 spin_unlock(&all_mddevs_lock);
634 kfree(new);
635 return mddev;
636 }
637
638 if (new) {
639 list_add(&new->all_mddevs, &all_mddevs);
1da177e4 640 spin_unlock(&all_mddevs_lock);
efeb53c0
N		 641 new->hold_active = UNTIL_IOCTL;
642 return new;
1da177e4 643 }
efeb53c0
N		 644 } else if (new) {
645 /* find an unused unit number */
646 static int next_minor = 512;
647 int start = next_minor;
648 int is_free = 0;
649 int dev = 0;
650 while (!is_free) {
651 dev = MKDEV(MD_MAJOR, next_minor);
652 next_minor++;
653 if (next_minor > MINORMASK)
654 next_minor = 0;
655 if (next_minor == start) {
656 /* Oh dear, all in use. */
657 spin_unlock(&all_mddevs_lock);
658 kfree(new);
659 return NULL;
660 }
661
662 is_free = 1;
663 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
664 if (mddev->unit == dev) {
665 is_free = 0;
666 break;
667 }
668 }
669 new->unit = dev;
670 new->md_minor = MINOR(dev);
671 new->hold_active = UNTIL_STOP;
1da177e4
LT		 672 list_add(&new->all_mddevs, &all_mddevs);
673 spin_unlock(&all_mddevs_lock);
674 return new;
675 }
676 spin_unlock(&all_mddevs_lock);
677
9ffae0cf 678 new = kzalloc(sizeof(*new), GFP_KERNEL);
1da177e4
LT		 679 if (!new)
680 return NULL;
681
1da177e4
LT		 682 new->unit = unit;
683 if (MAJOR(unit) == MD_MAJOR)
684 new->md_minor = MINOR(unit);
685 else
686 new->md_minor = MINOR(unit) >> MdpMinorShift;
687
fafd7fb0 688 mddev_init(new);
1da177e4 689
1da177e4
LT		 690 goto retry;
691}
692
fd01b88c 693static inline int mddev_lock(struct mddev * mddev)
1da177e4 694{
df5b89b3 695 return mutex_lock_interruptible(&mddev->reconfig_mutex);
1da177e4
LT		 696}
697
fd01b88c 698static inline int mddev_is_locked(struct mddev *mddev)
b522adcd
DW		 699{
700 return mutex_is_locked(&mddev->reconfig_mutex);
701}
702
fd01b88c 703static inline int mddev_trylock(struct mddev * mddev)
1da177e4 704{
df5b89b3 705 return mutex_trylock(&mddev->reconfig_mutex);
1da177e4
LT		 706}
707
b6eb127d
N		 708static struct attribute_group md_redundancy_group;
709
fd01b88c 710static void mddev_unlock(struct mddev * mddev)
1da177e4 711{
a64c876f 712 if (mddev->to_remove) {
b6eb127d
N		 713 /* These cannot be removed under reconfig_mutex as
714 * an access to the files will try to take reconfig_mutex
715 * while holding the file unremovable, which leads to
716 * a deadlock.
bb4f1e9d
N		 717 * So hold set sysfs_active while the remove in happeing,
718 * and anything else which might set ->to_remove or my
719 * otherwise change the sysfs namespace will fail with
720 * -EBUSY if sysfs_active is still set.
721 * We set sysfs_active under reconfig_mutex and elsewhere
722 * test it under the same mutex to ensure its correct value
723 * is seen.
b6eb127d 724 */
a64c876f
N		 725 struct attribute_group *to_remove = mddev->to_remove;
726 mddev->to_remove = NULL;
bb4f1e9d 727 mddev->sysfs_active = 1;
b6eb127d
N		 728 mutex_unlock(&mddev->reconfig_mutex);
729
00bcb4ac
N		 730 if (mddev->kobj.sd) {
731 if (to_remove != &md_redundancy_group)
732 sysfs_remove_group(&mddev->kobj, to_remove);
733 if (mddev->pers == NULL ||
734 mddev->pers->sync_request == NULL) {
735 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
736 if (mddev->sysfs_action)
737 sysfs_put(mddev->sysfs_action);
738 mddev->sysfs_action = NULL;
739 }
a64c876f 740 }
bb4f1e9d 741 mddev->sysfs_active = 0;
b6eb127d
N		 742 } else
743 mutex_unlock(&mddev->reconfig_mutex);
1da177e4 744
751e67ca
CD		 745 /* As we've dropped the mutex we need a spinlock to
746 * make sure the thread doesn't disappear
01f96c0a
N		 747 */
748 spin_lock(&pers_lock);
005eca5e 749 md_wakeup_thread(mddev->thread);
01f96c0a 750 spin_unlock(&pers_lock);
1da177e4
LT		 751}
752
fd01b88c 753static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
1da177e4 754{
3cb03002 755 struct md_rdev *rdev;
1da177e4 756
159ec1fc 757 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 758 if (rdev->desc_nr == nr)
759 return rdev;
159ec1fc 760
1da177e4
LT		 761 return NULL;
762}
763
fd01b88c 764static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
1da177e4 765{
3cb03002 766 struct md_rdev *rdev;
1da177e4 767
159ec1fc 768 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 769 if (rdev->bdev->bd_dev == dev)
770 return rdev;
159ec1fc 771
1da177e4
LT		 772 return NULL;
773}
774
84fc4b56 775static struct md_personality *find_pers(int level, char *clevel)
2604b703 776{
84fc4b56 777 struct md_personality *pers;
d9d166c2
N		 778 list_for_each_entry(pers, &pers_list, list) {
779 if (level != LEVEL_NONE && pers->level == level)
2604b703 780 return pers;
d9d166c2
N		 781 if (strcmp(pers->name, clevel)==0)
782 return pers;
783 }
2604b703
N		 784 return NULL;
785}
786
b73df2d3 787/* return the offset of the super block in 512byte sectors */
3cb03002 788static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
1da177e4 789{
57b2caa3 790 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
b73df2d3 791 return MD_NEW_SIZE_SECTORS(num_sectors);
1da177e4
LT		 792}
793
3cb03002 794static int alloc_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 795{
796 if (rdev->sb_page)
797 MD_BUG();
798
799 rdev->sb_page = alloc_page(GFP_KERNEL);
800 if (!rdev->sb_page) {
801 printk(KERN_ALERT "md: out of memory.\n");
ebc24337 802 return -ENOMEM;
1da177e4
LT		 803 }
804
805 return 0;
806}
807
3cb03002 808static void free_disk_sb(struct md_rdev * rdev)
1da177e4
LT		 809{
810 if (rdev->sb_page) {
2d1f3b5d 811 put_page(rdev->sb_page);
1da177e4
LT		 812 rdev->sb_loaded = 0;
813 rdev->sb_page = NULL;
0f420358 814 rdev->sb_start = 0;
dd8ac336 815 rdev->sectors = 0;
1da177e4 816 }
2699b672
N		 817 if (rdev->bb_page) {
818 put_page(rdev->bb_page);
819 rdev->bb_page = NULL;
820 }
1da177e4
LT		 821}
822
823
6712ecf8 824static void super_written(struct bio *bio, int error)
7bfa19f2 825{
3cb03002 826 struct md_rdev *rdev = bio->bi_private;
fd01b88c 827 struct mddev *mddev = rdev->mddev;
7bfa19f2 828
3a0f5bbb
N		 829 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
830 printk("md: super_written gets error=%d, uptodate=%d\n",
831 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
832 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
a9701a30 833 md_error(mddev, rdev);
3a0f5bbb 834 }
7bfa19f2 835
a9701a30
N		 836 if (atomic_dec_and_test(&mddev->pending_writes))
837 wake_up(&mddev->sb_wait);
f8b58edf 838 bio_put(bio);
7bfa19f2
N		 839}
840
fd01b88c 841void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
7bfa19f2
N		 842 sector_t sector, int size, struct page *page)
843{
844 /* write first size bytes of page to sector of rdev
845 * Increment mddev->pending_writes before returning
846 * and decrement it on completion, waking up sb_wait
847 * if zero is reached.
848 * If an error occurred, call md_error
849 */
a167f663 850 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
7bfa19f2 851
a6ff7e08 852 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
7bfa19f2
N		 853 bio->bi_sector = sector;
854 bio_add_page(bio, page, size, 0);
855 bio->bi_private = rdev;
856 bio->bi_end_io = super_written;
a9701a30 857
7bfa19f2 858 atomic_inc(&mddev->pending_writes);
a5bf4df0 859 submit_bio(WRITE_FLUSH_FUA, bio);
a9701a30
N		 860}
861
fd01b88c 862void md_super_wait(struct mddev *mddev)
a9701a30 863{
e9c7469b 864 /* wait for all superblock writes that were scheduled to complete */
a9701a30
N		 865 DEFINE_WAIT(wq);
866 for(;;) {
867 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
868 if (atomic_read(&mddev->pending_writes)==0)
869 break;
a9701a30
N		 870 schedule();
871 }
872 finish_wait(&mddev->sb_wait, &wq);
7bfa19f2
N		 873}
874
6712ecf8 875static void bi_complete(struct bio *bio, int error)
1da177e4 876{
1da177e4 877 complete((struct completion*)bio->bi_private);
1da177e4
LT		 878}
879
3cb03002 880int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
ccebd4c4 881 struct page *page, int rw, bool metadata_op)
1da177e4 882{
a167f663 883 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
1da177e4
LT		 884 struct completion event;
885 int ret;
886
721a9602 887 rw |= REQ_SYNC;
1da177e4 888
a6ff7e08
JB		 889 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
890 rdev->meta_bdev : rdev->bdev;
ccebd4c4
JB		 891 if (metadata_op)
892 bio->bi_sector = sector + rdev->sb_start;
893 else
894 bio->bi_sector = sector + rdev->data_offset;
1da177e4
LT		 895 bio_add_page(bio, page, size, 0);
896 init_completion(&event);
897 bio->bi_private = &event;
898 bio->bi_end_io = bi_complete;
899 submit_bio(rw, bio);
900 wait_for_completion(&event);
901
902 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
903 bio_put(bio);
904 return ret;
905}
a8745db2 906EXPORT_SYMBOL_GPL(sync_page_io);
1da177e4 907
3cb03002 908static int read_disk_sb(struct md_rdev * rdev, int size)
1da177e4
LT		 909{
910 char b[BDEVNAME_SIZE];
911 if (!rdev->sb_page) {
912 MD_BUG();
913 return -EINVAL;
914 }
915 if (rdev->sb_loaded)
916 return 0;
917
918
ccebd4c4 919 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
1da177e4
LT		 920 goto fail;
921 rdev->sb_loaded = 1;
922 return 0;
923
924fail:
925 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
926 bdevname(rdev->bdev,b));
927 return -EINVAL;
928}
929
930static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
931{
05710466
AN		 932 return sb1->set_uuid0 == sb2->set_uuid0 &&
933 sb1->set_uuid1 == sb2->set_uuid1 &&
934 sb1->set_uuid2 == sb2->set_uuid2 &&
935 sb1->set_uuid3 == sb2->set_uuid3;
1da177e4
LT		 936}
937
1da177e4
LT		 938static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
939{
940 int ret;
941 mdp_super_t *tmp1, *tmp2;
942
943 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
944 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
945
946 if (!tmp1 || !tmp2) {
947 ret = 0;
35020f1a 948 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
1da177e4
LT		 949 goto abort;
950 }
951
952 *tmp1 = *sb1;
953 *tmp2 = *sb2;
954
955 /*
956 * nr_disks is not constant
957 */
958 tmp1->nr_disks = 0;
959 tmp2->nr_disks = 0;
960
ce0c8e05 961 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1da177e4 962abort:
990a8baf
JJ		 963 kfree(tmp1);
964 kfree(tmp2);
1da177e4
LT		 965 return ret;
966}
967
4d167f09
N		 968
969static u32 md_csum_fold(u32 csum)
970{
971 csum = (csum & 0xffff) + (csum >> 16);
972 return (csum & 0xffff) + (csum >> 16);
973}
974
1da177e4
LT		 975static unsigned int calc_sb_csum(mdp_super_t * sb)
976{
4d167f09
N		 977 u64 newcsum = 0;
978 u32 *sb32 = (u32*)sb;
979 int i;
1da177e4
LT		 980 unsigned int disk_csum, csum;
981
982 disk_csum = sb->sb_csum;
983 sb->sb_csum = 0;
4d167f09
N		 984
985 for (i = 0; i < MD_SB_BYTES/4 ; i++)
986 newcsum += sb32[i];
987 csum = (newcsum & 0xffffffff) + (newcsum>>32);
988
989
990#ifdef CONFIG_ALPHA
991 /* This used to use csum_partial, which was wrong for several
992 * reasons including that different results are returned on
993 * different architectures. It isn't critical that we get exactly
994 * the same return value as before (we always csum_fold before
995 * testing, and that removes any differences). However as we
996 * know that csum_partial always returned a 16bit value on
997 * alphas, do a fold to maximise conformity to previous behaviour.
998 */
999 sb->sb_csum = md_csum_fold(disk_csum);
1000#else
1da177e4 1001 sb->sb_csum = disk_csum;
4d167f09 1002#endif
1da177e4
LT		 1003 return csum;
1004}
1005
1006
1007/*
1008 * Handle superblock details.
1009 * We want to be able to handle multiple superblock formats
1010 * so we have a common interface to them all, and an array of
1011 * different handlers.
1012 * We rely on user-space to write the initial superblock, and support
1013 * reading and updating of superblocks.
1014 * Interface methods are:
3cb03002 1015 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1016 * loads and validates a superblock on dev.
1017 * if refdev != NULL, compare superblocks on both devices
1018 * Return:
1019 * 0 - dev has a superblock that is compatible with refdev
1020 * 1 - dev has a superblock that is compatible and newer than refdev
1021 * so dev should be used as the refdev in future
1022 * -EINVAL superblock incompatible or invalid
1023 * -othererror e.g. -EIO
1024 *
fd01b88c 1025 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1026 * Verify that dev is acceptable into mddev.
1027 * The first time, mddev->raid_disks will be 0, and data from
1028 * dev should be merged in. Subsequent calls check that dev
1029 * is new enough. Return 0 or -EINVAL
1030 *
fd01b88c 1031 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1da177e4
LT		 1032 * Update the superblock for rdev with data in mddev
1033 * This does not write to disc.
1034 *
1035 */
1036
1037struct super_type {
0cd17fec
CW		 1038 char *name;
1039 struct module *owner;
3cb03002 1040 int (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
0cd17fec 1041 int minor_version);
fd01b88c
N		 1042 int (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1043 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
3cb03002 1044 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
15f4a5fd 1045 sector_t num_sectors);
1da177e4
LT		 1046};
1047
0894cc30
AN		 1048/*
1049 * Check that the given mddev has no bitmap.
1050 *
1051 * This function is called from the run method of all personalities that do not
1052 * support bitmaps. It prints an error message and returns non-zero if mddev
1053 * has a bitmap. Otherwise, it returns 0.
1054 *
1055 */
fd01b88c 1056int md_check_no_bitmap(struct mddev *mddev)
0894cc30 1057{
c3d9714e 1058 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
0894cc30
AN		 1059 return 0;
1060 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1061 mdname(mddev), mddev->pers->name);
1062 return 1;
1063}
1064EXPORT_SYMBOL(md_check_no_bitmap);
1065
1da177e4
LT		 1066/*
1067 * load_super for 0.90.0
1068 */
3cb03002 1069static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1070{
1071 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1072 mdp_super_t *sb;
1073 int ret;
1da177e4
LT		 1074
1075 /*
0f420358 1076 * Calculate the position of the superblock (512byte sectors),
1da177e4
LT		 1077 * it's at the end of the disk.
1078 *
1079 * It also happens to be a multiple of 4Kb.
1080 */
57b2caa3 1081 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 1082
0002b271 1083 ret = read_disk_sb(rdev, MD_SB_BYTES);
1da177e4
LT		 1084 if (ret) return ret;
1085
1086 ret = -EINVAL;
1087
1088 bdevname(rdev->bdev, b);
65a06f06 1089 sb = page_address(rdev->sb_page);
1da177e4
LT		 1090
1091 if (sb->md_magic != MD_SB_MAGIC) {
1092 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1093 b);
1094 goto abort;
1095 }
1096
1097 if (sb->major_version != 0 ||
f6705578
N		 1098 sb->minor_version < 90 ||
1099 sb->minor_version > 91) {
1da177e4
LT		 1100 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1101 sb->major_version, sb->minor_version,
1102 b);
1103 goto abort;
1104 }
1105
1106 if (sb->raid_disks <= 0)
1107 goto abort;
1108
4d167f09 1109 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1da177e4
LT		 1110 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1111 b);
1112 goto abort;
1113 }
1114
1115 rdev->preferred_minor = sb->md_minor;
1116 rdev->data_offset = 0;
0002b271 1117 rdev->sb_size = MD_SB_BYTES;
9f2f3830 1118 rdev->badblocks.shift = -1;
1da177e4
LT		 1119
1120 if (sb->level == LEVEL_MULTIPATH)
1121 rdev->desc_nr = -1;
1122 else
1123 rdev->desc_nr = sb->this_disk.number;
1124
9a7b2b0f 1125 if (!refdev) {
1da177e4 1126 ret = 1;
9a7b2b0f 1127 } else {
1da177e4 1128 __u64 ev1, ev2;
65a06f06 1129 mdp_super_t *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1130 if (!uuid_equal(refsb, sb)) {
1131 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1132 b, bdevname(refdev->bdev,b2));
1133 goto abort;
1134 }
1135 if (!sb_equal(refsb, sb)) {
1136 printk(KERN_WARNING "md: %s has same UUID"
1137 " but different superblock to %s\n",
1138 b, bdevname(refdev->bdev, b2));
1139 goto abort;
1140 }
1141 ev1 = md_event(sb);
1142 ev2 = md_event(refsb);
1143 if (ev1 > ev2)
1144 ret = 1;
1145 else
1146 ret = 0;
1147 }
8190e754 1148 rdev->sectors = rdev->sb_start;
27a7b260
N		 1149 /* Limit to 4TB as metadata cannot record more than that */
1150 if (rdev->sectors >= (2ULL << 32))
1151 rdev->sectors = (2ULL << 32) - 2;
1da177e4 1152
27a7b260 1153 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
2bf071bf
N		 1154 /* "this cannot possibly happen" ... */
1155 ret = -EINVAL;
1156
1da177e4
LT		 1157 abort:
1158 return ret;
1159}
1160
1161/*
1162 * validate_super for 0.90.0
1163 */
fd01b88c 1164static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1165{
1166 mdp_disk_t *desc;
65a06f06 1167 mdp_super_t *sb = page_address(rdev->sb_page);
07d84d10 1168 __u64 ev1 = md_event(sb);
1da177e4 1169
41158c7e 1170 rdev->raid_disk = -1;
c5d79adb
N		 1171 clear_bit(Faulty, &rdev->flags);
1172 clear_bit(In_sync, &rdev->flags);
1173 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1174
1da177e4
LT		 1175 if (mddev->raid_disks == 0) {
1176 mddev->major_version = 0;
1177 mddev->minor_version = sb->minor_version;
1178 mddev->patch_version = sb->patch_version;
e691063a 1179 mddev->external = 0;
9d8f0363 1180 mddev->chunk_sectors = sb->chunk_size >> 9;
1da177e4
LT		 1181 mddev->ctime = sb->ctime;
1182 mddev->utime = sb->utime;
1183 mddev->level = sb->level;
d9d166c2 1184 mddev->clevel[0] = 0;
1da177e4
LT		 1185 mddev->layout = sb->layout;
1186 mddev->raid_disks = sb->raid_disks;
27a7b260 1187 mddev->dev_sectors = ((sector_t)sb->size) * 2;
07d84d10 1188 mddev->events = ev1;
c3d9714e
N		 1189 mddev->bitmap_info.offset = 0;
1190 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1da177e4 1191
f6705578
N		 1192 if (mddev->minor_version >= 91) {
1193 mddev->reshape_position = sb->reshape_position;
1194 mddev->delta_disks = sb->delta_disks;
1195 mddev->new_level = sb->new_level;
1196 mddev->new_layout = sb->new_layout;
664e7c41 1197 mddev->new_chunk_sectors = sb->new_chunk >> 9;
f6705578
N		 1198 } else {
1199 mddev->reshape_position = MaxSector;
1200 mddev->delta_disks = 0;
1201 mddev->new_level = mddev->level;
1202 mddev->new_layout = mddev->layout;
664e7c41 1203 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1204 }
1205
1da177e4
LT		 1206 if (sb->state & (1<<MD_SB_CLEAN))
1207 mddev->recovery_cp = MaxSector;
1208 else {
1209 if (sb->events_hi == sb->cp_events_hi &&
1210 sb->events_lo == sb->cp_events_lo) {
1211 mddev->recovery_cp = sb->recovery_cp;
1212 } else
1213 mddev->recovery_cp = 0;
1214 }
1215
1216 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1217 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1218 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1219 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1220
1221 mddev->max_disks = MD_SB_DISKS;
a654b9d8
N		 1222
1223 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
c3d9714e
N		 1224 mddev->bitmap_info.file == NULL)
1225 mddev->bitmap_info.offset =
1226 mddev->bitmap_info.default_offset;
a654b9d8 1227
41158c7e 1228 } else if (mddev->pers == NULL) {
be6800a7
N		 1229 /* Insist on good event counter while assembling, except
1230 * for spares (which don't need an event count) */
1da177e4 1231 ++ev1;
be6800a7
N		 1232 if (sb->disks[rdev->desc_nr].state & (
1233 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1234 if (ev1 < mddev->events)
1235 return -EINVAL;
41158c7e
N		 1236 } else if (mddev->bitmap) {
1237 /* if adding to array with a bitmap, then we can accept an
1238 * older device ... but not too old.
1239 */
41158c7e
N		 1240 if (ev1 < mddev->bitmap->events_cleared)
1241 return 0;
07d84d10
N		 1242 } else {
1243 if (ev1 < mddev->events)
1244 /* just a hot-add of a new device, leave raid_disk at -1 */
1245 return 0;
1246 }
41158c7e 1247
1da177e4 1248 if (mddev->level != LEVEL_MULTIPATH) {
1da177e4
LT		 1249 desc = sb->disks + rdev->desc_nr;
1250
1251 if (desc->state & (1<<MD_DISK_FAULTY))
b2d444d7 1252 set_bit(Faulty, &rdev->flags);
7c7546cc
N		 1253 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1254 desc->raid_disk < mddev->raid_disks */) {
b2d444d7 1255 set_bit(In_sync, &rdev->flags);
1da177e4 1256 rdev->raid_disk = desc->raid_disk;
0261cd9f
N		 1257 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1258 /* active but not in sync implies recovery up to
1259 * reshape position. We don't know exactly where
1260 * that is, so set to zero for now */
1261 if (mddev->minor_version >= 91) {
1262 rdev->recovery_offset = 0;
1263 rdev->raid_disk = desc->raid_disk;
1264 }
1da177e4 1265 }
8ddf9efe
N		 1266 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1267 set_bit(WriteMostly, &rdev->flags);
41158c7e 1268 } else /* MULTIPATH are always insync */
b2d444d7 1269 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1270 return 0;
1271}
1272
1273/*
1274 * sync_super for 0.90.0
1275 */
fd01b88c 1276static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1277{
1278 mdp_super_t *sb;
3cb03002 1279 struct md_rdev *rdev2;
1da177e4 1280 int next_spare = mddev->raid_disks;
19133a42 1281
1da177e4
LT		 1282
1283 /* make rdev->sb match mddev data..
1284 *
1285 * 1/ zero out disks
1286 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1287 * 3/ any empty disks < next_spare become removed
1288 *
1289 * disks[0] gets initialised to REMOVED because
1290 * we cannot be sure from other fields if it has
1291 * been initialised or not.
1292 */
1293 int i;
1294 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1295
61181565
N		 1296 rdev->sb_size = MD_SB_BYTES;
1297
65a06f06 1298 sb = page_address(rdev->sb_page);
1da177e4
LT		 1299
1300 memset(sb, 0, sizeof(*sb));
1301
1302 sb->md_magic = MD_SB_MAGIC;
1303 sb->major_version = mddev->major_version;
1da177e4
LT		 1304 sb->patch_version = mddev->patch_version;
1305 sb->gvalid_words = 0; /* ignored */
1306 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1307 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1308 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1309 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1310
1311 sb->ctime = mddev->ctime;
1312 sb->level = mddev->level;
58c0fed4 1313 sb->size = mddev->dev_sectors / 2;
1da177e4
LT		 1314 sb->raid_disks = mddev->raid_disks;
1315 sb->md_minor = mddev->md_minor;
e691063a 1316 sb->not_persistent = 0;
1da177e4
LT		 1317 sb->utime = mddev->utime;
1318 sb->state = 0;
1319 sb->events_hi = (mddev->events>>32);
1320 sb->events_lo = (u32)mddev->events;
1321
f6705578
N		 1322 if (mddev->reshape_position == MaxSector)
1323 sb->minor_version = 90;
1324 else {
1325 sb->minor_version = 91;
1326 sb->reshape_position = mddev->reshape_position;
1327 sb->new_level = mddev->new_level;
1328 sb->delta_disks = mddev->delta_disks;
1329 sb->new_layout = mddev->new_layout;
664e7c41 1330 sb->new_chunk = mddev->new_chunk_sectors << 9;
f6705578
N		 1331 }
1332 mddev->minor_version = sb->minor_version;
1da177e4
LT		 1333 if (mddev->in_sync)
1334 {
1335 sb->recovery_cp = mddev->recovery_cp;
1336 sb->cp_events_hi = (mddev->events>>32);
1337 sb->cp_events_lo = (u32)mddev->events;
1338 if (mddev->recovery_cp == MaxSector)
1339 sb->state = (1<< MD_SB_CLEAN);
1340 } else
1341 sb->recovery_cp = 0;
1342
1343 sb->layout = mddev->layout;
9d8f0363 1344 sb->chunk_size = mddev->chunk_sectors << 9;
1da177e4 1345
c3d9714e 1346 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
a654b9d8
N		 1347 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1348
1da177e4 1349 sb->disks[0].state = (1<<MD_DISK_REMOVED);
159ec1fc 1350 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1da177e4 1351 mdp_disk_t *d;
86e6ffdd 1352 int desc_nr;
0261cd9f
N		 1353 int is_active = test_bit(In_sync, &rdev2->flags);
1354
1355 if (rdev2->raid_disk >= 0 &&
1356 sb->minor_version >= 91)
1357 /* we have nowhere to store the recovery_offset,
1358 * but if it is not below the reshape_position,
1359 * we can piggy-back on that.
1360 */
1361 is_active = 1;
1362 if (rdev2->raid_disk < 0 ||
1363 test_bit(Faulty, &rdev2->flags))
1364 is_active = 0;
1365 if (is_active)
86e6ffdd 1366 desc_nr = rdev2->raid_disk;
1da177e4 1367 else
86e6ffdd 1368 desc_nr = next_spare++;
19133a42 1369 rdev2->desc_nr = desc_nr;
1da177e4
LT		 1370 d = &sb->disks[rdev2->desc_nr];
1371 nr_disks++;
1372 d->number = rdev2->desc_nr;
1373 d->major = MAJOR(rdev2->bdev->bd_dev);
1374 d->minor = MINOR(rdev2->bdev->bd_dev);
0261cd9f 1375 if (is_active)
1da177e4
LT		 1376 d->raid_disk = rdev2->raid_disk;
1377 else
1378 d->raid_disk = rdev2->desc_nr; /* compatibility */
1be7892f 1379 if (test_bit(Faulty, &rdev2->flags))
1da177e4 1380 d->state = (1<<MD_DISK_FAULTY);
0261cd9f 1381 else if (is_active) {
1da177e4 1382 d->state = (1<<MD_DISK_ACTIVE);
0261cd9f
N		 1383 if (test_bit(In_sync, &rdev2->flags))
1384 d->state |= (1<<MD_DISK_SYNC);
1da177e4
LT		 1385 active++;
1386 working++;
1387 } else {
1388 d->state = 0;
1389 spare++;
1390 working++;
1391 }
8ddf9efe
N		 1392 if (test_bit(WriteMostly, &rdev2->flags))
1393 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4 1394 }
1da177e4
LT		 1395 /* now set the "removed" and "faulty" bits on any missing devices */
1396 for (i=0 ; i < mddev->raid_disks ; i++) {
1397 mdp_disk_t *d = &sb->disks[i];
1398 if (d->state == 0 && d->number == 0) {
1399 d->number = i;
1400 d->raid_disk = i;
1401 d->state = (1<<MD_DISK_REMOVED);
1402 d->state |= (1<<MD_DISK_FAULTY);
1403 failed++;
1404 }
1405 }
1406 sb->nr_disks = nr_disks;
1407 sb->active_disks = active;
1408 sb->working_disks = working;
1409 sb->failed_disks = failed;
1410 sb->spare_disks = spare;
1411
1412 sb->this_disk = sb->disks[rdev->desc_nr];
1413 sb->sb_csum = calc_sb_csum(sb);
1414}
1415
0cd17fec
CW		 1416/*
1417 * rdev_size_change for 0.90.0
1418 */
1419static unsigned long long
3cb03002 1420super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
0cd17fec 1421{
58c0fed4 1422 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
0cd17fec 1423 return 0; /* component must fit device */
c3d9714e 1424 if (rdev->mddev->bitmap_info.offset)
0cd17fec 1425 return 0; /* can't move bitmap */
57b2caa3 1426 rdev->sb_start = calc_dev_sboffset(rdev);
15f4a5fd
AN		 1427 if (!num_sectors || num_sectors > rdev->sb_start)
1428 num_sectors = rdev->sb_start;
27a7b260
N		 1429 /* Limit to 4TB as metadata cannot record more than that.
1430 * 4TB == 2^32 KB, or 2*2^32 sectors.
1431 */
1432 if (num_sectors >= (2ULL << 32))
1433 num_sectors = (2ULL << 32) - 2;
0f420358 1434 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
0cd17fec
CW		 1435 rdev->sb_page);
1436 md_super_wait(rdev->mddev);
c26a44ed 1437 return num_sectors;
0cd17fec
CW		 1438}
1439
1440
1da177e4
LT		 1441/*
1442 * version 1 superblock
1443 */
1444
1c05b4bc 1445static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1da177e4 1446{
1c05b4bc
N		 1447 __le32 disk_csum;
1448 u32 csum;
1da177e4
LT		 1449 unsigned long long newcsum;
1450 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1c05b4bc 1451 __le32 *isuper = (__le32*)sb;
1da177e4
LT		 1452 int i;
1453
1454 disk_csum = sb->sb_csum;
1455 sb->sb_csum = 0;
1456 newcsum = 0;
1457 for (i=0; size>=4; size -= 4 )
1458 newcsum += le32_to_cpu(*isuper++);
1459
1460 if (size == 2)
1c05b4bc 1461 newcsum += le16_to_cpu(*(__le16*) isuper);
1da177e4
LT		 1462
1463 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1464 sb->sb_csum = disk_csum;
1465 return cpu_to_le32(csum);
1466}
1467
2699b672
N		 1468static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1469 int acknowledged);
3cb03002 1470static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1da177e4
LT		 1471{
1472 struct mdp_superblock_1 *sb;
1473 int ret;
0f420358 1474 sector_t sb_start;
1da177e4 1475 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
0002b271 1476 int bmask;
1da177e4
LT		 1477
1478 /*
0f420358 1479 * Calculate the position of the superblock in 512byte sectors.
1da177e4
LT		 1480 * It is always aligned to a 4K boundary and
1481 * depeding on minor_version, it can be:
1482 * 0: At least 8K, but less than 12K, from end of device
1483 * 1: At start of device
1484 * 2: 4K from start of device.
1485 */
1486 switch(minor_version) {
1487 case 0:
77304d2a 1488 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
0f420358
AN		 1489 sb_start -= 8*2;
1490 sb_start &= ~(sector_t)(4*2-1);
1da177e4
LT		 1491 break;
1492 case 1:
0f420358 1493 sb_start = 0;
1da177e4
LT		 1494 break;
1495 case 2:
0f420358 1496 sb_start = 8;
1da177e4
LT		 1497 break;
1498 default:
1499 return -EINVAL;
1500 }
0f420358 1501 rdev->sb_start = sb_start;
1da177e4 1502
0002b271
N		 1503 /* superblock is rarely larger than 1K, but it can be larger,
1504 * and it is safe to read 4k, so we do that
1505 */
1506 ret = read_disk_sb(rdev, 4096);
1da177e4
LT		 1507 if (ret) return ret;
1508
1509
65a06f06 1510 sb = page_address(rdev->sb_page);
1da177e4
LT		 1511
1512 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1513 sb->major_version != cpu_to_le32(1) ||
1514 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
0f420358 1515 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
71c0805c 1516 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1da177e4
LT		 1517 return -EINVAL;
1518
1519 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1520 printk("md: invalid superblock checksum on %s\n",
1521 bdevname(rdev->bdev,b));
1522 return -EINVAL;
1523 }
1524 if (le64_to_cpu(sb->data_size) < 10) {
1525 printk("md: data_size too small on %s\n",
1526 bdevname(rdev->bdev,b));
1527 return -EINVAL;
1528 }
e11e93fa 1529
1da177e4
LT		 1530 rdev->preferred_minor = 0xffff;
1531 rdev->data_offset = le64_to_cpu(sb->data_offset);
4dbcdc75 1532 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1da177e4 1533
0002b271 1534 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
e1defc4f 1535 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
0002b271 1536 if (rdev->sb_size & bmask)
a1801f85
N		 1537 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1538
1539 if (minor_version
0f420358 1540 && rdev->data_offset < sb_start + (rdev->sb_size/512))
a1801f85 1541 return -EINVAL;
0002b271 1542
31b65a0d
N		 1543 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1544 rdev->desc_nr = -1;
1545 else
1546 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1547
2699b672
N		 1548 if (!rdev->bb_page) {
1549 rdev->bb_page = alloc_page(GFP_KERNEL);
1550 if (!rdev->bb_page)
1551 return -ENOMEM;
1552 }
1553 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1554 rdev->badblocks.count == 0) {
1555 /* need to load the bad block list.
1556 * Currently we limit it to one page.
1557 */
1558 s32 offset;
1559 sector_t bb_sector;
1560 u64 *bbp;
1561 int i;
1562 int sectors = le16_to_cpu(sb->bblog_size);
1563 if (sectors > (PAGE_SIZE / 512))
1564 return -EINVAL;
1565 offset = le32_to_cpu(sb->bblog_offset);
1566 if (offset == 0)
1567 return -EINVAL;
1568 bb_sector = (long long)offset;
1569 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1570 rdev->bb_page, READ, true))
1571 return -EIO;
1572 bbp = (u64 *)page_address(rdev->bb_page);
1573 rdev->badblocks.shift = sb->bblog_shift;
1574 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1575 u64 bb = le64_to_cpu(*bbp);
1576 int count = bb & (0x3ff);
1577 u64 sector = bb >> 10;
1578 sector <<= sb->bblog_shift;
1579 count <<= sb->bblog_shift;
1580 if (bb + 1 == 0)
1581 break;
1582 if (md_set_badblocks(&rdev->badblocks,
1583 sector, count, 1) == 0)
1584 return -EINVAL;
1585 }
1586 } else if (sb->bblog_offset == 0)
1587 rdev->badblocks.shift = -1;
1588
9a7b2b0f 1589 if (!refdev) {
8ed75463 1590 ret = 1;
9a7b2b0f 1591 } else {
1da177e4 1592 __u64 ev1, ev2;
65a06f06 1593 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1da177e4
LT		 1594
1595 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1596 sb->level != refsb->level ||
1597 sb->layout != refsb->layout ||
1598 sb->chunksize != refsb->chunksize) {
1599 printk(KERN_WARNING "md: %s has strangely different"
1600 " superblock to %s\n",
1601 bdevname(rdev->bdev,b),
1602 bdevname(refdev->bdev,b2));
1603 return -EINVAL;
1604 }
1605 ev1 = le64_to_cpu(sb->events);
1606 ev2 = le64_to_cpu(refsb->events);
1607
1608 if (ev1 > ev2)
8ed75463
N		 1609 ret = 1;
1610 else
1611 ret = 0;
1da177e4 1612 }
a1801f85 1613 if (minor_version)
77304d2a 1614 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 1615 le64_to_cpu(sb->data_offset);
1da177e4 1616 else
dd8ac336
AN		 1617 rdev->sectors = rdev->sb_start;
1618 if (rdev->sectors < le64_to_cpu(sb->data_size))
1da177e4 1619 return -EINVAL;
dd8ac336 1620 rdev->sectors = le64_to_cpu(sb->data_size);
dd8ac336 1621 if (le64_to_cpu(sb->size) > rdev->sectors)
2bf071bf 1622 return -EINVAL;
8ed75463 1623 return ret;
1da177e4
LT		 1624}
1625
fd01b88c 1626static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1da177e4 1627{
65a06f06 1628 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
07d84d10 1629 __u64 ev1 = le64_to_cpu(sb->events);
1da177e4 1630
41158c7e 1631 rdev->raid_disk = -1;
c5d79adb
N		 1632 clear_bit(Faulty, &rdev->flags);
1633 clear_bit(In_sync, &rdev->flags);
1634 clear_bit(WriteMostly, &rdev->flags);
c5d79adb 1635
1da177e4
LT		 1636 if (mddev->raid_disks == 0) {
1637 mddev->major_version = 1;
1638 mddev->patch_version = 0;
e691063a 1639 mddev->external = 0;
9d8f0363 1640 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1da177e4
LT		 1641 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1642 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1643 mddev->level = le32_to_cpu(sb->level);
d9d166c2 1644 mddev->clevel[0] = 0;
1da177e4
LT		 1645 mddev->layout = le32_to_cpu(sb->layout);
1646 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
58c0fed4 1647 mddev->dev_sectors = le64_to_cpu(sb->size);
07d84d10 1648 mddev->events = ev1;
c3d9714e
N		 1649 mddev->bitmap_info.offset = 0;
1650 mddev->bitmap_info.default_offset = 1024 >> 9;
1da177e4
LT		 1651
1652 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1653 memcpy(mddev->uuid, sb->set_uuid, 16);
1654
1655 mddev->max_disks = (4096-256)/2;
a654b9d8 1656
71c0805c 1657 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
c3d9714e
N		 1658 mddev->bitmap_info.file == NULL )
1659 mddev->bitmap_info.offset =
1660 (__s32)le32_to_cpu(sb->bitmap_offset);
e11e93fa 1661
f6705578
N		 1662 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1663 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1664 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1665 mddev->new_level = le32_to_cpu(sb->new_level);
1666 mddev->new_layout = le32_to_cpu(sb->new_layout);
664e7c41 1667 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
f6705578
N		 1668 } else {
1669 mddev->reshape_position = MaxSector;
1670 mddev->delta_disks = 0;
1671 mddev->new_level = mddev->level;
1672 mddev->new_layout = mddev->layout;
664e7c41 1673 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 1674 }
1675
41158c7e 1676 } else if (mddev->pers == NULL) {
be6800a7
N		 1677 /* Insist of good event counter while assembling, except for
1678 * spares (which don't need an event count) */
1da177e4 1679 ++ev1;
be6800a7
N		 1680 if (rdev->desc_nr >= 0 &&
1681 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1682 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1683 if (ev1 < mddev->events)
1684 return -EINVAL;
41158c7e
N		 1685 } else if (mddev->bitmap) {
1686 /* If adding to array with a bitmap, then we can accept an
1687 * older device, but not too old.
1688 */
41158c7e
N		 1689 if (ev1 < mddev->bitmap->events_cleared)
1690 return 0;
07d84d10
N		 1691 } else {
1692 if (ev1 < mddev->events)
1693 /* just a hot-add of a new device, leave raid_disk at -1 */
1694 return 0;
1695 }
1da177e4
LT		 1696 if (mddev->level != LEVEL_MULTIPATH) {
1697 int role;
3673f305
N		 1698 if (rdev->desc_nr < 0 ||
1699 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1700 role = 0xffff;
1701 rdev->desc_nr = -1;
1702 } else
1703 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1da177e4
LT		 1704 switch(role) {
1705 case 0xffff: /* spare */
1da177e4
LT		 1706 break;
1707 case 0xfffe: /* faulty */
b2d444d7 1708 set_bit(Faulty, &rdev->flags);
1da177e4
LT		 1709 break;
1710 default:
5fd6c1dc
N		 1711 if ((le32_to_cpu(sb->feature_map) &
1712 MD_FEATURE_RECOVERY_OFFSET))
1713 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1714 else
1715 set_bit(In_sync, &rdev->flags);
1da177e4
LT		 1716 rdev->raid_disk = role;
1717 break;
1718 }
8ddf9efe
N		 1719 if (sb->devflags & WriteMostly1)
1720 set_bit(WriteMostly, &rdev->flags);
41158c7e 1721 } else /* MULTIPATH are always insync */
b2d444d7 1722 set_bit(In_sync, &rdev->flags);
41158c7e 1723
1da177e4
LT		 1724 return 0;
1725}
1726
fd01b88c 1727static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 1728{
1729 struct mdp_superblock_1 *sb;
3cb03002 1730 struct md_rdev *rdev2;
1da177e4
LT		 1731 int max_dev, i;
1732 /* make rdev->sb match mddev and rdev data. */
1733
65a06f06 1734 sb = page_address(rdev->sb_page);
1da177e4
LT		 1735
1736 sb->feature_map = 0;
1737 sb->pad0 = 0;
5fd6c1dc 1738 sb->recovery_offset = cpu_to_le64(0);
1da177e4 1739 memset(sb->pad1, 0, sizeof(sb->pad1));
1da177e4
LT		 1740 memset(sb->pad3, 0, sizeof(sb->pad3));
1741
1742 sb->utime = cpu_to_le64((__u64)mddev->utime);
1743 sb->events = cpu_to_le64(mddev->events);
1744 if (mddev->in_sync)
1745 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1746 else
1747 sb->resync_offset = cpu_to_le64(0);
1748
1c05b4bc 1749 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
4dbcdc75 1750
f0ca340c 1751 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
58c0fed4 1752 sb->size = cpu_to_le64(mddev->dev_sectors);
9d8f0363 1753 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
62e1e389
N		 1754 sb->level = cpu_to_le32(mddev->level);
1755 sb->layout = cpu_to_le32(mddev->layout);
f0ca340c 1756
aeb9b211
N		 1757 if (test_bit(WriteMostly, &rdev->flags))
1758 sb->devflags |= WriteMostly1;
1759 else
1760 sb->devflags &= ~WriteMostly1;
1761
c3d9714e
N		 1762 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1763 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
71c0805c 1764 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
a654b9d8 1765 }
5fd6c1dc
N		 1766
1767 if (rdev->raid_disk >= 0 &&
97e4f42d 1768 !test_bit(In_sync, &rdev->flags)) {
93be75ff
N		 1769 sb->feature_map |=
1770 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1771 sb->recovery_offset =
1772 cpu_to_le64(rdev->recovery_offset);
5fd6c1dc
N		 1773 }
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;
159ec1fc 1819 list_for_each_entry(rdev2, &mddev->disks, same_set)
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
159ec1fc 1836 list_for_each_entry(rdev2, &mddev->disks, same_set) {
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 */
ac5e7113
AN		 1951 list_for_each_entry(rdev, &mddev->disks, same_set) {
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
d089c6af 2178 rdev_for_each(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));
159ec1fc 2310 list_for_each_entry(rdev, &mddev->disks, same_set)
1da177e4
LT		 2311 printk("<%s>", bdevname(rdev->bdev,b));
2312 printk("\n");
2313
159ec1fc 2314 list_for_each_entry(rdev, &mddev->disks, same_set)
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;
159ec1fc 2331 list_for_each_entry(rdev, &mddev->disks, same_set) {
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
N		 2353 /* First make sure individual recovery_offsets are correct */
2354 list_for_each_entry(rdev, &mddev->disks, same_set) {
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);
de393cde
N		 2367 list_for_each_entry(rdev, &mddev->disks, same_set) {
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
de393cde 2433 list_for_each_entry(rdev, &mddev->disks, same_set) {
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);
159ec1fc 2447 list_for_each_entry(rdev, &mddev->disks, same_set) {
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
de393cde
N		 2496 list_for_each_entry(rdev, &mddev->disks, same_set) {
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
N		 2552 if (test_bit(Blocked, &rdev->flags) ||
2553 rdev->badblocks.unacked_exist) {
6bfe0b49
DW		 2554 len += sprintf(page+len, "%sblocked", sep);
2555 sep = ",";
2556 }
b2d444d7
N		 2557 if (!test_bit(Faulty, &rdev->flags) &&
2558 !test_bit(In_sync, &rdev->flags)) {
86e6ffdd
N		 2559 len += sprintf(page+len, "%sspare", sep);
2560 sep = ",";
2561 }
d7a9d443
N		 2562 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2563 len += sprintf(page+len, "%swrite_error", sep);
2564 sep = ",";
2565 }
86e6ffdd
N		 2566 return len+sprintf(page+len, "\n");
2567}
2568
45dc2de1 2569static ssize_t
3cb03002 2570state_store(struct md_rdev *rdev, const char *buf, size_t len)
45dc2de1
N		 2571{
2572 /* can write
de393cde 2573 * faulty - simulates an error
45dc2de1 2574 * remove - disconnects the device
f655675b
N		 2575 * writemostly - sets write_mostly
2576 * -writemostly - clears write_mostly
de393cde
N		 2577 * blocked - sets the Blocked flags
2578 * -blocked - clears the Blocked and possibly simulates an error
6d56e278 2579 * insync - sets Insync providing device isn't active
d7a9d443
N		 2580 * write_error - sets WriteErrorSeen
2581 * -write_error - clears WriteErrorSeen
45dc2de1
N		 2582 */
2583 int err = -EINVAL;
2584 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2585 md_error(rdev->mddev, rdev);
5ef56c8f
N		 2586 if (test_bit(Faulty, &rdev->flags))
2587 err = 0;
2588 else
2589 err = -EBUSY;
45dc2de1
N		 2590 } else if (cmd_match(buf, "remove")) {
2591 if (rdev->raid_disk >= 0)
2592 err = -EBUSY;
2593 else {
fd01b88c 2594 struct mddev *mddev = rdev->mddev;
45dc2de1 2595 kick_rdev_from_array(rdev);
3f9d7b0d
N		 2596 if (mddev->pers)
2597 md_update_sb(mddev, 1);
45dc2de1
N		 2598 md_new_event(mddev);
2599 err = 0;
2600 }
f655675b
N		 2601 } else if (cmd_match(buf, "writemostly")) {
2602 set_bit(WriteMostly, &rdev->flags);
2603 err = 0;
2604 } else if (cmd_match(buf, "-writemostly")) {
2605 clear_bit(WriteMostly, &rdev->flags);
6bfe0b49
DW		 2606 err = 0;
2607 } else if (cmd_match(buf, "blocked")) {
2608 set_bit(Blocked, &rdev->flags);
2609 err = 0;
2610 } else if (cmd_match(buf, "-blocked")) {
de393cde 2611 if (!test_bit(Faulty, &rdev->flags) &&
7da64a0a 2612 rdev->badblocks.unacked_exist) {
de393cde
N		 2613 /* metadata handler doesn't understand badblocks,
2614 * so we need to fail the device
2615 */
2616 md_error(rdev->mddev, rdev);
2617 }
6bfe0b49 2618 clear_bit(Blocked, &rdev->flags);
de393cde 2619 clear_bit(BlockedBadBlocks, &rdev->flags);
6bfe0b49
DW		 2620 wake_up(&rdev->blocked_wait);
2621 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2622 md_wakeup_thread(rdev->mddev->thread);
2623
6d56e278
N		 2624 err = 0;
2625 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2626 set_bit(In_sync, &rdev->flags);
f655675b 2627 err = 0;
d7a9d443
N		 2628 } else if (cmd_match(buf, "write_error")) {
2629 set_bit(WriteErrorSeen, &rdev->flags);
2630 err = 0;
2631 } else if (cmd_match(buf, "-write_error")) {
2632 clear_bit(WriteErrorSeen, &rdev->flags);
2633 err = 0;
45dc2de1 2634 }
00bcb4ac
N		 2635 if (!err)
2636 sysfs_notify_dirent_safe(rdev->sysfs_state);
45dc2de1
N		 2637 return err ? err : len;
2638}
80ca3a44
N		 2639static struct rdev_sysfs_entry rdev_state =
2640__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
86e6ffdd 2641
4dbcdc75 2642static ssize_t
3cb03002 2643errors_show(struct md_rdev *rdev, char *page)
4dbcdc75
N		 2644{
2645 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2646}
2647
2648static ssize_t
3cb03002 2649errors_store(struct md_rdev *rdev, const char *buf, size_t len)
4dbcdc75
N		 2650{
2651 char *e;
2652 unsigned long n = simple_strtoul(buf, &e, 10);
2653 if (*buf && (*e == 0 || *e == '\n')) {
2654 atomic_set(&rdev->corrected_errors, n);
2655 return len;
2656 }
2657 return -EINVAL;
2658}
2659static struct rdev_sysfs_entry rdev_errors =
80ca3a44 2660__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
4dbcdc75 2661
014236d2 2662static ssize_t
3cb03002 2663slot_show(struct md_rdev *rdev, char *page)
014236d2
N		 2664{
2665 if (rdev->raid_disk < 0)
2666 return sprintf(page, "none\n");
2667 else
2668 return sprintf(page, "%d\n", rdev->raid_disk);
2669}
2670
2671static ssize_t
3cb03002 2672slot_store(struct md_rdev *rdev, const char *buf, size_t len)
014236d2
N		 2673{
2674 char *e;
c303da6d 2675 int err;
014236d2
N		 2676 int slot = simple_strtoul(buf, &e, 10);
2677 if (strncmp(buf, "none", 4)==0)
2678 slot = -1;
2679 else if (e==buf || (*e && *e!= '\n'))
2680 return -EINVAL;
6c2fce2e 2681 if (rdev->mddev->pers && slot == -1) {
c303da6d
N		 2682 /* Setting 'slot' on an active array requires also
2683 * updating the 'rd%d' link, and communicating
2684 * with the personality with ->hot_*_disk.
2685 * For now we only support removing
2686 * failed/spare devices. This normally happens automatically,
2687 * but not when the metadata is externally managed.
2688 */
c303da6d
N		 2689 if (rdev->raid_disk == -1)
2690 return -EEXIST;
2691 /* personality does all needed checks */
01393f3d 2692 if (rdev->mddev->pers->hot_remove_disk == NULL)
c303da6d
N		 2693 return -EINVAL;
2694 err = rdev->mddev->pers->
2695 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2696 if (err)
2697 return err;
36fad858 2698 sysfs_unlink_rdev(rdev->mddev, rdev);
b7103107 2699 rdev->raid_disk = -1;
c303da6d
N		 2700 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2701 md_wakeup_thread(rdev->mddev->thread);
6c2fce2e 2702 } else if (rdev->mddev->pers) {
3cb03002 2703 struct md_rdev *rdev2;
6c2fce2e 2704 /* Activating a spare .. or possibly reactivating
6d56e278 2705 * if we ever get bitmaps working here.
6c2fce2e
NB		 2706 */
2707
2708 if (rdev->raid_disk != -1)
2709 return -EBUSY;
2710
c6751b2b
N		 2711 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2712 return -EBUSY;
2713
6c2fce2e
NB		 2714 if (rdev->mddev->pers->hot_add_disk == NULL)
2715 return -EINVAL;
2716
159ec1fc 2717 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
6c2fce2e
NB		 2718 if (rdev2->raid_disk == slot)
2719 return -EEXIST;
2720
ba1b41b6
N		 2721 if (slot >= rdev->mddev->raid_disks &&
2722 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2723 return -ENOSPC;
2724
6c2fce2e
NB		 2725 rdev->raid_disk = slot;
2726 if (test_bit(In_sync, &rdev->flags))
2727 rdev->saved_raid_disk = slot;
2728 else
2729 rdev->saved_raid_disk = -1;
d30519fc 2730 clear_bit(In_sync, &rdev->flags);
6c2fce2e
NB		 2731 err = rdev->mddev->pers->
2732 hot_add_disk(rdev->mddev, rdev);
199050ea 2733 if (err) {
6c2fce2e 2734 rdev->raid_disk = -1;
6c2fce2e 2735 return err;
52664732 2736 } else
00bcb4ac 2737 sysfs_notify_dirent_safe(rdev->sysfs_state);
36fad858 2738 if (sysfs_link_rdev(rdev->mddev, rdev))
00bcb4ac 2739 /* failure here is OK */;
6c2fce2e 2740 /* don't wakeup anyone, leave that to userspace. */
c303da6d 2741 } else {
ba1b41b6
N		 2742 if (slot >= rdev->mddev->raid_disks &&
2743 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
c303da6d
N		 2744 return -ENOSPC;
2745 rdev->raid_disk = slot;
2746 /* assume it is working */
c5d79adb
N		 2747 clear_bit(Faulty, &rdev->flags);
2748 clear_bit(WriteMostly, &rdev->flags);
c303da6d 2749 set_bit(In_sync, &rdev->flags);
00bcb4ac 2750 sysfs_notify_dirent_safe(rdev->sysfs_state);
c303da6d 2751 }
014236d2
N		 2752 return len;
2753}
2754
2755
2756static struct rdev_sysfs_entry rdev_slot =
80ca3a44 2757__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
014236d2 2758
93c8cad0 2759static ssize_t
3cb03002 2760offset_show(struct md_rdev *rdev, char *page)
93c8cad0 2761{
6961ece4 2762 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
93c8cad0
N		 2763}
2764
2765static ssize_t
3cb03002 2766offset_store(struct md_rdev *rdev, const char *buf, size_t len)
93c8cad0
N		 2767{
2768 char *e;
2769 unsigned long long offset = simple_strtoull(buf, &e, 10);
2770 if (e==buf || (*e && *e != '\n'))
2771 return -EINVAL;
8ed0a521 2772 if (rdev->mddev->pers && rdev->raid_disk >= 0)
93c8cad0 2773 return -EBUSY;
dd8ac336 2774 if (rdev->sectors && rdev->mddev->external)
c5d79adb
N		 2775 /* Must set offset before size, so overlap checks
2776 * can be sane */
2777 return -EBUSY;
93c8cad0
N		 2778 rdev->data_offset = offset;
2779 return len;
2780}
2781
2782static struct rdev_sysfs_entry rdev_offset =
80ca3a44 2783__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
93c8cad0 2784
83303b61 2785static ssize_t
3cb03002 2786rdev_size_show(struct md_rdev *rdev, char *page)
83303b61 2787{
dd8ac336 2788 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
83303b61
N		 2789}
2790
c5d79adb
N		 2791static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2792{
2793 /* check if two start/length pairs overlap */
2794 if (s1+l1 <= s2)
2795 return 0;
2796 if (s2+l2 <= s1)
2797 return 0;
2798 return 1;
2799}
2800
b522adcd
DW		 2801static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2802{
2803 unsigned long long blocks;
2804 sector_t new;
2805
2806 if (strict_strtoull(buf, 10, &blocks) < 0)
2807 return -EINVAL;
2808
2809 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2810 return -EINVAL; /* sector conversion overflow */
2811
2812 new = blocks * 2;
2813 if (new != blocks * 2)
2814 return -EINVAL; /* unsigned long long to sector_t overflow */
2815
2816 *sectors = new;
2817 return 0;
2818}
2819
83303b61 2820static ssize_t
3cb03002 2821rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
83303b61 2822{
fd01b88c 2823 struct mddev *my_mddev = rdev->mddev;
dd8ac336 2824 sector_t oldsectors = rdev->sectors;
b522adcd 2825 sector_t sectors;
27c529bb 2826
b522adcd 2827 if (strict_blocks_to_sectors(buf, &sectors) < 0)
d7027458 2828 return -EINVAL;
0cd17fec 2829 if (my_mddev->pers && rdev->raid_disk >= 0) {
d7027458 2830 if (my_mddev->persistent) {
dd8ac336
AN		 2831 sectors = super_types[my_mddev->major_version].
2832 rdev_size_change(rdev, sectors);
2833 if (!sectors)
0cd17fec 2834 return -EBUSY;
dd8ac336 2835 } else if (!sectors)
77304d2a 2836 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
dd8ac336 2837 rdev->data_offset;
0cd17fec 2838 }
dd8ac336 2839 if (sectors < my_mddev->dev_sectors)
7d3c6f87 2840 return -EINVAL; /* component must fit device */
0cd17fec 2841
dd8ac336
AN		 2842 rdev->sectors = sectors;
2843 if (sectors > oldsectors && my_mddev->external) {
c5d79adb
N		 2844 /* need to check that all other rdevs with the same ->bdev
2845 * do not overlap. We need to unlock the mddev to avoid
dd8ac336 2846 * a deadlock. We have already changed rdev->sectors, and if
c5d79adb
N		 2847 * we have to change it back, we will have the lock again.
2848 */
fd01b88c 2849 struct mddev *mddev;
c5d79adb 2850 int overlap = 0;
159ec1fc 2851 struct list_head *tmp;
c5d79adb 2852
27c529bb 2853 mddev_unlock(my_mddev);
29ac4aa3 2854 for_each_mddev(mddev, tmp) {
3cb03002 2855 struct md_rdev *rdev2;
c5d79adb
N		 2856
2857 mddev_lock(mddev);
159ec1fc 2858 list_for_each_entry(rdev2, &mddev->disks, same_set)
f21e9ff7
N		 2859 if (rdev->bdev == rdev2->bdev &&
2860 rdev != rdev2 &&
2861 overlaps(rdev->data_offset, rdev->sectors,
2862 rdev2->data_offset,
2863 rdev2->sectors)) {
c5d79adb
N		 2864 overlap = 1;
2865 break;
2866 }
2867 mddev_unlock(mddev);
2868 if (overlap) {
2869 mddev_put(mddev);
2870 break;
2871 }
2872 }
27c529bb 2873 mddev_lock(my_mddev);
c5d79adb
N		 2874 if (overlap) {
2875 /* Someone else could have slipped in a size
2876 * change here, but doing so is just silly.
dd8ac336 2877 * We put oldsectors back because we *know* it is
c5d79adb
N		 2878 * safe, and trust userspace not to race with
2879 * itself
2880 */
dd8ac336 2881 rdev->sectors = oldsectors;
c5d79adb
N		 2882 return -EBUSY;
2883 }
2884 }
83303b61
N		 2885 return len;
2886}
2887
2888static struct rdev_sysfs_entry rdev_size =
80ca3a44 2889__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
83303b61 2890
06e3c817 2891
3cb03002 2892static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
06e3c817
DW		 2893{
2894 unsigned long long recovery_start = rdev->recovery_offset;
2895
2896 if (test_bit(In_sync, &rdev->flags) ||
2897 recovery_start == MaxSector)
2898 return sprintf(page, "none\n");
2899
2900 return sprintf(page, "%llu\n", recovery_start);
2901}
2902
3cb03002 2903static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
06e3c817
DW		 2904{
2905 unsigned long long recovery_start;
2906
2907 if (cmd_match(buf, "none"))
2908 recovery_start = MaxSector;
2909 else if (strict_strtoull(buf, 10, &recovery_start))
2910 return -EINVAL;
2911
2912 if (rdev->mddev->pers &&
2913 rdev->raid_disk >= 0)
2914 return -EBUSY;
2915
2916 rdev->recovery_offset = recovery_start;
2917 if (recovery_start == MaxSector)
2918 set_bit(In_sync, &rdev->flags);
2919 else
2920 clear_bit(In_sync, &rdev->flags);
2921 return len;
2922}
2923
2924static struct rdev_sysfs_entry rdev_recovery_start =
2925__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2926
16c791a5
N		 2927
2928static ssize_t
2929badblocks_show(struct badblocks *bb, char *page, int unack);
2930static ssize_t
2931badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2932
3cb03002 2933static ssize_t bb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2934{
2935 return badblocks_show(&rdev->badblocks, page, 0);
2936}
3cb03002 2937static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5 2938{
de393cde
N		 2939 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2940 /* Maybe that ack was all we needed */
2941 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2942 wake_up(&rdev->blocked_wait);
2943 return rv;
16c791a5
N		 2944}
2945static struct rdev_sysfs_entry rdev_bad_blocks =
2946__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2947
2948
3cb03002 2949static ssize_t ubb_show(struct md_rdev *rdev, char *page)
16c791a5
N		 2950{
2951 return badblocks_show(&rdev->badblocks, page, 1);
2952}
3cb03002 2953static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
16c791a5
N		 2954{
2955 return badblocks_store(&rdev->badblocks, page, len, 1);
2956}
2957static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2958__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2959
86e6ffdd
N		 2960static struct attribute *rdev_default_attrs[] = {
2961 &rdev_state.attr,
4dbcdc75 2962 &rdev_errors.attr,
014236d2 2963 &rdev_slot.attr,
93c8cad0 2964 &rdev_offset.attr,
83303b61 2965 &rdev_size.attr,
06e3c817 2966 &rdev_recovery_start.attr,
16c791a5
N		 2967 &rdev_bad_blocks.attr,
2968 &rdev_unack_bad_blocks.attr,
86e6ffdd
N		 2969 NULL,
2970};
2971static ssize_t
2972rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2973{
2974 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2975 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
fd01b88c 2976 struct mddev *mddev = rdev->mddev;
27c529bb 2977 ssize_t rv;
86e6ffdd
N		 2978
2979 if (!entry->show)
2980 return -EIO;
27c529bb
N		 2981
2982 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2983 if (!rv) {
2984 if (rdev->mddev == NULL)
2985 rv = -EBUSY;
2986 else
2987 rv = entry->show(rdev, page);
2988 mddev_unlock(mddev);
2989 }
2990 return rv;
86e6ffdd
N		 2991}
2992
2993static ssize_t
2994rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2995 const char *page, size_t length)
2996{
2997 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3cb03002 2998 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
27c529bb 2999 ssize_t rv;
fd01b88c 3000 struct mddev *mddev = rdev->mddev;
86e6ffdd
N		 3001
3002 if (!entry->store)
3003 return -EIO;
67463acb
N		 3004 if (!capable(CAP_SYS_ADMIN))
3005 return -EACCES;
27c529bb 3006 rv = mddev ? mddev_lock(mddev): -EBUSY;
ca388059 3007 if (!rv) {
27c529bb
N		 3008 if (rdev->mddev == NULL)
3009 rv = -EBUSY;
3010 else
3011 rv = entry->store(rdev, page, length);
6a51830e 3012 mddev_unlock(mddev);
ca388059
N		 3013 }
3014 return rv;
86e6ffdd
N		 3015}
3016
3017static void rdev_free(struct kobject *ko)
3018{
3cb03002 3019 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
86e6ffdd
N		 3020 kfree(rdev);
3021}
52cf25d0 3022static const struct sysfs_ops rdev_sysfs_ops = {
86e6ffdd
N		 3023 .show = rdev_attr_show,
3024 .store = rdev_attr_store,
3025};
3026static struct kobj_type rdev_ktype = {
3027 .release = rdev_free,
3028 .sysfs_ops = &rdev_sysfs_ops,
3029 .default_attrs = rdev_default_attrs,
3030};
3031
3cb03002 3032int md_rdev_init(struct md_rdev *rdev)
e8bb9a83
N		 3033{
3034 rdev->desc_nr = -1;
3035 rdev->saved_raid_disk = -1;
3036 rdev->raid_disk = -1;
3037 rdev->flags = 0;
3038 rdev->data_offset = 0;
3039 rdev->sb_events = 0;
3040 rdev->last_read_error.tv_sec = 0;
3041 rdev->last_read_error.tv_nsec = 0;
2699b672
N		 3042 rdev->sb_loaded = 0;
3043 rdev->bb_page = NULL;
e8bb9a83
N		 3044 atomic_set(&rdev->nr_pending, 0);
3045 atomic_set(&rdev->read_errors, 0);
3046 atomic_set(&rdev->corrected_errors, 0);
3047
3048 INIT_LIST_HEAD(&rdev->same_set);
3049 init_waitqueue_head(&rdev->blocked_wait);
2230dfe4
N		 3050
3051 /* Add space to store bad block list.
3052 * This reserves the space even on arrays where it cannot
3053 * be used - I wonder if that matters
3054 */
3055 rdev->badblocks.count = 0;
3056 rdev->badblocks.shift = 0;
3057 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3058 seqlock_init(&rdev->badblocks.lock);
3059 if (rdev->badblocks.page == NULL)
3060 return -ENOMEM;
3061
3062 return 0;
e8bb9a83
N		 3063}
3064EXPORT_SYMBOL_GPL(md_rdev_init);
1da177e4
LT		 3065/*
3066 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3067 *
3068 * mark the device faulty if:
3069 *
3070 * - the device is nonexistent (zero size)
3071 * - the device has no valid superblock
3072 *
3073 * a faulty rdev _never_ has rdev->sb set.
3074 */
3cb03002 3075static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
1da177e4
LT		 3076{
3077 char b[BDEVNAME_SIZE];
3078 int err;
3cb03002 3079 struct md_rdev *rdev;
1da177e4
LT		 3080 sector_t size;
3081
9ffae0cf 3082 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1da177e4
LT		 3083 if (!rdev) {
3084 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3085 return ERR_PTR(-ENOMEM);
3086 }
1da177e4 3087
2230dfe4
N		 3088 err = md_rdev_init(rdev);
3089 if (err)
3090 goto abort_free;
3091 err = alloc_disk_sb(rdev);
3092 if (err)
1da177e4
LT		 3093 goto abort_free;
3094
c5d79adb 3095 err = lock_rdev(rdev, newdev, super_format == -2);
1da177e4
LT		 3096 if (err)
3097 goto abort_free;
3098
f9cb074b 3099 kobject_init(&rdev->kobj, &rdev_ktype);
86e6ffdd 3100
77304d2a 3101 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
1da177e4
LT		 3102 if (!size) {
3103 printk(KERN_WARNING
3104 "md: %s has zero or unknown size, marking faulty!\n",
3105 bdevname(rdev->bdev,b));
3106 err = -EINVAL;
3107 goto abort_free;
3108 }
3109
3110 if (super_format >= 0) {
3111 err = super_types[super_format].
3112 load_super(rdev, NULL, super_minor);
3113 if (err == -EINVAL) {
df968c4e
N		 3114 printk(KERN_WARNING
3115 "md: %s does not have a valid v%d.%d "
3116 "superblock, not importing!\n",
3117 bdevname(rdev->bdev,b),
3118 super_format, super_minor);
1da177e4
LT		 3119 goto abort_free;
3120 }
3121 if (err < 0) {
3122 printk(KERN_WARNING
3123 "md: could not read %s's sb, not importing!\n",
3124 bdevname(rdev->bdev,b));
3125 goto abort_free;
3126 }
3127 }
9f2f3830
N		 3128 if (super_format == -1)
3129 /* hot-add for 0.90, or non-persistent: so no badblocks */
3130 rdev->badblocks.shift = -1;
6bfe0b49 3131
1da177e4
LT		 3132 return rdev;
3133
3134abort_free:
2699b672
N		 3135 if (rdev->bdev)
3136 unlock_rdev(rdev);
3137 free_disk_sb(rdev);
2230dfe4 3138 kfree(rdev->badblocks.page);
1da177e4
LT		 3139 kfree(rdev);
3140 return ERR_PTR(err);
3141}
3142
3143/*
3144 * Check a full RAID array for plausibility
3145 */
3146
3147
fd01b88c 3148static void analyze_sbs(struct mddev * mddev)
1da177e4
LT		 3149{
3150 int i;
3cb03002 3151 struct md_rdev *rdev, *freshest, *tmp;
1da177e4
LT		 3152 char b[BDEVNAME_SIZE];
3153
3154 freshest = NULL;
d089c6af 3155 rdev_for_each(rdev, tmp, mddev)
1da177e4
LT		 3156 switch (super_types[mddev->major_version].
3157 load_super(rdev, freshest, mddev->minor_version)) {
3158 case 1:
3159 freshest = rdev;
3160 break;
3161 case 0:
3162 break;
3163 default:
3164 printk( KERN_ERR \
3165 "md: fatal superblock inconsistency in %s"
3166 " -- removing from array\n",
3167 bdevname(rdev->bdev,b));
3168 kick_rdev_from_array(rdev);
3169 }
3170
3171
3172 super_types[mddev->major_version].
3173 validate_super(mddev, freshest);
3174
3175 i = 0;
d089c6af 3176 rdev_for_each(rdev, tmp, mddev) {
233fca36
N		 3177 if (mddev->max_disks &&
3178 (rdev->desc_nr >= mddev->max_disks ||
3179 i > mddev->max_disks)) {
de01dfad
N		 3180 printk(KERN_WARNING
3181 "md: %s: %s: only %d devices permitted\n",
3182 mdname(mddev), bdevname(rdev->bdev, b),
3183 mddev->max_disks);
3184 kick_rdev_from_array(rdev);
3185 continue;
3186 }
1da177e4
LT		 3187 if (rdev != freshest)
3188 if (super_types[mddev->major_version].
3189 validate_super(mddev, rdev)) {
3190 printk(KERN_WARNING "md: kicking non-fresh %s"
3191 " from array!\n",
3192 bdevname(rdev->bdev,b));
3193 kick_rdev_from_array(rdev);
3194 continue;
3195 }
3196 if (mddev->level == LEVEL_MULTIPATH) {
3197 rdev->desc_nr = i++;
3198 rdev->raid_disk = rdev->desc_nr;
b2d444d7 3199 set_bit(In_sync, &rdev->flags);
5e5e3e78 3200 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
a778b73f
N		 3201 rdev->raid_disk = -1;
3202 clear_bit(In_sync, &rdev->flags);
1da177e4
LT		 3203 }
3204 }
1da177e4
LT		 3205}
3206
72e02075
N		 3207/* Read a fixed-point number.
3208 * Numbers in sysfs attributes should be in "standard" units where
3209 * possible, so time should be in seconds.
3210 * However we internally use a a much smaller unit such as
3211 * milliseconds or jiffies.
3212 * This function takes a decimal number with a possible fractional
3213 * component, and produces an integer which is the result of
3214 * multiplying that number by 10^'scale'.
3215 * all without any floating-point arithmetic.
3216 */
3217int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3218{
3219 unsigned long result = 0;
3220 long decimals = -1;
3221 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3222 if (*cp == '.')
3223 decimals = 0;
3224 else if (decimals < scale) {
3225 unsigned int value;
3226 value = *cp - '0';
3227 result = result * 10 + value;
3228 if (decimals >= 0)
3229 decimals++;
3230 }
3231 cp++;
3232 }
3233 if (*cp == '\n')
3234 cp++;
3235 if (*cp)
3236 return -EINVAL;
3237 if (decimals < 0)
3238 decimals = 0;
3239 while (decimals < scale) {
3240 result *= 10;
3241 decimals ++;
3242 }
3243 *res = result;
3244 return 0;
3245}
3246
3247
19052c0e
N		 3248static void md_safemode_timeout(unsigned long data);
3249
16f17b39 3250static ssize_t
fd01b88c 3251safe_delay_show(struct mddev *mddev, char *page)
16f17b39
N		 3252{
3253 int msec = (mddev->safemode_delay*1000)/HZ;
3254 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3255}
3256static ssize_t
fd01b88c 3257safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
16f17b39 3258{
16f17b39 3259 unsigned long msec;
97ce0a7f 3260
72e02075 3261 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
16f17b39 3262 return -EINVAL;
16f17b39
N		 3263 if (msec == 0)
3264 mddev->safemode_delay = 0;
3265 else {
19052c0e 3266 unsigned long old_delay = mddev->safemode_delay;
16f17b39
N		 3267 mddev->safemode_delay = (msec*HZ)/1000;
3268 if (mddev->safemode_delay == 0)
3269 mddev->safemode_delay = 1;
19052c0e
N		 3270 if (mddev->safemode_delay < old_delay)
3271 md_safemode_timeout((unsigned long)mddev);
16f17b39
N		 3272 }
3273 return len;
3274}
3275static struct md_sysfs_entry md_safe_delay =
80ca3a44 3276__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
16f17b39 3277
eae1701f 3278static ssize_t
fd01b88c 3279level_show(struct mddev *mddev, char *page)
eae1701f 3280{
84fc4b56 3281 struct md_personality *p = mddev->pers;
d9d166c2 3282 if (p)
eae1701f 3283 return sprintf(page, "%s\n", p->name);
d9d166c2
N		 3284 else if (mddev->clevel[0])
3285 return sprintf(page, "%s\n", mddev->clevel);
3286 else if (mddev->level != LEVEL_NONE)
3287 return sprintf(page, "%d\n", mddev->level);
3288 else
3289 return 0;
eae1701f
N		 3290}
3291
d9d166c2 3292static ssize_t
fd01b88c 3293level_store(struct mddev *mddev, const char *buf, size_t len)
d9d166c2 3294{
f2859af6 3295 char clevel[16];
20a49ff6 3296 ssize_t rv = len;
84fc4b56 3297 struct md_personality *pers;
f2859af6 3298 long level;
245f46c2 3299 void *priv;
3cb03002 3300 struct md_rdev *rdev;
245f46c2
N		 3301
3302 if (mddev->pers == NULL) {
3303 if (len == 0)
3304 return 0;
3305 if (len >= sizeof(mddev->clevel))
3306 return -ENOSPC;
3307 strncpy(mddev->clevel, buf, len);
3308 if (mddev->clevel[len-1] == '\n')
3309 len--;
3310 mddev->clevel[len] = 0;
3311 mddev->level = LEVEL_NONE;
3312 return rv;
3313 }
3314
3315 /* request to change the personality. Need to ensure:
3316 * - array is not engaged in resync/recovery/reshape
3317 * - old personality can be suspended
3318 * - new personality will access other array.
3319 */
3320
bb4f1e9d
N		 3321 if (mddev->sync_thread ||
3322 mddev->reshape_position != MaxSector ||
3323 mddev->sysfs_active)
d9d166c2 3324 return -EBUSY;
245f46c2
N		 3325
3326 if (!mddev->pers->quiesce) {
3327 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3328 mdname(mddev), mddev->pers->name);
3329 return -EINVAL;
3330 }
3331
3332 /* Now find the new personality */
f2859af6 3333 if (len == 0 || len >= sizeof(clevel))
245f46c2 3334 return -EINVAL;
f2859af6
DW		 3335 strncpy(clevel, buf, len);
3336 if (clevel[len-1] == '\n')
d9d166c2 3337 len--;
f2859af6
DW		 3338 clevel[len] = 0;
3339 if (strict_strtol(clevel, 10, &level))
3340 level = LEVEL_NONE;
245f46c2 3341
f2859af6
DW		 3342 if (request_module("md-%s", clevel) != 0)
3343 request_module("md-level-%s", clevel);
245f46c2 3344 spin_lock(&pers_lock);
f2859af6 3345 pers = find_pers(level, clevel);
245f46c2
N		 3346 if (!pers || !try_module_get(pers->owner)) {
3347 spin_unlock(&pers_lock);
f2859af6 3348 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
245f46c2
N		 3349 return -EINVAL;
3350 }
3351 spin_unlock(&pers_lock);
3352
3353 if (pers == mddev->pers) {
3354 /* Nothing to do! */
3355 module_put(pers->owner);
3356 return rv;
3357 }
3358 if (!pers->takeover) {
3359 module_put(pers->owner);
3360 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
f2859af6 3361 mdname(mddev), clevel);
245f46c2
N		 3362 return -EINVAL;
3363 }
3364
e93f68a1
N		 3365 list_for_each_entry(rdev, &mddev->disks, same_set)
3366 rdev->new_raid_disk = rdev->raid_disk;
3367
245f46c2
N		 3368 /* ->takeover must set new_* and/or delta_disks
3369 * if it succeeds, and may set them when it fails.
3370 */
3371 priv = pers->takeover(mddev);
3372 if (IS_ERR(priv)) {
3373 mddev->new_level = mddev->level;
3374 mddev->new_layout = mddev->layout;
664e7c41 3375 mddev->new_chunk_sectors = mddev->chunk_sectors;
245f46c2
N		 3376 mddev->raid_disks -= mddev->delta_disks;
3377 mddev->delta_disks = 0;
3378 module_put(pers->owner);
3379 printk(KERN_WARNING "md: %s: %s would not accept array\n",
f2859af6 3380 mdname(mddev), clevel);
245f46c2
N		 3381 return PTR_ERR(priv);
3382 }
3383
3384 /* Looks like we have a winner */
3385 mddev_suspend(mddev);
3386 mddev->pers->stop(mddev);
a64c876f
N		 3387
3388 if (mddev->pers->sync_request == NULL &&
3389 pers->sync_request != NULL) {
3390 /* need to add the md_redundancy_group */
3391 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3392 printk(KERN_WARNING
3393 "md: cannot register extra attributes for %s\n",
3394 mdname(mddev));
19fdb9ee 3395 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
a64c876f
N		 3396 }
3397 if (mddev->pers->sync_request != NULL &&
3398 pers->sync_request == NULL) {
3399 /* need to remove the md_redundancy_group */
3400 if (mddev->to_remove == NULL)
3401 mddev->to_remove = &md_redundancy_group;
3402 }
3403
54071b38
TM		 3404 if (mddev->pers->sync_request == NULL &&
3405 mddev->external) {
3406 /* We are converting from a no-redundancy array
3407 * to a redundancy array and metadata is managed
3408 * externally so we need to be sure that writes
3409 * won't block due to a need to transition
3410 * clean->dirty
3411 * until external management is started.
3412 */
3413 mddev->in_sync = 0;
3414 mddev->safemode_delay = 0;
3415 mddev->safemode = 0;
3416 }
3417
e93f68a1 3418 list_for_each_entry(rdev, &mddev->disks, same_set) {
e93f68a1
N		 3419 if (rdev->raid_disk < 0)
3420 continue;
bf2cb0da 3421 if (rdev->new_raid_disk >= mddev->raid_disks)
e93f68a1
N		 3422 rdev->new_raid_disk = -1;
3423 if (rdev->new_raid_disk == rdev->raid_disk)
3424 continue;
36fad858 3425 sysfs_unlink_rdev(mddev, rdev);
e93f68a1
N		 3426 }
3427 list_for_each_entry(rdev, &mddev->disks, same_set) {
3428 if (rdev->raid_disk < 0)
3429 continue;
3430 if (rdev->new_raid_disk == rdev->raid_disk)
3431 continue;
3432 rdev->raid_disk = rdev->new_raid_disk;
3433 if (rdev->raid_disk < 0)
3a981b03 3434 clear_bit(In_sync, &rdev->flags);
e93f68a1 3435 else {
36fad858
NK		 3436 if (sysfs_link_rdev(mddev, rdev))
3437 printk(KERN_WARNING "md: cannot register rd%d"
3438 " for %s after level change\n",
3439 rdev->raid_disk, mdname(mddev));
3a981b03 3440 }
e93f68a1
N		 3441 }
3442
3443 module_put(mddev->pers->owner);
245f46c2
N		 3444 mddev->pers = pers;
3445 mddev->private = priv;
3446 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3447 mddev->level = mddev->new_level;
3448 mddev->layout = mddev->new_layout;
664e7c41 3449 mddev->chunk_sectors = mddev->new_chunk_sectors;
245f46c2 3450 mddev->delta_disks = 0;
fee68723 3451 mddev->degraded = 0;
9af204cf
TM		 3452 if (mddev->pers->sync_request == NULL) {
3453 /* this is now an array without redundancy, so
3454 * it must always be in_sync
3455 */
3456 mddev->in_sync = 1;
3457 del_timer_sync(&mddev->safemode_timer);
3458 }
245f46c2
N		 3459 pers->run(mddev);
3460 mddev_resume(mddev);
3461 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3462 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3463 md_wakeup_thread(mddev->thread);
5cac7861 3464 sysfs_notify(&mddev->kobj, NULL, "level");
bb7f8d22 3465 md_new_event(mddev);
d9d166c2
N		 3466 return rv;
3467}
3468
3469static struct md_sysfs_entry md_level =
80ca3a44 3470__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
eae1701f 3471
d4dbd025
N		 3472
3473static ssize_t
fd01b88c 3474layout_show(struct mddev *mddev, char *page)
d4dbd025
N		 3475{
3476 /* just a number, not meaningful for all levels */
08a02ecd
N		 3477 if (mddev->reshape_position != MaxSector &&
3478 mddev->layout != mddev->new_layout)
3479 return sprintf(page, "%d (%d)\n",
3480 mddev->new_layout, mddev->layout);
d4dbd025
N		 3481 return sprintf(page, "%d\n", mddev->layout);
3482}
3483
3484static ssize_t
fd01b88c 3485layout_store(struct mddev *mddev, const char *buf, size_t len)
d4dbd025
N		 3486{
3487 char *e;
3488 unsigned long n = simple_strtoul(buf, &e, 10);
d4dbd025
N		 3489
3490 if (!*buf || (*e && *e != '\n'))
3491 return -EINVAL;
3492
b3546035
N		 3493 if (mddev->pers) {
3494 int err;
50ac168a 3495 if (mddev->pers->check_reshape == NULL)
b3546035 3496 return -EBUSY;
597a711b 3497 mddev->new_layout = n;
50ac168a 3498 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3499 if (err) {
3500 mddev->new_layout = mddev->layout;
b3546035 3501 return err;
597a711b 3502 }
b3546035 3503 } else {
08a02ecd 3504 mddev->new_layout = n;
b3546035
N		 3505 if (mddev->reshape_position == MaxSector)
3506 mddev->layout = n;
3507 }
d4dbd025
N		 3508 return len;
3509}
3510static struct md_sysfs_entry md_layout =
80ca3a44 3511__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
d4dbd025
N		 3512
3513
eae1701f 3514static ssize_t
fd01b88c 3515raid_disks_show(struct mddev *mddev, char *page)
eae1701f 3516{
bb636547
N		 3517 if (mddev->raid_disks == 0)
3518 return 0;
08a02ecd
N		 3519 if (mddev->reshape_position != MaxSector &&
3520 mddev->delta_disks != 0)
3521 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3522 mddev->raid_disks - mddev->delta_disks);
eae1701f
N		 3523 return sprintf(page, "%d\n", mddev->raid_disks);
3524}
3525
fd01b88c 3526static int update_raid_disks(struct mddev *mddev, int raid_disks);
da943b99
N		 3527
3528static ssize_t
fd01b88c 3529raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
da943b99 3530{
da943b99
N		 3531 char *e;
3532 int rv = 0;
3533 unsigned long n = simple_strtoul(buf, &e, 10);
3534
3535 if (!*buf || (*e && *e != '\n'))
3536 return -EINVAL;
3537
3538 if (mddev->pers)
3539 rv = update_raid_disks(mddev, n);
08a02ecd
N		 3540 else if (mddev->reshape_position != MaxSector) {
3541 int olddisks = mddev->raid_disks - mddev->delta_disks;
3542 mddev->delta_disks = n - olddisks;
3543 mddev->raid_disks = n;
3544 } else
da943b99
N		 3545 mddev->raid_disks = n;
3546 return rv ? rv : len;
3547}
3548static struct md_sysfs_entry md_raid_disks =
80ca3a44 3549__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
eae1701f 3550
3b34380a 3551static ssize_t
fd01b88c 3552chunk_size_show(struct mddev *mddev, char *page)
3b34380a 3553{
08a02ecd 3554 if (mddev->reshape_position != MaxSector &&
664e7c41
AN		 3555 mddev->chunk_sectors != mddev->new_chunk_sectors)
3556 return sprintf(page, "%d (%d)\n",
3557 mddev->new_chunk_sectors << 9,
9d8f0363
AN		 3558 mddev->chunk_sectors << 9);
3559 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3b34380a
N		 3560}
3561
3562static ssize_t
fd01b88c 3563chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3b34380a 3564{
3b34380a
N		 3565 char *e;
3566 unsigned long n = simple_strtoul(buf, &e, 10);
3567
3b34380a
N		 3568 if (!*buf || (*e && *e != '\n'))
3569 return -EINVAL;
3570
b3546035
N		 3571 if (mddev->pers) {
3572 int err;
50ac168a 3573 if (mddev->pers->check_reshape == NULL)
b3546035 3574 return -EBUSY;
597a711b 3575 mddev->new_chunk_sectors = n >> 9;
50ac168a 3576 err = mddev->pers->check_reshape(mddev);
597a711b
N		 3577 if (err) {
3578 mddev->new_chunk_sectors = mddev->chunk_sectors;
b3546035 3579 return err;
597a711b 3580 }
b3546035 3581 } else {
664e7c41 3582 mddev->new_chunk_sectors = n >> 9;
b3546035 3583 if (mddev->reshape_position == MaxSector)
9d8f0363 3584 mddev->chunk_sectors = n >> 9;
b3546035 3585 }
3b34380a
N		 3586 return len;
3587}
3588static struct md_sysfs_entry md_chunk_size =
80ca3a44 3589__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3b34380a 3590
a94213b1 3591static ssize_t
fd01b88c 3592resync_start_show(struct mddev *mddev, char *page)
a94213b1 3593{
d1a7c503
N		 3594 if (mddev->recovery_cp == MaxSector)
3595 return sprintf(page, "none\n");
a94213b1
N		 3596 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3597}
3598
3599static ssize_t
fd01b88c 3600resync_start_store(struct mddev *mddev, const char *buf, size_t len)
a94213b1 3601{
a94213b1
N		 3602 char *e;
3603 unsigned long long n = simple_strtoull(buf, &e, 10);
3604
b098636c 3605 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
a94213b1 3606 return -EBUSY;
06e3c817
DW		 3607 if (cmd_match(buf, "none"))
3608 n = MaxSector;
3609 else if (!*buf || (*e && *e != '\n'))
a94213b1
N		 3610 return -EINVAL;
3611
3612 mddev->recovery_cp = n;
3613 return len;
3614}
3615static struct md_sysfs_entry md_resync_start =
80ca3a44 3616__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
a94213b1 3617
9e653b63
N		 3618/*
3619 * The array state can be:
3620 *
3621 * clear
3622 * No devices, no size, no level
3623 * Equivalent to STOP_ARRAY ioctl
3624 * inactive
3625 * May have some settings, but array is not active
3626 * all IO results in error
3627 * When written, doesn't tear down array, but just stops it
3628 * suspended (not supported yet)
3629 * All IO requests will block. The array can be reconfigured.
910d8cb3 3630 * Writing this, if accepted, will block until array is quiescent
9e653b63
N		 3631 * readonly
3632 * no resync can happen. no superblocks get written.
3633 * write requests fail
3634 * read-auto
3635 * like readonly, but behaves like 'clean' on a write request.
3636 *
3637 * clean - no pending writes, but otherwise active.
3638 * When written to inactive array, starts without resync
3639 * If a write request arrives then
3640 * if metadata is known, mark 'dirty' and switch to 'active'.
3641 * if not known, block and switch to write-pending
3642 * If written to an active array that has pending writes, then fails.
3643 * active
3644 * fully active: IO and resync can be happening.
3645 * When written to inactive array, starts with resync
3646 *
3647 * write-pending
3648 * clean, but writes are blocked waiting for 'active' to be written.
3649 *
3650 * active-idle
3651 * like active, but no writes have been seen for a while (100msec).
3652 *
3653 */
3654enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3655 write_pending, active_idle, bad_word};
05381954 3656static char *array_states[] = {
9e653b63
N		 3657 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3658 "write-pending", "active-idle", NULL };
3659
3660static int match_word(const char *word, char **list)
3661{
3662 int n;
3663 for (n=0; list[n]; n++)
3664 if (cmd_match(word, list[n]))
3665 break;
3666 return n;
3667}
3668
3669static ssize_t
fd01b88c 3670array_state_show(struct mddev *mddev, char *page)
9e653b63
N		 3671{
3672 enum array_state st = inactive;
3673
3674 if (mddev->pers)
3675 switch(mddev->ro) {
3676 case 1:
3677 st = readonly;
3678 break;
3679 case 2:
3680 st = read_auto;
3681 break;
3682 case 0:
3683 if (mddev->in_sync)
3684 st = clean;
070dc6dd 3685 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
e691063a 3686 st = write_pending;
9e653b63
N		 3687 else if (mddev->safemode)
3688 st = active_idle;
3689 else
3690 st = active;
3691 }
3692 else {
3693 if (list_empty(&mddev->disks) &&
3694 mddev->raid_disks == 0 &&
58c0fed4 3695 mddev->dev_sectors == 0)
9e653b63
N		 3696 st = clear;
3697 else
3698 st = inactive;
3699 }
3700 return sprintf(page, "%s\n", array_states[st]);
3701}
3702
fd01b88c
N		 3703static int do_md_stop(struct mddev * mddev, int ro, int is_open);
3704static int md_set_readonly(struct mddev * mddev, int is_open);
3705static int do_md_run(struct mddev * mddev);
3706static int restart_array(struct mddev *mddev);
9e653b63
N		 3707
3708static ssize_t
fd01b88c 3709array_state_store(struct mddev *mddev, const char *buf, size_t len)
9e653b63
N		 3710{
3711 int err = -EINVAL;
3712 enum array_state st = match_word(buf, array_states);
3713 switch(st) {
3714 case bad_word:
3715 break;
3716 case clear:
3717 /* stopping an active array */
f2ea68cf 3718 if (atomic_read(&mddev->openers) > 0)
e691063a 3719 return -EBUSY;
df5b20cf 3720 err = do_md_stop(mddev, 0, 0);
9e653b63
N		 3721 break;
3722 case inactive:
3723 /* stopping an active array */
3724 if (mddev->pers) {
f2ea68cf 3725 if (atomic_read(&mddev->openers) > 0)
9e653b63 3726 return -EBUSY;
df5b20cf 3727 err = do_md_stop(mddev, 2, 0);
e691063a
N		 3728 } else
3729 err = 0; /* already inactive */
9e653b63
N		 3730 break;
3731 case suspended:
3732 break; /* not supported yet */
3733 case readonly:
3734 if (mddev->pers)
a4bd82d0 3735 err = md_set_readonly(mddev, 0);
9e653b63
N		 3736 else {
3737 mddev->ro = 1;
648b629e 3738 set_disk_ro(mddev->gendisk, 1);
9e653b63
N		 3739 err = do_md_run(mddev);
3740 }
3741 break;
3742 case read_auto:
9e653b63 3743 if (mddev->pers) {
80268ee9 3744 if (mddev->ro == 0)
a4bd82d0 3745 err = md_set_readonly(mddev, 0);
80268ee9 3746 else if (mddev->ro == 1)
648b629e
N		 3747 err = restart_array(mddev);
3748 if (err == 0) {
3749 mddev->ro = 2;
3750 set_disk_ro(mddev->gendisk, 0);
3751 }
9e653b63
N		 3752 } else {
3753 mddev->ro = 2;
3754 err = do_md_run(mddev);
3755 }
3756 break;
3757 case clean:
3758 if (mddev->pers) {
3759 restart_array(mddev);
3760 spin_lock_irq(&mddev->write_lock);
3761 if (atomic_read(&mddev->writes_pending) == 0) {
e691063a
N		 3762 if (mddev->in_sync == 0) {
3763 mddev->in_sync = 1;
31a59e34
N		 3764 if (mddev->safemode == 1)
3765 mddev->safemode = 0;
070dc6dd 3766 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
e691063a
N		 3767 }
3768 err = 0;
3769 } else
3770 err = -EBUSY;
9e653b63 3771 spin_unlock_irq(&mddev->write_lock);
5bf29597
N		 3772 } else
3773 err = -EINVAL;
9e653b63
N		 3774 break;
3775 case active:
3776 if (mddev->pers) {
3777 restart_array(mddev);
070dc6dd 3778 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
9e653b63
N		 3779 wake_up(&mddev->sb_wait);
3780 err = 0;
3781 } else {
3782 mddev->ro = 0;
648b629e 3783 set_disk_ro(mddev->gendisk, 0);
9e653b63
N		 3784 err = do_md_run(mddev);
3785 }
3786 break;
3787 case write_pending:
3788 case active_idle:
3789 /* these cannot be set */
3790 break;
3791 }
3792 if (err)
3793 return err;
0fd62b86 3794 else {
00bcb4ac 3795 sysfs_notify_dirent_safe(mddev->sysfs_state);
9e653b63 3796 return len;
0fd62b86 3797 }
9e653b63 3798}
80ca3a44
N		 3799static struct md_sysfs_entry md_array_state =
3800__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
9e653b63 3801
1e50915f 3802static ssize_t
fd01b88c 3803max_corrected_read_errors_show(struct mddev *mddev, char *page) {
1e50915f
RB		 3804 return sprintf(page, "%d\n",
3805 atomic_read(&mddev->max_corr_read_errors));
3806}
3807
3808static ssize_t
fd01b88c 3809max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
1e50915f
RB		 3810{
3811 char *e;
3812 unsigned long n = simple_strtoul(buf, &e, 10);
3813
3814 if (*buf && (*e == 0 || *e == '\n')) {
3815 atomic_set(&mddev->max_corr_read_errors, n);
3816 return len;
3817 }
3818 return -EINVAL;
3819}
3820
3821static struct md_sysfs_entry max_corr_read_errors =
3822__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3823 max_corrected_read_errors_store);
3824
6d7ff738 3825static ssize_t
fd01b88c 3826null_show(struct mddev *mddev, char *page)
6d7ff738
N		 3827{
3828 return -EINVAL;
3829}
3830
3831static ssize_t
fd01b88c 3832new_dev_store(struct mddev *mddev, const char *buf, size_t len)
6d7ff738
N		 3833{
3834 /* buf must be %d:%d\n? giving major and minor numbers */
3835 /* The new device is added to the array.
3836 * If the array has a persistent superblock, we read the
3837 * superblock to initialise info and check validity.
3838 * Otherwise, only checking done is that in bind_rdev_to_array,
3839 * which mainly checks size.
3840 */
3841 char *e;
3842 int major = simple_strtoul(buf, &e, 10);
3843 int minor;
3844 dev_t dev;
3cb03002 3845 struct md_rdev *rdev;
6d7ff738
N		 3846 int err;
3847
3848 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3849 return -EINVAL;
3850 minor = simple_strtoul(e+1, &e, 10);
3851 if (*e && *e != '\n')
3852 return -EINVAL;
3853 dev = MKDEV(major, minor);
3854 if (major != MAJOR(dev) ||
3855 minor != MINOR(dev))
3856 return -EOVERFLOW;
3857
3858
3859 if (mddev->persistent) {
3860 rdev = md_import_device(dev, mddev->major_version,
3861 mddev->minor_version);
3862 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3cb03002
N		 3863 struct md_rdev *rdev0
3864 = list_entry(mddev->disks.next,
3865 struct md_rdev, same_set);
6d7ff738
N		 3866 err = super_types[mddev->major_version]
3867 .load_super(rdev, rdev0, mddev->minor_version);
3868 if (err < 0)
3869 goto out;
3870 }
c5d79adb
N		 3871 } else if (mddev->external)
3872 rdev = md_import_device(dev, -2, -1);
3873 else
6d7ff738
N		 3874 rdev = md_import_device(dev, -1, -1);
3875
3876 if (IS_ERR(rdev))
3877 return PTR_ERR(rdev);
3878 err = bind_rdev_to_array(rdev, mddev);
3879 out:
3880 if (err)
3881 export_rdev(rdev);
3882 return err ? err : len;
3883}
3884
3885static struct md_sysfs_entry md_new_device =
80ca3a44 3886__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3b34380a 3887
9b1d1dac 3888static ssize_t
fd01b88c 3889bitmap_store(struct mddev *mddev, const char *buf, size_t len)
9b1d1dac
PC		 3890{
3891 char *end;
3892 unsigned long chunk, end_chunk;
3893
3894 if (!mddev->bitmap)
3895 goto out;
3896 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3897 while (*buf) {
3898 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3899 if (buf == end) break;
3900 if (*end == '-') { /* range */
3901 buf = end + 1;
3902 end_chunk = simple_strtoul(buf, &end, 0);
3903 if (buf == end) break;
3904 }
3905 if (*end && !isspace(*end)) break;
3906 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
e7d2860b 3907 buf = skip_spaces(end);
9b1d1dac
PC		 3908 }
3909 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3910out:
3911 return len;
3912}
3913
3914static struct md_sysfs_entry md_bitmap =
3915__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3916
a35b0d69 3917static ssize_t
fd01b88c 3918size_show(struct mddev *mddev, char *page)
a35b0d69 3919{
58c0fed4
AN		 3920 return sprintf(page, "%llu\n",
3921 (unsigned long long)mddev->dev_sectors / 2);
a35b0d69
N		 3922}
3923
fd01b88c 3924static int update_size(struct mddev *mddev, sector_t num_sectors);
a35b0d69
N		 3925
3926static ssize_t
fd01b88c 3927size_store(struct mddev *mddev, const char *buf, size_t len)
a35b0d69
N		 3928{
3929 /* If array is inactive, we can reduce the component size, but
3930 * not increase it (except from 0).
3931 * If array is active, we can try an on-line resize
3932 */
b522adcd
DW		 3933 sector_t sectors;
3934 int err = strict_blocks_to_sectors(buf, &sectors);
a35b0d69 3935
58c0fed4
AN		 3936 if (err < 0)
3937 return err;
a35b0d69 3938 if (mddev->pers) {
58c0fed4 3939 err = update_size(mddev, sectors);
850b2b42 3940 md_update_sb(mddev, 1);
a35b0d69 3941 } else {
58c0fed4
AN		 3942 if (mddev->dev_sectors == 0 ||
3943 mddev->dev_sectors > sectors)
3944 mddev->dev_sectors = sectors;
a35b0d69
N		 3945 else
3946 err = -ENOSPC;
3947 }
3948 return err ? err : len;
3949}
3950
3951static struct md_sysfs_entry md_size =
80ca3a44 3952__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
a35b0d69 3953
8bb93aac
N		 3954
3955/* Metdata version.
e691063a
N		 3956 * This is one of
3957 * 'none' for arrays with no metadata (good luck...)
3958 * 'external' for arrays with externally managed metadata,
8bb93aac
N		 3959 * or N.M for internally known formats
3960 */
3961static ssize_t
fd01b88c 3962metadata_show(struct mddev *mddev, char *page)
8bb93aac
N		 3963{
3964 if (mddev->persistent)
3965 return sprintf(page, "%d.%d\n",
3966 mddev->major_version, mddev->minor_version);
e691063a
N		 3967 else if (mddev->external)
3968 return sprintf(page, "external:%s\n", mddev->metadata_type);
8bb93aac
N		 3969 else
3970 return sprintf(page, "none\n");
3971}
3972
3973static ssize_t
fd01b88c 3974metadata_store(struct mddev *mddev, const char *buf, size_t len)
8bb93aac
N		 3975{
3976 int major, minor;
3977 char *e;
ea43ddd8
N		 3978 /* Changing the details of 'external' metadata is
3979 * always permitted. Otherwise there must be
3980 * no devices attached to the array.
3981 */
3982 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3983 ;
3984 else if (!list_empty(&mddev->disks))
8bb93aac
N		 3985 return -EBUSY;
3986
3987 if (cmd_match(buf, "none")) {
3988 mddev->persistent = 0;
e691063a
N		 3989 mddev->external = 0;
3990 mddev->major_version = 0;
3991 mddev->minor_version = 90;
3992 return len;
3993 }
3994 if (strncmp(buf, "external:", 9) == 0) {
20a49ff6 3995 size_t namelen = len-9;
e691063a
N		 3996 if (namelen >= sizeof(mddev->metadata_type))
3997 namelen = sizeof(mddev->metadata_type)-1;
3998 strncpy(mddev->metadata_type, buf+9, namelen);
3999 mddev->metadata_type[namelen] = 0;
4000 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4001 mddev->metadata_type[--namelen] = 0;
4002 mddev->persistent = 0;
4003 mddev->external = 1;
8bb93aac
N		 4004 mddev->major_version = 0;
4005 mddev->minor_version = 90;
4006 return len;
4007 }
4008 major = simple_strtoul(buf, &e, 10);
4009 if (e==buf || *e != '.')
4010 return -EINVAL;
4011 buf = e+1;
4012 minor = simple_strtoul(buf, &e, 10);
3f9d7b0d 4013 if (e==buf || (*e && *e != '\n') )
8bb93aac 4014 return -EINVAL;
50511da3 4015 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
8bb93aac
N		 4016 return -ENOENT;
4017 mddev->major_version = major;
4018 mddev->minor_version = minor;
4019 mddev->persistent = 1;
e691063a 4020 mddev->external = 0;
8bb93aac
N		 4021 return len;
4022}
4023
4024static struct md_sysfs_entry md_metadata =
80ca3a44 4025__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
8bb93aac 4026
24dd469d 4027static ssize_t
fd01b88c 4028action_show(struct mddev *mddev, char *page)
24dd469d 4029{
7eec314d 4030 char *type = "idle";
b6a9ce68
N		 4031 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4032 type = "frozen";
4033 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2b12ab6d 4034 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
ccfcc3c1
N		 4035 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4036 type = "reshape";
4037 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
24dd469d
N		 4038 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4039 type = "resync";
4040 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4041 type = "check";
4042 else
4043 type = "repair";
72a23c21 4044 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
24dd469d
N		 4045 type = "recover";
4046 }
4047 return sprintf(page, "%s\n", type);
4048}
4049
fd01b88c 4050static void reap_sync_thread(struct mddev *mddev);
7ebc0be7 4051
24dd469d 4052static ssize_t
fd01b88c 4053action_store(struct mddev *mddev, const char *page, size_t len)
24dd469d 4054{
7eec314d
N		 4055 if (!mddev->pers || !mddev->pers->sync_request)
4056 return -EINVAL;
4057
b6a9ce68
N		 4058 if (cmd_match(page, "frozen"))
4059 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4060 else
4061 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4062
4063 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
7eec314d
N		 4064 if (mddev->sync_thread) {
4065 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 4066 reap_sync_thread(mddev);
7eec314d 4067 }
03c902e1
N		 4068 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4069 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
24dd469d 4070 return -EBUSY;
72a23c21
NB		 4071 else if (cmd_match(page, "resync"))
4072 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4073 else if (cmd_match(page, "recover")) {
4074 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7eec314d 4075 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
72a23c21 4076 } else if (cmd_match(page, "reshape")) {
16484bf5
N		 4077 int err;
4078 if (mddev->pers->start_reshape == NULL)
4079 return -EINVAL;
4080 err = mddev->pers->start_reshape(mddev);
4081 if (err)
4082 return err;
a99ac971 4083 sysfs_notify(&mddev->kobj, NULL, "degraded");
16484bf5 4084 } else {
bce74dac 4085 if (cmd_match(page, "check"))
7eec314d 4086 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2adc7d47 4087 else if (!cmd_match(page, "repair"))
7eec314d
N		 4088 return -EINVAL;
4089 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4090 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7eec314d 4091 }
03c902e1 4092 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d 4093 md_wakeup_thread(mddev->thread);
00bcb4ac 4094 sysfs_notify_dirent_safe(mddev->sysfs_action);
24dd469d
N		 4095 return len;
4096}
4097
9d88883e 4098static ssize_t
fd01b88c 4099mismatch_cnt_show(struct mddev *mddev, char *page)
9d88883e
N		 4100{
4101 return sprintf(page, "%llu\n",
4102 (unsigned long long) mddev->resync_mismatches);
4103}
4104
80ca3a44
N		 4105static struct md_sysfs_entry md_scan_mode =
4106__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
24dd469d 4107
96de1e66 4108
80ca3a44 4109static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
9d88883e 4110
88202a0c 4111static ssize_t
fd01b88c 4112sync_min_show(struct mddev *mddev, char *page)
88202a0c
N		 4113{
4114 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4115 mddev->sync_speed_min ? "local": "system");
4116}
4117
4118static ssize_t
fd01b88c 4119sync_min_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4120{
4121 int min;
4122 char *e;
4123 if (strncmp(buf, "system", 6)==0) {
4124 mddev->sync_speed_min = 0;
4125 return len;
4126 }
4127 min = simple_strtoul(buf, &e, 10);
4128 if (buf == e || (*e && *e != '\n') || min <= 0)
4129 return -EINVAL;
4130 mddev->sync_speed_min = min;
4131 return len;
4132}
4133
4134static struct md_sysfs_entry md_sync_min =
4135__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4136
4137static ssize_t
fd01b88c 4138sync_max_show(struct mddev *mddev, char *page)
88202a0c
N		 4139{
4140 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4141 mddev->sync_speed_max ? "local": "system");
4142}
4143
4144static ssize_t
fd01b88c 4145sync_max_store(struct mddev *mddev, const char *buf, size_t len)
88202a0c
N		 4146{
4147 int max;
4148 char *e;
4149 if (strncmp(buf, "system", 6)==0) {
4150 mddev->sync_speed_max = 0;
4151 return len;
4152 }
4153 max = simple_strtoul(buf, &e, 10);
4154 if (buf == e || (*e && *e != '\n') || max <= 0)
4155 return -EINVAL;
4156 mddev->sync_speed_max = max;
4157 return len;
4158}
4159
4160static struct md_sysfs_entry md_sync_max =
4161__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4162
d7f3d291 4163static ssize_t
fd01b88c 4164degraded_show(struct mddev *mddev, char *page)
d7f3d291
IP		 4165{
4166 return sprintf(page, "%d\n", mddev->degraded);
4167}
4168static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
88202a0c 4169
90b08710 4170static ssize_t
fd01b88c 4171sync_force_parallel_show(struct mddev *mddev, char *page)
90b08710
BS		 4172{
4173 return sprintf(page, "%d\n", mddev->parallel_resync);
4174}
4175
4176static ssize_t
fd01b88c 4177sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
90b08710
BS		 4178{
4179 long n;
4180
4181 if (strict_strtol(buf, 10, &n))
4182 return -EINVAL;
4183
4184 if (n != 0 && n != 1)
4185 return -EINVAL;
4186
4187 mddev->parallel_resync = n;
4188
4189 if (mddev->sync_thread)
4190 wake_up(&resync_wait);
4191
4192 return len;
4193}
4194
4195/* force parallel resync, even with shared block devices */
4196static struct md_sysfs_entry md_sync_force_parallel =
4197__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4198 sync_force_parallel_show, sync_force_parallel_store);
4199
88202a0c 4200static ssize_t
fd01b88c 4201sync_speed_show(struct mddev *mddev, char *page)
88202a0c
N		 4202{
4203 unsigned long resync, dt, db;
d1a7c503
N		 4204 if (mddev->curr_resync == 0)
4205 return sprintf(page, "none\n");
9687a60c
AN		 4206 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4207 dt = (jiffies - mddev->resync_mark) / HZ;
88202a0c 4208 if (!dt) dt++;
9687a60c
AN		 4209 db = resync - mddev->resync_mark_cnt;
4210 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
88202a0c
N		 4211}
4212
80ca3a44 4213static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
88202a0c
N		 4214
4215static ssize_t
fd01b88c 4216sync_completed_show(struct mddev *mddev, char *page)
88202a0c 4217{
13ae864b 4218 unsigned long long max_sectors, resync;
88202a0c 4219
acb180b0
N		 4220 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4221 return sprintf(page, "none\n");
4222
88202a0c 4223 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
58c0fed4 4224 max_sectors = mddev->resync_max_sectors;
88202a0c 4225 else
58c0fed4 4226 max_sectors = mddev->dev_sectors;
88202a0c 4227
acb180b0 4228 resync = mddev->curr_resync_completed;
13ae864b 4229 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
88202a0c
N		 4230}
4231
80ca3a44 4232static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
88202a0c 4233
5e96ee65 4234static ssize_t
fd01b88c 4235min_sync_show(struct mddev *mddev, char *page)
5e96ee65
NB		 4236{
4237 return sprintf(page, "%llu\n",
4238 (unsigned long long)mddev->resync_min);
4239}
4240static ssize_t
fd01b88c 4241min_sync_store(struct mddev *mddev, const char *buf, size_t len)
5e96ee65
NB		 4242{
4243 unsigned long long min;
4244 if (strict_strtoull(buf, 10, &min))
4245 return -EINVAL;
4246 if (min > mddev->resync_max)
4247 return -EINVAL;
4248 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4249 return -EBUSY;
4250
4251 /* Must be a multiple of chunk_size */
9d8f0363 4252 if (mddev->chunk_sectors) {
2ac06c33 4253 sector_t temp = min;
9d8f0363 4254 if (sector_div(temp, mddev->chunk_sectors))
5e96ee65
NB		 4255 return -EINVAL;
4256 }
4257 mddev->resync_min = min;
4258
4259 return len;
4260}
4261
4262static struct md_sysfs_entry md_min_sync =
4263__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4264
c6207277 4265static ssize_t
fd01b88c 4266max_sync_show(struct mddev *mddev, char *page)
c6207277
N		 4267{
4268 if (mddev->resync_max == MaxSector)
4269 return sprintf(page, "max\n");
4270 else
4271 return sprintf(page, "%llu\n",
4272 (unsigned long long)mddev->resync_max);
4273}
4274static ssize_t
fd01b88c 4275max_sync_store(struct mddev *mddev, const char *buf, size_t len)
c6207277
N		 4276{
4277 if (strncmp(buf, "max", 3) == 0)
4278 mddev->resync_max = MaxSector;
4279 else {
5e96ee65
NB		 4280 unsigned long long max;
4281 if (strict_strtoull(buf, 10, &max))
4282 return -EINVAL;
4283 if (max < mddev->resync_min)
c6207277
N		 4284 return -EINVAL;
4285 if (max < mddev->resync_max &&
4d484a4a 4286 mddev->ro == 0 &&
c6207277
N		 4287 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4288 return -EBUSY;
4289
4290 /* Must be a multiple of chunk_size */
9d8f0363 4291 if (mddev->chunk_sectors) {
2ac06c33 4292 sector_t temp = max;
9d8f0363 4293 if (sector_div(temp, mddev->chunk_sectors))
c6207277
N		 4294 return -EINVAL;
4295 }
4296 mddev->resync_max = max;
4297 }
4298 wake_up(&mddev->recovery_wait);
4299 return len;
4300}
4301
4302static struct md_sysfs_entry md_max_sync =
4303__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4304
e464eafd 4305static ssize_t
fd01b88c 4306suspend_lo_show(struct mddev *mddev, char *page)
e464eafd
N		 4307{
4308 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4309}
4310
4311static ssize_t
fd01b88c 4312suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4313{
4314 char *e;
4315 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4316 unsigned long long old = mddev->suspend_lo;
e464eafd 4317
b8d966ef
N		 4318 if (mddev->pers == NULL ||
4319 mddev->pers->quiesce == NULL)
e464eafd
N		 4320 return -EINVAL;
4321 if (buf == e || (*e && *e != '\n'))
4322 return -EINVAL;
23ddff37
N		 4323
4324 mddev->suspend_lo = new;
4325 if (new >= old)
4326 /* Shrinking suspended region */
e464eafd 4327 mddev->pers->quiesce(mddev, 2);
23ddff37
N		 4328 else {
4329 /* Expanding suspended region - need to wait */
4330 mddev->pers->quiesce(mddev, 1);
4331 mddev->pers->quiesce(mddev, 0);
4332 }
4333 return len;
e464eafd
N		 4334}
4335static struct md_sysfs_entry md_suspend_lo =
4336__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4337
4338
4339static ssize_t
fd01b88c 4340suspend_hi_show(struct mddev *mddev, char *page)
e464eafd
N		 4341{
4342 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4343}
4344
4345static ssize_t
fd01b88c 4346suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
e464eafd
N		 4347{
4348 char *e;
4349 unsigned long long new = simple_strtoull(buf, &e, 10);
23ddff37 4350 unsigned long long old = mddev->suspend_hi;
e464eafd 4351
b8d966ef
N		 4352 if (mddev->pers == NULL ||
4353 mddev->pers->quiesce == NULL)
e464eafd
N		 4354 return -EINVAL;
4355 if (buf == e || (*e && *e != '\n'))
4356 return -EINVAL;
23ddff37
N		 4357
4358 mddev->suspend_hi = new;
4359 if (new <= old)
4360 /* Shrinking suspended region */
4361 mddev->pers->quiesce(mddev, 2);
4362 else {
4363 /* Expanding suspended region - need to wait */
e464eafd
N		 4364 mddev->pers->quiesce(mddev, 1);
4365 mddev->pers->quiesce(mddev, 0);
23ddff37
N		 4366 }
4367 return len;
e464eafd
N		 4368}
4369static struct md_sysfs_entry md_suspend_hi =
4370__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4371
08a02ecd 4372static ssize_t
fd01b88c 4373reshape_position_show(struct mddev *mddev, char *page)
08a02ecd
N		 4374{
4375 if (mddev->reshape_position != MaxSector)
4376 return sprintf(page, "%llu\n",
4377 (unsigned long long)mddev->reshape_position);
4378 strcpy(page, "none\n");
4379 return 5;
4380}
4381
4382static ssize_t
fd01b88c 4383reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
08a02ecd
N		 4384{
4385 char *e;
4386 unsigned long long new = simple_strtoull(buf, &e, 10);
4387 if (mddev->pers)
4388 return -EBUSY;
4389 if (buf == e || (*e && *e != '\n'))
4390 return -EINVAL;
4391 mddev->reshape_position = new;
4392 mddev->delta_disks = 0;
4393 mddev->new_level = mddev->level;
4394 mddev->new_layout = mddev->layout;
664e7c41 4395 mddev->new_chunk_sectors = mddev->chunk_sectors;
08a02ecd
N		 4396 return len;
4397}
4398
4399static struct md_sysfs_entry md_reshape_position =
4400__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4401 reshape_position_store);
4402
b522adcd 4403static ssize_t
fd01b88c 4404array_size_show(struct mddev *mddev, char *page)
b522adcd
DW		 4405{
4406 if (mddev->external_size)
4407 return sprintf(page, "%llu\n",
4408 (unsigned long long)mddev->array_sectors/2);
4409 else
4410 return sprintf(page, "default\n");
4411}
4412
4413static ssize_t
fd01b88c 4414array_size_store(struct mddev *mddev, const char *buf, size_t len)
b522adcd
DW		 4415{
4416 sector_t sectors;
4417
4418 if (strncmp(buf, "default", 7) == 0) {
4419 if (mddev->pers)
4420 sectors = mddev->pers->size(mddev, 0, 0);
4421 else
4422 sectors = mddev->array_sectors;
4423
4424 mddev->external_size = 0;
4425 } else {
4426 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4427 return -EINVAL;
4428 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
2b69c839 4429 return -E2BIG;
b522adcd
DW		 4430
4431 mddev->external_size = 1;
4432 }
4433
4434 mddev->array_sectors = sectors;
cbe6ef1d
N		 4435 if (mddev->pers) {
4436 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 4437 revalidate_disk(mddev->gendisk);
cbe6ef1d 4438 }
b522adcd
DW		 4439 return len;
4440}
4441
4442static struct md_sysfs_entry md_array_size =
4443__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4444 array_size_store);
e464eafd 4445
eae1701f
N		 4446static struct attribute *md_default_attrs[] = {
4447 &md_level.attr,
d4dbd025 4448 &md_layout.attr,
eae1701f 4449 &md_raid_disks.attr,
3b34380a 4450 &md_chunk_size.attr,
a35b0d69 4451 &md_size.attr,
a94213b1 4452 &md_resync_start.attr,
8bb93aac 4453 &md_metadata.attr,
6d7ff738 4454 &md_new_device.attr,
16f17b39 4455 &md_safe_delay.attr,
9e653b63 4456 &md_array_state.attr,
08a02ecd 4457 &md_reshape_position.attr,
b522adcd 4458 &md_array_size.attr,
1e50915f 4459 &max_corr_read_errors.attr,
411036fa
N		 4460 NULL,
4461};
4462
4463static struct attribute *md_redundancy_attrs[] = {
24dd469d 4464 &md_scan_mode.attr,
9d88883e 4465 &md_mismatches.attr,
88202a0c
N		 4466 &md_sync_min.attr,
4467 &md_sync_max.attr,
4468 &md_sync_speed.attr,
90b08710 4469 &md_sync_force_parallel.attr,
88202a0c 4470 &md_sync_completed.attr,
5e96ee65 4471 &md_min_sync.attr,
c6207277 4472 &md_max_sync.attr,
e464eafd
N		 4473 &md_suspend_lo.attr,
4474 &md_suspend_hi.attr,
9b1d1dac 4475 &md_bitmap.attr,
d7f3d291 4476 &md_degraded.attr,
eae1701f
N		 4477 NULL,
4478};
411036fa
N		 4479static struct attribute_group md_redundancy_group = {
4480 .name = NULL,
4481 .attrs = md_redundancy_attrs,
4482};
4483
eae1701f
N		 4484
4485static ssize_t
4486md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4487{
4488 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4489 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4490 ssize_t rv;
eae1701f
N		 4491
4492 if (!entry->show)
4493 return -EIO;
5dc5cf7d
IM		 4494 rv = mddev_lock(mddev);
4495 if (!rv) {
4496 rv = entry->show(mddev, page);
4497 mddev_unlock(mddev);
4498 }
96de1e66 4499 return rv;
eae1701f
N		 4500}
4501
4502static ssize_t
4503md_attr_store(struct kobject *kobj, struct attribute *attr,
4504 const char *page, size_t length)
4505{
4506 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
fd01b88c 4507 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
96de1e66 4508 ssize_t rv;
eae1701f
N		 4509
4510 if (!entry->store)
4511 return -EIO;
67463acb
N		 4512 if (!capable(CAP_SYS_ADMIN))
4513 return -EACCES;
5dc5cf7d 4514 rv = mddev_lock(mddev);
d3374825
N		 4515 if (mddev->hold_active == UNTIL_IOCTL)
4516 mddev->hold_active = 0;
5dc5cf7d
IM		 4517 if (!rv) {
4518 rv = entry->store(mddev, page, length);
4519 mddev_unlock(mddev);
4520 }
96de1e66 4521 return rv;
eae1701f
N		 4522}
4523
4524static void md_free(struct kobject *ko)
4525{
fd01b88c 4526 struct mddev *mddev = container_of(ko, struct mddev, kobj);
a21d1504
N		 4527
4528 if (mddev->sysfs_state)
4529 sysfs_put(mddev->sysfs_state);
4530
4531 if (mddev->gendisk) {
4532 del_gendisk(mddev->gendisk);
4533 put_disk(mddev->gendisk);
4534 }
4535 if (mddev->queue)
4536 blk_cleanup_queue(mddev->queue);
4537
eae1701f
N		 4538 kfree(mddev);
4539}
4540
52cf25d0 4541static const struct sysfs_ops md_sysfs_ops = {
eae1701f
N		 4542 .show = md_attr_show,
4543 .store = md_attr_store,
4544};
4545static struct kobj_type md_ktype = {
4546 .release = md_free,
4547 .sysfs_ops = &md_sysfs_ops,
4548 .default_attrs = md_default_attrs,
4549};
4550
1da177e4
LT		 4551int mdp_major = 0;
4552
5fd3a17e
DW		 4553static void mddev_delayed_delete(struct work_struct *ws)
4554{
fd01b88c 4555 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5fd3a17e 4556
43a70507 4557 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5fd3a17e
DW		 4558 kobject_del(&mddev->kobj);
4559 kobject_put(&mddev->kobj);
4560}
4561
efeb53c0 4562static int md_alloc(dev_t dev, char *name)
1da177e4 4563{
48c9c27b 4564 static DEFINE_MUTEX(disks_mutex);
fd01b88c 4565 struct mddev *mddev = mddev_find(dev);
1da177e4 4566 struct gendisk *disk;
efeb53c0
N		 4567 int partitioned;
4568 int shift;
4569 int unit;
3830c62f 4570 int error;
1da177e4
LT		 4571
4572 if (!mddev)
efeb53c0
N		 4573 return -ENODEV;
4574
4575 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4576 shift = partitioned ? MdpMinorShift : 0;
4577 unit = MINOR(mddev->unit) >> shift;
1da177e4 4578
e804ac78
TH		 4579 /* wait for any previous instance of this device to be
4580 * completely removed (mddev_delayed_delete).
d3374825 4581 */
e804ac78 4582 flush_workqueue(md_misc_wq);
d3374825 4583
48c9c27b 4584 mutex_lock(&disks_mutex);
0909dc44
N		 4585 error = -EEXIST;
4586 if (mddev->gendisk)
4587 goto abort;
efeb53c0
N		 4588
4589 if (name) {
4590 /* Need to ensure that 'name' is not a duplicate.
4591 */
fd01b88c 4592 struct mddev *mddev2;
efeb53c0
N		 4593 spin_lock(&all_mddevs_lock);
4594
4595 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4596 if (mddev2->gendisk &&
4597 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4598 spin_unlock(&all_mddevs_lock);
0909dc44 4599 goto abort;
efeb53c0
N		 4600 }
4601 spin_unlock(&all_mddevs_lock);
1da177e4 4602 }
8b765398 4603
0909dc44 4604 error = -ENOMEM;
8b765398 4605 mddev->queue = blk_alloc_queue(GFP_KERNEL);
0909dc44
N		 4606 if (!mddev->queue)
4607 goto abort;
409c57f3
N		 4608 mddev->queue->queuedata = mddev;
4609
409c57f3 4610 blk_queue_make_request(mddev->queue, md_make_request);
8b765398 4611
1da177e4
LT		 4612 disk = alloc_disk(1 << shift);
4613 if (!disk) {
8b765398
N		 4614 blk_cleanup_queue(mddev->queue);
4615 mddev->queue = NULL;
0909dc44 4616 goto abort;
1da177e4 4617 }
efeb53c0 4618 disk->major = MAJOR(mddev->unit);
1da177e4 4619 disk->first_minor = unit << shift;
efeb53c0
N		 4620 if (name)
4621 strcpy(disk->disk_name, name);
4622 else if (partitioned)
1da177e4 4623 sprintf(disk->disk_name, "md_d%d", unit);
ce7b0f46 4624 else
1da177e4 4625 sprintf(disk->disk_name, "md%d", unit);
1da177e4
LT		 4626 disk->fops = &md_fops;
4627 disk->private_data = mddev;
4628 disk->queue = mddev->queue;
b0140891 4629 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
92850bbd 4630 /* Allow extended partitions. This makes the
d3374825 4631 * 'mdp' device redundant, but we can't really
92850bbd
N		 4632 * remove it now.
4633 */
4634 disk->flags |= GENHD_FL_EXT_DEVT;
1da177e4 4635 mddev->gendisk = disk;
b0140891
N		 4636 /* As soon as we call add_disk(), another thread could get
4637 * through to md_open, so make sure it doesn't get too far
4638 */
4639 mutex_lock(&mddev->open_mutex);
4640 add_disk(disk);
4641
ed9e1982
TH		 4642 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4643 &disk_to_dev(disk)->kobj, "%s", "md");
0909dc44
N		 4644 if (error) {
4645 /* This isn't possible, but as kobject_init_and_add is marked
4646 * __must_check, we must do something with the result
4647 */
5e55e2f5
N		 4648 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4649 disk->disk_name);
0909dc44
N		 4650 error = 0;
4651 }
00bcb4ac
N		 4652 if (mddev->kobj.sd &&
4653 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
43a70507 4654 printk(KERN_DEBUG "pointless warning\n");
b0140891 4655 mutex_unlock(&mddev->open_mutex);
0909dc44
N		 4656 abort:
4657 mutex_unlock(&disks_mutex);
00bcb4ac 4658 if (!error && mddev->kobj.sd) {
3830c62f 4659 kobject_uevent(&mddev->kobj, KOBJ_ADD);
00bcb4ac 4660 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
b62b7590 4661 }
d3374825 4662 mddev_put(mddev);
0909dc44 4663 return error;
efeb53c0
N		 4664}
4665
4666static struct kobject *md_probe(dev_t dev, int *part, void *data)
4667{
4668 md_alloc(dev, NULL);
1da177e4
LT		 4669 return NULL;
4670}
4671
efeb53c0
N		 4672static int add_named_array(const char *val, struct kernel_param *kp)
4673{
4674 /* val must be "md_*" where * is not all digits.
4675 * We allocate an array with a large free minor number, and
4676 * set the name to val. val must not already be an active name.
4677 */
4678 int len = strlen(val);
4679 char buf[DISK_NAME_LEN];
4680
4681 while (len && val[len-1] == '\n')
4682 len--;
4683 if (len >= DISK_NAME_LEN)
4684 return -E2BIG;
4685 strlcpy(buf, val, len+1);
4686 if (strncmp(buf, "md_", 3) != 0)
4687 return -EINVAL;
4688 return md_alloc(0, buf);
4689}
4690
1da177e4
LT		 4691static void md_safemode_timeout(unsigned long data)
4692{
fd01b88c 4693 struct mddev *mddev = (struct mddev *) data;
1da177e4 4694
0fd62b86
NB		 4695 if (!atomic_read(&mddev->writes_pending)) {
4696 mddev->safemode = 1;
4697 if (mddev->external)
00bcb4ac 4698 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86 4699 }
1da177e4
LT		 4700 md_wakeup_thread(mddev->thread);
4701}
4702
6ff8d8ec 4703static int start_dirty_degraded;
1da177e4 4704
fd01b88c 4705int md_run(struct mddev *mddev)
1da177e4 4706{
2604b703 4707 int err;
3cb03002 4708 struct md_rdev *rdev;
84fc4b56 4709 struct md_personality *pers;
1da177e4 4710
a757e64c
N		 4711 if (list_empty(&mddev->disks))
4712 /* cannot run an array with no devices.. */
1da177e4 4713 return -EINVAL;
1da177e4
LT		 4714
4715 if (mddev->pers)
4716 return -EBUSY;
bb4f1e9d
N		 4717 /* Cannot run until previous stop completes properly */
4718 if (mddev->sysfs_active)
4719 return -EBUSY;
b6eb127d 4720
1da177e4
LT		 4721 /*
4722 * Analyze all RAID superblock(s)
4723 */
1ec4a939
N		 4724 if (!mddev->raid_disks) {
4725 if (!mddev->persistent)
4726 return -EINVAL;
a757e64c 4727 analyze_sbs(mddev);
1ec4a939 4728 }
1da177e4 4729
d9d166c2
N		 4730 if (mddev->level != LEVEL_NONE)
4731 request_module("md-level-%d", mddev->level);
4732 else if (mddev->clevel[0])
4733 request_module("md-%s", mddev->clevel);
1da177e4
LT		 4734
4735 /*
4736 * Drop all container device buffers, from now on
4737 * the only valid external interface is through the md
4738 * device.
1da177e4 4739 */
159ec1fc 4740 list_for_each_entry(rdev, &mddev->disks, same_set) {
b2d444d7 4741 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 4742 continue;
4743 sync_blockdev(rdev->bdev);
f98393a6 4744 invalidate_bdev(rdev->bdev);
f0d76d70
N		 4745
4746 /* perform some consistency tests on the device.
4747 * We don't want the data to overlap the metadata,
58c0fed4 4748 * Internal Bitmap issues have been handled elsewhere.
f0d76d70 4749 */
a6ff7e08
JB		 4750 if (rdev->meta_bdev) {
4751 /* Nothing to check */;
4752 } else if (rdev->data_offset < rdev->sb_start) {
58c0fed4
AN		 4753 if (mddev->dev_sectors &&
4754 rdev->data_offset + mddev->dev_sectors
0f420358 4755 > rdev->sb_start) {
f0d76d70
N		 4756 printk("md: %s: data overlaps metadata\n",
4757 mdname(mddev));
4758 return -EINVAL;
4759 }
4760 } else {
0f420358 4761 if (rdev->sb_start + rdev->sb_size/512
f0d76d70
N		 4762 > rdev->data_offset) {
4763 printk("md: %s: metadata overlaps data\n",
4764 mdname(mddev));
4765 return -EINVAL;
4766 }
4767 }
00bcb4ac 4768 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 4769 }
4770
a167f663 4771 if (mddev->bio_set == NULL)
a519b26d 4772 mddev->bio_set = bioset_create(BIO_POOL_SIZE,
fd01b88c 4773 sizeof(struct mddev *));
a167f663 4774
1da177e4 4775 spin_lock(&pers_lock);
d9d166c2 4776 pers = find_pers(mddev->level, mddev->clevel);
2604b703 4777 if (!pers || !try_module_get(pers->owner)) {
1da177e4 4778 spin_unlock(&pers_lock);
d9d166c2
N		 4779 if (mddev->level != LEVEL_NONE)
4780 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4781 mddev->level);
4782 else
4783 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4784 mddev->clevel);
1da177e4
LT		 4785 return -EINVAL;
4786 }
2604b703 4787 mddev->pers = pers;
1da177e4 4788 spin_unlock(&pers_lock);
34817e8c
N		 4789 if (mddev->level != pers->level) {
4790 mddev->level = pers->level;
4791 mddev->new_level = pers->level;
4792 }
d9d166c2 4793 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
1da177e4 4794
f6705578 4795 if (mddev->reshape_position != MaxSector &&
63c70c4f 4796 pers->start_reshape == NULL) {
f6705578
N		 4797 /* This personality cannot handle reshaping... */
4798 mddev->pers = NULL;
4799 module_put(pers->owner);
4800 return -EINVAL;
4801 }
4802
7dd5e7c3
N		 4803 if (pers->sync_request) {
4804 /* Warn if this is a potentially silly
4805 * configuration.
4806 */
4807 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 4808 struct md_rdev *rdev2;
7dd5e7c3 4809 int warned = 0;
159ec1fc
CR		 4810
4811 list_for_each_entry(rdev, &mddev->disks, same_set)
4812 list_for_each_entry(rdev2, &mddev->disks, same_set) {
7dd5e7c3
N		 4813 if (rdev < rdev2 &&
4814 rdev->bdev->bd_contains ==
4815 rdev2->bdev->bd_contains) {
4816 printk(KERN_WARNING
4817 "%s: WARNING: %s appears to be"
4818 " on the same physical disk as"
4819 " %s.\n",
4820 mdname(mddev),
4821 bdevname(rdev->bdev,b),
4822 bdevname(rdev2->bdev,b2));
4823 warned = 1;
4824 }
4825 }
159ec1fc 4826
7dd5e7c3
N		 4827 if (warned)
4828 printk(KERN_WARNING
4829 "True protection against single-disk"
4830 " failure might be compromised.\n");
4831 }
4832
657390d2 4833 mddev->recovery = 0;
58c0fed4
AN		 4834 /* may be over-ridden by personality */
4835 mddev->resync_max_sectors = mddev->dev_sectors;
4836
6ff8d8ec 4837 mddev->ok_start_degraded = start_dirty_degraded;
1da177e4 4838
0f9552b5 4839 if (start_readonly && mddev->ro == 0)
f91de92e
N		 4840 mddev->ro = 2; /* read-only, but switch on first write */
4841
b15c2e57 4842 err = mddev->pers->run(mddev);
13e53df3
AN		 4843 if (err)
4844 printk(KERN_ERR "md: pers->run() failed ...\n");
b522adcd
DW		 4845 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4846 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4847 " but 'external_size' not in effect?\n", __func__);
4848 printk(KERN_ERR
4849 "md: invalid array_size %llu > default size %llu\n",
4850 (unsigned long long)mddev->array_sectors / 2,
4851 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4852 err = -EINVAL;
4853 mddev->pers->stop(mddev);
4854 }
4855 if (err == 0 && mddev->pers->sync_request) {
b15c2e57
N		 4856 err = bitmap_create(mddev);
4857 if (err) {
4858 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4859 mdname(mddev), err);
4860 mddev->pers->stop(mddev);
4861 }
4862 }
1da177e4 4863 if (err) {
1da177e4
LT		 4864 module_put(mddev->pers->owner);
4865 mddev->pers = NULL;
32a7627c
N		 4866 bitmap_destroy(mddev);
4867 return err;
1da177e4 4868 }
5e55e2f5 4869 if (mddev->pers->sync_request) {
00bcb4ac
N		 4870 if (mddev->kobj.sd &&
4871 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5e55e2f5
N		 4872 printk(KERN_WARNING
4873 "md: cannot register extra attributes for %s\n",
4874 mdname(mddev));
00bcb4ac 4875 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5e55e2f5 4876 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
fd9d49ca
N		 4877 mddev->ro = 0;
4878
1da177e4 4879 atomic_set(&mddev->writes_pending,0);
1e50915f
RB		 4880 atomic_set(&mddev->max_corr_read_errors,
4881 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
1da177e4
LT		 4882 mddev->safemode = 0;
4883 mddev->safemode_timer.function = md_safemode_timeout;
4884 mddev->safemode_timer.data = (unsigned long) mddev;
16f17b39 4885 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
1da177e4 4886 mddev->in_sync = 1;
0ca69886
N		 4887 smp_wmb();
4888 mddev->ready = 1;
159ec1fc 4889 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 4890 if (rdev->raid_disk >= 0)
4891 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 4892 /* failure here is OK */;
1da177e4
LT		 4893
4894 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4895
850b2b42
N		 4896 if (mddev->flags)
4897 md_update_sb(mddev, 0);
1da177e4 4898
d7603b7e 4899 md_new_event(mddev);
00bcb4ac
N		 4900 sysfs_notify_dirent_safe(mddev->sysfs_state);
4901 sysfs_notify_dirent_safe(mddev->sysfs_action);
a99ac971 4902 sysfs_notify(&mddev->kobj, NULL, "degraded");
1da177e4
LT		 4903 return 0;
4904}
390ee602 4905EXPORT_SYMBOL_GPL(md_run);
1da177e4 4906
fd01b88c 4907static int do_md_run(struct mddev *mddev)
fe60b014
N		 4908{
4909 int err;
4910
4911 err = md_run(mddev);
4912 if (err)
4913 goto out;
69e51b44
N		 4914 err = bitmap_load(mddev);
4915 if (err) {
4916 bitmap_destroy(mddev);
4917 goto out;
4918 }
0fd018af
JB		 4919
4920 md_wakeup_thread(mddev->thread);
4921 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4922
fe60b014
N		 4923 set_capacity(mddev->gendisk, mddev->array_sectors);
4924 revalidate_disk(mddev->gendisk);
f0b4f7e2 4925 mddev->changed = 1;
fe60b014
N		 4926 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4927out:
4928 return err;
4929}
4930
fd01b88c 4931static int restart_array(struct mddev *mddev)
1da177e4
LT		 4932{
4933 struct gendisk *disk = mddev->gendisk;
1da177e4 4934
80fab1d7 4935 /* Complain if it has no devices */
1da177e4 4936 if (list_empty(&mddev->disks))
80fab1d7
AN		 4937 return -ENXIO;
4938 if (!mddev->pers)
4939 return -EINVAL;
4940 if (!mddev->ro)
4941 return -EBUSY;
4942 mddev->safemode = 0;
4943 mddev->ro = 0;
4944 set_disk_ro(disk, 0);
4945 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4946 mdname(mddev));
4947 /* Kick recovery or resync if necessary */
4948 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4949 md_wakeup_thread(mddev->thread);
4950 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 4951 sysfs_notify_dirent_safe(mddev->sysfs_state);
80fab1d7 4952 return 0;
1da177e4
LT		 4953}
4954
acc55e22
N		 4955/* similar to deny_write_access, but accounts for our holding a reference
4956 * to the file ourselves */
4957static int deny_bitmap_write_access(struct file * file)
4958{
4959 struct inode *inode = file->f_mapping->host;
4960
4961 spin_lock(&inode->i_lock);
4962 if (atomic_read(&inode->i_writecount) > 1) {
4963 spin_unlock(&inode->i_lock);
4964 return -ETXTBSY;
4965 }
4966 atomic_set(&inode->i_writecount, -1);
4967 spin_unlock(&inode->i_lock);
4968
4969 return 0;
4970}
4971
43a70507 4972void restore_bitmap_write_access(struct file *file)
acc55e22
N		 4973{
4974 struct inode *inode = file->f_mapping->host;
4975
4976 spin_lock(&inode->i_lock);
4977 atomic_set(&inode->i_writecount, 1);
4978 spin_unlock(&inode->i_lock);
4979}
4980
fd01b88c 4981static void md_clean(struct mddev *mddev)
6177b472
N		 4982{
4983 mddev->array_sectors = 0;
4984 mddev->external_size = 0;
4985 mddev->dev_sectors = 0;
4986 mddev->raid_disks = 0;
4987 mddev->recovery_cp = 0;
4988 mddev->resync_min = 0;
4989 mddev->resync_max = MaxSector;
4990 mddev->reshape_position = MaxSector;
4991 mddev->external = 0;
4992 mddev->persistent = 0;
4993 mddev->level = LEVEL_NONE;
4994 mddev->clevel[0] = 0;
4995 mddev->flags = 0;
4996 mddev->ro = 0;
4997 mddev->metadata_type[0] = 0;
4998 mddev->chunk_sectors = 0;
4999 mddev->ctime = mddev->utime = 0;
5000 mddev->layout = 0;
5001 mddev->max_disks = 0;
5002 mddev->events = 0;
a8707c08 5003 mddev->can_decrease_events = 0;
6177b472
N		 5004 mddev->delta_disks = 0;
5005 mddev->new_level = LEVEL_NONE;
5006 mddev->new_layout = 0;
5007 mddev->new_chunk_sectors = 0;
5008 mddev->curr_resync = 0;
5009 mddev->resync_mismatches = 0;
5010 mddev->suspend_lo = mddev->suspend_hi = 0;
5011 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5012 mddev->recovery = 0;
5013 mddev->in_sync = 0;
f0b4f7e2 5014 mddev->changed = 0;
6177b472 5015 mddev->degraded = 0;
6177b472
N		 5016 mddev->safemode = 0;
5017 mddev->bitmap_info.offset = 0;
5018 mddev->bitmap_info.default_offset = 0;
5019 mddev->bitmap_info.chunksize = 0;
5020 mddev->bitmap_info.daemon_sleep = 0;
5021 mddev->bitmap_info.max_write_behind = 0;
5022}
5023
fd01b88c 5024static void __md_stop_writes(struct mddev *mddev)
a047e125
N		 5025{
5026 if (mddev->sync_thread) {
5027 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5028 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7ebc0be7 5029 reap_sync_thread(mddev);
a047e125
N		 5030 }
5031
5032 del_timer_sync(&mddev->safemode_timer);
5033
5034 bitmap_flush(mddev);
5035 md_super_wait(mddev);
5036
5037 if (!mddev->in_sync || mddev->flags) {
5038 /* mark array as shutdown cleanly */
5039 mddev->in_sync = 1;
5040 md_update_sb(mddev, 1);
5041 }
5042}
defad61a 5043
fd01b88c 5044void md_stop_writes(struct mddev *mddev)
defad61a
N		 5045{
5046 mddev_lock(mddev);
5047 __md_stop_writes(mddev);
5048 mddev_unlock(mddev);
5049}
390ee602 5050EXPORT_SYMBOL_GPL(md_stop_writes);
a047e125 5051
fd01b88c 5052void md_stop(struct mddev *mddev)
6177b472 5053{
0ca69886 5054 mddev->ready = 0;
6177b472
N		 5055 mddev->pers->stop(mddev);
5056 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5057 mddev->to_remove = &md_redundancy_group;
5058 module_put(mddev->pers->owner);
5059 mddev->pers = NULL;
cca9cf90 5060 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177b472 5061}
390ee602 5062EXPORT_SYMBOL_GPL(md_stop);
6177b472 5063
fd01b88c 5064static int md_set_readonly(struct mddev *mddev, int is_open)
a4bd82d0
N		 5065{
5066 int err = 0;
5067 mutex_lock(&mddev->open_mutex);
5068 if (atomic_read(&mddev->openers) > is_open) {
5069 printk("md: %s still in use.\n",mdname(mddev));
5070 err = -EBUSY;
5071 goto out;
5072 }
5073 if (mddev->pers) {
defad61a 5074 __md_stop_writes(mddev);
a4bd82d0
N		 5075
5076 err = -ENXIO;
5077 if (mddev->ro==1)
5078 goto out;
5079 mddev->ro = 1;
5080 set_disk_ro(mddev->gendisk, 1);
5081 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
00bcb4ac 5082 sysfs_notify_dirent_safe(mddev->sysfs_state);
a4bd82d0
N		 5083 err = 0;
5084 }
5085out:
5086 mutex_unlock(&mddev->open_mutex);
5087 return err;
5088}
5089
9e653b63
N		 5090/* mode:
5091 * 0 - completely stop and dis-assemble array
9e653b63
N		 5092 * 2 - stop but do not disassemble array
5093 */
fd01b88c 5094static int do_md_stop(struct mddev * mddev, int mode, int is_open)
1da177e4 5095{
1da177e4 5096 struct gendisk *disk = mddev->gendisk;
3cb03002 5097 struct md_rdev *rdev;
1da177e4 5098
c8c00a69 5099 mutex_lock(&mddev->open_mutex);
bb4f1e9d
N		 5100 if (atomic_read(&mddev->openers) > is_open ||
5101 mddev->sysfs_active) {
df5b20cf 5102 printk("md: %s still in use.\n",mdname(mddev));
6e17b027
N		 5103 mutex_unlock(&mddev->open_mutex);
5104 return -EBUSY;
5105 }
1da177e4 5106
6e17b027 5107 if (mddev->pers) {
a4bd82d0
N		 5108 if (mddev->ro)
5109 set_disk_ro(disk, 0);
409c57f3 5110
defad61a 5111 __md_stop_writes(mddev);
a4bd82d0
N		 5112 md_stop(mddev);
5113 mddev->queue->merge_bvec_fn = NULL;
a4bd82d0 5114 mddev->queue->backing_dev_info.congested_fn = NULL;
6177b472 5115
a4bd82d0 5116 /* tell userspace to handle 'inactive' */
00bcb4ac 5117 sysfs_notify_dirent_safe(mddev->sysfs_state);
0d4ca600 5118
a4bd82d0 5119 list_for_each_entry(rdev, &mddev->disks, same_set)
36fad858
NK		 5120 if (rdev->raid_disk >= 0)
5121 sysfs_unlink_rdev(mddev, rdev);
c4647292 5122
a4bd82d0 5123 set_capacity(disk, 0);
6e17b027 5124 mutex_unlock(&mddev->open_mutex);
f0b4f7e2 5125 mddev->changed = 1;
a4bd82d0 5126 revalidate_disk(disk);
0d4ca600 5127
a4bd82d0
N		 5128 if (mddev->ro)
5129 mddev->ro = 0;
6e17b027
N		 5130 } else
5131 mutex_unlock(&mddev->open_mutex);
1da177e4
LT		 5132 /*
5133 * Free resources if final stop
5134 */
9e653b63 5135 if (mode == 0) {
1da177e4
LT		 5136 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5137
978f946b 5138 bitmap_destroy(mddev);
c3d9714e
N		 5139 if (mddev->bitmap_info.file) {
5140 restore_bitmap_write_access(mddev->bitmap_info.file);
5141 fput(mddev->bitmap_info.file);
5142 mddev->bitmap_info.file = NULL;
978f946b 5143 }
c3d9714e 5144 mddev->bitmap_info.offset = 0;
978f946b 5145
1da177e4
LT		 5146 export_array(mddev);
5147
6177b472 5148 md_clean(mddev);
934d9c23 5149 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
efeb53c0
N		 5150 if (mddev->hold_active == UNTIL_STOP)
5151 mddev->hold_active = 0;
a4bd82d0 5152 }
3f9d99c1 5153 blk_integrity_unregister(disk);
d7603b7e 5154 md_new_event(mddev);
00bcb4ac 5155 sysfs_notify_dirent_safe(mddev->sysfs_state);
6e17b027 5156 return 0;
1da177e4
LT		 5157}
5158
fdee8ae4 5159#ifndef MODULE
fd01b88c 5160static void autorun_array(struct mddev *mddev)
1da177e4 5161{
3cb03002 5162 struct md_rdev *rdev;
1da177e4
LT		 5163 int err;
5164
a757e64c 5165 if (list_empty(&mddev->disks))
1da177e4 5166 return;
1da177e4
LT		 5167
5168 printk(KERN_INFO "md: running: ");
5169
159ec1fc 5170 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 5171 char b[BDEVNAME_SIZE];
5172 printk("<%s>", bdevname(rdev->bdev,b));
5173 }
5174 printk("\n");
5175
d710e138 5176 err = do_md_run(mddev);
1da177e4
LT		 5177 if (err) {
5178 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
d710e138 5179 do_md_stop(mddev, 0, 0);
1da177e4
LT		 5180 }
5181}
5182
5183/*
5184 * lets try to run arrays based on all disks that have arrived
5185 * until now. (those are in pending_raid_disks)
5186 *
5187 * the method: pick the first pending disk, collect all disks with
5188 * the same UUID, remove all from the pending list and put them into
5189 * the 'same_array' list. Then order this list based on superblock
5190 * update time (freshest comes first), kick out 'old' disks and
5191 * compare superblocks. If everything's fine then run it.
5192 *
5193 * If "unit" is allocated, then bump its reference count
5194 */
5195static void autorun_devices(int part)
5196{
3cb03002 5197 struct md_rdev *rdev0, *rdev, *tmp;
fd01b88c 5198 struct mddev *mddev;
1da177e4
LT		 5199 char b[BDEVNAME_SIZE];
5200
5201 printk(KERN_INFO "md: autorun ...\n");
5202 while (!list_empty(&pending_raid_disks)) {
e8703fe1 5203 int unit;
1da177e4 5204 dev_t dev;
ad01c9e3 5205 LIST_HEAD(candidates);
1da177e4 5206 rdev0 = list_entry(pending_raid_disks.next,
3cb03002 5207 struct md_rdev, same_set);
1da177e4
LT		 5208
5209 printk(KERN_INFO "md: considering %s ...\n",
5210 bdevname(rdev0->bdev,b));
5211 INIT_LIST_HEAD(&candidates);
159ec1fc 5212 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
1da177e4
LT		 5213 if (super_90_load(rdev, rdev0, 0) >= 0) {
5214 printk(KERN_INFO "md: adding %s ...\n",
5215 bdevname(rdev->bdev,b));
5216 list_move(&rdev->same_set, &candidates);
5217 }
5218 /*
5219 * now we have a set of devices, with all of them having
5220 * mostly sane superblocks. It's time to allocate the
5221 * mddev.
5222 */
e8703fe1
N		 5223 if (part) {
5224 dev = MKDEV(mdp_major,
5225 rdev0->preferred_minor << MdpMinorShift);
5226 unit = MINOR(dev) >> MdpMinorShift;
5227 } else {
5228 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5229 unit = MINOR(dev);
5230 }
5231 if (rdev0->preferred_minor != unit) {
1da177e4
LT		 5232 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5233 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5234 break;
5235 }
1da177e4
LT		 5236
5237 md_probe(dev, NULL, NULL);
5238 mddev = mddev_find(dev);
9bbbca3a
NB		 5239 if (!mddev || !mddev->gendisk) {
5240 if (mddev)
5241 mddev_put(mddev);
5242 printk(KERN_ERR
1da177e4
LT		 5243 "md: cannot allocate memory for md drive.\n");
5244 break;
5245 }
5246 if (mddev_lock(mddev))
5247 printk(KERN_WARNING "md: %s locked, cannot run\n",
5248 mdname(mddev));
5249 else if (mddev->raid_disks || mddev->major_version
5250 || !list_empty(&mddev->disks)) {
5251 printk(KERN_WARNING
5252 "md: %s already running, cannot run %s\n",
5253 mdname(mddev), bdevname(rdev0->bdev,b));
5254 mddev_unlock(mddev);
5255 } else {
5256 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1ec4a939 5257 mddev->persistent = 1;
159ec1fc 5258 rdev_for_each_list(rdev, tmp, &candidates) {
1da177e4
LT		 5259 list_del_init(&rdev->same_set);
5260 if (bind_rdev_to_array(rdev, mddev))
5261 export_rdev(rdev);
5262 }
5263 autorun_array(mddev);
5264 mddev_unlock(mddev);
5265 }
5266 /* on success, candidates will be empty, on error
5267 * it won't...
5268 */
159ec1fc 5269 rdev_for_each_list(rdev, tmp, &candidates) {
4b80991c 5270 list_del_init(&rdev->same_set);
1da177e4 5271 export_rdev(rdev);
4b80991c 5272 }
1da177e4
LT		 5273 mddev_put(mddev);
5274 }
5275 printk(KERN_INFO "md: ... autorun DONE.\n");
5276}
fdee8ae4 5277#endif /* !MODULE */
1da177e4 5278
1da177e4
LT		 5279static int get_version(void __user * arg)
5280{
5281 mdu_version_t ver;
5282
5283 ver.major = MD_MAJOR_VERSION;
5284 ver.minor = MD_MINOR_VERSION;
5285 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5286
5287 if (copy_to_user(arg, &ver, sizeof(ver)))
5288 return -EFAULT;
5289
5290 return 0;
5291}
5292
fd01b88c 5293static int get_array_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5294{
5295 mdu_array_info_t info;
a9f326eb 5296 int nr,working,insync,failed,spare;
3cb03002 5297 struct md_rdev *rdev;
1da177e4 5298
a9f326eb 5299 nr=working=insync=failed=spare=0;
159ec1fc 5300 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 5301 nr++;
b2d444d7 5302 if (test_bit(Faulty, &rdev->flags))
1da177e4
LT		 5303 failed++;
5304 else {
5305 working++;
b2d444d7 5306 if (test_bit(In_sync, &rdev->flags))
a9f326eb 5307 insync++;
1da177e4
LT		 5308 else
5309 spare++;
5310 }
5311 }
5312
5313 info.major_version = mddev->major_version;
5314 info.minor_version = mddev->minor_version;
5315 info.patch_version = MD_PATCHLEVEL_VERSION;
5316 info.ctime = mddev->ctime;
5317 info.level = mddev->level;
58c0fed4
AN		 5318 info.size = mddev->dev_sectors / 2;
5319 if (info.size != mddev->dev_sectors / 2) /* overflow */
284ae7ca 5320 info.size = -1;
1da177e4
LT		 5321 info.nr_disks = nr;
5322 info.raid_disks = mddev->raid_disks;
5323 info.md_minor = mddev->md_minor;
5324 info.not_persistent= !mddev->persistent;
5325
5326 info.utime = mddev->utime;
5327 info.state = 0;
5328 if (mddev->in_sync)
5329 info.state = (1<<MD_SB_CLEAN);
c3d9714e 5330 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5331 info.state = (1<<MD_SB_BITMAP_PRESENT);
a9f326eb 5332 info.active_disks = insync;
1da177e4
LT		 5333 info.working_disks = working;
5334 info.failed_disks = failed;
5335 info.spare_disks = spare;
5336
5337 info.layout = mddev->layout;
9d8f0363 5338 info.chunk_size = mddev->chunk_sectors << 9;
1da177e4
LT		 5339
5340 if (copy_to_user(arg, &info, sizeof(info)))
5341 return -EFAULT;
5342
5343 return 0;
5344}
5345
fd01b88c 5346static int get_bitmap_file(struct mddev * mddev, void __user * arg)
32a7627c
N		 5347{
5348 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5349 char *ptr, *buf = NULL;
5350 int err = -ENOMEM;
5351
b5470dc5
DW		 5352 if (md_allow_write(mddev))
5353 file = kmalloc(sizeof(*file), GFP_NOIO);
5354 else
5355 file = kmalloc(sizeof(*file), GFP_KERNEL);
2a2275d6 5356
32a7627c
N		 5357 if (!file)
5358 goto out;
5359
5360 /* bitmap disabled, zero the first byte and copy out */
5361 if (!mddev->bitmap || !mddev->bitmap->file) {
5362 file->pathname[0] = '\0';
5363 goto copy_out;
5364 }
5365
5366 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5367 if (!buf)
5368 goto out;
5369
6bcfd601
CH		 5370 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
5371 if (IS_ERR(ptr))
32a7627c
N		 5372 goto out;
5373
5374 strcpy(file->pathname, ptr);
5375
5376copy_out:
5377 err = 0;
5378 if (copy_to_user(arg, file, sizeof(*file)))
5379 err = -EFAULT;
5380out:
5381 kfree(buf);
5382 kfree(file);
5383 return err;
5384}
5385
fd01b88c 5386static int get_disk_info(struct mddev * mddev, void __user * arg)
1da177e4
LT		 5387{
5388 mdu_disk_info_t info;
3cb03002 5389 struct md_rdev *rdev;
1da177e4
LT		 5390
5391 if (copy_from_user(&info, arg, sizeof(info)))
5392 return -EFAULT;
5393
26ef379f 5394 rdev = find_rdev_nr(mddev, info.number);
1da177e4
LT		 5395 if (rdev) {
5396 info.major = MAJOR(rdev->bdev->bd_dev);
5397 info.minor = MINOR(rdev->bdev->bd_dev);
5398 info.raid_disk = rdev->raid_disk;
5399 info.state = 0;
b2d444d7 5400 if (test_bit(Faulty, &rdev->flags))
1da177e4 5401 info.state |= (1<<MD_DISK_FAULTY);
b2d444d7 5402 else if (test_bit(In_sync, &rdev->flags)) {
1da177e4
LT		 5403 info.state |= (1<<MD_DISK_ACTIVE);
5404 info.state |= (1<<MD_DISK_SYNC);
5405 }
8ddf9efe
N		 5406 if (test_bit(WriteMostly, &rdev->flags))
5407 info.state |= (1<<MD_DISK_WRITEMOSTLY);
1da177e4
LT		 5408 } else {
5409 info.major = info.minor = 0;
5410 info.raid_disk = -1;
5411 info.state = (1<<MD_DISK_REMOVED);
5412 }
5413
5414 if (copy_to_user(arg, &info, sizeof(info)))
5415 return -EFAULT;
5416
5417 return 0;
5418}
5419
fd01b88c 5420static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
1da177e4
LT		 5421{
5422 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3cb03002 5423 struct md_rdev *rdev;
1da177e4
LT		 5424 dev_t dev = MKDEV(info->major,info->minor);
5425
5426 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5427 return -EOVERFLOW;
5428
5429 if (!mddev->raid_disks) {
5430 int err;
5431 /* expecting a device which has a superblock */
5432 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5433 if (IS_ERR(rdev)) {
5434 printk(KERN_WARNING
5435 "md: md_import_device returned %ld\n",
5436 PTR_ERR(rdev));
5437 return PTR_ERR(rdev);
5438 }
5439 if (!list_empty(&mddev->disks)) {
3cb03002
N		 5440 struct md_rdev *rdev0
5441 = list_entry(mddev->disks.next,
5442 struct md_rdev, same_set);
a9f326eb 5443 err = super_types[mddev->major_version]
1da177e4
LT		 5444 .load_super(rdev, rdev0, mddev->minor_version);
5445 if (err < 0) {
5446 printk(KERN_WARNING
5447 "md: %s has different UUID to %s\n",
5448 bdevname(rdev->bdev,b),
5449 bdevname(rdev0->bdev,b2));
5450 export_rdev(rdev);
5451 return -EINVAL;
5452 }
5453 }
5454 err = bind_rdev_to_array(rdev, mddev);
5455 if (err)
5456 export_rdev(rdev);
5457 return err;
5458 }
5459
5460 /*
5461 * add_new_disk can be used once the array is assembled
5462 * to add "hot spares". They must already have a superblock
5463 * written
5464 */
5465 if (mddev->pers) {
5466 int err;
5467 if (!mddev->pers->hot_add_disk) {
5468 printk(KERN_WARNING
5469 "%s: personality does not support diskops!\n",
5470 mdname(mddev));
5471 return -EINVAL;
5472 }
7b1e35f6
N		 5473 if (mddev->persistent)
5474 rdev = md_import_device(dev, mddev->major_version,
5475 mddev->minor_version);
5476 else
5477 rdev = md_import_device(dev, -1, -1);
1da177e4
LT		 5478 if (IS_ERR(rdev)) {
5479 printk(KERN_WARNING
5480 "md: md_import_device returned %ld\n",
5481 PTR_ERR(rdev));
5482 return PTR_ERR(rdev);
5483 }
1a855a06 5484 /* set saved_raid_disk if appropriate */
41158c7e
N		 5485 if (!mddev->persistent) {
5486 if (info->state & (1<<MD_DISK_SYNC) &&
bf572541 5487 info->raid_disk < mddev->raid_disks) {
41158c7e 5488 rdev->raid_disk = info->raid_disk;
bf572541
N		 5489 set_bit(In_sync, &rdev->flags);
5490 } else
41158c7e
N		 5491 rdev->raid_disk = -1;
5492 } else
5493 super_types[mddev->major_version].
5494 validate_super(mddev, rdev);
bedd86b7
N		 5495 if ((info->state & (1<<MD_DISK_SYNC)) &&
5496 (!test_bit(In_sync, &rdev->flags) ||
5497 rdev->raid_disk != info->raid_disk)) {
5498 /* This was a hot-add request, but events doesn't
5499 * match, so reject it.
5500 */
5501 export_rdev(rdev);
5502 return -EINVAL;
5503 }
5504
1a855a06
N		 5505 if (test_bit(In_sync, &rdev->flags))
5506 rdev->saved_raid_disk = rdev->raid_disk;
5507 else
5508 rdev->saved_raid_disk = -1;
41158c7e 5509
b2d444d7 5510 clear_bit(In_sync, &rdev->flags); /* just to be sure */
8ddf9efe
N		 5511 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5512 set_bit(WriteMostly, &rdev->flags);
575a80fa
N		 5513 else
5514 clear_bit(WriteMostly, &rdev->flags);
8ddf9efe 5515
1da177e4
LT		 5516 rdev->raid_disk = -1;
5517 err = bind_rdev_to_array(rdev, mddev);
7c7546cc
N		 5518 if (!err && !mddev->pers->hot_remove_disk) {
5519 /* If there is hot_add_disk but no hot_remove_disk
5520 * then added disks for geometry changes,
5521 * and should be added immediately.
5522 */
5523 super_types[mddev->major_version].
5524 validate_super(mddev, rdev);
5525 err = mddev->pers->hot_add_disk(mddev, rdev);
5526 if (err)
5527 unbind_rdev_from_array(rdev);
5528 }
1da177e4
LT		 5529 if (err)
5530 export_rdev(rdev);
52664732 5531 else
00bcb4ac 5532 sysfs_notify_dirent_safe(rdev->sysfs_state);
c361777f 5533
17571284 5534 md_update_sb(mddev, 1);
72a23c21
NB		 5535 if (mddev->degraded)
5536 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
c361777f 5537 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9864c005 5538 if (!err)
5539 md_new_event(mddev);
005eca5e 5540 md_wakeup_thread(mddev->thread);
1da177e4
LT		 5541 return err;
5542 }
5543
5544 /* otherwise, add_new_disk is only allowed
5545 * for major_version==0 superblocks
5546 */
5547 if (mddev->major_version != 0) {
5548 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5549 mdname(mddev));
5550 return -EINVAL;
5551 }
5552
5553 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5554 int err;
d710e138 5555 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5556 if (IS_ERR(rdev)) {
5557 printk(KERN_WARNING
5558 "md: error, md_import_device() returned %ld\n",
5559 PTR_ERR(rdev));
5560 return PTR_ERR(rdev);
5561 }
5562 rdev->desc_nr = info->number;
5563 if (info->raid_disk < mddev->raid_disks)
5564 rdev->raid_disk = info->raid_disk;
5565 else
5566 rdev->raid_disk = -1;
5567
1da177e4 5568 if (rdev->raid_disk < mddev->raid_disks)
b2d444d7
N		 5569 if (info->state & (1<<MD_DISK_SYNC))
5570 set_bit(In_sync, &rdev->flags);
1da177e4 5571
8ddf9efe
N		 5572 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5573 set_bit(WriteMostly, &rdev->flags);
5574
1da177e4
LT		 5575 if (!mddev->persistent) {
5576 printk(KERN_INFO "md: nonpersistent superblock ...\n");
77304d2a
MS		 5577 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5578 } else
57b2caa3 5579 rdev->sb_start = calc_dev_sboffset(rdev);
8190e754 5580 rdev->sectors = rdev->sb_start;
1da177e4 5581
2bf071bf
N		 5582 err = bind_rdev_to_array(rdev, mddev);
5583 if (err) {
5584 export_rdev(rdev);
5585 return err;
5586 }
1da177e4
LT		 5587 }
5588
5589 return 0;
5590}
5591
fd01b88c 5592static int hot_remove_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5593{
5594 char b[BDEVNAME_SIZE];
3cb03002 5595 struct md_rdev *rdev;
1da177e4 5596
1da177e4
LT		 5597 rdev = find_rdev(mddev, dev);
5598 if (!rdev)
5599 return -ENXIO;
5600
5601 if (rdev->raid_disk >= 0)
5602 goto busy;
5603
5604 kick_rdev_from_array(rdev);
850b2b42 5605 md_update_sb(mddev, 1);
d7603b7e 5606 md_new_event(mddev);
1da177e4
LT		 5607
5608 return 0;
5609busy:
fdefa4d8 5610 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
1da177e4
LT		 5611 bdevname(rdev->bdev,b), mdname(mddev));
5612 return -EBUSY;
5613}
5614
fd01b88c 5615static int hot_add_disk(struct mddev * mddev, dev_t dev)
1da177e4
LT		 5616{
5617 char b[BDEVNAME_SIZE];
5618 int err;
3cb03002 5619 struct md_rdev *rdev;
1da177e4
LT		 5620
5621 if (!mddev->pers)
5622 return -ENODEV;
5623
5624 if (mddev->major_version != 0) {
5625 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5626 " version-0 superblocks.\n",
5627 mdname(mddev));
5628 return -EINVAL;
5629 }
5630 if (!mddev->pers->hot_add_disk) {
5631 printk(KERN_WARNING
5632 "%s: personality does not support diskops!\n",
5633 mdname(mddev));
5634 return -EINVAL;
5635 }
5636
d710e138 5637 rdev = md_import_device(dev, -1, 0);
1da177e4
LT		 5638 if (IS_ERR(rdev)) {
5639 printk(KERN_WARNING
5640 "md: error, md_import_device() returned %ld\n",
5641 PTR_ERR(rdev));
5642 return -EINVAL;
5643 }
5644
5645 if (mddev->persistent)
57b2caa3 5646 rdev->sb_start = calc_dev_sboffset(rdev);
1da177e4 5647 else
77304d2a 5648 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
1da177e4 5649
8190e754 5650 rdev->sectors = rdev->sb_start;
1da177e4 5651
b2d444d7 5652 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 5653 printk(KERN_WARNING
5654 "md: can not hot-add faulty %s disk to %s!\n",
5655 bdevname(rdev->bdev,b), mdname(mddev));
5656 err = -EINVAL;
5657 goto abort_export;
5658 }
b2d444d7 5659 clear_bit(In_sync, &rdev->flags);
1da177e4 5660 rdev->desc_nr = -1;
5842730d 5661 rdev->saved_raid_disk = -1;
2bf071bf
N		 5662 err = bind_rdev_to_array(rdev, mddev);
5663 if (err)
5664 goto abort_export;
1da177e4
LT		 5665
5666 /*
5667 * The rest should better be atomic, we can have disk failures
5668 * noticed in interrupt contexts ...
5669 */
5670
1da177e4
LT		 5671 rdev->raid_disk = -1;
5672
850b2b42 5673 md_update_sb(mddev, 1);
1da177e4
LT		 5674
5675 /*
5676 * Kick recovery, maybe this spare has to be added to the
5677 * array immediately.
5678 */
5679 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5680 md_wakeup_thread(mddev->thread);
d7603b7e 5681 md_new_event(mddev);
1da177e4
LT		 5682 return 0;
5683
1da177e4
LT		 5684abort_export:
5685 export_rdev(rdev);
5686 return err;
5687}
5688
fd01b88c 5689static int set_bitmap_file(struct mddev *mddev, int fd)
32a7627c
N		 5690{
5691 int err;
5692
36fa3063
N		 5693 if (mddev->pers) {
5694 if (!mddev->pers->quiesce)
5695 return -EBUSY;
5696 if (mddev->recovery || mddev->sync_thread)
5697 return -EBUSY;
5698 /* we should be able to change the bitmap.. */
5699 }
32a7627c 5700
32a7627c 5701
36fa3063
N		 5702 if (fd >= 0) {
5703 if (mddev->bitmap)
5704 return -EEXIST; /* cannot add when bitmap is present */
c3d9714e 5705 mddev->bitmap_info.file = fget(fd);
32a7627c 5706
c3d9714e 5707 if (mddev->bitmap_info.file == NULL) {
36fa3063
N		 5708 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5709 mdname(mddev));
5710 return -EBADF;
5711 }
5712
c3d9714e 5713 err = deny_bitmap_write_access(mddev->bitmap_info.file);
36fa3063
N		 5714 if (err) {
5715 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5716 mdname(mddev));
c3d9714e
N		 5717 fput(mddev->bitmap_info.file);
5718 mddev->bitmap_info.file = NULL;
36fa3063
N		 5719 return err;
5720 }
c3d9714e 5721 mddev->bitmap_info.offset = 0; /* file overrides offset */
36fa3063
N		 5722 } else if (mddev->bitmap == NULL)
5723 return -ENOENT; /* cannot remove what isn't there */
5724 err = 0;
5725 if (mddev->pers) {
5726 mddev->pers->quiesce(mddev, 1);
69e51b44 5727 if (fd >= 0) {
36fa3063 5728 err = bitmap_create(mddev);
69e51b44
N		 5729 if (!err)
5730 err = bitmap_load(mddev);
5731 }
d7375ab3 5732 if (fd < 0 || err) {
36fa3063 5733 bitmap_destroy(mddev);
d7375ab3
N		 5734 fd = -1; /* make sure to put the file */
5735 }
36fa3063 5736 mddev->pers->quiesce(mddev, 0);
d7375ab3
N		 5737 }
5738 if (fd < 0) {
c3d9714e
N		 5739 if (mddev->bitmap_info.file) {
5740 restore_bitmap_write_access(mddev->bitmap_info.file);
5741 fput(mddev->bitmap_info.file);
acc55e22 5742 }
c3d9714e 5743 mddev->bitmap_info.file = NULL;
36fa3063
N		 5744 }
5745
32a7627c
N		 5746 return err;
5747}
5748
1da177e4
LT		 5749/*
5750 * set_array_info is used two different ways
5751 * The original usage is when creating a new array.
5752 * In this usage, raid_disks is > 0 and it together with
5753 * level, size, not_persistent,layout,chunksize determine the
5754 * shape of the array.
5755 * This will always create an array with a type-0.90.0 superblock.
5756 * The newer usage is when assembling an array.
5757 * In this case raid_disks will be 0, and the major_version field is
5758 * use to determine which style super-blocks are to be found on the devices.
5759 * The minor and patch _version numbers are also kept incase the
5760 * super_block handler wishes to interpret them.
5761 */
fd01b88c 5762static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
1da177e4
LT		 5763{
5764
5765 if (info->raid_disks == 0) {
5766 /* just setting version number for superblock loading */
5767 if (info->major_version < 0 ||
50511da3 5768 info->major_version >= ARRAY_SIZE(super_types) ||
1da177e4
LT		 5769 super_types[info->major_version].name == NULL) {
5770 /* maybe try to auto-load a module? */
5771 printk(KERN_INFO
5772 "md: superblock version %d not known\n",
5773 info->major_version);
5774 return -EINVAL;
5775 }
5776 mddev->major_version = info->major_version;
5777 mddev->minor_version = info->minor_version;
5778 mddev->patch_version = info->patch_version;
3f9d7b0d 5779 mddev->persistent = !info->not_persistent;
cbd19983
N		 5780 /* ensure mddev_put doesn't delete this now that there
5781 * is some minimal configuration.
5782 */
5783 mddev->ctime = get_seconds();
1da177e4
LT		 5784 return 0;
5785 }
5786 mddev->major_version = MD_MAJOR_VERSION;
5787 mddev->minor_version = MD_MINOR_VERSION;
5788 mddev->patch_version = MD_PATCHLEVEL_VERSION;
5789 mddev->ctime = get_seconds();
5790
5791 mddev->level = info->level;
17115e03 5792 mddev->clevel[0] = 0;
58c0fed4 5793 mddev->dev_sectors = 2 * (sector_t)info->size;
1da177e4
LT		 5794 mddev->raid_disks = info->raid_disks;
5795 /* don't set md_minor, it is determined by which /dev/md* was
5796 * openned
5797 */
5798 if (info->state & (1<<MD_SB_CLEAN))
5799 mddev->recovery_cp = MaxSector;
5800 else
5801 mddev->recovery_cp = 0;
5802 mddev->persistent = ! info->not_persistent;
e691063a 5803 mddev->external = 0;
1da177e4
LT		 5804
5805 mddev->layout = info->layout;
9d8f0363 5806 mddev->chunk_sectors = info->chunk_size >> 9;
1da177e4
LT		 5807
5808 mddev->max_disks = MD_SB_DISKS;
5809
e691063a
N		 5810 if (mddev->persistent)
5811 mddev->flags = 0;
850b2b42 5812 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 5813
c3d9714e
N		 5814 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
5815 mddev->bitmap_info.offset = 0;
b2a2703c 5816
f6705578
N		 5817 mddev->reshape_position = MaxSector;
5818
1da177e4
LT		 5819 /*
5820 * Generate a 128 bit UUID
5821 */
5822 get_random_bytes(mddev->uuid, 16);
5823
f6705578 5824 mddev->new_level = mddev->level;
664e7c41 5825 mddev->new_chunk_sectors = mddev->chunk_sectors;
f6705578
N		 5826 mddev->new_layout = mddev->layout;
5827 mddev->delta_disks = 0;
5828
1da177e4
LT		 5829 return 0;
5830}
5831
fd01b88c 5832void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
1f403624 5833{
b522adcd
DW		 5834 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5835
5836 if (mddev->external_size)
5837 return;
5838
1f403624
DW		 5839 mddev->array_sectors = array_sectors;
5840}
5841EXPORT_SYMBOL(md_set_array_sectors);
5842
fd01b88c 5843static int update_size(struct mddev *mddev, sector_t num_sectors)
a35b0d69 5844{
3cb03002 5845 struct md_rdev *rdev;
a35b0d69 5846 int rv;
d71f9f88 5847 int fit = (num_sectors == 0);
a35b0d69
N		 5848
5849 if (mddev->pers->resize == NULL)
5850 return -EINVAL;
d71f9f88
AN		 5851 /* The "num_sectors" is the number of sectors of each device that
5852 * is used. This can only make sense for arrays with redundancy.
5853 * linear and raid0 always use whatever space is available. We can only
5854 * consider changing this number if no resync or reconstruction is
5855 * happening, and if the new size is acceptable. It must fit before the
0f420358 5856 * sb_start or, if that is <data_offset, it must fit before the size
d71f9f88
AN		 5857 * of each device. If num_sectors is zero, we find the largest size
5858 * that fits.
a35b0d69
N		 5859 */
5860 if (mddev->sync_thread)
5861 return -EBUSY;
dba034ee
N		 5862 if (mddev->bitmap)
5863 /* Sorry, cannot grow a bitmap yet, just remove it,
5864 * grow, and re-add.
5865 */
5866 return -EBUSY;
159ec1fc 5867 list_for_each_entry(rdev, &mddev->disks, same_set) {
dd8ac336 5868 sector_t avail = rdev->sectors;
01ab5662 5869
d71f9f88
AN		 5870 if (fit && (num_sectors == 0 || num_sectors > avail))
5871 num_sectors = avail;
5872 if (avail < num_sectors)
a35b0d69
N		 5873 return -ENOSPC;
5874 }
d71f9f88 5875 rv = mddev->pers->resize(mddev, num_sectors);
449aad3e
N		 5876 if (!rv)
5877 revalidate_disk(mddev->gendisk);
a35b0d69
N		 5878 return rv;
5879}
5880
fd01b88c 5881static int update_raid_disks(struct mddev *mddev, int raid_disks)
da943b99
N		 5882{
5883 int rv;
5884 /* change the number of raid disks */
63c70c4f 5885 if (mddev->pers->check_reshape == NULL)
da943b99
N		 5886 return -EINVAL;
5887 if (raid_disks <= 0 ||
233fca36 5888 (mddev->max_disks && raid_disks >= mddev->max_disks))
da943b99 5889 return -EINVAL;
63c70c4f 5890 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
da943b99 5891 return -EBUSY;
63c70c4f
N		 5892 mddev->delta_disks = raid_disks - mddev->raid_disks;
5893
5894 rv = mddev->pers->check_reshape(mddev);
de171cb9
N		 5895 if (rv < 0)
5896 mddev->delta_disks = 0;
da943b99
N		 5897 return rv;
5898}
5899
5900
1da177e4
LT		 5901/*
5902 * update_array_info is used to change the configuration of an
5903 * on-line array.
5904 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5905 * fields in the info are checked against the array.
5906 * Any differences that cannot be handled will cause an error.
5907 * Normally, only one change can be managed at a time.
5908 */
fd01b88c 5909static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
1da177e4
LT		 5910{
5911 int rv = 0;
5912 int cnt = 0;
36fa3063
N		 5913 int state = 0;
5914
5915 /* calculate expected state,ignoring low bits */
c3d9714e 5916 if (mddev->bitmap && mddev->bitmap_info.offset)
36fa3063 5917 state |= (1 << MD_SB_BITMAP_PRESENT);
1da177e4
LT		 5918
5919 if (mddev->major_version != info->major_version ||
5920 mddev->minor_version != info->minor_version ||
5921/* mddev->patch_version != info->patch_version || */
5922 mddev->ctime != info->ctime ||
5923 mddev->level != info->level ||
5924/* mddev->layout != info->layout || */
5925 !mddev->persistent != info->not_persistent||
9d8f0363 5926 mddev->chunk_sectors != info->chunk_size >> 9 ||
36fa3063
N		 5927 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5928 ((state^info->state) & 0xfffffe00)
5929 )
1da177e4
LT		 5930 return -EINVAL;
5931 /* Check there is only one change */
58c0fed4
AN		 5932 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5933 cnt++;
5934 if (mddev->raid_disks != info->raid_disks)
5935 cnt++;
5936 if (mddev->layout != info->layout)
5937 cnt++;
5938 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5939 cnt++;
5940 if (cnt == 0)
5941 return 0;
5942 if (cnt > 1)
5943 return -EINVAL;
1da177e4
LT		 5944
5945 if (mddev->layout != info->layout) {
5946 /* Change layout
5947 * we don't need to do anything at the md level, the
5948 * personality will take care of it all.
5949 */
50ac168a 5950 if (mddev->pers->check_reshape == NULL)
1da177e4 5951 return -EINVAL;
597a711b
N		 5952 else {
5953 mddev->new_layout = info->layout;
50ac168a 5954 rv = mddev->pers->check_reshape(mddev);
597a711b
N		 5955 if (rv)
5956 mddev->new_layout = mddev->layout;
5957 return rv;
5958 }
1da177e4 5959 }
58c0fed4 5960 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
d71f9f88 5961 rv = update_size(mddev, (sector_t)info->size * 2);
a35b0d69 5962
da943b99
N		 5963 if (mddev->raid_disks != info->raid_disks)
5964 rv = update_raid_disks(mddev, info->raid_disks);
5965
36fa3063
N		 5966 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5967 if (mddev->pers->quiesce == NULL)
5968 return -EINVAL;
5969 if (mddev->recovery || mddev->sync_thread)
5970 return -EBUSY;
5971 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5972 /* add the bitmap */
5973 if (mddev->bitmap)
5974 return -EEXIST;
c3d9714e 5975 if (mddev->bitmap_info.default_offset == 0)
36fa3063 5976 return -EINVAL;
c3d9714e
N		 5977 mddev->bitmap_info.offset =
5978 mddev->bitmap_info.default_offset;
36fa3063
N		 5979 mddev->pers->quiesce(mddev, 1);
5980 rv = bitmap_create(mddev);
69e51b44
N		 5981 if (!rv)
5982 rv = bitmap_load(mddev);
36fa3063
N		 5983 if (rv)
5984 bitmap_destroy(mddev);
5985 mddev->pers->quiesce(mddev, 0);
5986 } else {
5987 /* remove the bitmap */
5988 if (!mddev->bitmap)
5989 return -ENOENT;
5990 if (mddev->bitmap->file)
5991 return -EINVAL;
5992 mddev->pers->quiesce(mddev, 1);
5993 bitmap_destroy(mddev);
5994 mddev->pers->quiesce(mddev, 0);
c3d9714e 5995 mddev->bitmap_info.offset = 0;
36fa3063
N		 5996 }
5997 }
850b2b42 5998 md_update_sb(mddev, 1);
1da177e4
LT		 5999 return rv;
6000}
6001
fd01b88c 6002static int set_disk_faulty(struct mddev *mddev, dev_t dev)
1da177e4 6003{
3cb03002 6004 struct md_rdev *rdev;
1da177e4
LT		 6005
6006 if (mddev->pers == NULL)
6007 return -ENODEV;
6008
6009 rdev = find_rdev(mddev, dev);
6010 if (!rdev)
6011 return -ENODEV;
6012
6013 md_error(mddev, rdev);
5ef56c8f
N		 6014 if (!test_bit(Faulty, &rdev->flags))
6015 return -EBUSY;
1da177e4
LT		 6016 return 0;
6017}
6018
2f9618ce
AN		 6019/*
6020 * We have a problem here : there is no easy way to give a CHS
6021 * virtual geometry. We currently pretend that we have a 2 heads
6022 * 4 sectors (with a BIG number of cylinders...). This drives
6023 * dosfs just mad... ;-)
6024 */
a885c8c4
CH		 6025static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6026{
fd01b88c 6027 struct mddev *mddev = bdev->bd_disk->private_data;
a885c8c4
CH		 6028
6029 geo->heads = 2;
6030 geo->sectors = 4;
49ce6cea 6031 geo->cylinders = mddev->array_sectors / 8;
a885c8c4
CH		 6032 return 0;
6033}
6034
a39907fa 6035static int md_ioctl(struct block_device *bdev, fmode_t mode,
1da177e4
LT		 6036 unsigned int cmd, unsigned long arg)
6037{
6038 int err = 0;
6039 void __user *argp = (void __user *)arg;
fd01b88c 6040 struct mddev *mddev = NULL;
e2218350 6041 int ro;
1da177e4
LT		 6042
6043 if (!capable(CAP_SYS_ADMIN))
6044 return -EACCES;
6045
6046 /*
6047 * Commands dealing with the RAID driver but not any
6048 * particular array:
6049 */
6050 switch (cmd)
6051 {
6052 case RAID_VERSION:
6053 err = get_version(argp);
6054 goto done;
6055
6056 case PRINT_RAID_DEBUG:
6057 err = 0;
6058 md_print_devices();
6059 goto done;
6060
6061#ifndef MODULE
6062 case RAID_AUTORUN:
6063 err = 0;
6064 autostart_arrays(arg);
6065 goto done;
6066#endif
6067 default:;
6068 }
6069
6070 /*
6071 * Commands creating/starting a new array:
6072 */
6073
a39907fa 6074 mddev = bdev->bd_disk->private_data;
1da177e4
LT		 6075
6076 if (!mddev) {
6077 BUG();
6078 goto abort;
6079 }
6080
1da177e4
LT		 6081 err = mddev_lock(mddev);
6082 if (err) {
6083 printk(KERN_INFO
6084 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6085 err, cmd);
6086 goto abort;
6087 }
6088
6089 switch (cmd)
6090 {
6091 case SET_ARRAY_INFO:
6092 {
6093 mdu_array_info_t info;
6094 if (!arg)
6095 memset(&info, 0, sizeof(info));
6096 else if (copy_from_user(&info, argp, sizeof(info))) {
6097 err = -EFAULT;
6098 goto abort_unlock;
6099 }
6100 if (mddev->pers) {
6101 err = update_array_info(mddev, &info);
6102 if (err) {
6103 printk(KERN_WARNING "md: couldn't update"
6104 " array info. %d\n", err);
6105 goto abort_unlock;
6106 }
6107 goto done_unlock;
6108 }
6109 if (!list_empty(&mddev->disks)) {
6110 printk(KERN_WARNING
6111 "md: array %s already has disks!\n",
6112 mdname(mddev));
6113 err = -EBUSY;
6114 goto abort_unlock;
6115 }
6116 if (mddev->raid_disks) {
6117 printk(KERN_WARNING
6118 "md: array %s already initialised!\n",
6119 mdname(mddev));
6120 err = -EBUSY;
6121 goto abort_unlock;
6122 }
6123 err = set_array_info(mddev, &info);
6124 if (err) {
6125 printk(KERN_WARNING "md: couldn't set"
6126 " array info. %d\n", err);
6127 goto abort_unlock;
6128 }
6129 }
6130 goto done_unlock;
6131
6132 default:;
6133 }
6134
6135 /*
6136 * Commands querying/configuring an existing array:
6137 */
32a7627c 6138 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3f9d7b0d 6139 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
a17184a9
N		 6140 if ((!mddev->raid_disks && !mddev->external)
6141 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6142 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6143 && cmd != GET_BITMAP_FILE) {
1da177e4
LT		 6144 err = -ENODEV;
6145 goto abort_unlock;
6146 }
6147
6148 /*
6149 * Commands even a read-only array can execute:
6150 */
6151 switch (cmd)
6152 {
6153 case GET_ARRAY_INFO:
6154 err = get_array_info(mddev, argp);
6155 goto done_unlock;
6156
32a7627c 6157 case GET_BITMAP_FILE:
87162a28 6158 err = get_bitmap_file(mddev, argp);
32a7627c
N		 6159 goto done_unlock;
6160
1da177e4
LT		 6161 case GET_DISK_INFO:
6162 err = get_disk_info(mddev, argp);
6163 goto done_unlock;
6164
6165 case RESTART_ARRAY_RW:
6166 err = restart_array(mddev);
6167 goto done_unlock;
6168
6169 case STOP_ARRAY:
d710e138 6170 err = do_md_stop(mddev, 0, 1);
1da177e4
LT		 6171 goto done_unlock;
6172
6173 case STOP_ARRAY_RO:
a4bd82d0 6174 err = md_set_readonly(mddev, 1);
1da177e4
LT		 6175 goto done_unlock;
6176
e2218350
DW		 6177 case BLKROSET:
6178 if (get_user(ro, (int __user *)(arg))) {
6179 err = -EFAULT;
6180 goto done_unlock;
6181 }
6182 err = -EINVAL;
6183
6184 /* if the bdev is going readonly the value of mddev->ro
6185 * does not matter, no writes are coming
6186 */
6187 if (ro)
6188 goto done_unlock;
6189
6190 /* are we are already prepared for writes? */
6191 if (mddev->ro != 1)
6192 goto done_unlock;
6193
6194 /* transitioning to readauto need only happen for
6195 * arrays that call md_write_start
6196 */
6197 if (mddev->pers) {
6198 err = restart_array(mddev);
6199 if (err == 0) {
6200 mddev->ro = 2;
6201 set_disk_ro(mddev->gendisk, 0);
6202 }
6203 }
6204 goto done_unlock;
1da177e4
LT		 6205 }
6206
6207 /*
6208 * The remaining ioctls are changing the state of the
f91de92e
N		 6209 * superblock, so we do not allow them on read-only arrays.
6210 * However non-MD ioctls (e.g. get-size) will still come through
6211 * here and hit the 'default' below, so only disallow
6212 * 'md' ioctls, and switch to rw mode if started auto-readonly.
1da177e4 6213 */
bb57fc64 6214 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
f91de92e
N		 6215 if (mddev->ro == 2) {
6216 mddev->ro = 0;
00bcb4ac 6217 sysfs_notify_dirent_safe(mddev->sysfs_state);
0fd62b86
NB		 6218 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6219 md_wakeup_thread(mddev->thread);
f91de92e
N		 6220 } else {
6221 err = -EROFS;
6222 goto abort_unlock;
6223 }
1da177e4
LT		 6224 }
6225
6226 switch (cmd)
6227 {
6228 case ADD_NEW_DISK:
6229 {
6230 mdu_disk_info_t info;
6231 if (copy_from_user(&info, argp, sizeof(info)))
6232 err = -EFAULT;
6233 else
6234 err = add_new_disk(mddev, &info);
6235 goto done_unlock;
6236 }
6237
6238 case HOT_REMOVE_DISK:
6239 err = hot_remove_disk(mddev, new_decode_dev(arg));
6240 goto done_unlock;
6241
6242 case HOT_ADD_DISK:
6243 err = hot_add_disk(mddev, new_decode_dev(arg));
6244 goto done_unlock;
6245
6246 case SET_DISK_FAULTY:
6247 err = set_disk_faulty(mddev, new_decode_dev(arg));
6248 goto done_unlock;
6249
6250 case RUN_ARRAY:
d710e138 6251 err = do_md_run(mddev);
1da177e4
LT		 6252 goto done_unlock;
6253
32a7627c
N		 6254 case SET_BITMAP_FILE:
6255 err = set_bitmap_file(mddev, (int)arg);
6256 goto done_unlock;
6257
1da177e4 6258 default:
1da177e4
LT		 6259 err = -EINVAL;
6260 goto abort_unlock;
6261 }
6262
6263done_unlock:
6264abort_unlock:
d3374825
N		 6265 if (mddev->hold_active == UNTIL_IOCTL &&
6266 err != -EINVAL)
6267 mddev->hold_active = 0;
1da177e4
LT		 6268 mddev_unlock(mddev);
6269
6270 return err;
6271done:
6272 if (err)
6273 MD_BUG();
6274abort:
6275 return err;
6276}
aa98aa31
AB		 6277#ifdef CONFIG_COMPAT
6278static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6279 unsigned int cmd, unsigned long arg)
6280{
6281 switch (cmd) {
6282 case HOT_REMOVE_DISK:
6283 case HOT_ADD_DISK:
6284 case SET_DISK_FAULTY:
6285 case SET_BITMAP_FILE:
6286 /* These take in integer arg, do not convert */
6287 break;
6288 default:
6289 arg = (unsigned long)compat_ptr(arg);
6290 break;
6291 }
6292
6293 return md_ioctl(bdev, mode, cmd, arg);
6294}
6295#endif /* CONFIG_COMPAT */
1da177e4 6296
a39907fa 6297static int md_open(struct block_device *bdev, fmode_t mode)
1da177e4
LT		 6298{
6299 /*
6300 * Succeed if we can lock the mddev, which confirms that
6301 * it isn't being stopped right now.
6302 */
fd01b88c 6303 struct mddev *mddev = mddev_find(bdev->bd_dev);
1da177e4
LT		 6304 int err;
6305
d3374825
N		 6306 if (mddev->gendisk != bdev->bd_disk) {
6307 /* we are racing with mddev_put which is discarding this
6308 * bd_disk.
6309 */
6310 mddev_put(mddev);
6311 /* Wait until bdev->bd_disk is definitely gone */
e804ac78 6312 flush_workqueue(md_misc_wq);
d3374825
N		 6313 /* Then retry the open from the top */
6314 return -ERESTARTSYS;
6315 }
6316 BUG_ON(mddev != bdev->bd_disk->private_data);
6317
c8c00a69 6318 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
1da177e4
LT		 6319 goto out;
6320
6321 err = 0;
f2ea68cf 6322 atomic_inc(&mddev->openers);
c8c00a69 6323 mutex_unlock(&mddev->open_mutex);
1da177e4 6324
f0b4f7e2 6325 check_disk_change(bdev);
1da177e4
LT		 6326 out:
6327 return err;
6328}
6329
a39907fa 6330static int md_release(struct gendisk *disk, fmode_t mode)
1da177e4 6331{
fd01b88c 6332 struct mddev *mddev = disk->private_data;
1da177e4 6333
52e5f9d1 6334 BUG_ON(!mddev);
f2ea68cf 6335 atomic_dec(&mddev->openers);
1da177e4
LT		 6336 mddev_put(mddev);
6337
6338 return 0;
6339}
f0b4f7e2
N		 6340
6341static int md_media_changed(struct gendisk *disk)
6342{
fd01b88c 6343 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6344
6345 return mddev->changed;
6346}
6347
6348static int md_revalidate(struct gendisk *disk)
6349{
fd01b88c 6350 struct mddev *mddev = disk->private_data;
f0b4f7e2
N		 6351
6352 mddev->changed = 0;
6353 return 0;
6354}
83d5cde4 6355static const struct block_device_operations md_fops =
1da177e4
LT		 6356{
6357 .owner = THIS_MODULE,
a39907fa
AV		 6358 .open = md_open,
6359 .release = md_release,
b492b852 6360 .ioctl = md_ioctl,
aa98aa31
AB		 6361#ifdef CONFIG_COMPAT
6362 .compat_ioctl = md_compat_ioctl,
6363#endif
a885c8c4 6364 .getgeo = md_getgeo,
f0b4f7e2
N		 6365 .media_changed = md_media_changed,
6366 .revalidate_disk= md_revalidate,
1da177e4
LT		 6367};
6368
75c96f85 6369static int md_thread(void * arg)
1da177e4 6370{
2b8bf345 6371 struct md_thread *thread = arg;
1da177e4 6372
1da177e4
LT		 6373 /*
6374 * md_thread is a 'system-thread', it's priority should be very
6375 * high. We avoid resource deadlocks individually in each
6376 * raid personality. (RAID5 does preallocation) We also use RR and
6377 * the very same RT priority as kswapd, thus we will never get
6378 * into a priority inversion deadlock.
6379 *
6380 * we definitely have to have equal or higher priority than
6381 * bdflush, otherwise bdflush will deadlock if there are too
6382 * many dirty RAID5 blocks.
6383 */
1da177e4 6384
6985c43f 6385 allow_signal(SIGKILL);
a6fb0934 6386 while (!kthread_should_stop()) {
1da177e4 6387
93588e22
N		 6388 /* We need to wait INTERRUPTIBLE so that
6389 * we don't add to the load-average.
6390 * That means we need to be sure no signals are
6391 * pending
6392 */
6393 if (signal_pending(current))
6394 flush_signals(current);
6395
6396 wait_event_interruptible_timeout
6397 (thread->wqueue,
6398 test_bit(THREAD_WAKEUP, &thread->flags)
6399 || kthread_should_stop(),
6400 thread->timeout);
1da177e4 6401
6c987910
N		 6402 clear_bit(THREAD_WAKEUP, &thread->flags);
6403 if (!kthread_should_stop())
589a594b 6404 thread->run(thread->mddev);
1da177e4 6405 }
a6fb0934 6406
1da177e4
LT		 6407 return 0;
6408}
6409
2b8bf345 6410void md_wakeup_thread(struct md_thread *thread)
1da177e4
LT		 6411{
6412 if (thread) {
36a4e1fe 6413 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
1da177e4
LT		 6414 set_bit(THREAD_WAKEUP, &thread->flags);
6415 wake_up(&thread->wqueue);
6416 }
6417}
6418
2b8bf345 6419struct md_thread *md_register_thread(void (*run) (struct mddev *), struct mddev *mddev,
1da177e4
LT		 6420 const char *name)
6421{
2b8bf345 6422 struct md_thread *thread;
1da177e4 6423
2b8bf345 6424 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
1da177e4
LT		 6425 if (!thread)
6426 return NULL;
6427
1da177e4
LT		 6428 init_waitqueue_head(&thread->wqueue);
6429
1da177e4
LT		 6430 thread->run = run;
6431 thread->mddev = mddev;
32a7627c 6432 thread->timeout = MAX_SCHEDULE_TIMEOUT;
0da3c619
N		 6433 thread->tsk = kthread_run(md_thread, thread,
6434 "%s_%s",
6435 mdname(thread->mddev),
6436 name ?: mddev->pers->name);
a6fb0934 6437 if (IS_ERR(thread->tsk)) {
1da177e4
LT		 6438 kfree(thread);
6439 return NULL;
6440 }
1da177e4
LT		 6441 return thread;
6442}
6443
2b8bf345 6444void md_unregister_thread(struct md_thread **threadp)
1da177e4 6445{
2b8bf345 6446 struct md_thread *thread = *threadp;
e0cf8f04
N		 6447 if (!thread)
6448 return;
36a4e1fe 6449 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
01f96c0a
N		 6450 /* Locking ensures that mddev_unlock does not wake_up a
6451 * non-existent thread
6452 */
6453 spin_lock(&pers_lock);
6454 *threadp = NULL;
6455 spin_unlock(&pers_lock);
a6fb0934
N		 6456
6457 kthread_stop(thread->tsk);
1da177e4
LT		 6458 kfree(thread);
6459}
6460
fd01b88c 6461void md_error(struct mddev *mddev, struct md_rdev *rdev)
1da177e4
LT		 6462{
6463 if (!mddev) {
6464 MD_BUG();
6465 return;
6466 }
6467
b2d444d7 6468 if (!rdev || test_bit(Faulty, &rdev->flags))
1da177e4 6469 return;
6bfe0b49 6470
de393cde 6471 if (!mddev->pers || !mddev->pers->error_handler)
1da177e4
LT		 6472 return;
6473 mddev->pers->error_handler(mddev,rdev);
72a23c21
NB		 6474 if (mddev->degraded)
6475 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
00bcb4ac 6476 sysfs_notify_dirent_safe(rdev->sysfs_state);
1da177e4
LT		 6477 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6478 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6479 md_wakeup_thread(mddev->thread);
768a418d 6480 if (mddev->event_work.func)
e804ac78 6481 queue_work(md_misc_wq, &mddev->event_work);
c331eb04 6482 md_new_event_inintr(mddev);
1da177e4
LT		 6483}
6484
6485/* seq_file implementation /proc/mdstat */
6486
6487static void status_unused(struct seq_file *seq)
6488{
6489 int i = 0;
3cb03002 6490 struct md_rdev *rdev;
1da177e4
LT		 6491
6492 seq_printf(seq, "unused devices: ");
6493
159ec1fc 6494 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
1da177e4
LT		 6495 char b[BDEVNAME_SIZE];
6496 i++;
6497 seq_printf(seq, "%s ",
6498 bdevname(rdev->bdev,b));
6499 }
6500 if (!i)
6501 seq_printf(seq, "<none>");
6502
6503 seq_printf(seq, "\n");
6504}
6505
6506
fd01b88c 6507static void status_resync(struct seq_file *seq, struct mddev * mddev)
1da177e4 6508{
dd71cf6b
N		 6509 sector_t max_sectors, resync, res;
6510 unsigned long dt, db;
6511 sector_t rt;
4588b42e
N		 6512 int scale;
6513 unsigned int per_milli;
1da177e4 6514
dd71cf6b 6515 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
1da177e4
LT		 6516
6517 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
dd71cf6b 6518 max_sectors = mddev->resync_max_sectors;
1da177e4 6519 else
dd71cf6b 6520 max_sectors = mddev->dev_sectors;
1da177e4
LT		 6521
6522 /*
6523 * Should not happen.
6524 */
dd71cf6b 6525 if (!max_sectors) {
1da177e4
LT		 6526 MD_BUG();
6527 return;
6528 }
4588b42e 6529 /* Pick 'scale' such that (resync>>scale)*1000 will fit
dd71cf6b 6530 * in a sector_t, and (max_sectors>>scale) will fit in a
4588b42e
N		 6531 * u32, as those are the requirements for sector_div.
6532 * Thus 'scale' must be at least 10
6533 */
6534 scale = 10;
6535 if (sizeof(sector_t) > sizeof(unsigned long)) {
dd71cf6b 6536 while ( max_sectors/2 > (1ULL<<(scale+32)))
4588b42e
N		 6537 scale++;
6538 }
6539 res = (resync>>scale)*1000;
dd71cf6b 6540 sector_div(res, (u32)((max_sectors>>scale)+1));
4588b42e
N		 6541
6542 per_milli = res;
1da177e4 6543 {
4588b42e 6544 int i, x = per_milli/50, y = 20-x;
1da177e4
LT		 6545 seq_printf(seq, "[");
6546 for (i = 0; i < x; i++)
6547 seq_printf(seq, "=");
6548 seq_printf(seq, ">");
6549 for (i = 0; i < y; i++)
6550 seq_printf(seq, ".");
6551 seq_printf(seq, "] ");
6552 }
4588b42e 6553 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
ccfcc3c1
N		 6554 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6555 "reshape" :
61df9d91
N		 6556 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6557 "check" :
6558 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6559 "resync" : "recovery"))),
6560 per_milli/10, per_milli % 10,
dd71cf6b
N		 6561 (unsigned long long) resync/2,
6562 (unsigned long long) max_sectors/2);
1da177e4
LT		 6563
6564 /*
1da177e4
LT		 6565 * dt: time from mark until now
6566 * db: blocks written from mark until now
6567 * rt: remaining time
dd71cf6b
N		 6568 *
6569 * rt is a sector_t, so could be 32bit or 64bit.
6570 * So we divide before multiply in case it is 32bit and close
6571 * to the limit.
25985edc 6572 * We scale the divisor (db) by 32 to avoid losing precision
dd71cf6b
N		 6573 * near the end of resync when the number of remaining sectors
6574 * is close to 'db'.
6575 * We then divide rt by 32 after multiplying by db to compensate.
6576 * The '+1' avoids division by zero if db is very small.
1da177e4
LT		 6577 */
6578 dt = ((jiffies - mddev->resync_mark) / HZ);
6579 if (!dt) dt++;
ff4e8d9a
N		 6580 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6581 - mddev->resync_mark_cnt;
1da177e4 6582
dd71cf6b
N		 6583 rt = max_sectors - resync; /* number of remaining sectors */
6584 sector_div(rt, db/32+1);
6585 rt *= dt;
6586 rt >>= 5;
6587
6588 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6589 ((unsigned long)rt % 60)/6);
1da177e4 6590
ff4e8d9a 6591 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
1da177e4
LT		 6592}
6593
6594static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6595{
6596 struct list_head *tmp;
6597 loff_t l = *pos;
fd01b88c 6598 struct mddev *mddev;
1da177e4
LT		 6599
6600 if (l >= 0x10000)
6601 return NULL;
6602 if (!l--)
6603 /* header */
6604 return (void*)1;
6605
6606 spin_lock(&all_mddevs_lock);
6607 list_for_each(tmp,&all_mddevs)
6608 if (!l--) {
fd01b88c 6609 mddev = list_entry(tmp, struct mddev, all_mddevs);
1da177e4
LT		 6610 mddev_get(mddev);
6611 spin_unlock(&all_mddevs_lock);
6612 return mddev;
6613 }
6614 spin_unlock(&all_mddevs_lock);
6615 if (!l--)
6616 return (void*)2;/* tail */
6617 return NULL;
6618}
6619
6620static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6621{
6622 struct list_head *tmp;
fd01b88c 6623 struct mddev *next_mddev, *mddev = v;
1da177e4
LT		 6624
6625 ++*pos;
6626 if (v == (void*)2)
6627 return NULL;
6628
6629 spin_lock(&all_mddevs_lock);
6630 if (v == (void*)1)
6631 tmp = all_mddevs.next;
6632 else
6633 tmp = mddev->all_mddevs.next;
6634 if (tmp != &all_mddevs)
fd01b88c 6635 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
1da177e4
LT		 6636 else {
6637 next_mddev = (void*)2;
6638 *pos = 0x10000;
6639 }
6640 spin_unlock(&all_mddevs_lock);
6641
6642 if (v != (void*)1)
6643 mddev_put(mddev);
6644 return next_mddev;
6645
6646}
6647
6648static void md_seq_stop(struct seq_file *seq, void *v)
6649{
fd01b88c 6650 struct mddev *mddev = v;
1da177e4
LT		 6651
6652 if (mddev && v != (void*)1 && v != (void*)2)
6653 mddev_put(mddev);
6654}
6655
6656static int md_seq_show(struct seq_file *seq, void *v)
6657{
fd01b88c 6658 struct mddev *mddev = v;
dd8ac336 6659 sector_t sectors;
3cb03002 6660 struct md_rdev *rdev;
32a7627c 6661 struct bitmap *bitmap;
1da177e4
LT		 6662
6663 if (v == (void*)1) {
84fc4b56 6664 struct md_personality *pers;
1da177e4
LT		 6665 seq_printf(seq, "Personalities : ");
6666 spin_lock(&pers_lock);
2604b703
N		 6667 list_for_each_entry(pers, &pers_list, list)
6668 seq_printf(seq, "[%s] ", pers->name);
1da177e4
LT		 6669
6670 spin_unlock(&pers_lock);
6671 seq_printf(seq, "\n");
f1514638 6672 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6673 return 0;
6674 }
6675 if (v == (void*)2) {
6676 status_unused(seq);
6677 return 0;
6678 }
6679
5dc5cf7d 6680 if (mddev_lock(mddev) < 0)
1da177e4 6681 return -EINTR;
5dc5cf7d 6682
1da177e4
LT		 6683 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6684 seq_printf(seq, "%s : %sactive", mdname(mddev),
6685 mddev->pers ? "" : "in");
6686 if (mddev->pers) {
f91de92e 6687 if (mddev->ro==1)
1da177e4 6688 seq_printf(seq, " (read-only)");
f91de92e 6689 if (mddev->ro==2)
52720ae7 6690 seq_printf(seq, " (auto-read-only)");
1da177e4
LT		 6691 seq_printf(seq, " %s", mddev->pers->name);
6692 }
6693
dd8ac336 6694 sectors = 0;
159ec1fc 6695 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4
LT		 6696 char b[BDEVNAME_SIZE];
6697 seq_printf(seq, " %s[%d]",
6698 bdevname(rdev->bdev,b), rdev->desc_nr);
8ddf9efe
N		 6699 if (test_bit(WriteMostly, &rdev->flags))
6700 seq_printf(seq, "(W)");
b2d444d7 6701 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 6702 seq_printf(seq, "(F)");
6703 continue;
b325a32e
N		 6704 } else if (rdev->raid_disk < 0)
6705 seq_printf(seq, "(S)"); /* spare */
dd8ac336 6706 sectors += rdev->sectors;
1da177e4
LT		 6707 }
6708
6709 if (!list_empty(&mddev->disks)) {
6710 if (mddev->pers)
6711 seq_printf(seq, "\n %llu blocks",
f233ea5c
AN		 6712 (unsigned long long)
6713 mddev->array_sectors / 2);
1da177e4
LT		 6714 else
6715 seq_printf(seq, "\n %llu blocks",
dd8ac336 6716 (unsigned long long)sectors / 2);
1da177e4 6717 }
1cd6bf19
N		 6718 if (mddev->persistent) {
6719 if (mddev->major_version != 0 ||
6720 mddev->minor_version != 90) {
6721 seq_printf(seq," super %d.%d",
6722 mddev->major_version,
6723 mddev->minor_version);
6724 }
e691063a
N		 6725 } else if (mddev->external)
6726 seq_printf(seq, " super external:%s",
6727 mddev->metadata_type);
6728 else
1cd6bf19 6729 seq_printf(seq, " super non-persistent");
1da177e4
LT		 6730
6731 if (mddev->pers) {
d710e138 6732 mddev->pers->status(seq, mddev);
1da177e4 6733 seq_printf(seq, "\n ");
8e1b39d6
N		 6734 if (mddev->pers->sync_request) {
6735 if (mddev->curr_resync > 2) {
d710e138 6736 status_resync(seq, mddev);
8e1b39d6
N		 6737 seq_printf(seq, "\n ");
6738 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6739 seq_printf(seq, "\tresync=DELAYED\n ");
6740 else if (mddev->recovery_cp < MaxSector)
6741 seq_printf(seq, "\tresync=PENDING\n ");
6742 }
32a7627c
N		 6743 } else
6744 seq_printf(seq, "\n ");
6745
6746 if ((bitmap = mddev->bitmap)) {
32a7627c
N		 6747 unsigned long chunk_kb;
6748 unsigned long flags;
32a7627c 6749 spin_lock_irqsave(&bitmap->lock, flags);
42a04b50 6750 chunk_kb = mddev->bitmap_info.chunksize >> 10;
32a7627c
N		 6751 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6752 "%lu%s chunk",
6753 bitmap->pages - bitmap->missing_pages,
6754 bitmap->pages,
6755 (bitmap->pages - bitmap->missing_pages)
6756 << (PAGE_SHIFT - 10),
42a04b50 6757 chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
32a7627c 6758 chunk_kb ? "KB" : "B");
78d742d8
N		 6759 if (bitmap->file) {
6760 seq_printf(seq, ", file: ");
c32c2f63 6761 seq_path(seq, &bitmap->file->f_path, " \t\n");
32a7627c 6762 }
78d742d8 6763
32a7627c
N		 6764 seq_printf(seq, "\n");
6765 spin_unlock_irqrestore(&bitmap->lock, flags);
1da177e4
LT		 6766 }
6767
6768 seq_printf(seq, "\n");
6769 }
6770 mddev_unlock(mddev);
6771
6772 return 0;
6773}
6774
110518bc 6775static const struct seq_operations md_seq_ops = {
1da177e4
LT		 6776 .start = md_seq_start,
6777 .next = md_seq_next,
6778 .stop = md_seq_stop,
6779 .show = md_seq_show,
6780};
6781
6782static int md_seq_open(struct inode *inode, struct file *file)
6783{
f1514638 6784 struct seq_file *seq;
1da177e4
LT		 6785 int error;
6786
6787 error = seq_open(file, &md_seq_ops);
d7603b7e 6788 if (error)
f1514638
KS		 6789 return error;
6790
6791 seq = file->private_data;
6792 seq->poll_event = atomic_read(&md_event_count);
1da177e4
LT		 6793 return error;
6794}
6795
d7603b7e
N		 6796static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6797{
f1514638 6798 struct seq_file *seq = filp->private_data;
d7603b7e
N		 6799 int mask;
6800
6801 poll_wait(filp, &md_event_waiters, wait);
6802
6803 /* always allow read */
6804 mask = POLLIN | POLLRDNORM;
6805
f1514638 6806 if (seq->poll_event != atomic_read(&md_event_count))
d7603b7e
N		 6807 mask |= POLLERR | POLLPRI;
6808 return mask;
6809}
6810
fa027c2a 6811static const struct file_operations md_seq_fops = {
e24650c2 6812 .owner = THIS_MODULE,
1da177e4
LT		 6813 .open = md_seq_open,
6814 .read = seq_read,
6815 .llseek = seq_lseek,
c3f94b40 6816 .release = seq_release_private,
d7603b7e 6817 .poll = mdstat_poll,
1da177e4
LT		 6818};
6819
84fc4b56 6820int register_md_personality(struct md_personality *p)
1da177e4 6821{
1da177e4 6822 spin_lock(&pers_lock);
2604b703
N		 6823 list_add_tail(&p->list, &pers_list);
6824 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
1da177e4
LT		 6825 spin_unlock(&pers_lock);
6826 return 0;
6827}
6828
84fc4b56 6829int unregister_md_personality(struct md_personality *p)
1da177e4 6830{
2604b703 6831 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
1da177e4 6832 spin_lock(&pers_lock);
2604b703 6833 list_del_init(&p->list);
1da177e4
LT		 6834 spin_unlock(&pers_lock);
6835 return 0;
6836}
6837
fd01b88c 6838static int is_mddev_idle(struct mddev *mddev, int init)
1da177e4 6839{
3cb03002 6840 struct md_rdev * rdev;
1da177e4 6841 int idle;
eea1bf38 6842 int curr_events;
1da177e4
LT		 6843
6844 idle = 1;
4b80991c
N		 6845 rcu_read_lock();
6846 rdev_for_each_rcu(rdev, mddev) {
1da177e4 6847 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
eea1bf38
N		 6848 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6849 (int)part_stat_read(&disk->part0, sectors[1]) -
6850 atomic_read(&disk->sync_io);
713f6ab1
N		 6851 /* sync IO will cause sync_io to increase before the disk_stats
6852 * as sync_io is counted when a request starts, and
6853 * disk_stats is counted when it completes.
6854 * So resync activity will cause curr_events to be smaller than
6855 * when there was no such activity.
6856 * non-sync IO will cause disk_stat to increase without
6857 * increasing sync_io so curr_events will (eventually)
6858 * be larger than it was before. Once it becomes
6859 * substantially larger, the test below will cause
6860 * the array to appear non-idle, and resync will slow
6861 * down.
6862 * If there is a lot of outstanding resync activity when
6863 * we set last_event to curr_events, then all that activity
6864 * completing might cause the array to appear non-idle
6865 * and resync will be slowed down even though there might
6866 * not have been non-resync activity. This will only
6867 * happen once though. 'last_events' will soon reflect
6868 * the state where there is little or no outstanding
6869 * resync requests, and further resync activity will
6870 * always make curr_events less than last_events.
c0e48521 6871 *
1da177e4 6872 */
eea1bf38 6873 if (init || curr_events - rdev->last_events > 64) {
1da177e4
LT		 6874 rdev->last_events = curr_events;
6875 idle = 0;
6876 }
6877 }
4b80991c 6878 rcu_read_unlock();
1da177e4
LT		 6879 return idle;
6880}
6881
fd01b88c 6882void md_done_sync(struct mddev *mddev, int blocks, int ok)
1da177e4
LT		 6883{
6884 /* another "blocks" (512byte) blocks have been synced */
6885 atomic_sub(blocks, &mddev->recovery_active);
6886 wake_up(&mddev->recovery_wait);
6887 if (!ok) {
dfc70645 6888 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 6889 md_wakeup_thread(mddev->thread);
6890 // stop recovery, signal do_sync ....
6891 }
6892}
6893
6894
06d91a5f
N		 6895/* md_write_start(mddev, bi)
6896 * If we need to update some array metadata (e.g. 'active' flag
3d310eb7
N		 6897 * in superblock) before writing, schedule a superblock update
6898 * and wait for it to complete.
06d91a5f 6899 */
fd01b88c 6900void md_write_start(struct mddev *mddev, struct bio *bi)
1da177e4 6901{
0fd62b86 6902 int did_change = 0;
06d91a5f 6903 if (bio_data_dir(bi) != WRITE)
3d310eb7 6904 return;
06d91a5f 6905
f91de92e
N		 6906 BUG_ON(mddev->ro == 1);
6907 if (mddev->ro == 2) {
6908 /* need to switch to read/write */
6909 mddev->ro = 0;
6910 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6911 md_wakeup_thread(mddev->thread);
25156198 6912 md_wakeup_thread(mddev->sync_thread);
0fd62b86 6913 did_change = 1;
f91de92e 6914 }
06d91a5f 6915 atomic_inc(&mddev->writes_pending);
31a59e34
N		 6916 if (mddev->safemode == 1)
6917 mddev->safemode = 0;
06d91a5f 6918 if (mddev->in_sync) {
a9701a30 6919 spin_lock_irq(&mddev->write_lock);
3d310eb7
N		 6920 if (mddev->in_sync) {
6921 mddev->in_sync = 0;
850b2b42 6922 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6923 set_bit(MD_CHANGE_PENDING, &mddev->flags);
3d310eb7 6924 md_wakeup_thread(mddev->thread);
0fd62b86 6925 did_change = 1;
3d310eb7 6926 }
a9701a30 6927 spin_unlock_irq(&mddev->write_lock);
06d91a5f 6928 }
0fd62b86 6929 if (did_change)
00bcb4ac 6930 sysfs_notify_dirent_safe(mddev->sysfs_state);
09a44cc1 6931 wait_event(mddev->sb_wait,
09a44cc1 6932 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
1da177e4
LT		 6933}
6934
fd01b88c 6935void md_write_end(struct mddev *mddev)
1da177e4
LT		 6936{
6937 if (atomic_dec_and_test(&mddev->writes_pending)) {
6938 if (mddev->safemode == 2)
6939 md_wakeup_thread(mddev->thread);
16f17b39 6940 else if (mddev->safemode_delay)
1da177e4
LT		 6941 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6942 }
6943}
6944
2a2275d6
N		 6945/* md_allow_write(mddev)
6946 * Calling this ensures that the array is marked 'active' so that writes
6947 * may proceed without blocking. It is important to call this before
6948 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6949 * Must be called with mddev_lock held.
b5470dc5
DW		 6950 *
6951 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6952 * is dropped, so return -EAGAIN after notifying userspace.
2a2275d6 6953 */
fd01b88c 6954int md_allow_write(struct mddev *mddev)
2a2275d6
N		 6955{
6956 if (!mddev->pers)
b5470dc5 6957 return 0;
2a2275d6 6958 if (mddev->ro)
b5470dc5 6959 return 0;
1a0fd497 6960 if (!mddev->pers->sync_request)
b5470dc5 6961 return 0;
2a2275d6
N		 6962
6963 spin_lock_irq(&mddev->write_lock);
6964 if (mddev->in_sync) {
6965 mddev->in_sync = 0;
6966 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
070dc6dd 6967 set_bit(MD_CHANGE_PENDING, &mddev->flags);
2a2275d6
N		 6968 if (mddev->safemode_delay &&
6969 mddev->safemode == 0)
6970 mddev->safemode = 1;
6971 spin_unlock_irq(&mddev->write_lock);
6972 md_update_sb(mddev, 0);
00bcb4ac 6973 sysfs_notify_dirent_safe(mddev->sysfs_state);
2a2275d6
N		 6974 } else
6975 spin_unlock_irq(&mddev->write_lock);
b5470dc5 6976
070dc6dd 6977 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
b5470dc5
DW		 6978 return -EAGAIN;
6979 else
6980 return 0;
2a2275d6
N		 6981}
6982EXPORT_SYMBOL_GPL(md_allow_write);
6983
1da177e4
LT		 6984#define SYNC_MARKS 10
6985#define SYNC_MARK_STEP (3*HZ)
fd01b88c 6986void md_do_sync(struct mddev *mddev)
1da177e4 6987{
fd01b88c 6988 struct mddev *mddev2;
1da177e4
LT		 6989 unsigned int currspeed = 0,
6990 window;
57afd89f 6991 sector_t max_sectors,j, io_sectors;
1da177e4
LT		 6992 unsigned long mark[SYNC_MARKS];
6993 sector_t mark_cnt[SYNC_MARKS];
6994 int last_mark,m;
6995 struct list_head *tmp;
6996 sector_t last_check;
57afd89f 6997 int skipped = 0;
3cb03002 6998 struct md_rdev *rdev;
61df9d91 6999 char *desc;
1da177e4
LT		 7000
7001 /* just incase thread restarts... */
7002 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7003 return;
5fd6c1dc
N		 7004 if (mddev->ro) /* never try to sync a read-only array */
7005 return;
1da177e4 7006
61df9d91
N		 7007 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7008 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7009 desc = "data-check";
7010 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7011 desc = "requested-resync";
7012 else
7013 desc = "resync";
7014 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7015 desc = "reshape";
7016 else
7017 desc = "recovery";
7018
1da177e4
LT		 7019 /* we overload curr_resync somewhat here.
7020 * 0 == not engaged in resync at all
7021 * 2 == checking that there is no conflict with another sync
7022 * 1 == like 2, but have yielded to allow conflicting resync to
7023 * commense
7024 * other == active in resync - this many blocks
7025 *
7026 * Before starting a resync we must have set curr_resync to
7027 * 2, and then checked that every "conflicting" array has curr_resync
7028 * less than ours. When we find one that is the same or higher
7029 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7030 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7031 * This will mean we have to start checking from the beginning again.
7032 *
7033 */
7034
7035 do {
7036 mddev->curr_resync = 2;
7037
7038 try_again:
404e4b43 7039 if (kthread_should_stop())
6985c43f 7040 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
404e4b43
N		 7041
7042 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
1da177e4 7043 goto skip;
29ac4aa3 7044 for_each_mddev(mddev2, tmp) {
1da177e4
LT		 7045 if (mddev2 == mddev)
7046 continue;
90b08710
BS		 7047 if (!mddev->parallel_resync
7048 && mddev2->curr_resync
7049 && match_mddev_units(mddev, mddev2)) {
1da177e4
LT		 7050 DEFINE_WAIT(wq);
7051 if (mddev < mddev2 && mddev->curr_resync == 2) {
7052 /* arbitrarily yield */
7053 mddev->curr_resync = 1;
7054 wake_up(&resync_wait);
7055 }
7056 if (mddev > mddev2 && mddev->curr_resync == 1)
7057 /* no need to wait here, we can wait the next
7058 * time 'round when curr_resync == 2
7059 */
7060 continue;
9744197c
N		 7061 /* We need to wait 'interruptible' so as not to
7062 * contribute to the load average, and not to
7063 * be caught by 'softlockup'
7064 */
7065 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
787453c2 7066 if (!kthread_should_stop() &&
8712e553 7067 mddev2->curr_resync >= mddev->curr_resync) {
61df9d91
N		 7068 printk(KERN_INFO "md: delaying %s of %s"
7069 " until %s has finished (they"
1da177e4 7070 " share one or more physical units)\n",
61df9d91 7071 desc, mdname(mddev), mdname(mddev2));
1da177e4 7072 mddev_put(mddev2);
9744197c
N		 7073 if (signal_pending(current))
7074 flush_signals(current);
1da177e4
LT		 7075 schedule();
7076 finish_wait(&resync_wait, &wq);
7077 goto try_again;
7078 }
7079 finish_wait(&resync_wait, &wq);
7080 }
7081 }
7082 } while (mddev->curr_resync < 2);
7083
5fd6c1dc 7084 j = 0;
9d88883e 7085 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1da177e4 7086 /* resync follows the size requested by the personality,
57afd89f 7087 * which defaults to physical size, but can be virtual size
1da177e4
LT		 7088 */
7089 max_sectors = mddev->resync_max_sectors;
9d88883e 7090 mddev->resync_mismatches = 0;
5fd6c1dc 7091 /* we don't use the checkpoint if there's a bitmap */
5e96ee65
NB		 7092 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7093 j = mddev->resync_min;
7094 else if (!mddev->bitmap)
5fd6c1dc 7095 j = mddev->recovery_cp;
5e96ee65 7096
ccfcc3c1 7097 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
58c0fed4 7098 max_sectors = mddev->dev_sectors;
5fd6c1dc 7099 else {
1da177e4 7100 /* recovery follows the physical size of devices */
58c0fed4 7101 max_sectors = mddev->dev_sectors;
5fd6c1dc 7102 j = MaxSector;
4e59ca7d
DW		 7103 rcu_read_lock();
7104 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc
N		 7105 if (rdev->raid_disk >= 0 &&
7106 !test_bit(Faulty, &rdev->flags) &&
7107 !test_bit(In_sync, &rdev->flags) &&
7108 rdev->recovery_offset < j)
7109 j = rdev->recovery_offset;
4e59ca7d 7110 rcu_read_unlock();
5fd6c1dc 7111 }
1da177e4 7112
61df9d91
N		 7113 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7114 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7115 " %d KB/sec/disk.\n", speed_min(mddev));
338cec32 7116 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
61df9d91
N		 7117 "(but not more than %d KB/sec) for %s.\n",
7118 speed_max(mddev), desc);
1da177e4 7119
eea1bf38 7120 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
5fd6c1dc 7121
57afd89f 7122 io_sectors = 0;
1da177e4
LT		 7123 for (m = 0; m < SYNC_MARKS; m++) {
7124 mark[m] = jiffies;
57afd89f 7125 mark_cnt[m] = io_sectors;
1da177e4
LT		 7126 }
7127 last_mark = 0;
7128 mddev->resync_mark = mark[last_mark];
7129 mddev->resync_mark_cnt = mark_cnt[last_mark];
7130
7131 /*
7132 * Tune reconstruction:
7133 */
7134 window = 32*(PAGE_SIZE/512);
ac42450c
JB		 7135 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7136 window/2, (unsigned long long)max_sectors/2);
1da177e4
LT		 7137
7138 atomic_set(&mddev->recovery_active, 0);
1da177e4
LT		 7139 last_check = 0;
7140
7141 if (j>2) {
7142 printk(KERN_INFO
61df9d91
N		 7143 "md: resuming %s of %s from checkpoint.\n",
7144 desc, mdname(mddev));
1da177e4
LT		 7145 mddev->curr_resync = j;
7146 }
75d3da43 7147 mddev->curr_resync_completed = j;
1da177e4
LT		 7148
7149 while (j < max_sectors) {
57afd89f 7150 sector_t sectors;
1da177e4 7151
57afd89f 7152 skipped = 0;
97e4f42d 7153
7a91ee1f
N		 7154 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7155 ((mddev->curr_resync > mddev->curr_resync_completed &&
7156 (mddev->curr_resync - mddev->curr_resync_completed)
7157 > (max_sectors >> 4)) ||
7158 (j - mddev->curr_resync_completed)*2
7159 >= mddev->resync_max - mddev->curr_resync_completed
7160 )) {
97e4f42d 7161 /* time to update curr_resync_completed */
97e4f42d
N		 7162 wait_event(mddev->recovery_wait,
7163 atomic_read(&mddev->recovery_active) == 0);
75d3da43 7164 mddev->curr_resync_completed = j;
070dc6dd 7165 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
acb180b0 7166 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
97e4f42d 7167 }
acb180b0 7168
e62e58a5
N		 7169 while (j >= mddev->resync_max && !kthread_should_stop()) {
7170 /* As this condition is controlled by user-space,
7171 * we can block indefinitely, so use '_interruptible'
7172 * to avoid triggering warnings.
7173 */
7174 flush_signals(current); /* just in case */
7175 wait_event_interruptible(mddev->recovery_wait,
7176 mddev->resync_max > j
7177 || kthread_should_stop());
7178 }
acb180b0
N		 7179
7180 if (kthread_should_stop())
7181 goto interrupted;
7182
57afd89f 7183 sectors = mddev->pers->sync_request(mddev, j, &skipped,
c6207277 7184 currspeed < speed_min(mddev));
57afd89f 7185 if (sectors == 0) {
dfc70645 7186 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1da177e4
LT		 7187 goto out;
7188 }
57afd89f
N		 7189
7190 if (!skipped) { /* actual IO requested */
7191 io_sectors += sectors;
7192 atomic_add(sectors, &mddev->recovery_active);
7193 }
7194
e875ecea
N		 7195 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7196 break;
7197
1da177e4
LT		 7198 j += sectors;
7199 if (j>1) mddev->curr_resync = j;
ff4e8d9a 7200 mddev->curr_mark_cnt = io_sectors;
d7603b7e 7201 if (last_check == 0)
e875ecea 7202 /* this is the earliest that rebuild will be
d7603b7e
N		 7203 * visible in /proc/mdstat
7204 */
7205 md_new_event(mddev);
57afd89f
N		 7206
7207 if (last_check + window > io_sectors || j == max_sectors)
1da177e4
LT		 7208 continue;
7209
57afd89f 7210 last_check = io_sectors;
1da177e4
LT		 7211 repeat:
7212 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7213 /* step marks */
7214 int next = (last_mark+1) % SYNC_MARKS;
7215
7216 mddev->resync_mark = mark[next];
7217 mddev->resync_mark_cnt = mark_cnt[next];
7218 mark[next] = jiffies;
57afd89f 7219 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
1da177e4
LT		 7220 last_mark = next;
7221 }
7222
7223
c6207277
N		 7224 if (kthread_should_stop())
7225 goto interrupted;
7226
1da177e4
LT		 7227
7228 /*
7229 * this loop exits only if either when we are slower than
7230 * the 'hard' speed limit, or the system was IO-idle for
7231 * a jiffy.
7232 * the system might be non-idle CPU-wise, but we only care
7233 * about not overloading the IO subsystem. (things like an
7234 * e2fsck being done on the RAID array should execute fast)
7235 */
1da177e4
LT		 7236 cond_resched();
7237
57afd89f
N		 7238 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7239 /((jiffies-mddev->resync_mark)/HZ +1) +1;
1da177e4 7240
88202a0c
N		 7241 if (currspeed > speed_min(mddev)) {
7242 if ((currspeed > speed_max(mddev)) ||
eea1bf38 7243 !is_mddev_idle(mddev, 0)) {
c0e48521 7244 msleep(500);
1da177e4
LT		 7245 goto repeat;
7246 }
7247 }
7248 }
61df9d91 7249 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
1da177e4
LT		 7250 /*
7251 * this also signals 'finished resyncing' to md_stop
7252 */
7253 out:
1da177e4
LT		 7254 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7255
7256 /* tell personality that we are finished */
57afd89f 7257 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
1da177e4 7258
dfc70645 7259 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5fd6c1dc
N		 7260 mddev->curr_resync > 2) {
7261 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7262 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7263 if (mddev->curr_resync >= mddev->recovery_cp) {
7264 printk(KERN_INFO
61df9d91
N		 7265 "md: checkpointing %s of %s.\n",
7266 desc, mdname(mddev));
5fd6c1dc
N		 7267 mddev->recovery_cp = mddev->curr_resync;
7268 }
7269 } else
7270 mddev->recovery_cp = MaxSector;
7271 } else {
7272 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7273 mddev->curr_resync = MaxSector;
4e59ca7d
DW		 7274 rcu_read_lock();
7275 list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
5fd6c1dc 7276 if (rdev->raid_disk >= 0 &&
70fffd0b 7277 mddev->delta_disks >= 0 &&
5fd6c1dc
N		 7278 !test_bit(Faulty, &rdev->flags) &&
7279 !test_bit(In_sync, &rdev->flags) &&
7280 rdev->recovery_offset < mddev->curr_resync)
7281 rdev->recovery_offset = mddev->curr_resync;
4e59ca7d 7282 rcu_read_unlock();
5fd6c1dc 7283 }
1da177e4 7284 }
17571284 7285 set_bit(MD_CHANGE_DEVS, &mddev->flags);
1da177e4 7286
1da177e4 7287 skip:
c07b70ad
N		 7288 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7289 /* We completed so min/max setting can be forgotten if used. */
7290 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7291 mddev->resync_min = 0;
7292 mddev->resync_max = MaxSector;
7293 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7294 mddev->resync_min = mddev->curr_resync_completed;
1da177e4
LT		 7295 mddev->curr_resync = 0;
7296 wake_up(&resync_wait);
7297 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7298 md_wakeup_thread(mddev->thread);
c6207277
N		 7299 return;
7300
7301 interrupted:
7302 /*
7303 * got a signal, exit.
7304 */
7305 printk(KERN_INFO
7306 "md: md_do_sync() got signal ... exiting\n");
7307 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7308 goto out;
7309
1da177e4 7310}
29269553 7311EXPORT_SYMBOL_GPL(md_do_sync);
1da177e4 7312
fd01b88c 7313static int remove_and_add_spares(struct mddev *mddev)
b4c4c7b8 7314{
3cb03002 7315 struct md_rdev *rdev;
b4c4c7b8
N		 7316 int spares = 0;
7317
97e4f42d
N		 7318 mddev->curr_resync_completed = 0;
7319
159ec1fc 7320 list_for_each_entry(rdev, &mddev->disks, same_set)
b4c4c7b8 7321 if (rdev->raid_disk >= 0 &&
6bfe0b49 7322 !test_bit(Blocked, &rdev->flags) &&
b4c4c7b8
N		 7323 (test_bit(Faulty, &rdev->flags) ||
7324 ! test_bit(In_sync, &rdev->flags)) &&
7325 atomic_read(&rdev->nr_pending)==0) {
7326 if (mddev->pers->hot_remove_disk(
7327 mddev, rdev->raid_disk)==0) {
36fad858 7328 sysfs_unlink_rdev(mddev, rdev);
b4c4c7b8
N		 7329 rdev->raid_disk = -1;
7330 }
7331 }
7332
5389042f 7333 if (mddev->degraded) {
159ec1fc 7334 list_for_each_entry(rdev, &mddev->disks, same_set) {
dfc70645 7335 if (rdev->raid_disk >= 0 &&
e5427135 7336 !test_bit(In_sync, &rdev->flags) &&
de393cde 7337 !test_bit(Faulty, &rdev->flags))
dfc70645 7338 spares++;
b4c4c7b8
N		 7339 if (rdev->raid_disk < 0
7340 && !test_bit(Faulty, &rdev->flags)) {
7341 rdev->recovery_offset = 0;
199050ea
NB		 7342 if (mddev->pers->
7343 hot_add_disk(mddev, rdev) == 0) {
36fad858 7344 if (sysfs_link_rdev(mddev, rdev))
00bcb4ac 7345 /* failure here is OK */;
b4c4c7b8
N		 7346 spares++;
7347 md_new_event(mddev);
93be75ff 7348 set_bit(MD_CHANGE_DEVS, &mddev->flags);
b4c4c7b8
N		 7349 } else
7350 break;
7351 }
dfc70645 7352 }
b4c4c7b8
N		 7353 }
7354 return spares;
7355}
7ebc0be7 7356
fd01b88c 7357static void reap_sync_thread(struct mddev *mddev)
7ebc0be7 7358{
3cb03002 7359 struct md_rdev *rdev;
7ebc0be7
N		 7360
7361 /* resync has finished, collect result */
01f96c0a 7362 md_unregister_thread(&mddev->sync_thread);
7ebc0be7
N		 7363 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7364 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7365 /* success...*/
7366 /* activate any spares */
7367 if (mddev->pers->spare_active(mddev))
7368 sysfs_notify(&mddev->kobj, NULL,
7369 "degraded");
7370 }
7371 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7372 mddev->pers->finish_reshape)
7373 mddev->pers->finish_reshape(mddev);
7ebc0be7 7374
d70ed2e4
AW		 7375 /* If array is no-longer degraded, then any saved_raid_disk
7376 * information must be scrapped. Also if any device is now
7377 * In_sync we must scrape the saved_raid_disk for that device
7378 * do the superblock for an incrementally recovered device
7379 * written out.
7ebc0be7 7380 */
d70ed2e4
AW		 7381 list_for_each_entry(rdev, &mddev->disks, same_set)
7382 if (!mddev->degraded ||
7383 test_bit(In_sync, &rdev->flags))
7ebc0be7
N		 7384 rdev->saved_raid_disk = -1;
7385
d70ed2e4 7386 md_update_sb(mddev, 1);
7ebc0be7
N		 7387 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7388 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7389 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7390 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7391 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7392 /* flag recovery needed just to double check */
7393 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7394 sysfs_notify_dirent_safe(mddev->sysfs_action);
7395 md_new_event(mddev);
768e587e
JB		 7396 if (mddev->event_work.func)
7397 queue_work(md_misc_wq, &mddev->event_work);
7ebc0be7
N		 7398}
7399
1da177e4
LT		 7400/*
7401 * This routine is regularly called by all per-raid-array threads to
7402 * deal with generic issues like resync and super-block update.
7403 * Raid personalities that don't have a thread (linear/raid0) do not
7404 * need this as they never do any recovery or update the superblock.
7405 *
7406 * It does not do any resync itself, but rather "forks" off other threads
7407 * to do that as needed.
7408 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7409 * "->recovery" and create a thread at ->sync_thread.
dfc70645 7410 * When the thread finishes it sets MD_RECOVERY_DONE
1da177e4
LT		 7411 * and wakeups up this thread which will reap the thread and finish up.
7412 * This thread also removes any faulty devices (with nr_pending == 0).
7413 *
7414 * The overall approach is:
7415 * 1/ if the superblock needs updating, update it.
7416 * 2/ If a recovery thread is running, don't do anything else.
7417 * 3/ If recovery has finished, clean up, possibly marking spares active.
7418 * 4/ If there are any faulty devices, remove them.
7419 * 5/ If array is degraded, try to add spares devices
7420 * 6/ If array has spares or is not in-sync, start a resync thread.
7421 */
fd01b88c 7422void md_check_recovery(struct mddev *mddev)
1da177e4 7423{
68866e42
JB		 7424 if (mddev->suspended)
7425 return;
7426
5f40402d 7427 if (mddev->bitmap)
aa5cbd10 7428 bitmap_daemon_work(mddev);
1da177e4 7429
fca4d848 7430 if (signal_pending(current)) {
31a59e34 7431 if (mddev->pers->sync_request && !mddev->external) {
fca4d848
N		 7432 printk(KERN_INFO "md: %s in immediate safe mode\n",
7433 mdname(mddev));
7434 mddev->safemode = 2;
7435 }
7436 flush_signals(current);
7437 }
7438
c89a8eee
N		 7439 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7440 return;
1da177e4 7441 if ( ! (
126925c0 7442 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
1da177e4 7443 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
fca4d848 7444 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
31a59e34 7445 (mddev->external == 0 && mddev->safemode == 1) ||
fca4d848
N		 7446 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7447 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
1da177e4
LT		 7448 ))
7449 return;
fca4d848 7450
df5b89b3 7451 if (mddev_trylock(mddev)) {
b4c4c7b8 7452 int spares = 0;
fca4d848 7453
c89a8eee
N		 7454 if (mddev->ro) {
7455 /* Only thing we do on a ro array is remove
7456 * failed devices.
7457 */
3cb03002 7458 struct md_rdev *rdev;
a8c42c7f
N		 7459 list_for_each_entry(rdev, &mddev->disks, same_set)
7460 if (rdev->raid_disk >= 0 &&
7461 !test_bit(Blocked, &rdev->flags) &&
7462 test_bit(Faulty, &rdev->flags) &&
7463 atomic_read(&rdev->nr_pending)==0) {
7464 if (mddev->pers->hot_remove_disk(
7465 mddev, rdev->raid_disk)==0) {
36fad858 7466 sysfs_unlink_rdev(mddev, rdev);
a8c42c7f
N		 7467 rdev->raid_disk = -1;
7468 }
7469 }
c89a8eee
N		 7470 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7471 goto unlock;
7472 }
7473
31a59e34 7474 if (!mddev->external) {
0fd62b86 7475 int did_change = 0;
31a59e34
N		 7476 spin_lock_irq(&mddev->write_lock);
7477 if (mddev->safemode &&
7478 !atomic_read(&mddev->writes_pending) &&
7479 !mddev->in_sync &&
7480 mddev->recovery_cp == MaxSector) {
7481 mddev->in_sync = 1;
0fd62b86 7482 did_change = 1;
070dc6dd 7483 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
31a59e34
N		 7484 }
7485 if (mddev->safemode == 1)
7486 mddev->safemode = 0;
7487 spin_unlock_irq(&mddev->write_lock);
0fd62b86 7488 if (did_change)
00bcb4ac 7489 sysfs_notify_dirent_safe(mddev->sysfs_state);
fca4d848 7490 }
fca4d848 7491
850b2b42
N		 7492 if (mddev->flags)
7493 md_update_sb(mddev, 0);
06d91a5f 7494
1da177e4
LT		 7495 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7496 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7497 /* resync/recovery still happening */
7498 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7499 goto unlock;
7500 }
7501 if (mddev->sync_thread) {
7ebc0be7 7502 reap_sync_thread(mddev);
1da177e4
LT		 7503 goto unlock;
7504 }
72a23c21
NB		 7505 /* Set RUNNING before clearing NEEDED to avoid
7506 * any transients in the value of "sync_action".
7507 */
7508 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7509 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
24dd469d
N		 7510 /* Clear some bits that don't mean anything, but
7511 * might be left set
7512 */
24dd469d
N		 7513 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7514 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
1da177e4 7515
5fd6c1dc
N		 7516 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7517 goto unlock;
1da177e4
LT		 7518 /* no recovery is running.
7519 * remove any failed drives, then
7520 * add spares if possible.
7521 * Spare are also removed and re-added, to allow
7522 * the personality to fail the re-add.
7523 */
1da177e4 7524
b4c4c7b8 7525 if (mddev->reshape_position != MaxSector) {
50ac168a
N		 7526 if (mddev->pers->check_reshape == NULL ||
7527 mddev->pers->check_reshape(mddev) != 0)
b4c4c7b8
N		 7528 /* Cannot proceed */
7529 goto unlock;
7530 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
72a23c21 7531 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
b4c4c7b8 7532 } else if ((spares = remove_and_add_spares(mddev))) {
24dd469d
N		 7533 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7534 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
56ac36d7 7535 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
72a23c21 7536 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7537 } else if (mddev->recovery_cp < MaxSector) {
7538 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
72a23c21 7539 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
24dd469d
N		 7540 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7541 /* nothing to be done ... */
1da177e4 7542 goto unlock;
24dd469d 7543
1da177e4 7544 if (mddev->pers->sync_request) {
a654b9d8
N		 7545 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7546 /* We are adding a device or devices to an array
7547 * which has the bitmap stored on all devices.
7548 * So make sure all bitmap pages get written
7549 */
7550 bitmap_write_all(mddev->bitmap);
7551 }
1da177e4
LT		 7552 mddev->sync_thread = md_register_thread(md_do_sync,
7553 mddev,
0da3c619 7554 "resync");
1da177e4
LT		 7555 if (!mddev->sync_thread) {
7556 printk(KERN_ERR "%s: could not start resync"
7557 " thread...\n",
7558 mdname(mddev));
7559 /* leave the spares where they are, it shouldn't hurt */
7ebc0be7
N		 7560 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7561 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7562 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7563 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7564 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
d7603b7e 7565 } else
1da177e4 7566 md_wakeup_thread(mddev->sync_thread);
00bcb4ac 7567 sysfs_notify_dirent_safe(mddev->sysfs_action);
d7603b7e 7568 md_new_event(mddev);
1da177e4
LT		 7569 }
7570 unlock:
72a23c21
NB		 7571 if (!mddev->sync_thread) {
7572 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7573 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7574 &mddev->recovery))
0c3573f1 7575 if (mddev->sysfs_action)
00bcb4ac 7576 sysfs_notify_dirent_safe(mddev->sysfs_action);
72a23c21 7577 }
1da177e4
LT		 7578 mddev_unlock(mddev);
7579 }
7580}
7581
fd01b88c 7582void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
6bfe0b49 7583{
00bcb4ac 7584 sysfs_notify_dirent_safe(rdev->sysfs_state);
6bfe0b49 7585 wait_event_timeout(rdev->blocked_wait,
de393cde
N		 7586 !test_bit(Blocked, &rdev->flags) &&
7587 !test_bit(BlockedBadBlocks, &rdev->flags),
6bfe0b49
DW		 7588 msecs_to_jiffies(5000));
7589 rdev_dec_pending(rdev, mddev);
7590}
7591EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7592
2230dfe4
N		 7593
7594/* Bad block management.
7595 * We can record which blocks on each device are 'bad' and so just
7596 * fail those blocks, or that stripe, rather than the whole device.
7597 * Entries in the bad-block table are 64bits wide. This comprises:
7598 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7599 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7600 * A 'shift' can be set so that larger blocks are tracked and
7601 * consequently larger devices can be covered.
7602 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7603 *
7604 * Locking of the bad-block table uses a seqlock so md_is_badblock
7605 * might need to retry if it is very unlucky.
7606 * We will sometimes want to check for bad blocks in a bi_end_io function,
7607 * so we use the write_seqlock_irq variant.
7608 *
7609 * When looking for a bad block we specify a range and want to
7610 * know if any block in the range is bad. So we binary-search
7611 * to the last range that starts at-or-before the given endpoint,
7612 * (or "before the sector after the target range")
7613 * then see if it ends after the given start.
7614 * We return
7615 * 0 if there are no known bad blocks in the range
7616 * 1 if there are known bad block which are all acknowledged
7617 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7618 * plus the start/length of the first bad section we overlap.
7619 */
7620int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7621 sector_t *first_bad, int *bad_sectors)
7622{
7623 int hi;
7624 int lo = 0;
7625 u64 *p = bb->page;
7626 int rv = 0;
7627 sector_t target = s + sectors;
7628 unsigned seq;
7629
7630 if (bb->shift > 0) {
7631 /* round the start down, and the end up */
7632 s >>= bb->shift;
7633 target += (1<<bb->shift) - 1;
7634 target >>= bb->shift;
7635 sectors = target - s;
7636 }
7637 /* 'target' is now the first block after the bad range */
7638
7639retry:
7640 seq = read_seqbegin(&bb->lock);
7641
7642 hi = bb->count;
7643
7644 /* Binary search between lo and hi for 'target'
7645 * i.e. for the last range that starts before 'target'
7646 */
7647 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7648 * are known not to be the last range before target.
7649 * VARIANT: hi-lo is the number of possible
7650 * ranges, and decreases until it reaches 1
7651 */
7652 while (hi - lo > 1) {
7653 int mid = (lo + hi) / 2;
7654 sector_t a = BB_OFFSET(p[mid]);
7655 if (a < target)
7656 /* This could still be the one, earlier ranges
7657 * could not. */
7658 lo = mid;
7659 else
7660 /* This and later ranges are definitely out. */
7661 hi = mid;
7662 }
7663 /* 'lo' might be the last that started before target, but 'hi' isn't */
7664 if (hi > lo) {
7665 /* need to check all range that end after 's' to see if
7666 * any are unacknowledged.
7667 */
7668 while (lo >= 0 &&
7669 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7670 if (BB_OFFSET(p[lo]) < target) {
7671 /* starts before the end, and finishes after
7672 * the start, so they must overlap
7673 */
7674 if (rv != -1 && BB_ACK(p[lo]))
7675 rv = 1;
7676 else
7677 rv = -1;
7678 *first_bad = BB_OFFSET(p[lo]);
7679 *bad_sectors = BB_LEN(p[lo]);
7680 }
7681 lo--;
7682 }
7683 }
7684
7685 if (read_seqretry(&bb->lock, seq))
7686 goto retry;
7687
7688 return rv;
7689}
7690EXPORT_SYMBOL_GPL(md_is_badblock);
7691
7692/*
7693 * Add a range of bad blocks to the table.
7694 * This might extend the table, or might contract it
7695 * if two adjacent ranges can be merged.
7696 * We binary-search to find the 'insertion' point, then
7697 * decide how best to handle it.
7698 */
7699static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7700 int acknowledged)
7701{
7702 u64 *p;
7703 int lo, hi;
7704 int rv = 1;
7705
7706 if (bb->shift < 0)
7707 /* badblocks are disabled */
7708 return 0;
7709
7710 if (bb->shift) {
7711 /* round the start down, and the end up */
7712 sector_t next = s + sectors;
7713 s >>= bb->shift;
7714 next += (1<<bb->shift) - 1;
7715 next >>= bb->shift;
7716 sectors = next - s;
7717 }
7718
7719 write_seqlock_irq(&bb->lock);
7720
7721 p = bb->page;
7722 lo = 0;
7723 hi = bb->count;
7724 /* Find the last range that starts at-or-before 's' */
7725 while (hi - lo > 1) {
7726 int mid = (lo + hi) / 2;
7727 sector_t a = BB_OFFSET(p[mid]);
7728 if (a <= s)
7729 lo = mid;
7730 else
7731 hi = mid;
7732 }
7733 if (hi > lo && BB_OFFSET(p[lo]) > s)
7734 hi = lo;
7735
7736 if (hi > lo) {
7737 /* we found a range that might merge with the start
7738 * of our new range
7739 */
7740 sector_t a = BB_OFFSET(p[lo]);
7741 sector_t e = a + BB_LEN(p[lo]);
7742 int ack = BB_ACK(p[lo]);
7743 if (e >= s) {
7744 /* Yes, we can merge with a previous range */
7745 if (s == a && s + sectors >= e)
7746 /* new range covers old */
7747 ack = acknowledged;
7748 else
7749 ack = ack && acknowledged;
7750
7751 if (e < s + sectors)
7752 e = s + sectors;
7753 if (e - a <= BB_MAX_LEN) {
7754 p[lo] = BB_MAKE(a, e-a, ack);
7755 s = e;
7756 } else {
7757 /* does not all fit in one range,
7758 * make p[lo] maximal
7759 */
7760 if (BB_LEN(p[lo]) != BB_MAX_LEN)
7761 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7762 s = a + BB_MAX_LEN;
7763 }
7764 sectors = e - s;
7765 }
7766 }
7767 if (sectors && hi < bb->count) {
7768 /* 'hi' points to the first range that starts after 's'.
7769 * Maybe we can merge with the start of that range */
7770 sector_t a = BB_OFFSET(p[hi]);
7771 sector_t e = a + BB_LEN(p[hi]);
7772 int ack = BB_ACK(p[hi]);
7773 if (a <= s + sectors) {
7774 /* merging is possible */
7775 if (e <= s + sectors) {
7776 /* full overlap */
7777 e = s + sectors;
7778 ack = acknowledged;
7779 } else
7780 ack = ack && acknowledged;
7781
7782 a = s;
7783 if (e - a <= BB_MAX_LEN) {
7784 p[hi] = BB_MAKE(a, e-a, ack);
7785 s = e;
7786 } else {
7787 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7788 s = a + BB_MAX_LEN;
7789 }
7790 sectors = e - s;
7791 lo = hi;
7792 hi++;
7793 }
7794 }
7795 if (sectors == 0 && hi < bb->count) {
7796 /* we might be able to combine lo and hi */
7797 /* Note: 's' is at the end of 'lo' */
7798 sector_t a = BB_OFFSET(p[hi]);
7799 int lolen = BB_LEN(p[lo]);
7800 int hilen = BB_LEN(p[hi]);
7801 int newlen = lolen + hilen - (s - a);
7802 if (s >= a && newlen < BB_MAX_LEN) {
7803 /* yes, we can combine them */
7804 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7805 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7806 memmove(p + hi, p + hi + 1,
7807 (bb->count - hi - 1) * 8);
7808 bb->count--;
7809 }
7810 }
7811 while (sectors) {
7812 /* didn't merge (it all).
7813 * Need to add a range just before 'hi' */
7814 if (bb->count >= MD_MAX_BADBLOCKS) {
7815 /* No room for more */
7816 rv = 0;
7817 break;
7818 } else {
7819 int this_sectors = sectors;
7820 memmove(p + hi + 1, p + hi,
7821 (bb->count - hi) * 8);
7822 bb->count++;
7823
7824 if (this_sectors > BB_MAX_LEN)
7825 this_sectors = BB_MAX_LEN;
7826 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7827 sectors -= this_sectors;
7828 s += this_sectors;
7829 }
7830 }
7831
7832 bb->changed = 1;
de393cde
N		 7833 if (!acknowledged)
7834 bb->unacked_exist = 1;
2230dfe4
N		 7835 write_sequnlock_irq(&bb->lock);
7836
7837 return rv;
7838}
7839
3cb03002 7840int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
2230dfe4
N		 7841 int acknowledged)
7842{
7843 int rv = md_set_badblocks(&rdev->badblocks,
7844 s + rdev->data_offset, sectors, acknowledged);
7845 if (rv) {
7846 /* Make sure they get written out promptly */
7847 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7848 md_wakeup_thread(rdev->mddev->thread);
7849 }
7850 return rv;
7851}
7852EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7853
7854/*
7855 * Remove a range of bad blocks from the table.
7856 * This may involve extending the table if we spilt a region,
7857 * but it must not fail. So if the table becomes full, we just
7858 * drop the remove request.
7859 */
7860static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7861{
7862 u64 *p;
7863 int lo, hi;
7864 sector_t target = s + sectors;
7865 int rv = 0;
7866
7867 if (bb->shift > 0) {
7868 /* When clearing we round the start up and the end down.
7869 * This should not matter as the shift should align with
7870 * the block size and no rounding should ever be needed.
7871 * However it is better the think a block is bad when it
7872 * isn't than to think a block is not bad when it is.
7873 */
7874 s += (1<<bb->shift) - 1;
7875 s >>= bb->shift;
7876 target >>= bb->shift;
7877 sectors = target - s;
7878 }
7879
7880 write_seqlock_irq(&bb->lock);
7881
7882 p = bb->page;
7883 lo = 0;
7884 hi = bb->count;
7885 /* Find the last range that starts before 'target' */
7886 while (hi - lo > 1) {
7887 int mid = (lo + hi) / 2;
7888 sector_t a = BB_OFFSET(p[mid]);
7889 if (a < target)
7890 lo = mid;
7891 else
7892 hi = mid;
7893 }
7894 if (hi > lo) {
7895 /* p[lo] is the last range that could overlap the
7896 * current range. Earlier ranges could also overlap,
7897 * but only this one can overlap the end of the range.
7898 */
7899 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7900 /* Partial overlap, leave the tail of this range */
7901 int ack = BB_ACK(p[lo]);
7902 sector_t a = BB_OFFSET(p[lo]);
7903 sector_t end = a + BB_LEN(p[lo]);
7904
7905 if (a < s) {
7906 /* we need to split this range */
7907 if (bb->count >= MD_MAX_BADBLOCKS) {
7908 rv = 0;
7909 goto out;
7910 }
7911 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7912 bb->count++;
7913 p[lo] = BB_MAKE(a, s-a, ack);
7914 lo++;
7915 }
7916 p[lo] = BB_MAKE(target, end - target, ack);
7917 /* there is no longer an overlap */
7918 hi = lo;
7919 lo--;
7920 }
7921 while (lo >= 0 &&
7922 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7923 /* This range does overlap */
7924 if (BB_OFFSET(p[lo]) < s) {
7925 /* Keep the early parts of this range. */
7926 int ack = BB_ACK(p[lo]);
7927 sector_t start = BB_OFFSET(p[lo]);
7928 p[lo] = BB_MAKE(start, s - start, ack);
7929 /* now low doesn't overlap, so.. */
7930 break;
7931 }
7932 lo--;
7933 }
7934 /* 'lo' is strictly before, 'hi' is strictly after,
7935 * anything between needs to be discarded
7936 */
7937 if (hi - lo > 1) {
7938 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7939 bb->count -= (hi - lo - 1);
7940 }
7941 }
7942
7943 bb->changed = 1;
7944out:
7945 write_sequnlock_irq(&bb->lock);
7946 return rv;
7947}
7948
3cb03002 7949int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors)
2230dfe4
N		 7950{
7951 return md_clear_badblocks(&rdev->badblocks,
7952 s + rdev->data_offset,
7953 sectors);
7954}
7955EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7956
7957/*
7958 * Acknowledge all bad blocks in a list.
7959 * This only succeeds if ->changed is clear. It is used by
7960 * in-kernel metadata updates
7961 */
7962void md_ack_all_badblocks(struct badblocks *bb)
7963{
7964 if (bb->page == NULL || bb->changed)
7965 /* no point even trying */
7966 return;
7967 write_seqlock_irq(&bb->lock);
7968
7969 if (bb->changed == 0) {
7970 u64 *p = bb->page;
7971 int i;
7972 for (i = 0; i < bb->count ; i++) {
7973 if (!BB_ACK(p[i])) {
7974 sector_t start = BB_OFFSET(p[i]);
7975 int len = BB_LEN(p[i]);
7976 p[i] = BB_MAKE(start, len, 1);
7977 }
7978 }
de393cde 7979 bb->unacked_exist = 0;
2230dfe4
N		 7980 }
7981 write_sequnlock_irq(&bb->lock);
7982}
7983EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
7984
16c791a5
N		 7985/* sysfs access to bad-blocks list.
7986 * We present two files.
7987 * 'bad-blocks' lists sector numbers and lengths of ranges that
7988 * are recorded as bad. The list is truncated to fit within
7989 * the one-page limit of sysfs.
7990 * Writing "sector length" to this file adds an acknowledged
7991 * bad block list.
7992 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7993 * been acknowledged. Writing to this file adds bad blocks
7994 * without acknowledging them. This is largely for testing.
7995 */
7996
7997static ssize_t
7998badblocks_show(struct badblocks *bb, char *page, int unack)
7999{
8000 size_t len;
8001 int i;
8002 u64 *p = bb->page;
8003 unsigned seq;
8004
8005 if (bb->shift < 0)
8006 return 0;
8007
8008retry:
8009 seq = read_seqbegin(&bb->lock);
8010
8011 len = 0;
8012 i = 0;
8013
8014 while (len < PAGE_SIZE && i < bb->count) {
8015 sector_t s = BB_OFFSET(p[i]);
8016 unsigned int length = BB_LEN(p[i]);
8017 int ack = BB_ACK(p[i]);
8018 i++;
8019
8020 if (unack && ack)
8021 continue;
8022
8023 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8024 (unsigned long long)s << bb->shift,
8025 length << bb->shift);
8026 }
de393cde
N		 8027 if (unack && len == 0)
8028 bb->unacked_exist = 0;
16c791a5
N		 8029
8030 if (read_seqretry(&bb->lock, seq))
8031 goto retry;
8032
8033 return len;
8034}
8035
8036#define DO_DEBUG 1
8037
8038static ssize_t
8039badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8040{
8041 unsigned long long sector;
8042 int length;
8043 char newline;
8044#ifdef DO_DEBUG
8045 /* Allow clearing via sysfs *only* for testing/debugging.
8046 * Normally only a successful write may clear a badblock
8047 */
8048 int clear = 0;
8049 if (page[0] == '-') {
8050 clear = 1;
8051 page++;
8052 }
8053#endif /* DO_DEBUG */
8054
8055 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8056 case 3:
8057 if (newline != '\n')
8058 return -EINVAL;
8059 case 2:
8060 if (length <= 0)
8061 return -EINVAL;
8062 break;
8063 default:
8064 return -EINVAL;
8065 }
8066
8067#ifdef DO_DEBUG
8068 if (clear) {
8069 md_clear_badblocks(bb, sector, length);
8070 return len;
8071 }
8072#endif /* DO_DEBUG */
8073 if (md_set_badblocks(bb, sector, length, !unack))
8074 return len;
8075 else
8076 return -ENOSPC;
8077}
8078
75c96f85
AB		 8079static int md_notify_reboot(struct notifier_block *this,
8080 unsigned long code, void *x)
1da177e4
LT		 8081{
8082 struct list_head *tmp;
fd01b88c 8083 struct mddev *mddev;
2dba6a91 8084 int need_delay = 0;
1da177e4
LT		 8085
8086 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8087
8088 printk(KERN_INFO "md: stopping all md devices.\n");
8089
2dba6a91 8090 for_each_mddev(mddev, tmp) {
c71d4887 8091 if (mddev_trylock(mddev)) {
2b25000b
N		 8092 /* Force a switch to readonly even array
8093 * appears to still be in use. Hence
8094 * the '100'.
8095 */
a4bd82d0 8096 md_set_readonly(mddev, 100);
c71d4887
NB		 8097 mddev_unlock(mddev);
8098 }
2dba6a91
DB		 8099 need_delay = 1;
8100 }
1da177e4
LT		 8101 /*
8102 * certain more exotic SCSI devices are known to be
8103 * volatile wrt too early system reboots. While the
8104 * right place to handle this issue is the given
8105 * driver, we do want to have a safe RAID driver ...
8106 */
2dba6a91
DB		 8107 if (need_delay)
8108 mdelay(1000*1);
1da177e4
LT		 8109 }
8110 return NOTIFY_DONE;
8111}
8112
75c96f85 8113static struct notifier_block md_notifier = {
1da177e4
LT		 8114 .notifier_call = md_notify_reboot,
8115 .next = NULL,
8116 .priority = INT_MAX, /* before any real devices */
8117};
8118
8119static void md_geninit(void)
8120{
36a4e1fe 8121 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
1da177e4 8122
c7705f34 8123 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
1da177e4
LT		 8124}
8125
75c96f85 8126static int __init md_init(void)
1da177e4 8127{
e804ac78
TH		 8128 int ret = -ENOMEM;
8129
ada609ee 8130 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
e804ac78
TH		 8131 if (!md_wq)
8132 goto err_wq;
8133
8134 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8135 if (!md_misc_wq)
8136 goto err_misc_wq;
8137
8138 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8139 goto err_md;
8140
8141 if ((ret = register_blkdev(0, "mdp")) < 0)
8142 goto err_mdp;
8143 mdp_major = ret;
8144
3dbd8c2e 8145 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
e8703fe1
N		 8146 md_probe, NULL, NULL);
8147 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
1da177e4
LT		 8148 md_probe, NULL, NULL);
8149
1da177e4 8150 register_reboot_notifier(&md_notifier);
0b4d4147 8151 raid_table_header = register_sysctl_table(raid_root_table);
1da177e4
LT		 8152
8153 md_geninit();
d710e138 8154 return 0;
1da177e4 8155
e804ac78
TH		 8156err_mdp:
8157 unregister_blkdev(MD_MAJOR, "md");
8158err_md:
8159 destroy_workqueue(md_misc_wq);
8160err_misc_wq:
8161 destroy_workqueue(md_wq);
8162err_wq:
8163 return ret;
8164}
1da177e4
LT		 8165
8166#ifndef MODULE
8167
8168/*
8169 * Searches all registered partitions for autorun RAID arrays
8170 * at boot time.
8171 */
4d936ec1
ME		 8172
8173static LIST_HEAD(all_detected_devices);
8174struct detected_devices_node {
8175 struct list_head list;
8176 dev_t dev;
8177};
1da177e4
LT		 8178
8179void md_autodetect_dev(dev_t dev)
8180{
4d936ec1
ME		 8181 struct detected_devices_node *node_detected_dev;
8182
8183 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8184 if (node_detected_dev) {
8185 node_detected_dev->dev = dev;
8186 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8187 } else {
8188 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8189 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8190 }
1da177e4
LT		 8191}
8192
8193
8194static void autostart_arrays(int part)
8195{
3cb03002 8196 struct md_rdev *rdev;
4d936ec1
ME		 8197 struct detected_devices_node *node_detected_dev;
8198 dev_t dev;
8199 int i_scanned, i_passed;
1da177e4 8200
4d936ec1
ME		 8201 i_scanned = 0;
8202 i_passed = 0;
1da177e4 8203
4d936ec1 8204 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
1da177e4 8205
4d936ec1
ME		 8206 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8207 i_scanned++;
8208 node_detected_dev = list_entry(all_detected_devices.next,
8209 struct detected_devices_node, list);
8210 list_del(&node_detected_dev->list);
8211 dev = node_detected_dev->dev;
8212 kfree(node_detected_dev);
df968c4e 8213 rdev = md_import_device(dev,0, 90);
1da177e4
LT		 8214 if (IS_ERR(rdev))
8215 continue;
8216
b2d444d7 8217 if (test_bit(Faulty, &rdev->flags)) {
1da177e4
LT		 8218 MD_BUG();
8219 continue;
8220 }
d0fae18f 8221 set_bit(AutoDetected, &rdev->flags);
1da177e4 8222 list_add(&rdev->same_set, &pending_raid_disks);
4d936ec1 8223 i_passed++;
1da177e4 8224 }
4d936ec1
ME		 8225
8226 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8227 i_scanned, i_passed);
1da177e4
LT		 8228
8229 autorun_devices(part);
8230}
8231
fdee8ae4 8232#endif /* !MODULE */
1da177e4
LT		 8233
8234static __exit void md_exit(void)
8235{
fd01b88c 8236 struct mddev *mddev;
1da177e4 8237 struct list_head *tmp;
8ab5e4c1 8238
3dbd8c2e 8239 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
e8703fe1 8240 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
1da177e4 8241
3dbd8c2e 8242 unregister_blkdev(MD_MAJOR,"md");
1da177e4
LT		 8243 unregister_blkdev(mdp_major, "mdp");
8244 unregister_reboot_notifier(&md_notifier);
8245 unregister_sysctl_table(raid_table_header);
8246 remove_proc_entry("mdstat", NULL);
29ac4aa3 8247 for_each_mddev(mddev, tmp) {
1da177e4 8248 export_array(mddev);
d3374825 8249 mddev->hold_active = 0;
1da177e4 8250 }
e804ac78
TH		 8251 destroy_workqueue(md_misc_wq);
8252 destroy_workqueue(md_wq);
1da177e4
LT		 8253}
8254
685784aa 8255subsys_initcall(md_init);
1da177e4
LT		 8256module_exit(md_exit)
8257
f91de92e
N		 8258static int get_ro(char *buffer, struct kernel_param *kp)
8259{
8260 return sprintf(buffer, "%d", start_readonly);
8261}
8262static int set_ro(const char *val, struct kernel_param *kp)
8263{
8264 char *e;
8265 int num = simple_strtoul(val, &e, 10);
8266 if (*val && (*e == '\0' || *e == '\n')) {
8267 start_readonly = num;
4dbcdc75 8268 return 0;
f91de92e
N		 8269 }
8270 return -EINVAL;
8271}
8272
80ca3a44
N		 8273module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8274module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6ff8d8ec 8275
efeb53c0 8276module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
f91de92e 8277
1da177e4
LT		 8278EXPORT_SYMBOL(register_md_personality);
8279EXPORT_SYMBOL(unregister_md_personality);
8280EXPORT_SYMBOL(md_error);
8281EXPORT_SYMBOL(md_done_sync);
8282EXPORT_SYMBOL(md_write_start);
8283EXPORT_SYMBOL(md_write_end);
1da177e4
LT		 8284EXPORT_SYMBOL(md_register_thread);
8285EXPORT_SYMBOL(md_unregister_thread);
8286EXPORT_SYMBOL(md_wakeup_thread);
1da177e4
LT		 8287EXPORT_SYMBOL(md_check_recovery);
8288MODULE_LICENSE("GPL");
0efb9e61 8289MODULE_DESCRIPTION("MD RAID framework");
aa1595e9 8290MODULE_ALIAS("md");
72008652 8291MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);
Linux kernel - Deliverables
This page took 1.572169 seconds and 5 git commands to generate.